Loading...
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
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.
6 *
7 * Definitions for the TCP module.
8 *
9 * Version: @(#)tcp.h 1.0.5 05/23/93
10 *
11 * Authors: Ross Biro
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 */
14#ifndef _TCP_H
15#define _TCP_H
16
17#define FASTRETRANS_DEBUG 1
18
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/cryptohash.h>
27#include <linux/kref.h>
28#include <linux/ktime.h>
29
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>
36#include <net/sock.h>
37#include <net/snmp.h>
38#include <net/ip.h>
39#include <net/tcp_states.h>
40#include <net/inet_ecn.h>
41#include <net/dst.h>
42
43#include <linux/seq_file.h>
44#include <linux/memcontrol.h>
45#include <linux/bpf-cgroup.h>
46#include <linux/siphash.h>
47
48extern struct inet_hashinfo tcp_hashinfo;
49
50extern struct percpu_counter tcp_orphan_count;
51void tcp_time_wait(struct sock *sk, int state, int timeo);
52
53#define MAX_TCP_HEADER (128 + MAX_HEADER)
54#define MAX_TCP_OPTION_SPACE 40
55#define TCP_MIN_SND_MSS 48
56#define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
57
58/*
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
61 */
62#define MAX_TCP_WINDOW 32767U
63
64/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65#define TCP_MIN_MSS 88U
66
67/* The initial MTU to use for probing */
68#define TCP_BASE_MSS 1024
69
70/* probing interval, default to 10 minutes as per RFC4821 */
71#define TCP_PROBE_INTERVAL 600
72
73/* Specify interval when tcp mtu probing will stop */
74#define TCP_PROBE_THRESHOLD 8
75
76/* After receiving this amount of duplicate ACKs fast retransmit starts. */
77#define TCP_FASTRETRANS_THRESH 3
78
79/* Maximal number of ACKs sent quickly to accelerate slow-start. */
80#define TCP_MAX_QUICKACKS 16U
81
82/* Maximal number of window scale according to RFC1323 */
83#define TCP_MAX_WSCALE 14U
84
85/* urg_data states */
86#define TCP_URG_VALID 0x0100
87#define TCP_URG_NOTYET 0x0200
88#define TCP_URG_READ 0x0400
89
90#define TCP_RETR1 3 /*
91 * This is how many retries it does before it
92 * tries to figure out if the gateway is
93 * down. Minimal RFC value is 3; it corresponds
94 * to ~3sec-8min depending on RTO.
95 */
96
97#define TCP_RETR2 15 /*
98 * This should take at least
99 * 90 minutes to time out.
100 * RFC1122 says that the limit is 100 sec.
101 * 15 is ~13-30min depending on RTO.
102 */
103
104#define TCP_SYN_RETRIES 6 /* This is how many retries are done
105 * when active opening a connection.
106 * RFC1122 says the minimum retry MUST
107 * be at least 180secs. Nevertheless
108 * this value is corresponding to
109 * 63secs of retransmission with the
110 * current initial RTO.
111 */
112
113#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
114 * when passive opening a connection.
115 * This is corresponding to 31secs of
116 * retransmission with the current
117 * initial RTO.
118 */
119
120#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
121 * state, about 60 seconds */
122#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
123 /* BSD style FIN_WAIT2 deadlock breaker.
124 * It used to be 3min, new value is 60sec,
125 * to combine FIN-WAIT-2 timeout with
126 * TIME-WAIT timer.
127 */
128
129#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
130#if HZ >= 100
131#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
132#define TCP_ATO_MIN ((unsigned)(HZ/25))
133#else
134#define TCP_DELACK_MIN 4U
135#define TCP_ATO_MIN 4U
136#endif
137#define TCP_RTO_MAX ((unsigned)(120*HZ))
138#define TCP_RTO_MIN ((unsigned)(HZ/5))
139#define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
140#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
141#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
142 * used as a fallback RTO for the
143 * initial data transmission if no
144 * valid RTT sample has been acquired,
145 * most likely due to retrans in 3WHS.
146 */
147
148#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
149 * for local resources.
150 */
151#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
152#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
153#define TCP_KEEPALIVE_INTVL (75*HZ)
154
155#define MAX_TCP_KEEPIDLE 32767
156#define MAX_TCP_KEEPINTVL 32767
157#define MAX_TCP_KEEPCNT 127
158#define MAX_TCP_SYNCNT 127
159
160#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
161
162#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
163#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
164 * after this time. It should be equal
165 * (or greater than) TCP_TIMEWAIT_LEN
166 * to provide reliability equal to one
167 * provided by timewait state.
168 */
169#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
170 * timestamps. It must be less than
171 * minimal timewait lifetime.
172 */
173/*
174 * TCP option
175 */
176
177#define TCPOPT_NOP 1 /* Padding */
178#define TCPOPT_EOL 0 /* End of options */
179#define TCPOPT_MSS 2 /* Segment size negotiating */
180#define TCPOPT_WINDOW 3 /* Window scaling */
181#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
182#define TCPOPT_SACK 5 /* SACK Block */
183#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
184#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
185#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
186#define TCPOPT_EXP 254 /* Experimental */
187/* Magic number to be after the option value for sharing TCP
188 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
189 */
190#define TCPOPT_FASTOPEN_MAGIC 0xF989
191#define TCPOPT_SMC_MAGIC 0xE2D4C3D9
192
193/*
194 * TCP option lengths
195 */
196
197#define TCPOLEN_MSS 4
198#define TCPOLEN_WINDOW 3
199#define TCPOLEN_SACK_PERM 2
200#define TCPOLEN_TIMESTAMP 10
201#define TCPOLEN_MD5SIG 18
202#define TCPOLEN_FASTOPEN_BASE 2
203#define TCPOLEN_EXP_FASTOPEN_BASE 4
204#define TCPOLEN_EXP_SMC_BASE 6
205
206/* But this is what stacks really send out. */
207#define TCPOLEN_TSTAMP_ALIGNED 12
208#define TCPOLEN_WSCALE_ALIGNED 4
209#define TCPOLEN_SACKPERM_ALIGNED 4
210#define TCPOLEN_SACK_BASE 2
211#define TCPOLEN_SACK_BASE_ALIGNED 4
212#define TCPOLEN_SACK_PERBLOCK 8
213#define TCPOLEN_MD5SIG_ALIGNED 20
214#define TCPOLEN_MSS_ALIGNED 4
215#define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
216
217/* Flags in tp->nonagle */
218#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
219#define TCP_NAGLE_CORK 2 /* Socket is corked */
220#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
221
222/* TCP thin-stream limits */
223#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
224
225/* TCP initial congestion window as per rfc6928 */
226#define TCP_INIT_CWND 10
227
228/* Bit Flags for sysctl_tcp_fastopen */
229#define TFO_CLIENT_ENABLE 1
230#define TFO_SERVER_ENABLE 2
231#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
232
233/* Accept SYN data w/o any cookie option */
234#define TFO_SERVER_COOKIE_NOT_REQD 0x200
235
236/* Force enable TFO on all listeners, i.e., not requiring the
237 * TCP_FASTOPEN socket option.
238 */
239#define TFO_SERVER_WO_SOCKOPT1 0x400
240
241
242/* sysctl variables for tcp */
243extern int sysctl_tcp_max_orphans;
244extern long sysctl_tcp_mem[3];
245
246#define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
247#define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
248#define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
249
250extern atomic_long_t tcp_memory_allocated;
251extern struct percpu_counter tcp_sockets_allocated;
252extern unsigned long tcp_memory_pressure;
253
254/* optimized version of sk_under_memory_pressure() for TCP sockets */
255static inline bool tcp_under_memory_pressure(const struct sock *sk)
256{
257 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
258 mem_cgroup_under_socket_pressure(sk->sk_memcg))
259 return true;
260
261 return READ_ONCE(tcp_memory_pressure);
262}
263/*
264 * The next routines deal with comparing 32 bit unsigned ints
265 * and worry about wraparound (automatic with unsigned arithmetic).
266 */
267
268static inline bool before(__u32 seq1, __u32 seq2)
269{
270 return (__s32)(seq1-seq2) < 0;
271}
272#define after(seq2, seq1) before(seq1, seq2)
273
274/* is s2<=s1<=s3 ? */
275static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
276{
277 return seq3 - seq2 >= seq1 - seq2;
278}
279
280static inline bool tcp_out_of_memory(struct sock *sk)
281{
282 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
283 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
284 return true;
285 return false;
286}
287
288void sk_forced_mem_schedule(struct sock *sk, int size);
289
290static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
291{
292 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
293 int orphans = percpu_counter_read_positive(ocp);
294
295 if (orphans << shift > sysctl_tcp_max_orphans) {
296 orphans = percpu_counter_sum_positive(ocp);
297 if (orphans << shift > sysctl_tcp_max_orphans)
298 return true;
299 }
300 return false;
301}
302
303bool tcp_check_oom(struct sock *sk, int shift);
304
305
306extern struct proto tcp_prot;
307
308#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
309#define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
310#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
311#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
312
313void tcp_tasklet_init(void);
314
315int tcp_v4_err(struct sk_buff *skb, u32);
316
317void tcp_shutdown(struct sock *sk, int how);
318
319int tcp_v4_early_demux(struct sk_buff *skb);
320int tcp_v4_rcv(struct sk_buff *skb);
321
322int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
323int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
324int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
325int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
326 int flags);
327int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
328 size_t size, int flags);
329ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
330 size_t size, int flags);
331void tcp_release_cb(struct sock *sk);
332void tcp_wfree(struct sk_buff *skb);
333void tcp_write_timer_handler(struct sock *sk);
334void tcp_delack_timer_handler(struct sock *sk);
335int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
336int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
337void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
338void tcp_rcv_space_adjust(struct sock *sk);
339int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
340void tcp_twsk_destructor(struct sock *sk);
341ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
342 struct pipe_inode_info *pipe, size_t len,
343 unsigned int flags);
344
345void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
346static inline void tcp_dec_quickack_mode(struct sock *sk,
347 const unsigned int pkts)
348{
349 struct inet_connection_sock *icsk = inet_csk(sk);
350
351 if (icsk->icsk_ack.quick) {
352 if (pkts >= icsk->icsk_ack.quick) {
353 icsk->icsk_ack.quick = 0;
354 /* Leaving quickack mode we deflate ATO. */
355 icsk->icsk_ack.ato = TCP_ATO_MIN;
356 } else
357 icsk->icsk_ack.quick -= pkts;
358 }
359}
360
361#define TCP_ECN_OK 1
362#define TCP_ECN_QUEUE_CWR 2
363#define TCP_ECN_DEMAND_CWR 4
364#define TCP_ECN_SEEN 8
365
366enum tcp_tw_status {
367 TCP_TW_SUCCESS = 0,
368 TCP_TW_RST = 1,
369 TCP_TW_ACK = 2,
370 TCP_TW_SYN = 3
371};
372
373
374enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
375 struct sk_buff *skb,
376 const struct tcphdr *th);
377struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
378 struct request_sock *req, bool fastopen,
379 bool *lost_race);
380int tcp_child_process(struct sock *parent, struct sock *child,
381 struct sk_buff *skb);
382void tcp_enter_loss(struct sock *sk);
383void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
384void tcp_clear_retrans(struct tcp_sock *tp);
385void tcp_update_metrics(struct sock *sk);
386void tcp_init_metrics(struct sock *sk);
387void tcp_metrics_init(void);
388bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
389void tcp_close(struct sock *sk, long timeout);
390void tcp_init_sock(struct sock *sk);
391void tcp_init_transfer(struct sock *sk, int bpf_op);
392__poll_t tcp_poll(struct file *file, struct socket *sock,
393 struct poll_table_struct *wait);
394int tcp_getsockopt(struct sock *sk, int level, int optname,
395 char __user *optval, int __user *optlen);
396int tcp_setsockopt(struct sock *sk, int level, int optname,
397 char __user *optval, unsigned int optlen);
398int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
399 char __user *optval, int __user *optlen);
400int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
401 char __user *optval, unsigned int optlen);
402void tcp_set_keepalive(struct sock *sk, int val);
403void tcp_syn_ack_timeout(const struct request_sock *req);
404int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
405 int flags, int *addr_len);
406int tcp_set_rcvlowat(struct sock *sk, int val);
407void tcp_data_ready(struct sock *sk);
408#ifdef CONFIG_MMU
409int tcp_mmap(struct file *file, struct socket *sock,
410 struct vm_area_struct *vma);
411#endif
412void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
413 struct tcp_options_received *opt_rx,
414 int estab, struct tcp_fastopen_cookie *foc);
415const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
416
417/*
418 * BPF SKB-less helpers
419 */
420u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
421 struct tcphdr *th, u32 *cookie);
422u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
423 struct tcphdr *th, u32 *cookie);
424u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
425 const struct tcp_request_sock_ops *af_ops,
426 struct sock *sk, struct tcphdr *th);
427/*
428 * TCP v4 functions exported for the inet6 API
429 */
430
431void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
432void tcp_v4_mtu_reduced(struct sock *sk);
433void tcp_req_err(struct sock *sk, u32 seq, bool abort);
434int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
435struct sock *tcp_create_openreq_child(const struct sock *sk,
436 struct request_sock *req,
437 struct sk_buff *skb);
438void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
439struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
440 struct request_sock *req,
441 struct dst_entry *dst,
442 struct request_sock *req_unhash,
443 bool *own_req);
444int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
445int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
446int tcp_connect(struct sock *sk);
447enum tcp_synack_type {
448 TCP_SYNACK_NORMAL,
449 TCP_SYNACK_FASTOPEN,
450 TCP_SYNACK_COOKIE,
451};
452struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
453 struct request_sock *req,
454 struct tcp_fastopen_cookie *foc,
455 enum tcp_synack_type synack_type);
456int tcp_disconnect(struct sock *sk, int flags);
457
458void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
459int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
460void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
461
462/* From syncookies.c */
463struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
464 struct request_sock *req,
465 struct dst_entry *dst, u32 tsoff);
466int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
467 u32 cookie);
468struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
469#ifdef CONFIG_SYN_COOKIES
470
471/* Syncookies use a monotonic timer which increments every 60 seconds.
472 * This counter is used both as a hash input and partially encoded into
473 * the cookie value. A cookie is only validated further if the delta
474 * between the current counter value and the encoded one is less than this,
475 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
476 * the counter advances immediately after a cookie is generated).
477 */
478#define MAX_SYNCOOKIE_AGE 2
479#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
480#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
481
482/* syncookies: remember time of last synqueue overflow
483 * But do not dirty this field too often (once per second is enough)
484 * It is racy as we do not hold a lock, but race is very minor.
485 */
486static inline void tcp_synq_overflow(const struct sock *sk)
487{
488 unsigned int last_overflow;
489 unsigned int now = jiffies;
490
491 if (sk->sk_reuseport) {
492 struct sock_reuseport *reuse;
493
494 reuse = rcu_dereference(sk->sk_reuseport_cb);
495 if (likely(reuse)) {
496 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
497 if (time_after32(now, last_overflow + HZ))
498 WRITE_ONCE(reuse->synq_overflow_ts, now);
499 return;
500 }
501 }
502
503 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
504 if (time_after32(now, last_overflow + HZ))
505 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
506}
507
508/* syncookies: no recent synqueue overflow on this listening socket? */
509static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
510{
511 unsigned int last_overflow;
512 unsigned int now = jiffies;
513
514 if (sk->sk_reuseport) {
515 struct sock_reuseport *reuse;
516
517 reuse = rcu_dereference(sk->sk_reuseport_cb);
518 if (likely(reuse)) {
519 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
520 return time_after32(now, last_overflow +
521 TCP_SYNCOOKIE_VALID);
522 }
523 }
524
525 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
526 return time_after32(now, last_overflow + TCP_SYNCOOKIE_VALID);
527}
528
529static inline u32 tcp_cookie_time(void)
530{
531 u64 val = get_jiffies_64();
532
533 do_div(val, TCP_SYNCOOKIE_PERIOD);
534 return val;
535}
536
537u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
538 u16 *mssp);
539__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
540u64 cookie_init_timestamp(struct request_sock *req);
541bool cookie_timestamp_decode(const struct net *net,
542 struct tcp_options_received *opt);
543bool cookie_ecn_ok(const struct tcp_options_received *opt,
544 const struct net *net, const struct dst_entry *dst);
545
546/* From net/ipv6/syncookies.c */
547int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
548 u32 cookie);
549struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
550
551u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
552 const struct tcphdr *th, u16 *mssp);
553__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
554#endif
555/* tcp_output.c */
556
557void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
558 int nonagle);
559int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
560int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
561void tcp_retransmit_timer(struct sock *sk);
562void tcp_xmit_retransmit_queue(struct sock *);
563void tcp_simple_retransmit(struct sock *);
564void tcp_enter_recovery(struct sock *sk, bool ece_ack);
565int tcp_trim_head(struct sock *, struct sk_buff *, u32);
566enum tcp_queue {
567 TCP_FRAG_IN_WRITE_QUEUE,
568 TCP_FRAG_IN_RTX_QUEUE,
569};
570int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
571 struct sk_buff *skb, u32 len,
572 unsigned int mss_now, gfp_t gfp);
573
574void tcp_send_probe0(struct sock *);
575void tcp_send_partial(struct sock *);
576int tcp_write_wakeup(struct sock *, int mib);
577void tcp_send_fin(struct sock *sk);
578void tcp_send_active_reset(struct sock *sk, gfp_t priority);
579int tcp_send_synack(struct sock *);
580void tcp_push_one(struct sock *, unsigned int mss_now);
581void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
582void tcp_send_ack(struct sock *sk);
583void tcp_send_delayed_ack(struct sock *sk);
584void tcp_send_loss_probe(struct sock *sk);
585bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
586void tcp_skb_collapse_tstamp(struct sk_buff *skb,
587 const struct sk_buff *next_skb);
588
589/* tcp_input.c */
590void tcp_rearm_rto(struct sock *sk);
591void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
592void tcp_reset(struct sock *sk);
593void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
594void tcp_fin(struct sock *sk);
595
596/* tcp_timer.c */
597void tcp_init_xmit_timers(struct sock *);
598static inline void tcp_clear_xmit_timers(struct sock *sk)
599{
600 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
601 __sock_put(sk);
602
603 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
604 __sock_put(sk);
605
606 inet_csk_clear_xmit_timers(sk);
607}
608
609unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
610unsigned int tcp_current_mss(struct sock *sk);
611
612/* Bound MSS / TSO packet size with the half of the window */
613static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
614{
615 int cutoff;
616
617 /* When peer uses tiny windows, there is no use in packetizing
618 * to sub-MSS pieces for the sake of SWS or making sure there
619 * are enough packets in the pipe for fast recovery.
620 *
621 * On the other hand, for extremely large MSS devices, handling
622 * smaller than MSS windows in this way does make sense.
623 */
624 if (tp->max_window > TCP_MSS_DEFAULT)
625 cutoff = (tp->max_window >> 1);
626 else
627 cutoff = tp->max_window;
628
629 if (cutoff && pktsize > cutoff)
630 return max_t(int, cutoff, 68U - tp->tcp_header_len);
631 else
632 return pktsize;
633}
634
635/* tcp.c */
636void tcp_get_info(struct sock *, struct tcp_info *);
637
638/* Read 'sendfile()'-style from a TCP socket */
639int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
640 sk_read_actor_t recv_actor);
641
642void tcp_initialize_rcv_mss(struct sock *sk);
643
644int tcp_mtu_to_mss(struct sock *sk, int pmtu);
645int tcp_mss_to_mtu(struct sock *sk, int mss);
646void tcp_mtup_init(struct sock *sk);
647void tcp_init_buffer_space(struct sock *sk);
648
649static inline void tcp_bound_rto(const struct sock *sk)
650{
651 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
652 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
653}
654
655static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
656{
657 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
658}
659
660static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
661{
662 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
663 ntohl(TCP_FLAG_ACK) |
664 snd_wnd);
665}
666
667static inline void tcp_fast_path_on(struct tcp_sock *tp)
668{
669 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
670}
671
672static inline void tcp_fast_path_check(struct sock *sk)
673{
674 struct tcp_sock *tp = tcp_sk(sk);
675
676 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
677 tp->rcv_wnd &&
678 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
679 !tp->urg_data)
680 tcp_fast_path_on(tp);
681}
682
683/* Compute the actual rto_min value */
684static inline u32 tcp_rto_min(struct sock *sk)
685{
686 const struct dst_entry *dst = __sk_dst_get(sk);
687 u32 rto_min = TCP_RTO_MIN;
688
689 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
690 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
691 return rto_min;
692}
693
694static inline u32 tcp_rto_min_us(struct sock *sk)
695{
696 return jiffies_to_usecs(tcp_rto_min(sk));
697}
698
699static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
700{
701 return dst_metric_locked(dst, RTAX_CC_ALGO);
702}
703
704/* Minimum RTT in usec. ~0 means not available. */
705static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
706{
707 return minmax_get(&tp->rtt_min);
708}
709
710/* Compute the actual receive window we are currently advertising.
711 * Rcv_nxt can be after the window if our peer push more data
712 * than the offered window.
713 */
714static inline u32 tcp_receive_window(const struct tcp_sock *tp)
715{
716 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
717
718 if (win < 0)
719 win = 0;
720 return (u32) win;
721}
722
723/* Choose a new window, without checks for shrinking, and without
724 * scaling applied to the result. The caller does these things
725 * if necessary. This is a "raw" window selection.
726 */
727u32 __tcp_select_window(struct sock *sk);
728
729void tcp_send_window_probe(struct sock *sk);
730
731/* TCP uses 32bit jiffies to save some space.
732 * Note that this is different from tcp_time_stamp, which
733 * historically has been the same until linux-4.13.
734 */
735#define tcp_jiffies32 ((u32)jiffies)
736
737/*
738 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
739 * It is no longer tied to jiffies, but to 1 ms clock.
740 * Note: double check if you want to use tcp_jiffies32 instead of this.
741 */
742#define TCP_TS_HZ 1000
743
744static inline u64 tcp_clock_ns(void)
745{
746 return ktime_get_ns();
747}
748
749static inline u64 tcp_clock_us(void)
750{
751 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
752}
753
754/* This should only be used in contexts where tp->tcp_mstamp is up to date */
755static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
756{
757 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
758}
759
760/* Could use tcp_clock_us() / 1000, but this version uses a single divide */
761static inline u32 tcp_time_stamp_raw(void)
762{
763 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
764}
765
766void tcp_mstamp_refresh(struct tcp_sock *tp);
767
768static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
769{
770 return max_t(s64, t1 - t0, 0);
771}
772
773static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
774{
775 return div_u64(skb->skb_mstamp_ns, NSEC_PER_SEC / TCP_TS_HZ);
776}
777
778/* provide the departure time in us unit */
779static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
780{
781 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
782}
783
784
785#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
786
787#define TCPHDR_FIN 0x01
788#define TCPHDR_SYN 0x02
789#define TCPHDR_RST 0x04
790#define TCPHDR_PSH 0x08
791#define TCPHDR_ACK 0x10
792#define TCPHDR_URG 0x20
793#define TCPHDR_ECE 0x40
794#define TCPHDR_CWR 0x80
795
796#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
797
798/* This is what the send packet queuing engine uses to pass
799 * TCP per-packet control information to the transmission code.
800 * We also store the host-order sequence numbers in here too.
801 * This is 44 bytes if IPV6 is enabled.
802 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
803 */
804struct tcp_skb_cb {
805 __u32 seq; /* Starting sequence number */
806 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
807 union {
808 /* Note : tcp_tw_isn is used in input path only
809 * (isn chosen by tcp_timewait_state_process())
810 *
811 * tcp_gso_segs/size are used in write queue only,
812 * cf tcp_skb_pcount()/tcp_skb_mss()
813 */
814 __u32 tcp_tw_isn;
815 struct {
816 u16 tcp_gso_segs;
817 u16 tcp_gso_size;
818 };
819 };
820 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
821
822 __u8 sacked; /* State flags for SACK. */
823#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
824#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
825#define TCPCB_LOST 0x04 /* SKB is lost */
826#define TCPCB_TAGBITS 0x07 /* All tag bits */
827#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
828#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
829#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
830 TCPCB_REPAIRED)
831
832 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
833 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
834 eor:1, /* Is skb MSG_EOR marked? */
835 has_rxtstamp:1, /* SKB has a RX timestamp */
836 unused:5;
837 __u32 ack_seq; /* Sequence number ACK'd */
838 union {
839 struct {
840 /* There is space for up to 24 bytes */
841 __u32 in_flight:30,/* Bytes in flight at transmit */
842 is_app_limited:1, /* cwnd not fully used? */
843 unused:1;
844 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
845 __u32 delivered;
846 /* start of send pipeline phase */
847 u64 first_tx_mstamp;
848 /* when we reached the "delivered" count */
849 u64 delivered_mstamp;
850 } tx; /* only used for outgoing skbs */
851 union {
852 struct inet_skb_parm h4;
853#if IS_ENABLED(CONFIG_IPV6)
854 struct inet6_skb_parm h6;
855#endif
856 } header; /* For incoming skbs */
857 struct {
858 __u32 flags;
859 struct sock *sk_redir;
860 void *data_end;
861 } bpf;
862 };
863};
864
865#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
866
867static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
868{
869 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
870}
871
872static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb)
873{
874 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS;
875}
876
877static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb)
878{
879 return TCP_SKB_CB(skb)->bpf.sk_redir;
880}
881
882static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb)
883{
884 TCP_SKB_CB(skb)->bpf.sk_redir = NULL;
885}
886
887#if IS_ENABLED(CONFIG_IPV6)
888/* This is the variant of inet6_iif() that must be used by TCP,
889 * as TCP moves IP6CB into a different location in skb->cb[]
890 */
891static inline int tcp_v6_iif(const struct sk_buff *skb)
892{
893 return TCP_SKB_CB(skb)->header.h6.iif;
894}
895
896static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
897{
898 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
899
900 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
901}
902
903/* TCP_SKB_CB reference means this can not be used from early demux */
904static inline int tcp_v6_sdif(const struct sk_buff *skb)
905{
906#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
907 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
908 return TCP_SKB_CB(skb)->header.h6.iif;
909#endif
910 return 0;
911}
912#endif
913
914static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
915{
916#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
917 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
918 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
919 return true;
920#endif
921 return false;
922}
923
924/* TCP_SKB_CB reference means this can not be used from early demux */
925static inline int tcp_v4_sdif(struct sk_buff *skb)
926{
927#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
928 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
929 return TCP_SKB_CB(skb)->header.h4.iif;
930#endif
931 return 0;
932}
933
934/* Due to TSO, an SKB can be composed of multiple actual
935 * packets. To keep these tracked properly, we use this.
936 */
937static inline int tcp_skb_pcount(const struct sk_buff *skb)
938{
939 return TCP_SKB_CB(skb)->tcp_gso_segs;
940}
941
942static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
943{
944 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
945}
946
947static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
948{
949 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
950}
951
952/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
953static inline int tcp_skb_mss(const struct sk_buff *skb)
954{
955 return TCP_SKB_CB(skb)->tcp_gso_size;
956}
957
958static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
959{
960 return likely(!TCP_SKB_CB(skb)->eor);
961}
962
963/* Events passed to congestion control interface */
964enum tcp_ca_event {
965 CA_EVENT_TX_START, /* first transmit when no packets in flight */
966 CA_EVENT_CWND_RESTART, /* congestion window restart */
967 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
968 CA_EVENT_LOSS, /* loss timeout */
969 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
970 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
971};
972
973/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
974enum tcp_ca_ack_event_flags {
975 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
976 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
977 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
978};
979
980/*
981 * Interface for adding new TCP congestion control handlers
982 */
983#define TCP_CA_NAME_MAX 16
984#define TCP_CA_MAX 128
985#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
986
987#define TCP_CA_UNSPEC 0
988
989/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
990#define TCP_CONG_NON_RESTRICTED 0x1
991/* Requires ECN/ECT set on all packets */
992#define TCP_CONG_NEEDS_ECN 0x2
993
994union tcp_cc_info;
995
996struct ack_sample {
997 u32 pkts_acked;
998 s32 rtt_us;
999 u32 in_flight;
1000};
1001
1002/* A rate sample measures the number of (original/retransmitted) data
1003 * packets delivered "delivered" over an interval of time "interval_us".
1004 * The tcp_rate.c code fills in the rate sample, and congestion
1005 * control modules that define a cong_control function to run at the end
1006 * of ACK processing can optionally chose to consult this sample when
1007 * setting cwnd and pacing rate.
1008 * A sample is invalid if "delivered" or "interval_us" is negative.
1009 */
1010struct rate_sample {
1011 u64 prior_mstamp; /* starting timestamp for interval */
1012 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1013 s32 delivered; /* number of packets delivered over interval */
1014 long interval_us; /* time for tp->delivered to incr "delivered" */
1015 u32 snd_interval_us; /* snd interval for delivered packets */
1016 u32 rcv_interval_us; /* rcv interval for delivered packets */
1017 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1018 int losses; /* number of packets marked lost upon ACK */
1019 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1020 u32 prior_in_flight; /* in flight before this ACK */
1021 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1022 bool is_retrans; /* is sample from retransmission? */
1023 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1024};
1025
1026struct tcp_congestion_ops {
1027 struct list_head list;
1028 u32 key;
1029 u32 flags;
1030
1031 /* initialize private data (optional) */
1032 void (*init)(struct sock *sk);
1033 /* cleanup private data (optional) */
1034 void (*release)(struct sock *sk);
1035
1036 /* return slow start threshold (required) */
1037 u32 (*ssthresh)(struct sock *sk);
1038 /* do new cwnd calculation (required) */
1039 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1040 /* call before changing ca_state (optional) */
1041 void (*set_state)(struct sock *sk, u8 new_state);
1042 /* call when cwnd event occurs (optional) */
1043 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1044 /* call when ack arrives (optional) */
1045 void (*in_ack_event)(struct sock *sk, u32 flags);
1046 /* new value of cwnd after loss (required) */
1047 u32 (*undo_cwnd)(struct sock *sk);
1048 /* hook for packet ack accounting (optional) */
1049 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1050 /* override sysctl_tcp_min_tso_segs */
1051 u32 (*min_tso_segs)(struct sock *sk);
1052 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1053 u32 (*sndbuf_expand)(struct sock *sk);
1054 /* call when packets are delivered to update cwnd and pacing rate,
1055 * after all the ca_state processing. (optional)
1056 */
1057 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1058 /* get info for inet_diag (optional) */
1059 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1060 union tcp_cc_info *info);
1061
1062 char name[TCP_CA_NAME_MAX];
1063 struct module *owner;
1064};
1065
1066int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1067void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1068
1069void tcp_assign_congestion_control(struct sock *sk);
1070void tcp_init_congestion_control(struct sock *sk);
1071void tcp_cleanup_congestion_control(struct sock *sk);
1072int tcp_set_default_congestion_control(struct net *net, const char *name);
1073void tcp_get_default_congestion_control(struct net *net, char *name);
1074void tcp_get_available_congestion_control(char *buf, size_t len);
1075void tcp_get_allowed_congestion_control(char *buf, size_t len);
1076int tcp_set_allowed_congestion_control(char *allowed);
1077int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1078 bool reinit, bool cap_net_admin);
1079u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1080void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1081
1082u32 tcp_reno_ssthresh(struct sock *sk);
1083u32 tcp_reno_undo_cwnd(struct sock *sk);
1084void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1085extern struct tcp_congestion_ops tcp_reno;
1086
1087struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1088u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1089#ifdef CONFIG_INET
1090char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1091#else
1092static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1093{
1094 return NULL;
1095}
1096#endif
1097
1098static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1099{
1100 const struct inet_connection_sock *icsk = inet_csk(sk);
1101
1102 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1103}
1104
1105static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1106{
1107 struct inet_connection_sock *icsk = inet_csk(sk);
1108
1109 if (icsk->icsk_ca_ops->set_state)
1110 icsk->icsk_ca_ops->set_state(sk, ca_state);
1111 icsk->icsk_ca_state = ca_state;
1112}
1113
1114static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1115{
1116 const struct inet_connection_sock *icsk = inet_csk(sk);
1117
1118 if (icsk->icsk_ca_ops->cwnd_event)
1119 icsk->icsk_ca_ops->cwnd_event(sk, event);
1120}
1121
1122/* From tcp_rate.c */
1123void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1124void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1125 struct rate_sample *rs);
1126void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1127 bool is_sack_reneg, struct rate_sample *rs);
1128void tcp_rate_check_app_limited(struct sock *sk);
1129
1130/* These functions determine how the current flow behaves in respect of SACK
1131 * handling. SACK is negotiated with the peer, and therefore it can vary
1132 * between different flows.
1133 *
1134 * tcp_is_sack - SACK enabled
1135 * tcp_is_reno - No SACK
1136 */
1137static inline int tcp_is_sack(const struct tcp_sock *tp)
1138{
1139 return likely(tp->rx_opt.sack_ok);
1140}
1141
1142static inline bool tcp_is_reno(const struct tcp_sock *tp)
1143{
1144 return !tcp_is_sack(tp);
1145}
1146
1147static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1148{
1149 return tp->sacked_out + tp->lost_out;
1150}
1151
1152/* This determines how many packets are "in the network" to the best
1153 * of our knowledge. In many cases it is conservative, but where
1154 * detailed information is available from the receiver (via SACK
1155 * blocks etc.) we can make more aggressive calculations.
1156 *
1157 * Use this for decisions involving congestion control, use just
1158 * tp->packets_out to determine if the send queue is empty or not.
1159 *
1160 * Read this equation as:
1161 *
1162 * "Packets sent once on transmission queue" MINUS
1163 * "Packets left network, but not honestly ACKed yet" PLUS
1164 * "Packets fast retransmitted"
1165 */
1166static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1167{
1168 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1169}
1170
1171#define TCP_INFINITE_SSTHRESH 0x7fffffff
1172
1173static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1174{
1175 return tp->snd_cwnd < tp->snd_ssthresh;
1176}
1177
1178static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1179{
1180 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1181}
1182
1183static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1184{
1185 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1186 (1 << inet_csk(sk)->icsk_ca_state);
1187}
1188
1189/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1190 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1191 * ssthresh.
1192 */
1193static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1194{
1195 const struct tcp_sock *tp = tcp_sk(sk);
1196
1197 if (tcp_in_cwnd_reduction(sk))
1198 return tp->snd_ssthresh;
1199 else
1200 return max(tp->snd_ssthresh,
1201 ((tp->snd_cwnd >> 1) +
1202 (tp->snd_cwnd >> 2)));
1203}
1204
1205/* Use define here intentionally to get WARN_ON location shown at the caller */
1206#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1207
1208void tcp_enter_cwr(struct sock *sk);
1209__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1210
1211/* The maximum number of MSS of available cwnd for which TSO defers
1212 * sending if not using sysctl_tcp_tso_win_divisor.
1213 */
1214static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1215{
1216 return 3;
1217}
1218
1219/* Returns end sequence number of the receiver's advertised window */
1220static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1221{
1222 return tp->snd_una + tp->snd_wnd;
1223}
1224
1225/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1226 * flexible approach. The RFC suggests cwnd should not be raised unless
1227 * it was fully used previously. And that's exactly what we do in
1228 * congestion avoidance mode. But in slow start we allow cwnd to grow
1229 * as long as the application has used half the cwnd.
1230 * Example :
1231 * cwnd is 10 (IW10), but application sends 9 frames.
1232 * We allow cwnd to reach 18 when all frames are ACKed.
1233 * This check is safe because it's as aggressive as slow start which already
1234 * risks 100% overshoot. The advantage is that we discourage application to
1235 * either send more filler packets or data to artificially blow up the cwnd
1236 * usage, and allow application-limited process to probe bw more aggressively.
1237 */
1238static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1239{
1240 const struct tcp_sock *tp = tcp_sk(sk);
1241
1242 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1243 if (tcp_in_slow_start(tp))
1244 return tp->snd_cwnd < 2 * tp->max_packets_out;
1245
1246 return tp->is_cwnd_limited;
1247}
1248
1249/* BBR congestion control needs pacing.
1250 * Same remark for SO_MAX_PACING_RATE.
1251 * sch_fq packet scheduler is efficiently handling pacing,
1252 * but is not always installed/used.
1253 * Return true if TCP stack should pace packets itself.
1254 */
1255static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1256{
1257 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1258}
1259
1260/* Return in jiffies the delay before one skb is sent.
1261 * If @skb is NULL, we look at EDT for next packet being sent on the socket.
1262 */
1263static inline unsigned long tcp_pacing_delay(const struct sock *sk,
1264 const struct sk_buff *skb)
1265{
1266 s64 pacing_delay = skb ? skb->tstamp : tcp_sk(sk)->tcp_wstamp_ns;
1267
1268 pacing_delay -= tcp_sk(sk)->tcp_clock_cache;
1269
1270 return pacing_delay > 0 ? nsecs_to_jiffies(pacing_delay) : 0;
1271}
1272
1273static inline void tcp_reset_xmit_timer(struct sock *sk,
1274 const int what,
1275 unsigned long when,
1276 const unsigned long max_when,
1277 const struct sk_buff *skb)
1278{
1279 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk, skb),
1280 max_when);
1281}
1282
1283/* Something is really bad, we could not queue an additional packet,
1284 * because qdisc is full or receiver sent a 0 window, or we are paced.
1285 * We do not want to add fuel to the fire, or abort too early,
1286 * so make sure the timer we arm now is at least 200ms in the future,
1287 * regardless of current icsk_rto value (as it could be ~2ms)
1288 */
1289static inline unsigned long tcp_probe0_base(const struct sock *sk)
1290{
1291 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1292}
1293
1294/* Variant of inet_csk_rto_backoff() used for zero window probes */
1295static inline unsigned long tcp_probe0_when(const struct sock *sk,
1296 unsigned long max_when)
1297{
1298 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1299
1300 return (unsigned long)min_t(u64, when, max_when);
1301}
1302
1303static inline void tcp_check_probe_timer(struct sock *sk)
1304{
1305 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1306 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1307 tcp_probe0_base(sk), TCP_RTO_MAX,
1308 NULL);
1309}
1310
1311static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1312{
1313 tp->snd_wl1 = seq;
1314}
1315
1316static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1317{
1318 tp->snd_wl1 = seq;
1319}
1320
1321/*
1322 * Calculate(/check) TCP checksum
1323 */
1324static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1325 __be32 daddr, __wsum base)
1326{
1327 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1328}
1329
1330static inline bool tcp_checksum_complete(struct sk_buff *skb)
1331{
1332 return !skb_csum_unnecessary(skb) &&
1333 __skb_checksum_complete(skb);
1334}
1335
1336bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1337int tcp_filter(struct sock *sk, struct sk_buff *skb);
1338void tcp_set_state(struct sock *sk, int state);
1339void tcp_done(struct sock *sk);
1340int tcp_abort(struct sock *sk, int err);
1341
1342static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1343{
1344 rx_opt->dsack = 0;
1345 rx_opt->num_sacks = 0;
1346}
1347
1348u32 tcp_default_init_rwnd(u32 mss);
1349void tcp_cwnd_restart(struct sock *sk, s32 delta);
1350
1351static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1352{
1353 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1354 struct tcp_sock *tp = tcp_sk(sk);
1355 s32 delta;
1356
1357 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1358 ca_ops->cong_control)
1359 return;
1360 delta = tcp_jiffies32 - tp->lsndtime;
1361 if (delta > inet_csk(sk)->icsk_rto)
1362 tcp_cwnd_restart(sk, delta);
1363}
1364
1365/* Determine a window scaling and initial window to offer. */
1366void tcp_select_initial_window(const struct sock *sk, int __space,
1367 __u32 mss, __u32 *rcv_wnd,
1368 __u32 *window_clamp, int wscale_ok,
1369 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1370
1371static inline int tcp_win_from_space(const struct sock *sk, int space)
1372{
1373 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1374
1375 return tcp_adv_win_scale <= 0 ?
1376 (space>>(-tcp_adv_win_scale)) :
1377 space - (space>>tcp_adv_win_scale);
1378}
1379
1380/* Note: caller must be prepared to deal with negative returns */
1381static inline int tcp_space(const struct sock *sk)
1382{
1383 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1384 READ_ONCE(sk->sk_backlog.len) -
1385 atomic_read(&sk->sk_rmem_alloc));
1386}
1387
1388static inline int tcp_full_space(const struct sock *sk)
1389{
1390 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1391}
1392
1393extern void tcp_openreq_init_rwin(struct request_sock *req,
1394 const struct sock *sk_listener,
1395 const struct dst_entry *dst);
1396
1397void tcp_enter_memory_pressure(struct sock *sk);
1398void tcp_leave_memory_pressure(struct sock *sk);
1399
1400static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1401{
1402 struct net *net = sock_net((struct sock *)tp);
1403
1404 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1405}
1406
1407static inline int keepalive_time_when(const struct tcp_sock *tp)
1408{
1409 struct net *net = sock_net((struct sock *)tp);
1410
1411 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1412}
1413
1414static inline int keepalive_probes(const struct tcp_sock *tp)
1415{
1416 struct net *net = sock_net((struct sock *)tp);
1417
1418 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1419}
1420
1421static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1422{
1423 const struct inet_connection_sock *icsk = &tp->inet_conn;
1424
1425 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1426 tcp_jiffies32 - tp->rcv_tstamp);
1427}
1428
1429static inline int tcp_fin_time(const struct sock *sk)
1430{
1431 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1432 const int rto = inet_csk(sk)->icsk_rto;
1433
1434 if (fin_timeout < (rto << 2) - (rto >> 1))
1435 fin_timeout = (rto << 2) - (rto >> 1);
1436
1437 return fin_timeout;
1438}
1439
1440static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1441 int paws_win)
1442{
1443 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1444 return true;
1445 if (unlikely(!time_before32(ktime_get_seconds(),
1446 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1447 return true;
1448 /*
1449 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1450 * then following tcp messages have valid values. Ignore 0 value,
1451 * or else 'negative' tsval might forbid us to accept their packets.
1452 */
1453 if (!rx_opt->ts_recent)
1454 return true;
1455 return false;
1456}
1457
1458static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1459 int rst)
1460{
1461 if (tcp_paws_check(rx_opt, 0))
1462 return false;
1463
1464 /* RST segments are not recommended to carry timestamp,
1465 and, if they do, it is recommended to ignore PAWS because
1466 "their cleanup function should take precedence over timestamps."
1467 Certainly, it is mistake. It is necessary to understand the reasons
1468 of this constraint to relax it: if peer reboots, clock may go
1469 out-of-sync and half-open connections will not be reset.
1470 Actually, the problem would be not existing if all
1471 the implementations followed draft about maintaining clock
1472 via reboots. Linux-2.2 DOES NOT!
1473
1474 However, we can relax time bounds for RST segments to MSL.
1475 */
1476 if (rst && !time_before32(ktime_get_seconds(),
1477 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1478 return false;
1479 return true;
1480}
1481
1482bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1483 int mib_idx, u32 *last_oow_ack_time);
1484
1485static inline void tcp_mib_init(struct net *net)
1486{
1487 /* See RFC 2012 */
1488 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1489 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1490 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1491 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1492}
1493
1494/* from STCP */
1495static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1496{
1497 tp->lost_skb_hint = NULL;
1498}
1499
1500static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1501{
1502 tcp_clear_retrans_hints_partial(tp);
1503 tp->retransmit_skb_hint = NULL;
1504}
1505
1506union tcp_md5_addr {
1507 struct in_addr a4;
1508#if IS_ENABLED(CONFIG_IPV6)
1509 struct in6_addr a6;
1510#endif
1511};
1512
1513/* - key database */
1514struct tcp_md5sig_key {
1515 struct hlist_node node;
1516 u8 keylen;
1517 u8 family; /* AF_INET or AF_INET6 */
1518 union tcp_md5_addr addr;
1519 u8 prefixlen;
1520 u8 key[TCP_MD5SIG_MAXKEYLEN];
1521 struct rcu_head rcu;
1522};
1523
1524/* - sock block */
1525struct tcp_md5sig_info {
1526 struct hlist_head head;
1527 struct rcu_head rcu;
1528};
1529
1530/* - pseudo header */
1531struct tcp4_pseudohdr {
1532 __be32 saddr;
1533 __be32 daddr;
1534 __u8 pad;
1535 __u8 protocol;
1536 __be16 len;
1537};
1538
1539struct tcp6_pseudohdr {
1540 struct in6_addr saddr;
1541 struct in6_addr daddr;
1542 __be32 len;
1543 __be32 protocol; /* including padding */
1544};
1545
1546union tcp_md5sum_block {
1547 struct tcp4_pseudohdr ip4;
1548#if IS_ENABLED(CONFIG_IPV6)
1549 struct tcp6_pseudohdr ip6;
1550#endif
1551};
1552
1553/* - pool: digest algorithm, hash description and scratch buffer */
1554struct tcp_md5sig_pool {
1555 struct ahash_request *md5_req;
1556 void *scratch;
1557};
1558
1559/* - functions */
1560int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1561 const struct sock *sk, const struct sk_buff *skb);
1562int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1563 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1564 gfp_t gfp);
1565int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1566 int family, u8 prefixlen);
1567struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1568 const struct sock *addr_sk);
1569
1570#ifdef CONFIG_TCP_MD5SIG
1571#include <linux/jump_label.h>
1572extern struct static_key_false tcp_md5_needed;
1573struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk,
1574 const union tcp_md5_addr *addr,
1575 int family);
1576static inline struct tcp_md5sig_key *
1577tcp_md5_do_lookup(const struct sock *sk,
1578 const union tcp_md5_addr *addr,
1579 int family)
1580{
1581 if (!static_branch_unlikely(&tcp_md5_needed))
1582 return NULL;
1583 return __tcp_md5_do_lookup(sk, addr, family);
1584}
1585
1586#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1587#else
1588static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1589 const union tcp_md5_addr *addr,
1590 int family)
1591{
1592 return NULL;
1593}
1594#define tcp_twsk_md5_key(twsk) NULL
1595#endif
1596
1597bool tcp_alloc_md5sig_pool(void);
1598
1599struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1600static inline void tcp_put_md5sig_pool(void)
1601{
1602 local_bh_enable();
1603}
1604
1605int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1606 unsigned int header_len);
1607int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1608 const struct tcp_md5sig_key *key);
1609
1610/* From tcp_fastopen.c */
1611void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1612 struct tcp_fastopen_cookie *cookie);
1613void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1614 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1615 u16 try_exp);
1616struct tcp_fastopen_request {
1617 /* Fast Open cookie. Size 0 means a cookie request */
1618 struct tcp_fastopen_cookie cookie;
1619 struct msghdr *data; /* data in MSG_FASTOPEN */
1620 size_t size;
1621 int copied; /* queued in tcp_connect() */
1622 struct ubuf_info *uarg;
1623};
1624void tcp_free_fastopen_req(struct tcp_sock *tp);
1625void tcp_fastopen_destroy_cipher(struct sock *sk);
1626void tcp_fastopen_ctx_destroy(struct net *net);
1627int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1628 void *primary_key, void *backup_key);
1629void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1630struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1631 struct request_sock *req,
1632 struct tcp_fastopen_cookie *foc,
1633 const struct dst_entry *dst);
1634void tcp_fastopen_init_key_once(struct net *net);
1635bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1636 struct tcp_fastopen_cookie *cookie);
1637bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1638#define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1639#define TCP_FASTOPEN_KEY_MAX 2
1640#define TCP_FASTOPEN_KEY_BUF_LENGTH \
1641 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1642
1643/* Fastopen key context */
1644struct tcp_fastopen_context {
1645 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
1646 int num;
1647 struct rcu_head rcu;
1648};
1649
1650extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1651void tcp_fastopen_active_disable(struct sock *sk);
1652bool tcp_fastopen_active_should_disable(struct sock *sk);
1653void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1654void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1655
1656/* Caller needs to wrap with rcu_read_(un)lock() */
1657static inline
1658struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1659{
1660 struct tcp_fastopen_context *ctx;
1661
1662 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1663 if (!ctx)
1664 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1665 return ctx;
1666}
1667
1668static inline
1669bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1670 const struct tcp_fastopen_cookie *orig)
1671{
1672 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1673 orig->len == foc->len &&
1674 !memcmp(orig->val, foc->val, foc->len))
1675 return true;
1676 return false;
1677}
1678
1679static inline
1680int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1681{
1682 return ctx->num;
1683}
1684
1685/* Latencies incurred by various limits for a sender. They are
1686 * chronograph-like stats that are mutually exclusive.
1687 */
1688enum tcp_chrono {
1689 TCP_CHRONO_UNSPEC,
1690 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1691 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1692 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1693 __TCP_CHRONO_MAX,
1694};
1695
1696void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1697void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1698
1699/* This helper is needed, because skb->tcp_tsorted_anchor uses
1700 * the same memory storage than skb->destructor/_skb_refdst
1701 */
1702static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1703{
1704 skb->destructor = NULL;
1705 skb->_skb_refdst = 0UL;
1706}
1707
1708#define tcp_skb_tsorted_save(skb) { \
1709 unsigned long _save = skb->_skb_refdst; \
1710 skb->_skb_refdst = 0UL;
1711
1712#define tcp_skb_tsorted_restore(skb) \
1713 skb->_skb_refdst = _save; \
1714}
1715
1716void tcp_write_queue_purge(struct sock *sk);
1717
1718static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1719{
1720 return skb_rb_first(&sk->tcp_rtx_queue);
1721}
1722
1723static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1724{
1725 return skb_rb_last(&sk->tcp_rtx_queue);
1726}
1727
1728static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1729{
1730 return skb_peek(&sk->sk_write_queue);
1731}
1732
1733static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1734{
1735 return skb_peek_tail(&sk->sk_write_queue);
1736}
1737
1738#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1739 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1740
1741static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1742{
1743 return skb_peek(&sk->sk_write_queue);
1744}
1745
1746static inline bool tcp_skb_is_last(const struct sock *sk,
1747 const struct sk_buff *skb)
1748{
1749 return skb_queue_is_last(&sk->sk_write_queue, skb);
1750}
1751
1752static inline bool tcp_write_queue_empty(const struct sock *sk)
1753{
1754 return skb_queue_empty(&sk->sk_write_queue);
1755}
1756
1757static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1758{
1759 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1760}
1761
1762static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1763{
1764 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1765}
1766
1767static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1768{
1769 __skb_queue_tail(&sk->sk_write_queue, skb);
1770
1771 /* Queue it, remembering where we must start sending. */
1772 if (sk->sk_write_queue.next == skb)
1773 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1774}
1775
1776/* Insert new before skb on the write queue of sk. */
1777static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1778 struct sk_buff *skb,
1779 struct sock *sk)
1780{
1781 __skb_queue_before(&sk->sk_write_queue, skb, new);
1782}
1783
1784static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1785{
1786 tcp_skb_tsorted_anchor_cleanup(skb);
1787 __skb_unlink(skb, &sk->sk_write_queue);
1788}
1789
1790void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1791
1792static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1793{
1794 tcp_skb_tsorted_anchor_cleanup(skb);
1795 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1796}
1797
1798static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1799{
1800 list_del(&skb->tcp_tsorted_anchor);
1801 tcp_rtx_queue_unlink(skb, sk);
1802 sk_wmem_free_skb(sk, skb);
1803}
1804
1805static inline void tcp_push_pending_frames(struct sock *sk)
1806{
1807 if (tcp_send_head(sk)) {
1808 struct tcp_sock *tp = tcp_sk(sk);
1809
1810 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1811 }
1812}
1813
1814/* Start sequence of the skb just after the highest skb with SACKed
1815 * bit, valid only if sacked_out > 0 or when the caller has ensured
1816 * validity by itself.
1817 */
1818static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1819{
1820 if (!tp->sacked_out)
1821 return tp->snd_una;
1822
1823 if (tp->highest_sack == NULL)
1824 return tp->snd_nxt;
1825
1826 return TCP_SKB_CB(tp->highest_sack)->seq;
1827}
1828
1829static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1830{
1831 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1832}
1833
1834static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1835{
1836 return tcp_sk(sk)->highest_sack;
1837}
1838
1839static inline void tcp_highest_sack_reset(struct sock *sk)
1840{
1841 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1842}
1843
1844/* Called when old skb is about to be deleted and replaced by new skb */
1845static inline void tcp_highest_sack_replace(struct sock *sk,
1846 struct sk_buff *old,
1847 struct sk_buff *new)
1848{
1849 if (old == tcp_highest_sack(sk))
1850 tcp_sk(sk)->highest_sack = new;
1851}
1852
1853/* This helper checks if socket has IP_TRANSPARENT set */
1854static inline bool inet_sk_transparent(const struct sock *sk)
1855{
1856 switch (sk->sk_state) {
1857 case TCP_TIME_WAIT:
1858 return inet_twsk(sk)->tw_transparent;
1859 case TCP_NEW_SYN_RECV:
1860 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1861 }
1862 return inet_sk(sk)->transparent;
1863}
1864
1865/* Determines whether this is a thin stream (which may suffer from
1866 * increased latency). Used to trigger latency-reducing mechanisms.
1867 */
1868static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1869{
1870 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1871}
1872
1873/* /proc */
1874enum tcp_seq_states {
1875 TCP_SEQ_STATE_LISTENING,
1876 TCP_SEQ_STATE_ESTABLISHED,
1877};
1878
1879void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1880void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1881void tcp_seq_stop(struct seq_file *seq, void *v);
1882
1883struct tcp_seq_afinfo {
1884 sa_family_t family;
1885};
1886
1887struct tcp_iter_state {
1888 struct seq_net_private p;
1889 enum tcp_seq_states state;
1890 struct sock *syn_wait_sk;
1891 int bucket, offset, sbucket, num;
1892 loff_t last_pos;
1893};
1894
1895extern struct request_sock_ops tcp_request_sock_ops;
1896extern struct request_sock_ops tcp6_request_sock_ops;
1897
1898void tcp_v4_destroy_sock(struct sock *sk);
1899
1900struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1901 netdev_features_t features);
1902struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1903int tcp_gro_complete(struct sk_buff *skb);
1904
1905void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1906
1907static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1908{
1909 struct net *net = sock_net((struct sock *)tp);
1910 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1911}
1912
1913/* @wake is one when sk_stream_write_space() calls us.
1914 * This sends EPOLLOUT only if notsent_bytes is half the limit.
1915 * This mimics the strategy used in sock_def_write_space().
1916 */
1917static inline bool tcp_stream_memory_free(const struct sock *sk, int wake)
1918{
1919 const struct tcp_sock *tp = tcp_sk(sk);
1920 u32 notsent_bytes = READ_ONCE(tp->write_seq) -
1921 READ_ONCE(tp->snd_nxt);
1922
1923 return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
1924}
1925
1926#ifdef CONFIG_PROC_FS
1927int tcp4_proc_init(void);
1928void tcp4_proc_exit(void);
1929#endif
1930
1931int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1932int tcp_conn_request(struct request_sock_ops *rsk_ops,
1933 const struct tcp_request_sock_ops *af_ops,
1934 struct sock *sk, struct sk_buff *skb);
1935
1936/* TCP af-specific functions */
1937struct tcp_sock_af_ops {
1938#ifdef CONFIG_TCP_MD5SIG
1939 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1940 const struct sock *addr_sk);
1941 int (*calc_md5_hash)(char *location,
1942 const struct tcp_md5sig_key *md5,
1943 const struct sock *sk,
1944 const struct sk_buff *skb);
1945 int (*md5_parse)(struct sock *sk,
1946 int optname,
1947 char __user *optval,
1948 int optlen);
1949#endif
1950};
1951
1952struct tcp_request_sock_ops {
1953 u16 mss_clamp;
1954#ifdef CONFIG_TCP_MD5SIG
1955 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1956 const struct sock *addr_sk);
1957 int (*calc_md5_hash) (char *location,
1958 const struct tcp_md5sig_key *md5,
1959 const struct sock *sk,
1960 const struct sk_buff *skb);
1961#endif
1962 void (*init_req)(struct request_sock *req,
1963 const struct sock *sk_listener,
1964 struct sk_buff *skb);
1965#ifdef CONFIG_SYN_COOKIES
1966 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1967 __u16 *mss);
1968#endif
1969 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1970 const struct request_sock *req);
1971 u32 (*init_seq)(const struct sk_buff *skb);
1972 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1973 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1974 struct flowi *fl, struct request_sock *req,
1975 struct tcp_fastopen_cookie *foc,
1976 enum tcp_synack_type synack_type);
1977};
1978
1979#ifdef CONFIG_SYN_COOKIES
1980static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1981 const struct sock *sk, struct sk_buff *skb,
1982 __u16 *mss)
1983{
1984 tcp_synq_overflow(sk);
1985 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1986 return ops->cookie_init_seq(skb, mss);
1987}
1988#else
1989static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1990 const struct sock *sk, struct sk_buff *skb,
1991 __u16 *mss)
1992{
1993 return 0;
1994}
1995#endif
1996
1997int tcpv4_offload_init(void);
1998
1999void tcp_v4_init(void);
2000void tcp_init(void);
2001
2002/* tcp_recovery.c */
2003void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2004void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2005extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2006 u32 reo_wnd);
2007extern void tcp_rack_mark_lost(struct sock *sk);
2008extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2009 u64 xmit_time);
2010extern void tcp_rack_reo_timeout(struct sock *sk);
2011extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2012
2013/* At how many usecs into the future should the RTO fire? */
2014static inline s64 tcp_rto_delta_us(const struct sock *sk)
2015{
2016 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2017 u32 rto = inet_csk(sk)->icsk_rto;
2018 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2019
2020 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2021}
2022
2023/*
2024 * Save and compile IPv4 options, return a pointer to it
2025 */
2026static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2027 struct sk_buff *skb)
2028{
2029 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2030 struct ip_options_rcu *dopt = NULL;
2031
2032 if (opt->optlen) {
2033 int opt_size = sizeof(*dopt) + opt->optlen;
2034
2035 dopt = kmalloc(opt_size, GFP_ATOMIC);
2036 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2037 kfree(dopt);
2038 dopt = NULL;
2039 }
2040 }
2041 return dopt;
2042}
2043
2044/* locally generated TCP pure ACKs have skb->truesize == 2
2045 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2046 * This is much faster than dissecting the packet to find out.
2047 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2048 */
2049static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2050{
2051 return skb->truesize == 2;
2052}
2053
2054static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2055{
2056 skb->truesize = 2;
2057}
2058
2059static inline int tcp_inq(struct sock *sk)
2060{
2061 struct tcp_sock *tp = tcp_sk(sk);
2062 int answ;
2063
2064 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2065 answ = 0;
2066 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2067 !tp->urg_data ||
2068 before(tp->urg_seq, tp->copied_seq) ||
2069 !before(tp->urg_seq, tp->rcv_nxt)) {
2070
2071 answ = tp->rcv_nxt - tp->copied_seq;
2072
2073 /* Subtract 1, if FIN was received */
2074 if (answ && sock_flag(sk, SOCK_DONE))
2075 answ--;
2076 } else {
2077 answ = tp->urg_seq - tp->copied_seq;
2078 }
2079
2080 return answ;
2081}
2082
2083int tcp_peek_len(struct socket *sock);
2084
2085static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2086{
2087 u16 segs_in;
2088
2089 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2090 tp->segs_in += segs_in;
2091 if (skb->len > tcp_hdrlen(skb))
2092 tp->data_segs_in += segs_in;
2093}
2094
2095/*
2096 * TCP listen path runs lockless.
2097 * We forced "struct sock" to be const qualified to make sure
2098 * we don't modify one of its field by mistake.
2099 * Here, we increment sk_drops which is an atomic_t, so we can safely
2100 * make sock writable again.
2101 */
2102static inline void tcp_listendrop(const struct sock *sk)
2103{
2104 atomic_inc(&((struct sock *)sk)->sk_drops);
2105 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2106}
2107
2108enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2109
2110/*
2111 * Interface for adding Upper Level Protocols over TCP
2112 */
2113
2114#define TCP_ULP_NAME_MAX 16
2115#define TCP_ULP_MAX 128
2116#define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2117
2118struct tcp_ulp_ops {
2119 struct list_head list;
2120
2121 /* initialize ulp */
2122 int (*init)(struct sock *sk);
2123 /* update ulp */
2124 void (*update)(struct sock *sk, struct proto *p);
2125 /* cleanup ulp */
2126 void (*release)(struct sock *sk);
2127 /* diagnostic */
2128 int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2129 size_t (*get_info_size)(const struct sock *sk);
2130
2131 char name[TCP_ULP_NAME_MAX];
2132 struct module *owner;
2133};
2134int tcp_register_ulp(struct tcp_ulp_ops *type);
2135void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2136int tcp_set_ulp(struct sock *sk, const char *name);
2137void tcp_get_available_ulp(char *buf, size_t len);
2138void tcp_cleanup_ulp(struct sock *sk);
2139void tcp_update_ulp(struct sock *sk, struct proto *p);
2140
2141#define MODULE_ALIAS_TCP_ULP(name) \
2142 __MODULE_INFO(alias, alias_userspace, name); \
2143 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2144
2145struct sk_msg;
2146struct sk_psock;
2147
2148int tcp_bpf_init(struct sock *sk);
2149void tcp_bpf_reinit(struct sock *sk);
2150int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2151 int flags);
2152int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
2153 int nonblock, int flags, int *addr_len);
2154int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock,
2155 struct msghdr *msg, int len, int flags);
2156
2157/* Call BPF_SOCK_OPS program that returns an int. If the return value
2158 * is < 0, then the BPF op failed (for example if the loaded BPF
2159 * program does not support the chosen operation or there is no BPF
2160 * program loaded).
2161 */
2162#ifdef CONFIG_BPF
2163static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2164{
2165 struct bpf_sock_ops_kern sock_ops;
2166 int ret;
2167
2168 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2169 if (sk_fullsock(sk)) {
2170 sock_ops.is_fullsock = 1;
2171 sock_owned_by_me(sk);
2172 }
2173
2174 sock_ops.sk = sk;
2175 sock_ops.op = op;
2176 if (nargs > 0)
2177 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2178
2179 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2180 if (ret == 0)
2181 ret = sock_ops.reply;
2182 else
2183 ret = -1;
2184 return ret;
2185}
2186
2187static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2188{
2189 u32 args[2] = {arg1, arg2};
2190
2191 return tcp_call_bpf(sk, op, 2, args);
2192}
2193
2194static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2195 u32 arg3)
2196{
2197 u32 args[3] = {arg1, arg2, arg3};
2198
2199 return tcp_call_bpf(sk, op, 3, args);
2200}
2201
2202#else
2203static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2204{
2205 return -EPERM;
2206}
2207
2208static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2209{
2210 return -EPERM;
2211}
2212
2213static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2214 u32 arg3)
2215{
2216 return -EPERM;
2217}
2218
2219#endif
2220
2221static inline u32 tcp_timeout_init(struct sock *sk)
2222{
2223 int timeout;
2224
2225 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2226
2227 if (timeout <= 0)
2228 timeout = TCP_TIMEOUT_INIT;
2229 return timeout;
2230}
2231
2232static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2233{
2234 int rwnd;
2235
2236 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2237
2238 if (rwnd < 0)
2239 rwnd = 0;
2240 return rwnd;
2241}
2242
2243static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2244{
2245 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2246}
2247
2248static inline void tcp_bpf_rtt(struct sock *sk)
2249{
2250 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2251 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2252}
2253
2254#if IS_ENABLED(CONFIG_SMC)
2255extern struct static_key_false tcp_have_smc;
2256#endif
2257
2258#if IS_ENABLED(CONFIG_TLS_DEVICE)
2259void clean_acked_data_enable(struct inet_connection_sock *icsk,
2260 void (*cad)(struct sock *sk, u32 ack_seq));
2261void clean_acked_data_disable(struct inet_connection_sock *icsk);
2262void clean_acked_data_flush(void);
2263#endif
2264
2265DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2266static inline void tcp_add_tx_delay(struct sk_buff *skb,
2267 const struct tcp_sock *tp)
2268{
2269 if (static_branch_unlikely(&tcp_tx_delay_enabled))
2270 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2271}
2272
2273/* Compute Earliest Departure Time for some control packets
2274 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2275 */
2276static inline u64 tcp_transmit_time(const struct sock *sk)
2277{
2278 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2279 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2280 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2281
2282 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
2283 }
2284 return 0;
2285}
2286
2287#endif /* _TCP_H */
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.h>
40#include <net/snmp.h>
41#include <net/ip.h>
42#include <net/tcp_states.h>
43#include <net/inet_ecn.h>
44#include <net/dst.h>
45
46#include <linux/seq_file.h>
47#include <linux/memcontrol.h>
48#include <linux/bpf-cgroup.h>
49
50extern struct inet_hashinfo tcp_hashinfo;
51
52extern struct percpu_counter tcp_orphan_count;
53void tcp_time_wait(struct sock *sk, int state, int timeo);
54
55#define MAX_TCP_HEADER (128 + MAX_HEADER)
56#define MAX_TCP_OPTION_SPACE 40
57
58/*
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
61 */
62#define MAX_TCP_WINDOW 32767U
63
64/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65#define TCP_MIN_MSS 88U
66
67/* The least MTU to use for probing */
68#define TCP_BASE_MSS 1024
69
70/* probing interval, default to 10 minutes as per RFC4821 */
71#define TCP_PROBE_INTERVAL 600
72
73/* Specify interval when tcp mtu probing will stop */
74#define TCP_PROBE_THRESHOLD 8
75
76/* After receiving this amount of duplicate ACKs fast retransmit starts. */
77#define TCP_FASTRETRANS_THRESH 3
78
79/* Maximal number of ACKs sent quickly to accelerate slow-start. */
80#define TCP_MAX_QUICKACKS 16U
81
82/* Maximal number of window scale according to RFC1323 */
83#define TCP_MAX_WSCALE 14U
84
85/* urg_data states */
86#define TCP_URG_VALID 0x0100
87#define TCP_URG_NOTYET 0x0200
88#define TCP_URG_READ 0x0400
89
90#define TCP_RETR1 3 /*
91 * This is how many retries it does before it
92 * tries to figure out if the gateway is
93 * down. Minimal RFC value is 3; it corresponds
94 * to ~3sec-8min depending on RTO.
95 */
96
97#define TCP_RETR2 15 /*
98 * This should take at least
99 * 90 minutes to time out.
100 * RFC1122 says that the limit is 100 sec.
101 * 15 is ~13-30min depending on RTO.
102 */
103
104#define TCP_SYN_RETRIES 6 /* This is how many retries are done
105 * when active opening a connection.
106 * RFC1122 says the minimum retry MUST
107 * be at least 180secs. Nevertheless
108 * this value is corresponding to
109 * 63secs of retransmission with the
110 * current initial RTO.
111 */
112
113#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
114 * when passive opening a connection.
115 * This is corresponding to 31secs of
116 * retransmission with the current
117 * initial RTO.
118 */
119
120#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
121 * state, about 60 seconds */
122#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
123 /* BSD style FIN_WAIT2 deadlock breaker.
124 * It used to be 3min, new value is 60sec,
125 * to combine FIN-WAIT-2 timeout with
126 * TIME-WAIT timer.
127 */
128
129#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
130#if HZ >= 100
131#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
132#define TCP_ATO_MIN ((unsigned)(HZ/25))
133#else
134#define TCP_DELACK_MIN 4U
135#define TCP_ATO_MIN 4U
136#endif
137#define TCP_RTO_MAX ((unsigned)(120*HZ))
138#define TCP_RTO_MIN ((unsigned)(HZ/5))
139#define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
140#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
141#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
142 * used as a fallback RTO for the
143 * initial data transmission if no
144 * valid RTT sample has been acquired,
145 * most likely due to retrans in 3WHS.
146 */
147
148#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
149 * for local resources.
150 */
151#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
152#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
153#define TCP_KEEPALIVE_INTVL (75*HZ)
154
155#define MAX_TCP_KEEPIDLE 32767
156#define MAX_TCP_KEEPINTVL 32767
157#define MAX_TCP_KEEPCNT 127
158#define MAX_TCP_SYNCNT 127
159
160#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
161
162#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
163#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
164 * after this time. It should be equal
165 * (or greater than) TCP_TIMEWAIT_LEN
166 * to provide reliability equal to one
167 * provided by timewait state.
168 */
169#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
170 * timestamps. It must be less than
171 * minimal timewait lifetime.
172 */
173/*
174 * TCP option
175 */
176
177#define TCPOPT_NOP 1 /* Padding */
178#define TCPOPT_EOL 0 /* End of options */
179#define TCPOPT_MSS 2 /* Segment size negotiating */
180#define TCPOPT_WINDOW 3 /* Window scaling */
181#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
182#define TCPOPT_SACK 5 /* SACK Block */
183#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
184#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
185#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
186#define TCPOPT_EXP 254 /* Experimental */
187/* Magic number to be after the option value for sharing TCP
188 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
189 */
190#define TCPOPT_FASTOPEN_MAGIC 0xF989
191#define TCPOPT_SMC_MAGIC 0xE2D4C3D9
192
193/*
194 * TCP option lengths
195 */
196
197#define TCPOLEN_MSS 4
198#define TCPOLEN_WINDOW 3
199#define TCPOLEN_SACK_PERM 2
200#define TCPOLEN_TIMESTAMP 10
201#define TCPOLEN_MD5SIG 18
202#define TCPOLEN_FASTOPEN_BASE 2
203#define TCPOLEN_EXP_FASTOPEN_BASE 4
204#define TCPOLEN_EXP_SMC_BASE 6
205
206/* But this is what stacks really send out. */
207#define TCPOLEN_TSTAMP_ALIGNED 12
208#define TCPOLEN_WSCALE_ALIGNED 4
209#define TCPOLEN_SACKPERM_ALIGNED 4
210#define TCPOLEN_SACK_BASE 2
211#define TCPOLEN_SACK_BASE_ALIGNED 4
212#define TCPOLEN_SACK_PERBLOCK 8
213#define TCPOLEN_MD5SIG_ALIGNED 20
214#define TCPOLEN_MSS_ALIGNED 4
215#define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
216
217/* Flags in tp->nonagle */
218#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
219#define TCP_NAGLE_CORK 2 /* Socket is corked */
220#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
221
222/* TCP thin-stream limits */
223#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
224
225/* TCP initial congestion window as per rfc6928 */
226#define TCP_INIT_CWND 10
227
228/* Bit Flags for sysctl_tcp_fastopen */
229#define TFO_CLIENT_ENABLE 1
230#define TFO_SERVER_ENABLE 2
231#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
232
233/* Accept SYN data w/o any cookie option */
234#define TFO_SERVER_COOKIE_NOT_REQD 0x200
235
236/* Force enable TFO on all listeners, i.e., not requiring the
237 * TCP_FASTOPEN socket option.
238 */
239#define TFO_SERVER_WO_SOCKOPT1 0x400
240
241
242/* sysctl variables for tcp */
243extern int sysctl_tcp_max_orphans;
244extern long sysctl_tcp_mem[3];
245
246#define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
247#define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
248
249extern atomic_long_t tcp_memory_allocated;
250extern struct percpu_counter tcp_sockets_allocated;
251extern unsigned long tcp_memory_pressure;
252
253/* optimized version of sk_under_memory_pressure() for TCP sockets */
254static inline bool tcp_under_memory_pressure(const struct sock *sk)
255{
256 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
257 mem_cgroup_under_socket_pressure(sk->sk_memcg))
258 return true;
259
260 return tcp_memory_pressure;
261}
262/*
263 * The next routines deal with comparing 32 bit unsigned ints
264 * and worry about wraparound (automatic with unsigned arithmetic).
265 */
266
267static inline bool before(__u32 seq1, __u32 seq2)
268{
269 return (__s32)(seq1-seq2) < 0;
270}
271#define after(seq2, seq1) before(seq1, seq2)
272
273/* is s2<=s1<=s3 ? */
274static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
275{
276 return seq3 - seq2 >= seq1 - seq2;
277}
278
279static inline bool tcp_out_of_memory(struct sock *sk)
280{
281 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
282 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
283 return true;
284 return false;
285}
286
287void sk_forced_mem_schedule(struct sock *sk, int size);
288
289static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
290{
291 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
292 int orphans = percpu_counter_read_positive(ocp);
293
294 if (orphans << shift > sysctl_tcp_max_orphans) {
295 orphans = percpu_counter_sum_positive(ocp);
296 if (orphans << shift > sysctl_tcp_max_orphans)
297 return true;
298 }
299 return false;
300}
301
302bool tcp_check_oom(struct sock *sk, int shift);
303
304
305extern struct proto tcp_prot;
306
307#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
308#define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
309#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
310#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
311
312void tcp_tasklet_init(void);
313
314void tcp_v4_err(struct sk_buff *skb, u32);
315
316void tcp_shutdown(struct sock *sk, int how);
317
318int tcp_v4_early_demux(struct sk_buff *skb);
319int tcp_v4_rcv(struct sk_buff *skb);
320
321int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
322int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
323int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
324int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
325 int flags);
326int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
327 size_t size, int flags);
328ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
329 size_t size, int flags);
330void tcp_release_cb(struct sock *sk);
331void tcp_wfree(struct sk_buff *skb);
332void tcp_write_timer_handler(struct sock *sk);
333void tcp_delack_timer_handler(struct sock *sk);
334int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
335int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
336void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
337 const struct tcphdr *th);
338void tcp_rcv_space_adjust(struct sock *sk);
339int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
340void tcp_twsk_destructor(struct sock *sk);
341ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
342 struct pipe_inode_info *pipe, size_t len,
343 unsigned int flags);
344
345static inline void tcp_dec_quickack_mode(struct sock *sk,
346 const unsigned int pkts)
347{
348 struct inet_connection_sock *icsk = inet_csk(sk);
349
350 if (icsk->icsk_ack.quick) {
351 if (pkts >= icsk->icsk_ack.quick) {
352 icsk->icsk_ack.quick = 0;
353 /* Leaving quickack mode we deflate ATO. */
354 icsk->icsk_ack.ato = TCP_ATO_MIN;
355 } else
356 icsk->icsk_ack.quick -= pkts;
357 }
358}
359
360#define TCP_ECN_OK 1
361#define TCP_ECN_QUEUE_CWR 2
362#define TCP_ECN_DEMAND_CWR 4
363#define TCP_ECN_SEEN 8
364
365enum tcp_tw_status {
366 TCP_TW_SUCCESS = 0,
367 TCP_TW_RST = 1,
368 TCP_TW_ACK = 2,
369 TCP_TW_SYN = 3
370};
371
372
373enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
374 struct sk_buff *skb,
375 const struct tcphdr *th);
376struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
377 struct request_sock *req, bool fastopen,
378 bool *lost_race);
379int tcp_child_process(struct sock *parent, struct sock *child,
380 struct sk_buff *skb);
381void tcp_enter_loss(struct sock *sk);
382void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
383void tcp_clear_retrans(struct tcp_sock *tp);
384void tcp_update_metrics(struct sock *sk);
385void tcp_init_metrics(struct sock *sk);
386void tcp_metrics_init(void);
387bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
388void tcp_close(struct sock *sk, long timeout);
389void tcp_init_sock(struct sock *sk);
390void tcp_init_transfer(struct sock *sk, int bpf_op);
391__poll_t tcp_poll(struct file *file, struct socket *sock,
392 struct poll_table_struct *wait);
393int tcp_getsockopt(struct sock *sk, int level, int optname,
394 char __user *optval, int __user *optlen);
395int tcp_setsockopt(struct sock *sk, int level, int optname,
396 char __user *optval, unsigned int optlen);
397int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
398 char __user *optval, int __user *optlen);
399int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
400 char __user *optval, unsigned int optlen);
401void tcp_set_keepalive(struct sock *sk, int val);
402void tcp_syn_ack_timeout(const struct request_sock *req);
403int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
404 int flags, int *addr_len);
405void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
406 struct tcp_options_received *opt_rx,
407 int estab, struct tcp_fastopen_cookie *foc);
408const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
409
410/*
411 * TCP v4 functions exported for the inet6 API
412 */
413
414void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
415void tcp_v4_mtu_reduced(struct sock *sk);
416void tcp_req_err(struct sock *sk, u32 seq, bool abort);
417int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
418struct sock *tcp_create_openreq_child(const struct sock *sk,
419 struct request_sock *req,
420 struct sk_buff *skb);
421void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
422struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
423 struct request_sock *req,
424 struct dst_entry *dst,
425 struct request_sock *req_unhash,
426 bool *own_req);
427int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
428int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
429int tcp_connect(struct sock *sk);
430enum tcp_synack_type {
431 TCP_SYNACK_NORMAL,
432 TCP_SYNACK_FASTOPEN,
433 TCP_SYNACK_COOKIE,
434};
435struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
436 struct request_sock *req,
437 struct tcp_fastopen_cookie *foc,
438 enum tcp_synack_type synack_type);
439int tcp_disconnect(struct sock *sk, int flags);
440
441void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
442int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
443void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
444
445/* From syncookies.c */
446struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
447 struct request_sock *req,
448 struct dst_entry *dst, u32 tsoff);
449int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
450 u32 cookie);
451struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
452#ifdef CONFIG_SYN_COOKIES
453
454/* Syncookies use a monotonic timer which increments every 60 seconds.
455 * This counter is used both as a hash input and partially encoded into
456 * the cookie value. A cookie is only validated further if the delta
457 * between the current counter value and the encoded one is less than this,
458 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
459 * the counter advances immediately after a cookie is generated).
460 */
461#define MAX_SYNCOOKIE_AGE 2
462#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
463#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
464
465/* syncookies: remember time of last synqueue overflow
466 * But do not dirty this field too often (once per second is enough)
467 * It is racy as we do not hold a lock, but race is very minor.
468 */
469static inline void tcp_synq_overflow(const struct sock *sk)
470{
471 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
472 unsigned long now = jiffies;
473
474 if (time_after(now, last_overflow + HZ))
475 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
476}
477
478/* syncookies: no recent synqueue overflow on this listening socket? */
479static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
480{
481 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
482
483 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
484}
485
486static inline u32 tcp_cookie_time(void)
487{
488 u64 val = get_jiffies_64();
489
490 do_div(val, TCP_SYNCOOKIE_PERIOD);
491 return val;
492}
493
494u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
495 u16 *mssp);
496__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
497u64 cookie_init_timestamp(struct request_sock *req);
498bool cookie_timestamp_decode(const struct net *net,
499 struct tcp_options_received *opt);
500bool cookie_ecn_ok(const struct tcp_options_received *opt,
501 const struct net *net, const struct dst_entry *dst);
502
503/* From net/ipv6/syncookies.c */
504int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
505 u32 cookie);
506struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
507
508u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
509 const struct tcphdr *th, u16 *mssp);
510__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
511#endif
512/* tcp_output.c */
513
514void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
515 int nonagle);
516int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
517int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
518void tcp_retransmit_timer(struct sock *sk);
519void tcp_xmit_retransmit_queue(struct sock *);
520void tcp_simple_retransmit(struct sock *);
521void tcp_enter_recovery(struct sock *sk, bool ece_ack);
522int tcp_trim_head(struct sock *, struct sk_buff *, u32);
523enum tcp_queue {
524 TCP_FRAG_IN_WRITE_QUEUE,
525 TCP_FRAG_IN_RTX_QUEUE,
526};
527int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
528 struct sk_buff *skb, u32 len,
529 unsigned int mss_now, gfp_t gfp);
530
531void tcp_send_probe0(struct sock *);
532void tcp_send_partial(struct sock *);
533int tcp_write_wakeup(struct sock *, int mib);
534void tcp_send_fin(struct sock *sk);
535void tcp_send_active_reset(struct sock *sk, gfp_t priority);
536int tcp_send_synack(struct sock *);
537void tcp_push_one(struct sock *, unsigned int mss_now);
538void tcp_send_ack(struct sock *sk);
539void tcp_send_delayed_ack(struct sock *sk);
540void tcp_send_loss_probe(struct sock *sk);
541bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
542void tcp_skb_collapse_tstamp(struct sk_buff *skb,
543 const struct sk_buff *next_skb);
544
545/* tcp_input.c */
546void tcp_rearm_rto(struct sock *sk);
547void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
548void tcp_reset(struct sock *sk);
549void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
550void tcp_fin(struct sock *sk);
551
552/* tcp_timer.c */
553void tcp_init_xmit_timers(struct sock *);
554static inline void tcp_clear_xmit_timers(struct sock *sk)
555{
556 hrtimer_cancel(&tcp_sk(sk)->pacing_timer);
557 inet_csk_clear_xmit_timers(sk);
558}
559
560unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
561unsigned int tcp_current_mss(struct sock *sk);
562
563/* Bound MSS / TSO packet size with the half of the window */
564static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
565{
566 int cutoff;
567
568 /* When peer uses tiny windows, there is no use in packetizing
569 * to sub-MSS pieces for the sake of SWS or making sure there
570 * are enough packets in the pipe for fast recovery.
571 *
572 * On the other hand, for extremely large MSS devices, handling
573 * smaller than MSS windows in this way does make sense.
574 */
575 if (tp->max_window > TCP_MSS_DEFAULT)
576 cutoff = (tp->max_window >> 1);
577 else
578 cutoff = tp->max_window;
579
580 if (cutoff && pktsize > cutoff)
581 return max_t(int, cutoff, 68U - tp->tcp_header_len);
582 else
583 return pktsize;
584}
585
586/* tcp.c */
587void tcp_get_info(struct sock *, struct tcp_info *);
588
589/* Read 'sendfile()'-style from a TCP socket */
590int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
591 sk_read_actor_t recv_actor);
592
593void tcp_initialize_rcv_mss(struct sock *sk);
594
595int tcp_mtu_to_mss(struct sock *sk, int pmtu);
596int tcp_mss_to_mtu(struct sock *sk, int mss);
597void tcp_mtup_init(struct sock *sk);
598void tcp_init_buffer_space(struct sock *sk);
599
600static inline void tcp_bound_rto(const struct sock *sk)
601{
602 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
603 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
604}
605
606static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
607{
608 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
609}
610
611static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
612{
613 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
614 ntohl(TCP_FLAG_ACK) |
615 snd_wnd);
616}
617
618static inline void tcp_fast_path_on(struct tcp_sock *tp)
619{
620 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
621}
622
623static inline void tcp_fast_path_check(struct sock *sk)
624{
625 struct tcp_sock *tp = tcp_sk(sk);
626
627 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
628 tp->rcv_wnd &&
629 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
630 !tp->urg_data)
631 tcp_fast_path_on(tp);
632}
633
634/* Compute the actual rto_min value */
635static inline u32 tcp_rto_min(struct sock *sk)
636{
637 const struct dst_entry *dst = __sk_dst_get(sk);
638 u32 rto_min = TCP_RTO_MIN;
639
640 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
641 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
642 return rto_min;
643}
644
645static inline u32 tcp_rto_min_us(struct sock *sk)
646{
647 return jiffies_to_usecs(tcp_rto_min(sk));
648}
649
650static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
651{
652 return dst_metric_locked(dst, RTAX_CC_ALGO);
653}
654
655/* Minimum RTT in usec. ~0 means not available. */
656static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
657{
658 return minmax_get(&tp->rtt_min);
659}
660
661/* Compute the actual receive window we are currently advertising.
662 * Rcv_nxt can be after the window if our peer push more data
663 * than the offered window.
664 */
665static inline u32 tcp_receive_window(const struct tcp_sock *tp)
666{
667 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
668
669 if (win < 0)
670 win = 0;
671 return (u32) win;
672}
673
674/* Choose a new window, without checks for shrinking, and without
675 * scaling applied to the result. The caller does these things
676 * if necessary. This is a "raw" window selection.
677 */
678u32 __tcp_select_window(struct sock *sk);
679
680void tcp_send_window_probe(struct sock *sk);
681
682/* TCP uses 32bit jiffies to save some space.
683 * Note that this is different from tcp_time_stamp, which
684 * historically has been the same until linux-4.13.
685 */
686#define tcp_jiffies32 ((u32)jiffies)
687
688/*
689 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
690 * It is no longer tied to jiffies, but to 1 ms clock.
691 * Note: double check if you want to use tcp_jiffies32 instead of this.
692 */
693#define TCP_TS_HZ 1000
694
695static inline u64 tcp_clock_ns(void)
696{
697 return local_clock();
698}
699
700static inline u64 tcp_clock_us(void)
701{
702 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
703}
704
705/* This should only be used in contexts where tp->tcp_mstamp is up to date */
706static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
707{
708 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
709}
710
711/* Could use tcp_clock_us() / 1000, but this version uses a single divide */
712static inline u32 tcp_time_stamp_raw(void)
713{
714 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
715}
716
717
718/* Refresh 1us clock of a TCP socket,
719 * ensuring monotically increasing values.
720 */
721static inline void tcp_mstamp_refresh(struct tcp_sock *tp)
722{
723 u64 val = tcp_clock_us();
724
725 if (val > tp->tcp_mstamp)
726 tp->tcp_mstamp = val;
727}
728
729static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
730{
731 return max_t(s64, t1 - t0, 0);
732}
733
734static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
735{
736 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ);
737}
738
739
740#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
741
742#define TCPHDR_FIN 0x01
743#define TCPHDR_SYN 0x02
744#define TCPHDR_RST 0x04
745#define TCPHDR_PSH 0x08
746#define TCPHDR_ACK 0x10
747#define TCPHDR_URG 0x20
748#define TCPHDR_ECE 0x40
749#define TCPHDR_CWR 0x80
750
751#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
752
753/* This is what the send packet queuing engine uses to pass
754 * TCP per-packet control information to the transmission code.
755 * We also store the host-order sequence numbers in here too.
756 * This is 44 bytes if IPV6 is enabled.
757 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
758 */
759struct tcp_skb_cb {
760 __u32 seq; /* Starting sequence number */
761 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
762 union {
763 /* Note : tcp_tw_isn is used in input path only
764 * (isn chosen by tcp_timewait_state_process())
765 *
766 * tcp_gso_segs/size are used in write queue only,
767 * cf tcp_skb_pcount()/tcp_skb_mss()
768 */
769 __u32 tcp_tw_isn;
770 struct {
771 u16 tcp_gso_segs;
772 u16 tcp_gso_size;
773 };
774 };
775 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
776
777 __u8 sacked; /* State flags for SACK. */
778#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
779#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
780#define TCPCB_LOST 0x04 /* SKB is lost */
781#define TCPCB_TAGBITS 0x07 /* All tag bits */
782#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
783#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
784#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
785 TCPCB_REPAIRED)
786
787 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
788 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
789 eor:1, /* Is skb MSG_EOR marked? */
790 has_rxtstamp:1, /* SKB has a RX timestamp */
791 unused:5;
792 __u32 ack_seq; /* Sequence number ACK'd */
793 union {
794 struct {
795 /* There is space for up to 24 bytes */
796 __u32 in_flight:30,/* Bytes in flight at transmit */
797 is_app_limited:1, /* cwnd not fully used? */
798 unused:1;
799 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
800 __u32 delivered;
801 /* start of send pipeline phase */
802 u64 first_tx_mstamp;
803 /* when we reached the "delivered" count */
804 u64 delivered_mstamp;
805 } tx; /* only used for outgoing skbs */
806 union {
807 struct inet_skb_parm h4;
808#if IS_ENABLED(CONFIG_IPV6)
809 struct inet6_skb_parm h6;
810#endif
811 } header; /* For incoming skbs */
812 struct {
813 __u32 key;
814 __u32 flags;
815 struct bpf_map *map;
816 void *data_end;
817 } bpf;
818 };
819};
820
821#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
822
823
824#if IS_ENABLED(CONFIG_IPV6)
825/* This is the variant of inet6_iif() that must be used by TCP,
826 * as TCP moves IP6CB into a different location in skb->cb[]
827 */
828static inline int tcp_v6_iif(const struct sk_buff *skb)
829{
830 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
831
832 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
833}
834
835/* TCP_SKB_CB reference means this can not be used from early demux */
836static inline int tcp_v6_sdif(const struct sk_buff *skb)
837{
838#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
839 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
840 return TCP_SKB_CB(skb)->header.h6.iif;
841#endif
842 return 0;
843}
844#endif
845
846static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
847{
848#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
849 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
850 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
851 return true;
852#endif
853 return false;
854}
855
856/* TCP_SKB_CB reference means this can not be used from early demux */
857static inline int tcp_v4_sdif(struct sk_buff *skb)
858{
859#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
860 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
861 return TCP_SKB_CB(skb)->header.h4.iif;
862#endif
863 return 0;
864}
865
866/* Due to TSO, an SKB can be composed of multiple actual
867 * packets. To keep these tracked properly, we use this.
868 */
869static inline int tcp_skb_pcount(const struct sk_buff *skb)
870{
871 return TCP_SKB_CB(skb)->tcp_gso_segs;
872}
873
874static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
875{
876 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
877}
878
879static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
880{
881 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
882}
883
884/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
885static inline int tcp_skb_mss(const struct sk_buff *skb)
886{
887 return TCP_SKB_CB(skb)->tcp_gso_size;
888}
889
890static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
891{
892 return likely(!TCP_SKB_CB(skb)->eor);
893}
894
895/* Events passed to congestion control interface */
896enum tcp_ca_event {
897 CA_EVENT_TX_START, /* first transmit when no packets in flight */
898 CA_EVENT_CWND_RESTART, /* congestion window restart */
899 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
900 CA_EVENT_LOSS, /* loss timeout */
901 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
902 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
903 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
904 CA_EVENT_NON_DELAYED_ACK,
905};
906
907/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
908enum tcp_ca_ack_event_flags {
909 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
910 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
911 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
912};
913
914/*
915 * Interface for adding new TCP congestion control handlers
916 */
917#define TCP_CA_NAME_MAX 16
918#define TCP_CA_MAX 128
919#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
920
921#define TCP_CA_UNSPEC 0
922
923/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
924#define TCP_CONG_NON_RESTRICTED 0x1
925/* Requires ECN/ECT set on all packets */
926#define TCP_CONG_NEEDS_ECN 0x2
927
928union tcp_cc_info;
929
930struct ack_sample {
931 u32 pkts_acked;
932 s32 rtt_us;
933 u32 in_flight;
934};
935
936/* A rate sample measures the number of (original/retransmitted) data
937 * packets delivered "delivered" over an interval of time "interval_us".
938 * The tcp_rate.c code fills in the rate sample, and congestion
939 * control modules that define a cong_control function to run at the end
940 * of ACK processing can optionally chose to consult this sample when
941 * setting cwnd and pacing rate.
942 * A sample is invalid if "delivered" or "interval_us" is negative.
943 */
944struct rate_sample {
945 u64 prior_mstamp; /* starting timestamp for interval */
946 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
947 s32 delivered; /* number of packets delivered over interval */
948 long interval_us; /* time for tp->delivered to incr "delivered" */
949 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
950 int losses; /* number of packets marked lost upon ACK */
951 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
952 u32 prior_in_flight; /* in flight before this ACK */
953 bool is_app_limited; /* is sample from packet with bubble in pipe? */
954 bool is_retrans; /* is sample from retransmission? */
955 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
956};
957
958struct tcp_congestion_ops {
959 struct list_head list;
960 u32 key;
961 u32 flags;
962
963 /* initialize private data (optional) */
964 void (*init)(struct sock *sk);
965 /* cleanup private data (optional) */
966 void (*release)(struct sock *sk);
967
968 /* return slow start threshold (required) */
969 u32 (*ssthresh)(struct sock *sk);
970 /* do new cwnd calculation (required) */
971 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
972 /* call before changing ca_state (optional) */
973 void (*set_state)(struct sock *sk, u8 new_state);
974 /* call when cwnd event occurs (optional) */
975 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
976 /* call when ack arrives (optional) */
977 void (*in_ack_event)(struct sock *sk, u32 flags);
978 /* new value of cwnd after loss (required) */
979 u32 (*undo_cwnd)(struct sock *sk);
980 /* hook for packet ack accounting (optional) */
981 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
982 /* override sysctl_tcp_min_tso_segs */
983 u32 (*min_tso_segs)(struct sock *sk);
984 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
985 u32 (*sndbuf_expand)(struct sock *sk);
986 /* call when packets are delivered to update cwnd and pacing rate,
987 * after all the ca_state processing. (optional)
988 */
989 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
990 /* get info for inet_diag (optional) */
991 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
992 union tcp_cc_info *info);
993
994 char name[TCP_CA_NAME_MAX];
995 struct module *owner;
996};
997
998int tcp_register_congestion_control(struct tcp_congestion_ops *type);
999void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1000
1001void tcp_assign_congestion_control(struct sock *sk);
1002void tcp_init_congestion_control(struct sock *sk);
1003void tcp_cleanup_congestion_control(struct sock *sk);
1004int tcp_set_default_congestion_control(struct net *net, const char *name);
1005void tcp_get_default_congestion_control(struct net *net, char *name);
1006void tcp_get_available_congestion_control(char *buf, size_t len);
1007void tcp_get_allowed_congestion_control(char *buf, size_t len);
1008int tcp_set_allowed_congestion_control(char *allowed);
1009int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit);
1010u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1011void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1012
1013u32 tcp_reno_ssthresh(struct sock *sk);
1014u32 tcp_reno_undo_cwnd(struct sock *sk);
1015void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1016extern struct tcp_congestion_ops tcp_reno;
1017
1018struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1019u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1020#ifdef CONFIG_INET
1021char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1022#else
1023static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1024{
1025 return NULL;
1026}
1027#endif
1028
1029static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1030{
1031 const struct inet_connection_sock *icsk = inet_csk(sk);
1032
1033 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1034}
1035
1036static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1037{
1038 struct inet_connection_sock *icsk = inet_csk(sk);
1039
1040 if (icsk->icsk_ca_ops->set_state)
1041 icsk->icsk_ca_ops->set_state(sk, ca_state);
1042 icsk->icsk_ca_state = ca_state;
1043}
1044
1045static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1046{
1047 const struct inet_connection_sock *icsk = inet_csk(sk);
1048
1049 if (icsk->icsk_ca_ops->cwnd_event)
1050 icsk->icsk_ca_ops->cwnd_event(sk, event);
1051}
1052
1053/* From tcp_rate.c */
1054void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1055void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1056 struct rate_sample *rs);
1057void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1058 bool is_sack_reneg, struct rate_sample *rs);
1059void tcp_rate_check_app_limited(struct sock *sk);
1060
1061/* These functions determine how the current flow behaves in respect of SACK
1062 * handling. SACK is negotiated with the peer, and therefore it can vary
1063 * between different flows.
1064 *
1065 * tcp_is_sack - SACK enabled
1066 * tcp_is_reno - No SACK
1067 */
1068static inline int tcp_is_sack(const struct tcp_sock *tp)
1069{
1070 return tp->rx_opt.sack_ok;
1071}
1072
1073static inline bool tcp_is_reno(const struct tcp_sock *tp)
1074{
1075 return !tcp_is_sack(tp);
1076}
1077
1078static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1079{
1080 return tp->sacked_out + tp->lost_out;
1081}
1082
1083/* This determines how many packets are "in the network" to the best
1084 * of our knowledge. In many cases it is conservative, but where
1085 * detailed information is available from the receiver (via SACK
1086 * blocks etc.) we can make more aggressive calculations.
1087 *
1088 * Use this for decisions involving congestion control, use just
1089 * tp->packets_out to determine if the send queue is empty or not.
1090 *
1091 * Read this equation as:
1092 *
1093 * "Packets sent once on transmission queue" MINUS
1094 * "Packets left network, but not honestly ACKed yet" PLUS
1095 * "Packets fast retransmitted"
1096 */
1097static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1098{
1099 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1100}
1101
1102#define TCP_INFINITE_SSTHRESH 0x7fffffff
1103
1104static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1105{
1106 return tp->snd_cwnd < tp->snd_ssthresh;
1107}
1108
1109static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1110{
1111 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1112}
1113
1114static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1115{
1116 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1117 (1 << inet_csk(sk)->icsk_ca_state);
1118}
1119
1120/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1121 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1122 * ssthresh.
1123 */
1124static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1125{
1126 const struct tcp_sock *tp = tcp_sk(sk);
1127
1128 if (tcp_in_cwnd_reduction(sk))
1129 return tp->snd_ssthresh;
1130 else
1131 return max(tp->snd_ssthresh,
1132 ((tp->snd_cwnd >> 1) +
1133 (tp->snd_cwnd >> 2)));
1134}
1135
1136/* Use define here intentionally to get WARN_ON location shown at the caller */
1137#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1138
1139void tcp_enter_cwr(struct sock *sk);
1140__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1141
1142/* The maximum number of MSS of available cwnd for which TSO defers
1143 * sending if not using sysctl_tcp_tso_win_divisor.
1144 */
1145static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1146{
1147 return 3;
1148}
1149
1150/* Returns end sequence number of the receiver's advertised window */
1151static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1152{
1153 return tp->snd_una + tp->snd_wnd;
1154}
1155
1156/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1157 * flexible approach. The RFC suggests cwnd should not be raised unless
1158 * it was fully used previously. And that's exactly what we do in
1159 * congestion avoidance mode. But in slow start we allow cwnd to grow
1160 * as long as the application has used half the cwnd.
1161 * Example :
1162 * cwnd is 10 (IW10), but application sends 9 frames.
1163 * We allow cwnd to reach 18 when all frames are ACKed.
1164 * This check is safe because it's as aggressive as slow start which already
1165 * risks 100% overshoot. The advantage is that we discourage application to
1166 * either send more filler packets or data to artificially blow up the cwnd
1167 * usage, and allow application-limited process to probe bw more aggressively.
1168 */
1169static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1170{
1171 const struct tcp_sock *tp = tcp_sk(sk);
1172
1173 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1174 if (tcp_in_slow_start(tp))
1175 return tp->snd_cwnd < 2 * tp->max_packets_out;
1176
1177 return tp->is_cwnd_limited;
1178}
1179
1180/* Something is really bad, we could not queue an additional packet,
1181 * because qdisc is full or receiver sent a 0 window.
1182 * We do not want to add fuel to the fire, or abort too early,
1183 * so make sure the timer we arm now is at least 200ms in the future,
1184 * regardless of current icsk_rto value (as it could be ~2ms)
1185 */
1186static inline unsigned long tcp_probe0_base(const struct sock *sk)
1187{
1188 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1189}
1190
1191/* Variant of inet_csk_rto_backoff() used for zero window probes */
1192static inline unsigned long tcp_probe0_when(const struct sock *sk,
1193 unsigned long max_when)
1194{
1195 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1196
1197 return (unsigned long)min_t(u64, when, max_when);
1198}
1199
1200static inline void tcp_check_probe_timer(struct sock *sk)
1201{
1202 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1203 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1204 tcp_probe0_base(sk), TCP_RTO_MAX);
1205}
1206
1207static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1208{
1209 tp->snd_wl1 = seq;
1210}
1211
1212static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1213{
1214 tp->snd_wl1 = seq;
1215}
1216
1217/*
1218 * Calculate(/check) TCP checksum
1219 */
1220static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1221 __be32 daddr, __wsum base)
1222{
1223 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1224}
1225
1226static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1227{
1228 return __skb_checksum_complete(skb);
1229}
1230
1231static inline bool tcp_checksum_complete(struct sk_buff *skb)
1232{
1233 return !skb_csum_unnecessary(skb) &&
1234 __tcp_checksum_complete(skb);
1235}
1236
1237bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1238int tcp_filter(struct sock *sk, struct sk_buff *skb);
1239
1240#undef STATE_TRACE
1241
1242#ifdef STATE_TRACE
1243static const char *statename[]={
1244 "Unused","Established","Syn Sent","Syn Recv",
1245 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1246 "Close Wait","Last ACK","Listen","Closing"
1247};
1248#endif
1249void tcp_set_state(struct sock *sk, int state);
1250
1251void tcp_done(struct sock *sk);
1252
1253int tcp_abort(struct sock *sk, int err);
1254
1255static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1256{
1257 rx_opt->dsack = 0;
1258 rx_opt->num_sacks = 0;
1259}
1260
1261u32 tcp_default_init_rwnd(u32 mss);
1262void tcp_cwnd_restart(struct sock *sk, s32 delta);
1263
1264static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1265{
1266 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1267 struct tcp_sock *tp = tcp_sk(sk);
1268 s32 delta;
1269
1270 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1271 ca_ops->cong_control)
1272 return;
1273 delta = tcp_jiffies32 - tp->lsndtime;
1274 if (delta > inet_csk(sk)->icsk_rto)
1275 tcp_cwnd_restart(sk, delta);
1276}
1277
1278/* Determine a window scaling and initial window to offer. */
1279void tcp_select_initial_window(const struct sock *sk, int __space,
1280 __u32 mss, __u32 *rcv_wnd,
1281 __u32 *window_clamp, int wscale_ok,
1282 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1283
1284static inline int tcp_win_from_space(const struct sock *sk, int space)
1285{
1286 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1287
1288 return tcp_adv_win_scale <= 0 ?
1289 (space>>(-tcp_adv_win_scale)) :
1290 space - (space>>tcp_adv_win_scale);
1291}
1292
1293/* Note: caller must be prepared to deal with negative returns */
1294static inline int tcp_space(const struct sock *sk)
1295{
1296 return tcp_win_from_space(sk, sk->sk_rcvbuf -
1297 atomic_read(&sk->sk_rmem_alloc));
1298}
1299
1300static inline int tcp_full_space(const struct sock *sk)
1301{
1302 return tcp_win_from_space(sk, sk->sk_rcvbuf);
1303}
1304
1305extern void tcp_openreq_init_rwin(struct request_sock *req,
1306 const struct sock *sk_listener,
1307 const struct dst_entry *dst);
1308
1309void tcp_enter_memory_pressure(struct sock *sk);
1310void tcp_leave_memory_pressure(struct sock *sk);
1311
1312static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1313{
1314 struct net *net = sock_net((struct sock *)tp);
1315
1316 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1317}
1318
1319static inline int keepalive_time_when(const struct tcp_sock *tp)
1320{
1321 struct net *net = sock_net((struct sock *)tp);
1322
1323 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1324}
1325
1326static inline int keepalive_probes(const struct tcp_sock *tp)
1327{
1328 struct net *net = sock_net((struct sock *)tp);
1329
1330 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1331}
1332
1333static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1334{
1335 const struct inet_connection_sock *icsk = &tp->inet_conn;
1336
1337 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1338 tcp_jiffies32 - tp->rcv_tstamp);
1339}
1340
1341static inline int tcp_fin_time(const struct sock *sk)
1342{
1343 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1344 const int rto = inet_csk(sk)->icsk_rto;
1345
1346 if (fin_timeout < (rto << 2) - (rto >> 1))
1347 fin_timeout = (rto << 2) - (rto >> 1);
1348
1349 return fin_timeout;
1350}
1351
1352static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1353 int paws_win)
1354{
1355 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1356 return true;
1357 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1358 return true;
1359 /*
1360 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1361 * then following tcp messages have valid values. Ignore 0 value,
1362 * or else 'negative' tsval might forbid us to accept their packets.
1363 */
1364 if (!rx_opt->ts_recent)
1365 return true;
1366 return false;
1367}
1368
1369static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1370 int rst)
1371{
1372 if (tcp_paws_check(rx_opt, 0))
1373 return false;
1374
1375 /* RST segments are not recommended to carry timestamp,
1376 and, if they do, it is recommended to ignore PAWS because
1377 "their cleanup function should take precedence over timestamps."
1378 Certainly, it is mistake. It is necessary to understand the reasons
1379 of this constraint to relax it: if peer reboots, clock may go
1380 out-of-sync and half-open connections will not be reset.
1381 Actually, the problem would be not existing if all
1382 the implementations followed draft about maintaining clock
1383 via reboots. Linux-2.2 DOES NOT!
1384
1385 However, we can relax time bounds for RST segments to MSL.
1386 */
1387 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1388 return false;
1389 return true;
1390}
1391
1392bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1393 int mib_idx, u32 *last_oow_ack_time);
1394
1395static inline void tcp_mib_init(struct net *net)
1396{
1397 /* See RFC 2012 */
1398 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1399 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1400 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1401 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1402}
1403
1404/* from STCP */
1405static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1406{
1407 tp->lost_skb_hint = NULL;
1408}
1409
1410static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1411{
1412 tcp_clear_retrans_hints_partial(tp);
1413 tp->retransmit_skb_hint = NULL;
1414}
1415
1416union tcp_md5_addr {
1417 struct in_addr a4;
1418#if IS_ENABLED(CONFIG_IPV6)
1419 struct in6_addr a6;
1420#endif
1421};
1422
1423/* - key database */
1424struct tcp_md5sig_key {
1425 struct hlist_node node;
1426 u8 keylen;
1427 u8 family; /* AF_INET or AF_INET6 */
1428 union tcp_md5_addr addr;
1429 u8 prefixlen;
1430 u8 key[TCP_MD5SIG_MAXKEYLEN];
1431 struct rcu_head rcu;
1432};
1433
1434/* - sock block */
1435struct tcp_md5sig_info {
1436 struct hlist_head head;
1437 struct rcu_head rcu;
1438};
1439
1440/* - pseudo header */
1441struct tcp4_pseudohdr {
1442 __be32 saddr;
1443 __be32 daddr;
1444 __u8 pad;
1445 __u8 protocol;
1446 __be16 len;
1447};
1448
1449struct tcp6_pseudohdr {
1450 struct in6_addr saddr;
1451 struct in6_addr daddr;
1452 __be32 len;
1453 __be32 protocol; /* including padding */
1454};
1455
1456union tcp_md5sum_block {
1457 struct tcp4_pseudohdr ip4;
1458#if IS_ENABLED(CONFIG_IPV6)
1459 struct tcp6_pseudohdr ip6;
1460#endif
1461};
1462
1463/* - pool: digest algorithm, hash description and scratch buffer */
1464struct tcp_md5sig_pool {
1465 struct ahash_request *md5_req;
1466 void *scratch;
1467};
1468
1469/* - functions */
1470int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1471 const struct sock *sk, const struct sk_buff *skb);
1472int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1473 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1474 gfp_t gfp);
1475int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1476 int family, u8 prefixlen);
1477struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1478 const struct sock *addr_sk);
1479
1480#ifdef CONFIG_TCP_MD5SIG
1481struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1482 const union tcp_md5_addr *addr,
1483 int family);
1484#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1485#else
1486static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1487 const union tcp_md5_addr *addr,
1488 int family)
1489{
1490 return NULL;
1491}
1492#define tcp_twsk_md5_key(twsk) NULL
1493#endif
1494
1495bool tcp_alloc_md5sig_pool(void);
1496
1497struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1498static inline void tcp_put_md5sig_pool(void)
1499{
1500 local_bh_enable();
1501}
1502
1503int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1504 unsigned int header_len);
1505int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1506 const struct tcp_md5sig_key *key);
1507
1508/* From tcp_fastopen.c */
1509void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1510 struct tcp_fastopen_cookie *cookie);
1511void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1512 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1513 u16 try_exp);
1514struct tcp_fastopen_request {
1515 /* Fast Open cookie. Size 0 means a cookie request */
1516 struct tcp_fastopen_cookie cookie;
1517 struct msghdr *data; /* data in MSG_FASTOPEN */
1518 size_t size;
1519 int copied; /* queued in tcp_connect() */
1520};
1521void tcp_free_fastopen_req(struct tcp_sock *tp);
1522void tcp_fastopen_destroy_cipher(struct sock *sk);
1523void tcp_fastopen_ctx_destroy(struct net *net);
1524int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1525 void *key, unsigned int len);
1526void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1527struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1528 struct request_sock *req,
1529 struct tcp_fastopen_cookie *foc,
1530 const struct dst_entry *dst);
1531void tcp_fastopen_init_key_once(struct net *net);
1532bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1533 struct tcp_fastopen_cookie *cookie);
1534bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1535#define TCP_FASTOPEN_KEY_LENGTH 16
1536
1537/* Fastopen key context */
1538struct tcp_fastopen_context {
1539 struct crypto_cipher *tfm;
1540 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1541 struct rcu_head rcu;
1542};
1543
1544extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1545void tcp_fastopen_active_disable(struct sock *sk);
1546bool tcp_fastopen_active_should_disable(struct sock *sk);
1547void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1548void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1549
1550/* Latencies incurred by various limits for a sender. They are
1551 * chronograph-like stats that are mutually exclusive.
1552 */
1553enum tcp_chrono {
1554 TCP_CHRONO_UNSPEC,
1555 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1556 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1557 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1558 __TCP_CHRONO_MAX,
1559};
1560
1561void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1562void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1563
1564/* This helper is needed, because skb->tcp_tsorted_anchor uses
1565 * the same memory storage than skb->destructor/_skb_refdst
1566 */
1567static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1568{
1569 skb->destructor = NULL;
1570 skb->_skb_refdst = 0UL;
1571}
1572
1573#define tcp_skb_tsorted_save(skb) { \
1574 unsigned long _save = skb->_skb_refdst; \
1575 skb->_skb_refdst = 0UL;
1576
1577#define tcp_skb_tsorted_restore(skb) \
1578 skb->_skb_refdst = _save; \
1579}
1580
1581void tcp_write_queue_purge(struct sock *sk);
1582
1583static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1584{
1585 return skb_rb_first(&sk->tcp_rtx_queue);
1586}
1587
1588static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1589{
1590 return skb_peek(&sk->sk_write_queue);
1591}
1592
1593static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1594{
1595 return skb_peek_tail(&sk->sk_write_queue);
1596}
1597
1598#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1599 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1600
1601static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1602{
1603 return skb_peek(&sk->sk_write_queue);
1604}
1605
1606static inline bool tcp_skb_is_last(const struct sock *sk,
1607 const struct sk_buff *skb)
1608{
1609 return skb_queue_is_last(&sk->sk_write_queue, skb);
1610}
1611
1612static inline bool tcp_write_queue_empty(const struct sock *sk)
1613{
1614 return skb_queue_empty(&sk->sk_write_queue);
1615}
1616
1617static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1618{
1619 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1620}
1621
1622static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1623{
1624 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1625}
1626
1627static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1628{
1629 if (tcp_write_queue_empty(sk))
1630 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1631}
1632
1633static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1634{
1635 __skb_queue_tail(&sk->sk_write_queue, skb);
1636}
1637
1638static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1639{
1640 __tcp_add_write_queue_tail(sk, skb);
1641
1642 /* Queue it, remembering where we must start sending. */
1643 if (sk->sk_write_queue.next == skb)
1644 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1645}
1646
1647/* Insert new before skb on the write queue of sk. */
1648static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1649 struct sk_buff *skb,
1650 struct sock *sk)
1651{
1652 __skb_queue_before(&sk->sk_write_queue, skb, new);
1653}
1654
1655static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1656{
1657 tcp_skb_tsorted_anchor_cleanup(skb);
1658 __skb_unlink(skb, &sk->sk_write_queue);
1659}
1660
1661void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1662
1663static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1664{
1665 tcp_skb_tsorted_anchor_cleanup(skb);
1666 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1667}
1668
1669static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1670{
1671 list_del(&skb->tcp_tsorted_anchor);
1672 tcp_rtx_queue_unlink(skb, sk);
1673 sk_wmem_free_skb(sk, skb);
1674}
1675
1676static inline void tcp_push_pending_frames(struct sock *sk)
1677{
1678 if (tcp_send_head(sk)) {
1679 struct tcp_sock *tp = tcp_sk(sk);
1680
1681 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1682 }
1683}
1684
1685/* Start sequence of the skb just after the highest skb with SACKed
1686 * bit, valid only if sacked_out > 0 or when the caller has ensured
1687 * validity by itself.
1688 */
1689static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1690{
1691 if (!tp->sacked_out)
1692 return tp->snd_una;
1693
1694 if (tp->highest_sack == NULL)
1695 return tp->snd_nxt;
1696
1697 return TCP_SKB_CB(tp->highest_sack)->seq;
1698}
1699
1700static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1701{
1702 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1703}
1704
1705static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1706{
1707 return tcp_sk(sk)->highest_sack;
1708}
1709
1710static inline void tcp_highest_sack_reset(struct sock *sk)
1711{
1712 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1713}
1714
1715/* Called when old skb is about to be deleted and replaced by new skb */
1716static inline void tcp_highest_sack_replace(struct sock *sk,
1717 struct sk_buff *old,
1718 struct sk_buff *new)
1719{
1720 if (old == tcp_highest_sack(sk))
1721 tcp_sk(sk)->highest_sack = new;
1722}
1723
1724/* This helper checks if socket has IP_TRANSPARENT set */
1725static inline bool inet_sk_transparent(const struct sock *sk)
1726{
1727 switch (sk->sk_state) {
1728 case TCP_TIME_WAIT:
1729 return inet_twsk(sk)->tw_transparent;
1730 case TCP_NEW_SYN_RECV:
1731 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1732 }
1733 return inet_sk(sk)->transparent;
1734}
1735
1736/* Determines whether this is a thin stream (which may suffer from
1737 * increased latency). Used to trigger latency-reducing mechanisms.
1738 */
1739static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1740{
1741 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1742}
1743
1744/* /proc */
1745enum tcp_seq_states {
1746 TCP_SEQ_STATE_LISTENING,
1747 TCP_SEQ_STATE_ESTABLISHED,
1748};
1749
1750int tcp_seq_open(struct inode *inode, struct file *file);
1751
1752struct tcp_seq_afinfo {
1753 char *name;
1754 sa_family_t family;
1755 const struct file_operations *seq_fops;
1756 struct seq_operations seq_ops;
1757};
1758
1759struct tcp_iter_state {
1760 struct seq_net_private p;
1761 sa_family_t family;
1762 enum tcp_seq_states state;
1763 struct sock *syn_wait_sk;
1764 int bucket, offset, sbucket, num;
1765 loff_t last_pos;
1766};
1767
1768int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1769void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1770
1771extern struct request_sock_ops tcp_request_sock_ops;
1772extern struct request_sock_ops tcp6_request_sock_ops;
1773
1774void tcp_v4_destroy_sock(struct sock *sk);
1775
1776struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1777 netdev_features_t features);
1778struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1779int tcp_gro_complete(struct sk_buff *skb);
1780
1781void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1782
1783static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1784{
1785 struct net *net = sock_net((struct sock *)tp);
1786 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1787}
1788
1789static inline bool tcp_stream_memory_free(const struct sock *sk)
1790{
1791 const struct tcp_sock *tp = tcp_sk(sk);
1792 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1793
1794 return notsent_bytes < tcp_notsent_lowat(tp);
1795}
1796
1797#ifdef CONFIG_PROC_FS
1798int tcp4_proc_init(void);
1799void tcp4_proc_exit(void);
1800#endif
1801
1802int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1803int tcp_conn_request(struct request_sock_ops *rsk_ops,
1804 const struct tcp_request_sock_ops *af_ops,
1805 struct sock *sk, struct sk_buff *skb);
1806
1807/* TCP af-specific functions */
1808struct tcp_sock_af_ops {
1809#ifdef CONFIG_TCP_MD5SIG
1810 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1811 const struct sock *addr_sk);
1812 int (*calc_md5_hash)(char *location,
1813 const struct tcp_md5sig_key *md5,
1814 const struct sock *sk,
1815 const struct sk_buff *skb);
1816 int (*md5_parse)(struct sock *sk,
1817 int optname,
1818 char __user *optval,
1819 int optlen);
1820#endif
1821};
1822
1823struct tcp_request_sock_ops {
1824 u16 mss_clamp;
1825#ifdef CONFIG_TCP_MD5SIG
1826 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1827 const struct sock *addr_sk);
1828 int (*calc_md5_hash) (char *location,
1829 const struct tcp_md5sig_key *md5,
1830 const struct sock *sk,
1831 const struct sk_buff *skb);
1832#endif
1833 void (*init_req)(struct request_sock *req,
1834 const struct sock *sk_listener,
1835 struct sk_buff *skb);
1836#ifdef CONFIG_SYN_COOKIES
1837 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1838 __u16 *mss);
1839#endif
1840 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1841 const struct request_sock *req);
1842 u32 (*init_seq)(const struct sk_buff *skb);
1843 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1844 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1845 struct flowi *fl, struct request_sock *req,
1846 struct tcp_fastopen_cookie *foc,
1847 enum tcp_synack_type synack_type);
1848};
1849
1850#ifdef CONFIG_SYN_COOKIES
1851static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1852 const struct sock *sk, struct sk_buff *skb,
1853 __u16 *mss)
1854{
1855 tcp_synq_overflow(sk);
1856 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1857 return ops->cookie_init_seq(skb, mss);
1858}
1859#else
1860static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1861 const struct sock *sk, struct sk_buff *skb,
1862 __u16 *mss)
1863{
1864 return 0;
1865}
1866#endif
1867
1868int tcpv4_offload_init(void);
1869
1870void tcp_v4_init(void);
1871void tcp_init(void);
1872
1873/* tcp_recovery.c */
1874extern void tcp_rack_mark_lost(struct sock *sk);
1875extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1876 u64 xmit_time);
1877extern void tcp_rack_reo_timeout(struct sock *sk);
1878extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
1879
1880/* At how many usecs into the future should the RTO fire? */
1881static inline s64 tcp_rto_delta_us(const struct sock *sk)
1882{
1883 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
1884 u32 rto = inet_csk(sk)->icsk_rto;
1885 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto);
1886
1887 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
1888}
1889
1890/*
1891 * Save and compile IPv4 options, return a pointer to it
1892 */
1893static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
1894 struct sk_buff *skb)
1895{
1896 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1897 struct ip_options_rcu *dopt = NULL;
1898
1899 if (opt->optlen) {
1900 int opt_size = sizeof(*dopt) + opt->optlen;
1901
1902 dopt = kmalloc(opt_size, GFP_ATOMIC);
1903 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
1904 kfree(dopt);
1905 dopt = NULL;
1906 }
1907 }
1908 return dopt;
1909}
1910
1911/* locally generated TCP pure ACKs have skb->truesize == 2
1912 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1913 * This is much faster than dissecting the packet to find out.
1914 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1915 */
1916static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1917{
1918 return skb->truesize == 2;
1919}
1920
1921static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1922{
1923 skb->truesize = 2;
1924}
1925
1926static inline int tcp_inq(struct sock *sk)
1927{
1928 struct tcp_sock *tp = tcp_sk(sk);
1929 int answ;
1930
1931 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1932 answ = 0;
1933 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1934 !tp->urg_data ||
1935 before(tp->urg_seq, tp->copied_seq) ||
1936 !before(tp->urg_seq, tp->rcv_nxt)) {
1937
1938 answ = tp->rcv_nxt - tp->copied_seq;
1939
1940 /* Subtract 1, if FIN was received */
1941 if (answ && sock_flag(sk, SOCK_DONE))
1942 answ--;
1943 } else {
1944 answ = tp->urg_seq - tp->copied_seq;
1945 }
1946
1947 return answ;
1948}
1949
1950int tcp_peek_len(struct socket *sock);
1951
1952static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1953{
1954 u16 segs_in;
1955
1956 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1957 tp->segs_in += segs_in;
1958 if (skb->len > tcp_hdrlen(skb))
1959 tp->data_segs_in += segs_in;
1960}
1961
1962/*
1963 * TCP listen path runs lockless.
1964 * We forced "struct sock" to be const qualified to make sure
1965 * we don't modify one of its field by mistake.
1966 * Here, we increment sk_drops which is an atomic_t, so we can safely
1967 * make sock writable again.
1968 */
1969static inline void tcp_listendrop(const struct sock *sk)
1970{
1971 atomic_inc(&((struct sock *)sk)->sk_drops);
1972 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
1973}
1974
1975enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
1976
1977/*
1978 * Interface for adding Upper Level Protocols over TCP
1979 */
1980
1981#define TCP_ULP_NAME_MAX 16
1982#define TCP_ULP_MAX 128
1983#define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
1984
1985enum {
1986 TCP_ULP_TLS,
1987 TCP_ULP_BPF,
1988};
1989
1990struct tcp_ulp_ops {
1991 struct list_head list;
1992
1993 /* initialize ulp */
1994 int (*init)(struct sock *sk);
1995 /* cleanup ulp */
1996 void (*release)(struct sock *sk);
1997
1998 int uid;
1999 char name[TCP_ULP_NAME_MAX];
2000 bool user_visible;
2001 struct module *owner;
2002};
2003int tcp_register_ulp(struct tcp_ulp_ops *type);
2004void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2005int tcp_set_ulp(struct sock *sk, const char *name);
2006int tcp_set_ulp_id(struct sock *sk, const int ulp);
2007void tcp_get_available_ulp(char *buf, size_t len);
2008void tcp_cleanup_ulp(struct sock *sk);
2009
2010/* Call BPF_SOCK_OPS program that returns an int. If the return value
2011 * is < 0, then the BPF op failed (for example if the loaded BPF
2012 * program does not support the chosen operation or there is no BPF
2013 * program loaded).
2014 */
2015#ifdef CONFIG_BPF
2016static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2017{
2018 struct bpf_sock_ops_kern sock_ops;
2019 int ret;
2020
2021 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2022 if (sk_fullsock(sk)) {
2023 sock_ops.is_fullsock = 1;
2024 sock_owned_by_me(sk);
2025 }
2026
2027 sock_ops.sk = sk;
2028 sock_ops.op = op;
2029 if (nargs > 0)
2030 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2031
2032 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2033 if (ret == 0)
2034 ret = sock_ops.reply;
2035 else
2036 ret = -1;
2037 return ret;
2038}
2039
2040static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2041{
2042 u32 args[2] = {arg1, arg2};
2043
2044 return tcp_call_bpf(sk, op, 2, args);
2045}
2046
2047static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2048 u32 arg3)
2049{
2050 u32 args[3] = {arg1, arg2, arg3};
2051
2052 return tcp_call_bpf(sk, op, 3, args);
2053}
2054
2055#else
2056static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2057{
2058 return -EPERM;
2059}
2060
2061static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2062{
2063 return -EPERM;
2064}
2065
2066static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2067 u32 arg3)
2068{
2069 return -EPERM;
2070}
2071
2072#endif
2073
2074static inline u32 tcp_timeout_init(struct sock *sk)
2075{
2076 int timeout;
2077
2078 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2079
2080 if (timeout <= 0)
2081 timeout = TCP_TIMEOUT_INIT;
2082 return timeout;
2083}
2084
2085static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2086{
2087 int rwnd;
2088
2089 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2090
2091 if (rwnd < 0)
2092 rwnd = 0;
2093 return rwnd;
2094}
2095
2096static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2097{
2098 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2099}
2100
2101#if IS_ENABLED(CONFIG_SMC)
2102extern struct static_key_false tcp_have_smc;
2103#endif
2104#endif /* _TCP_H */