1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Definitions for the TCP module.
9 * Version: @(#)tcp.h 1.0.5 05/23/93
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
17 #define FASTRETRANS_DEBUG 1
19 #include <linux/list.h>
20 #include <linux/tcp.h>
21 #include <linux/bug.h>
22 #include <linux/slab.h>
23 #include <linux/cache.h>
24 #include <linux/percpu.h>
25 #include <linux/skbuff.h>
26 #include <linux/kref.h>
27 #include <linux/ktime.h>
28 #include <linux/indirect_call_wrapper.h>
30 #include <net/inet_connection_sock.h>
31 #include <net/inet_timewait_sock.h>
32 #include <net/inet_hashtables.h>
33 #include <net/checksum.h>
34 #include <net/request_sock.h>
35 #include <net/sock_reuseport.h>
39 #include <net/tcp_states.h>
40 #include <net/inet_ecn.h>
42 #include <net/mptcp.h>
44 #include <linux/seq_file.h>
45 #include <linux/memcontrol.h>
46 #include <linux/bpf-cgroup.h>
47 #include <linux/siphash.h>
49 extern struct inet_hashinfo tcp_hashinfo;
51 DECLARE_PER_CPU(unsigned int, tcp_orphan_count);
52 int tcp_orphan_count_sum(void);
54 void tcp_time_wait(struct sock *sk, int state, int timeo);
56 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER)
57 #define MAX_TCP_OPTION_SPACE 40
58 #define TCP_MIN_SND_MSS 48
59 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
62 * Never offer a window over 32767 without using window scaling. Some
63 * poor stacks do signed 16bit maths!
65 #define MAX_TCP_WINDOW 32767U
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS 88U
70 /* The initial MTU to use for probing */
71 #define TCP_BASE_MSS 1024
73 /* probing interval, default to 10 minutes as per RFC4821 */
74 #define TCP_PROBE_INTERVAL 600
76 /* Specify interval when tcp mtu probing will stop */
77 #define TCP_PROBE_THRESHOLD 8
79 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
80 #define TCP_FASTRETRANS_THRESH 3
82 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
83 #define TCP_MAX_QUICKACKS 16U
85 /* Maximal number of window scale according to RFC1323 */
86 #define TCP_MAX_WSCALE 14U
89 #define TCP_URG_VALID 0x0100
90 #define TCP_URG_NOTYET 0x0200
91 #define TCP_URG_READ 0x0400
93 #define TCP_RETR1 3 /*
94 * This is how many retries it does before it
95 * tries to figure out if the gateway is
96 * down. Minimal RFC value is 3; it corresponds
97 * to ~3sec-8min depending on RTO.
100 #define TCP_RETR2 15 /*
101 * This should take at least
102 * 90 minutes to time out.
103 * RFC1122 says that the limit is 100 sec.
104 * 15 is ~13-30min depending on RTO.
107 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
108 * when active opening a connection.
109 * RFC1122 says the minimum retry MUST
110 * be at least 180secs. Nevertheless
111 * this value is corresponding to
112 * 63secs of retransmission with the
113 * current initial RTO.
116 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
117 * when passive opening a connection.
118 * This is corresponding to 31secs of
119 * retransmission with the current
123 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
124 * state, about 60 seconds */
125 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
126 /* BSD style FIN_WAIT2 deadlock breaker.
127 * It used to be 3min, new value is 60sec,
128 * to combine FIN-WAIT-2 timeout with
131 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
133 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
135 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
136 #define TCP_ATO_MIN ((unsigned)(HZ/25))
138 #define TCP_DELACK_MIN 4U
139 #define TCP_ATO_MIN 4U
141 #define TCP_RTO_MAX ((unsigned)(120*HZ))
142 #define TCP_RTO_MIN ((unsigned)(HZ/5))
143 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
144 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
145 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
146 * used as a fallback RTO for the
147 * initial data transmission if no
148 * valid RTT sample has been acquired,
149 * most likely due to retrans in 3WHS.
152 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
153 * for local resources.
155 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
156 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
157 #define TCP_KEEPALIVE_INTVL (75*HZ)
159 #define MAX_TCP_KEEPIDLE 32767
160 #define MAX_TCP_KEEPINTVL 32767
161 #define MAX_TCP_KEEPCNT 127
162 #define MAX_TCP_SYNCNT 127
164 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
166 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
167 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
168 * after this time. It should be equal
169 * (or greater than) TCP_TIMEWAIT_LEN
170 * to provide reliability equal to one
171 * provided by timewait state.
173 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
174 * timestamps. It must be less than
175 * minimal timewait lifetime.
181 #define TCPOPT_NOP 1 /* Padding */
182 #define TCPOPT_EOL 0 /* End of options */
183 #define TCPOPT_MSS 2 /* Segment size negotiating */
184 #define TCPOPT_WINDOW 3 /* Window scaling */
185 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
186 #define TCPOPT_SACK 5 /* SACK Block */
187 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
188 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
189 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */
190 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
191 #define TCPOPT_EXP 254 /* Experimental */
192 /* Magic number to be after the option value for sharing TCP
193 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
195 #define TCPOPT_FASTOPEN_MAGIC 0xF989
196 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9
202 #define TCPOLEN_MSS 4
203 #define TCPOLEN_WINDOW 3
204 #define TCPOLEN_SACK_PERM 2
205 #define TCPOLEN_TIMESTAMP 10
206 #define TCPOLEN_MD5SIG 18
207 #define TCPOLEN_FASTOPEN_BASE 2
208 #define TCPOLEN_EXP_FASTOPEN_BASE 4
209 #define TCPOLEN_EXP_SMC_BASE 6
211 /* But this is what stacks really send out. */
212 #define TCPOLEN_TSTAMP_ALIGNED 12
213 #define TCPOLEN_WSCALE_ALIGNED 4
214 #define TCPOLEN_SACKPERM_ALIGNED 4
215 #define TCPOLEN_SACK_BASE 2
216 #define TCPOLEN_SACK_BASE_ALIGNED 4
217 #define TCPOLEN_SACK_PERBLOCK 8
218 #define TCPOLEN_MD5SIG_ALIGNED 20
219 #define TCPOLEN_MSS_ALIGNED 4
220 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
222 /* Flags in tp->nonagle */
223 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
224 #define TCP_NAGLE_CORK 2 /* Socket is corked */
225 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
227 /* TCP thin-stream limits */
228 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
230 /* TCP initial congestion window as per rfc6928 */
231 #define TCP_INIT_CWND 10
233 /* Bit Flags for sysctl_tcp_fastopen */
234 #define TFO_CLIENT_ENABLE 1
235 #define TFO_SERVER_ENABLE 2
236 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
238 /* Accept SYN data w/o any cookie option */
239 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
241 /* Force enable TFO on all listeners, i.e., not requiring the
242 * TCP_FASTOPEN socket option.
244 #define TFO_SERVER_WO_SOCKOPT1 0x400
247 /* sysctl variables for tcp */
248 extern int sysctl_tcp_max_orphans;
249 extern long sysctl_tcp_mem[3];
251 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
252 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
253 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
255 extern atomic_long_t tcp_memory_allocated;
256 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc);
258 extern struct percpu_counter tcp_sockets_allocated;
259 extern unsigned long tcp_memory_pressure;
261 /* optimized version of sk_under_memory_pressure() for TCP sockets */
262 static inline bool tcp_under_memory_pressure(const struct sock *sk)
264 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
265 mem_cgroup_under_socket_pressure(sk->sk_memcg))
268 return READ_ONCE(tcp_memory_pressure);
271 * The next routines deal with comparing 32 bit unsigned ints
272 * and worry about wraparound (automatic with unsigned arithmetic).
275 static inline bool before(__u32 seq1, __u32 seq2)
277 return (__s32)(seq1-seq2) < 0;
279 #define after(seq2, seq1) before(seq1, seq2)
281 /* is s2<=s1<=s3 ? */
282 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
284 return seq3 - seq2 >= seq1 - seq2;
287 static inline bool tcp_out_of_memory(struct sock *sk)
289 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
290 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
295 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
297 sk_wmem_queued_add(sk, -skb->truesize);
298 if (!skb_zcopy_pure(skb))
299 sk_mem_uncharge(sk, skb->truesize);
301 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb)));
305 void sk_forced_mem_schedule(struct sock *sk, int size);
307 bool tcp_check_oom(struct sock *sk, int shift);
310 extern struct proto tcp_prot;
312 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
313 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
314 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
315 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
317 void tcp_tasklet_init(void);
319 int tcp_v4_err(struct sk_buff *skb, u32);
321 void tcp_shutdown(struct sock *sk, int how);
323 int tcp_v4_early_demux(struct sk_buff *skb);
324 int tcp_v4_rcv(struct sk_buff *skb);
326 void tcp_remove_empty_skb(struct sock *sk);
327 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
328 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
329 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
330 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
332 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
333 size_t size, int flags);
334 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
335 size_t size, int flags);
336 int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
337 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
339 void tcp_release_cb(struct sock *sk);
340 void tcp_wfree(struct sk_buff *skb);
341 void tcp_write_timer_handler(struct sock *sk);
342 void tcp_delack_timer_handler(struct sock *sk);
343 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
344 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
345 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
346 void tcp_rcv_space_adjust(struct sock *sk);
347 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
348 void tcp_twsk_destructor(struct sock *sk);
349 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
350 struct pipe_inode_info *pipe, size_t len,
352 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
353 bool force_schedule);
355 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
356 static inline void tcp_dec_quickack_mode(struct sock *sk,
357 const unsigned int pkts)
359 struct inet_connection_sock *icsk = inet_csk(sk);
361 if (icsk->icsk_ack.quick) {
362 if (pkts >= icsk->icsk_ack.quick) {
363 icsk->icsk_ack.quick = 0;
364 /* Leaving quickack mode we deflate ATO. */
365 icsk->icsk_ack.ato = TCP_ATO_MIN;
367 icsk->icsk_ack.quick -= pkts;
372 #define TCP_ECN_QUEUE_CWR 2
373 #define TCP_ECN_DEMAND_CWR 4
374 #define TCP_ECN_SEEN 8
384 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
386 const struct tcphdr *th);
387 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
388 struct request_sock *req, bool fastopen,
390 int tcp_child_process(struct sock *parent, struct sock *child,
391 struct sk_buff *skb);
392 void tcp_enter_loss(struct sock *sk);
393 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
394 void tcp_clear_retrans(struct tcp_sock *tp);
395 void tcp_update_metrics(struct sock *sk);
396 void tcp_init_metrics(struct sock *sk);
397 void tcp_metrics_init(void);
398 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
399 void __tcp_close(struct sock *sk, long timeout);
400 void tcp_close(struct sock *sk, long timeout);
401 void tcp_init_sock(struct sock *sk);
402 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
403 __poll_t tcp_poll(struct file *file, struct socket *sock,
404 struct poll_table_struct *wait);
405 int tcp_getsockopt(struct sock *sk, int level, int optname,
406 char __user *optval, int __user *optlen);
407 bool tcp_bpf_bypass_getsockopt(int level, int optname);
408 int do_tcp_setsockopt(struct sock *sk, int level, int optname,
409 sockptr_t optval, unsigned int optlen);
410 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
411 unsigned int optlen);
412 void tcp_set_keepalive(struct sock *sk, int val);
413 void tcp_syn_ack_timeout(const struct request_sock *req);
414 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
415 int flags, int *addr_len);
416 int tcp_set_rcvlowat(struct sock *sk, int val);
417 int tcp_set_window_clamp(struct sock *sk, int val);
418 void tcp_update_recv_tstamps(struct sk_buff *skb,
419 struct scm_timestamping_internal *tss);
420 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
421 struct scm_timestamping_internal *tss);
422 void tcp_data_ready(struct sock *sk);
424 int tcp_mmap(struct file *file, struct socket *sock,
425 struct vm_area_struct *vma);
427 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
428 struct tcp_options_received *opt_rx,
429 int estab, struct tcp_fastopen_cookie *foc);
430 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
433 * BPF SKB-less helpers
435 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
436 struct tcphdr *th, u32 *cookie);
437 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
438 struct tcphdr *th, u32 *cookie);
439 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss);
440 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
441 const struct tcp_request_sock_ops *af_ops,
442 struct sock *sk, struct tcphdr *th);
444 * TCP v4 functions exported for the inet6 API
447 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
448 void tcp_v4_mtu_reduced(struct sock *sk);
449 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
450 void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
451 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
452 struct sock *tcp_create_openreq_child(const struct sock *sk,
453 struct request_sock *req,
454 struct sk_buff *skb);
455 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
456 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
457 struct request_sock *req,
458 struct dst_entry *dst,
459 struct request_sock *req_unhash,
461 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
462 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
463 int tcp_connect(struct sock *sk);
464 enum tcp_synack_type {
469 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
470 struct request_sock *req,
471 struct tcp_fastopen_cookie *foc,
472 enum tcp_synack_type synack_type,
473 struct sk_buff *syn_skb);
474 int tcp_disconnect(struct sock *sk, int flags);
476 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
477 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
478 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
480 /* From syncookies.c */
481 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
482 struct request_sock *req,
483 struct dst_entry *dst, u32 tsoff);
484 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
486 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
487 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
488 const struct tcp_request_sock_ops *af_ops,
489 struct sock *sk, struct sk_buff *skb);
490 #ifdef CONFIG_SYN_COOKIES
492 /* Syncookies use a monotonic timer which increments every 60 seconds.
493 * This counter is used both as a hash input and partially encoded into
494 * the cookie value. A cookie is only validated further if the delta
495 * between the current counter value and the encoded one is less than this,
496 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
497 * the counter advances immediately after a cookie is generated).
499 #define MAX_SYNCOOKIE_AGE 2
500 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
501 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
503 /* syncookies: remember time of last synqueue overflow
504 * But do not dirty this field too often (once per second is enough)
505 * It is racy as we do not hold a lock, but race is very minor.
507 static inline void tcp_synq_overflow(const struct sock *sk)
509 unsigned int last_overflow;
510 unsigned int now = jiffies;
512 if (sk->sk_reuseport) {
513 struct sock_reuseport *reuse;
515 reuse = rcu_dereference(sk->sk_reuseport_cb);
517 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
518 if (!time_between32(now, last_overflow,
520 WRITE_ONCE(reuse->synq_overflow_ts, now);
525 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
526 if (!time_between32(now, last_overflow, last_overflow + HZ))
527 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
530 /* syncookies: no recent synqueue overflow on this listening socket? */
531 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
533 unsigned int last_overflow;
534 unsigned int now = jiffies;
536 if (sk->sk_reuseport) {
537 struct sock_reuseport *reuse;
539 reuse = rcu_dereference(sk->sk_reuseport_cb);
541 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
542 return !time_between32(now, last_overflow - HZ,
544 TCP_SYNCOOKIE_VALID);
548 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
550 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
551 * then we're under synflood. However, we have to use
552 * 'last_overflow - HZ' as lower bound. That's because a concurrent
553 * tcp_synq_overflow() could update .ts_recent_stamp after we read
554 * jiffies but before we store .ts_recent_stamp into last_overflow,
555 * which could lead to rejecting a valid syncookie.
557 return !time_between32(now, last_overflow - HZ,
558 last_overflow + TCP_SYNCOOKIE_VALID);
561 static inline u32 tcp_cookie_time(void)
563 u64 val = get_jiffies_64();
565 do_div(val, TCP_SYNCOOKIE_PERIOD);
569 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
571 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
572 u64 cookie_init_timestamp(struct request_sock *req, u64 now);
573 bool cookie_timestamp_decode(const struct net *net,
574 struct tcp_options_received *opt);
575 bool cookie_ecn_ok(const struct tcp_options_received *opt,
576 const struct net *net, const struct dst_entry *dst);
578 /* From net/ipv6/syncookies.c */
579 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
581 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
583 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
584 const struct tcphdr *th, u16 *mssp);
585 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
589 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb);
590 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb);
591 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
593 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
594 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
595 void tcp_retransmit_timer(struct sock *sk);
596 void tcp_xmit_retransmit_queue(struct sock *);
597 void tcp_simple_retransmit(struct sock *);
598 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
599 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
601 TCP_FRAG_IN_WRITE_QUEUE,
602 TCP_FRAG_IN_RTX_QUEUE,
604 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
605 struct sk_buff *skb, u32 len,
606 unsigned int mss_now, gfp_t gfp);
608 void tcp_send_probe0(struct sock *);
609 void tcp_send_partial(struct sock *);
610 int tcp_write_wakeup(struct sock *, int mib);
611 void tcp_send_fin(struct sock *sk);
612 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
613 int tcp_send_synack(struct sock *);
614 void tcp_push_one(struct sock *, unsigned int mss_now);
615 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
616 void tcp_send_ack(struct sock *sk);
617 void tcp_send_delayed_ack(struct sock *sk);
618 void tcp_send_loss_probe(struct sock *sk);
619 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
620 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
621 const struct sk_buff *next_skb);
624 void tcp_rearm_rto(struct sock *sk);
625 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
626 void tcp_reset(struct sock *sk, struct sk_buff *skb);
627 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
628 void tcp_fin(struct sock *sk);
629 void tcp_check_space(struct sock *sk);
632 void tcp_init_xmit_timers(struct sock *);
633 static inline void tcp_clear_xmit_timers(struct sock *sk)
635 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
638 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
641 inet_csk_clear_xmit_timers(sk);
644 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
645 unsigned int tcp_current_mss(struct sock *sk);
646 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
648 /* Bound MSS / TSO packet size with the half of the window */
649 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
653 /* When peer uses tiny windows, there is no use in packetizing
654 * to sub-MSS pieces for the sake of SWS or making sure there
655 * are enough packets in the pipe for fast recovery.
657 * On the other hand, for extremely large MSS devices, handling
658 * smaller than MSS windows in this way does make sense.
660 if (tp->max_window > TCP_MSS_DEFAULT)
661 cutoff = (tp->max_window >> 1);
663 cutoff = tp->max_window;
665 if (cutoff && pktsize > cutoff)
666 return max_t(int, cutoff, 68U - tp->tcp_header_len);
672 void tcp_get_info(struct sock *, struct tcp_info *);
674 /* Read 'sendfile()'-style from a TCP socket */
675 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
676 sk_read_actor_t recv_actor);
677 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
678 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off);
679 void tcp_read_done(struct sock *sk, size_t len);
681 void tcp_initialize_rcv_mss(struct sock *sk);
683 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
684 int tcp_mss_to_mtu(struct sock *sk, int mss);
685 void tcp_mtup_init(struct sock *sk);
687 static inline void tcp_bound_rto(const struct sock *sk)
689 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
690 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
693 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
695 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
698 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
700 /* mptcp hooks are only on the slow path */
701 if (sk_is_mptcp((struct sock *)tp))
704 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
705 ntohl(TCP_FLAG_ACK) |
709 static inline void tcp_fast_path_on(struct tcp_sock *tp)
711 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
714 static inline void tcp_fast_path_check(struct sock *sk)
716 struct tcp_sock *tp = tcp_sk(sk);
718 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
720 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
722 tcp_fast_path_on(tp);
725 /* Compute the actual rto_min value */
726 static inline u32 tcp_rto_min(struct sock *sk)
728 const struct dst_entry *dst = __sk_dst_get(sk);
729 u32 rto_min = inet_csk(sk)->icsk_rto_min;
731 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
732 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
736 static inline u32 tcp_rto_min_us(struct sock *sk)
738 return jiffies_to_usecs(tcp_rto_min(sk));
741 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
743 return dst_metric_locked(dst, RTAX_CC_ALGO);
746 /* Minimum RTT in usec. ~0 means not available. */
747 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
749 return minmax_get(&tp->rtt_min);
752 /* Compute the actual receive window we are currently advertising.
753 * Rcv_nxt can be after the window if our peer push more data
754 * than the offered window.
756 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
758 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
765 /* Choose a new window, without checks for shrinking, and without
766 * scaling applied to the result. The caller does these things
767 * if necessary. This is a "raw" window selection.
769 u32 __tcp_select_window(struct sock *sk);
771 void tcp_send_window_probe(struct sock *sk);
773 /* TCP uses 32bit jiffies to save some space.
774 * Note that this is different from tcp_time_stamp, which
775 * historically has been the same until linux-4.13.
777 #define tcp_jiffies32 ((u32)jiffies)
780 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
781 * It is no longer tied to jiffies, but to 1 ms clock.
782 * Note: double check if you want to use tcp_jiffies32 instead of this.
784 #define TCP_TS_HZ 1000
786 static inline u64 tcp_clock_ns(void)
788 return ktime_get_ns();
791 static inline u64 tcp_clock_us(void)
793 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
796 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
797 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
799 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
802 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
803 static inline u32 tcp_ns_to_ts(u64 ns)
805 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
808 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
809 static inline u32 tcp_time_stamp_raw(void)
811 return tcp_ns_to_ts(tcp_clock_ns());
814 void tcp_mstamp_refresh(struct tcp_sock *tp);
816 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
818 return max_t(s64, t1 - t0, 0);
821 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
823 return tcp_ns_to_ts(skb->skb_mstamp_ns);
826 /* provide the departure time in us unit */
827 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
829 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
833 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
835 #define TCPHDR_FIN 0x01
836 #define TCPHDR_SYN 0x02
837 #define TCPHDR_RST 0x04
838 #define TCPHDR_PSH 0x08
839 #define TCPHDR_ACK 0x10
840 #define TCPHDR_URG 0x20
841 #define TCPHDR_ECE 0x40
842 #define TCPHDR_CWR 0x80
844 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
846 /* This is what the send packet queuing engine uses to pass
847 * TCP per-packet control information to the transmission code.
848 * We also store the host-order sequence numbers in here too.
849 * This is 44 bytes if IPV6 is enabled.
850 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
853 __u32 seq; /* Starting sequence number */
854 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
856 /* Note : tcp_tw_isn is used in input path only
857 * (isn chosen by tcp_timewait_state_process())
859 * tcp_gso_segs/size are used in write queue only,
860 * cf tcp_skb_pcount()/tcp_skb_mss()
868 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
870 __u8 sacked; /* State flags for SACK. */
871 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
872 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
873 #define TCPCB_LOST 0x04 /* SKB is lost */
874 #define TCPCB_TAGBITS 0x07 /* All tag bits */
875 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
876 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
877 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
880 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
881 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
882 eor:1, /* Is skb MSG_EOR marked? */
883 has_rxtstamp:1, /* SKB has a RX timestamp */
885 __u32 ack_seq; /* Sequence number ACK'd */
888 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1)
889 /* There is space for up to 24 bytes */
890 __u32 is_app_limited:1, /* cwnd not fully used? */
893 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
895 /* start of send pipeline phase */
897 /* when we reached the "delivered" count */
898 u64 delivered_mstamp;
899 } tx; /* only used for outgoing skbs */
901 struct inet_skb_parm h4;
902 #if IS_ENABLED(CONFIG_IPV6)
903 struct inet6_skb_parm h6;
905 } header; /* For incoming skbs */
909 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
911 extern const struct inet_connection_sock_af_ops ipv4_specific;
913 #if IS_ENABLED(CONFIG_IPV6)
914 /* This is the variant of inet6_iif() that must be used by TCP,
915 * as TCP moves IP6CB into a different location in skb->cb[]
917 static inline int tcp_v6_iif(const struct sk_buff *skb)
919 return TCP_SKB_CB(skb)->header.h6.iif;
922 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
924 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
926 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
929 /* TCP_SKB_CB reference means this can not be used from early demux */
930 static inline int tcp_v6_sdif(const struct sk_buff *skb)
932 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
933 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
934 return TCP_SKB_CB(skb)->header.h6.iif;
939 extern const struct inet_connection_sock_af_ops ipv6_specific;
941 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
942 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
943 void tcp_v6_early_demux(struct sk_buff *skb);
947 /* TCP_SKB_CB reference means this can not be used from early demux */
948 static inline int tcp_v4_sdif(struct sk_buff *skb)
950 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
951 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
952 return TCP_SKB_CB(skb)->header.h4.iif;
957 /* Due to TSO, an SKB can be composed of multiple actual
958 * packets. To keep these tracked properly, we use this.
960 static inline int tcp_skb_pcount(const struct sk_buff *skb)
962 return TCP_SKB_CB(skb)->tcp_gso_segs;
965 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
967 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
970 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
972 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
975 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
976 static inline int tcp_skb_mss(const struct sk_buff *skb)
978 return TCP_SKB_CB(skb)->tcp_gso_size;
981 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
983 return likely(!TCP_SKB_CB(skb)->eor);
986 static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
987 const struct sk_buff *from)
989 return likely(tcp_skb_can_collapse_to(to) &&
990 mptcp_skb_can_collapse(to, from) &&
991 skb_pure_zcopy_same(to, from));
994 /* Events passed to congestion control interface */
996 CA_EVENT_TX_START, /* first transmit when no packets in flight */
997 CA_EVENT_CWND_RESTART, /* congestion window restart */
998 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
999 CA_EVENT_LOSS, /* loss timeout */
1000 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
1001 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
1004 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1005 enum tcp_ca_ack_event_flags {
1006 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
1007 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
1008 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
1012 * Interface for adding new TCP congestion control handlers
1014 #define TCP_CA_NAME_MAX 16
1015 #define TCP_CA_MAX 128
1016 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
1018 #define TCP_CA_UNSPEC 0
1020 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1021 #define TCP_CONG_NON_RESTRICTED 0x1
1022 /* Requires ECN/ECT set on all packets */
1023 #define TCP_CONG_NEEDS_ECN 0x2
1024 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1034 /* A rate sample measures the number of (original/retransmitted) data
1035 * packets delivered "delivered" over an interval of time "interval_us".
1036 * The tcp_rate.c code fills in the rate sample, and congestion
1037 * control modules that define a cong_control function to run at the end
1038 * of ACK processing can optionally chose to consult this sample when
1039 * setting cwnd and pacing rate.
1040 * A sample is invalid if "delivered" or "interval_us" is negative.
1042 struct rate_sample {
1043 u64 prior_mstamp; /* starting timestamp for interval */
1044 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1045 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */
1046 s32 delivered; /* number of packets delivered over interval */
1047 s32 delivered_ce; /* number of packets delivered w/ CE marks*/
1048 long interval_us; /* time for tp->delivered to incr "delivered" */
1049 u32 snd_interval_us; /* snd interval for delivered packets */
1050 u32 rcv_interval_us; /* rcv interval for delivered packets */
1051 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1052 int losses; /* number of packets marked lost upon ACK */
1053 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1054 u32 prior_in_flight; /* in flight before this ACK */
1055 u32 last_end_seq; /* end_seq of most recently ACKed packet */
1056 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1057 bool is_retrans; /* is sample from retransmission? */
1058 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1061 struct tcp_congestion_ops {
1062 /* fast path fields are put first to fill one cache line */
1064 /* return slow start threshold (required) */
1065 u32 (*ssthresh)(struct sock *sk);
1067 /* do new cwnd calculation (required) */
1068 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1070 /* call before changing ca_state (optional) */
1071 void (*set_state)(struct sock *sk, u8 new_state);
1073 /* call when cwnd event occurs (optional) */
1074 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1076 /* call when ack arrives (optional) */
1077 void (*in_ack_event)(struct sock *sk, u32 flags);
1079 /* hook for packet ack accounting (optional) */
1080 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1082 /* override sysctl_tcp_min_tso_segs */
1083 u32 (*min_tso_segs)(struct sock *sk);
1085 /* call when packets are delivered to update cwnd and pacing rate,
1086 * after all the ca_state processing. (optional)
1088 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1091 /* new value of cwnd after loss (required) */
1092 u32 (*undo_cwnd)(struct sock *sk);
1093 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1094 u32 (*sndbuf_expand)(struct sock *sk);
1096 /* control/slow paths put last */
1097 /* get info for inet_diag (optional) */
1098 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1099 union tcp_cc_info *info);
1101 char name[TCP_CA_NAME_MAX];
1102 struct module *owner;
1103 struct list_head list;
1107 /* initialize private data (optional) */
1108 void (*init)(struct sock *sk);
1109 /* cleanup private data (optional) */
1110 void (*release)(struct sock *sk);
1111 } ____cacheline_aligned_in_smp;
1113 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1114 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1116 void tcp_assign_congestion_control(struct sock *sk);
1117 void tcp_init_congestion_control(struct sock *sk);
1118 void tcp_cleanup_congestion_control(struct sock *sk);
1119 int tcp_set_default_congestion_control(struct net *net, const char *name);
1120 void tcp_get_default_congestion_control(struct net *net, char *name);
1121 void tcp_get_available_congestion_control(char *buf, size_t len);
1122 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1123 int tcp_set_allowed_congestion_control(char *allowed);
1124 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1125 bool cap_net_admin);
1126 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1127 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1129 u32 tcp_reno_ssthresh(struct sock *sk);
1130 u32 tcp_reno_undo_cwnd(struct sock *sk);
1131 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1132 extern struct tcp_congestion_ops tcp_reno;
1134 struct tcp_congestion_ops *tcp_ca_find(const char *name);
1135 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1136 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1138 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1140 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1146 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1148 const struct inet_connection_sock *icsk = inet_csk(sk);
1150 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1153 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1155 const struct inet_connection_sock *icsk = inet_csk(sk);
1157 if (icsk->icsk_ca_ops->cwnd_event)
1158 icsk->icsk_ca_ops->cwnd_event(sk, event);
1161 /* From tcp_cong.c */
1162 void tcp_set_ca_state(struct sock *sk, const u8 ca_state);
1164 /* From tcp_rate.c */
1165 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1166 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1167 struct rate_sample *rs);
1168 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1169 bool is_sack_reneg, struct rate_sample *rs);
1170 void tcp_rate_check_app_limited(struct sock *sk);
1172 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
1174 return t1 > t2 || (t1 == t2 && after(seq1, seq2));
1177 /* These functions determine how the current flow behaves in respect of SACK
1178 * handling. SACK is negotiated with the peer, and therefore it can vary
1179 * between different flows.
1181 * tcp_is_sack - SACK enabled
1182 * tcp_is_reno - No SACK
1184 static inline int tcp_is_sack(const struct tcp_sock *tp)
1186 return likely(tp->rx_opt.sack_ok);
1189 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1191 return !tcp_is_sack(tp);
1194 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1196 return tp->sacked_out + tp->lost_out;
1199 /* This determines how many packets are "in the network" to the best
1200 * of our knowledge. In many cases it is conservative, but where
1201 * detailed information is available from the receiver (via SACK
1202 * blocks etc.) we can make more aggressive calculations.
1204 * Use this for decisions involving congestion control, use just
1205 * tp->packets_out to determine if the send queue is empty or not.
1207 * Read this equation as:
1209 * "Packets sent once on transmission queue" MINUS
1210 * "Packets left network, but not honestly ACKed yet" PLUS
1211 * "Packets fast retransmitted"
1213 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1215 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1218 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1220 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp)
1222 return tp->snd_cwnd;
1225 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val)
1227 WARN_ON_ONCE((int)val <= 0);
1231 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1233 return tcp_snd_cwnd(tp) < tp->snd_ssthresh;
1236 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1238 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1241 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1243 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1244 (1 << inet_csk(sk)->icsk_ca_state);
1247 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1248 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1251 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1253 const struct tcp_sock *tp = tcp_sk(sk);
1255 if (tcp_in_cwnd_reduction(sk))
1256 return tp->snd_ssthresh;
1258 return max(tp->snd_ssthresh,
1259 ((tcp_snd_cwnd(tp) >> 1) +
1260 (tcp_snd_cwnd(tp) >> 2)));
1263 /* Use define here intentionally to get WARN_ON location shown at the caller */
1264 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1266 void tcp_enter_cwr(struct sock *sk);
1267 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1269 /* The maximum number of MSS of available cwnd for which TSO defers
1270 * sending if not using sysctl_tcp_tso_win_divisor.
1272 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1277 /* Returns end sequence number of the receiver's advertised window */
1278 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1280 return tp->snd_una + tp->snd_wnd;
1283 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1284 * flexible approach. The RFC suggests cwnd should not be raised unless
1285 * it was fully used previously. And that's exactly what we do in
1286 * congestion avoidance mode. But in slow start we allow cwnd to grow
1287 * as long as the application has used half the cwnd.
1289 * cwnd is 10 (IW10), but application sends 9 frames.
1290 * We allow cwnd to reach 18 when all frames are ACKed.
1291 * This check is safe because it's as aggressive as slow start which already
1292 * risks 100% overshoot. The advantage is that we discourage application to
1293 * either send more filler packets or data to artificially blow up the cwnd
1294 * usage, and allow application-limited process to probe bw more aggressively.
1296 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1298 const struct tcp_sock *tp = tcp_sk(sk);
1300 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1301 if (tcp_in_slow_start(tp))
1302 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out;
1304 return tp->is_cwnd_limited;
1307 /* BBR congestion control needs pacing.
1308 * Same remark for SO_MAX_PACING_RATE.
1309 * sch_fq packet scheduler is efficiently handling pacing,
1310 * but is not always installed/used.
1311 * Return true if TCP stack should pace packets itself.
1313 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1315 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1318 /* Estimates in how many jiffies next packet for this flow can be sent.
1319 * Scheduling a retransmit timer too early would be silly.
1321 static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1323 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1325 return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1328 static inline void tcp_reset_xmit_timer(struct sock *sk,
1331 const unsigned long max_when)
1333 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
1337 /* Something is really bad, we could not queue an additional packet,
1338 * because qdisc is full or receiver sent a 0 window, or we are paced.
1339 * We do not want to add fuel to the fire, or abort too early,
1340 * so make sure the timer we arm now is at least 200ms in the future,
1341 * regardless of current icsk_rto value (as it could be ~2ms)
1343 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1345 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1348 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1349 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1350 unsigned long max_when)
1352 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1353 inet_csk(sk)->icsk_backoff);
1354 u64 when = (u64)tcp_probe0_base(sk) << backoff;
1356 return (unsigned long)min_t(u64, when, max_when);
1359 static inline void tcp_check_probe_timer(struct sock *sk)
1361 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1362 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1363 tcp_probe0_base(sk), TCP_RTO_MAX);
1366 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1371 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1377 * Calculate(/check) TCP checksum
1379 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1380 __be32 daddr, __wsum base)
1382 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1385 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1387 return !skb_csum_unnecessary(skb) &&
1388 __skb_checksum_complete(skb);
1391 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1392 enum skb_drop_reason *reason);
1395 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1396 void tcp_set_state(struct sock *sk, int state);
1397 void tcp_done(struct sock *sk);
1398 int tcp_abort(struct sock *sk, int err);
1400 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1403 rx_opt->num_sacks = 0;
1406 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1408 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1410 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1411 struct tcp_sock *tp = tcp_sk(sk);
1414 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) ||
1415 tp->packets_out || ca_ops->cong_control)
1417 delta = tcp_jiffies32 - tp->lsndtime;
1418 if (delta > inet_csk(sk)->icsk_rto)
1419 tcp_cwnd_restart(sk, delta);
1422 /* Determine a window scaling and initial window to offer. */
1423 void tcp_select_initial_window(const struct sock *sk, int __space,
1424 __u32 mss, __u32 *rcv_wnd,
1425 __u32 *window_clamp, int wscale_ok,
1426 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1428 static inline int tcp_win_from_space(const struct sock *sk, int space)
1430 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale);
1432 return tcp_adv_win_scale <= 0 ?
1433 (space>>(-tcp_adv_win_scale)) :
1434 space - (space>>tcp_adv_win_scale);
1437 /* Note: caller must be prepared to deal with negative returns */
1438 static inline int tcp_space(const struct sock *sk)
1440 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1441 READ_ONCE(sk->sk_backlog.len) -
1442 atomic_read(&sk->sk_rmem_alloc));
1445 static inline int tcp_full_space(const struct sock *sk)
1447 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1450 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk)
1452 int unused_mem = sk_unused_reserved_mem(sk);
1453 struct tcp_sock *tp = tcp_sk(sk);
1455 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss);
1457 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh,
1458 tcp_win_from_space(sk, unused_mem));
1461 void tcp_cleanup_rbuf(struct sock *sk, int copied);
1463 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1464 * If 87.5 % (7/8) of the space has been consumed, we want to override
1465 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1466 * len/truesize ratio.
1468 static inline bool tcp_rmem_pressure(const struct sock *sk)
1470 int rcvbuf, threshold;
1472 if (tcp_under_memory_pressure(sk))
1475 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1476 threshold = rcvbuf - (rcvbuf >> 3);
1478 return atomic_read(&sk->sk_rmem_alloc) > threshold;
1481 static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1483 const struct tcp_sock *tp = tcp_sk(sk);
1484 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1489 return (avail >= target) || tcp_rmem_pressure(sk) ||
1490 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1493 extern void tcp_openreq_init_rwin(struct request_sock *req,
1494 const struct sock *sk_listener,
1495 const struct dst_entry *dst);
1497 void tcp_enter_memory_pressure(struct sock *sk);
1498 void tcp_leave_memory_pressure(struct sock *sk);
1500 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1502 struct net *net = sock_net((struct sock *)tp);
1504 return tp->keepalive_intvl ? :
1505 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl);
1508 static inline int keepalive_time_when(const struct tcp_sock *tp)
1510 struct net *net = sock_net((struct sock *)tp);
1512 return tp->keepalive_time ? :
1513 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time);
1516 static inline int keepalive_probes(const struct tcp_sock *tp)
1518 struct net *net = sock_net((struct sock *)tp);
1520 return tp->keepalive_probes ? :
1521 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes);
1524 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1526 const struct inet_connection_sock *icsk = &tp->inet_conn;
1528 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1529 tcp_jiffies32 - tp->rcv_tstamp);
1532 static inline int tcp_fin_time(const struct sock *sk)
1534 int fin_timeout = tcp_sk(sk)->linger2 ? :
1535 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout);
1536 const int rto = inet_csk(sk)->icsk_rto;
1538 if (fin_timeout < (rto << 2) - (rto >> 1))
1539 fin_timeout = (rto << 2) - (rto >> 1);
1544 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1547 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1549 if (unlikely(!time_before32(ktime_get_seconds(),
1550 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1553 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1554 * then following tcp messages have valid values. Ignore 0 value,
1555 * or else 'negative' tsval might forbid us to accept their packets.
1557 if (!rx_opt->ts_recent)
1562 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1565 if (tcp_paws_check(rx_opt, 0))
1568 /* RST segments are not recommended to carry timestamp,
1569 and, if they do, it is recommended to ignore PAWS because
1570 "their cleanup function should take precedence over timestamps."
1571 Certainly, it is mistake. It is necessary to understand the reasons
1572 of this constraint to relax it: if peer reboots, clock may go
1573 out-of-sync and half-open connections will not be reset.
1574 Actually, the problem would be not existing if all
1575 the implementations followed draft about maintaining clock
1576 via reboots. Linux-2.2 DOES NOT!
1578 However, we can relax time bounds for RST segments to MSL.
1580 if (rst && !time_before32(ktime_get_seconds(),
1581 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1586 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1587 int mib_idx, u32 *last_oow_ack_time);
1589 static inline void tcp_mib_init(struct net *net)
1592 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1593 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1594 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1595 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1599 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1601 tp->lost_skb_hint = NULL;
1604 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1606 tcp_clear_retrans_hints_partial(tp);
1607 tp->retransmit_skb_hint = NULL;
1610 union tcp_md5_addr {
1612 #if IS_ENABLED(CONFIG_IPV6)
1617 /* - key database */
1618 struct tcp_md5sig_key {
1619 struct hlist_node node;
1621 u8 family; /* AF_INET or AF_INET6 */
1624 union tcp_md5_addr addr;
1625 int l3index; /* set if key added with L3 scope */
1626 u8 key[TCP_MD5SIG_MAXKEYLEN];
1627 struct rcu_head rcu;
1631 struct tcp_md5sig_info {
1632 struct hlist_head head;
1633 struct rcu_head rcu;
1636 /* - pseudo header */
1637 struct tcp4_pseudohdr {
1645 struct tcp6_pseudohdr {
1646 struct in6_addr saddr;
1647 struct in6_addr daddr;
1649 __be32 protocol; /* including padding */
1652 union tcp_md5sum_block {
1653 struct tcp4_pseudohdr ip4;
1654 #if IS_ENABLED(CONFIG_IPV6)
1655 struct tcp6_pseudohdr ip6;
1659 /* - pool: digest algorithm, hash description and scratch buffer */
1660 struct tcp_md5sig_pool {
1661 struct ahash_request *md5_req;
1666 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1667 const struct sock *sk, const struct sk_buff *skb);
1668 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1669 int family, u8 prefixlen, int l3index, u8 flags,
1670 const u8 *newkey, u8 newkeylen, gfp_t gfp);
1671 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1672 int family, u8 prefixlen, int l3index, u8 flags);
1673 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1674 const struct sock *addr_sk);
1676 #ifdef CONFIG_TCP_MD5SIG
1677 #include <linux/jump_label.h>
1678 extern struct static_key_false tcp_md5_needed;
1679 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1680 const union tcp_md5_addr *addr,
1682 static inline struct tcp_md5sig_key *
1683 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1684 const union tcp_md5_addr *addr, int family)
1686 if (!static_branch_unlikely(&tcp_md5_needed))
1688 return __tcp_md5_do_lookup(sk, l3index, addr, family);
1691 enum skb_drop_reason
1692 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1693 const void *saddr, const void *daddr,
1694 int family, int dif, int sdif);
1697 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1699 static inline struct tcp_md5sig_key *
1700 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1701 const union tcp_md5_addr *addr, int family)
1706 static inline enum skb_drop_reason
1707 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1708 const void *saddr, const void *daddr,
1709 int family, int dif, int sdif)
1711 return SKB_NOT_DROPPED_YET;
1713 #define tcp_twsk_md5_key(twsk) NULL
1716 bool tcp_alloc_md5sig_pool(void);
1718 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1719 static inline void tcp_put_md5sig_pool(void)
1724 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1725 unsigned int header_len);
1726 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1727 const struct tcp_md5sig_key *key);
1729 /* From tcp_fastopen.c */
1730 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1731 struct tcp_fastopen_cookie *cookie);
1732 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1733 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1735 struct tcp_fastopen_request {
1736 /* Fast Open cookie. Size 0 means a cookie request */
1737 struct tcp_fastopen_cookie cookie;
1738 struct msghdr *data; /* data in MSG_FASTOPEN */
1740 int copied; /* queued in tcp_connect() */
1741 struct ubuf_info *uarg;
1743 void tcp_free_fastopen_req(struct tcp_sock *tp);
1744 void tcp_fastopen_destroy_cipher(struct sock *sk);
1745 void tcp_fastopen_ctx_destroy(struct net *net);
1746 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1747 void *primary_key, void *backup_key);
1748 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1750 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1751 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1752 struct request_sock *req,
1753 struct tcp_fastopen_cookie *foc,
1754 const struct dst_entry *dst);
1755 void tcp_fastopen_init_key_once(struct net *net);
1756 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1757 struct tcp_fastopen_cookie *cookie);
1758 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1759 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1760 #define TCP_FASTOPEN_KEY_MAX 2
1761 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1762 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1764 /* Fastopen key context */
1765 struct tcp_fastopen_context {
1766 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
1768 struct rcu_head rcu;
1771 void tcp_fastopen_active_disable(struct sock *sk);
1772 bool tcp_fastopen_active_should_disable(struct sock *sk);
1773 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1774 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1776 /* Caller needs to wrap with rcu_read_(un)lock() */
1778 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1780 struct tcp_fastopen_context *ctx;
1782 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1784 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1789 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1790 const struct tcp_fastopen_cookie *orig)
1792 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1793 orig->len == foc->len &&
1794 !memcmp(orig->val, foc->val, foc->len))
1800 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1805 /* Latencies incurred by various limits for a sender. They are
1806 * chronograph-like stats that are mutually exclusive.
1810 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1811 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1812 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1816 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1817 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1819 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1820 * the same memory storage than skb->destructor/_skb_refdst
1822 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1824 skb->destructor = NULL;
1825 skb->_skb_refdst = 0UL;
1828 #define tcp_skb_tsorted_save(skb) { \
1829 unsigned long _save = skb->_skb_refdst; \
1830 skb->_skb_refdst = 0UL;
1832 #define tcp_skb_tsorted_restore(skb) \
1833 skb->_skb_refdst = _save; \
1836 void tcp_write_queue_purge(struct sock *sk);
1838 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1840 return skb_rb_first(&sk->tcp_rtx_queue);
1843 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1845 return skb_rb_last(&sk->tcp_rtx_queue);
1848 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1850 return skb_peek_tail(&sk->sk_write_queue);
1853 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1854 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1856 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1858 return skb_peek(&sk->sk_write_queue);
1861 static inline bool tcp_skb_is_last(const struct sock *sk,
1862 const struct sk_buff *skb)
1864 return skb_queue_is_last(&sk->sk_write_queue, skb);
1868 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1871 * Since the write queue can have a temporary empty skb in it,
1872 * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1874 static inline bool tcp_write_queue_empty(const struct sock *sk)
1876 const struct tcp_sock *tp = tcp_sk(sk);
1878 return tp->write_seq == tp->snd_nxt;
1881 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1883 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1886 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1888 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1891 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1893 __skb_queue_tail(&sk->sk_write_queue, skb);
1895 /* Queue it, remembering where we must start sending. */
1896 if (sk->sk_write_queue.next == skb)
1897 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1900 /* Insert new before skb on the write queue of sk. */
1901 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1902 struct sk_buff *skb,
1905 __skb_queue_before(&sk->sk_write_queue, skb, new);
1908 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1910 tcp_skb_tsorted_anchor_cleanup(skb);
1911 __skb_unlink(skb, &sk->sk_write_queue);
1914 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1916 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1918 tcp_skb_tsorted_anchor_cleanup(skb);
1919 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1922 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1924 list_del(&skb->tcp_tsorted_anchor);
1925 tcp_rtx_queue_unlink(skb, sk);
1926 tcp_wmem_free_skb(sk, skb);
1929 static inline void tcp_push_pending_frames(struct sock *sk)
1931 if (tcp_send_head(sk)) {
1932 struct tcp_sock *tp = tcp_sk(sk);
1934 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1938 /* Start sequence of the skb just after the highest skb with SACKed
1939 * bit, valid only if sacked_out > 0 or when the caller has ensured
1940 * validity by itself.
1942 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1944 if (!tp->sacked_out)
1947 if (tp->highest_sack == NULL)
1950 return TCP_SKB_CB(tp->highest_sack)->seq;
1953 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1955 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1958 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1960 return tcp_sk(sk)->highest_sack;
1963 static inline void tcp_highest_sack_reset(struct sock *sk)
1965 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1968 /* Called when old skb is about to be deleted and replaced by new skb */
1969 static inline void tcp_highest_sack_replace(struct sock *sk,
1970 struct sk_buff *old,
1971 struct sk_buff *new)
1973 if (old == tcp_highest_sack(sk))
1974 tcp_sk(sk)->highest_sack = new;
1977 /* This helper checks if socket has IP_TRANSPARENT set */
1978 static inline bool inet_sk_transparent(const struct sock *sk)
1980 switch (sk->sk_state) {
1982 return inet_twsk(sk)->tw_transparent;
1983 case TCP_NEW_SYN_RECV:
1984 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1986 return inet_sk(sk)->transparent;
1989 /* Determines whether this is a thin stream (which may suffer from
1990 * increased latency). Used to trigger latency-reducing mechanisms.
1992 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1994 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1998 enum tcp_seq_states {
1999 TCP_SEQ_STATE_LISTENING,
2000 TCP_SEQ_STATE_ESTABLISHED,
2003 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
2004 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
2005 void tcp_seq_stop(struct seq_file *seq, void *v);
2007 struct tcp_seq_afinfo {
2011 struct tcp_iter_state {
2012 struct seq_net_private p;
2013 enum tcp_seq_states state;
2014 struct sock *syn_wait_sk;
2015 int bucket, offset, sbucket, num;
2019 extern struct request_sock_ops tcp_request_sock_ops;
2020 extern struct request_sock_ops tcp6_request_sock_ops;
2022 void tcp_v4_destroy_sock(struct sock *sk);
2024 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
2025 netdev_features_t features);
2026 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
2027 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
2028 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
2029 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
2030 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
2031 int tcp_gro_complete(struct sk_buff *skb);
2033 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
2035 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
2037 struct net *net = sock_net((struct sock *)tp);
2038 return tp->notsent_lowat ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
2041 bool tcp_stream_memory_free(const struct sock *sk, int wake);
2043 #ifdef CONFIG_PROC_FS
2044 int tcp4_proc_init(void);
2045 void tcp4_proc_exit(void);
2048 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
2049 int tcp_conn_request(struct request_sock_ops *rsk_ops,
2050 const struct tcp_request_sock_ops *af_ops,
2051 struct sock *sk, struct sk_buff *skb);
2053 /* TCP af-specific functions */
2054 struct tcp_sock_af_ops {
2055 #ifdef CONFIG_TCP_MD5SIG
2056 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
2057 const struct sock *addr_sk);
2058 int (*calc_md5_hash)(char *location,
2059 const struct tcp_md5sig_key *md5,
2060 const struct sock *sk,
2061 const struct sk_buff *skb);
2062 int (*md5_parse)(struct sock *sk,
2069 struct tcp_request_sock_ops {
2071 #ifdef CONFIG_TCP_MD5SIG
2072 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2073 const struct sock *addr_sk);
2074 int (*calc_md5_hash) (char *location,
2075 const struct tcp_md5sig_key *md5,
2076 const struct sock *sk,
2077 const struct sk_buff *skb);
2079 #ifdef CONFIG_SYN_COOKIES
2080 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
2083 struct dst_entry *(*route_req)(const struct sock *sk,
2084 struct sk_buff *skb,
2086 struct request_sock *req);
2087 u32 (*init_seq)(const struct sk_buff *skb);
2088 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2089 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2090 struct flowi *fl, struct request_sock *req,
2091 struct tcp_fastopen_cookie *foc,
2092 enum tcp_synack_type synack_type,
2093 struct sk_buff *syn_skb);
2096 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2097 #if IS_ENABLED(CONFIG_IPV6)
2098 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2101 #ifdef CONFIG_SYN_COOKIES
2102 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2103 const struct sock *sk, struct sk_buff *skb,
2106 tcp_synq_overflow(sk);
2107 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2108 return ops->cookie_init_seq(skb, mss);
2111 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2112 const struct sock *sk, struct sk_buff *skb,
2119 int tcpv4_offload_init(void);
2121 void tcp_v4_init(void);
2122 void tcp_init(void);
2124 /* tcp_recovery.c */
2125 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2126 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2127 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2129 extern bool tcp_rack_mark_lost(struct sock *sk);
2130 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2132 extern void tcp_rack_reo_timeout(struct sock *sk);
2133 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2135 /* At how many usecs into the future should the RTO fire? */
2136 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2138 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2139 u32 rto = inet_csk(sk)->icsk_rto;
2140 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2142 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2146 * Save and compile IPv4 options, return a pointer to it
2148 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2149 struct sk_buff *skb)
2151 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2152 struct ip_options_rcu *dopt = NULL;
2155 int opt_size = sizeof(*dopt) + opt->optlen;
2157 dopt = kmalloc(opt_size, GFP_ATOMIC);
2158 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2166 /* locally generated TCP pure ACKs have skb->truesize == 2
2167 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2168 * This is much faster than dissecting the packet to find out.
2169 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2171 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2173 return skb->truesize == 2;
2176 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2181 static inline int tcp_inq(struct sock *sk)
2183 struct tcp_sock *tp = tcp_sk(sk);
2186 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2188 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2190 before(tp->urg_seq, tp->copied_seq) ||
2191 !before(tp->urg_seq, tp->rcv_nxt)) {
2193 answ = tp->rcv_nxt - tp->copied_seq;
2195 /* Subtract 1, if FIN was received */
2196 if (answ && sock_flag(sk, SOCK_DONE))
2199 answ = tp->urg_seq - tp->copied_seq;
2205 int tcp_peek_len(struct socket *sock);
2207 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2211 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2213 /* We update these fields while other threads might
2214 * read them from tcp_get_info()
2216 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in);
2217 if (skb->len > tcp_hdrlen(skb))
2218 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in);
2222 * TCP listen path runs lockless.
2223 * We forced "struct sock" to be const qualified to make sure
2224 * we don't modify one of its field by mistake.
2225 * Here, we increment sk_drops which is an atomic_t, so we can safely
2226 * make sock writable again.
2228 static inline void tcp_listendrop(const struct sock *sk)
2230 atomic_inc(&((struct sock *)sk)->sk_drops);
2231 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2234 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2237 * Interface for adding Upper Level Protocols over TCP
2240 #define TCP_ULP_NAME_MAX 16
2241 #define TCP_ULP_MAX 128
2242 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2244 struct tcp_ulp_ops {
2245 struct list_head list;
2247 /* initialize ulp */
2248 int (*init)(struct sock *sk);
2250 void (*update)(struct sock *sk, struct proto *p,
2251 void (*write_space)(struct sock *sk));
2253 void (*release)(struct sock *sk);
2255 int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2256 size_t (*get_info_size)(const struct sock *sk);
2258 void (*clone)(const struct request_sock *req, struct sock *newsk,
2259 const gfp_t priority);
2261 char name[TCP_ULP_NAME_MAX];
2262 struct module *owner;
2264 int tcp_register_ulp(struct tcp_ulp_ops *type);
2265 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2266 int tcp_set_ulp(struct sock *sk, const char *name);
2267 void tcp_get_available_ulp(char *buf, size_t len);
2268 void tcp_cleanup_ulp(struct sock *sk);
2269 void tcp_update_ulp(struct sock *sk, struct proto *p,
2270 void (*write_space)(struct sock *sk));
2272 #define MODULE_ALIAS_TCP_ULP(name) \
2273 __MODULE_INFO(alias, alias_userspace, name); \
2274 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2276 #ifdef CONFIG_NET_SOCK_MSG
2280 #ifdef CONFIG_BPF_SYSCALL
2281 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
2282 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
2283 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2284 #endif /* CONFIG_BPF_SYSCALL */
2286 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2288 #endif /* CONFIG_NET_SOCK_MSG */
2290 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2291 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2296 #ifdef CONFIG_CGROUP_BPF
2297 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2298 struct sk_buff *skb,
2299 unsigned int end_offset)
2302 skops->skb_data_end = skb->data + end_offset;
2305 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2306 struct sk_buff *skb,
2307 unsigned int end_offset)
2312 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2313 * is < 0, then the BPF op failed (for example if the loaded BPF
2314 * program does not support the chosen operation or there is no BPF
2318 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2320 struct bpf_sock_ops_kern sock_ops;
2323 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2324 if (sk_fullsock(sk)) {
2325 sock_ops.is_fullsock = 1;
2326 sock_owned_by_me(sk);
2332 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2334 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2336 ret = sock_ops.reply;
2342 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2344 u32 args[2] = {arg1, arg2};
2346 return tcp_call_bpf(sk, op, 2, args);
2349 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2352 u32 args[3] = {arg1, arg2, arg3};
2354 return tcp_call_bpf(sk, op, 3, args);
2358 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2363 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2368 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2376 static inline u32 tcp_timeout_init(struct sock *sk)
2380 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2383 timeout = TCP_TIMEOUT_INIT;
2384 return min_t(int, timeout, TCP_RTO_MAX);
2387 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2391 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2398 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2400 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2403 static inline void tcp_bpf_rtt(struct sock *sk)
2405 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2406 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2409 #if IS_ENABLED(CONFIG_SMC)
2410 extern struct static_key_false tcp_have_smc;
2413 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2414 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2415 void (*cad)(struct sock *sk, u32 ack_seq));
2416 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2417 void clean_acked_data_flush(void);
2420 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2421 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2422 const struct tcp_sock *tp)
2424 if (static_branch_unlikely(&tcp_tx_delay_enabled))
2425 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2428 /* Compute Earliest Departure Time for some control packets
2429 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2431 static inline u64 tcp_transmit_time(const struct sock *sk)
2433 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2434 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2435 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2437 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;