Loading...
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/dmaengine.h>
31#include <linux/crypto.h>
32#include <linux/cryptohash.h>
33#include <linux/kref.h>
34#include <linux/ktime.h>
35
36#include <net/inet_connection_sock.h>
37#include <net/inet_timewait_sock.h>
38#include <net/inet_hashtables.h>
39#include <net/checksum.h>
40#include <net/request_sock.h>
41#include <net/sock.h>
42#include <net/snmp.h>
43#include <net/ip.h>
44#include <net/tcp_states.h>
45#include <net/inet_ecn.h>
46#include <net/dst.h>
47
48#include <linux/seq_file.h>
49#include <linux/memcontrol.h>
50
51extern struct inet_hashinfo tcp_hashinfo;
52
53extern struct percpu_counter tcp_orphan_count;
54void tcp_time_wait(struct sock *sk, int state, int timeo);
55
56#define MAX_TCP_HEADER (128 + MAX_HEADER)
57#define MAX_TCP_OPTION_SPACE 40
58
59/*
60 * Never offer a window over 32767 without using window scaling. Some
61 * poor stacks do signed 16bit maths!
62 */
63#define MAX_TCP_WINDOW 32767U
64
65/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
66#define TCP_MIN_MSS 88U
67
68/* The least MTU to use for probing */
69#define TCP_BASE_MSS 512
70
71/* After receiving this amount of duplicate ACKs fast retransmit starts. */
72#define TCP_FASTRETRANS_THRESH 3
73
74/* Maximal reordering. */
75#define TCP_MAX_REORDERING 127
76
77/* Maximal number of ACKs sent quickly to accelerate slow-start. */
78#define TCP_MAX_QUICKACKS 16U
79
80/* urg_data states */
81#define TCP_URG_VALID 0x0100
82#define TCP_URG_NOTYET 0x0200
83#define TCP_URG_READ 0x0400
84
85#define TCP_RETR1 3 /*
86 * This is how many retries it does before it
87 * tries to figure out if the gateway is
88 * down. Minimal RFC value is 3; it corresponds
89 * to ~3sec-8min depending on RTO.
90 */
91
92#define TCP_RETR2 15 /*
93 * This should take at least
94 * 90 minutes to time out.
95 * RFC1122 says that the limit is 100 sec.
96 * 15 is ~13-30min depending on RTO.
97 */
98
99#define TCP_SYN_RETRIES 6 /* This is how many retries are done
100 * when active opening a connection.
101 * RFC1122 says the minimum retry MUST
102 * be at least 180secs. Nevertheless
103 * this value is corresponding to
104 * 63secs of retransmission with the
105 * current initial RTO.
106 */
107
108#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
109 * when passive opening a connection.
110 * This is corresponding to 31secs of
111 * retransmission with the current
112 * initial RTO.
113 */
114
115#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
116 * state, about 60 seconds */
117#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
118 /* BSD style FIN_WAIT2 deadlock breaker.
119 * It used to be 3min, new value is 60sec,
120 * to combine FIN-WAIT-2 timeout with
121 * TIME-WAIT timer.
122 */
123
124#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
125#if HZ >= 100
126#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
127#define TCP_ATO_MIN ((unsigned)(HZ/25))
128#else
129#define TCP_DELACK_MIN 4U
130#define TCP_ATO_MIN 4U
131#endif
132#define TCP_RTO_MAX ((unsigned)(120*HZ))
133#define TCP_RTO_MIN ((unsigned)(HZ/5))
134#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
135#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
136 * used as a fallback RTO for the
137 * initial data transmission if no
138 * valid RTT sample has been acquired,
139 * most likely due to retrans in 3WHS.
140 */
141
142#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
143 * for local resources.
144 */
145
146#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
147#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
148#define TCP_KEEPALIVE_INTVL (75*HZ)
149
150#define MAX_TCP_KEEPIDLE 32767
151#define MAX_TCP_KEEPINTVL 32767
152#define MAX_TCP_KEEPCNT 127
153#define MAX_TCP_SYNCNT 127
154
155#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
156
157#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
158#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
159 * after this time. It should be equal
160 * (or greater than) TCP_TIMEWAIT_LEN
161 * to provide reliability equal to one
162 * provided by timewait state.
163 */
164#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
165 * timestamps. It must be less than
166 * minimal timewait lifetime.
167 */
168/*
169 * TCP option
170 */
171
172#define TCPOPT_NOP 1 /* Padding */
173#define TCPOPT_EOL 0 /* End of options */
174#define TCPOPT_MSS 2 /* Segment size negotiating */
175#define TCPOPT_WINDOW 3 /* Window scaling */
176#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
177#define TCPOPT_SACK 5 /* SACK Block */
178#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
179#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
180#define TCPOPT_EXP 254 /* Experimental */
181/* Magic number to be after the option value for sharing TCP
182 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
183 */
184#define TCPOPT_FASTOPEN_MAGIC 0xF989
185
186/*
187 * TCP option lengths
188 */
189
190#define TCPOLEN_MSS 4
191#define TCPOLEN_WINDOW 3
192#define TCPOLEN_SACK_PERM 2
193#define TCPOLEN_TIMESTAMP 10
194#define TCPOLEN_MD5SIG 18
195#define TCPOLEN_EXP_FASTOPEN_BASE 4
196
197/* But this is what stacks really send out. */
198#define TCPOLEN_TSTAMP_ALIGNED 12
199#define TCPOLEN_WSCALE_ALIGNED 4
200#define TCPOLEN_SACKPERM_ALIGNED 4
201#define TCPOLEN_SACK_BASE 2
202#define TCPOLEN_SACK_BASE_ALIGNED 4
203#define TCPOLEN_SACK_PERBLOCK 8
204#define TCPOLEN_MD5SIG_ALIGNED 20
205#define TCPOLEN_MSS_ALIGNED 4
206
207/* Flags in tp->nonagle */
208#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
209#define TCP_NAGLE_CORK 2 /* Socket is corked */
210#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
211
212/* TCP thin-stream limits */
213#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
214
215/* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
216#define TCP_INIT_CWND 10
217
218/* Bit Flags for sysctl_tcp_fastopen */
219#define TFO_CLIENT_ENABLE 1
220#define TFO_SERVER_ENABLE 2
221#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
222
223/* Process SYN data but skip cookie validation */
224#define TFO_SERVER_COOKIE_NOT_CHKED 0x100
225/* Accept SYN data w/o any cookie option */
226#define TFO_SERVER_COOKIE_NOT_REQD 0x200
227
228/* Force enable TFO on all listeners, i.e., not requiring the
229 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
230 */
231#define TFO_SERVER_WO_SOCKOPT1 0x400
232#define TFO_SERVER_WO_SOCKOPT2 0x800
233/* Always create TFO child sockets on a TFO listener even when
234 * cookie/data not present. (For testing purpose!)
235 */
236#define TFO_SERVER_ALWAYS 0x1000
237
238extern struct inet_timewait_death_row tcp_death_row;
239
240/* sysctl variables for tcp */
241extern int sysctl_tcp_timestamps;
242extern int sysctl_tcp_window_scaling;
243extern int sysctl_tcp_sack;
244extern int sysctl_tcp_fin_timeout;
245extern int sysctl_tcp_keepalive_time;
246extern int sysctl_tcp_keepalive_probes;
247extern int sysctl_tcp_keepalive_intvl;
248extern int sysctl_tcp_syn_retries;
249extern int sysctl_tcp_synack_retries;
250extern int sysctl_tcp_retries1;
251extern int sysctl_tcp_retries2;
252extern int sysctl_tcp_orphan_retries;
253extern int sysctl_tcp_syncookies;
254extern int sysctl_tcp_fastopen;
255extern int sysctl_tcp_retrans_collapse;
256extern int sysctl_tcp_stdurg;
257extern int sysctl_tcp_rfc1337;
258extern int sysctl_tcp_abort_on_overflow;
259extern int sysctl_tcp_max_orphans;
260extern int sysctl_tcp_fack;
261extern int sysctl_tcp_reordering;
262extern int sysctl_tcp_dsack;
263extern long sysctl_tcp_mem[3];
264extern int sysctl_tcp_wmem[3];
265extern int sysctl_tcp_rmem[3];
266extern int sysctl_tcp_app_win;
267extern int sysctl_tcp_adv_win_scale;
268extern int sysctl_tcp_tw_reuse;
269extern int sysctl_tcp_frto;
270extern int sysctl_tcp_low_latency;
271extern int sysctl_tcp_dma_copybreak;
272extern int sysctl_tcp_nometrics_save;
273extern int sysctl_tcp_moderate_rcvbuf;
274extern int sysctl_tcp_tso_win_divisor;
275extern int sysctl_tcp_mtu_probing;
276extern int sysctl_tcp_base_mss;
277extern int sysctl_tcp_workaround_signed_windows;
278extern int sysctl_tcp_slow_start_after_idle;
279extern int sysctl_tcp_thin_linear_timeouts;
280extern int sysctl_tcp_thin_dupack;
281extern int sysctl_tcp_early_retrans;
282extern int sysctl_tcp_limit_output_bytes;
283extern int sysctl_tcp_challenge_ack_limit;
284extern unsigned int sysctl_tcp_notsent_lowat;
285extern int sysctl_tcp_min_tso_segs;
286extern int sysctl_tcp_autocorking;
287
288extern atomic_long_t tcp_memory_allocated;
289extern struct percpu_counter tcp_sockets_allocated;
290extern int tcp_memory_pressure;
291
292/*
293 * The next routines deal with comparing 32 bit unsigned ints
294 * and worry about wraparound (automatic with unsigned arithmetic).
295 */
296
297static inline bool before(__u32 seq1, __u32 seq2)
298{
299 return (__s32)(seq1-seq2) < 0;
300}
301#define after(seq2, seq1) before(seq1, seq2)
302
303/* is s2<=s1<=s3 ? */
304static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
305{
306 return seq3 - seq2 >= seq1 - seq2;
307}
308
309static inline bool tcp_out_of_memory(struct sock *sk)
310{
311 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
312 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
313 return true;
314 return false;
315}
316
317static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
318{
319 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
320 int orphans = percpu_counter_read_positive(ocp);
321
322 if (orphans << shift > sysctl_tcp_max_orphans) {
323 orphans = percpu_counter_sum_positive(ocp);
324 if (orphans << shift > sysctl_tcp_max_orphans)
325 return true;
326 }
327 return false;
328}
329
330bool tcp_check_oom(struct sock *sk, int shift);
331
332/* syncookies: remember time of last synqueue overflow */
333static inline void tcp_synq_overflow(struct sock *sk)
334{
335 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
336}
337
338/* syncookies: no recent synqueue overflow on this listening socket? */
339static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
340{
341 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
342 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
343}
344
345extern struct proto tcp_prot;
346
347#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
348#define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
349#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
350#define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
351#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
352
353void tcp_tasklet_init(void);
354
355void tcp_v4_err(struct sk_buff *skb, u32);
356
357void tcp_shutdown(struct sock *sk, int how);
358
359void tcp_v4_early_demux(struct sk_buff *skb);
360int tcp_v4_rcv(struct sk_buff *skb);
361
362int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
363int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
364 size_t size);
365int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
366 int flags);
367void tcp_release_cb(struct sock *sk);
368void tcp_wfree(struct sk_buff *skb);
369void tcp_write_timer_handler(struct sock *sk);
370void tcp_delack_timer_handler(struct sock *sk);
371int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
372int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
373 const struct tcphdr *th, unsigned int len);
374void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
375 const struct tcphdr *th, unsigned int len);
376void tcp_rcv_space_adjust(struct sock *sk);
377void tcp_cleanup_rbuf(struct sock *sk, int copied);
378int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
379void tcp_twsk_destructor(struct sock *sk);
380ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
381 struct pipe_inode_info *pipe, size_t len,
382 unsigned int flags);
383
384static inline void tcp_dec_quickack_mode(struct sock *sk,
385 const unsigned int pkts)
386{
387 struct inet_connection_sock *icsk = inet_csk(sk);
388
389 if (icsk->icsk_ack.quick) {
390 if (pkts >= icsk->icsk_ack.quick) {
391 icsk->icsk_ack.quick = 0;
392 /* Leaving quickack mode we deflate ATO. */
393 icsk->icsk_ack.ato = TCP_ATO_MIN;
394 } else
395 icsk->icsk_ack.quick -= pkts;
396 }
397}
398
399#define TCP_ECN_OK 1
400#define TCP_ECN_QUEUE_CWR 2
401#define TCP_ECN_DEMAND_CWR 4
402#define TCP_ECN_SEEN 8
403
404enum tcp_tw_status {
405 TCP_TW_SUCCESS = 0,
406 TCP_TW_RST = 1,
407 TCP_TW_ACK = 2,
408 TCP_TW_SYN = 3
409};
410
411
412enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
413 struct sk_buff *skb,
414 const struct tcphdr *th);
415struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
416 struct request_sock *req, struct request_sock **prev,
417 bool fastopen);
418int tcp_child_process(struct sock *parent, struct sock *child,
419 struct sk_buff *skb);
420void tcp_enter_loss(struct sock *sk, int how);
421void tcp_clear_retrans(struct tcp_sock *tp);
422void tcp_update_metrics(struct sock *sk);
423void tcp_init_metrics(struct sock *sk);
424void tcp_metrics_init(void);
425bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
426 bool paws_check);
427bool tcp_remember_stamp(struct sock *sk);
428bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
429void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
430void tcp_disable_fack(struct tcp_sock *tp);
431void tcp_close(struct sock *sk, long timeout);
432void tcp_init_sock(struct sock *sk);
433unsigned int tcp_poll(struct file *file, struct socket *sock,
434 struct poll_table_struct *wait);
435int tcp_getsockopt(struct sock *sk, int level, int optname,
436 char __user *optval, int __user *optlen);
437int tcp_setsockopt(struct sock *sk, int level, int optname,
438 char __user *optval, unsigned int optlen);
439int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
440 char __user *optval, int __user *optlen);
441int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
442 char __user *optval, unsigned int optlen);
443void tcp_set_keepalive(struct sock *sk, int val);
444void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
445int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
446 size_t len, int nonblock, int flags, int *addr_len);
447void tcp_parse_options(const struct sk_buff *skb,
448 struct tcp_options_received *opt_rx,
449 int estab, struct tcp_fastopen_cookie *foc);
450const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
451
452/*
453 * TCP v4 functions exported for the inet6 API
454 */
455
456void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
457int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
458struct sock *tcp_create_openreq_child(struct sock *sk,
459 struct request_sock *req,
460 struct sk_buff *skb);
461struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
462 struct request_sock *req,
463 struct dst_entry *dst);
464int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
465int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
466int tcp_connect(struct sock *sk);
467struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
468 struct request_sock *req,
469 struct tcp_fastopen_cookie *foc);
470int tcp_disconnect(struct sock *sk, int flags);
471
472void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
473int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
474void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
475
476/* From syncookies.c */
477int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
478 u32 cookie);
479struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
480 struct ip_options *opt);
481#ifdef CONFIG_SYN_COOKIES
482
483/* Syncookies use a monotonic timer which increments every 60 seconds.
484 * This counter is used both as a hash input and partially encoded into
485 * the cookie value. A cookie is only validated further if the delta
486 * between the current counter value and the encoded one is less than this,
487 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
488 * the counter advances immediately after a cookie is generated).
489 */
490#define MAX_SYNCOOKIE_AGE 2
491
492static inline u32 tcp_cookie_time(void)
493{
494 u64 val = get_jiffies_64();
495
496 do_div(val, 60 * HZ);
497 return val;
498}
499
500u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
501 u16 *mssp);
502__u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mss);
503#else
504static inline __u32 cookie_v4_init_sequence(struct sock *sk,
505 struct sk_buff *skb,
506 __u16 *mss)
507{
508 return 0;
509}
510#endif
511
512__u32 cookie_init_timestamp(struct request_sock *req);
513bool cookie_check_timestamp(struct tcp_options_received *opt, struct net *net,
514 bool *ecn_ok);
515
516/* From net/ipv6/syncookies.c */
517int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
518 u32 cookie);
519struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
520#ifdef CONFIG_SYN_COOKIES
521u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
522 const struct tcphdr *th, u16 *mssp);
523__u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
524 __u16 *mss);
525#else
526static inline __u32 cookie_v6_init_sequence(struct sock *sk,
527 struct sk_buff *skb,
528 __u16 *mss)
529{
530 return 0;
531}
532#endif
533/* tcp_output.c */
534
535void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
536 int nonagle);
537bool tcp_may_send_now(struct sock *sk);
538int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
539int tcp_retransmit_skb(struct sock *, struct sk_buff *);
540void tcp_retransmit_timer(struct sock *sk);
541void tcp_xmit_retransmit_queue(struct sock *);
542void tcp_simple_retransmit(struct sock *);
543int tcp_trim_head(struct sock *, struct sk_buff *, u32);
544int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
545
546void tcp_send_probe0(struct sock *);
547void tcp_send_partial(struct sock *);
548int tcp_write_wakeup(struct sock *);
549void tcp_send_fin(struct sock *sk);
550void tcp_send_active_reset(struct sock *sk, gfp_t priority);
551int tcp_send_synack(struct sock *);
552bool tcp_syn_flood_action(struct sock *sk, const struct sk_buff *skb,
553 const char *proto);
554void tcp_push_one(struct sock *, unsigned int mss_now);
555void tcp_send_ack(struct sock *sk);
556void tcp_send_delayed_ack(struct sock *sk);
557void tcp_send_loss_probe(struct sock *sk);
558bool tcp_schedule_loss_probe(struct sock *sk);
559
560/* tcp_input.c */
561void tcp_cwnd_application_limited(struct sock *sk);
562void tcp_resume_early_retransmit(struct sock *sk);
563void tcp_rearm_rto(struct sock *sk);
564void tcp_reset(struct sock *sk);
565
566/* tcp_timer.c */
567void tcp_init_xmit_timers(struct sock *);
568static inline void tcp_clear_xmit_timers(struct sock *sk)
569{
570 inet_csk_clear_xmit_timers(sk);
571}
572
573unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
574unsigned int tcp_current_mss(struct sock *sk);
575
576/* Bound MSS / TSO packet size with the half of the window */
577static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
578{
579 int cutoff;
580
581 /* When peer uses tiny windows, there is no use in packetizing
582 * to sub-MSS pieces for the sake of SWS or making sure there
583 * are enough packets in the pipe for fast recovery.
584 *
585 * On the other hand, for extremely large MSS devices, handling
586 * smaller than MSS windows in this way does make sense.
587 */
588 if (tp->max_window >= 512)
589 cutoff = (tp->max_window >> 1);
590 else
591 cutoff = tp->max_window;
592
593 if (cutoff && pktsize > cutoff)
594 return max_t(int, cutoff, 68U - tp->tcp_header_len);
595 else
596 return pktsize;
597}
598
599/* tcp.c */
600void tcp_get_info(const struct sock *, struct tcp_info *);
601
602/* Read 'sendfile()'-style from a TCP socket */
603typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
604 unsigned int, size_t);
605int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
606 sk_read_actor_t recv_actor);
607
608void tcp_initialize_rcv_mss(struct sock *sk);
609
610int tcp_mtu_to_mss(struct sock *sk, int pmtu);
611int tcp_mss_to_mtu(struct sock *sk, int mss);
612void tcp_mtup_init(struct sock *sk);
613void tcp_init_buffer_space(struct sock *sk);
614
615static inline void tcp_bound_rto(const struct sock *sk)
616{
617 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
618 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
619}
620
621static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
622{
623 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
624}
625
626static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
627{
628 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
629 ntohl(TCP_FLAG_ACK) |
630 snd_wnd);
631}
632
633static inline void tcp_fast_path_on(struct tcp_sock *tp)
634{
635 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
636}
637
638static inline void tcp_fast_path_check(struct sock *sk)
639{
640 struct tcp_sock *tp = tcp_sk(sk);
641
642 if (skb_queue_empty(&tp->out_of_order_queue) &&
643 tp->rcv_wnd &&
644 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
645 !tp->urg_data)
646 tcp_fast_path_on(tp);
647}
648
649/* Compute the actual rto_min value */
650static inline u32 tcp_rto_min(struct sock *sk)
651{
652 const struct dst_entry *dst = __sk_dst_get(sk);
653 u32 rto_min = TCP_RTO_MIN;
654
655 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
656 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
657 return rto_min;
658}
659
660static inline u32 tcp_rto_min_us(struct sock *sk)
661{
662 return jiffies_to_usecs(tcp_rto_min(sk));
663}
664
665/* Compute the actual receive window we are currently advertising.
666 * Rcv_nxt can be after the window if our peer push more data
667 * than the offered window.
668 */
669static inline u32 tcp_receive_window(const struct tcp_sock *tp)
670{
671 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
672
673 if (win < 0)
674 win = 0;
675 return (u32) win;
676}
677
678/* Choose a new window, without checks for shrinking, and without
679 * scaling applied to the result. The caller does these things
680 * if necessary. This is a "raw" window selection.
681 */
682u32 __tcp_select_window(struct sock *sk);
683
684void tcp_send_window_probe(struct sock *sk);
685
686/* TCP timestamps are only 32-bits, this causes a slight
687 * complication on 64-bit systems since we store a snapshot
688 * of jiffies in the buffer control blocks below. We decided
689 * to use only the low 32-bits of jiffies and hide the ugly
690 * casts with the following macro.
691 */
692#define tcp_time_stamp ((__u32)(jiffies))
693
694#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
695
696#define TCPHDR_FIN 0x01
697#define TCPHDR_SYN 0x02
698#define TCPHDR_RST 0x04
699#define TCPHDR_PSH 0x08
700#define TCPHDR_ACK 0x10
701#define TCPHDR_URG 0x20
702#define TCPHDR_ECE 0x40
703#define TCPHDR_CWR 0x80
704
705/* This is what the send packet queuing engine uses to pass
706 * TCP per-packet control information to the transmission code.
707 * We also store the host-order sequence numbers in here too.
708 * This is 44 bytes if IPV6 is enabled.
709 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
710 */
711struct tcp_skb_cb {
712 union {
713 struct inet_skb_parm h4;
714#if IS_ENABLED(CONFIG_IPV6)
715 struct inet6_skb_parm h6;
716#endif
717 } header; /* For incoming frames */
718 __u32 seq; /* Starting sequence number */
719 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
720 __u32 when; /* used to compute rtt's */
721 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
722
723 __u8 sacked; /* State flags for SACK/FACK. */
724#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
725#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
726#define TCPCB_LOST 0x04 /* SKB is lost */
727#define TCPCB_TAGBITS 0x07 /* All tag bits */
728#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
729#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
730
731 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
732 /* 1 byte hole */
733 __u32 ack_seq; /* Sequence number ACK'd */
734};
735
736#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
737
738/* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
739 *
740 * If we receive a SYN packet with these bits set, it means a network is
741 * playing bad games with TOS bits. In order to avoid possible false congestion
742 * notifications, we disable TCP ECN negociation.
743 */
744static inline void
745TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb,
746 struct net *net)
747{
748 const struct tcphdr *th = tcp_hdr(skb);
749
750 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr &&
751 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
752 inet_rsk(req)->ecn_ok = 1;
753}
754
755/* Due to TSO, an SKB can be composed of multiple actual
756 * packets. To keep these tracked properly, we use this.
757 */
758static inline int tcp_skb_pcount(const struct sk_buff *skb)
759{
760 return skb_shinfo(skb)->gso_segs;
761}
762
763/* This is valid iff tcp_skb_pcount() > 1. */
764static inline int tcp_skb_mss(const struct sk_buff *skb)
765{
766 return skb_shinfo(skb)->gso_size;
767}
768
769/* Events passed to congestion control interface */
770enum tcp_ca_event {
771 CA_EVENT_TX_START, /* first transmit when no packets in flight */
772 CA_EVENT_CWND_RESTART, /* congestion window restart */
773 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
774 CA_EVENT_LOSS, /* loss timeout */
775 CA_EVENT_FAST_ACK, /* in sequence ack */
776 CA_EVENT_SLOW_ACK, /* other ack */
777};
778
779/*
780 * Interface for adding new TCP congestion control handlers
781 */
782#define TCP_CA_NAME_MAX 16
783#define TCP_CA_MAX 128
784#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
785
786#define TCP_CONG_NON_RESTRICTED 0x1
787
788struct tcp_congestion_ops {
789 struct list_head list;
790 unsigned long flags;
791
792 /* initialize private data (optional) */
793 void (*init)(struct sock *sk);
794 /* cleanup private data (optional) */
795 void (*release)(struct sock *sk);
796
797 /* return slow start threshold (required) */
798 u32 (*ssthresh)(struct sock *sk);
799 /* do new cwnd calculation (required) */
800 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked, u32 in_flight);
801 /* call before changing ca_state (optional) */
802 void (*set_state)(struct sock *sk, u8 new_state);
803 /* call when cwnd event occurs (optional) */
804 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
805 /* new value of cwnd after loss (optional) */
806 u32 (*undo_cwnd)(struct sock *sk);
807 /* hook for packet ack accounting (optional) */
808 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
809 /* get info for inet_diag (optional) */
810 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
811
812 char name[TCP_CA_NAME_MAX];
813 struct module *owner;
814};
815
816int tcp_register_congestion_control(struct tcp_congestion_ops *type);
817void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
818
819void tcp_init_congestion_control(struct sock *sk);
820void tcp_cleanup_congestion_control(struct sock *sk);
821int tcp_set_default_congestion_control(const char *name);
822void tcp_get_default_congestion_control(char *name);
823void tcp_get_available_congestion_control(char *buf, size_t len);
824void tcp_get_allowed_congestion_control(char *buf, size_t len);
825int tcp_set_allowed_congestion_control(char *allowed);
826int tcp_set_congestion_control(struct sock *sk, const char *name);
827int tcp_slow_start(struct tcp_sock *tp, u32 acked);
828void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
829
830extern struct tcp_congestion_ops tcp_init_congestion_ops;
831u32 tcp_reno_ssthresh(struct sock *sk);
832void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked, u32 in_flight);
833extern struct tcp_congestion_ops tcp_reno;
834
835static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
836{
837 struct inet_connection_sock *icsk = inet_csk(sk);
838
839 if (icsk->icsk_ca_ops->set_state)
840 icsk->icsk_ca_ops->set_state(sk, ca_state);
841 icsk->icsk_ca_state = ca_state;
842}
843
844static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
845{
846 const struct inet_connection_sock *icsk = inet_csk(sk);
847
848 if (icsk->icsk_ca_ops->cwnd_event)
849 icsk->icsk_ca_ops->cwnd_event(sk, event);
850}
851
852/* These functions determine how the current flow behaves in respect of SACK
853 * handling. SACK is negotiated with the peer, and therefore it can vary
854 * between different flows.
855 *
856 * tcp_is_sack - SACK enabled
857 * tcp_is_reno - No SACK
858 * tcp_is_fack - FACK enabled, implies SACK enabled
859 */
860static inline int tcp_is_sack(const struct tcp_sock *tp)
861{
862 return tp->rx_opt.sack_ok;
863}
864
865static inline bool tcp_is_reno(const struct tcp_sock *tp)
866{
867 return !tcp_is_sack(tp);
868}
869
870static inline bool tcp_is_fack(const struct tcp_sock *tp)
871{
872 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
873}
874
875static inline void tcp_enable_fack(struct tcp_sock *tp)
876{
877 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
878}
879
880/* TCP early-retransmit (ER) is similar to but more conservative than
881 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
882 */
883static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
884{
885 tp->do_early_retrans = sysctl_tcp_early_retrans &&
886 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
887 sysctl_tcp_reordering == 3;
888}
889
890static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
891{
892 tp->do_early_retrans = 0;
893}
894
895static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
896{
897 return tp->sacked_out + tp->lost_out;
898}
899
900/* This determines how many packets are "in the network" to the best
901 * of our knowledge. In many cases it is conservative, but where
902 * detailed information is available from the receiver (via SACK
903 * blocks etc.) we can make more aggressive calculations.
904 *
905 * Use this for decisions involving congestion control, use just
906 * tp->packets_out to determine if the send queue is empty or not.
907 *
908 * Read this equation as:
909 *
910 * "Packets sent once on transmission queue" MINUS
911 * "Packets left network, but not honestly ACKed yet" PLUS
912 * "Packets fast retransmitted"
913 */
914static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
915{
916 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
917}
918
919#define TCP_INFINITE_SSTHRESH 0x7fffffff
920
921static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
922{
923 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
924}
925
926static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
927{
928 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
929 (1 << inet_csk(sk)->icsk_ca_state);
930}
931
932/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
933 * The exception is cwnd reduction phase, when cwnd is decreasing towards
934 * ssthresh.
935 */
936static inline __u32 tcp_current_ssthresh(const struct sock *sk)
937{
938 const struct tcp_sock *tp = tcp_sk(sk);
939
940 if (tcp_in_cwnd_reduction(sk))
941 return tp->snd_ssthresh;
942 else
943 return max(tp->snd_ssthresh,
944 ((tp->snd_cwnd >> 1) +
945 (tp->snd_cwnd >> 2)));
946}
947
948/* Use define here intentionally to get WARN_ON location shown at the caller */
949#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
950
951void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
952__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
953
954/* The maximum number of MSS of available cwnd for which TSO defers
955 * sending if not using sysctl_tcp_tso_win_divisor.
956 */
957static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
958{
959 return 3;
960}
961
962/* Slow start with delack produces 3 packets of burst, so that
963 * it is safe "de facto". This will be the default - same as
964 * the default reordering threshold - but if reordering increases,
965 * we must be able to allow cwnd to burst at least this much in order
966 * to not pull it back when holes are filled.
967 */
968static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
969{
970 return tp->reordering;
971}
972
973/* Returns end sequence number of the receiver's advertised window */
974static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
975{
976 return tp->snd_una + tp->snd_wnd;
977}
978bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
979
980static inline void tcp_check_probe_timer(struct sock *sk)
981{
982 const struct tcp_sock *tp = tcp_sk(sk);
983 const struct inet_connection_sock *icsk = inet_csk(sk);
984
985 if (!tp->packets_out && !icsk->icsk_pending)
986 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
987 icsk->icsk_rto, TCP_RTO_MAX);
988}
989
990static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
991{
992 tp->snd_wl1 = seq;
993}
994
995static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
996{
997 tp->snd_wl1 = seq;
998}
999
1000/*
1001 * Calculate(/check) TCP checksum
1002 */
1003static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1004 __be32 daddr, __wsum base)
1005{
1006 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1007}
1008
1009static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1010{
1011 return __skb_checksum_complete(skb);
1012}
1013
1014static inline bool tcp_checksum_complete(struct sk_buff *skb)
1015{
1016 return !skb_csum_unnecessary(skb) &&
1017 __tcp_checksum_complete(skb);
1018}
1019
1020/* Prequeue for VJ style copy to user, combined with checksumming. */
1021
1022static inline void tcp_prequeue_init(struct tcp_sock *tp)
1023{
1024 tp->ucopy.task = NULL;
1025 tp->ucopy.len = 0;
1026 tp->ucopy.memory = 0;
1027 skb_queue_head_init(&tp->ucopy.prequeue);
1028#ifdef CONFIG_NET_DMA
1029 tp->ucopy.dma_chan = NULL;
1030 tp->ucopy.wakeup = 0;
1031 tp->ucopy.pinned_list = NULL;
1032 tp->ucopy.dma_cookie = 0;
1033#endif
1034}
1035
1036bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1037
1038#undef STATE_TRACE
1039
1040#ifdef STATE_TRACE
1041static const char *statename[]={
1042 "Unused","Established","Syn Sent","Syn Recv",
1043 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1044 "Close Wait","Last ACK","Listen","Closing"
1045};
1046#endif
1047void tcp_set_state(struct sock *sk, int state);
1048
1049void tcp_done(struct sock *sk);
1050
1051static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1052{
1053 rx_opt->dsack = 0;
1054 rx_opt->num_sacks = 0;
1055}
1056
1057u32 tcp_default_init_rwnd(u32 mss);
1058
1059/* Determine a window scaling and initial window to offer. */
1060void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1061 __u32 *window_clamp, int wscale_ok,
1062 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1063
1064static inline int tcp_win_from_space(int space)
1065{
1066 return sysctl_tcp_adv_win_scale<=0 ?
1067 (space>>(-sysctl_tcp_adv_win_scale)) :
1068 space - (space>>sysctl_tcp_adv_win_scale);
1069}
1070
1071/* Note: caller must be prepared to deal with negative returns */
1072static inline int tcp_space(const struct sock *sk)
1073{
1074 return tcp_win_from_space(sk->sk_rcvbuf -
1075 atomic_read(&sk->sk_rmem_alloc));
1076}
1077
1078static inline int tcp_full_space(const struct sock *sk)
1079{
1080 return tcp_win_from_space(sk->sk_rcvbuf);
1081}
1082
1083static inline void tcp_openreq_init(struct request_sock *req,
1084 struct tcp_options_received *rx_opt,
1085 struct sk_buff *skb)
1086{
1087 struct inet_request_sock *ireq = inet_rsk(req);
1088
1089 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1090 req->cookie_ts = 0;
1091 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1092 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
1093 tcp_rsk(req)->snt_synack = 0;
1094 req->mss = rx_opt->mss_clamp;
1095 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1096 ireq->tstamp_ok = rx_opt->tstamp_ok;
1097 ireq->sack_ok = rx_opt->sack_ok;
1098 ireq->snd_wscale = rx_opt->snd_wscale;
1099 ireq->wscale_ok = rx_opt->wscale_ok;
1100 ireq->acked = 0;
1101 ireq->ecn_ok = 0;
1102 ireq->ir_rmt_port = tcp_hdr(skb)->source;
1103 ireq->ir_num = ntohs(tcp_hdr(skb)->dest);
1104}
1105
1106void tcp_enter_memory_pressure(struct sock *sk);
1107
1108static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1109{
1110 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1111}
1112
1113static inline int keepalive_time_when(const struct tcp_sock *tp)
1114{
1115 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1116}
1117
1118static inline int keepalive_probes(const struct tcp_sock *tp)
1119{
1120 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1121}
1122
1123static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1124{
1125 const struct inet_connection_sock *icsk = &tp->inet_conn;
1126
1127 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1128 tcp_time_stamp - tp->rcv_tstamp);
1129}
1130
1131static inline int tcp_fin_time(const struct sock *sk)
1132{
1133 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1134 const int rto = inet_csk(sk)->icsk_rto;
1135
1136 if (fin_timeout < (rto << 2) - (rto >> 1))
1137 fin_timeout = (rto << 2) - (rto >> 1);
1138
1139 return fin_timeout;
1140}
1141
1142static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1143 int paws_win)
1144{
1145 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1146 return true;
1147 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1148 return true;
1149 /*
1150 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1151 * then following tcp messages have valid values. Ignore 0 value,
1152 * or else 'negative' tsval might forbid us to accept their packets.
1153 */
1154 if (!rx_opt->ts_recent)
1155 return true;
1156 return false;
1157}
1158
1159static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1160 int rst)
1161{
1162 if (tcp_paws_check(rx_opt, 0))
1163 return false;
1164
1165 /* RST segments are not recommended to carry timestamp,
1166 and, if they do, it is recommended to ignore PAWS because
1167 "their cleanup function should take precedence over timestamps."
1168 Certainly, it is mistake. It is necessary to understand the reasons
1169 of this constraint to relax it: if peer reboots, clock may go
1170 out-of-sync and half-open connections will not be reset.
1171 Actually, the problem would be not existing if all
1172 the implementations followed draft about maintaining clock
1173 via reboots. Linux-2.2 DOES NOT!
1174
1175 However, we can relax time bounds for RST segments to MSL.
1176 */
1177 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1178 return false;
1179 return true;
1180}
1181
1182static inline void tcp_mib_init(struct net *net)
1183{
1184 /* See RFC 2012 */
1185 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1186 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1187 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1188 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1189}
1190
1191/* from STCP */
1192static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1193{
1194 tp->lost_skb_hint = NULL;
1195}
1196
1197static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1198{
1199 tcp_clear_retrans_hints_partial(tp);
1200 tp->retransmit_skb_hint = NULL;
1201}
1202
1203/* MD5 Signature */
1204struct crypto_hash;
1205
1206union tcp_md5_addr {
1207 struct in_addr a4;
1208#if IS_ENABLED(CONFIG_IPV6)
1209 struct in6_addr a6;
1210#endif
1211};
1212
1213/* - key database */
1214struct tcp_md5sig_key {
1215 struct hlist_node node;
1216 u8 keylen;
1217 u8 family; /* AF_INET or AF_INET6 */
1218 union tcp_md5_addr addr;
1219 u8 key[TCP_MD5SIG_MAXKEYLEN];
1220 struct rcu_head rcu;
1221};
1222
1223/* - sock block */
1224struct tcp_md5sig_info {
1225 struct hlist_head head;
1226 struct rcu_head rcu;
1227};
1228
1229/* - pseudo header */
1230struct tcp4_pseudohdr {
1231 __be32 saddr;
1232 __be32 daddr;
1233 __u8 pad;
1234 __u8 protocol;
1235 __be16 len;
1236};
1237
1238struct tcp6_pseudohdr {
1239 struct in6_addr saddr;
1240 struct in6_addr daddr;
1241 __be32 len;
1242 __be32 protocol; /* including padding */
1243};
1244
1245union tcp_md5sum_block {
1246 struct tcp4_pseudohdr ip4;
1247#if IS_ENABLED(CONFIG_IPV6)
1248 struct tcp6_pseudohdr ip6;
1249#endif
1250};
1251
1252/* - pool: digest algorithm, hash description and scratch buffer */
1253struct tcp_md5sig_pool {
1254 struct hash_desc md5_desc;
1255 union tcp_md5sum_block md5_blk;
1256};
1257
1258/* - functions */
1259int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1260 const struct sock *sk, const struct request_sock *req,
1261 const struct sk_buff *skb);
1262int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1263 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1264int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1265 int family);
1266struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1267 struct sock *addr_sk);
1268
1269#ifdef CONFIG_TCP_MD5SIG
1270struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1271 const union tcp_md5_addr *addr,
1272 int family);
1273#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1274#else
1275static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1276 const union tcp_md5_addr *addr,
1277 int family)
1278{
1279 return NULL;
1280}
1281#define tcp_twsk_md5_key(twsk) NULL
1282#endif
1283
1284bool tcp_alloc_md5sig_pool(void);
1285
1286struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1287static inline void tcp_put_md5sig_pool(void)
1288{
1289 local_bh_enable();
1290}
1291
1292int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1293int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1294 unsigned int header_len);
1295int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1296 const struct tcp_md5sig_key *key);
1297
1298/* From tcp_fastopen.c */
1299void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1300 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1301 unsigned long *last_syn_loss);
1302void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1303 struct tcp_fastopen_cookie *cookie, bool syn_lost);
1304struct tcp_fastopen_request {
1305 /* Fast Open cookie. Size 0 means a cookie request */
1306 struct tcp_fastopen_cookie cookie;
1307 struct msghdr *data; /* data in MSG_FASTOPEN */
1308 size_t size;
1309 int copied; /* queued in tcp_connect() */
1310};
1311void tcp_free_fastopen_req(struct tcp_sock *tp);
1312
1313extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1314int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1315void tcp_fastopen_cookie_gen(__be32 src, __be32 dst,
1316 struct tcp_fastopen_cookie *foc);
1317void tcp_fastopen_init_key_once(bool publish);
1318#define TCP_FASTOPEN_KEY_LENGTH 16
1319
1320/* Fastopen key context */
1321struct tcp_fastopen_context {
1322 struct crypto_cipher *tfm;
1323 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1324 struct rcu_head rcu;
1325};
1326
1327/* write queue abstraction */
1328static inline void tcp_write_queue_purge(struct sock *sk)
1329{
1330 struct sk_buff *skb;
1331
1332 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1333 sk_wmem_free_skb(sk, skb);
1334 sk_mem_reclaim(sk);
1335 tcp_clear_all_retrans_hints(tcp_sk(sk));
1336}
1337
1338static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1339{
1340 return skb_peek(&sk->sk_write_queue);
1341}
1342
1343static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1344{
1345 return skb_peek_tail(&sk->sk_write_queue);
1346}
1347
1348static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1349 const struct sk_buff *skb)
1350{
1351 return skb_queue_next(&sk->sk_write_queue, skb);
1352}
1353
1354static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1355 const struct sk_buff *skb)
1356{
1357 return skb_queue_prev(&sk->sk_write_queue, skb);
1358}
1359
1360#define tcp_for_write_queue(skb, sk) \
1361 skb_queue_walk(&(sk)->sk_write_queue, skb)
1362
1363#define tcp_for_write_queue_from(skb, sk) \
1364 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1365
1366#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1367 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1368
1369static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1370{
1371 return sk->sk_send_head;
1372}
1373
1374static inline bool tcp_skb_is_last(const struct sock *sk,
1375 const struct sk_buff *skb)
1376{
1377 return skb_queue_is_last(&sk->sk_write_queue, skb);
1378}
1379
1380static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1381{
1382 if (tcp_skb_is_last(sk, skb))
1383 sk->sk_send_head = NULL;
1384 else
1385 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1386}
1387
1388static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1389{
1390 if (sk->sk_send_head == skb_unlinked)
1391 sk->sk_send_head = NULL;
1392}
1393
1394static inline void tcp_init_send_head(struct sock *sk)
1395{
1396 sk->sk_send_head = NULL;
1397}
1398
1399static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1400{
1401 __skb_queue_tail(&sk->sk_write_queue, skb);
1402}
1403
1404static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1405{
1406 __tcp_add_write_queue_tail(sk, skb);
1407
1408 /* Queue it, remembering where we must start sending. */
1409 if (sk->sk_send_head == NULL) {
1410 sk->sk_send_head = skb;
1411
1412 if (tcp_sk(sk)->highest_sack == NULL)
1413 tcp_sk(sk)->highest_sack = skb;
1414 }
1415}
1416
1417static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1418{
1419 __skb_queue_head(&sk->sk_write_queue, skb);
1420}
1421
1422/* Insert buff after skb on the write queue of sk. */
1423static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1424 struct sk_buff *buff,
1425 struct sock *sk)
1426{
1427 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1428}
1429
1430/* Insert new before skb on the write queue of sk. */
1431static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1432 struct sk_buff *skb,
1433 struct sock *sk)
1434{
1435 __skb_queue_before(&sk->sk_write_queue, skb, new);
1436
1437 if (sk->sk_send_head == skb)
1438 sk->sk_send_head = new;
1439}
1440
1441static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1442{
1443 __skb_unlink(skb, &sk->sk_write_queue);
1444}
1445
1446static inline bool tcp_write_queue_empty(struct sock *sk)
1447{
1448 return skb_queue_empty(&sk->sk_write_queue);
1449}
1450
1451static inline void tcp_push_pending_frames(struct sock *sk)
1452{
1453 if (tcp_send_head(sk)) {
1454 struct tcp_sock *tp = tcp_sk(sk);
1455
1456 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1457 }
1458}
1459
1460/* Start sequence of the skb just after the highest skb with SACKed
1461 * bit, valid only if sacked_out > 0 or when the caller has ensured
1462 * validity by itself.
1463 */
1464static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1465{
1466 if (!tp->sacked_out)
1467 return tp->snd_una;
1468
1469 if (tp->highest_sack == NULL)
1470 return tp->snd_nxt;
1471
1472 return TCP_SKB_CB(tp->highest_sack)->seq;
1473}
1474
1475static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1476{
1477 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1478 tcp_write_queue_next(sk, skb);
1479}
1480
1481static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1482{
1483 return tcp_sk(sk)->highest_sack;
1484}
1485
1486static inline void tcp_highest_sack_reset(struct sock *sk)
1487{
1488 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1489}
1490
1491/* Called when old skb is about to be deleted (to be combined with new skb) */
1492static inline void tcp_highest_sack_combine(struct sock *sk,
1493 struct sk_buff *old,
1494 struct sk_buff *new)
1495{
1496 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1497 tcp_sk(sk)->highest_sack = new;
1498}
1499
1500/* Determines whether this is a thin stream (which may suffer from
1501 * increased latency). Used to trigger latency-reducing mechanisms.
1502 */
1503static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1504{
1505 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1506}
1507
1508/* /proc */
1509enum tcp_seq_states {
1510 TCP_SEQ_STATE_LISTENING,
1511 TCP_SEQ_STATE_OPENREQ,
1512 TCP_SEQ_STATE_ESTABLISHED,
1513};
1514
1515int tcp_seq_open(struct inode *inode, struct file *file);
1516
1517struct tcp_seq_afinfo {
1518 char *name;
1519 sa_family_t family;
1520 const struct file_operations *seq_fops;
1521 struct seq_operations seq_ops;
1522};
1523
1524struct tcp_iter_state {
1525 struct seq_net_private p;
1526 sa_family_t family;
1527 enum tcp_seq_states state;
1528 struct sock *syn_wait_sk;
1529 int bucket, offset, sbucket, num;
1530 kuid_t uid;
1531 loff_t last_pos;
1532};
1533
1534int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1535void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1536
1537extern struct request_sock_ops tcp_request_sock_ops;
1538extern struct request_sock_ops tcp6_request_sock_ops;
1539
1540void tcp_v4_destroy_sock(struct sock *sk);
1541
1542struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1543 netdev_features_t features);
1544struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1545int tcp_gro_complete(struct sk_buff *skb);
1546
1547void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1548
1549static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1550{
1551 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1552}
1553
1554static inline bool tcp_stream_memory_free(const struct sock *sk)
1555{
1556 const struct tcp_sock *tp = tcp_sk(sk);
1557 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1558
1559 return notsent_bytes < tcp_notsent_lowat(tp);
1560}
1561
1562#ifdef CONFIG_PROC_FS
1563int tcp4_proc_init(void);
1564void tcp4_proc_exit(void);
1565#endif
1566
1567/* TCP af-specific functions */
1568struct tcp_sock_af_ops {
1569#ifdef CONFIG_TCP_MD5SIG
1570 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1571 struct sock *addr_sk);
1572 int (*calc_md5_hash) (char *location,
1573 struct tcp_md5sig_key *md5,
1574 const struct sock *sk,
1575 const struct request_sock *req,
1576 const struct sk_buff *skb);
1577 int (*md5_parse) (struct sock *sk,
1578 char __user *optval,
1579 int optlen);
1580#endif
1581};
1582
1583struct tcp_request_sock_ops {
1584#ifdef CONFIG_TCP_MD5SIG
1585 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1586 struct request_sock *req);
1587 int (*calc_md5_hash) (char *location,
1588 struct tcp_md5sig_key *md5,
1589 const struct sock *sk,
1590 const struct request_sock *req,
1591 const struct sk_buff *skb);
1592#endif
1593};
1594
1595int tcpv4_offload_init(void);
1596
1597void tcp_v4_init(void);
1598void tcp_init(void);
1599
1600#endif /* _TCP_H */
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/kref.h>
27#include <linux/ktime.h>
28#include <linux/indirect_call_wrapper.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#include <net/mptcp.h>
43
44#include <linux/seq_file.h>
45#include <linux/memcontrol.h>
46#include <linux/bpf-cgroup.h>
47#include <linux/siphash.h>
48
49extern struct inet_hashinfo tcp_hashinfo;
50
51DECLARE_PER_CPU(unsigned int, tcp_orphan_count);
52int tcp_orphan_count_sum(void);
53
54void tcp_time_wait(struct sock *sk, int state, int timeo);
55
56#define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER)
57#define MAX_TCP_OPTION_SPACE 40
58#define TCP_MIN_SND_MSS 48
59#define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
60
61/*
62 * Never offer a window over 32767 without using window scaling. Some
63 * poor stacks do signed 16bit maths!
64 */
65#define MAX_TCP_WINDOW 32767U
66
67/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68#define TCP_MIN_MSS 88U
69
70/* The initial MTU to use for probing */
71#define TCP_BASE_MSS 1024
72
73/* probing interval, default to 10 minutes as per RFC4821 */
74#define TCP_PROBE_INTERVAL 600
75
76/* Specify interval when tcp mtu probing will stop */
77#define TCP_PROBE_THRESHOLD 8
78
79/* After receiving this amount of duplicate ACKs fast retransmit starts. */
80#define TCP_FASTRETRANS_THRESH 3
81
82/* Maximal number of ACKs sent quickly to accelerate slow-start. */
83#define TCP_MAX_QUICKACKS 16U
84
85/* Maximal number of window scale according to RFC1323 */
86#define TCP_MAX_WSCALE 14U
87
88/* urg_data states */
89#define TCP_URG_VALID 0x0100
90#define TCP_URG_NOTYET 0x0200
91#define TCP_URG_READ 0x0400
92
93#define TCP_RETR1 3 /*
94 * This is how many retries it does before it
95 * tries to figure out if the gateway is
96 * down. Minimal RFC value is 3; it corresponds
97 * to ~3sec-8min depending on RTO.
98 */
99
100#define TCP_RETR2 15 /*
101 * This should take at least
102 * 90 minutes to time out.
103 * RFC1122 says that the limit is 100 sec.
104 * 15 is ~13-30min depending on RTO.
105 */
106
107#define TCP_SYN_RETRIES 6 /* This is how many retries are done
108 * when active opening a connection.
109 * RFC1122 says the minimum retry MUST
110 * be at least 180secs. Nevertheless
111 * this value is corresponding to
112 * 63secs of retransmission with the
113 * current initial RTO.
114 */
115
116#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
117 * when passive opening a connection.
118 * This is corresponding to 31secs of
119 * retransmission with the current
120 * initial RTO.
121 */
122
123#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
124 * state, about 60 seconds */
125#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
126 /* BSD style FIN_WAIT2 deadlock breaker.
127 * It used to be 3min, new value is 60sec,
128 * to combine FIN-WAIT-2 timeout with
129 * TIME-WAIT timer.
130 */
131#define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
132
133#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
134#if HZ >= 100
135#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
136#define TCP_ATO_MIN ((unsigned)(HZ/25))
137#else
138#define TCP_DELACK_MIN 4U
139#define TCP_ATO_MIN 4U
140#endif
141#define TCP_RTO_MAX ((unsigned)(120*HZ))
142#define TCP_RTO_MIN ((unsigned)(HZ/5))
143#define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
144#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
145#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
146 * used as a fallback RTO for the
147 * initial data transmission if no
148 * valid RTT sample has been acquired,
149 * most likely due to retrans in 3WHS.
150 */
151
152#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
153 * for local resources.
154 */
155#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
156#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
157#define TCP_KEEPALIVE_INTVL (75*HZ)
158
159#define MAX_TCP_KEEPIDLE 32767
160#define MAX_TCP_KEEPINTVL 32767
161#define MAX_TCP_KEEPCNT 127
162#define MAX_TCP_SYNCNT 127
163
164#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
165
166#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
167#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
168 * after this time. It should be equal
169 * (or greater than) TCP_TIMEWAIT_LEN
170 * to provide reliability equal to one
171 * provided by timewait state.
172 */
173#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
174 * timestamps. It must be less than
175 * minimal timewait lifetime.
176 */
177/*
178 * TCP option
179 */
180
181#define TCPOPT_NOP 1 /* Padding */
182#define TCPOPT_EOL 0 /* End of options */
183#define TCPOPT_MSS 2 /* Segment size negotiating */
184#define TCPOPT_WINDOW 3 /* Window scaling */
185#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
186#define TCPOPT_SACK 5 /* SACK Block */
187#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
188#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
189#define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */
190#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
191#define TCPOPT_EXP 254 /* Experimental */
192/* Magic number to be after the option value for sharing TCP
193 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
194 */
195#define TCPOPT_FASTOPEN_MAGIC 0xF989
196#define TCPOPT_SMC_MAGIC 0xE2D4C3D9
197
198/*
199 * TCP option lengths
200 */
201
202#define TCPOLEN_MSS 4
203#define TCPOLEN_WINDOW 3
204#define TCPOLEN_SACK_PERM 2
205#define TCPOLEN_TIMESTAMP 10
206#define TCPOLEN_MD5SIG 18
207#define TCPOLEN_FASTOPEN_BASE 2
208#define TCPOLEN_EXP_FASTOPEN_BASE 4
209#define TCPOLEN_EXP_SMC_BASE 6
210
211/* But this is what stacks really send out. */
212#define TCPOLEN_TSTAMP_ALIGNED 12
213#define TCPOLEN_WSCALE_ALIGNED 4
214#define TCPOLEN_SACKPERM_ALIGNED 4
215#define TCPOLEN_SACK_BASE 2
216#define TCPOLEN_SACK_BASE_ALIGNED 4
217#define TCPOLEN_SACK_PERBLOCK 8
218#define TCPOLEN_MD5SIG_ALIGNED 20
219#define TCPOLEN_MSS_ALIGNED 4
220#define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
221
222/* Flags in tp->nonagle */
223#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
224#define TCP_NAGLE_CORK 2 /* Socket is corked */
225#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
226
227/* TCP thin-stream limits */
228#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
229
230/* TCP initial congestion window as per rfc6928 */
231#define TCP_INIT_CWND 10
232
233/* Bit Flags for sysctl_tcp_fastopen */
234#define TFO_CLIENT_ENABLE 1
235#define TFO_SERVER_ENABLE 2
236#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
237
238/* Accept SYN data w/o any cookie option */
239#define TFO_SERVER_COOKIE_NOT_REQD 0x200
240
241/* Force enable TFO on all listeners, i.e., not requiring the
242 * TCP_FASTOPEN socket option.
243 */
244#define TFO_SERVER_WO_SOCKOPT1 0x400
245
246
247/* sysctl variables for tcp */
248extern int sysctl_tcp_max_orphans;
249extern long sysctl_tcp_mem[3];
250
251#define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
252#define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
253#define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
254
255extern atomic_long_t tcp_memory_allocated;
256DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc);
257
258extern struct percpu_counter tcp_sockets_allocated;
259extern unsigned long tcp_memory_pressure;
260
261/* optimized version of sk_under_memory_pressure() for TCP sockets */
262static inline bool tcp_under_memory_pressure(const struct sock *sk)
263{
264 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
265 mem_cgroup_under_socket_pressure(sk->sk_memcg))
266 return true;
267
268 return READ_ONCE(tcp_memory_pressure);
269}
270/*
271 * The next routines deal with comparing 32 bit unsigned ints
272 * and worry about wraparound (automatic with unsigned arithmetic).
273 */
274
275static inline bool before(__u32 seq1, __u32 seq2)
276{
277 return (__s32)(seq1-seq2) < 0;
278}
279#define after(seq2, seq1) before(seq1, seq2)
280
281/* is s2<=s1<=s3 ? */
282static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
283{
284 return seq3 - seq2 >= seq1 - seq2;
285}
286
287static inline bool tcp_out_of_memory(struct sock *sk)
288{
289 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
290 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
291 return true;
292 return false;
293}
294
295static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
296{
297 sk_wmem_queued_add(sk, -skb->truesize);
298 if (!skb_zcopy_pure(skb))
299 sk_mem_uncharge(sk, skb->truesize);
300 else
301 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb)));
302 __kfree_skb(skb);
303}
304
305void sk_forced_mem_schedule(struct sock *sk, int size);
306
307bool tcp_check_oom(struct sock *sk, int shift);
308
309
310extern struct proto tcp_prot;
311
312#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
313#define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
314#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
315#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
316
317void tcp_tasklet_init(void);
318
319int tcp_v4_err(struct sk_buff *skb, u32);
320
321void tcp_shutdown(struct sock *sk, int how);
322
323int tcp_v4_early_demux(struct sk_buff *skb);
324int tcp_v4_rcv(struct sk_buff *skb);
325
326void tcp_remove_empty_skb(struct sock *sk);
327int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
328int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
329int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
330int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied,
331 size_t size, struct ubuf_info *uarg);
332int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
333 int flags);
334int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
335 size_t size, int flags);
336ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
337 size_t size, int flags);
338int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
339void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
340 int size_goal);
341void tcp_release_cb(struct sock *sk);
342void tcp_wfree(struct sk_buff *skb);
343void tcp_write_timer_handler(struct sock *sk);
344void tcp_delack_timer_handler(struct sock *sk);
345int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
346int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
347void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
348void tcp_rcv_space_adjust(struct sock *sk);
349int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
350void tcp_twsk_destructor(struct sock *sk);
351void tcp_twsk_purge(struct list_head *net_exit_list, int family);
352ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
353 struct pipe_inode_info *pipe, size_t len,
354 unsigned int flags);
355struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
356 bool force_schedule);
357
358void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
359static inline void tcp_dec_quickack_mode(struct sock *sk,
360 const unsigned int pkts)
361{
362 struct inet_connection_sock *icsk = inet_csk(sk);
363
364 if (icsk->icsk_ack.quick) {
365 if (pkts >= icsk->icsk_ack.quick) {
366 icsk->icsk_ack.quick = 0;
367 /* Leaving quickack mode we deflate ATO. */
368 icsk->icsk_ack.ato = TCP_ATO_MIN;
369 } else
370 icsk->icsk_ack.quick -= pkts;
371 }
372}
373
374#define TCP_ECN_OK 1
375#define TCP_ECN_QUEUE_CWR 2
376#define TCP_ECN_DEMAND_CWR 4
377#define TCP_ECN_SEEN 8
378
379enum tcp_tw_status {
380 TCP_TW_SUCCESS = 0,
381 TCP_TW_RST = 1,
382 TCP_TW_ACK = 2,
383 TCP_TW_SYN = 3
384};
385
386
387enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
388 struct sk_buff *skb,
389 const struct tcphdr *th);
390struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
391 struct request_sock *req, bool fastopen,
392 bool *lost_race);
393int tcp_child_process(struct sock *parent, struct sock *child,
394 struct sk_buff *skb);
395void tcp_enter_loss(struct sock *sk);
396void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
397void tcp_clear_retrans(struct tcp_sock *tp);
398void tcp_update_metrics(struct sock *sk);
399void tcp_init_metrics(struct sock *sk);
400void tcp_metrics_init(void);
401bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
402void __tcp_close(struct sock *sk, long timeout);
403void tcp_close(struct sock *sk, long timeout);
404void tcp_init_sock(struct sock *sk);
405void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
406__poll_t tcp_poll(struct file *file, struct socket *sock,
407 struct poll_table_struct *wait);
408int do_tcp_getsockopt(struct sock *sk, int level,
409 int optname, sockptr_t optval, sockptr_t optlen);
410int tcp_getsockopt(struct sock *sk, int level, int optname,
411 char __user *optval, int __user *optlen);
412bool tcp_bpf_bypass_getsockopt(int level, int optname);
413int do_tcp_setsockopt(struct sock *sk, int level, int optname,
414 sockptr_t optval, unsigned int optlen);
415int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
416 unsigned int optlen);
417void tcp_set_keepalive(struct sock *sk, int val);
418void tcp_syn_ack_timeout(const struct request_sock *req);
419int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
420 int flags, int *addr_len);
421int tcp_set_rcvlowat(struct sock *sk, int val);
422int tcp_set_window_clamp(struct sock *sk, int val);
423void tcp_update_recv_tstamps(struct sk_buff *skb,
424 struct scm_timestamping_internal *tss);
425void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
426 struct scm_timestamping_internal *tss);
427void tcp_data_ready(struct sock *sk);
428#ifdef CONFIG_MMU
429int tcp_mmap(struct file *file, struct socket *sock,
430 struct vm_area_struct *vma);
431#endif
432void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
433 struct tcp_options_received *opt_rx,
434 int estab, struct tcp_fastopen_cookie *foc);
435const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
436
437/*
438 * BPF SKB-less helpers
439 */
440u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
441 struct tcphdr *th, u32 *cookie);
442u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
443 struct tcphdr *th, u32 *cookie);
444u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss);
445u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
446 const struct tcp_request_sock_ops *af_ops,
447 struct sock *sk, struct tcphdr *th);
448/*
449 * TCP v4 functions exported for the inet6 API
450 */
451
452void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
453void tcp_v4_mtu_reduced(struct sock *sk);
454void tcp_req_err(struct sock *sk, u32 seq, bool abort);
455void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
456int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
457struct sock *tcp_create_openreq_child(const struct sock *sk,
458 struct request_sock *req,
459 struct sk_buff *skb);
460void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
461struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
462 struct request_sock *req,
463 struct dst_entry *dst,
464 struct request_sock *req_unhash,
465 bool *own_req);
466int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
467int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
468int tcp_connect(struct sock *sk);
469enum tcp_synack_type {
470 TCP_SYNACK_NORMAL,
471 TCP_SYNACK_FASTOPEN,
472 TCP_SYNACK_COOKIE,
473};
474struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
475 struct request_sock *req,
476 struct tcp_fastopen_cookie *foc,
477 enum tcp_synack_type synack_type,
478 struct sk_buff *syn_skb);
479int tcp_disconnect(struct sock *sk, int flags);
480
481void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
482int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
483void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
484
485/* From syncookies.c */
486struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
487 struct request_sock *req,
488 struct dst_entry *dst, u32 tsoff);
489int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
490 u32 cookie);
491struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
492struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
493 const struct tcp_request_sock_ops *af_ops,
494 struct sock *sk, struct sk_buff *skb);
495#ifdef CONFIG_SYN_COOKIES
496
497/* Syncookies use a monotonic timer which increments every 60 seconds.
498 * This counter is used both as a hash input and partially encoded into
499 * the cookie value. A cookie is only validated further if the delta
500 * between the current counter value and the encoded one is less than this,
501 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
502 * the counter advances immediately after a cookie is generated).
503 */
504#define MAX_SYNCOOKIE_AGE 2
505#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
506#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
507
508/* syncookies: remember time of last synqueue overflow
509 * But do not dirty this field too often (once per second is enough)
510 * It is racy as we do not hold a lock, but race is very minor.
511 */
512static inline void tcp_synq_overflow(const struct sock *sk)
513{
514 unsigned int last_overflow;
515 unsigned int now = jiffies;
516
517 if (sk->sk_reuseport) {
518 struct sock_reuseport *reuse;
519
520 reuse = rcu_dereference(sk->sk_reuseport_cb);
521 if (likely(reuse)) {
522 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
523 if (!time_between32(now, last_overflow,
524 last_overflow + HZ))
525 WRITE_ONCE(reuse->synq_overflow_ts, now);
526 return;
527 }
528 }
529
530 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
531 if (!time_between32(now, last_overflow, last_overflow + HZ))
532 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
533}
534
535/* syncookies: no recent synqueue overflow on this listening socket? */
536static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
537{
538 unsigned int last_overflow;
539 unsigned int now = jiffies;
540
541 if (sk->sk_reuseport) {
542 struct sock_reuseport *reuse;
543
544 reuse = rcu_dereference(sk->sk_reuseport_cb);
545 if (likely(reuse)) {
546 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
547 return !time_between32(now, last_overflow - HZ,
548 last_overflow +
549 TCP_SYNCOOKIE_VALID);
550 }
551 }
552
553 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
554
555 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
556 * then we're under synflood. However, we have to use
557 * 'last_overflow - HZ' as lower bound. That's because a concurrent
558 * tcp_synq_overflow() could update .ts_recent_stamp after we read
559 * jiffies but before we store .ts_recent_stamp into last_overflow,
560 * which could lead to rejecting a valid syncookie.
561 */
562 return !time_between32(now, last_overflow - HZ,
563 last_overflow + TCP_SYNCOOKIE_VALID);
564}
565
566static inline u32 tcp_cookie_time(void)
567{
568 u64 val = get_jiffies_64();
569
570 do_div(val, TCP_SYNCOOKIE_PERIOD);
571 return val;
572}
573
574u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
575 u16 *mssp);
576__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
577u64 cookie_init_timestamp(struct request_sock *req, u64 now);
578bool cookie_timestamp_decode(const struct net *net,
579 struct tcp_options_received *opt);
580bool cookie_ecn_ok(const struct tcp_options_received *opt,
581 const struct net *net, const struct dst_entry *dst);
582
583/* From net/ipv6/syncookies.c */
584int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
585 u32 cookie);
586struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
587
588u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
589 const struct tcphdr *th, u16 *mssp);
590__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
591#endif
592/* tcp_output.c */
593
594void tcp_skb_entail(struct sock *sk, struct sk_buff *skb);
595void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb);
596void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
597 int nonagle);
598int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
599int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
600void tcp_retransmit_timer(struct sock *sk);
601void tcp_xmit_retransmit_queue(struct sock *);
602void tcp_simple_retransmit(struct sock *);
603void tcp_enter_recovery(struct sock *sk, bool ece_ack);
604int tcp_trim_head(struct sock *, struct sk_buff *, u32);
605enum tcp_queue {
606 TCP_FRAG_IN_WRITE_QUEUE,
607 TCP_FRAG_IN_RTX_QUEUE,
608};
609int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
610 struct sk_buff *skb, u32 len,
611 unsigned int mss_now, gfp_t gfp);
612
613void tcp_send_probe0(struct sock *);
614void tcp_send_partial(struct sock *);
615int tcp_write_wakeup(struct sock *, int mib);
616void tcp_send_fin(struct sock *sk);
617void tcp_send_active_reset(struct sock *sk, gfp_t priority);
618int tcp_send_synack(struct sock *);
619void tcp_push_one(struct sock *, unsigned int mss_now);
620void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
621void tcp_send_ack(struct sock *sk);
622void tcp_send_delayed_ack(struct sock *sk);
623void tcp_send_loss_probe(struct sock *sk);
624bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
625void tcp_skb_collapse_tstamp(struct sk_buff *skb,
626 const struct sk_buff *next_skb);
627
628/* tcp_input.c */
629void tcp_rearm_rto(struct sock *sk);
630void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
631void tcp_reset(struct sock *sk, struct sk_buff *skb);
632void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
633void tcp_fin(struct sock *sk);
634void tcp_check_space(struct sock *sk);
635
636/* tcp_timer.c */
637void tcp_init_xmit_timers(struct sock *);
638static inline void tcp_clear_xmit_timers(struct sock *sk)
639{
640 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
641 __sock_put(sk);
642
643 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
644 __sock_put(sk);
645
646 inet_csk_clear_xmit_timers(sk);
647}
648
649unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
650unsigned int tcp_current_mss(struct sock *sk);
651u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
652
653/* Bound MSS / TSO packet size with the half of the window */
654static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
655{
656 int cutoff;
657
658 /* When peer uses tiny windows, there is no use in packetizing
659 * to sub-MSS pieces for the sake of SWS or making sure there
660 * are enough packets in the pipe for fast recovery.
661 *
662 * On the other hand, for extremely large MSS devices, handling
663 * smaller than MSS windows in this way does make sense.
664 */
665 if (tp->max_window > TCP_MSS_DEFAULT)
666 cutoff = (tp->max_window >> 1);
667 else
668 cutoff = tp->max_window;
669
670 if (cutoff && pktsize > cutoff)
671 return max_t(int, cutoff, 68U - tp->tcp_header_len);
672 else
673 return pktsize;
674}
675
676/* tcp.c */
677void tcp_get_info(struct sock *, struct tcp_info *);
678
679/* Read 'sendfile()'-style from a TCP socket */
680int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
681 sk_read_actor_t recv_actor);
682int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
683struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off);
684void tcp_read_done(struct sock *sk, size_t len);
685
686void tcp_initialize_rcv_mss(struct sock *sk);
687
688int tcp_mtu_to_mss(struct sock *sk, int pmtu);
689int tcp_mss_to_mtu(struct sock *sk, int mss);
690void tcp_mtup_init(struct sock *sk);
691
692static inline void tcp_bound_rto(const struct sock *sk)
693{
694 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
695 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
696}
697
698static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
699{
700 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
701}
702
703static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
704{
705 /* mptcp hooks are only on the slow path */
706 if (sk_is_mptcp((struct sock *)tp))
707 return;
708
709 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
710 ntohl(TCP_FLAG_ACK) |
711 snd_wnd);
712}
713
714static inline void tcp_fast_path_on(struct tcp_sock *tp)
715{
716 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
717}
718
719static inline void tcp_fast_path_check(struct sock *sk)
720{
721 struct tcp_sock *tp = tcp_sk(sk);
722
723 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
724 tp->rcv_wnd &&
725 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
726 !tp->urg_data)
727 tcp_fast_path_on(tp);
728}
729
730/* Compute the actual rto_min value */
731static inline u32 tcp_rto_min(struct sock *sk)
732{
733 const struct dst_entry *dst = __sk_dst_get(sk);
734 u32 rto_min = inet_csk(sk)->icsk_rto_min;
735
736 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
737 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
738 return rto_min;
739}
740
741static inline u32 tcp_rto_min_us(struct sock *sk)
742{
743 return jiffies_to_usecs(tcp_rto_min(sk));
744}
745
746static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
747{
748 return dst_metric_locked(dst, RTAX_CC_ALGO);
749}
750
751/* Minimum RTT in usec. ~0 means not available. */
752static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
753{
754 return minmax_get(&tp->rtt_min);
755}
756
757/* Compute the actual receive window we are currently advertising.
758 * Rcv_nxt can be after the window if our peer push more data
759 * than the offered window.
760 */
761static inline u32 tcp_receive_window(const struct tcp_sock *tp)
762{
763 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
764
765 if (win < 0)
766 win = 0;
767 return (u32) win;
768}
769
770/* Choose a new window, without checks for shrinking, and without
771 * scaling applied to the result. The caller does these things
772 * if necessary. This is a "raw" window selection.
773 */
774u32 __tcp_select_window(struct sock *sk);
775
776void tcp_send_window_probe(struct sock *sk);
777
778/* TCP uses 32bit jiffies to save some space.
779 * Note that this is different from tcp_time_stamp, which
780 * historically has been the same until linux-4.13.
781 */
782#define tcp_jiffies32 ((u32)jiffies)
783
784/*
785 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
786 * It is no longer tied to jiffies, but to 1 ms clock.
787 * Note: double check if you want to use tcp_jiffies32 instead of this.
788 */
789#define TCP_TS_HZ 1000
790
791static inline u64 tcp_clock_ns(void)
792{
793 return ktime_get_ns();
794}
795
796static inline u64 tcp_clock_us(void)
797{
798 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
799}
800
801/* This should only be used in contexts where tp->tcp_mstamp is up to date */
802static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
803{
804 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
805}
806
807/* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
808static inline u32 tcp_ns_to_ts(u64 ns)
809{
810 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
811}
812
813/* Could use tcp_clock_us() / 1000, but this version uses a single divide */
814static inline u32 tcp_time_stamp_raw(void)
815{
816 return tcp_ns_to_ts(tcp_clock_ns());
817}
818
819void tcp_mstamp_refresh(struct tcp_sock *tp);
820
821static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
822{
823 return max_t(s64, t1 - t0, 0);
824}
825
826static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
827{
828 return tcp_ns_to_ts(skb->skb_mstamp_ns);
829}
830
831/* provide the departure time in us unit */
832static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
833{
834 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
835}
836
837
838#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
839
840#define TCPHDR_FIN 0x01
841#define TCPHDR_SYN 0x02
842#define TCPHDR_RST 0x04
843#define TCPHDR_PSH 0x08
844#define TCPHDR_ACK 0x10
845#define TCPHDR_URG 0x20
846#define TCPHDR_ECE 0x40
847#define TCPHDR_CWR 0x80
848
849#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
850
851/* This is what the send packet queuing engine uses to pass
852 * TCP per-packet control information to the transmission code.
853 * We also store the host-order sequence numbers in here too.
854 * This is 44 bytes if IPV6 is enabled.
855 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
856 */
857struct tcp_skb_cb {
858 __u32 seq; /* Starting sequence number */
859 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
860 union {
861 /* Note : tcp_tw_isn is used in input path only
862 * (isn chosen by tcp_timewait_state_process())
863 *
864 * tcp_gso_segs/size are used in write queue only,
865 * cf tcp_skb_pcount()/tcp_skb_mss()
866 */
867 __u32 tcp_tw_isn;
868 struct {
869 u16 tcp_gso_segs;
870 u16 tcp_gso_size;
871 };
872 };
873 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
874
875 __u8 sacked; /* State flags for SACK. */
876#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
877#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
878#define TCPCB_LOST 0x04 /* SKB is lost */
879#define TCPCB_TAGBITS 0x07 /* All tag bits */
880#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
881#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
882#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
883 TCPCB_REPAIRED)
884
885 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
886 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
887 eor:1, /* Is skb MSG_EOR marked? */
888 has_rxtstamp:1, /* SKB has a RX timestamp */
889 unused:5;
890 __u32 ack_seq; /* Sequence number ACK'd */
891 union {
892 struct {
893#define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1)
894 /* There is space for up to 24 bytes */
895 __u32 is_app_limited:1, /* cwnd not fully used? */
896 delivered_ce:20,
897 unused:11;
898 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
899 __u32 delivered;
900 /* start of send pipeline phase */
901 u64 first_tx_mstamp;
902 /* when we reached the "delivered" count */
903 u64 delivered_mstamp;
904 } tx; /* only used for outgoing skbs */
905 union {
906 struct inet_skb_parm h4;
907#if IS_ENABLED(CONFIG_IPV6)
908 struct inet6_skb_parm h6;
909#endif
910 } header; /* For incoming skbs */
911 };
912};
913
914#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
915
916extern const struct inet_connection_sock_af_ops ipv4_specific;
917
918#if IS_ENABLED(CONFIG_IPV6)
919/* This is the variant of inet6_iif() that must be used by TCP,
920 * as TCP moves IP6CB into a different location in skb->cb[]
921 */
922static inline int tcp_v6_iif(const struct sk_buff *skb)
923{
924 return TCP_SKB_CB(skb)->header.h6.iif;
925}
926
927static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
928{
929 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
930
931 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
932}
933
934/* TCP_SKB_CB reference means this can not be used from early demux */
935static inline int tcp_v6_sdif(const struct sk_buff *skb)
936{
937#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
938 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
939 return TCP_SKB_CB(skb)->header.h6.iif;
940#endif
941 return 0;
942}
943
944extern const struct inet_connection_sock_af_ops ipv6_specific;
945
946INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
947INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
948void tcp_v6_early_demux(struct sk_buff *skb);
949
950#endif
951
952/* TCP_SKB_CB reference means this can not be used from early demux */
953static inline int tcp_v4_sdif(struct sk_buff *skb)
954{
955#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
956 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
957 return TCP_SKB_CB(skb)->header.h4.iif;
958#endif
959 return 0;
960}
961
962/* Due to TSO, an SKB can be composed of multiple actual
963 * packets. To keep these tracked properly, we use this.
964 */
965static inline int tcp_skb_pcount(const struct sk_buff *skb)
966{
967 return TCP_SKB_CB(skb)->tcp_gso_segs;
968}
969
970static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
971{
972 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
973}
974
975static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
976{
977 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
978}
979
980/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
981static inline int tcp_skb_mss(const struct sk_buff *skb)
982{
983 return TCP_SKB_CB(skb)->tcp_gso_size;
984}
985
986static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
987{
988 return likely(!TCP_SKB_CB(skb)->eor);
989}
990
991static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
992 const struct sk_buff *from)
993{
994 return likely(tcp_skb_can_collapse_to(to) &&
995 mptcp_skb_can_collapse(to, from) &&
996 skb_pure_zcopy_same(to, from));
997}
998
999/* Events passed to congestion control interface */
1000enum tcp_ca_event {
1001 CA_EVENT_TX_START, /* first transmit when no packets in flight */
1002 CA_EVENT_CWND_RESTART, /* congestion window restart */
1003 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
1004 CA_EVENT_LOSS, /* loss timeout */
1005 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
1006 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
1007};
1008
1009/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1010enum tcp_ca_ack_event_flags {
1011 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
1012 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
1013 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
1014};
1015
1016/*
1017 * Interface for adding new TCP congestion control handlers
1018 */
1019#define TCP_CA_NAME_MAX 16
1020#define TCP_CA_MAX 128
1021#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
1022
1023#define TCP_CA_UNSPEC 0
1024
1025/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1026#define TCP_CONG_NON_RESTRICTED 0x1
1027/* Requires ECN/ECT set on all packets */
1028#define TCP_CONG_NEEDS_ECN 0x2
1029#define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1030
1031union tcp_cc_info;
1032
1033struct ack_sample {
1034 u32 pkts_acked;
1035 s32 rtt_us;
1036 u32 in_flight;
1037};
1038
1039/* A rate sample measures the number of (original/retransmitted) data
1040 * packets delivered "delivered" over an interval of time "interval_us".
1041 * The tcp_rate.c code fills in the rate sample, and congestion
1042 * control modules that define a cong_control function to run at the end
1043 * of ACK processing can optionally chose to consult this sample when
1044 * setting cwnd and pacing rate.
1045 * A sample is invalid if "delivered" or "interval_us" is negative.
1046 */
1047struct rate_sample {
1048 u64 prior_mstamp; /* starting timestamp for interval */
1049 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1050 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */
1051 s32 delivered; /* number of packets delivered over interval */
1052 s32 delivered_ce; /* number of packets delivered w/ CE marks*/
1053 long interval_us; /* time for tp->delivered to incr "delivered" */
1054 u32 snd_interval_us; /* snd interval for delivered packets */
1055 u32 rcv_interval_us; /* rcv interval for delivered packets */
1056 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1057 int losses; /* number of packets marked lost upon ACK */
1058 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1059 u32 prior_in_flight; /* in flight before this ACK */
1060 u32 last_end_seq; /* end_seq of most recently ACKed packet */
1061 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1062 bool is_retrans; /* is sample from retransmission? */
1063 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1064};
1065
1066struct tcp_congestion_ops {
1067/* fast path fields are put first to fill one cache line */
1068
1069 /* return slow start threshold (required) */
1070 u32 (*ssthresh)(struct sock *sk);
1071
1072 /* do new cwnd calculation (required) */
1073 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1074
1075 /* call before changing ca_state (optional) */
1076 void (*set_state)(struct sock *sk, u8 new_state);
1077
1078 /* call when cwnd event occurs (optional) */
1079 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1080
1081 /* call when ack arrives (optional) */
1082 void (*in_ack_event)(struct sock *sk, u32 flags);
1083
1084 /* hook for packet ack accounting (optional) */
1085 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1086
1087 /* override sysctl_tcp_min_tso_segs */
1088 u32 (*min_tso_segs)(struct sock *sk);
1089
1090 /* call when packets are delivered to update cwnd and pacing rate,
1091 * after all the ca_state processing. (optional)
1092 */
1093 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1094
1095
1096 /* new value of cwnd after loss (required) */
1097 u32 (*undo_cwnd)(struct sock *sk);
1098 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1099 u32 (*sndbuf_expand)(struct sock *sk);
1100
1101/* control/slow paths put last */
1102 /* get info for inet_diag (optional) */
1103 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1104 union tcp_cc_info *info);
1105
1106 char name[TCP_CA_NAME_MAX];
1107 struct module *owner;
1108 struct list_head list;
1109 u32 key;
1110 u32 flags;
1111
1112 /* initialize private data (optional) */
1113 void (*init)(struct sock *sk);
1114 /* cleanup private data (optional) */
1115 void (*release)(struct sock *sk);
1116} ____cacheline_aligned_in_smp;
1117
1118int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1119void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1120
1121void tcp_assign_congestion_control(struct sock *sk);
1122void tcp_init_congestion_control(struct sock *sk);
1123void tcp_cleanup_congestion_control(struct sock *sk);
1124int tcp_set_default_congestion_control(struct net *net, const char *name);
1125void tcp_get_default_congestion_control(struct net *net, char *name);
1126void tcp_get_available_congestion_control(char *buf, size_t len);
1127void tcp_get_allowed_congestion_control(char *buf, size_t len);
1128int tcp_set_allowed_congestion_control(char *allowed);
1129int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1130 bool cap_net_admin);
1131u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1132void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1133
1134u32 tcp_reno_ssthresh(struct sock *sk);
1135u32 tcp_reno_undo_cwnd(struct sock *sk);
1136void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1137extern struct tcp_congestion_ops tcp_reno;
1138
1139struct tcp_congestion_ops *tcp_ca_find(const char *name);
1140struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1141u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1142#ifdef CONFIG_INET
1143char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1144#else
1145static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1146{
1147 return NULL;
1148}
1149#endif
1150
1151static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1152{
1153 const struct inet_connection_sock *icsk = inet_csk(sk);
1154
1155 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1156}
1157
1158static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1159{
1160 const struct inet_connection_sock *icsk = inet_csk(sk);
1161
1162 if (icsk->icsk_ca_ops->cwnd_event)
1163 icsk->icsk_ca_ops->cwnd_event(sk, event);
1164}
1165
1166/* From tcp_cong.c */
1167void tcp_set_ca_state(struct sock *sk, const u8 ca_state);
1168
1169/* From tcp_rate.c */
1170void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1171void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1172 struct rate_sample *rs);
1173void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1174 bool is_sack_reneg, struct rate_sample *rs);
1175void tcp_rate_check_app_limited(struct sock *sk);
1176
1177static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
1178{
1179 return t1 > t2 || (t1 == t2 && after(seq1, seq2));
1180}
1181
1182/* These functions determine how the current flow behaves in respect of SACK
1183 * handling. SACK is negotiated with the peer, and therefore it can vary
1184 * between different flows.
1185 *
1186 * tcp_is_sack - SACK enabled
1187 * tcp_is_reno - No SACK
1188 */
1189static inline int tcp_is_sack(const struct tcp_sock *tp)
1190{
1191 return likely(tp->rx_opt.sack_ok);
1192}
1193
1194static inline bool tcp_is_reno(const struct tcp_sock *tp)
1195{
1196 return !tcp_is_sack(tp);
1197}
1198
1199static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1200{
1201 return tp->sacked_out + tp->lost_out;
1202}
1203
1204/* This determines how many packets are "in the network" to the best
1205 * of our knowledge. In many cases it is conservative, but where
1206 * detailed information is available from the receiver (via SACK
1207 * blocks etc.) we can make more aggressive calculations.
1208 *
1209 * Use this for decisions involving congestion control, use just
1210 * tp->packets_out to determine if the send queue is empty or not.
1211 *
1212 * Read this equation as:
1213 *
1214 * "Packets sent once on transmission queue" MINUS
1215 * "Packets left network, but not honestly ACKed yet" PLUS
1216 * "Packets fast retransmitted"
1217 */
1218static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1219{
1220 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1221}
1222
1223#define TCP_INFINITE_SSTHRESH 0x7fffffff
1224
1225static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp)
1226{
1227 return tp->snd_cwnd;
1228}
1229
1230static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val)
1231{
1232 WARN_ON_ONCE((int)val <= 0);
1233 tp->snd_cwnd = val;
1234}
1235
1236static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1237{
1238 return tcp_snd_cwnd(tp) < tp->snd_ssthresh;
1239}
1240
1241static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1242{
1243 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1244}
1245
1246static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1247{
1248 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1249 (1 << inet_csk(sk)->icsk_ca_state);
1250}
1251
1252/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1253 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1254 * ssthresh.
1255 */
1256static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1257{
1258 const struct tcp_sock *tp = tcp_sk(sk);
1259
1260 if (tcp_in_cwnd_reduction(sk))
1261 return tp->snd_ssthresh;
1262 else
1263 return max(tp->snd_ssthresh,
1264 ((tcp_snd_cwnd(tp) >> 1) +
1265 (tcp_snd_cwnd(tp) >> 2)));
1266}
1267
1268/* Use define here intentionally to get WARN_ON location shown at the caller */
1269#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1270
1271void tcp_enter_cwr(struct sock *sk);
1272__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1273
1274/* The maximum number of MSS of available cwnd for which TSO defers
1275 * sending if not using sysctl_tcp_tso_win_divisor.
1276 */
1277static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1278{
1279 return 3;
1280}
1281
1282/* Returns end sequence number of the receiver's advertised window */
1283static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1284{
1285 return tp->snd_una + tp->snd_wnd;
1286}
1287
1288/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1289 * flexible approach. The RFC suggests cwnd should not be raised unless
1290 * it was fully used previously. And that's exactly what we do in
1291 * congestion avoidance mode. But in slow start we allow cwnd to grow
1292 * as long as the application has used half the cwnd.
1293 * Example :
1294 * cwnd is 10 (IW10), but application sends 9 frames.
1295 * We allow cwnd to reach 18 when all frames are ACKed.
1296 * This check is safe because it's as aggressive as slow start which already
1297 * risks 100% overshoot. The advantage is that we discourage application to
1298 * either send more filler packets or data to artificially blow up the cwnd
1299 * usage, and allow application-limited process to probe bw more aggressively.
1300 */
1301static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1302{
1303 const struct tcp_sock *tp = tcp_sk(sk);
1304
1305 if (tp->is_cwnd_limited)
1306 return true;
1307
1308 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1309 if (tcp_in_slow_start(tp))
1310 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out;
1311
1312 return false;
1313}
1314
1315/* BBR congestion control needs pacing.
1316 * Same remark for SO_MAX_PACING_RATE.
1317 * sch_fq packet scheduler is efficiently handling pacing,
1318 * but is not always installed/used.
1319 * Return true if TCP stack should pace packets itself.
1320 */
1321static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1322{
1323 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1324}
1325
1326/* Estimates in how many jiffies next packet for this flow can be sent.
1327 * Scheduling a retransmit timer too early would be silly.
1328 */
1329static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1330{
1331 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1332
1333 return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1334}
1335
1336static inline void tcp_reset_xmit_timer(struct sock *sk,
1337 const int what,
1338 unsigned long when,
1339 const unsigned long max_when)
1340{
1341 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
1342 max_when);
1343}
1344
1345/* Something is really bad, we could not queue an additional packet,
1346 * because qdisc is full or receiver sent a 0 window, or we are paced.
1347 * We do not want to add fuel to the fire, or abort too early,
1348 * so make sure the timer we arm now is at least 200ms in the future,
1349 * regardless of current icsk_rto value (as it could be ~2ms)
1350 */
1351static inline unsigned long tcp_probe0_base(const struct sock *sk)
1352{
1353 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1354}
1355
1356/* Variant of inet_csk_rto_backoff() used for zero window probes */
1357static inline unsigned long tcp_probe0_when(const struct sock *sk,
1358 unsigned long max_when)
1359{
1360 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1361 inet_csk(sk)->icsk_backoff);
1362 u64 when = (u64)tcp_probe0_base(sk) << backoff;
1363
1364 return (unsigned long)min_t(u64, when, max_when);
1365}
1366
1367static inline void tcp_check_probe_timer(struct sock *sk)
1368{
1369 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1370 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1371 tcp_probe0_base(sk), TCP_RTO_MAX);
1372}
1373
1374static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1375{
1376 tp->snd_wl1 = seq;
1377}
1378
1379static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1380{
1381 tp->snd_wl1 = seq;
1382}
1383
1384/*
1385 * Calculate(/check) TCP checksum
1386 */
1387static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1388 __be32 daddr, __wsum base)
1389{
1390 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1391}
1392
1393static inline bool tcp_checksum_complete(struct sk_buff *skb)
1394{
1395 return !skb_csum_unnecessary(skb) &&
1396 __skb_checksum_complete(skb);
1397}
1398
1399bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1400 enum skb_drop_reason *reason);
1401
1402
1403int tcp_filter(struct sock *sk, struct sk_buff *skb);
1404void tcp_set_state(struct sock *sk, int state);
1405void tcp_done(struct sock *sk);
1406int tcp_abort(struct sock *sk, int err);
1407
1408static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1409{
1410 rx_opt->dsack = 0;
1411 rx_opt->num_sacks = 0;
1412}
1413
1414void tcp_cwnd_restart(struct sock *sk, s32 delta);
1415
1416static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1417{
1418 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1419 struct tcp_sock *tp = tcp_sk(sk);
1420 s32 delta;
1421
1422 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) ||
1423 tp->packets_out || ca_ops->cong_control)
1424 return;
1425 delta = tcp_jiffies32 - tp->lsndtime;
1426 if (delta > inet_csk(sk)->icsk_rto)
1427 tcp_cwnd_restart(sk, delta);
1428}
1429
1430/* Determine a window scaling and initial window to offer. */
1431void tcp_select_initial_window(const struct sock *sk, int __space,
1432 __u32 mss, __u32 *rcv_wnd,
1433 __u32 *window_clamp, int wscale_ok,
1434 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1435
1436static inline int tcp_win_from_space(const struct sock *sk, int space)
1437{
1438 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale);
1439
1440 return tcp_adv_win_scale <= 0 ?
1441 (space>>(-tcp_adv_win_scale)) :
1442 space - (space>>tcp_adv_win_scale);
1443}
1444
1445/* Note: caller must be prepared to deal with negative returns */
1446static inline int tcp_space(const struct sock *sk)
1447{
1448 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1449 READ_ONCE(sk->sk_backlog.len) -
1450 atomic_read(&sk->sk_rmem_alloc));
1451}
1452
1453static inline int tcp_full_space(const struct sock *sk)
1454{
1455 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1456}
1457
1458static inline void tcp_adjust_rcv_ssthresh(struct sock *sk)
1459{
1460 int unused_mem = sk_unused_reserved_mem(sk);
1461 struct tcp_sock *tp = tcp_sk(sk);
1462
1463 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss);
1464 if (unused_mem)
1465 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh,
1466 tcp_win_from_space(sk, unused_mem));
1467}
1468
1469void tcp_cleanup_rbuf(struct sock *sk, int copied);
1470
1471/* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1472 * If 87.5 % (7/8) of the space has been consumed, we want to override
1473 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1474 * len/truesize ratio.
1475 */
1476static inline bool tcp_rmem_pressure(const struct sock *sk)
1477{
1478 int rcvbuf, threshold;
1479
1480 if (tcp_under_memory_pressure(sk))
1481 return true;
1482
1483 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1484 threshold = rcvbuf - (rcvbuf >> 3);
1485
1486 return atomic_read(&sk->sk_rmem_alloc) > threshold;
1487}
1488
1489static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1490{
1491 const struct tcp_sock *tp = tcp_sk(sk);
1492 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1493
1494 if (avail <= 0)
1495 return false;
1496
1497 return (avail >= target) || tcp_rmem_pressure(sk) ||
1498 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1499}
1500
1501extern void tcp_openreq_init_rwin(struct request_sock *req,
1502 const struct sock *sk_listener,
1503 const struct dst_entry *dst);
1504
1505void tcp_enter_memory_pressure(struct sock *sk);
1506void tcp_leave_memory_pressure(struct sock *sk);
1507
1508static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1509{
1510 struct net *net = sock_net((struct sock *)tp);
1511
1512 return tp->keepalive_intvl ? :
1513 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl);
1514}
1515
1516static inline int keepalive_time_when(const struct tcp_sock *tp)
1517{
1518 struct net *net = sock_net((struct sock *)tp);
1519
1520 return tp->keepalive_time ? :
1521 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time);
1522}
1523
1524static inline int keepalive_probes(const struct tcp_sock *tp)
1525{
1526 struct net *net = sock_net((struct sock *)tp);
1527
1528 return tp->keepalive_probes ? :
1529 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes);
1530}
1531
1532static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1533{
1534 const struct inet_connection_sock *icsk = &tp->inet_conn;
1535
1536 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1537 tcp_jiffies32 - tp->rcv_tstamp);
1538}
1539
1540static inline int tcp_fin_time(const struct sock *sk)
1541{
1542 int fin_timeout = tcp_sk(sk)->linger2 ? :
1543 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout);
1544 const int rto = inet_csk(sk)->icsk_rto;
1545
1546 if (fin_timeout < (rto << 2) - (rto >> 1))
1547 fin_timeout = (rto << 2) - (rto >> 1);
1548
1549 return fin_timeout;
1550}
1551
1552static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1553 int paws_win)
1554{
1555 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1556 return true;
1557 if (unlikely(!time_before32(ktime_get_seconds(),
1558 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1559 return true;
1560 /*
1561 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1562 * then following tcp messages have valid values. Ignore 0 value,
1563 * or else 'negative' tsval might forbid us to accept their packets.
1564 */
1565 if (!rx_opt->ts_recent)
1566 return true;
1567 return false;
1568}
1569
1570static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1571 int rst)
1572{
1573 if (tcp_paws_check(rx_opt, 0))
1574 return false;
1575
1576 /* RST segments are not recommended to carry timestamp,
1577 and, if they do, it is recommended to ignore PAWS because
1578 "their cleanup function should take precedence over timestamps."
1579 Certainly, it is mistake. It is necessary to understand the reasons
1580 of this constraint to relax it: if peer reboots, clock may go
1581 out-of-sync and half-open connections will not be reset.
1582 Actually, the problem would be not existing if all
1583 the implementations followed draft about maintaining clock
1584 via reboots. Linux-2.2 DOES NOT!
1585
1586 However, we can relax time bounds for RST segments to MSL.
1587 */
1588 if (rst && !time_before32(ktime_get_seconds(),
1589 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1590 return false;
1591 return true;
1592}
1593
1594bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1595 int mib_idx, u32 *last_oow_ack_time);
1596
1597static inline void tcp_mib_init(struct net *net)
1598{
1599 /* See RFC 2012 */
1600 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1601 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1602 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1603 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1604}
1605
1606/* from STCP */
1607static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1608{
1609 tp->lost_skb_hint = NULL;
1610}
1611
1612static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1613{
1614 tcp_clear_retrans_hints_partial(tp);
1615 tp->retransmit_skb_hint = NULL;
1616}
1617
1618union tcp_md5_addr {
1619 struct in_addr a4;
1620#if IS_ENABLED(CONFIG_IPV6)
1621 struct in6_addr a6;
1622#endif
1623};
1624
1625/* - key database */
1626struct tcp_md5sig_key {
1627 struct hlist_node node;
1628 u8 keylen;
1629 u8 family; /* AF_INET or AF_INET6 */
1630 u8 prefixlen;
1631 u8 flags;
1632 union tcp_md5_addr addr;
1633 int l3index; /* set if key added with L3 scope */
1634 u8 key[TCP_MD5SIG_MAXKEYLEN];
1635 struct rcu_head rcu;
1636};
1637
1638/* - sock block */
1639struct tcp_md5sig_info {
1640 struct hlist_head head;
1641 struct rcu_head rcu;
1642};
1643
1644/* - pseudo header */
1645struct tcp4_pseudohdr {
1646 __be32 saddr;
1647 __be32 daddr;
1648 __u8 pad;
1649 __u8 protocol;
1650 __be16 len;
1651};
1652
1653struct tcp6_pseudohdr {
1654 struct in6_addr saddr;
1655 struct in6_addr daddr;
1656 __be32 len;
1657 __be32 protocol; /* including padding */
1658};
1659
1660union tcp_md5sum_block {
1661 struct tcp4_pseudohdr ip4;
1662#if IS_ENABLED(CONFIG_IPV6)
1663 struct tcp6_pseudohdr ip6;
1664#endif
1665};
1666
1667/* - pool: digest algorithm, hash description and scratch buffer */
1668struct tcp_md5sig_pool {
1669 struct ahash_request *md5_req;
1670 void *scratch;
1671};
1672
1673/* - functions */
1674int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1675 const struct sock *sk, const struct sk_buff *skb);
1676int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1677 int family, u8 prefixlen, int l3index, u8 flags,
1678 const u8 *newkey, u8 newkeylen);
1679int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr,
1680 int family, u8 prefixlen, int l3index,
1681 struct tcp_md5sig_key *key);
1682
1683int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1684 int family, u8 prefixlen, int l3index, u8 flags);
1685struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1686 const struct sock *addr_sk);
1687
1688#ifdef CONFIG_TCP_MD5SIG
1689#include <linux/jump_label.h>
1690extern struct static_key_false_deferred tcp_md5_needed;
1691struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1692 const union tcp_md5_addr *addr,
1693 int family);
1694static inline struct tcp_md5sig_key *
1695tcp_md5_do_lookup(const struct sock *sk, int l3index,
1696 const union tcp_md5_addr *addr, int family)
1697{
1698 if (!static_branch_unlikely(&tcp_md5_needed.key))
1699 return NULL;
1700 return __tcp_md5_do_lookup(sk, l3index, addr, family);
1701}
1702
1703enum skb_drop_reason
1704tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1705 const void *saddr, const void *daddr,
1706 int family, int dif, int sdif);
1707
1708
1709#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1710#else
1711static inline struct tcp_md5sig_key *
1712tcp_md5_do_lookup(const struct sock *sk, int l3index,
1713 const union tcp_md5_addr *addr, int family)
1714{
1715 return NULL;
1716}
1717
1718static inline enum skb_drop_reason
1719tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1720 const void *saddr, const void *daddr,
1721 int family, int dif, int sdif)
1722{
1723 return SKB_NOT_DROPPED_YET;
1724}
1725#define tcp_twsk_md5_key(twsk) NULL
1726#endif
1727
1728bool tcp_alloc_md5sig_pool(void);
1729
1730struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1731static inline void tcp_put_md5sig_pool(void)
1732{
1733 local_bh_enable();
1734}
1735
1736int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1737 unsigned int header_len);
1738int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1739 const struct tcp_md5sig_key *key);
1740
1741/* From tcp_fastopen.c */
1742void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1743 struct tcp_fastopen_cookie *cookie);
1744void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1745 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1746 u16 try_exp);
1747struct tcp_fastopen_request {
1748 /* Fast Open cookie. Size 0 means a cookie request */
1749 struct tcp_fastopen_cookie cookie;
1750 struct msghdr *data; /* data in MSG_FASTOPEN */
1751 size_t size;
1752 int copied; /* queued in tcp_connect() */
1753 struct ubuf_info *uarg;
1754};
1755void tcp_free_fastopen_req(struct tcp_sock *tp);
1756void tcp_fastopen_destroy_cipher(struct sock *sk);
1757void tcp_fastopen_ctx_destroy(struct net *net);
1758int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1759 void *primary_key, void *backup_key);
1760int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1761 u64 *key);
1762void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1763struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1764 struct request_sock *req,
1765 struct tcp_fastopen_cookie *foc,
1766 const struct dst_entry *dst);
1767void tcp_fastopen_init_key_once(struct net *net);
1768bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1769 struct tcp_fastopen_cookie *cookie);
1770bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1771#define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1772#define TCP_FASTOPEN_KEY_MAX 2
1773#define TCP_FASTOPEN_KEY_BUF_LENGTH \
1774 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1775
1776/* Fastopen key context */
1777struct tcp_fastopen_context {
1778 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
1779 int num;
1780 struct rcu_head rcu;
1781};
1782
1783void tcp_fastopen_active_disable(struct sock *sk);
1784bool tcp_fastopen_active_should_disable(struct sock *sk);
1785void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1786void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1787
1788/* Caller needs to wrap with rcu_read_(un)lock() */
1789static inline
1790struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1791{
1792 struct tcp_fastopen_context *ctx;
1793
1794 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1795 if (!ctx)
1796 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1797 return ctx;
1798}
1799
1800static inline
1801bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1802 const struct tcp_fastopen_cookie *orig)
1803{
1804 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1805 orig->len == foc->len &&
1806 !memcmp(orig->val, foc->val, foc->len))
1807 return true;
1808 return false;
1809}
1810
1811static inline
1812int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1813{
1814 return ctx->num;
1815}
1816
1817/* Latencies incurred by various limits for a sender. They are
1818 * chronograph-like stats that are mutually exclusive.
1819 */
1820enum tcp_chrono {
1821 TCP_CHRONO_UNSPEC,
1822 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1823 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1824 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1825 __TCP_CHRONO_MAX,
1826};
1827
1828void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1829void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1830
1831/* This helper is needed, because skb->tcp_tsorted_anchor uses
1832 * the same memory storage than skb->destructor/_skb_refdst
1833 */
1834static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1835{
1836 skb->destructor = NULL;
1837 skb->_skb_refdst = 0UL;
1838}
1839
1840#define tcp_skb_tsorted_save(skb) { \
1841 unsigned long _save = skb->_skb_refdst; \
1842 skb->_skb_refdst = 0UL;
1843
1844#define tcp_skb_tsorted_restore(skb) \
1845 skb->_skb_refdst = _save; \
1846}
1847
1848void tcp_write_queue_purge(struct sock *sk);
1849
1850static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1851{
1852 return skb_rb_first(&sk->tcp_rtx_queue);
1853}
1854
1855static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1856{
1857 return skb_rb_last(&sk->tcp_rtx_queue);
1858}
1859
1860static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1861{
1862 return skb_peek_tail(&sk->sk_write_queue);
1863}
1864
1865#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1866 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1867
1868static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1869{
1870 return skb_peek(&sk->sk_write_queue);
1871}
1872
1873static inline bool tcp_skb_is_last(const struct sock *sk,
1874 const struct sk_buff *skb)
1875{
1876 return skb_queue_is_last(&sk->sk_write_queue, skb);
1877}
1878
1879/**
1880 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1881 * @sk: socket
1882 *
1883 * Since the write queue can have a temporary empty skb in it,
1884 * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1885 */
1886static inline bool tcp_write_queue_empty(const struct sock *sk)
1887{
1888 const struct tcp_sock *tp = tcp_sk(sk);
1889
1890 return tp->write_seq == tp->snd_nxt;
1891}
1892
1893static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1894{
1895 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1896}
1897
1898static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1899{
1900 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1901}
1902
1903static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1904{
1905 __skb_queue_tail(&sk->sk_write_queue, skb);
1906
1907 /* Queue it, remembering where we must start sending. */
1908 if (sk->sk_write_queue.next == skb)
1909 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1910}
1911
1912/* Insert new before skb on the write queue of sk. */
1913static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1914 struct sk_buff *skb,
1915 struct sock *sk)
1916{
1917 __skb_queue_before(&sk->sk_write_queue, skb, new);
1918}
1919
1920static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1921{
1922 tcp_skb_tsorted_anchor_cleanup(skb);
1923 __skb_unlink(skb, &sk->sk_write_queue);
1924}
1925
1926void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1927
1928static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1929{
1930 tcp_skb_tsorted_anchor_cleanup(skb);
1931 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1932}
1933
1934static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1935{
1936 list_del(&skb->tcp_tsorted_anchor);
1937 tcp_rtx_queue_unlink(skb, sk);
1938 tcp_wmem_free_skb(sk, skb);
1939}
1940
1941static inline void tcp_push_pending_frames(struct sock *sk)
1942{
1943 if (tcp_send_head(sk)) {
1944 struct tcp_sock *tp = tcp_sk(sk);
1945
1946 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1947 }
1948}
1949
1950/* Start sequence of the skb just after the highest skb with SACKed
1951 * bit, valid only if sacked_out > 0 or when the caller has ensured
1952 * validity by itself.
1953 */
1954static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1955{
1956 if (!tp->sacked_out)
1957 return tp->snd_una;
1958
1959 if (tp->highest_sack == NULL)
1960 return tp->snd_nxt;
1961
1962 return TCP_SKB_CB(tp->highest_sack)->seq;
1963}
1964
1965static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1966{
1967 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1968}
1969
1970static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1971{
1972 return tcp_sk(sk)->highest_sack;
1973}
1974
1975static inline void tcp_highest_sack_reset(struct sock *sk)
1976{
1977 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1978}
1979
1980/* Called when old skb is about to be deleted and replaced by new skb */
1981static inline void tcp_highest_sack_replace(struct sock *sk,
1982 struct sk_buff *old,
1983 struct sk_buff *new)
1984{
1985 if (old == tcp_highest_sack(sk))
1986 tcp_sk(sk)->highest_sack = new;
1987}
1988
1989/* This helper checks if socket has IP_TRANSPARENT set */
1990static inline bool inet_sk_transparent(const struct sock *sk)
1991{
1992 switch (sk->sk_state) {
1993 case TCP_TIME_WAIT:
1994 return inet_twsk(sk)->tw_transparent;
1995 case TCP_NEW_SYN_RECV:
1996 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1997 }
1998 return inet_sk(sk)->transparent;
1999}
2000
2001/* Determines whether this is a thin stream (which may suffer from
2002 * increased latency). Used to trigger latency-reducing mechanisms.
2003 */
2004static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
2005{
2006 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
2007}
2008
2009/* /proc */
2010enum tcp_seq_states {
2011 TCP_SEQ_STATE_LISTENING,
2012 TCP_SEQ_STATE_ESTABLISHED,
2013};
2014
2015void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
2016void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
2017void tcp_seq_stop(struct seq_file *seq, void *v);
2018
2019struct tcp_seq_afinfo {
2020 sa_family_t family;
2021};
2022
2023struct tcp_iter_state {
2024 struct seq_net_private p;
2025 enum tcp_seq_states state;
2026 struct sock *syn_wait_sk;
2027 int bucket, offset, sbucket, num;
2028 loff_t last_pos;
2029};
2030
2031extern struct request_sock_ops tcp_request_sock_ops;
2032extern struct request_sock_ops tcp6_request_sock_ops;
2033
2034void tcp_v4_destroy_sock(struct sock *sk);
2035
2036struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
2037 netdev_features_t features);
2038struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
2039INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
2040INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
2041INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
2042INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
2043int tcp_gro_complete(struct sk_buff *skb);
2044
2045void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
2046
2047static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
2048{
2049 struct net *net = sock_net((struct sock *)tp);
2050 return tp->notsent_lowat ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
2051}
2052
2053bool tcp_stream_memory_free(const struct sock *sk, int wake);
2054
2055#ifdef CONFIG_PROC_FS
2056int tcp4_proc_init(void);
2057void tcp4_proc_exit(void);
2058#endif
2059
2060int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
2061int tcp_conn_request(struct request_sock_ops *rsk_ops,
2062 const struct tcp_request_sock_ops *af_ops,
2063 struct sock *sk, struct sk_buff *skb);
2064
2065/* TCP af-specific functions */
2066struct tcp_sock_af_ops {
2067#ifdef CONFIG_TCP_MD5SIG
2068 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
2069 const struct sock *addr_sk);
2070 int (*calc_md5_hash)(char *location,
2071 const struct tcp_md5sig_key *md5,
2072 const struct sock *sk,
2073 const struct sk_buff *skb);
2074 int (*md5_parse)(struct sock *sk,
2075 int optname,
2076 sockptr_t optval,
2077 int optlen);
2078#endif
2079};
2080
2081struct tcp_request_sock_ops {
2082 u16 mss_clamp;
2083#ifdef CONFIG_TCP_MD5SIG
2084 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2085 const struct sock *addr_sk);
2086 int (*calc_md5_hash) (char *location,
2087 const struct tcp_md5sig_key *md5,
2088 const struct sock *sk,
2089 const struct sk_buff *skb);
2090#endif
2091#ifdef CONFIG_SYN_COOKIES
2092 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
2093 __u16 *mss);
2094#endif
2095 struct dst_entry *(*route_req)(const struct sock *sk,
2096 struct sk_buff *skb,
2097 struct flowi *fl,
2098 struct request_sock *req);
2099 u32 (*init_seq)(const struct sk_buff *skb);
2100 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2101 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2102 struct flowi *fl, struct request_sock *req,
2103 struct tcp_fastopen_cookie *foc,
2104 enum tcp_synack_type synack_type,
2105 struct sk_buff *syn_skb);
2106};
2107
2108extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2109#if IS_ENABLED(CONFIG_IPV6)
2110extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2111#endif
2112
2113#ifdef CONFIG_SYN_COOKIES
2114static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2115 const struct sock *sk, struct sk_buff *skb,
2116 __u16 *mss)
2117{
2118 tcp_synq_overflow(sk);
2119 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2120 return ops->cookie_init_seq(skb, mss);
2121}
2122#else
2123static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2124 const struct sock *sk, struct sk_buff *skb,
2125 __u16 *mss)
2126{
2127 return 0;
2128}
2129#endif
2130
2131int tcpv4_offload_init(void);
2132
2133void tcp_v4_init(void);
2134void tcp_init(void);
2135
2136/* tcp_recovery.c */
2137void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2138void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2139extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2140 u32 reo_wnd);
2141extern bool tcp_rack_mark_lost(struct sock *sk);
2142extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2143 u64 xmit_time);
2144extern void tcp_rack_reo_timeout(struct sock *sk);
2145extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2146
2147/* tcp_plb.c */
2148
2149/*
2150 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state
2151 * expects cong_ratio which represents fraction of traffic that experienced
2152 * congestion over a single RTT. In order to avoid floating point operations,
2153 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in.
2154 */
2155#define TCP_PLB_SCALE 8
2156
2157/* State for PLB (Protective Load Balancing) for a single TCP connection. */
2158struct tcp_plb_state {
2159 u8 consec_cong_rounds:5, /* consecutive congested rounds */
2160 unused:3;
2161 u32 pause_until; /* jiffies32 when PLB can resume rerouting */
2162};
2163
2164static inline void tcp_plb_init(const struct sock *sk,
2165 struct tcp_plb_state *plb)
2166{
2167 plb->consec_cong_rounds = 0;
2168 plb->pause_until = 0;
2169}
2170void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb,
2171 const int cong_ratio);
2172void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb);
2173void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb);
2174
2175/* At how many usecs into the future should the RTO fire? */
2176static inline s64 tcp_rto_delta_us(const struct sock *sk)
2177{
2178 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2179 u32 rto = inet_csk(sk)->icsk_rto;
2180 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2181
2182 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2183}
2184
2185/*
2186 * Save and compile IPv4 options, return a pointer to it
2187 */
2188static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2189 struct sk_buff *skb)
2190{
2191 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2192 struct ip_options_rcu *dopt = NULL;
2193
2194 if (opt->optlen) {
2195 int opt_size = sizeof(*dopt) + opt->optlen;
2196
2197 dopt = kmalloc(opt_size, GFP_ATOMIC);
2198 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2199 kfree(dopt);
2200 dopt = NULL;
2201 }
2202 }
2203 return dopt;
2204}
2205
2206/* locally generated TCP pure ACKs have skb->truesize == 2
2207 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2208 * This is much faster than dissecting the packet to find out.
2209 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2210 */
2211static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2212{
2213 return skb->truesize == 2;
2214}
2215
2216static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2217{
2218 skb->truesize = 2;
2219}
2220
2221static inline int tcp_inq(struct sock *sk)
2222{
2223 struct tcp_sock *tp = tcp_sk(sk);
2224 int answ;
2225
2226 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2227 answ = 0;
2228 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2229 !tp->urg_data ||
2230 before(tp->urg_seq, tp->copied_seq) ||
2231 !before(tp->urg_seq, tp->rcv_nxt)) {
2232
2233 answ = tp->rcv_nxt - tp->copied_seq;
2234
2235 /* Subtract 1, if FIN was received */
2236 if (answ && sock_flag(sk, SOCK_DONE))
2237 answ--;
2238 } else {
2239 answ = tp->urg_seq - tp->copied_seq;
2240 }
2241
2242 return answ;
2243}
2244
2245int tcp_peek_len(struct socket *sock);
2246
2247static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2248{
2249 u16 segs_in;
2250
2251 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2252
2253 /* We update these fields while other threads might
2254 * read them from tcp_get_info()
2255 */
2256 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in);
2257 if (skb->len > tcp_hdrlen(skb))
2258 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in);
2259}
2260
2261/*
2262 * TCP listen path runs lockless.
2263 * We forced "struct sock" to be const qualified to make sure
2264 * we don't modify one of its field by mistake.
2265 * Here, we increment sk_drops which is an atomic_t, so we can safely
2266 * make sock writable again.
2267 */
2268static inline void tcp_listendrop(const struct sock *sk)
2269{
2270 atomic_inc(&((struct sock *)sk)->sk_drops);
2271 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2272}
2273
2274enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2275
2276/*
2277 * Interface for adding Upper Level Protocols over TCP
2278 */
2279
2280#define TCP_ULP_NAME_MAX 16
2281#define TCP_ULP_MAX 128
2282#define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2283
2284struct tcp_ulp_ops {
2285 struct list_head list;
2286
2287 /* initialize ulp */
2288 int (*init)(struct sock *sk);
2289 /* update ulp */
2290 void (*update)(struct sock *sk, struct proto *p,
2291 void (*write_space)(struct sock *sk));
2292 /* cleanup ulp */
2293 void (*release)(struct sock *sk);
2294 /* diagnostic */
2295 int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2296 size_t (*get_info_size)(const struct sock *sk);
2297 /* clone ulp */
2298 void (*clone)(const struct request_sock *req, struct sock *newsk,
2299 const gfp_t priority);
2300
2301 char name[TCP_ULP_NAME_MAX];
2302 struct module *owner;
2303};
2304int tcp_register_ulp(struct tcp_ulp_ops *type);
2305void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2306int tcp_set_ulp(struct sock *sk, const char *name);
2307void tcp_get_available_ulp(char *buf, size_t len);
2308void tcp_cleanup_ulp(struct sock *sk);
2309void tcp_update_ulp(struct sock *sk, struct proto *p,
2310 void (*write_space)(struct sock *sk));
2311
2312#define MODULE_ALIAS_TCP_ULP(name) \
2313 __MODULE_INFO(alias, alias_userspace, name); \
2314 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2315
2316#ifdef CONFIG_NET_SOCK_MSG
2317struct sk_msg;
2318struct sk_psock;
2319
2320#ifdef CONFIG_BPF_SYSCALL
2321struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
2322int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
2323void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2324#endif /* CONFIG_BPF_SYSCALL */
2325
2326int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress,
2327 struct sk_msg *msg, u32 bytes, int flags);
2328#endif /* CONFIG_NET_SOCK_MSG */
2329
2330#if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2331static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2332{
2333}
2334#endif
2335
2336#ifdef CONFIG_CGROUP_BPF
2337static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2338 struct sk_buff *skb,
2339 unsigned int end_offset)
2340{
2341 skops->skb = skb;
2342 skops->skb_data_end = skb->data + end_offset;
2343}
2344#else
2345static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2346 struct sk_buff *skb,
2347 unsigned int end_offset)
2348{
2349}
2350#endif
2351
2352/* Call BPF_SOCK_OPS program that returns an int. If the return value
2353 * is < 0, then the BPF op failed (for example if the loaded BPF
2354 * program does not support the chosen operation or there is no BPF
2355 * program loaded).
2356 */
2357#ifdef CONFIG_BPF
2358static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2359{
2360 struct bpf_sock_ops_kern sock_ops;
2361 int ret;
2362
2363 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2364 if (sk_fullsock(sk)) {
2365 sock_ops.is_fullsock = 1;
2366 sock_owned_by_me(sk);
2367 }
2368
2369 sock_ops.sk = sk;
2370 sock_ops.op = op;
2371 if (nargs > 0)
2372 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2373
2374 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2375 if (ret == 0)
2376 ret = sock_ops.reply;
2377 else
2378 ret = -1;
2379 return ret;
2380}
2381
2382static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2383{
2384 u32 args[2] = {arg1, arg2};
2385
2386 return tcp_call_bpf(sk, op, 2, args);
2387}
2388
2389static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2390 u32 arg3)
2391{
2392 u32 args[3] = {arg1, arg2, arg3};
2393
2394 return tcp_call_bpf(sk, op, 3, args);
2395}
2396
2397#else
2398static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2399{
2400 return -EPERM;
2401}
2402
2403static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2404{
2405 return -EPERM;
2406}
2407
2408static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2409 u32 arg3)
2410{
2411 return -EPERM;
2412}
2413
2414#endif
2415
2416static inline u32 tcp_timeout_init(struct sock *sk)
2417{
2418 int timeout;
2419
2420 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2421
2422 if (timeout <= 0)
2423 timeout = TCP_TIMEOUT_INIT;
2424 return min_t(int, timeout, TCP_RTO_MAX);
2425}
2426
2427static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2428{
2429 int rwnd;
2430
2431 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2432
2433 if (rwnd < 0)
2434 rwnd = 0;
2435 return rwnd;
2436}
2437
2438static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2439{
2440 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2441}
2442
2443static inline void tcp_bpf_rtt(struct sock *sk)
2444{
2445 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2446 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2447}
2448
2449#if IS_ENABLED(CONFIG_SMC)
2450extern struct static_key_false tcp_have_smc;
2451#endif
2452
2453#if IS_ENABLED(CONFIG_TLS_DEVICE)
2454void clean_acked_data_enable(struct inet_connection_sock *icsk,
2455 void (*cad)(struct sock *sk, u32 ack_seq));
2456void clean_acked_data_disable(struct inet_connection_sock *icsk);
2457void clean_acked_data_flush(void);
2458#endif
2459
2460DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2461static inline void tcp_add_tx_delay(struct sk_buff *skb,
2462 const struct tcp_sock *tp)
2463{
2464 if (static_branch_unlikely(&tcp_tx_delay_enabled))
2465 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2466}
2467
2468/* Compute Earliest Departure Time for some control packets
2469 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2470 */
2471static inline u64 tcp_transmit_time(const struct sock *sk)
2472{
2473 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2474 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2475 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2476
2477 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
2478 }
2479 return 0;
2480}
2481
2482#endif /* _TCP_H */