1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Definitions for the TCP module.
7 *
8 * Version: @(#)tcp.h 1.0.5 05/23/93
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
17 */
18#ifndef _TCP_H
19#define _TCP_H
20
21#define FASTRETRANS_DEBUG 1
22
23#include <linux/list.h>
24#include <linux/tcp.h>
25#include <linux/bug.h>
26#include <linux/slab.h>
27#include <linux/cache.h>
28#include <linux/percpu.h>
29#include <linux/skbuff.h>
30#include <linux/cryptohash.h>
31#include <linux/kref.h>
32#include <linux/ktime.h>
33
34#include <net/inet_connection_sock.h>
35#include <net/inet_timewait_sock.h>
36#include <net/inet_hashtables.h>
37#include <net/checksum.h>
38#include <net/request_sock.h>
39#include <net/sock_reuseport.h>
40#include <net/sock.h>
41#include <net/snmp.h>
42#include <net/ip.h>
43#include <net/tcp_states.h>
44#include <net/inet_ecn.h>
45#include <net/dst.h>
46
47#include <linux/seq_file.h>
48#include <linux/memcontrol.h>
49#include <linux/bpf-cgroup.h>
50
51extern struct inet_hashinfo tcp_hashinfo;
52
53extern struct percpu_counter tcp_orphan_count;
54void tcp_time_wait(struct sock *sk, int state, int timeo);
55
56#define MAX_TCP_HEADER (128 + MAX_HEADER)
57#define MAX_TCP_OPTION_SPACE 40
58
59/*
60 * Never offer a window over 32767 without using window scaling. Some
61 * poor stacks do signed 16bit maths!
62 */
63#define MAX_TCP_WINDOW 32767U
64
65/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
66#define TCP_MIN_MSS 88U
67
68/* The least MTU to use for probing */
69#define TCP_BASE_MSS 1024
70
71/* probing interval, default to 10 minutes as per RFC4821 */
72#define TCP_PROBE_INTERVAL 600
73
74/* Specify interval when tcp mtu probing will stop */
75#define TCP_PROBE_THRESHOLD 8
76
77/* After receiving this amount of duplicate ACKs fast retransmit starts. */
78#define TCP_FASTRETRANS_THRESH 3
79
80/* Maximal number of ACKs sent quickly to accelerate slow-start. */
81#define TCP_MAX_QUICKACKS 16U
82
83/* Maximal number of window scale according to RFC1323 */
84#define TCP_MAX_WSCALE 14U
85
86/* urg_data states */
87#define TCP_URG_VALID 0x0100
88#define TCP_URG_NOTYET 0x0200
89#define TCP_URG_READ 0x0400
90
91#define TCP_RETR1 3 /*
92 * This is how many retries it does before it
93 * tries to figure out if the gateway is
94 * down. Minimal RFC value is 3; it corresponds
95 * to ~3sec-8min depending on RTO.
96 */
97
98#define TCP_RETR2 15 /*
99 * This should take at least
100 * 90 minutes to time out.
101 * RFC1122 says that the limit is 100 sec.
102 * 15 is ~13-30min depending on RTO.
103 */
104
105#define TCP_SYN_RETRIES 6 /* This is how many retries are done
106 * when active opening a connection.
107 * RFC1122 says the minimum retry MUST
108 * be at least 180secs. Nevertheless
109 * this value is corresponding to
110 * 63secs of retransmission with the
111 * current initial RTO.
112 */
113
114#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
115 * when passive opening a connection.
116 * This is corresponding to 31secs of
117 * retransmission with the current
118 * initial RTO.
119 */
120
121#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
122 * state, about 60 seconds */
123#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
124 /* BSD style FIN_WAIT2 deadlock breaker.
125 * It used to be 3min, new value is 60sec,
126 * to combine FIN-WAIT-2 timeout with
127 * TIME-WAIT timer.
128 */
129
130#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
131#if HZ >= 100
132#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
133#define TCP_ATO_MIN ((unsigned)(HZ/25))
134#else
135#define TCP_DELACK_MIN 4U
136#define TCP_ATO_MIN 4U
137#endif
138#define TCP_RTO_MAX ((unsigned)(120*HZ))
139#define TCP_RTO_MIN ((unsigned)(HZ/5))
140#define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
141#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
142#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
143 * used as a fallback RTO for the
144 * initial data transmission if no
145 * valid RTT sample has been acquired,
146 * most likely due to retrans in 3WHS.
147 */
148
149#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
150 * for local resources.
151 */
152#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
153#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
154#define TCP_KEEPALIVE_INTVL (75*HZ)
155
156#define MAX_TCP_KEEPIDLE 32767
157#define MAX_TCP_KEEPINTVL 32767
158#define MAX_TCP_KEEPCNT 127
159#define MAX_TCP_SYNCNT 127
160
161#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
162
163#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
164#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
165 * after this time. It should be equal
166 * (or greater than) TCP_TIMEWAIT_LEN
167 * to provide reliability equal to one
168 * provided by timewait state.
169 */
170#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
171 * timestamps. It must be less than
172 * minimal timewait lifetime.
173 */
174/*
175 * TCP option
176 */
177
178#define TCPOPT_NOP 1 /* Padding */
179#define TCPOPT_EOL 0 /* End of options */
180#define TCPOPT_MSS 2 /* Segment size negotiating */
181#define TCPOPT_WINDOW 3 /* Window scaling */
182#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
183#define TCPOPT_SACK 5 /* SACK Block */
184#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
185#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
186#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
187#define TCPOPT_EXP 254 /* Experimental */
188/* Magic number to be after the option value for sharing TCP
189 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
190 */
191#define TCPOPT_FASTOPEN_MAGIC 0xF989
192#define TCPOPT_SMC_MAGIC 0xE2D4C3D9
193
194/*
195 * TCP option lengths
196 */
197
198#define TCPOLEN_MSS 4
199#define TCPOLEN_WINDOW 3
200#define TCPOLEN_SACK_PERM 2
201#define TCPOLEN_TIMESTAMP 10
202#define TCPOLEN_MD5SIG 18
203#define TCPOLEN_FASTOPEN_BASE 2
204#define TCPOLEN_EXP_FASTOPEN_BASE 4
205#define TCPOLEN_EXP_SMC_BASE 6
206
207/* But this is what stacks really send out. */
208#define TCPOLEN_TSTAMP_ALIGNED 12
209#define TCPOLEN_WSCALE_ALIGNED 4
210#define TCPOLEN_SACKPERM_ALIGNED 4
211#define TCPOLEN_SACK_BASE 2
212#define TCPOLEN_SACK_BASE_ALIGNED 4
213#define TCPOLEN_SACK_PERBLOCK 8
214#define TCPOLEN_MD5SIG_ALIGNED 20
215#define TCPOLEN_MSS_ALIGNED 4
216#define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
217
218/* Flags in tp->nonagle */
219#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
220#define TCP_NAGLE_CORK 2 /* Socket is corked */
221#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
222
223/* TCP thin-stream limits */
224#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
225
226/* TCP initial congestion window as per rfc6928 */
227#define TCP_INIT_CWND 10
228
229/* Bit Flags for sysctl_tcp_fastopen */
230#define TFO_CLIENT_ENABLE 1
231#define TFO_SERVER_ENABLE 2
232#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
233
234/* Accept SYN data w/o any cookie option */
235#define TFO_SERVER_COOKIE_NOT_REQD 0x200
236
237/* Force enable TFO on all listeners, i.e., not requiring the
238 * TCP_FASTOPEN socket option.
239 */
240#define TFO_SERVER_WO_SOCKOPT1 0x400
241
242
243/* sysctl variables for tcp */
244extern int sysctl_tcp_max_orphans;
245extern long sysctl_tcp_mem[3];
246
247#define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
248#define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
249#define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
250
251extern atomic_long_t tcp_memory_allocated;
252extern struct percpu_counter tcp_sockets_allocated;
253extern unsigned long tcp_memory_pressure;
254
255/* optimized version of sk_under_memory_pressure() for TCP sockets */
256static inline bool tcp_under_memory_pressure(const struct sock *sk)
257{
258 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
259 mem_cgroup_under_socket_pressure(sk->sk_memcg))
260 return true;
261
262 return tcp_memory_pressure;
263}
264/*
265 * The next routines deal with comparing 32 bit unsigned ints
266 * and worry about wraparound (automatic with unsigned arithmetic).
267 */
268
269static inline bool before(__u32 seq1, __u32 seq2)
270{
271 return (__s32)(seq1-seq2) < 0;
272}
273#define after(seq2, seq1) before(seq1, seq2)
274
275/* is s2<=s1<=s3 ? */
276static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
277{
278 return seq3 - seq2 >= seq1 - seq2;
279}
280
281static inline bool tcp_out_of_memory(struct sock *sk)
282{
283 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
284 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
285 return true;
286 return false;
287}
288
289void sk_forced_mem_schedule(struct sock *sk, int size);
290
291static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
292{
293 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
294 int orphans = percpu_counter_read_positive(ocp);
295
296 if (orphans << shift > sysctl_tcp_max_orphans) {
297 orphans = percpu_counter_sum_positive(ocp);
298 if (orphans << shift > sysctl_tcp_max_orphans)
299 return true;
300 }
301 return false;
302}
303
304bool tcp_check_oom(struct sock *sk, int shift);
305
306
307extern struct proto tcp_prot;
308
309#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
310#define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
311#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
312#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
313
314void tcp_tasklet_init(void);
315
316int tcp_v4_err(struct sk_buff *skb, u32);
317
318void tcp_shutdown(struct sock *sk, int how);
319
320int tcp_v4_early_demux(struct sk_buff *skb);
321int tcp_v4_rcv(struct sk_buff *skb);
322
323int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
324int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
325int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
326int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
327 int flags);
328int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
329 size_t size, int flags);
330ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
331 size_t size, int flags);
332void tcp_release_cb(struct sock *sk);
333void tcp_wfree(struct sk_buff *skb);
334void tcp_write_timer_handler(struct sock *sk);
335void tcp_delack_timer_handler(struct sock *sk);
336int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
337int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
338void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
339void tcp_rcv_space_adjust(struct sock *sk);
340int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
341void tcp_twsk_destructor(struct sock *sk);
342ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
343 struct pipe_inode_info *pipe, size_t len,
344 unsigned int flags);
345
346void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
347static inline void tcp_dec_quickack_mode(struct sock *sk,
348 const unsigned int pkts)
349{
350 struct inet_connection_sock *icsk = inet_csk(sk);
351
352 if (icsk->icsk_ack.quick) {
353 if (pkts >= icsk->icsk_ack.quick) {
354 icsk->icsk_ack.quick = 0;
355 /* Leaving quickack mode we deflate ATO. */
356 icsk->icsk_ack.ato = TCP_ATO_MIN;
357 } else
358 icsk->icsk_ack.quick -= pkts;
359 }
360}
361
362#define TCP_ECN_OK 1
363#define TCP_ECN_QUEUE_CWR 2
364#define TCP_ECN_DEMAND_CWR 4
365#define TCP_ECN_SEEN 8
366
367enum tcp_tw_status {
368 TCP_TW_SUCCESS = 0,
369 TCP_TW_RST = 1,
370 TCP_TW_ACK = 2,
371 TCP_TW_SYN = 3
372};
373
374
375enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
376 struct sk_buff *skb,
377 const struct tcphdr *th);
378struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
379 struct request_sock *req, bool fastopen,
380 bool *lost_race);
381int tcp_child_process(struct sock *parent, struct sock *child,
382 struct sk_buff *skb);
383void tcp_enter_loss(struct sock *sk);
384void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
385void tcp_clear_retrans(struct tcp_sock *tp);
386void tcp_update_metrics(struct sock *sk);
387void tcp_init_metrics(struct sock *sk);
388void tcp_metrics_init(void);
389bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
390void tcp_close(struct sock *sk, long timeout);
391void tcp_init_sock(struct sock *sk);
392void tcp_init_transfer(struct sock *sk, int bpf_op);
393__poll_t tcp_poll(struct file *file, struct socket *sock,
394 struct poll_table_struct *wait);
395int tcp_getsockopt(struct sock *sk, int level, int optname,
396 char __user *optval, int __user *optlen);
397int tcp_setsockopt(struct sock *sk, int level, int optname,
398 char __user *optval, unsigned int optlen);
399int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
400 char __user *optval, int __user *optlen);
401int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
402 char __user *optval, unsigned int optlen);
403void tcp_set_keepalive(struct sock *sk, int val);
404void tcp_syn_ack_timeout(const struct request_sock *req);
405int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
406 int flags, int *addr_len);
407int tcp_set_rcvlowat(struct sock *sk, int val);
408void tcp_data_ready(struct sock *sk);
409#ifdef CONFIG_MMU
410int tcp_mmap(struct file *file, struct socket *sock,
411 struct vm_area_struct *vma);
412#endif
413void 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);
416const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
417
418/*
419 * TCP v4 functions exported for the inet6 API
420 */
421
422void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
423void tcp_v4_mtu_reduced(struct sock *sk);
424void tcp_req_err(struct sock *sk, u32 seq, bool abort);
425int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
426struct sock *tcp_create_openreq_child(const struct sock *sk,
427 struct request_sock *req,
428 struct sk_buff *skb);
429void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
430struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
431 struct request_sock *req,
432 struct dst_entry *dst,
433 struct request_sock *req_unhash,
434 bool *own_req);
435int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
436int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
437int tcp_connect(struct sock *sk);
438enum tcp_synack_type {
439 TCP_SYNACK_NORMAL,
440 TCP_SYNACK_FASTOPEN,
441 TCP_SYNACK_COOKIE,
442};
443struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
444 struct request_sock *req,
445 struct tcp_fastopen_cookie *foc,
446 enum tcp_synack_type synack_type);
447int tcp_disconnect(struct sock *sk, int flags);
448
449void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
450int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
451void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
452
453/* From syncookies.c */
454struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
455 struct request_sock *req,
456 struct dst_entry *dst, u32 tsoff);
457int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
458 u32 cookie);
459struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
460#ifdef CONFIG_SYN_COOKIES
461
462/* Syncookies use a monotonic timer which increments every 60 seconds.
463 * This counter is used both as a hash input and partially encoded into
464 * the cookie value. A cookie is only validated further if the delta
465 * between the current counter value and the encoded one is less than this,
466 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
467 * the counter advances immediately after a cookie is generated).
468 */
469#define MAX_SYNCOOKIE_AGE 2
470#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
471#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
472
473/* syncookies: remember time of last synqueue overflow
474 * But do not dirty this field too often (once per second is enough)
475 * It is racy as we do not hold a lock, but race is very minor.
476 */
477static inline void tcp_synq_overflow(const struct sock *sk)
478{
479 unsigned int last_overflow;
480 unsigned int now = jiffies;
481
482 if (sk->sk_reuseport) {
483 struct sock_reuseport *reuse;
484
485 reuse = rcu_dereference(sk->sk_reuseport_cb);
486 if (likely(reuse)) {
487 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
488 if (time_after32(now, last_overflow + HZ))
489 WRITE_ONCE(reuse->synq_overflow_ts, now);
490 return;
491 }
492 }
493
494 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
495 if (time_after32(now, last_overflow + HZ))
496 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
497}
498
499/* syncookies: no recent synqueue overflow on this listening socket? */
500static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
501{
502 unsigned int last_overflow;
503 unsigned int now = jiffies;
504
505 if (sk->sk_reuseport) {
506 struct sock_reuseport *reuse;
507
508 reuse = rcu_dereference(sk->sk_reuseport_cb);
509 if (likely(reuse)) {
510 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
511 return time_after32(now, last_overflow +
512 TCP_SYNCOOKIE_VALID);
513 }
514 }
515
516 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
517 return time_after32(now, last_overflow + TCP_SYNCOOKIE_VALID);
518}
519
520static inline u32 tcp_cookie_time(void)
521{
522 u64 val = get_jiffies_64();
523
524 do_div(val, TCP_SYNCOOKIE_PERIOD);
525 return val;
526}
527
528u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
529 u16 *mssp);
530__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
531u64 cookie_init_timestamp(struct request_sock *req);
532bool cookie_timestamp_decode(const struct net *net,
533 struct tcp_options_received *opt);
534bool cookie_ecn_ok(const struct tcp_options_received *opt,
535 const struct net *net, const struct dst_entry *dst);
536
537/* From net/ipv6/syncookies.c */
538int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
539 u32 cookie);
540struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
541
542u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
543 const struct tcphdr *th, u16 *mssp);
544__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
545#endif
546/* tcp_output.c */
547
548void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
549 int nonagle);
550int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
551int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
552void tcp_retransmit_timer(struct sock *sk);
553void tcp_xmit_retransmit_queue(struct sock *);
554void tcp_simple_retransmit(struct sock *);
555void tcp_enter_recovery(struct sock *sk, bool ece_ack);
556int tcp_trim_head(struct sock *, struct sk_buff *, u32);
557enum tcp_queue {
558 TCP_FRAG_IN_WRITE_QUEUE,
559 TCP_FRAG_IN_RTX_QUEUE,
560};
561int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
562 struct sk_buff *skb, u32 len,
563 unsigned int mss_now, gfp_t gfp);
564
565void tcp_send_probe0(struct sock *);
566void tcp_send_partial(struct sock *);
567int tcp_write_wakeup(struct sock *, int mib);
568void tcp_send_fin(struct sock *sk);
569void tcp_send_active_reset(struct sock *sk, gfp_t priority);
570int tcp_send_synack(struct sock *);
571void tcp_push_one(struct sock *, unsigned int mss_now);
572void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
573void tcp_send_ack(struct sock *sk);
574void tcp_send_delayed_ack(struct sock *sk);
575void tcp_send_loss_probe(struct sock *sk);
576bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
577void tcp_skb_collapse_tstamp(struct sk_buff *skb,
578 const struct sk_buff *next_skb);
579
580/* tcp_input.c */
581void tcp_rearm_rto(struct sock *sk);
582void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
583void tcp_reset(struct sock *sk);
584void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
585void tcp_fin(struct sock *sk);
586
587/* tcp_timer.c */
588void tcp_init_xmit_timers(struct sock *);
589static inline void tcp_clear_xmit_timers(struct sock *sk)
590{
591 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
592 __sock_put(sk);
593
594 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
595 __sock_put(sk);
596
597 inet_csk_clear_xmit_timers(sk);
598}
599
600unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
601unsigned int tcp_current_mss(struct sock *sk);
602
603/* Bound MSS / TSO packet size with the half of the window */
604static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
605{
606 int cutoff;
607
608 /* When peer uses tiny windows, there is no use in packetizing
609 * to sub-MSS pieces for the sake of SWS or making sure there
610 * are enough packets in the pipe for fast recovery.
611 *
612 * On the other hand, for extremely large MSS devices, handling
613 * smaller than MSS windows in this way does make sense.
614 */
615 if (tp->max_window > TCP_MSS_DEFAULT)
616 cutoff = (tp->max_window >> 1);
617 else
618 cutoff = tp->max_window;
619
620 if (cutoff && pktsize > cutoff)
621 return max_t(int, cutoff, 68U - tp->tcp_header_len);
622 else
623 return pktsize;
624}
625
626/* tcp.c */
627void tcp_get_info(struct sock *, struct tcp_info *);
628
629/* Read 'sendfile()'-style from a TCP socket */
630int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
631 sk_read_actor_t recv_actor);
632
633void tcp_initialize_rcv_mss(struct sock *sk);
634
635int tcp_mtu_to_mss(struct sock *sk, int pmtu);
636int tcp_mss_to_mtu(struct sock *sk, int mss);
637void tcp_mtup_init(struct sock *sk);
638void tcp_init_buffer_space(struct sock *sk);
639
640static inline void tcp_bound_rto(const struct sock *sk)
641{
642 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
643 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
644}
645
646static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
647{
648 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
649}
650
651static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
652{
653 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
654 ntohl(TCP_FLAG_ACK) |
655 snd_wnd);
656}
657
658static inline void tcp_fast_path_on(struct tcp_sock *tp)
659{
660 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
661}
662
663static inline void tcp_fast_path_check(struct sock *sk)
664{
665 struct tcp_sock *tp = tcp_sk(sk);
666
667 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
668 tp->rcv_wnd &&
669 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
670 !tp->urg_data)
671 tcp_fast_path_on(tp);
672}
673
674/* Compute the actual rto_min value */
675static inline u32 tcp_rto_min(struct sock *sk)
676{
677 const struct dst_entry *dst = __sk_dst_get(sk);
678 u32 rto_min = TCP_RTO_MIN;
679
680 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
681 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
682 return rto_min;
683}
684
685static inline u32 tcp_rto_min_us(struct sock *sk)
686{
687 return jiffies_to_usecs(tcp_rto_min(sk));
688}
689
690static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
691{
692 return dst_metric_locked(dst, RTAX_CC_ALGO);
693}
694
695/* Minimum RTT in usec. ~0 means not available. */
696static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
697{
698 return minmax_get(&tp->rtt_min);
699}
700
701/* Compute the actual receive window we are currently advertising.
702 * Rcv_nxt can be after the window if our peer push more data
703 * than the offered window.
704 */
705static inline u32 tcp_receive_window(const struct tcp_sock *tp)
706{
707 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
708
709 if (win < 0)
710 win = 0;
711 return (u32) win;
712}
713
714/* Choose a new window, without checks for shrinking, and without
715 * scaling applied to the result. The caller does these things
716 * if necessary. This is a "raw" window selection.
717 */
718u32 __tcp_select_window(struct sock *sk);
719
720void tcp_send_window_probe(struct sock *sk);
721
722/* TCP uses 32bit jiffies to save some space.
723 * Note that this is different from tcp_time_stamp, which
724 * historically has been the same until linux-4.13.
725 */
726#define tcp_jiffies32 ((u32)jiffies)
727
728/*
729 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
730 * It is no longer tied to jiffies, but to 1 ms clock.
731 * Note: double check if you want to use tcp_jiffies32 instead of this.
732 */
733#define TCP_TS_HZ 1000
734
735static inline u64 tcp_clock_ns(void)
736{
737 return ktime_get_ns();
738}
739
740static inline u64 tcp_clock_us(void)
741{
742 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
743}
744
745/* This should only be used in contexts where tp->tcp_mstamp is up to date */
746static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
747{
748 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
749}
750
751/* Could use tcp_clock_us() / 1000, but this version uses a single divide */
752static inline u32 tcp_time_stamp_raw(void)
753{
754 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
755}
756
757void tcp_mstamp_refresh(struct tcp_sock *tp);
758
759static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
760{
761 return max_t(s64, t1 - t0, 0);
762}
763
764static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
765{
766 return div_u64(skb->skb_mstamp_ns, NSEC_PER_SEC / TCP_TS_HZ);
767}
768
769/* provide the departure time in us unit */
770static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
771{
772 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
773}
774
775
776#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
777
778#define TCPHDR_FIN 0x01
779#define TCPHDR_SYN 0x02
780#define TCPHDR_RST 0x04
781#define TCPHDR_PSH 0x08
782#define TCPHDR_ACK 0x10
783#define TCPHDR_URG 0x20
784#define TCPHDR_ECE 0x40
785#define TCPHDR_CWR 0x80
786
787#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
788
789/* This is what the send packet queuing engine uses to pass
790 * TCP per-packet control information to the transmission code.
791 * We also store the host-order sequence numbers in here too.
792 * This is 44 bytes if IPV6 is enabled.
793 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
794 */
795struct tcp_skb_cb {
796 __u32 seq; /* Starting sequence number */
797 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
798 union {
799 /* Note : tcp_tw_isn is used in input path only
800 * (isn chosen by tcp_timewait_state_process())
801 *
802 * tcp_gso_segs/size are used in write queue only,
803 * cf tcp_skb_pcount()/tcp_skb_mss()
804 */
805 __u32 tcp_tw_isn;
806 struct {
807 u16 tcp_gso_segs;
808 u16 tcp_gso_size;
809 };
810 };
811 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
812
813 __u8 sacked; /* State flags for SACK. */
814#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
815#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
816#define TCPCB_LOST 0x04 /* SKB is lost */
817#define TCPCB_TAGBITS 0x07 /* All tag bits */
818#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
819#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
820#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
821 TCPCB_REPAIRED)
822
823 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
824 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
825 eor:1, /* Is skb MSG_EOR marked? */
826 has_rxtstamp:1, /* SKB has a RX timestamp */
827 unused:5;
828 __u32 ack_seq; /* Sequence number ACK'd */
829 union {
830 struct {
831 /* There is space for up to 24 bytes */
832 __u32 in_flight:30,/* Bytes in flight at transmit */
833 is_app_limited:1, /* cwnd not fully used? */
834 unused:1;
835 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
836 __u32 delivered;
837 /* start of send pipeline phase */
838 u64 first_tx_mstamp;
839 /* when we reached the "delivered" count */
840 u64 delivered_mstamp;
841 } tx; /* only used for outgoing skbs */
842 union {
843 struct inet_skb_parm h4;
844#if IS_ENABLED(CONFIG_IPV6)
845 struct inet6_skb_parm h6;
846#endif
847 } header; /* For incoming skbs */
848 struct {
849 __u32 flags;
850 struct sock *sk_redir;
851 void *data_end;
852 } bpf;
853 };
854};
855
856#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
857
858static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
859{
860 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
861}
862
863static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb)
864{
865 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS;
866}
867
868static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb)
869{
870 return TCP_SKB_CB(skb)->bpf.sk_redir;
871}
872
873static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb)
874{
875 TCP_SKB_CB(skb)->bpf.sk_redir = NULL;
876}
877
878#if IS_ENABLED(CONFIG_IPV6)
879/* This is the variant of inet6_iif() that must be used by TCP,
880 * as TCP moves IP6CB into a different location in skb->cb[]
881 */
882static inline int tcp_v6_iif(const struct sk_buff *skb)
883{
884 return TCP_SKB_CB(skb)->header.h6.iif;
885}
886
887static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
888{
889 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
890
891 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
892}
893
894/* TCP_SKB_CB reference means this can not be used from early demux */
895static inline int tcp_v6_sdif(const struct sk_buff *skb)
896{
897#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
898 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
899 return TCP_SKB_CB(skb)->header.h6.iif;
900#endif
901 return 0;
902}
903#endif
904
905static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
906{
907#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
908 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
909 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
910 return true;
911#endif
912 return false;
913}
914
915/* TCP_SKB_CB reference means this can not be used from early demux */
916static inline int tcp_v4_sdif(struct sk_buff *skb)
917{
918#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
919 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
920 return TCP_SKB_CB(skb)->header.h4.iif;
921#endif
922 return 0;
923}
924
925/* Due to TSO, an SKB can be composed of multiple actual
926 * packets. To keep these tracked properly, we use this.
927 */
928static inline int tcp_skb_pcount(const struct sk_buff *skb)
929{
930 return TCP_SKB_CB(skb)->tcp_gso_segs;
931}
932
933static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
934{
935 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
936}
937
938static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
939{
940 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
941}
942
943/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
944static inline int tcp_skb_mss(const struct sk_buff *skb)
945{
946 return TCP_SKB_CB(skb)->tcp_gso_size;
947}
948
949static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
950{
951 return likely(!TCP_SKB_CB(skb)->eor);
952}
953
954/* Events passed to congestion control interface */
955enum tcp_ca_event {
956 CA_EVENT_TX_START, /* first transmit when no packets in flight */
957 CA_EVENT_CWND_RESTART, /* congestion window restart */
958 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
959 CA_EVENT_LOSS, /* loss timeout */
960 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
961 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
962};
963
964/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
965enum tcp_ca_ack_event_flags {
966 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
967 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
968 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
969};
970
971/*
972 * Interface for adding new TCP congestion control handlers
973 */
974#define TCP_CA_NAME_MAX 16
975#define TCP_CA_MAX 128
976#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
977
978#define TCP_CA_UNSPEC 0
979
980/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
981#define TCP_CONG_NON_RESTRICTED 0x1
982/* Requires ECN/ECT set on all packets */
983#define TCP_CONG_NEEDS_ECN 0x2
984
985union tcp_cc_info;
986
987struct ack_sample {
988 u32 pkts_acked;
989 s32 rtt_us;
990 u32 in_flight;
991};
992
993/* A rate sample measures the number of (original/retransmitted) data
994 * packets delivered "delivered" over an interval of time "interval_us".
995 * The tcp_rate.c code fills in the rate sample, and congestion
996 * control modules that define a cong_control function to run at the end
997 * of ACK processing can optionally chose to consult this sample when
998 * setting cwnd and pacing rate.
999 * A sample is invalid if "delivered" or "interval_us" is negative.
1000 */
1001struct rate_sample {
1002 u64 prior_mstamp; /* starting timestamp for interval */
1003 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1004 s32 delivered; /* number of packets delivered over interval */
1005 long interval_us; /* time for tp->delivered to incr "delivered" */
1006 u32 snd_interval_us; /* snd interval for delivered packets */
1007 u32 rcv_interval_us; /* rcv interval for delivered packets */
1008 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1009 int losses; /* number of packets marked lost upon ACK */
1010 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1011 u32 prior_in_flight; /* in flight before this ACK */
1012 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1013 bool is_retrans; /* is sample from retransmission? */
1014 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1015};
1016
1017struct tcp_congestion_ops {
1018 struct list_head list;
1019 u32 key;
1020 u32 flags;
1021
1022 /* initialize private data (optional) */
1023 void (*init)(struct sock *sk);
1024 /* cleanup private data (optional) */
1025 void (*release)(struct sock *sk);
1026
1027 /* return slow start threshold (required) */
1028 u32 (*ssthresh)(struct sock *sk);
1029 /* do new cwnd calculation (required) */
1030 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1031 /* call before changing ca_state (optional) */
1032 void (*set_state)(struct sock *sk, u8 new_state);
1033 /* call when cwnd event occurs (optional) */
1034 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1035 /* call when ack arrives (optional) */
1036 void (*in_ack_event)(struct sock *sk, u32 flags);
1037 /* new value of cwnd after loss (required) */
1038 u32 (*undo_cwnd)(struct sock *sk);
1039 /* hook for packet ack accounting (optional) */
1040 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1041 /* override sysctl_tcp_min_tso_segs */
1042 u32 (*min_tso_segs)(struct sock *sk);
1043 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1044 u32 (*sndbuf_expand)(struct sock *sk);
1045 /* call when packets are delivered to update cwnd and pacing rate,
1046 * after all the ca_state processing. (optional)
1047 */
1048 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1049 /* get info for inet_diag (optional) */
1050 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1051 union tcp_cc_info *info);
1052
1053 char name[TCP_CA_NAME_MAX];
1054 struct module *owner;
1055};
1056
1057int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1058void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1059
1060void tcp_assign_congestion_control(struct sock *sk);
1061void tcp_init_congestion_control(struct sock *sk);
1062void tcp_cleanup_congestion_control(struct sock *sk);
1063int tcp_set_default_congestion_control(struct net *net, const char *name);
1064void tcp_get_default_congestion_control(struct net *net, char *name);
1065void tcp_get_available_congestion_control(char *buf, size_t len);
1066void tcp_get_allowed_congestion_control(char *buf, size_t len);
1067int tcp_set_allowed_congestion_control(char *allowed);
1068int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit);
1069u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1070void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1071
1072u32 tcp_reno_ssthresh(struct sock *sk);
1073u32 tcp_reno_undo_cwnd(struct sock *sk);
1074void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1075extern struct tcp_congestion_ops tcp_reno;
1076
1077struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1078u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1079#ifdef CONFIG_INET
1080char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1081#else
1082static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1083{
1084 return NULL;
1085}
1086#endif
1087
1088static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1089{
1090 const struct inet_connection_sock *icsk = inet_csk(sk);
1091
1092 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1093}
1094
1095static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1096{
1097 struct inet_connection_sock *icsk = inet_csk(sk);
1098
1099 if (icsk->icsk_ca_ops->set_state)
1100 icsk->icsk_ca_ops->set_state(sk, ca_state);
1101 icsk->icsk_ca_state = ca_state;
1102}
1103
1104static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1105{
1106 const struct inet_connection_sock *icsk = inet_csk(sk);
1107
1108 if (icsk->icsk_ca_ops->cwnd_event)
1109 icsk->icsk_ca_ops->cwnd_event(sk, event);
1110}
1111
1112/* From tcp_rate.c */
1113void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1114void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1115 struct rate_sample *rs);
1116void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1117 bool is_sack_reneg, struct rate_sample *rs);
1118void tcp_rate_check_app_limited(struct sock *sk);
1119
1120/* These functions determine how the current flow behaves in respect of SACK
1121 * handling. SACK is negotiated with the peer, and therefore it can vary
1122 * between different flows.
1123 *
1124 * tcp_is_sack - SACK enabled
1125 * tcp_is_reno - No SACK
1126 */
1127static inline int tcp_is_sack(const struct tcp_sock *tp)
1128{
1129 return likely(tp->rx_opt.sack_ok);
1130}
1131
1132static inline bool tcp_is_reno(const struct tcp_sock *tp)
1133{
1134 return !tcp_is_sack(tp);
1135}
1136
1137static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1138{
1139 return tp->sacked_out + tp->lost_out;
1140}
1141
1142/* This determines how many packets are "in the network" to the best
1143 * of our knowledge. In many cases it is conservative, but where
1144 * detailed information is available from the receiver (via SACK
1145 * blocks etc.) we can make more aggressive calculations.
1146 *
1147 * Use this for decisions involving congestion control, use just
1148 * tp->packets_out to determine if the send queue is empty or not.
1149 *
1150 * Read this equation as:
1151 *
1152 * "Packets sent once on transmission queue" MINUS
1153 * "Packets left network, but not honestly ACKed yet" PLUS
1154 * "Packets fast retransmitted"
1155 */
1156static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1157{
1158 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1159}
1160
1161#define TCP_INFINITE_SSTHRESH 0x7fffffff
1162
1163static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1164{
1165 return tp->snd_cwnd < tp->snd_ssthresh;
1166}
1167
1168static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1169{
1170 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1171}
1172
1173static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1174{
1175 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1176 (1 << inet_csk(sk)->icsk_ca_state);
1177}
1178
1179/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1180 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1181 * ssthresh.
1182 */
1183static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1184{
1185 const struct tcp_sock *tp = tcp_sk(sk);
1186
1187 if (tcp_in_cwnd_reduction(sk))
1188 return tp->snd_ssthresh;
1189 else
1190 return max(tp->snd_ssthresh,
1191 ((tp->snd_cwnd >> 1) +
1192 (tp->snd_cwnd >> 2)));
1193}
1194
1195/* Use define here intentionally to get WARN_ON location shown at the caller */
1196#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1197
1198void tcp_enter_cwr(struct sock *sk);
1199__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1200
1201/* The maximum number of MSS of available cwnd for which TSO defers
1202 * sending if not using sysctl_tcp_tso_win_divisor.
1203 */
1204static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1205{
1206 return 3;
1207}
1208
1209/* Returns end sequence number of the receiver's advertised window */
1210static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1211{
1212 return tp->snd_una + tp->snd_wnd;
1213}
1214
1215/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1216 * flexible approach. The RFC suggests cwnd should not be raised unless
1217 * it was fully used previously. And that's exactly what we do in
1218 * congestion avoidance mode. But in slow start we allow cwnd to grow
1219 * as long as the application has used half the cwnd.
1220 * Example :
1221 * cwnd is 10 (IW10), but application sends 9 frames.
1222 * We allow cwnd to reach 18 when all frames are ACKed.
1223 * This check is safe because it's as aggressive as slow start which already
1224 * risks 100% overshoot. The advantage is that we discourage application to
1225 * either send more filler packets or data to artificially blow up the cwnd
1226 * usage, and allow application-limited process to probe bw more aggressively.
1227 */
1228static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1229{
1230 const struct tcp_sock *tp = tcp_sk(sk);
1231
1232 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1233 if (tcp_in_slow_start(tp))
1234 return tp->snd_cwnd < 2 * tp->max_packets_out;
1235
1236 return tp->is_cwnd_limited;
1237}
1238
1239/* BBR congestion control needs pacing.
1240 * Same remark for SO_MAX_PACING_RATE.
1241 * sch_fq packet scheduler is efficiently handling pacing,
1242 * but is not always installed/used.
1243 * Return true if TCP stack should pace packets itself.
1244 */
1245static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1246{
1247 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1248}
1249
1250/* Return in jiffies the delay before one skb is sent.
1251 * If @skb is NULL, we look at EDT for next packet being sent on the socket.
1252 */
1253static inline unsigned long tcp_pacing_delay(const struct sock *sk,
1254 const struct sk_buff *skb)
1255{
1256 s64 pacing_delay = skb ? skb->tstamp : tcp_sk(sk)->tcp_wstamp_ns;
1257
1258 pacing_delay -= tcp_sk(sk)->tcp_clock_cache;
1259
1260 return pacing_delay > 0 ? nsecs_to_jiffies(pacing_delay) : 0;
1261}
1262
1263static inline void tcp_reset_xmit_timer(struct sock *sk,
1264 const int what,
1265 unsigned long when,
1266 const unsigned long max_when,
1267 const struct sk_buff *skb)
1268{
1269 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk, skb),
1270 max_when);
1271}
1272
1273/* Something is really bad, we could not queue an additional packet,
1274 * because qdisc is full or receiver sent a 0 window, or we are paced.
1275 * We do not want to add fuel to the fire, or abort too early,
1276 * so make sure the timer we arm now is at least 200ms in the future,
1277 * regardless of current icsk_rto value (as it could be ~2ms)
1278 */
1279static inline unsigned long tcp_probe0_base(const struct sock *sk)
1280{
1281 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1282}
1283
1284/* Variant of inet_csk_rto_backoff() used for zero window probes */
1285static inline unsigned long tcp_probe0_when(const struct sock *sk,
1286 unsigned long max_when)
1287{
1288 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1289
1290 return (unsigned long)min_t(u64, when, max_when);
1291}
1292
1293static inline void tcp_check_probe_timer(struct sock *sk)
1294{
1295 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1296 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1297 tcp_probe0_base(sk), TCP_RTO_MAX,
1298 NULL);
1299}
1300
1301static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1302{
1303 tp->snd_wl1 = seq;
1304}
1305
1306static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1307{
1308 tp->snd_wl1 = seq;
1309}
1310
1311/*
1312 * Calculate(/check) TCP checksum
1313 */
1314static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1315 __be32 daddr, __wsum base)
1316{
1317 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1318}
1319
1320static inline bool tcp_checksum_complete(struct sk_buff *skb)
1321{
1322 return !skb_csum_unnecessary(skb) &&
1323 __skb_checksum_complete(skb);
1324}
1325
1326bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1327int tcp_filter(struct sock *sk, struct sk_buff *skb);
1328void tcp_set_state(struct sock *sk, int state);
1329void tcp_done(struct sock *sk);
1330int tcp_abort(struct sock *sk, int err);
1331
1332static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1333{
1334 rx_opt->dsack = 0;
1335 rx_opt->num_sacks = 0;
1336}
1337
1338u32 tcp_default_init_rwnd(u32 mss);
1339void tcp_cwnd_restart(struct sock *sk, s32 delta);
1340
1341static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1342{
1343 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1344 struct tcp_sock *tp = tcp_sk(sk);
1345 s32 delta;
1346
1347 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1348 ca_ops->cong_control)
1349 return;
1350 delta = tcp_jiffies32 - tp->lsndtime;
1351 if (delta > inet_csk(sk)->icsk_rto)
1352 tcp_cwnd_restart(sk, delta);
1353}
1354
1355/* Determine a window scaling and initial window to offer. */
1356void tcp_select_initial_window(const struct sock *sk, int __space,
1357 __u32 mss, __u32 *rcv_wnd,
1358 __u32 *window_clamp, int wscale_ok,
1359 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1360
1361static inline int tcp_win_from_space(const struct sock *sk, int space)
1362{
1363 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1364
1365 return tcp_adv_win_scale <= 0 ?
1366 (space>>(-tcp_adv_win_scale)) :
1367 space - (space>>tcp_adv_win_scale);
1368}
1369
1370/* Note: caller must be prepared to deal with negative returns */
1371static inline int tcp_space(const struct sock *sk)
1372{
1373 return tcp_win_from_space(sk, sk->sk_rcvbuf - sk->sk_backlog.len -
1374 atomic_read(&sk->sk_rmem_alloc));
1375}
1376
1377static inline int tcp_full_space(const struct sock *sk)
1378{
1379 return tcp_win_from_space(sk, sk->sk_rcvbuf);
1380}
1381
1382extern void tcp_openreq_init_rwin(struct request_sock *req,
1383 const struct sock *sk_listener,
1384 const struct dst_entry *dst);
1385
1386void tcp_enter_memory_pressure(struct sock *sk);
1387void tcp_leave_memory_pressure(struct sock *sk);
1388
1389static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1390{
1391 struct net *net = sock_net((struct sock *)tp);
1392
1393 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1394}
1395
1396static inline int keepalive_time_when(const struct tcp_sock *tp)
1397{
1398 struct net *net = sock_net((struct sock *)tp);
1399
1400 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1401}
1402
1403static inline int keepalive_probes(const struct tcp_sock *tp)
1404{
1405 struct net *net = sock_net((struct sock *)tp);
1406
1407 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1408}
1409
1410static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1411{
1412 const struct inet_connection_sock *icsk = &tp->inet_conn;
1413
1414 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1415 tcp_jiffies32 - tp->rcv_tstamp);
1416}
1417
1418static inline int tcp_fin_time(const struct sock *sk)
1419{
1420 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1421 const int rto = inet_csk(sk)->icsk_rto;
1422
1423 if (fin_timeout < (rto << 2) - (rto >> 1))
1424 fin_timeout = (rto << 2) - (rto >> 1);
1425
1426 return fin_timeout;
1427}
1428
1429static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1430 int paws_win)
1431{
1432 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1433 return true;
1434 if (unlikely(!time_before32(ktime_get_seconds(),
1435 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1436 return true;
1437 /*
1438 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1439 * then following tcp messages have valid values. Ignore 0 value,
1440 * or else 'negative' tsval might forbid us to accept their packets.
1441 */
1442 if (!rx_opt->ts_recent)
1443 return true;
1444 return false;
1445}
1446
1447static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1448 int rst)
1449{
1450 if (tcp_paws_check(rx_opt, 0))
1451 return false;
1452
1453 /* RST segments are not recommended to carry timestamp,
1454 and, if they do, it is recommended to ignore PAWS because
1455 "their cleanup function should take precedence over timestamps."
1456 Certainly, it is mistake. It is necessary to understand the reasons
1457 of this constraint to relax it: if peer reboots, clock may go
1458 out-of-sync and half-open connections will not be reset.
1459 Actually, the problem would be not existing if all
1460 the implementations followed draft about maintaining clock
1461 via reboots. Linux-2.2 DOES NOT!
1462
1463 However, we can relax time bounds for RST segments to MSL.
1464 */
1465 if (rst && !time_before32(ktime_get_seconds(),
1466 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1467 return false;
1468 return true;
1469}
1470
1471bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1472 int mib_idx, u32 *last_oow_ack_time);
1473
1474static inline void tcp_mib_init(struct net *net)
1475{
1476 /* See RFC 2012 */
1477 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1478 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1479 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1480 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1481}
1482
1483/* from STCP */
1484static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1485{
1486 tp->lost_skb_hint = NULL;
1487}
1488
1489static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1490{
1491 tcp_clear_retrans_hints_partial(tp);
1492 tp->retransmit_skb_hint = NULL;
1493}
1494
1495union tcp_md5_addr {
1496 struct in_addr a4;
1497#if IS_ENABLED(CONFIG_IPV6)
1498 struct in6_addr a6;
1499#endif
1500};
1501
1502/* - key database */
1503struct tcp_md5sig_key {
1504 struct hlist_node node;
1505 u8 keylen;
1506 u8 family; /* AF_INET or AF_INET6 */
1507 union tcp_md5_addr addr;
1508 u8 prefixlen;
1509 u8 key[TCP_MD5SIG_MAXKEYLEN];
1510 struct rcu_head rcu;
1511};
1512
1513/* - sock block */
1514struct tcp_md5sig_info {
1515 struct hlist_head head;
1516 struct rcu_head rcu;
1517};
1518
1519/* - pseudo header */
1520struct tcp4_pseudohdr {
1521 __be32 saddr;
1522 __be32 daddr;
1523 __u8 pad;
1524 __u8 protocol;
1525 __be16 len;
1526};
1527
1528struct tcp6_pseudohdr {
1529 struct in6_addr saddr;
1530 struct in6_addr daddr;
1531 __be32 len;
1532 __be32 protocol; /* including padding */
1533};
1534
1535union tcp_md5sum_block {
1536 struct tcp4_pseudohdr ip4;
1537#if IS_ENABLED(CONFIG_IPV6)
1538 struct tcp6_pseudohdr ip6;
1539#endif
1540};
1541
1542/* - pool: digest algorithm, hash description and scratch buffer */
1543struct tcp_md5sig_pool {
1544 struct ahash_request *md5_req;
1545 void *scratch;
1546};
1547
1548/* - functions */
1549int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1550 const struct sock *sk, const struct sk_buff *skb);
1551int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1552 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1553 gfp_t gfp);
1554int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1555 int family, u8 prefixlen);
1556struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1557 const struct sock *addr_sk);
1558
1559#ifdef CONFIG_TCP_MD5SIG
1560#include <linux/jump_label.h>
1561extern struct static_key_false tcp_md5_needed;
1562struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk,
1563 const union tcp_md5_addr *addr,
1564 int family);
1565static inline struct tcp_md5sig_key *
1566tcp_md5_do_lookup(const struct sock *sk,
1567 const union tcp_md5_addr *addr,
1568 int family)
1569{
1570 if (!static_branch_unlikely(&tcp_md5_needed))
1571 return NULL;
1572 return __tcp_md5_do_lookup(sk, addr, family);
1573}
1574
1575#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1576#else
1577static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1578 const union tcp_md5_addr *addr,
1579 int family)
1580{
1581 return NULL;
1582}
1583#define tcp_twsk_md5_key(twsk) NULL
1584#endif
1585
1586bool tcp_alloc_md5sig_pool(void);
1587
1588struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1589static inline void tcp_put_md5sig_pool(void)
1590{
1591 local_bh_enable();
1592}
1593
1594int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1595 unsigned int header_len);
1596int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1597 const struct tcp_md5sig_key *key);
1598
1599/* From tcp_fastopen.c */
1600void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1601 struct tcp_fastopen_cookie *cookie);
1602void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1603 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1604 u16 try_exp);
1605struct tcp_fastopen_request {
1606 /* Fast Open cookie. Size 0 means a cookie request */
1607 struct tcp_fastopen_cookie cookie;
1608 struct msghdr *data; /* data in MSG_FASTOPEN */
1609 size_t size;
1610 int copied; /* queued in tcp_connect() */
1611 struct ubuf_info *uarg;
1612};
1613void tcp_free_fastopen_req(struct tcp_sock *tp);
1614void tcp_fastopen_destroy_cipher(struct sock *sk);
1615void tcp_fastopen_ctx_destroy(struct net *net);
1616int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1617 void *key, unsigned int len);
1618void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1619struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1620 struct request_sock *req,
1621 struct tcp_fastopen_cookie *foc,
1622 const struct dst_entry *dst);
1623void tcp_fastopen_init_key_once(struct net *net);
1624bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1625 struct tcp_fastopen_cookie *cookie);
1626bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1627#define TCP_FASTOPEN_KEY_LENGTH 16
1628
1629/* Fastopen key context */
1630struct tcp_fastopen_context {
1631 struct crypto_cipher *tfm;
1632 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1633 struct rcu_head rcu;
1634};
1635
1636extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1637void tcp_fastopen_active_disable(struct sock *sk);
1638bool tcp_fastopen_active_should_disable(struct sock *sk);
1639void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1640void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1641
1642/* Latencies incurred by various limits for a sender. They are
1643 * chronograph-like stats that are mutually exclusive.
1644 */
1645enum tcp_chrono {
1646 TCP_CHRONO_UNSPEC,
1647 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1648 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1649 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1650 __TCP_CHRONO_MAX,
1651};
1652
1653void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1654void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1655
1656/* This helper is needed, because skb->tcp_tsorted_anchor uses
1657 * the same memory storage than skb->destructor/_skb_refdst
1658 */
1659static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1660{
1661 skb->destructor = NULL;
1662 skb->_skb_refdst = 0UL;
1663}
1664
1665#define tcp_skb_tsorted_save(skb) { \
1666 unsigned long _save = skb->_skb_refdst; \
1667 skb->_skb_refdst = 0UL;
1668
1669#define tcp_skb_tsorted_restore(skb) \
1670 skb->_skb_refdst = _save; \
1671}
1672
1673void tcp_write_queue_purge(struct sock *sk);
1674
1675static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1676{
1677 return skb_rb_first(&sk->tcp_rtx_queue);
1678}
1679
1680static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1681{
1682 return skb_peek(&sk->sk_write_queue);
1683}
1684
1685static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1686{
1687 return skb_peek_tail(&sk->sk_write_queue);
1688}
1689
1690#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1691 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1692
1693static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1694{
1695 return skb_peek(&sk->sk_write_queue);
1696}
1697
1698static inline bool tcp_skb_is_last(const struct sock *sk,
1699 const struct sk_buff *skb)
1700{
1701 return skb_queue_is_last(&sk->sk_write_queue, skb);
1702}
1703
1704static inline bool tcp_write_queue_empty(const struct sock *sk)
1705{
1706 return skb_queue_empty(&sk->sk_write_queue);
1707}
1708
1709static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1710{
1711 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1712}
1713
1714static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1715{
1716 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1717}
1718
1719static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1720{
1721 __skb_queue_tail(&sk->sk_write_queue, skb);
1722
1723 /* Queue it, remembering where we must start sending. */
1724 if (sk->sk_write_queue.next == skb)
1725 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1726}
1727
1728/* Insert new before skb on the write queue of sk. */
1729static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1730 struct sk_buff *skb,
1731 struct sock *sk)
1732{
1733 __skb_queue_before(&sk->sk_write_queue, skb, new);
1734}
1735
1736static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1737{
1738 tcp_skb_tsorted_anchor_cleanup(skb);
1739 __skb_unlink(skb, &sk->sk_write_queue);
1740}
1741
1742void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1743
1744static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1745{
1746 tcp_skb_tsorted_anchor_cleanup(skb);
1747 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1748}
1749
1750static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1751{
1752 list_del(&skb->tcp_tsorted_anchor);
1753 tcp_rtx_queue_unlink(skb, sk);
1754 sk_wmem_free_skb(sk, skb);
1755}
1756
1757static inline void tcp_push_pending_frames(struct sock *sk)
1758{
1759 if (tcp_send_head(sk)) {
1760 struct tcp_sock *tp = tcp_sk(sk);
1761
1762 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1763 }
1764}
1765
1766/* Start sequence of the skb just after the highest skb with SACKed
1767 * bit, valid only if sacked_out > 0 or when the caller has ensured
1768 * validity by itself.
1769 */
1770static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1771{
1772 if (!tp->sacked_out)
1773 return tp->snd_una;
1774
1775 if (tp->highest_sack == NULL)
1776 return tp->snd_nxt;
1777
1778 return TCP_SKB_CB(tp->highest_sack)->seq;
1779}
1780
1781static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1782{
1783 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1784}
1785
1786static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1787{
1788 return tcp_sk(sk)->highest_sack;
1789}
1790
1791static inline void tcp_highest_sack_reset(struct sock *sk)
1792{
1793 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1794}
1795
1796/* Called when old skb is about to be deleted and replaced by new skb */
1797static inline void tcp_highest_sack_replace(struct sock *sk,
1798 struct sk_buff *old,
1799 struct sk_buff *new)
1800{
1801 if (old == tcp_highest_sack(sk))
1802 tcp_sk(sk)->highest_sack = new;
1803}
1804
1805/* This helper checks if socket has IP_TRANSPARENT set */
1806static inline bool inet_sk_transparent(const struct sock *sk)
1807{
1808 switch (sk->sk_state) {
1809 case TCP_TIME_WAIT:
1810 return inet_twsk(sk)->tw_transparent;
1811 case TCP_NEW_SYN_RECV:
1812 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1813 }
1814 return inet_sk(sk)->transparent;
1815}
1816
1817/* Determines whether this is a thin stream (which may suffer from
1818 * increased latency). Used to trigger latency-reducing mechanisms.
1819 */
1820static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1821{
1822 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1823}
1824
1825/* /proc */
1826enum tcp_seq_states {
1827 TCP_SEQ_STATE_LISTENING,
1828 TCP_SEQ_STATE_ESTABLISHED,
1829};
1830
1831void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1832void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1833void tcp_seq_stop(struct seq_file *seq, void *v);
1834
1835struct tcp_seq_afinfo {
1836 sa_family_t family;
1837};
1838
1839struct tcp_iter_state {
1840 struct seq_net_private p;
1841 enum tcp_seq_states state;
1842 struct sock *syn_wait_sk;
1843 int bucket, offset, sbucket, num;
1844 loff_t last_pos;
1845};
1846
1847extern struct request_sock_ops tcp_request_sock_ops;
1848extern struct request_sock_ops tcp6_request_sock_ops;
1849
1850void tcp_v4_destroy_sock(struct sock *sk);
1851
1852struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1853 netdev_features_t features);
1854struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1855int tcp_gro_complete(struct sk_buff *skb);
1856
1857void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1858
1859static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1860{
1861 struct net *net = sock_net((struct sock *)tp);
1862 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1863}
1864
1865/* @wake is one when sk_stream_write_space() calls us.
1866 * This sends EPOLLOUT only if notsent_bytes is half the limit.
1867 * This mimics the strategy used in sock_def_write_space().
1868 */
1869static inline bool tcp_stream_memory_free(const struct sock *sk, int wake)
1870{
1871 const struct tcp_sock *tp = tcp_sk(sk);
1872 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1873
1874 return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
1875}
1876
1877#ifdef CONFIG_PROC_FS
1878int tcp4_proc_init(void);
1879void tcp4_proc_exit(void);
1880#endif
1881
1882int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1883int tcp_conn_request(struct request_sock_ops *rsk_ops,
1884 const struct tcp_request_sock_ops *af_ops,
1885 struct sock *sk, struct sk_buff *skb);
1886
1887/* TCP af-specific functions */
1888struct tcp_sock_af_ops {
1889#ifdef CONFIG_TCP_MD5SIG
1890 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1891 const struct sock *addr_sk);
1892 int (*calc_md5_hash)(char *location,
1893 const struct tcp_md5sig_key *md5,
1894 const struct sock *sk,
1895 const struct sk_buff *skb);
1896 int (*md5_parse)(struct sock *sk,
1897 int optname,
1898 char __user *optval,
1899 int optlen);
1900#endif
1901};
1902
1903struct tcp_request_sock_ops {
1904 u16 mss_clamp;
1905#ifdef CONFIG_TCP_MD5SIG
1906 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1907 const struct sock *addr_sk);
1908 int (*calc_md5_hash) (char *location,
1909 const struct tcp_md5sig_key *md5,
1910 const struct sock *sk,
1911 const struct sk_buff *skb);
1912#endif
1913 void (*init_req)(struct request_sock *req,
1914 const struct sock *sk_listener,
1915 struct sk_buff *skb);
1916#ifdef CONFIG_SYN_COOKIES
1917 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1918 __u16 *mss);
1919#endif
1920 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1921 const struct request_sock *req);
1922 u32 (*init_seq)(const struct sk_buff *skb);
1923 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1924 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1925 struct flowi *fl, struct request_sock *req,
1926 struct tcp_fastopen_cookie *foc,
1927 enum tcp_synack_type synack_type);
1928};
1929
1930#ifdef CONFIG_SYN_COOKIES
1931static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1932 const struct sock *sk, struct sk_buff *skb,
1933 __u16 *mss)
1934{
1935 tcp_synq_overflow(sk);
1936 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1937 return ops->cookie_init_seq(skb, mss);
1938}
1939#else
1940static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1941 const struct sock *sk, struct sk_buff *skb,
1942 __u16 *mss)
1943{
1944 return 0;
1945}
1946#endif
1947
1948int tcpv4_offload_init(void);
1949
1950void tcp_v4_init(void);
1951void tcp_init(void);
1952
1953/* tcp_recovery.c */
1954void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
1955void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
1956extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
1957 u32 reo_wnd);
1958extern void tcp_rack_mark_lost(struct sock *sk);
1959extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1960 u64 xmit_time);
1961extern void tcp_rack_reo_timeout(struct sock *sk);
1962extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
1963
1964/* At how many usecs into the future should the RTO fire? */
1965static inline s64 tcp_rto_delta_us(const struct sock *sk)
1966{
1967 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
1968 u32 rto = inet_csk(sk)->icsk_rto;
1969 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
1970
1971 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
1972}
1973
1974/*
1975 * Save and compile IPv4 options, return a pointer to it
1976 */
1977static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
1978 struct sk_buff *skb)
1979{
1980 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1981 struct ip_options_rcu *dopt = NULL;
1982
1983 if (opt->optlen) {
1984 int opt_size = sizeof(*dopt) + opt->optlen;
1985
1986 dopt = kmalloc(opt_size, GFP_ATOMIC);
1987 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
1988 kfree(dopt);
1989 dopt = NULL;
1990 }
1991 }
1992 return dopt;
1993}
1994
1995/* locally generated TCP pure ACKs have skb->truesize == 2
1996 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1997 * This is much faster than dissecting the packet to find out.
1998 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1999 */
2000static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2001{
2002 return skb->truesize == 2;
2003}
2004
2005static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2006{
2007 skb->truesize = 2;
2008}
2009
2010static inline int tcp_inq(struct sock *sk)
2011{
2012 struct tcp_sock *tp = tcp_sk(sk);
2013 int answ;
2014
2015 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2016 answ = 0;
2017 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2018 !tp->urg_data ||
2019 before(tp->urg_seq, tp->copied_seq) ||
2020 !before(tp->urg_seq, tp->rcv_nxt)) {
2021
2022 answ = tp->rcv_nxt - tp->copied_seq;
2023
2024 /* Subtract 1, if FIN was received */
2025 if (answ && sock_flag(sk, SOCK_DONE))
2026 answ--;
2027 } else {
2028 answ = tp->urg_seq - tp->copied_seq;
2029 }
2030
2031 return answ;
2032}
2033
2034int tcp_peek_len(struct socket *sock);
2035
2036static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2037{
2038 u16 segs_in;
2039
2040 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2041 tp->segs_in += segs_in;
2042 if (skb->len > tcp_hdrlen(skb))
2043 tp->data_segs_in += segs_in;
2044}
2045
2046/*
2047 * TCP listen path runs lockless.
2048 * We forced "struct sock" to be const qualified to make sure
2049 * we don't modify one of its field by mistake.
2050 * Here, we increment sk_drops which is an atomic_t, so we can safely
2051 * make sock writable again.
2052 */
2053static inline void tcp_listendrop(const struct sock *sk)
2054{
2055 atomic_inc(&((struct sock *)sk)->sk_drops);
2056 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2057}
2058
2059enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2060
2061/*
2062 * Interface for adding Upper Level Protocols over TCP
2063 */
2064
2065#define TCP_ULP_NAME_MAX 16
2066#define TCP_ULP_MAX 128
2067#define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2068
2069struct tcp_ulp_ops {
2070 struct list_head list;
2071
2072 /* initialize ulp */
2073 int (*init)(struct sock *sk);
2074 /* cleanup ulp */
2075 void (*release)(struct sock *sk);
2076
2077 char name[TCP_ULP_NAME_MAX];
2078 struct module *owner;
2079};
2080int tcp_register_ulp(struct tcp_ulp_ops *type);
2081void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2082int tcp_set_ulp(struct sock *sk, const char *name);
2083void tcp_get_available_ulp(char *buf, size_t len);
2084void tcp_cleanup_ulp(struct sock *sk);
2085
2086#define MODULE_ALIAS_TCP_ULP(name) \
2087 __MODULE_INFO(alias, alias_userspace, name); \
2088 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2089
2090struct sk_msg;
2091struct sk_psock;
2092
2093int tcp_bpf_init(struct sock *sk);
2094void tcp_bpf_reinit(struct sock *sk);
2095int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2096 int flags);
2097int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
2098 int nonblock, int flags, int *addr_len);
2099int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock,
2100 struct msghdr *msg, int len, int flags);
2101
2102/* Call BPF_SOCK_OPS program that returns an int. If the return value
2103 * is < 0, then the BPF op failed (for example if the loaded BPF
2104 * program does not support the chosen operation or there is no BPF
2105 * program loaded).
2106 */
2107#ifdef CONFIG_BPF
2108static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2109{
2110 struct bpf_sock_ops_kern sock_ops;
2111 int ret;
2112
2113 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2114 if (sk_fullsock(sk)) {
2115 sock_ops.is_fullsock = 1;
2116 sock_owned_by_me(sk);
2117 }
2118
2119 sock_ops.sk = sk;
2120 sock_ops.op = op;
2121 if (nargs > 0)
2122 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2123
2124 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2125 if (ret == 0)
2126 ret = sock_ops.reply;
2127 else
2128 ret = -1;
2129 return ret;
2130}
2131
2132static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2133{
2134 u32 args[2] = {arg1, arg2};
2135
2136 return tcp_call_bpf(sk, op, 2, args);
2137}
2138
2139static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2140 u32 arg3)
2141{
2142 u32 args[3] = {arg1, arg2, arg3};
2143
2144 return tcp_call_bpf(sk, op, 3, args);
2145}
2146
2147#else
2148static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2149{
2150 return -EPERM;
2151}
2152
2153static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2154{
2155 return -EPERM;
2156}
2157
2158static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2159 u32 arg3)
2160{
2161 return -EPERM;
2162}
2163
2164#endif
2165
2166static inline u32 tcp_timeout_init(struct sock *sk)
2167{
2168 int timeout;
2169
2170 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2171
2172 if (timeout <= 0)
2173 timeout = TCP_TIMEOUT_INIT;
2174 return timeout;
2175}
2176
2177static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2178{
2179 int rwnd;
2180
2181 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2182
2183 if (rwnd < 0)
2184 rwnd = 0;
2185 return rwnd;
2186}
2187
2188static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2189{
2190 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2191}
2192
2193#if IS_ENABLED(CONFIG_SMC)
2194extern struct static_key_false tcp_have_smc;
2195#endif
2196
2197#if IS_ENABLED(CONFIG_TLS_DEVICE)
2198void clean_acked_data_enable(struct inet_connection_sock *icsk,
2199 void (*cad)(struct sock *sk, u32 ack_seq));
2200void clean_acked_data_disable(struct inet_connection_sock *icsk);
2201
2202#endif
2203
2204#endif /* _TCP_H */
2205