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/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Definitions for the TCP module.
7 *
8 * Version: @(#)tcp.h 1.0.5 05/23/93
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
17 */
18#ifndef _TCP_H
19#define _TCP_H
20
21#define FASTRETRANS_DEBUG 1
22
23#include <linux/list.h>
24#include <linux/tcp.h>
25#include <linux/bug.h>
26#include <linux/slab.h>
27#include <linux/cache.h>
28#include <linux/percpu.h>
29#include <linux/skbuff.h>
30#include <linux/cryptohash.h>
31#include <linux/kref.h>
32#include <linux/ktime.h>
33
34#include <net/inet_connection_sock.h>
35#include <net/inet_timewait_sock.h>
36#include <net/inet_hashtables.h>
37#include <net/checksum.h>
38#include <net/request_sock.h>
39#include <net/sock.h>
40#include <net/snmp.h>
41#include <net/ip.h>
42#include <net/tcp_states.h>
43#include <net/inet_ecn.h>
44#include <net/dst.h>
45
46#include <linux/seq_file.h>
47#include <linux/memcontrol.h>
48
49extern struct inet_hashinfo tcp_hashinfo;
50
51extern struct percpu_counter tcp_orphan_count;
52void tcp_time_wait(struct sock *sk, int state, int timeo);
53
54#define MAX_TCP_HEADER (128 + MAX_HEADER)
55#define MAX_TCP_OPTION_SPACE 40
56
57/*
58 * Never offer a window over 32767 without using window scaling. Some
59 * poor stacks do signed 16bit maths!
60 */
61#define MAX_TCP_WINDOW 32767U
62
63/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
64#define TCP_MIN_MSS 88U
65
66/* The least MTU to use for probing */
67#define TCP_BASE_MSS 1024
68
69/* probing interval, default to 10 minutes as per RFC4821 */
70#define TCP_PROBE_INTERVAL 600
71
72/* Specify interval when tcp mtu probing will stop */
73#define TCP_PROBE_THRESHOLD 8
74
75/* After receiving this amount of duplicate ACKs fast retransmit starts. */
76#define TCP_FASTRETRANS_THRESH 3
77
78/* Maximal number of ACKs sent quickly to accelerate slow-start. */
79#define TCP_MAX_QUICKACKS 16U
80
81/* urg_data states */
82#define TCP_URG_VALID 0x0100
83#define TCP_URG_NOTYET 0x0200
84#define TCP_URG_READ 0x0400
85
86#define TCP_RETR1 3 /*
87 * This is how many retries it does before it
88 * tries to figure out if the gateway is
89 * down. Minimal RFC value is 3; it corresponds
90 * to ~3sec-8min depending on RTO.
91 */
92
93#define TCP_RETR2 15 /*
94 * This should take at least
95 * 90 minutes to time out.
96 * RFC1122 says that the limit is 100 sec.
97 * 15 is ~13-30min depending on RTO.
98 */
99
100#define TCP_SYN_RETRIES 6 /* This is how many retries are done
101 * when active opening a connection.
102 * RFC1122 says the minimum retry MUST
103 * be at least 180secs. Nevertheless
104 * this value is corresponding to
105 * 63secs of retransmission with the
106 * current initial RTO.
107 */
108
109#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
110 * when passive opening a connection.
111 * This is corresponding to 31secs of
112 * retransmission with the current
113 * initial RTO.
114 */
115
116#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
117 * state, about 60 seconds */
118#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
119 /* BSD style FIN_WAIT2 deadlock breaker.
120 * It used to be 3min, new value is 60sec,
121 * to combine FIN-WAIT-2 timeout with
122 * TIME-WAIT timer.
123 */
124
125#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
126#if HZ >= 100
127#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
128#define TCP_ATO_MIN ((unsigned)(HZ/25))
129#else
130#define TCP_DELACK_MIN 4U
131#define TCP_ATO_MIN 4U
132#endif
133#define TCP_RTO_MAX ((unsigned)(120*HZ))
134#define TCP_RTO_MIN ((unsigned)(HZ/5))
135#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
136#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
137 * used as a fallback RTO for the
138 * initial data transmission if no
139 * valid RTT sample has been acquired,
140 * most likely due to retrans in 3WHS.
141 */
142
143#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
144 * for local resources.
145 */
146
147#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
148#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
149#define TCP_KEEPALIVE_INTVL (75*HZ)
150
151#define MAX_TCP_KEEPIDLE 32767
152#define MAX_TCP_KEEPINTVL 32767
153#define MAX_TCP_KEEPCNT 127
154#define MAX_TCP_SYNCNT 127
155
156#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
157
158#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
159#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
160 * after this time. It should be equal
161 * (or greater than) TCP_TIMEWAIT_LEN
162 * to provide reliability equal to one
163 * provided by timewait state.
164 */
165#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
166 * timestamps. It must be less than
167 * minimal timewait lifetime.
168 */
169/*
170 * TCP option
171 */
172
173#define TCPOPT_NOP 1 /* Padding */
174#define TCPOPT_EOL 0 /* End of options */
175#define TCPOPT_MSS 2 /* Segment size negotiating */
176#define TCPOPT_WINDOW 3 /* Window scaling */
177#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
178#define TCPOPT_SACK 5 /* SACK Block */
179#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
180#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
181#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
182#define TCPOPT_EXP 254 /* Experimental */
183/* Magic number to be after the option value for sharing TCP
184 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
185 */
186#define TCPOPT_FASTOPEN_MAGIC 0xF989
187
188/*
189 * TCP option lengths
190 */
191
192#define TCPOLEN_MSS 4
193#define TCPOLEN_WINDOW 3
194#define TCPOLEN_SACK_PERM 2
195#define TCPOLEN_TIMESTAMP 10
196#define TCPOLEN_MD5SIG 18
197#define TCPOLEN_FASTOPEN_BASE 2
198#define TCPOLEN_EXP_FASTOPEN_BASE 4
199
200/* But this is what stacks really send out. */
201#define TCPOLEN_TSTAMP_ALIGNED 12
202#define TCPOLEN_WSCALE_ALIGNED 4
203#define TCPOLEN_SACKPERM_ALIGNED 4
204#define TCPOLEN_SACK_BASE 2
205#define TCPOLEN_SACK_BASE_ALIGNED 4
206#define TCPOLEN_SACK_PERBLOCK 8
207#define TCPOLEN_MD5SIG_ALIGNED 20
208#define TCPOLEN_MSS_ALIGNED 4
209
210/* Flags in tp->nonagle */
211#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
212#define TCP_NAGLE_CORK 2 /* Socket is corked */
213#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
214
215/* TCP thin-stream limits */
216#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
217
218/* TCP initial congestion window as per rfc6928 */
219#define TCP_INIT_CWND 10
220
221/* Bit Flags for sysctl_tcp_fastopen */
222#define TFO_CLIENT_ENABLE 1
223#define TFO_SERVER_ENABLE 2
224#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
225
226/* Accept SYN data w/o any cookie option */
227#define TFO_SERVER_COOKIE_NOT_REQD 0x200
228
229/* Force enable TFO on all listeners, i.e., not requiring the
230 * TCP_FASTOPEN socket option.
231 */
232#define TFO_SERVER_WO_SOCKOPT1 0x400
233
234extern struct inet_timewait_death_row tcp_death_row;
235
236/* sysctl variables for tcp */
237extern int sysctl_tcp_timestamps;
238extern int sysctl_tcp_window_scaling;
239extern int sysctl_tcp_sack;
240extern int sysctl_tcp_fastopen;
241extern int sysctl_tcp_retrans_collapse;
242extern int sysctl_tcp_stdurg;
243extern int sysctl_tcp_rfc1337;
244extern int sysctl_tcp_abort_on_overflow;
245extern int sysctl_tcp_max_orphans;
246extern int sysctl_tcp_fack;
247extern int sysctl_tcp_reordering;
248extern int sysctl_tcp_max_reordering;
249extern int sysctl_tcp_dsack;
250extern long sysctl_tcp_mem[3];
251extern int sysctl_tcp_wmem[3];
252extern int sysctl_tcp_rmem[3];
253extern int sysctl_tcp_app_win;
254extern int sysctl_tcp_adv_win_scale;
255extern int sysctl_tcp_frto;
256extern int sysctl_tcp_low_latency;
257extern int sysctl_tcp_nometrics_save;
258extern int sysctl_tcp_moderate_rcvbuf;
259extern int sysctl_tcp_tso_win_divisor;
260extern int sysctl_tcp_workaround_signed_windows;
261extern int sysctl_tcp_slow_start_after_idle;
262extern int sysctl_tcp_thin_linear_timeouts;
263extern int sysctl_tcp_thin_dupack;
264extern int sysctl_tcp_early_retrans;
265extern int sysctl_tcp_limit_output_bytes;
266extern int sysctl_tcp_challenge_ack_limit;
267extern int sysctl_tcp_min_tso_segs;
268extern int sysctl_tcp_min_rtt_wlen;
269extern int sysctl_tcp_autocorking;
270extern int sysctl_tcp_invalid_ratelimit;
271extern int sysctl_tcp_pacing_ss_ratio;
272extern int sysctl_tcp_pacing_ca_ratio;
273
274extern atomic_long_t tcp_memory_allocated;
275extern struct percpu_counter tcp_sockets_allocated;
276extern int tcp_memory_pressure;
277
278/* optimized version of sk_under_memory_pressure() for TCP sockets */
279static inline bool tcp_under_memory_pressure(const struct sock *sk)
280{
281 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
282 mem_cgroup_under_socket_pressure(sk->sk_memcg))
283 return true;
284
285 return tcp_memory_pressure;
286}
287/*
288 * The next routines deal with comparing 32 bit unsigned ints
289 * and worry about wraparound (automatic with unsigned arithmetic).
290 */
291
292static inline bool before(__u32 seq1, __u32 seq2)
293{
294 return (__s32)(seq1-seq2) < 0;
295}
296#define after(seq2, seq1) before(seq1, seq2)
297
298/* is s2<=s1<=s3 ? */
299static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
300{
301 return seq3 - seq2 >= seq1 - seq2;
302}
303
304static inline bool tcp_out_of_memory(struct sock *sk)
305{
306 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
307 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
308 return true;
309 return false;
310}
311
312void sk_forced_mem_schedule(struct sock *sk, int size);
313
314static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
315{
316 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
317 int orphans = percpu_counter_read_positive(ocp);
318
319 if (orphans << shift > sysctl_tcp_max_orphans) {
320 orphans = percpu_counter_sum_positive(ocp);
321 if (orphans << shift > sysctl_tcp_max_orphans)
322 return true;
323 }
324 return false;
325}
326
327bool tcp_check_oom(struct sock *sk, int shift);
328
329
330extern struct proto tcp_prot;
331
332#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
333#define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
334#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
335#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
336
337void tcp_tasklet_init(void);
338
339void tcp_v4_err(struct sk_buff *skb, u32);
340
341void tcp_shutdown(struct sock *sk, int how);
342
343void tcp_v4_early_demux(struct sk_buff *skb);
344int tcp_v4_rcv(struct sk_buff *skb);
345
346int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
347int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
348int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
349 int flags);
350void tcp_release_cb(struct sock *sk);
351void tcp_wfree(struct sk_buff *skb);
352void tcp_write_timer_handler(struct sock *sk);
353void tcp_delack_timer_handler(struct sock *sk);
354int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
355int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
356void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
357 const struct tcphdr *th, unsigned int len);
358void tcp_rcv_space_adjust(struct sock *sk);
359int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
360void tcp_twsk_destructor(struct sock *sk);
361ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
362 struct pipe_inode_info *pipe, size_t len,
363 unsigned int flags);
364
365static inline void tcp_dec_quickack_mode(struct sock *sk,
366 const unsigned int pkts)
367{
368 struct inet_connection_sock *icsk = inet_csk(sk);
369
370 if (icsk->icsk_ack.quick) {
371 if (pkts >= icsk->icsk_ack.quick) {
372 icsk->icsk_ack.quick = 0;
373 /* Leaving quickack mode we deflate ATO. */
374 icsk->icsk_ack.ato = TCP_ATO_MIN;
375 } else
376 icsk->icsk_ack.quick -= pkts;
377 }
378}
379
380#define TCP_ECN_OK 1
381#define TCP_ECN_QUEUE_CWR 2
382#define TCP_ECN_DEMAND_CWR 4
383#define TCP_ECN_SEEN 8
384
385enum tcp_tw_status {
386 TCP_TW_SUCCESS = 0,
387 TCP_TW_RST = 1,
388 TCP_TW_ACK = 2,
389 TCP_TW_SYN = 3
390};
391
392
393enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
394 struct sk_buff *skb,
395 const struct tcphdr *th);
396struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
397 struct request_sock *req, bool fastopen);
398int tcp_child_process(struct sock *parent, struct sock *child,
399 struct sk_buff *skb);
400void tcp_enter_loss(struct sock *sk);
401void tcp_clear_retrans(struct tcp_sock *tp);
402void tcp_update_metrics(struct sock *sk);
403void tcp_init_metrics(struct sock *sk);
404void tcp_metrics_init(void);
405bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
406 bool paws_check, bool timestamps);
407bool tcp_remember_stamp(struct sock *sk);
408bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
409void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
410void tcp_disable_fack(struct tcp_sock *tp);
411void tcp_close(struct sock *sk, long timeout);
412void tcp_init_sock(struct sock *sk);
413unsigned int tcp_poll(struct file *file, struct socket *sock,
414 struct poll_table_struct *wait);
415int tcp_getsockopt(struct sock *sk, int level, int optname,
416 char __user *optval, int __user *optlen);
417int tcp_setsockopt(struct sock *sk, int level, int optname,
418 char __user *optval, unsigned int optlen);
419int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
420 char __user *optval, int __user *optlen);
421int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
422 char __user *optval, unsigned int optlen);
423void tcp_set_keepalive(struct sock *sk, int val);
424void tcp_syn_ack_timeout(const struct request_sock *req);
425int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
426 int flags, int *addr_len);
427void tcp_parse_options(const struct sk_buff *skb,
428 struct tcp_options_received *opt_rx,
429 int estab, struct tcp_fastopen_cookie *foc);
430const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
431
432/*
433 * TCP v4 functions exported for the inet6 API
434 */
435
436void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
437void tcp_v4_mtu_reduced(struct sock *sk);
438void tcp_req_err(struct sock *sk, u32 seq, bool abort);
439int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
440struct sock *tcp_create_openreq_child(const struct sock *sk,
441 struct request_sock *req,
442 struct sk_buff *skb);
443void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
444struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
445 struct request_sock *req,
446 struct dst_entry *dst,
447 struct request_sock *req_unhash,
448 bool *own_req);
449int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
450int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
451int tcp_connect(struct sock *sk);
452enum tcp_synack_type {
453 TCP_SYNACK_NORMAL,
454 TCP_SYNACK_FASTOPEN,
455 TCP_SYNACK_COOKIE,
456};
457struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
458 struct request_sock *req,
459 struct tcp_fastopen_cookie *foc,
460 enum tcp_synack_type synack_type);
461int tcp_disconnect(struct sock *sk, int flags);
462
463void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
464int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
465void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
466
467/* From syncookies.c */
468struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
469 struct request_sock *req,
470 struct dst_entry *dst);
471int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
472 u32 cookie);
473struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
474#ifdef CONFIG_SYN_COOKIES
475
476/* Syncookies use a monotonic timer which increments every 60 seconds.
477 * This counter is used both as a hash input and partially encoded into
478 * the cookie value. A cookie is only validated further if the delta
479 * between the current counter value and the encoded one is less than this,
480 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
481 * the counter advances immediately after a cookie is generated).
482 */
483#define MAX_SYNCOOKIE_AGE 2
484#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
485#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
486
487/* syncookies: remember time of last synqueue overflow
488 * But do not dirty this field too often (once per second is enough)
489 * It is racy as we do not hold a lock, but race is very minor.
490 */
491static inline void tcp_synq_overflow(const struct sock *sk)
492{
493 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
494 unsigned long now = jiffies;
495
496 if (time_after(now, last_overflow + HZ))
497 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
498}
499
500/* syncookies: no recent synqueue overflow on this listening socket? */
501static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
502{
503 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
504
505 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
506}
507
508static inline u32 tcp_cookie_time(void)
509{
510 u64 val = get_jiffies_64();
511
512 do_div(val, TCP_SYNCOOKIE_PERIOD);
513 return val;
514}
515
516u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
517 u16 *mssp);
518__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
519__u32 cookie_init_timestamp(struct request_sock *req);
520bool cookie_timestamp_decode(struct tcp_options_received *opt);
521bool cookie_ecn_ok(const struct tcp_options_received *opt,
522 const struct net *net, const struct dst_entry *dst);
523
524/* From net/ipv6/syncookies.c */
525int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
526 u32 cookie);
527struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
528
529u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
530 const struct tcphdr *th, u16 *mssp);
531__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
532#endif
533/* tcp_output.c */
534
535u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
536 int min_tso_segs);
537void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
538 int nonagle);
539bool tcp_may_send_now(struct sock *sk);
540int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
541int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
542void tcp_retransmit_timer(struct sock *sk);
543void tcp_xmit_retransmit_queue(struct sock *);
544void tcp_simple_retransmit(struct sock *);
545int tcp_trim_head(struct sock *, struct sk_buff *, u32);
546int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
547
548void tcp_send_probe0(struct sock *);
549void tcp_send_partial(struct sock *);
550int tcp_write_wakeup(struct sock *, int mib);
551void tcp_send_fin(struct sock *sk);
552void tcp_send_active_reset(struct sock *sk, gfp_t priority);
553int tcp_send_synack(struct sock *);
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);
559void tcp_skb_collapse_tstamp(struct sk_buff *skb,
560 const struct sk_buff *next_skb);
561
562/* tcp_input.c */
563void tcp_resume_early_retransmit(struct sock *sk);
564void tcp_rearm_rto(struct sock *sk);
565void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
566void tcp_reset(struct sock *sk);
567void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
568void tcp_fin(struct sock *sk);
569
570/* tcp_timer.c */
571void tcp_init_xmit_timers(struct sock *);
572static inline void tcp_clear_xmit_timers(struct sock *sk)
573{
574 inet_csk_clear_xmit_timers(sk);
575}
576
577unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
578unsigned int tcp_current_mss(struct sock *sk);
579
580/* Bound MSS / TSO packet size with the half of the window */
581static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
582{
583 int cutoff;
584
585 /* When peer uses tiny windows, there is no use in packetizing
586 * to sub-MSS pieces for the sake of SWS or making sure there
587 * are enough packets in the pipe for fast recovery.
588 *
589 * On the other hand, for extremely large MSS devices, handling
590 * smaller than MSS windows in this way does make sense.
591 */
592 if (tp->max_window > TCP_MSS_DEFAULT)
593 cutoff = (tp->max_window >> 1);
594 else
595 cutoff = tp->max_window;
596
597 if (cutoff && pktsize > cutoff)
598 return max_t(int, cutoff, 68U - tp->tcp_header_len);
599 else
600 return pktsize;
601}
602
603/* tcp.c */
604void tcp_get_info(struct sock *, struct tcp_info *);
605
606/* Read 'sendfile()'-style from a TCP socket */
607int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
608 sk_read_actor_t recv_actor);
609
610void tcp_initialize_rcv_mss(struct sock *sk);
611
612int tcp_mtu_to_mss(struct sock *sk, int pmtu);
613int tcp_mss_to_mtu(struct sock *sk, int mss);
614void tcp_mtup_init(struct sock *sk);
615void tcp_init_buffer_space(struct sock *sk);
616
617static inline void tcp_bound_rto(const struct sock *sk)
618{
619 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
620 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
621}
622
623static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
624{
625 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
626}
627
628static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
629{
630 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
631 ntohl(TCP_FLAG_ACK) |
632 snd_wnd);
633}
634
635static inline void tcp_fast_path_on(struct tcp_sock *tp)
636{
637 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
638}
639
640static inline void tcp_fast_path_check(struct sock *sk)
641{
642 struct tcp_sock *tp = tcp_sk(sk);
643
644 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
645 tp->rcv_wnd &&
646 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
647 !tp->urg_data)
648 tcp_fast_path_on(tp);
649}
650
651/* Compute the actual rto_min value */
652static inline u32 tcp_rto_min(struct sock *sk)
653{
654 const struct dst_entry *dst = __sk_dst_get(sk);
655 u32 rto_min = TCP_RTO_MIN;
656
657 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
658 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
659 return rto_min;
660}
661
662static inline u32 tcp_rto_min_us(struct sock *sk)
663{
664 return jiffies_to_usecs(tcp_rto_min(sk));
665}
666
667static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
668{
669 return dst_metric_locked(dst, RTAX_CC_ALGO);
670}
671
672/* Minimum RTT in usec. ~0 means not available. */
673static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
674{
675 return minmax_get(&tp->rtt_min);
676}
677
678/* Compute the actual receive window we are currently advertising.
679 * Rcv_nxt can be after the window if our peer push more data
680 * than the offered window.
681 */
682static inline u32 tcp_receive_window(const struct tcp_sock *tp)
683{
684 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
685
686 if (win < 0)
687 win = 0;
688 return (u32) win;
689}
690
691/* Choose a new window, without checks for shrinking, and without
692 * scaling applied to the result. The caller does these things
693 * if necessary. This is a "raw" window selection.
694 */
695u32 __tcp_select_window(struct sock *sk);
696
697void tcp_send_window_probe(struct sock *sk);
698
699/* TCP timestamps are only 32-bits, this causes a slight
700 * complication on 64-bit systems since we store a snapshot
701 * of jiffies in the buffer control blocks below. We decided
702 * to use only the low 32-bits of jiffies and hide the ugly
703 * casts with the following macro.
704 */
705#define tcp_time_stamp ((__u32)(jiffies))
706
707static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
708{
709 return skb->skb_mstamp.stamp_jiffies;
710}
711
712
713#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
714
715#define TCPHDR_FIN 0x01
716#define TCPHDR_SYN 0x02
717#define TCPHDR_RST 0x04
718#define TCPHDR_PSH 0x08
719#define TCPHDR_ACK 0x10
720#define TCPHDR_URG 0x20
721#define TCPHDR_ECE 0x40
722#define TCPHDR_CWR 0x80
723
724#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
725
726/* This is what the send packet queuing engine uses to pass
727 * TCP per-packet control information to the transmission code.
728 * We also store the host-order sequence numbers in here too.
729 * This is 44 bytes if IPV6 is enabled.
730 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
731 */
732struct tcp_skb_cb {
733 __u32 seq; /* Starting sequence number */
734 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
735 union {
736 /* Note : tcp_tw_isn is used in input path only
737 * (isn chosen by tcp_timewait_state_process())
738 *
739 * tcp_gso_segs/size are used in write queue only,
740 * cf tcp_skb_pcount()/tcp_skb_mss()
741 */
742 __u32 tcp_tw_isn;
743 struct {
744 u16 tcp_gso_segs;
745 u16 tcp_gso_size;
746 };
747 };
748 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
749
750 __u8 sacked; /* State flags for SACK/FACK. */
751#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
752#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
753#define TCPCB_LOST 0x04 /* SKB is lost */
754#define TCPCB_TAGBITS 0x07 /* All tag bits */
755#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
756#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
757#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
758 TCPCB_REPAIRED)
759
760 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
761 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
762 eor:1, /* Is skb MSG_EOR marked? */
763 unused:6;
764 __u32 ack_seq; /* Sequence number ACK'd */
765 union {
766 struct {
767 /* There is space for up to 24 bytes */
768 __u32 in_flight:30,/* Bytes in flight at transmit */
769 is_app_limited:1, /* cwnd not fully used? */
770 unused:1;
771 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
772 __u32 delivered;
773 /* start of send pipeline phase */
774 struct skb_mstamp first_tx_mstamp;
775 /* when we reached the "delivered" count */
776 struct skb_mstamp delivered_mstamp;
777 } tx; /* only used for outgoing skbs */
778 union {
779 struct inet_skb_parm h4;
780#if IS_ENABLED(CONFIG_IPV6)
781 struct inet6_skb_parm h6;
782#endif
783 } header; /* For incoming skbs */
784 };
785};
786
787#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
788
789
790#if IS_ENABLED(CONFIG_IPV6)
791/* This is the variant of inet6_iif() that must be used by TCP,
792 * as TCP moves IP6CB into a different location in skb->cb[]
793 */
794static inline int tcp_v6_iif(const struct sk_buff *skb)
795{
796 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
797
798 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
799}
800#endif
801
802/* TCP_SKB_CB reference means this can not be used from early demux */
803static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
804{
805#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
806 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
807 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
808 return true;
809#endif
810 return false;
811}
812
813/* Due to TSO, an SKB can be composed of multiple actual
814 * packets. To keep these tracked properly, we use this.
815 */
816static inline int tcp_skb_pcount(const struct sk_buff *skb)
817{
818 return TCP_SKB_CB(skb)->tcp_gso_segs;
819}
820
821static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
822{
823 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
824}
825
826static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
827{
828 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
829}
830
831/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
832static inline int tcp_skb_mss(const struct sk_buff *skb)
833{
834 return TCP_SKB_CB(skb)->tcp_gso_size;
835}
836
837static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
838{
839 return likely(!TCP_SKB_CB(skb)->eor);
840}
841
842/* Events passed to congestion control interface */
843enum tcp_ca_event {
844 CA_EVENT_TX_START, /* first transmit when no packets in flight */
845 CA_EVENT_CWND_RESTART, /* congestion window restart */
846 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
847 CA_EVENT_LOSS, /* loss timeout */
848 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
849 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
850 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
851 CA_EVENT_NON_DELAYED_ACK,
852};
853
854/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
855enum tcp_ca_ack_event_flags {
856 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
857 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
858 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
859};
860
861/*
862 * Interface for adding new TCP congestion control handlers
863 */
864#define TCP_CA_NAME_MAX 16
865#define TCP_CA_MAX 128
866#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
867
868#define TCP_CA_UNSPEC 0
869
870/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
871#define TCP_CONG_NON_RESTRICTED 0x1
872/* Requires ECN/ECT set on all packets */
873#define TCP_CONG_NEEDS_ECN 0x2
874
875union tcp_cc_info;
876
877struct ack_sample {
878 u32 pkts_acked;
879 s32 rtt_us;
880 u32 in_flight;
881};
882
883/* A rate sample measures the number of (original/retransmitted) data
884 * packets delivered "delivered" over an interval of time "interval_us".
885 * The tcp_rate.c code fills in the rate sample, and congestion
886 * control modules that define a cong_control function to run at the end
887 * of ACK processing can optionally chose to consult this sample when
888 * setting cwnd and pacing rate.
889 * A sample is invalid if "delivered" or "interval_us" is negative.
890 */
891struct rate_sample {
892 struct skb_mstamp prior_mstamp; /* starting timestamp for interval */
893 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
894 s32 delivered; /* number of packets delivered over interval */
895 long interval_us; /* time for tp->delivered to incr "delivered" */
896 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
897 int losses; /* number of packets marked lost upon ACK */
898 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
899 u32 prior_in_flight; /* in flight before this ACK */
900 bool is_app_limited; /* is sample from packet with bubble in pipe? */
901 bool is_retrans; /* is sample from retransmission? */
902};
903
904struct tcp_congestion_ops {
905 struct list_head list;
906 u32 key;
907 u32 flags;
908
909 /* initialize private data (optional) */
910 void (*init)(struct sock *sk);
911 /* cleanup private data (optional) */
912 void (*release)(struct sock *sk);
913
914 /* return slow start threshold (required) */
915 u32 (*ssthresh)(struct sock *sk);
916 /* do new cwnd calculation (required) */
917 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
918 /* call before changing ca_state (optional) */
919 void (*set_state)(struct sock *sk, u8 new_state);
920 /* call when cwnd event occurs (optional) */
921 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
922 /* call when ack arrives (optional) */
923 void (*in_ack_event)(struct sock *sk, u32 flags);
924 /* new value of cwnd after loss (optional) */
925 u32 (*undo_cwnd)(struct sock *sk);
926 /* hook for packet ack accounting (optional) */
927 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
928 /* suggest number of segments for each skb to transmit (optional) */
929 u32 (*tso_segs_goal)(struct sock *sk);
930 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
931 u32 (*sndbuf_expand)(struct sock *sk);
932 /* call when packets are delivered to update cwnd and pacing rate,
933 * after all the ca_state processing. (optional)
934 */
935 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
936 /* get info for inet_diag (optional) */
937 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
938 union tcp_cc_info *info);
939
940 char name[TCP_CA_NAME_MAX];
941 struct module *owner;
942};
943
944int tcp_register_congestion_control(struct tcp_congestion_ops *type);
945void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
946
947void tcp_assign_congestion_control(struct sock *sk);
948void tcp_init_congestion_control(struct sock *sk);
949void tcp_cleanup_congestion_control(struct sock *sk);
950int tcp_set_default_congestion_control(const char *name);
951void tcp_get_default_congestion_control(char *name);
952void tcp_get_available_congestion_control(char *buf, size_t len);
953void tcp_get_allowed_congestion_control(char *buf, size_t len);
954int tcp_set_allowed_congestion_control(char *allowed);
955int tcp_set_congestion_control(struct sock *sk, const char *name);
956u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
957void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
958
959u32 tcp_reno_ssthresh(struct sock *sk);
960u32 tcp_reno_undo_cwnd(struct sock *sk);
961void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
962extern struct tcp_congestion_ops tcp_reno;
963
964struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
965u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
966#ifdef CONFIG_INET
967char *tcp_ca_get_name_by_key(u32 key, char *buffer);
968#else
969static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
970{
971 return NULL;
972}
973#endif
974
975static inline bool tcp_ca_needs_ecn(const struct sock *sk)
976{
977 const struct inet_connection_sock *icsk = inet_csk(sk);
978
979 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
980}
981
982static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
983{
984 struct inet_connection_sock *icsk = inet_csk(sk);
985
986 if (icsk->icsk_ca_ops->set_state)
987 icsk->icsk_ca_ops->set_state(sk, ca_state);
988 icsk->icsk_ca_state = ca_state;
989}
990
991static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
992{
993 const struct inet_connection_sock *icsk = inet_csk(sk);
994
995 if (icsk->icsk_ca_ops->cwnd_event)
996 icsk->icsk_ca_ops->cwnd_event(sk, event);
997}
998
999/* From tcp_rate.c */
1000void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1001void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1002 struct rate_sample *rs);
1003void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1004 struct skb_mstamp *now, struct rate_sample *rs);
1005void tcp_rate_check_app_limited(struct sock *sk);
1006
1007/* These functions determine how the current flow behaves in respect of SACK
1008 * handling. SACK is negotiated with the peer, and therefore it can vary
1009 * between different flows.
1010 *
1011 * tcp_is_sack - SACK enabled
1012 * tcp_is_reno - No SACK
1013 * tcp_is_fack - FACK enabled, implies SACK enabled
1014 */
1015static inline int tcp_is_sack(const struct tcp_sock *tp)
1016{
1017 return tp->rx_opt.sack_ok;
1018}
1019
1020static inline bool tcp_is_reno(const struct tcp_sock *tp)
1021{
1022 return !tcp_is_sack(tp);
1023}
1024
1025static inline bool tcp_is_fack(const struct tcp_sock *tp)
1026{
1027 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
1028}
1029
1030static inline void tcp_enable_fack(struct tcp_sock *tp)
1031{
1032 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
1033}
1034
1035/* TCP early-retransmit (ER) is similar to but more conservative than
1036 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
1037 */
1038static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
1039{
1040 struct net *net = sock_net((struct sock *)tp);
1041
1042 tp->do_early_retrans = sysctl_tcp_early_retrans &&
1043 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
1044 net->ipv4.sysctl_tcp_reordering == 3;
1045}
1046
1047static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
1048{
1049 tp->do_early_retrans = 0;
1050}
1051
1052static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1053{
1054 return tp->sacked_out + tp->lost_out;
1055}
1056
1057/* This determines how many packets are "in the network" to the best
1058 * of our knowledge. In many cases it is conservative, but where
1059 * detailed information is available from the receiver (via SACK
1060 * blocks etc.) we can make more aggressive calculations.
1061 *
1062 * Use this for decisions involving congestion control, use just
1063 * tp->packets_out to determine if the send queue is empty or not.
1064 *
1065 * Read this equation as:
1066 *
1067 * "Packets sent once on transmission queue" MINUS
1068 * "Packets left network, but not honestly ACKed yet" PLUS
1069 * "Packets fast retransmitted"
1070 */
1071static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1072{
1073 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1074}
1075
1076#define TCP_INFINITE_SSTHRESH 0x7fffffff
1077
1078static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1079{
1080 return tp->snd_cwnd < tp->snd_ssthresh;
1081}
1082
1083static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1084{
1085 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1086}
1087
1088static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1089{
1090 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1091 (1 << inet_csk(sk)->icsk_ca_state);
1092}
1093
1094/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1095 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1096 * ssthresh.
1097 */
1098static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1099{
1100 const struct tcp_sock *tp = tcp_sk(sk);
1101
1102 if (tcp_in_cwnd_reduction(sk))
1103 return tp->snd_ssthresh;
1104 else
1105 return max(tp->snd_ssthresh,
1106 ((tp->snd_cwnd >> 1) +
1107 (tp->snd_cwnd >> 2)));
1108}
1109
1110/* Use define here intentionally to get WARN_ON location shown at the caller */
1111#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1112
1113void tcp_enter_cwr(struct sock *sk);
1114__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1115
1116/* The maximum number of MSS of available cwnd for which TSO defers
1117 * sending if not using sysctl_tcp_tso_win_divisor.
1118 */
1119static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1120{
1121 return 3;
1122}
1123
1124/* Returns end sequence number of the receiver's advertised window */
1125static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1126{
1127 return tp->snd_una + tp->snd_wnd;
1128}
1129
1130/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1131 * flexible approach. The RFC suggests cwnd should not be raised unless
1132 * it was fully used previously. And that's exactly what we do in
1133 * congestion avoidance mode. But in slow start we allow cwnd to grow
1134 * as long as the application has used half the cwnd.
1135 * Example :
1136 * cwnd is 10 (IW10), but application sends 9 frames.
1137 * We allow cwnd to reach 18 when all frames are ACKed.
1138 * This check is safe because it's as aggressive as slow start which already
1139 * risks 100% overshoot. The advantage is that we discourage application to
1140 * either send more filler packets or data to artificially blow up the cwnd
1141 * usage, and allow application-limited process to probe bw more aggressively.
1142 */
1143static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1144{
1145 const struct tcp_sock *tp = tcp_sk(sk);
1146
1147 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1148 if (tcp_in_slow_start(tp))
1149 return tp->snd_cwnd < 2 * tp->max_packets_out;
1150
1151 return tp->is_cwnd_limited;
1152}
1153
1154/* Something is really bad, we could not queue an additional packet,
1155 * because qdisc is full or receiver sent a 0 window.
1156 * We do not want to add fuel to the fire, or abort too early,
1157 * so make sure the timer we arm now is at least 200ms in the future,
1158 * regardless of current icsk_rto value (as it could be ~2ms)
1159 */
1160static inline unsigned long tcp_probe0_base(const struct sock *sk)
1161{
1162 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1163}
1164
1165/* Variant of inet_csk_rto_backoff() used for zero window probes */
1166static inline unsigned long tcp_probe0_when(const struct sock *sk,
1167 unsigned long max_when)
1168{
1169 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1170
1171 return (unsigned long)min_t(u64, when, max_when);
1172}
1173
1174static inline void tcp_check_probe_timer(struct sock *sk)
1175{
1176 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1177 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1178 tcp_probe0_base(sk), TCP_RTO_MAX);
1179}
1180
1181static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1182{
1183 tp->snd_wl1 = seq;
1184}
1185
1186static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1187{
1188 tp->snd_wl1 = seq;
1189}
1190
1191/*
1192 * Calculate(/check) TCP checksum
1193 */
1194static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1195 __be32 daddr, __wsum base)
1196{
1197 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1198}
1199
1200static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1201{
1202 return __skb_checksum_complete(skb);
1203}
1204
1205static inline bool tcp_checksum_complete(struct sk_buff *skb)
1206{
1207 return !skb_csum_unnecessary(skb) &&
1208 __tcp_checksum_complete(skb);
1209}
1210
1211/* Prequeue for VJ style copy to user, combined with checksumming. */
1212
1213static inline void tcp_prequeue_init(struct tcp_sock *tp)
1214{
1215 tp->ucopy.task = NULL;
1216 tp->ucopy.len = 0;
1217 tp->ucopy.memory = 0;
1218 skb_queue_head_init(&tp->ucopy.prequeue);
1219}
1220
1221bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1222bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1223int tcp_filter(struct sock *sk, struct sk_buff *skb);
1224
1225#undef STATE_TRACE
1226
1227#ifdef STATE_TRACE
1228static const char *statename[]={
1229 "Unused","Established","Syn Sent","Syn Recv",
1230 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1231 "Close Wait","Last ACK","Listen","Closing"
1232};
1233#endif
1234void tcp_set_state(struct sock *sk, int state);
1235
1236void tcp_done(struct sock *sk);
1237
1238int tcp_abort(struct sock *sk, int err);
1239
1240static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1241{
1242 rx_opt->dsack = 0;
1243 rx_opt->num_sacks = 0;
1244}
1245
1246u32 tcp_default_init_rwnd(u32 mss);
1247void tcp_cwnd_restart(struct sock *sk, s32 delta);
1248
1249static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1250{
1251 struct tcp_sock *tp = tcp_sk(sk);
1252 s32 delta;
1253
1254 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1255 return;
1256 delta = tcp_time_stamp - tp->lsndtime;
1257 if (delta > inet_csk(sk)->icsk_rto)
1258 tcp_cwnd_restart(sk, delta);
1259}
1260
1261/* Determine a window scaling and initial window to offer. */
1262void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1263 __u32 *window_clamp, int wscale_ok,
1264 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1265
1266static inline int tcp_win_from_space(int space)
1267{
1268 return sysctl_tcp_adv_win_scale<=0 ?
1269 (space>>(-sysctl_tcp_adv_win_scale)) :
1270 space - (space>>sysctl_tcp_adv_win_scale);
1271}
1272
1273/* Note: caller must be prepared to deal with negative returns */
1274static inline int tcp_space(const struct sock *sk)
1275{
1276 return tcp_win_from_space(sk->sk_rcvbuf -
1277 atomic_read(&sk->sk_rmem_alloc));
1278}
1279
1280static inline int tcp_full_space(const struct sock *sk)
1281{
1282 return tcp_win_from_space(sk->sk_rcvbuf);
1283}
1284
1285extern void tcp_openreq_init_rwin(struct request_sock *req,
1286 const struct sock *sk_listener,
1287 const struct dst_entry *dst);
1288
1289void tcp_enter_memory_pressure(struct sock *sk);
1290
1291static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1292{
1293 struct net *net = sock_net((struct sock *)tp);
1294
1295 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1296}
1297
1298static inline int keepalive_time_when(const struct tcp_sock *tp)
1299{
1300 struct net *net = sock_net((struct sock *)tp);
1301
1302 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1303}
1304
1305static inline int keepalive_probes(const struct tcp_sock *tp)
1306{
1307 struct net *net = sock_net((struct sock *)tp);
1308
1309 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1310}
1311
1312static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1313{
1314 const struct inet_connection_sock *icsk = &tp->inet_conn;
1315
1316 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1317 tcp_time_stamp - tp->rcv_tstamp);
1318}
1319
1320static inline int tcp_fin_time(const struct sock *sk)
1321{
1322 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1323 const int rto = inet_csk(sk)->icsk_rto;
1324
1325 if (fin_timeout < (rto << 2) - (rto >> 1))
1326 fin_timeout = (rto << 2) - (rto >> 1);
1327
1328 return fin_timeout;
1329}
1330
1331static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1332 int paws_win)
1333{
1334 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1335 return true;
1336 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1337 return true;
1338 /*
1339 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1340 * then following tcp messages have valid values. Ignore 0 value,
1341 * or else 'negative' tsval might forbid us to accept their packets.
1342 */
1343 if (!rx_opt->ts_recent)
1344 return true;
1345 return false;
1346}
1347
1348static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1349 int rst)
1350{
1351 if (tcp_paws_check(rx_opt, 0))
1352 return false;
1353
1354 /* RST segments are not recommended to carry timestamp,
1355 and, if they do, it is recommended to ignore PAWS because
1356 "their cleanup function should take precedence over timestamps."
1357 Certainly, it is mistake. It is necessary to understand the reasons
1358 of this constraint to relax it: if peer reboots, clock may go
1359 out-of-sync and half-open connections will not be reset.
1360 Actually, the problem would be not existing if all
1361 the implementations followed draft about maintaining clock
1362 via reboots. Linux-2.2 DOES NOT!
1363
1364 However, we can relax time bounds for RST segments to MSL.
1365 */
1366 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1367 return false;
1368 return true;
1369}
1370
1371bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1372 int mib_idx, u32 *last_oow_ack_time);
1373
1374static inline void tcp_mib_init(struct net *net)
1375{
1376 /* See RFC 2012 */
1377 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1378 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1379 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1380 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1381}
1382
1383/* from STCP */
1384static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1385{
1386 tp->lost_skb_hint = NULL;
1387}
1388
1389static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1390{
1391 tcp_clear_retrans_hints_partial(tp);
1392 tp->retransmit_skb_hint = NULL;
1393}
1394
1395union tcp_md5_addr {
1396 struct in_addr a4;
1397#if IS_ENABLED(CONFIG_IPV6)
1398 struct in6_addr a6;
1399#endif
1400};
1401
1402/* - key database */
1403struct tcp_md5sig_key {
1404 struct hlist_node node;
1405 u8 keylen;
1406 u8 family; /* AF_INET or AF_INET6 */
1407 union tcp_md5_addr addr;
1408 u8 key[TCP_MD5SIG_MAXKEYLEN];
1409 struct rcu_head rcu;
1410};
1411
1412/* - sock block */
1413struct tcp_md5sig_info {
1414 struct hlist_head head;
1415 struct rcu_head rcu;
1416};
1417
1418/* - pseudo header */
1419struct tcp4_pseudohdr {
1420 __be32 saddr;
1421 __be32 daddr;
1422 __u8 pad;
1423 __u8 protocol;
1424 __be16 len;
1425};
1426
1427struct tcp6_pseudohdr {
1428 struct in6_addr saddr;
1429 struct in6_addr daddr;
1430 __be32 len;
1431 __be32 protocol; /* including padding */
1432};
1433
1434union tcp_md5sum_block {
1435 struct tcp4_pseudohdr ip4;
1436#if IS_ENABLED(CONFIG_IPV6)
1437 struct tcp6_pseudohdr ip6;
1438#endif
1439};
1440
1441/* - pool: digest algorithm, hash description and scratch buffer */
1442struct tcp_md5sig_pool {
1443 struct ahash_request *md5_req;
1444 void *scratch;
1445};
1446
1447/* - functions */
1448int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1449 const struct sock *sk, const struct sk_buff *skb);
1450int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1451 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1452int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1453 int family);
1454struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1455 const struct sock *addr_sk);
1456
1457#ifdef CONFIG_TCP_MD5SIG
1458struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1459 const union tcp_md5_addr *addr,
1460 int family);
1461#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1462#else
1463static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1464 const union tcp_md5_addr *addr,
1465 int family)
1466{
1467 return NULL;
1468}
1469#define tcp_twsk_md5_key(twsk) NULL
1470#endif
1471
1472bool tcp_alloc_md5sig_pool(void);
1473
1474struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1475static inline void tcp_put_md5sig_pool(void)
1476{
1477 local_bh_enable();
1478}
1479
1480int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1481 unsigned int header_len);
1482int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1483 const struct tcp_md5sig_key *key);
1484
1485/* From tcp_fastopen.c */
1486void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1487 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1488 unsigned long *last_syn_loss);
1489void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1490 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1491 u16 try_exp);
1492struct tcp_fastopen_request {
1493 /* Fast Open cookie. Size 0 means a cookie request */
1494 struct tcp_fastopen_cookie cookie;
1495 struct msghdr *data; /* data in MSG_FASTOPEN */
1496 size_t size;
1497 int copied; /* queued in tcp_connect() */
1498};
1499void tcp_free_fastopen_req(struct tcp_sock *tp);
1500
1501extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1502int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1503void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1504struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1505 struct request_sock *req,
1506 struct tcp_fastopen_cookie *foc,
1507 struct dst_entry *dst);
1508void tcp_fastopen_init_key_once(bool publish);
1509#define TCP_FASTOPEN_KEY_LENGTH 16
1510
1511/* Fastopen key context */
1512struct tcp_fastopen_context {
1513 struct crypto_cipher *tfm;
1514 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1515 struct rcu_head rcu;
1516};
1517
1518/* Latencies incurred by various limits for a sender. They are
1519 * chronograph-like stats that are mutually exclusive.
1520 */
1521enum tcp_chrono {
1522 TCP_CHRONO_UNSPEC,
1523 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1524 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1525 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1526 __TCP_CHRONO_MAX,
1527};
1528
1529void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1530void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1531
1532/* write queue abstraction */
1533static inline void tcp_write_queue_purge(struct sock *sk)
1534{
1535 struct sk_buff *skb;
1536
1537 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1538 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1539 sk_wmem_free_skb(sk, skb);
1540 sk_mem_reclaim(sk);
1541 tcp_clear_all_retrans_hints(tcp_sk(sk));
1542}
1543
1544static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1545{
1546 return skb_peek(&sk->sk_write_queue);
1547}
1548
1549static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1550{
1551 return skb_peek_tail(&sk->sk_write_queue);
1552}
1553
1554static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1555 const struct sk_buff *skb)
1556{
1557 return skb_queue_next(&sk->sk_write_queue, skb);
1558}
1559
1560static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1561 const struct sk_buff *skb)
1562{
1563 return skb_queue_prev(&sk->sk_write_queue, skb);
1564}
1565
1566#define tcp_for_write_queue(skb, sk) \
1567 skb_queue_walk(&(sk)->sk_write_queue, skb)
1568
1569#define tcp_for_write_queue_from(skb, sk) \
1570 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1571
1572#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1573 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1574
1575static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1576{
1577 return sk->sk_send_head;
1578}
1579
1580static inline bool tcp_skb_is_last(const struct sock *sk,
1581 const struct sk_buff *skb)
1582{
1583 return skb_queue_is_last(&sk->sk_write_queue, skb);
1584}
1585
1586static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1587{
1588 if (tcp_skb_is_last(sk, skb))
1589 sk->sk_send_head = NULL;
1590 else
1591 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1592}
1593
1594static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1595{
1596 if (sk->sk_send_head == skb_unlinked) {
1597 sk->sk_send_head = NULL;
1598 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1599 }
1600 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1601 tcp_sk(sk)->highest_sack = NULL;
1602}
1603
1604static inline void tcp_init_send_head(struct sock *sk)
1605{
1606 sk->sk_send_head = NULL;
1607}
1608
1609static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1610{
1611 __skb_queue_tail(&sk->sk_write_queue, skb);
1612}
1613
1614static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1615{
1616 __tcp_add_write_queue_tail(sk, skb);
1617
1618 /* Queue it, remembering where we must start sending. */
1619 if (sk->sk_send_head == NULL) {
1620 sk->sk_send_head = skb;
1621 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1622
1623 if (tcp_sk(sk)->highest_sack == NULL)
1624 tcp_sk(sk)->highest_sack = skb;
1625 }
1626}
1627
1628static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1629{
1630 __skb_queue_head(&sk->sk_write_queue, skb);
1631}
1632
1633/* Insert buff after skb on the write queue of sk. */
1634static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1635 struct sk_buff *buff,
1636 struct sock *sk)
1637{
1638 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1639}
1640
1641/* Insert new before skb on the write queue of sk. */
1642static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1643 struct sk_buff *skb,
1644 struct sock *sk)
1645{
1646 __skb_queue_before(&sk->sk_write_queue, skb, new);
1647
1648 if (sk->sk_send_head == skb)
1649 sk->sk_send_head = new;
1650}
1651
1652static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1653{
1654 __skb_unlink(skb, &sk->sk_write_queue);
1655}
1656
1657static inline bool tcp_write_queue_empty(struct sock *sk)
1658{
1659 return skb_queue_empty(&sk->sk_write_queue);
1660}
1661
1662static inline void tcp_push_pending_frames(struct sock *sk)
1663{
1664 if (tcp_send_head(sk)) {
1665 struct tcp_sock *tp = tcp_sk(sk);
1666
1667 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1668 }
1669}
1670
1671/* Start sequence of the skb just after the highest skb with SACKed
1672 * bit, valid only if sacked_out > 0 or when the caller has ensured
1673 * validity by itself.
1674 */
1675static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1676{
1677 if (!tp->sacked_out)
1678 return tp->snd_una;
1679
1680 if (tp->highest_sack == NULL)
1681 return tp->snd_nxt;
1682
1683 return TCP_SKB_CB(tp->highest_sack)->seq;
1684}
1685
1686static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1687{
1688 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1689 tcp_write_queue_next(sk, skb);
1690}
1691
1692static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1693{
1694 return tcp_sk(sk)->highest_sack;
1695}
1696
1697static inline void tcp_highest_sack_reset(struct sock *sk)
1698{
1699 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1700}
1701
1702/* Called when old skb is about to be deleted (to be combined with new skb) */
1703static inline void tcp_highest_sack_combine(struct sock *sk,
1704 struct sk_buff *old,
1705 struct sk_buff *new)
1706{
1707 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1708 tcp_sk(sk)->highest_sack = new;
1709}
1710
1711/* This helper checks if socket has IP_TRANSPARENT set */
1712static inline bool inet_sk_transparent(const struct sock *sk)
1713{
1714 switch (sk->sk_state) {
1715 case TCP_TIME_WAIT:
1716 return inet_twsk(sk)->tw_transparent;
1717 case TCP_NEW_SYN_RECV:
1718 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1719 }
1720 return inet_sk(sk)->transparent;
1721}
1722
1723/* Determines whether this is a thin stream (which may suffer from
1724 * increased latency). Used to trigger latency-reducing mechanisms.
1725 */
1726static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1727{
1728 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1729}
1730
1731/* /proc */
1732enum tcp_seq_states {
1733 TCP_SEQ_STATE_LISTENING,
1734 TCP_SEQ_STATE_ESTABLISHED,
1735};
1736
1737int tcp_seq_open(struct inode *inode, struct file *file);
1738
1739struct tcp_seq_afinfo {
1740 char *name;
1741 sa_family_t family;
1742 const struct file_operations *seq_fops;
1743 struct seq_operations seq_ops;
1744};
1745
1746struct tcp_iter_state {
1747 struct seq_net_private p;
1748 sa_family_t family;
1749 enum tcp_seq_states state;
1750 struct sock *syn_wait_sk;
1751 int bucket, offset, sbucket, num;
1752 loff_t last_pos;
1753};
1754
1755int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1756void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1757
1758extern struct request_sock_ops tcp_request_sock_ops;
1759extern struct request_sock_ops tcp6_request_sock_ops;
1760
1761void tcp_v4_destroy_sock(struct sock *sk);
1762
1763struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1764 netdev_features_t features);
1765struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1766int tcp_gro_complete(struct sk_buff *skb);
1767
1768void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1769
1770static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1771{
1772 struct net *net = sock_net((struct sock *)tp);
1773 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1774}
1775
1776static inline bool tcp_stream_memory_free(const struct sock *sk)
1777{
1778 const struct tcp_sock *tp = tcp_sk(sk);
1779 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1780
1781 return notsent_bytes < tcp_notsent_lowat(tp);
1782}
1783
1784#ifdef CONFIG_PROC_FS
1785int tcp4_proc_init(void);
1786void tcp4_proc_exit(void);
1787#endif
1788
1789int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1790int tcp_conn_request(struct request_sock_ops *rsk_ops,
1791 const struct tcp_request_sock_ops *af_ops,
1792 struct sock *sk, struct sk_buff *skb);
1793
1794/* TCP af-specific functions */
1795struct tcp_sock_af_ops {
1796#ifdef CONFIG_TCP_MD5SIG
1797 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1798 const struct sock *addr_sk);
1799 int (*calc_md5_hash)(char *location,
1800 const struct tcp_md5sig_key *md5,
1801 const struct sock *sk,
1802 const struct sk_buff *skb);
1803 int (*md5_parse)(struct sock *sk,
1804 char __user *optval,
1805 int optlen);
1806#endif
1807};
1808
1809struct tcp_request_sock_ops {
1810 u16 mss_clamp;
1811#ifdef CONFIG_TCP_MD5SIG
1812 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1813 const struct sock *addr_sk);
1814 int (*calc_md5_hash) (char *location,
1815 const struct tcp_md5sig_key *md5,
1816 const struct sock *sk,
1817 const struct sk_buff *skb);
1818#endif
1819 void (*init_req)(struct request_sock *req,
1820 const struct sock *sk_listener,
1821 struct sk_buff *skb);
1822#ifdef CONFIG_SYN_COOKIES
1823 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1824 __u16 *mss);
1825#endif
1826 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1827 const struct request_sock *req,
1828 bool *strict);
1829 __u32 (*init_seq)(const struct sk_buff *skb, u32 *tsoff);
1830 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1831 struct flowi *fl, struct request_sock *req,
1832 struct tcp_fastopen_cookie *foc,
1833 enum tcp_synack_type synack_type);
1834};
1835
1836#ifdef CONFIG_SYN_COOKIES
1837static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1838 const struct sock *sk, struct sk_buff *skb,
1839 __u16 *mss)
1840{
1841 tcp_synq_overflow(sk);
1842 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1843 return ops->cookie_init_seq(skb, mss);
1844}
1845#else
1846static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1847 const struct sock *sk, struct sk_buff *skb,
1848 __u16 *mss)
1849{
1850 return 0;
1851}
1852#endif
1853
1854int tcpv4_offload_init(void);
1855
1856void tcp_v4_init(void);
1857void tcp_init(void);
1858
1859/* tcp_recovery.c */
1860
1861/* Flags to enable various loss recovery features. See below */
1862extern int sysctl_tcp_recovery;
1863
1864/* Use TCP RACK to detect (some) tail and retransmit losses */
1865#define TCP_RACK_LOST_RETRANS 0x1
1866
1867extern int tcp_rack_mark_lost(struct sock *sk);
1868
1869extern void tcp_rack_advance(struct tcp_sock *tp,
1870 const struct skb_mstamp *xmit_time, u8 sacked);
1871
1872/*
1873 * Save and compile IPv4 options, return a pointer to it
1874 */
1875static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1876{
1877 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1878 struct ip_options_rcu *dopt = NULL;
1879
1880 if (opt->optlen) {
1881 int opt_size = sizeof(*dopt) + opt->optlen;
1882
1883 dopt = kmalloc(opt_size, GFP_ATOMIC);
1884 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1885 kfree(dopt);
1886 dopt = NULL;
1887 }
1888 }
1889 return dopt;
1890}
1891
1892/* locally generated TCP pure ACKs have skb->truesize == 2
1893 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1894 * This is much faster than dissecting the packet to find out.
1895 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1896 */
1897static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1898{
1899 return skb->truesize == 2;
1900}
1901
1902static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1903{
1904 skb->truesize = 2;
1905}
1906
1907static inline int tcp_inq(struct sock *sk)
1908{
1909 struct tcp_sock *tp = tcp_sk(sk);
1910 int answ;
1911
1912 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1913 answ = 0;
1914 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1915 !tp->urg_data ||
1916 before(tp->urg_seq, tp->copied_seq) ||
1917 !before(tp->urg_seq, tp->rcv_nxt)) {
1918
1919 answ = tp->rcv_nxt - tp->copied_seq;
1920
1921 /* Subtract 1, if FIN was received */
1922 if (answ && sock_flag(sk, SOCK_DONE))
1923 answ--;
1924 } else {
1925 answ = tp->urg_seq - tp->copied_seq;
1926 }
1927
1928 return answ;
1929}
1930
1931int tcp_peek_len(struct socket *sock);
1932
1933static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1934{
1935 u16 segs_in;
1936
1937 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1938 tp->segs_in += segs_in;
1939 if (skb->len > tcp_hdrlen(skb))
1940 tp->data_segs_in += segs_in;
1941}
1942
1943/*
1944 * TCP listen path runs lockless.
1945 * We forced "struct sock" to be const qualified to make sure
1946 * we don't modify one of its field by mistake.
1947 * Here, we increment sk_drops which is an atomic_t, so we can safely
1948 * make sock writable again.
1949 */
1950static inline void tcp_listendrop(const struct sock *sk)
1951{
1952 atomic_inc(&((struct sock *)sk)->sk_drops);
1953 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
1954}
1955
1956#endif /* _TCP_H */