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 TCP_DEBUG 1
22#define FASTRETRANS_DEBUG 1
23
24#include <linux/list.h>
25#include <linux/tcp.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
35#include <net/inet_connection_sock.h>
36#include <net/inet_timewait_sock.h>
37#include <net/inet_hashtables.h>
38#include <net/checksum.h>
39#include <net/request_sock.h>
40#include <net/sock.h>
41#include <net/snmp.h>
42#include <net/ip.h>
43#include <net/tcp_states.h>
44#include <net/inet_ecn.h>
45#include <net/dst.h>
46
47#include <linux/seq_file.h>
48
49extern struct inet_hashinfo tcp_hashinfo;
50
51extern struct percpu_counter tcp_orphan_count;
52extern void 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/* Offer an initial receive window of 10 mss. */
64#define TCP_DEFAULT_INIT_RCVWND 10
65
66/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
67#define TCP_MIN_MSS 88U
68
69/* The least MTU to use for probing */
70#define TCP_BASE_MSS 512
71
72/* After receiving this amount of duplicate ACKs fast retransmit starts. */
73#define TCP_FASTRETRANS_THRESH 3
74
75/* Maximal reordering. */
76#define TCP_MAX_REORDERING 127
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 5 /* number of times to retry active opening a
101 * connection: ~180sec is RFC minimum */
102
103#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
104 * connection: ~180sec is RFC minimum */
105
106#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
107 * state, about 60 seconds */
108#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
109 /* BSD style FIN_WAIT2 deadlock breaker.
110 * It used to be 3min, new value is 60sec,
111 * to combine FIN-WAIT-2 timeout with
112 * TIME-WAIT timer.
113 */
114
115#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
116#if HZ >= 100
117#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
118#define TCP_ATO_MIN ((unsigned)(HZ/25))
119#else
120#define TCP_DELACK_MIN 4U
121#define TCP_ATO_MIN 4U
122#endif
123#define TCP_RTO_MAX ((unsigned)(120*HZ))
124#define TCP_RTO_MIN ((unsigned)(HZ/5))
125#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC2988bis initial RTO value */
126#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
127 * used as a fallback RTO for the
128 * initial data transmission if no
129 * valid RTT sample has been acquired,
130 * most likely due to retrans in 3WHS.
131 */
132
133#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
134 * for local resources.
135 */
136
137#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
138#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
139#define TCP_KEEPALIVE_INTVL (75*HZ)
140
141#define MAX_TCP_KEEPIDLE 32767
142#define MAX_TCP_KEEPINTVL 32767
143#define MAX_TCP_KEEPCNT 127
144#define MAX_TCP_SYNCNT 127
145
146#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
147
148#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
149#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
150 * after this time. It should be equal
151 * (or greater than) TCP_TIMEWAIT_LEN
152 * to provide reliability equal to one
153 * provided by timewait state.
154 */
155#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
156 * timestamps. It must be less than
157 * minimal timewait lifetime.
158 */
159/*
160 * TCP option
161 */
162
163#define TCPOPT_NOP 1 /* Padding */
164#define TCPOPT_EOL 0 /* End of options */
165#define TCPOPT_MSS 2 /* Segment size negotiating */
166#define TCPOPT_WINDOW 3 /* Window scaling */
167#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
168#define TCPOPT_SACK 5 /* SACK Block */
169#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
170#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
171#define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
172
173/*
174 * TCP option lengths
175 */
176
177#define TCPOLEN_MSS 4
178#define TCPOLEN_WINDOW 3
179#define TCPOLEN_SACK_PERM 2
180#define TCPOLEN_TIMESTAMP 10
181#define TCPOLEN_MD5SIG 18
182#define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
183#define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
184#define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
185#define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
186
187/* But this is what stacks really send out. */
188#define TCPOLEN_TSTAMP_ALIGNED 12
189#define TCPOLEN_WSCALE_ALIGNED 4
190#define TCPOLEN_SACKPERM_ALIGNED 4
191#define TCPOLEN_SACK_BASE 2
192#define TCPOLEN_SACK_BASE_ALIGNED 4
193#define TCPOLEN_SACK_PERBLOCK 8
194#define TCPOLEN_MD5SIG_ALIGNED 20
195#define TCPOLEN_MSS_ALIGNED 4
196
197/* Flags in tp->nonagle */
198#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
199#define TCP_NAGLE_CORK 2 /* Socket is corked */
200#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
201
202/* TCP thin-stream limits */
203#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
204
205/* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
206#define TCP_INIT_CWND 10
207
208extern struct inet_timewait_death_row tcp_death_row;
209
210/* sysctl variables for tcp */
211extern int sysctl_tcp_timestamps;
212extern int sysctl_tcp_window_scaling;
213extern int sysctl_tcp_sack;
214extern int sysctl_tcp_fin_timeout;
215extern int sysctl_tcp_keepalive_time;
216extern int sysctl_tcp_keepalive_probes;
217extern int sysctl_tcp_keepalive_intvl;
218extern int sysctl_tcp_syn_retries;
219extern int sysctl_tcp_synack_retries;
220extern int sysctl_tcp_retries1;
221extern int sysctl_tcp_retries2;
222extern int sysctl_tcp_orphan_retries;
223extern int sysctl_tcp_syncookies;
224extern int sysctl_tcp_retrans_collapse;
225extern int sysctl_tcp_stdurg;
226extern int sysctl_tcp_rfc1337;
227extern int sysctl_tcp_abort_on_overflow;
228extern int sysctl_tcp_max_orphans;
229extern int sysctl_tcp_fack;
230extern int sysctl_tcp_reordering;
231extern int sysctl_tcp_ecn;
232extern int sysctl_tcp_dsack;
233extern long sysctl_tcp_mem[3];
234extern int sysctl_tcp_wmem[3];
235extern int sysctl_tcp_rmem[3];
236extern int sysctl_tcp_app_win;
237extern int sysctl_tcp_adv_win_scale;
238extern int sysctl_tcp_tw_reuse;
239extern int sysctl_tcp_frto;
240extern int sysctl_tcp_frto_response;
241extern int sysctl_tcp_low_latency;
242extern int sysctl_tcp_dma_copybreak;
243extern int sysctl_tcp_nometrics_save;
244extern int sysctl_tcp_moderate_rcvbuf;
245extern int sysctl_tcp_tso_win_divisor;
246extern int sysctl_tcp_abc;
247extern int sysctl_tcp_mtu_probing;
248extern int sysctl_tcp_base_mss;
249extern int sysctl_tcp_workaround_signed_windows;
250extern int sysctl_tcp_slow_start_after_idle;
251extern int sysctl_tcp_max_ssthresh;
252extern int sysctl_tcp_cookie_size;
253extern int sysctl_tcp_thin_linear_timeouts;
254extern int sysctl_tcp_thin_dupack;
255
256extern atomic_long_t tcp_memory_allocated;
257extern struct percpu_counter tcp_sockets_allocated;
258extern int tcp_memory_pressure;
259
260/*
261 * The next routines deal with comparing 32 bit unsigned ints
262 * and worry about wraparound (automatic with unsigned arithmetic).
263 */
264
265static inline int before(__u32 seq1, __u32 seq2)
266{
267 return (__s32)(seq1-seq2) < 0;
268}
269#define after(seq2, seq1) before(seq1, seq2)
270
271/* is s2<=s1<=s3 ? */
272static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
273{
274 return seq3 - seq2 >= seq1 - seq2;
275}
276
277static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
278{
279 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
280 int orphans = percpu_counter_read_positive(ocp);
281
282 if (orphans << shift > sysctl_tcp_max_orphans) {
283 orphans = percpu_counter_sum_positive(ocp);
284 if (orphans << shift > sysctl_tcp_max_orphans)
285 return true;
286 }
287
288 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
289 atomic_long_read(&tcp_memory_allocated) > sysctl_tcp_mem[2])
290 return true;
291 return false;
292}
293
294/* syncookies: remember time of last synqueue overflow */
295static inline void tcp_synq_overflow(struct sock *sk)
296{
297 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
298}
299
300/* syncookies: no recent synqueue overflow on this listening socket? */
301static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
302{
303 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
304 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
305}
306
307extern struct proto tcp_prot;
308
309#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
310#define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
311#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
312#define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
313#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
314
315extern void tcp_v4_err(struct sk_buff *skb, u32);
316
317extern void tcp_shutdown (struct sock *sk, int how);
318
319extern int tcp_v4_rcv(struct sk_buff *skb);
320
321extern struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it);
322extern void *tcp_v4_tw_get_peer(struct sock *sk);
323extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
324extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
325 size_t size);
326extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
327 size_t size, int flags);
328extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
329extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
330 struct tcphdr *th, unsigned len);
331extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
332 struct tcphdr *th, unsigned len);
333extern void tcp_rcv_space_adjust(struct sock *sk);
334extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
335extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
336extern void tcp_twsk_destructor(struct sock *sk);
337extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
338 struct pipe_inode_info *pipe, size_t len,
339 unsigned int flags);
340
341static inline void tcp_dec_quickack_mode(struct sock *sk,
342 const unsigned int pkts)
343{
344 struct inet_connection_sock *icsk = inet_csk(sk);
345
346 if (icsk->icsk_ack.quick) {
347 if (pkts >= icsk->icsk_ack.quick) {
348 icsk->icsk_ack.quick = 0;
349 /* Leaving quickack mode we deflate ATO. */
350 icsk->icsk_ack.ato = TCP_ATO_MIN;
351 } else
352 icsk->icsk_ack.quick -= pkts;
353 }
354}
355
356#define TCP_ECN_OK 1
357#define TCP_ECN_QUEUE_CWR 2
358#define TCP_ECN_DEMAND_CWR 4
359
360static __inline__ void
361TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
362{
363 if (sysctl_tcp_ecn && th->ece && th->cwr)
364 inet_rsk(req)->ecn_ok = 1;
365}
366
367enum tcp_tw_status {
368 TCP_TW_SUCCESS = 0,
369 TCP_TW_RST = 1,
370 TCP_TW_ACK = 2,
371 TCP_TW_SYN = 3
372};
373
374
375extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
376 struct sk_buff *skb,
377 const struct tcphdr *th);
378extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
379 struct request_sock *req,
380 struct request_sock **prev);
381extern int tcp_child_process(struct sock *parent, struct sock *child,
382 struct sk_buff *skb);
383extern int tcp_use_frto(struct sock *sk);
384extern void tcp_enter_frto(struct sock *sk);
385extern void tcp_enter_loss(struct sock *sk, int how);
386extern void tcp_clear_retrans(struct tcp_sock *tp);
387extern void tcp_update_metrics(struct sock *sk);
388extern void tcp_close(struct sock *sk, long timeout);
389extern unsigned int tcp_poll(struct file * file, struct socket *sock,
390 struct poll_table_struct *wait);
391extern int tcp_getsockopt(struct sock *sk, int level, int optname,
392 char __user *optval, int __user *optlen);
393extern int tcp_setsockopt(struct sock *sk, int level, int optname,
394 char __user *optval, unsigned int optlen);
395extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
396 char __user *optval, int __user *optlen);
397extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
398 char __user *optval, unsigned int optlen);
399extern void tcp_set_keepalive(struct sock *sk, int val);
400extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
401extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
402 size_t len, int nonblock, int flags, int *addr_len);
403extern void tcp_parse_options(struct sk_buff *skb,
404 struct tcp_options_received *opt_rx, u8 **hvpp,
405 int estab);
406extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
407
408/*
409 * TCP v4 functions exported for the inet6 API
410 */
411
412extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
413extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
414extern struct sock * tcp_create_openreq_child(struct sock *sk,
415 struct request_sock *req,
416 struct sk_buff *skb);
417extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
418 struct request_sock *req,
419 struct dst_entry *dst);
420extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
421extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
422 int addr_len);
423extern int tcp_connect(struct sock *sk);
424extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
425 struct request_sock *req,
426 struct request_values *rvp);
427extern int tcp_disconnect(struct sock *sk, int flags);
428
429
430/* From syncookies.c */
431extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
432extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
433 struct ip_options *opt);
434#ifdef CONFIG_SYN_COOKIES
435extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
436 __u16 *mss);
437#else
438static inline __u32 cookie_v4_init_sequence(struct sock *sk,
439 struct sk_buff *skb,
440 __u16 *mss)
441{
442 return 0;
443}
444#endif
445
446extern __u32 cookie_init_timestamp(struct request_sock *req);
447extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
448
449/* From net/ipv6/syncookies.c */
450extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
451#ifdef CONFIG_SYN_COOKIES
452extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
453 __u16 *mss);
454#else
455static inline __u32 cookie_v6_init_sequence(struct sock *sk,
456 struct sk_buff *skb,
457 __u16 *mss)
458{
459 return 0;
460}
461#endif
462/* tcp_output.c */
463
464extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
465 int nonagle);
466extern int tcp_may_send_now(struct sock *sk);
467extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
468extern void tcp_retransmit_timer(struct sock *sk);
469extern void tcp_xmit_retransmit_queue(struct sock *);
470extern void tcp_simple_retransmit(struct sock *);
471extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
472extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
473
474extern void tcp_send_probe0(struct sock *);
475extern void tcp_send_partial(struct sock *);
476extern int tcp_write_wakeup(struct sock *);
477extern void tcp_send_fin(struct sock *sk);
478extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
479extern int tcp_send_synack(struct sock *);
480extern int tcp_syn_flood_action(struct sock *sk,
481 const struct sk_buff *skb,
482 const char *proto);
483extern void tcp_push_one(struct sock *, unsigned int mss_now);
484extern void tcp_send_ack(struct sock *sk);
485extern void tcp_send_delayed_ack(struct sock *sk);
486
487/* tcp_input.c */
488extern void tcp_cwnd_application_limited(struct sock *sk);
489
490/* tcp_timer.c */
491extern void tcp_init_xmit_timers(struct sock *);
492static inline void tcp_clear_xmit_timers(struct sock *sk)
493{
494 inet_csk_clear_xmit_timers(sk);
495}
496
497extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
498extern unsigned int tcp_current_mss(struct sock *sk);
499
500/* Bound MSS / TSO packet size with the half of the window */
501static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
502{
503 int cutoff;
504
505 /* When peer uses tiny windows, there is no use in packetizing
506 * to sub-MSS pieces for the sake of SWS or making sure there
507 * are enough packets in the pipe for fast recovery.
508 *
509 * On the other hand, for extremely large MSS devices, handling
510 * smaller than MSS windows in this way does make sense.
511 */
512 if (tp->max_window >= 512)
513 cutoff = (tp->max_window >> 1);
514 else
515 cutoff = tp->max_window;
516
517 if (cutoff && pktsize > cutoff)
518 return max_t(int, cutoff, 68U - tp->tcp_header_len);
519 else
520 return pktsize;
521}
522
523/* tcp.c */
524extern void tcp_get_info(struct sock *, struct tcp_info *);
525
526/* Read 'sendfile()'-style from a TCP socket */
527typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
528 unsigned int, size_t);
529extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
530 sk_read_actor_t recv_actor);
531
532extern void tcp_initialize_rcv_mss(struct sock *sk);
533
534extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
535extern int tcp_mss_to_mtu(struct sock *sk, int mss);
536extern void tcp_mtup_init(struct sock *sk);
537extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
538
539static inline void tcp_bound_rto(const struct sock *sk)
540{
541 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
542 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
543}
544
545static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
546{
547 return (tp->srtt >> 3) + tp->rttvar;
548}
549
550static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
551{
552 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
553 ntohl(TCP_FLAG_ACK) |
554 snd_wnd);
555}
556
557static inline void tcp_fast_path_on(struct tcp_sock *tp)
558{
559 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
560}
561
562static inline void tcp_fast_path_check(struct sock *sk)
563{
564 struct tcp_sock *tp = tcp_sk(sk);
565
566 if (skb_queue_empty(&tp->out_of_order_queue) &&
567 tp->rcv_wnd &&
568 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
569 !tp->urg_data)
570 tcp_fast_path_on(tp);
571}
572
573/* Compute the actual rto_min value */
574static inline u32 tcp_rto_min(struct sock *sk)
575{
576 struct dst_entry *dst = __sk_dst_get(sk);
577 u32 rto_min = TCP_RTO_MIN;
578
579 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
580 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
581 return rto_min;
582}
583
584/* Compute the actual receive window we are currently advertising.
585 * Rcv_nxt can be after the window if our peer push more data
586 * than the offered window.
587 */
588static inline u32 tcp_receive_window(const struct tcp_sock *tp)
589{
590 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
591
592 if (win < 0)
593 win = 0;
594 return (u32) win;
595}
596
597/* Choose a new window, without checks for shrinking, and without
598 * scaling applied to the result. The caller does these things
599 * if necessary. This is a "raw" window selection.
600 */
601extern u32 __tcp_select_window(struct sock *sk);
602
603/* TCP timestamps are only 32-bits, this causes a slight
604 * complication on 64-bit systems since we store a snapshot
605 * of jiffies in the buffer control blocks below. We decided
606 * to use only the low 32-bits of jiffies and hide the ugly
607 * casts with the following macro.
608 */
609#define tcp_time_stamp ((__u32)(jiffies))
610
611#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
612
613#define TCPHDR_FIN 0x01
614#define TCPHDR_SYN 0x02
615#define TCPHDR_RST 0x04
616#define TCPHDR_PSH 0x08
617#define TCPHDR_ACK 0x10
618#define TCPHDR_URG 0x20
619#define TCPHDR_ECE 0x40
620#define TCPHDR_CWR 0x80
621
622/* This is what the send packet queuing engine uses to pass
623 * TCP per-packet control information to the transmission code.
624 * We also store the host-order sequence numbers in here too.
625 * This is 44 bytes if IPV6 is enabled.
626 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
627 */
628struct tcp_skb_cb {
629 union {
630 struct inet_skb_parm h4;
631#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
632 struct inet6_skb_parm h6;
633#endif
634 } header; /* For incoming frames */
635 __u32 seq; /* Starting sequence number */
636 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
637 __u32 when; /* used to compute rtt's */
638 __u8 flags; /* TCP header flags. */
639 __u8 sacked; /* State flags for SACK/FACK. */
640#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
641#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
642#define TCPCB_LOST 0x04 /* SKB is lost */
643#define TCPCB_TAGBITS 0x07 /* All tag bits */
644
645#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
646#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
647
648 __u32 ack_seq; /* Sequence number ACK'd */
649};
650
651#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
652
653/* Due to TSO, an SKB can be composed of multiple actual
654 * packets. To keep these tracked properly, we use this.
655 */
656static inline int tcp_skb_pcount(const struct sk_buff *skb)
657{
658 return skb_shinfo(skb)->gso_segs;
659}
660
661/* This is valid iff tcp_skb_pcount() > 1. */
662static inline int tcp_skb_mss(const struct sk_buff *skb)
663{
664 return skb_shinfo(skb)->gso_size;
665}
666
667/* Events passed to congestion control interface */
668enum tcp_ca_event {
669 CA_EVENT_TX_START, /* first transmit when no packets in flight */
670 CA_EVENT_CWND_RESTART, /* congestion window restart */
671 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
672 CA_EVENT_FRTO, /* fast recovery timeout */
673 CA_EVENT_LOSS, /* loss timeout */
674 CA_EVENT_FAST_ACK, /* in sequence ack */
675 CA_EVENT_SLOW_ACK, /* other ack */
676};
677
678/*
679 * Interface for adding new TCP congestion control handlers
680 */
681#define TCP_CA_NAME_MAX 16
682#define TCP_CA_MAX 128
683#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
684
685#define TCP_CONG_NON_RESTRICTED 0x1
686#define TCP_CONG_RTT_STAMP 0x2
687
688struct tcp_congestion_ops {
689 struct list_head list;
690 unsigned long flags;
691
692 /* initialize private data (optional) */
693 void (*init)(struct sock *sk);
694 /* cleanup private data (optional) */
695 void (*release)(struct sock *sk);
696
697 /* return slow start threshold (required) */
698 u32 (*ssthresh)(struct sock *sk);
699 /* lower bound for congestion window (optional) */
700 u32 (*min_cwnd)(const struct sock *sk);
701 /* do new cwnd calculation (required) */
702 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
703 /* call before changing ca_state (optional) */
704 void (*set_state)(struct sock *sk, u8 new_state);
705 /* call when cwnd event occurs (optional) */
706 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
707 /* new value of cwnd after loss (optional) */
708 u32 (*undo_cwnd)(struct sock *sk);
709 /* hook for packet ack accounting (optional) */
710 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
711 /* get info for inet_diag (optional) */
712 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
713
714 char name[TCP_CA_NAME_MAX];
715 struct module *owner;
716};
717
718extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
719extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
720
721extern void tcp_init_congestion_control(struct sock *sk);
722extern void tcp_cleanup_congestion_control(struct sock *sk);
723extern int tcp_set_default_congestion_control(const char *name);
724extern void tcp_get_default_congestion_control(char *name);
725extern void tcp_get_available_congestion_control(char *buf, size_t len);
726extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
727extern int tcp_set_allowed_congestion_control(char *allowed);
728extern int tcp_set_congestion_control(struct sock *sk, const char *name);
729extern void tcp_slow_start(struct tcp_sock *tp);
730extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
731
732extern struct tcp_congestion_ops tcp_init_congestion_ops;
733extern u32 tcp_reno_ssthresh(struct sock *sk);
734extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
735extern u32 tcp_reno_min_cwnd(const struct sock *sk);
736extern struct tcp_congestion_ops tcp_reno;
737
738static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
739{
740 struct inet_connection_sock *icsk = inet_csk(sk);
741
742 if (icsk->icsk_ca_ops->set_state)
743 icsk->icsk_ca_ops->set_state(sk, ca_state);
744 icsk->icsk_ca_state = ca_state;
745}
746
747static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
748{
749 const struct inet_connection_sock *icsk = inet_csk(sk);
750
751 if (icsk->icsk_ca_ops->cwnd_event)
752 icsk->icsk_ca_ops->cwnd_event(sk, event);
753}
754
755/* These functions determine how the current flow behaves in respect of SACK
756 * handling. SACK is negotiated with the peer, and therefore it can vary
757 * between different flows.
758 *
759 * tcp_is_sack - SACK enabled
760 * tcp_is_reno - No SACK
761 * tcp_is_fack - FACK enabled, implies SACK enabled
762 */
763static inline int tcp_is_sack(const struct tcp_sock *tp)
764{
765 return tp->rx_opt.sack_ok;
766}
767
768static inline int tcp_is_reno(const struct tcp_sock *tp)
769{
770 return !tcp_is_sack(tp);
771}
772
773static inline int tcp_is_fack(const struct tcp_sock *tp)
774{
775 return tp->rx_opt.sack_ok & 2;
776}
777
778static inline void tcp_enable_fack(struct tcp_sock *tp)
779{
780 tp->rx_opt.sack_ok |= 2;
781}
782
783static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
784{
785 return tp->sacked_out + tp->lost_out;
786}
787
788/* This determines how many packets are "in the network" to the best
789 * of our knowledge. In many cases it is conservative, but where
790 * detailed information is available from the receiver (via SACK
791 * blocks etc.) we can make more aggressive calculations.
792 *
793 * Use this for decisions involving congestion control, use just
794 * tp->packets_out to determine if the send queue is empty or not.
795 *
796 * Read this equation as:
797 *
798 * "Packets sent once on transmission queue" MINUS
799 * "Packets left network, but not honestly ACKed yet" PLUS
800 * "Packets fast retransmitted"
801 */
802static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
803{
804 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
805}
806
807#define TCP_INFINITE_SSTHRESH 0x7fffffff
808
809static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
810{
811 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
812}
813
814/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
815 * The exception is rate halving phase, when cwnd is decreasing towards
816 * ssthresh.
817 */
818static inline __u32 tcp_current_ssthresh(const struct sock *sk)
819{
820 const struct tcp_sock *tp = tcp_sk(sk);
821 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
822 return tp->snd_ssthresh;
823 else
824 return max(tp->snd_ssthresh,
825 ((tp->snd_cwnd >> 1) +
826 (tp->snd_cwnd >> 2)));
827}
828
829/* Use define here intentionally to get WARN_ON location shown at the caller */
830#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
831
832extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
833extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
834
835/* Slow start with delack produces 3 packets of burst, so that
836 * it is safe "de facto". This will be the default - same as
837 * the default reordering threshold - but if reordering increases,
838 * we must be able to allow cwnd to burst at least this much in order
839 * to not pull it back when holes are filled.
840 */
841static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
842{
843 return tp->reordering;
844}
845
846/* Returns end sequence number of the receiver's advertised window */
847static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
848{
849 return tp->snd_una + tp->snd_wnd;
850}
851extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
852
853static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
854 const struct sk_buff *skb)
855{
856 if (skb->len < mss)
857 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
858}
859
860static inline void tcp_check_probe_timer(struct sock *sk)
861{
862 struct tcp_sock *tp = tcp_sk(sk);
863 const struct inet_connection_sock *icsk = inet_csk(sk);
864
865 if (!tp->packets_out && !icsk->icsk_pending)
866 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
867 icsk->icsk_rto, TCP_RTO_MAX);
868}
869
870static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
871{
872 tp->snd_wl1 = seq;
873}
874
875static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
876{
877 tp->snd_wl1 = seq;
878}
879
880/*
881 * Calculate(/check) TCP checksum
882 */
883static inline __sum16 tcp_v4_check(int len, __be32 saddr,
884 __be32 daddr, __wsum base)
885{
886 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
887}
888
889static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
890{
891 return __skb_checksum_complete(skb);
892}
893
894static inline int tcp_checksum_complete(struct sk_buff *skb)
895{
896 return !skb_csum_unnecessary(skb) &&
897 __tcp_checksum_complete(skb);
898}
899
900/* Prequeue for VJ style copy to user, combined with checksumming. */
901
902static inline void tcp_prequeue_init(struct tcp_sock *tp)
903{
904 tp->ucopy.task = NULL;
905 tp->ucopy.len = 0;
906 tp->ucopy.memory = 0;
907 skb_queue_head_init(&tp->ucopy.prequeue);
908#ifdef CONFIG_NET_DMA
909 tp->ucopy.dma_chan = NULL;
910 tp->ucopy.wakeup = 0;
911 tp->ucopy.pinned_list = NULL;
912 tp->ucopy.dma_cookie = 0;
913#endif
914}
915
916/* Packet is added to VJ-style prequeue for processing in process
917 * context, if a reader task is waiting. Apparently, this exciting
918 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
919 * failed somewhere. Latency? Burstiness? Well, at least now we will
920 * see, why it failed. 8)8) --ANK
921 *
922 * NOTE: is this not too big to inline?
923 */
924static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
925{
926 struct tcp_sock *tp = tcp_sk(sk);
927
928 if (sysctl_tcp_low_latency || !tp->ucopy.task)
929 return 0;
930
931 __skb_queue_tail(&tp->ucopy.prequeue, skb);
932 tp->ucopy.memory += skb->truesize;
933 if (tp->ucopy.memory > sk->sk_rcvbuf) {
934 struct sk_buff *skb1;
935
936 BUG_ON(sock_owned_by_user(sk));
937
938 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
939 sk_backlog_rcv(sk, skb1);
940 NET_INC_STATS_BH(sock_net(sk),
941 LINUX_MIB_TCPPREQUEUEDROPPED);
942 }
943
944 tp->ucopy.memory = 0;
945 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
946 wake_up_interruptible_sync_poll(sk_sleep(sk),
947 POLLIN | POLLRDNORM | POLLRDBAND);
948 if (!inet_csk_ack_scheduled(sk))
949 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
950 (3 * tcp_rto_min(sk)) / 4,
951 TCP_RTO_MAX);
952 }
953 return 1;
954}
955
956
957#undef STATE_TRACE
958
959#ifdef STATE_TRACE
960static const char *statename[]={
961 "Unused","Established","Syn Sent","Syn Recv",
962 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
963 "Close Wait","Last ACK","Listen","Closing"
964};
965#endif
966extern void tcp_set_state(struct sock *sk, int state);
967
968extern void tcp_done(struct sock *sk);
969
970static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
971{
972 rx_opt->dsack = 0;
973 rx_opt->num_sacks = 0;
974}
975
976/* Determine a window scaling and initial window to offer. */
977extern void tcp_select_initial_window(int __space, __u32 mss,
978 __u32 *rcv_wnd, __u32 *window_clamp,
979 int wscale_ok, __u8 *rcv_wscale,
980 __u32 init_rcv_wnd);
981
982static inline int tcp_win_from_space(int space)
983{
984 return sysctl_tcp_adv_win_scale<=0 ?
985 (space>>(-sysctl_tcp_adv_win_scale)) :
986 space - (space>>sysctl_tcp_adv_win_scale);
987}
988
989/* Note: caller must be prepared to deal with negative returns */
990static inline int tcp_space(const struct sock *sk)
991{
992 return tcp_win_from_space(sk->sk_rcvbuf -
993 atomic_read(&sk->sk_rmem_alloc));
994}
995
996static inline int tcp_full_space(const struct sock *sk)
997{
998 return tcp_win_from_space(sk->sk_rcvbuf);
999}
1000
1001static inline void tcp_openreq_init(struct request_sock *req,
1002 struct tcp_options_received *rx_opt,
1003 struct sk_buff *skb)
1004{
1005 struct inet_request_sock *ireq = inet_rsk(req);
1006
1007 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1008 req->cookie_ts = 0;
1009 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1010 req->mss = rx_opt->mss_clamp;
1011 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1012 ireq->tstamp_ok = rx_opt->tstamp_ok;
1013 ireq->sack_ok = rx_opt->sack_ok;
1014 ireq->snd_wscale = rx_opt->snd_wscale;
1015 ireq->wscale_ok = rx_opt->wscale_ok;
1016 ireq->acked = 0;
1017 ireq->ecn_ok = 0;
1018 ireq->rmt_port = tcp_hdr(skb)->source;
1019 ireq->loc_port = tcp_hdr(skb)->dest;
1020}
1021
1022extern void tcp_enter_memory_pressure(struct sock *sk);
1023
1024static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1025{
1026 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1027}
1028
1029static inline int keepalive_time_when(const struct tcp_sock *tp)
1030{
1031 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1032}
1033
1034static inline int keepalive_probes(const struct tcp_sock *tp)
1035{
1036 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1037}
1038
1039static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1040{
1041 const struct inet_connection_sock *icsk = &tp->inet_conn;
1042
1043 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1044 tcp_time_stamp - tp->rcv_tstamp);
1045}
1046
1047static inline int tcp_fin_time(const struct sock *sk)
1048{
1049 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1050 const int rto = inet_csk(sk)->icsk_rto;
1051
1052 if (fin_timeout < (rto << 2) - (rto >> 1))
1053 fin_timeout = (rto << 2) - (rto >> 1);
1054
1055 return fin_timeout;
1056}
1057
1058static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1059 int paws_win)
1060{
1061 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1062 return 1;
1063 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1064 return 1;
1065 /*
1066 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1067 * then following tcp messages have valid values. Ignore 0 value,
1068 * or else 'negative' tsval might forbid us to accept their packets.
1069 */
1070 if (!rx_opt->ts_recent)
1071 return 1;
1072 return 0;
1073}
1074
1075static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1076 int rst)
1077{
1078 if (tcp_paws_check(rx_opt, 0))
1079 return 0;
1080
1081 /* RST segments are not recommended to carry timestamp,
1082 and, if they do, it is recommended to ignore PAWS because
1083 "their cleanup function should take precedence over timestamps."
1084 Certainly, it is mistake. It is necessary to understand the reasons
1085 of this constraint to relax it: if peer reboots, clock may go
1086 out-of-sync and half-open connections will not be reset.
1087 Actually, the problem would be not existing if all
1088 the implementations followed draft about maintaining clock
1089 via reboots. Linux-2.2 DOES NOT!
1090
1091 However, we can relax time bounds for RST segments to MSL.
1092 */
1093 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1094 return 0;
1095 return 1;
1096}
1097
1098static inline void tcp_mib_init(struct net *net)
1099{
1100 /* See RFC 2012 */
1101 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1102 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1103 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1104 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1105}
1106
1107/* from STCP */
1108static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1109{
1110 tp->lost_skb_hint = NULL;
1111 tp->scoreboard_skb_hint = NULL;
1112}
1113
1114static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1115{
1116 tcp_clear_retrans_hints_partial(tp);
1117 tp->retransmit_skb_hint = NULL;
1118}
1119
1120/* MD5 Signature */
1121struct crypto_hash;
1122
1123/* - key database */
1124struct tcp_md5sig_key {
1125 u8 *key;
1126 u8 keylen;
1127};
1128
1129struct tcp4_md5sig_key {
1130 struct tcp_md5sig_key base;
1131 __be32 addr;
1132};
1133
1134struct tcp6_md5sig_key {
1135 struct tcp_md5sig_key base;
1136#if 0
1137 u32 scope_id; /* XXX */
1138#endif
1139 struct in6_addr addr;
1140};
1141
1142/* - sock block */
1143struct tcp_md5sig_info {
1144 struct tcp4_md5sig_key *keys4;
1145#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1146 struct tcp6_md5sig_key *keys6;
1147 u32 entries6;
1148 u32 alloced6;
1149#endif
1150 u32 entries4;
1151 u32 alloced4;
1152};
1153
1154/* - pseudo header */
1155struct tcp4_pseudohdr {
1156 __be32 saddr;
1157 __be32 daddr;
1158 __u8 pad;
1159 __u8 protocol;
1160 __be16 len;
1161};
1162
1163struct tcp6_pseudohdr {
1164 struct in6_addr saddr;
1165 struct in6_addr daddr;
1166 __be32 len;
1167 __be32 protocol; /* including padding */
1168};
1169
1170union tcp_md5sum_block {
1171 struct tcp4_pseudohdr ip4;
1172#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1173 struct tcp6_pseudohdr ip6;
1174#endif
1175};
1176
1177/* - pool: digest algorithm, hash description and scratch buffer */
1178struct tcp_md5sig_pool {
1179 struct hash_desc md5_desc;
1180 union tcp_md5sum_block md5_blk;
1181};
1182
1183/* - functions */
1184extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1185 struct sock *sk, struct request_sock *req,
1186 struct sk_buff *skb);
1187extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk,
1188 struct sock *addr_sk);
1189extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey,
1190 u8 newkeylen);
1191extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr);
1192
1193#ifdef CONFIG_TCP_MD5SIG
1194#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1195 &(struct tcp_md5sig_key) { \
1196 .key = (twsk)->tw_md5_key, \
1197 .keylen = (twsk)->tw_md5_keylen, \
1198 } : NULL)
1199#else
1200#define tcp_twsk_md5_key(twsk) NULL
1201#endif
1202
1203extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *);
1204extern void tcp_free_md5sig_pool(void);
1205
1206extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1207extern void tcp_put_md5sig_pool(void);
1208
1209extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1210extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1211 unsigned header_len);
1212extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1213 struct tcp_md5sig_key *key);
1214
1215/* write queue abstraction */
1216static inline void tcp_write_queue_purge(struct sock *sk)
1217{
1218 struct sk_buff *skb;
1219
1220 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1221 sk_wmem_free_skb(sk, skb);
1222 sk_mem_reclaim(sk);
1223 tcp_clear_all_retrans_hints(tcp_sk(sk));
1224}
1225
1226static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1227{
1228 return skb_peek(&sk->sk_write_queue);
1229}
1230
1231static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1232{
1233 return skb_peek_tail(&sk->sk_write_queue);
1234}
1235
1236static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1237{
1238 return skb_queue_next(&sk->sk_write_queue, skb);
1239}
1240
1241static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1242{
1243 return skb_queue_prev(&sk->sk_write_queue, skb);
1244}
1245
1246#define tcp_for_write_queue(skb, sk) \
1247 skb_queue_walk(&(sk)->sk_write_queue, skb)
1248
1249#define tcp_for_write_queue_from(skb, sk) \
1250 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1251
1252#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1253 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1254
1255static inline struct sk_buff *tcp_send_head(struct sock *sk)
1256{
1257 return sk->sk_send_head;
1258}
1259
1260static inline bool tcp_skb_is_last(const struct sock *sk,
1261 const struct sk_buff *skb)
1262{
1263 return skb_queue_is_last(&sk->sk_write_queue, skb);
1264}
1265
1266static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1267{
1268 if (tcp_skb_is_last(sk, skb))
1269 sk->sk_send_head = NULL;
1270 else
1271 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1272}
1273
1274static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1275{
1276 if (sk->sk_send_head == skb_unlinked)
1277 sk->sk_send_head = NULL;
1278}
1279
1280static inline void tcp_init_send_head(struct sock *sk)
1281{
1282 sk->sk_send_head = NULL;
1283}
1284
1285static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1286{
1287 __skb_queue_tail(&sk->sk_write_queue, skb);
1288}
1289
1290static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1291{
1292 __tcp_add_write_queue_tail(sk, skb);
1293
1294 /* Queue it, remembering where we must start sending. */
1295 if (sk->sk_send_head == NULL) {
1296 sk->sk_send_head = skb;
1297
1298 if (tcp_sk(sk)->highest_sack == NULL)
1299 tcp_sk(sk)->highest_sack = skb;
1300 }
1301}
1302
1303static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1304{
1305 __skb_queue_head(&sk->sk_write_queue, skb);
1306}
1307
1308/* Insert buff after skb on the write queue of sk. */
1309static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1310 struct sk_buff *buff,
1311 struct sock *sk)
1312{
1313 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1314}
1315
1316/* Insert new before skb on the write queue of sk. */
1317static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1318 struct sk_buff *skb,
1319 struct sock *sk)
1320{
1321 __skb_queue_before(&sk->sk_write_queue, skb, new);
1322
1323 if (sk->sk_send_head == skb)
1324 sk->sk_send_head = new;
1325}
1326
1327static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1328{
1329 __skb_unlink(skb, &sk->sk_write_queue);
1330}
1331
1332static inline int tcp_write_queue_empty(struct sock *sk)
1333{
1334 return skb_queue_empty(&sk->sk_write_queue);
1335}
1336
1337static inline void tcp_push_pending_frames(struct sock *sk)
1338{
1339 if (tcp_send_head(sk)) {
1340 struct tcp_sock *tp = tcp_sk(sk);
1341
1342 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1343 }
1344}
1345
1346/* Start sequence of the highest skb with SACKed bit, valid only if
1347 * sacked > 0 or when the caller has ensured validity by itself.
1348 */
1349static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1350{
1351 if (!tp->sacked_out)
1352 return tp->snd_una;
1353
1354 if (tp->highest_sack == NULL)
1355 return tp->snd_nxt;
1356
1357 return TCP_SKB_CB(tp->highest_sack)->seq;
1358}
1359
1360static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1361{
1362 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1363 tcp_write_queue_next(sk, skb);
1364}
1365
1366static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1367{
1368 return tcp_sk(sk)->highest_sack;
1369}
1370
1371static inline void tcp_highest_sack_reset(struct sock *sk)
1372{
1373 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1374}
1375
1376/* Called when old skb is about to be deleted (to be combined with new skb) */
1377static inline void tcp_highest_sack_combine(struct sock *sk,
1378 struct sk_buff *old,
1379 struct sk_buff *new)
1380{
1381 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1382 tcp_sk(sk)->highest_sack = new;
1383}
1384
1385/* Determines whether this is a thin stream (which may suffer from
1386 * increased latency). Used to trigger latency-reducing mechanisms.
1387 */
1388static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp)
1389{
1390 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1391}
1392
1393/* /proc */
1394enum tcp_seq_states {
1395 TCP_SEQ_STATE_LISTENING,
1396 TCP_SEQ_STATE_OPENREQ,
1397 TCP_SEQ_STATE_ESTABLISHED,
1398 TCP_SEQ_STATE_TIME_WAIT,
1399};
1400
1401struct tcp_seq_afinfo {
1402 char *name;
1403 sa_family_t family;
1404 struct file_operations seq_fops;
1405 struct seq_operations seq_ops;
1406};
1407
1408struct tcp_iter_state {
1409 struct seq_net_private p;
1410 sa_family_t family;
1411 enum tcp_seq_states state;
1412 struct sock *syn_wait_sk;
1413 int bucket, offset, sbucket, num, uid;
1414 loff_t last_pos;
1415};
1416
1417extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1418extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1419
1420extern struct request_sock_ops tcp_request_sock_ops;
1421extern struct request_sock_ops tcp6_request_sock_ops;
1422
1423extern void tcp_v4_destroy_sock(struct sock *sk);
1424
1425extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1426extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, u32 features);
1427extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1428 struct sk_buff *skb);
1429extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1430 struct sk_buff *skb);
1431extern int tcp_gro_complete(struct sk_buff *skb);
1432extern int tcp4_gro_complete(struct sk_buff *skb);
1433
1434#ifdef CONFIG_PROC_FS
1435extern int tcp4_proc_init(void);
1436extern void tcp4_proc_exit(void);
1437#endif
1438
1439/* TCP af-specific functions */
1440struct tcp_sock_af_ops {
1441#ifdef CONFIG_TCP_MD5SIG
1442 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1443 struct sock *addr_sk);
1444 int (*calc_md5_hash) (char *location,
1445 struct tcp_md5sig_key *md5,
1446 struct sock *sk,
1447 struct request_sock *req,
1448 struct sk_buff *skb);
1449 int (*md5_add) (struct sock *sk,
1450 struct sock *addr_sk,
1451 u8 *newkey,
1452 u8 len);
1453 int (*md5_parse) (struct sock *sk,
1454 char __user *optval,
1455 int optlen);
1456#endif
1457};
1458
1459struct tcp_request_sock_ops {
1460#ifdef CONFIG_TCP_MD5SIG
1461 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1462 struct request_sock *req);
1463 int (*calc_md5_hash) (char *location,
1464 struct tcp_md5sig_key *md5,
1465 struct sock *sk,
1466 struct request_sock *req,
1467 struct sk_buff *skb);
1468#endif
1469};
1470
1471/* Using SHA1 for now, define some constants.
1472 */
1473#define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1474#define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1475#define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1476
1477extern int tcp_cookie_generator(u32 *bakery);
1478
1479/**
1480 * struct tcp_cookie_values - each socket needs extra space for the
1481 * cookies, together with (optional) space for any SYN data.
1482 *
1483 * A tcp_sock contains a pointer to the current value, and this is
1484 * cloned to the tcp_timewait_sock.
1485 *
1486 * @cookie_pair: variable data from the option exchange.
1487 *
1488 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1489 * indicates default (sysctl_tcp_cookie_size).
1490 * After cookie sent, remembers size of cookie.
1491 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1492 *
1493 * @s_data_desired: user specified tcpct_s_data_desired. When the
1494 * constant payload is specified (@s_data_constant),
1495 * holds its length instead.
1496 * Range 0 to TCP_MSS_DESIRED.
1497 *
1498 * @s_data_payload: constant data that is to be included in the
1499 * payload of SYN or SYNACK segments when the
1500 * cookie option is present.
1501 */
1502struct tcp_cookie_values {
1503 struct kref kref;
1504 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1505 u8 cookie_pair_size;
1506 u8 cookie_desired;
1507 u16 s_data_desired:11,
1508 s_data_constant:1,
1509 s_data_in:1,
1510 s_data_out:1,
1511 s_data_unused:2;
1512 u8 s_data_payload[0];
1513};
1514
1515static inline void tcp_cookie_values_release(struct kref *kref)
1516{
1517 kfree(container_of(kref, struct tcp_cookie_values, kref));
1518}
1519
1520/* The length of constant payload data. Note that s_data_desired is
1521 * overloaded, depending on s_data_constant: either the length of constant
1522 * data (returned here) or the limit on variable data.
1523 */
1524static inline int tcp_s_data_size(const struct tcp_sock *tp)
1525{
1526 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1527 ? tp->cookie_values->s_data_desired
1528 : 0;
1529}
1530
1531/**
1532 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1533 *
1534 * As tcp_request_sock has already been extended in other places, the
1535 * only remaining method is to pass stack values along as function
1536 * parameters. These parameters are not needed after sending SYNACK.
1537 *
1538 * @cookie_bakery: cryptographic secret and message workspace.
1539 *
1540 * @cookie_plus: bytes in authenticator/cookie option, copied from
1541 * struct tcp_options_received (above).
1542 */
1543struct tcp_extend_values {
1544 struct request_values rv;
1545 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1546 u8 cookie_plus:6,
1547 cookie_out_never:1,
1548 cookie_in_always:1;
1549};
1550
1551static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1552{
1553 return (struct tcp_extend_values *)rvp;
1554}
1555
1556extern void tcp_v4_init(void);
1557extern void tcp_init(void);
1558
1559#endif /* _TCP_H */
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Definitions for the TCP module.
8 *
9 * Version: @(#)tcp.h 1.0.5 05/23/93
10 *
11 * Authors: Ross Biro
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 */
14#ifndef _TCP_H
15#define _TCP_H
16
17#define FASTRETRANS_DEBUG 1
18
19#include <linux/list.h>
20#include <linux/tcp.h>
21#include <linux/bug.h>
22#include <linux/slab.h>
23#include <linux/cache.h>
24#include <linux/percpu.h>
25#include <linux/skbuff.h>
26#include <linux/kref.h>
27#include <linux/ktime.h>
28#include <linux/indirect_call_wrapper.h>
29
30#include <net/inet_connection_sock.h>
31#include <net/inet_timewait_sock.h>
32#include <net/inet_hashtables.h>
33#include <net/checksum.h>
34#include <net/request_sock.h>
35#include <net/sock_reuseport.h>
36#include <net/sock.h>
37#include <net/snmp.h>
38#include <net/ip.h>
39#include <net/tcp_states.h>
40#include <net/inet_ecn.h>
41#include <net/dst.h>
42#include <net/mptcp.h>
43
44#include <linux/seq_file.h>
45#include <linux/memcontrol.h>
46#include <linux/bpf-cgroup.h>
47#include <linux/siphash.h>
48
49extern struct inet_hashinfo tcp_hashinfo;
50
51extern struct percpu_counter tcp_orphan_count;
52void tcp_time_wait(struct sock *sk, int state, int timeo);
53
54#define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER)
55#define MAX_TCP_OPTION_SPACE 40
56#define TCP_MIN_SND_MSS 48
57#define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
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 initial MTU to use for probing */
69#define TCP_BASE_MSS 1024
70
71/* probing interval, default to 10 minutes as per RFC4821 */
72#define TCP_PROBE_INTERVAL 600
73
74/* Specify interval when tcp mtu probing will stop */
75#define TCP_PROBE_THRESHOLD 8
76
77/* After receiving this amount of duplicate ACKs fast retransmit starts. */
78#define TCP_FASTRETRANS_THRESH 3
79
80/* Maximal number of ACKs sent quickly to accelerate slow-start. */
81#define TCP_MAX_QUICKACKS 16U
82
83/* Maximal number of window scale according to RFC1323 */
84#define TCP_MAX_WSCALE 14U
85
86/* urg_data states */
87#define TCP_URG_VALID 0x0100
88#define TCP_URG_NOTYET 0x0200
89#define TCP_URG_READ 0x0400
90
91#define TCP_RETR1 3 /*
92 * This is how many retries it does before it
93 * tries to figure out if the gateway is
94 * down. Minimal RFC value is 3; it corresponds
95 * to ~3sec-8min depending on RTO.
96 */
97
98#define TCP_RETR2 15 /*
99 * This should take at least
100 * 90 minutes to time out.
101 * RFC1122 says that the limit is 100 sec.
102 * 15 is ~13-30min depending on RTO.
103 */
104
105#define TCP_SYN_RETRIES 6 /* This is how many retries are done
106 * when active opening a connection.
107 * RFC1122 says the minimum retry MUST
108 * be at least 180secs. Nevertheless
109 * this value is corresponding to
110 * 63secs of retransmission with the
111 * current initial RTO.
112 */
113
114#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
115 * when passive opening a connection.
116 * This is corresponding to 31secs of
117 * retransmission with the current
118 * initial RTO.
119 */
120
121#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
122 * state, about 60 seconds */
123#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
124 /* BSD style FIN_WAIT2 deadlock breaker.
125 * It used to be 3min, new value is 60sec,
126 * to combine FIN-WAIT-2 timeout with
127 * TIME-WAIT timer.
128 */
129#define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
130
131#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
132#if HZ >= 100
133#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
134#define TCP_ATO_MIN ((unsigned)(HZ/25))
135#else
136#define TCP_DELACK_MIN 4U
137#define TCP_ATO_MIN 4U
138#endif
139#define TCP_RTO_MAX ((unsigned)(120*HZ))
140#define TCP_RTO_MIN ((unsigned)(HZ/5))
141#define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
142#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
143#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
144 * used as a fallback RTO for the
145 * initial data transmission if no
146 * valid RTT sample has been acquired,
147 * most likely due to retrans in 3WHS.
148 */
149
150#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
151 * for local resources.
152 */
153#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
154#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
155#define TCP_KEEPALIVE_INTVL (75*HZ)
156
157#define MAX_TCP_KEEPIDLE 32767
158#define MAX_TCP_KEEPINTVL 32767
159#define MAX_TCP_KEEPCNT 127
160#define MAX_TCP_SYNCNT 127
161
162#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
163
164#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
165#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
166 * after this time. It should be equal
167 * (or greater than) TCP_TIMEWAIT_LEN
168 * to provide reliability equal to one
169 * provided by timewait state.
170 */
171#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
172 * timestamps. It must be less than
173 * minimal timewait lifetime.
174 */
175/*
176 * TCP option
177 */
178
179#define TCPOPT_NOP 1 /* Padding */
180#define TCPOPT_EOL 0 /* End of options */
181#define TCPOPT_MSS 2 /* Segment size negotiating */
182#define TCPOPT_WINDOW 3 /* Window scaling */
183#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
184#define TCPOPT_SACK 5 /* SACK Block */
185#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
186#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
187#define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */
188#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
189#define TCPOPT_EXP 254 /* Experimental */
190/* Magic number to be after the option value for sharing TCP
191 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
192 */
193#define TCPOPT_FASTOPEN_MAGIC 0xF989
194#define TCPOPT_SMC_MAGIC 0xE2D4C3D9
195
196/*
197 * TCP option lengths
198 */
199
200#define TCPOLEN_MSS 4
201#define TCPOLEN_WINDOW 3
202#define TCPOLEN_SACK_PERM 2
203#define TCPOLEN_TIMESTAMP 10
204#define TCPOLEN_MD5SIG 18
205#define TCPOLEN_FASTOPEN_BASE 2
206#define TCPOLEN_EXP_FASTOPEN_BASE 4
207#define TCPOLEN_EXP_SMC_BASE 6
208
209/* But this is what stacks really send out. */
210#define TCPOLEN_TSTAMP_ALIGNED 12
211#define TCPOLEN_WSCALE_ALIGNED 4
212#define TCPOLEN_SACKPERM_ALIGNED 4
213#define TCPOLEN_SACK_BASE 2
214#define TCPOLEN_SACK_BASE_ALIGNED 4
215#define TCPOLEN_SACK_PERBLOCK 8
216#define TCPOLEN_MD5SIG_ALIGNED 20
217#define TCPOLEN_MSS_ALIGNED 4
218#define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
219
220/* Flags in tp->nonagle */
221#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
222#define TCP_NAGLE_CORK 2 /* Socket is corked */
223#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
224
225/* TCP thin-stream limits */
226#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
227
228/* TCP initial congestion window as per rfc6928 */
229#define TCP_INIT_CWND 10
230
231/* Bit Flags for sysctl_tcp_fastopen */
232#define TFO_CLIENT_ENABLE 1
233#define TFO_SERVER_ENABLE 2
234#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
235
236/* Accept SYN data w/o any cookie option */
237#define TFO_SERVER_COOKIE_NOT_REQD 0x200
238
239/* Force enable TFO on all listeners, i.e., not requiring the
240 * TCP_FASTOPEN socket option.
241 */
242#define TFO_SERVER_WO_SOCKOPT1 0x400
243
244
245/* sysctl variables for tcp */
246extern int sysctl_tcp_max_orphans;
247extern long sysctl_tcp_mem[3];
248
249#define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
250#define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
251#define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
252
253extern atomic_long_t tcp_memory_allocated;
254extern struct percpu_counter tcp_sockets_allocated;
255extern unsigned long tcp_memory_pressure;
256
257/* optimized version of sk_under_memory_pressure() for TCP sockets */
258static inline bool tcp_under_memory_pressure(const struct sock *sk)
259{
260 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
261 mem_cgroup_under_socket_pressure(sk->sk_memcg))
262 return true;
263
264 return READ_ONCE(tcp_memory_pressure);
265}
266/*
267 * The next routines deal with comparing 32 bit unsigned ints
268 * and worry about wraparound (automatic with unsigned arithmetic).
269 */
270
271static inline bool before(__u32 seq1, __u32 seq2)
272{
273 return (__s32)(seq1-seq2) < 0;
274}
275#define after(seq2, seq1) before(seq1, seq2)
276
277/* is s2<=s1<=s3 ? */
278static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
279{
280 return seq3 - seq2 >= seq1 - seq2;
281}
282
283static inline bool tcp_out_of_memory(struct sock *sk)
284{
285 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
286 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
287 return true;
288 return false;
289}
290
291void sk_forced_mem_schedule(struct sock *sk, int size);
292
293static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
294{
295 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
296 int orphans = percpu_counter_read_positive(ocp);
297
298 if (orphans << shift > sysctl_tcp_max_orphans) {
299 orphans = percpu_counter_sum_positive(ocp);
300 if (orphans << shift > sysctl_tcp_max_orphans)
301 return true;
302 }
303 return false;
304}
305
306bool tcp_check_oom(struct sock *sk, int shift);
307
308
309extern struct proto tcp_prot;
310
311#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
312#define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
313#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
314#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
315
316void tcp_tasklet_init(void);
317
318int tcp_v4_err(struct sk_buff *skb, u32);
319
320void tcp_shutdown(struct sock *sk, int how);
321
322int tcp_v4_early_demux(struct sk_buff *skb);
323int tcp_v4_rcv(struct sk_buff *skb);
324
325void tcp_remove_empty_skb(struct sock *sk, struct sk_buff *skb);
326int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
327int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
328int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
329int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
330 int flags);
331int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
332 size_t size, int flags);
333struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags,
334 struct page *page, int offset, size_t *size);
335ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
336 size_t size, int flags);
337int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
338void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
339 int size_goal);
340void tcp_release_cb(struct sock *sk);
341void tcp_wfree(struct sk_buff *skb);
342void tcp_write_timer_handler(struct sock *sk);
343void tcp_delack_timer_handler(struct sock *sk);
344int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
345int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
346void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
347void tcp_rcv_space_adjust(struct sock *sk);
348int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
349void tcp_twsk_destructor(struct sock *sk);
350ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
351 struct pipe_inode_info *pipe, size_t len,
352 unsigned int flags);
353
354void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
355static inline void tcp_dec_quickack_mode(struct sock *sk,
356 const unsigned int pkts)
357{
358 struct inet_connection_sock *icsk = inet_csk(sk);
359
360 if (icsk->icsk_ack.quick) {
361 if (pkts >= icsk->icsk_ack.quick) {
362 icsk->icsk_ack.quick = 0;
363 /* Leaving quickack mode we deflate ATO. */
364 icsk->icsk_ack.ato = TCP_ATO_MIN;
365 } else
366 icsk->icsk_ack.quick -= pkts;
367 }
368}
369
370#define TCP_ECN_OK 1
371#define TCP_ECN_QUEUE_CWR 2
372#define TCP_ECN_DEMAND_CWR 4
373#define TCP_ECN_SEEN 8
374
375enum tcp_tw_status {
376 TCP_TW_SUCCESS = 0,
377 TCP_TW_RST = 1,
378 TCP_TW_ACK = 2,
379 TCP_TW_SYN = 3
380};
381
382
383enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
384 struct sk_buff *skb,
385 const struct tcphdr *th);
386struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
387 struct request_sock *req, bool fastopen,
388 bool *lost_race);
389int tcp_child_process(struct sock *parent, struct sock *child,
390 struct sk_buff *skb);
391void tcp_enter_loss(struct sock *sk);
392void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
393void tcp_clear_retrans(struct tcp_sock *tp);
394void tcp_update_metrics(struct sock *sk);
395void tcp_init_metrics(struct sock *sk);
396void tcp_metrics_init(void);
397bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
398void __tcp_close(struct sock *sk, long timeout);
399void tcp_close(struct sock *sk, long timeout);
400void tcp_init_sock(struct sock *sk);
401void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
402__poll_t tcp_poll(struct file *file, struct socket *sock,
403 struct poll_table_struct *wait);
404int tcp_getsockopt(struct sock *sk, int level, int optname,
405 char __user *optval, int __user *optlen);
406bool tcp_bpf_bypass_getsockopt(int level, int optname);
407int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
408 unsigned int optlen);
409void tcp_set_keepalive(struct sock *sk, int val);
410void tcp_syn_ack_timeout(const struct request_sock *req);
411int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
412 int flags, int *addr_len);
413int tcp_set_rcvlowat(struct sock *sk, int val);
414int tcp_set_window_clamp(struct sock *sk, int val);
415void tcp_update_recv_tstamps(struct sk_buff *skb,
416 struct scm_timestamping_internal *tss);
417void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
418 struct scm_timestamping_internal *tss);
419void tcp_data_ready(struct sock *sk);
420#ifdef CONFIG_MMU
421int tcp_mmap(struct file *file, struct socket *sock,
422 struct vm_area_struct *vma);
423#endif
424void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
425 struct tcp_options_received *opt_rx,
426 int estab, struct tcp_fastopen_cookie *foc);
427const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
428
429/*
430 * BPF SKB-less helpers
431 */
432u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
433 struct tcphdr *th, u32 *cookie);
434u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
435 struct tcphdr *th, u32 *cookie);
436u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
437 const struct tcp_request_sock_ops *af_ops,
438 struct sock *sk, struct tcphdr *th);
439/*
440 * TCP v4 functions exported for the inet6 API
441 */
442
443void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
444void tcp_v4_mtu_reduced(struct sock *sk);
445void tcp_req_err(struct sock *sk, u32 seq, bool abort);
446void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
447int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
448struct sock *tcp_create_openreq_child(const struct sock *sk,
449 struct request_sock *req,
450 struct sk_buff *skb);
451void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
452struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
453 struct request_sock *req,
454 struct dst_entry *dst,
455 struct request_sock *req_unhash,
456 bool *own_req);
457int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
458int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
459int tcp_connect(struct sock *sk);
460enum tcp_synack_type {
461 TCP_SYNACK_NORMAL,
462 TCP_SYNACK_FASTOPEN,
463 TCP_SYNACK_COOKIE,
464};
465struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
466 struct request_sock *req,
467 struct tcp_fastopen_cookie *foc,
468 enum tcp_synack_type synack_type,
469 struct sk_buff *syn_skb);
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 */
477struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
478 struct request_sock *req,
479 struct dst_entry *dst, u32 tsoff);
480int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
481 u32 cookie);
482struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
483struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
484 struct sock *sk, struct sk_buff *skb);
485#ifdef CONFIG_SYN_COOKIES
486
487/* Syncookies use a monotonic timer which increments every 60 seconds.
488 * This counter is used both as a hash input and partially encoded into
489 * the cookie value. A cookie is only validated further if the delta
490 * between the current counter value and the encoded one is less than this,
491 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
492 * the counter advances immediately after a cookie is generated).
493 */
494#define MAX_SYNCOOKIE_AGE 2
495#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
496#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
497
498/* syncookies: remember time of last synqueue overflow
499 * But do not dirty this field too often (once per second is enough)
500 * It is racy as we do not hold a lock, but race is very minor.
501 */
502static inline void tcp_synq_overflow(const struct sock *sk)
503{
504 unsigned int last_overflow;
505 unsigned int now = jiffies;
506
507 if (sk->sk_reuseport) {
508 struct sock_reuseport *reuse;
509
510 reuse = rcu_dereference(sk->sk_reuseport_cb);
511 if (likely(reuse)) {
512 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
513 if (!time_between32(now, last_overflow,
514 last_overflow + HZ))
515 WRITE_ONCE(reuse->synq_overflow_ts, now);
516 return;
517 }
518 }
519
520 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
521 if (!time_between32(now, last_overflow, last_overflow + HZ))
522 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
523}
524
525/* syncookies: no recent synqueue overflow on this listening socket? */
526static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
527{
528 unsigned int last_overflow;
529 unsigned int now = jiffies;
530
531 if (sk->sk_reuseport) {
532 struct sock_reuseport *reuse;
533
534 reuse = rcu_dereference(sk->sk_reuseport_cb);
535 if (likely(reuse)) {
536 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
537 return !time_between32(now, last_overflow - HZ,
538 last_overflow +
539 TCP_SYNCOOKIE_VALID);
540 }
541 }
542
543 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
544
545 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
546 * then we're under synflood. However, we have to use
547 * 'last_overflow - HZ' as lower bound. That's because a concurrent
548 * tcp_synq_overflow() could update .ts_recent_stamp after we read
549 * jiffies but before we store .ts_recent_stamp into last_overflow,
550 * which could lead to rejecting a valid syncookie.
551 */
552 return !time_between32(now, last_overflow - HZ,
553 last_overflow + TCP_SYNCOOKIE_VALID);
554}
555
556static inline u32 tcp_cookie_time(void)
557{
558 u64 val = get_jiffies_64();
559
560 do_div(val, TCP_SYNCOOKIE_PERIOD);
561 return val;
562}
563
564u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
565 u16 *mssp);
566__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
567u64 cookie_init_timestamp(struct request_sock *req, u64 now);
568bool cookie_timestamp_decode(const struct net *net,
569 struct tcp_options_received *opt);
570bool cookie_ecn_ok(const struct tcp_options_received *opt,
571 const struct net *net, const struct dst_entry *dst);
572
573/* From net/ipv6/syncookies.c */
574int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
575 u32 cookie);
576struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
577
578u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
579 const struct tcphdr *th, u16 *mssp);
580__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
581#endif
582/* tcp_output.c */
583
584void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
585 int nonagle);
586int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
587int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
588void tcp_retransmit_timer(struct sock *sk);
589void tcp_xmit_retransmit_queue(struct sock *);
590void tcp_simple_retransmit(struct sock *);
591void tcp_enter_recovery(struct sock *sk, bool ece_ack);
592int tcp_trim_head(struct sock *, struct sk_buff *, u32);
593enum tcp_queue {
594 TCP_FRAG_IN_WRITE_QUEUE,
595 TCP_FRAG_IN_RTX_QUEUE,
596};
597int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
598 struct sk_buff *skb, u32 len,
599 unsigned int mss_now, gfp_t gfp);
600
601void tcp_send_probe0(struct sock *);
602void tcp_send_partial(struct sock *);
603int tcp_write_wakeup(struct sock *, int mib);
604void tcp_send_fin(struct sock *sk);
605void tcp_send_active_reset(struct sock *sk, gfp_t priority);
606int tcp_send_synack(struct sock *);
607void tcp_push_one(struct sock *, unsigned int mss_now);
608void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
609void tcp_send_ack(struct sock *sk);
610void tcp_send_delayed_ack(struct sock *sk);
611void tcp_send_loss_probe(struct sock *sk);
612bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
613void tcp_skb_collapse_tstamp(struct sk_buff *skb,
614 const struct sk_buff *next_skb);
615
616/* tcp_input.c */
617void tcp_rearm_rto(struct sock *sk);
618void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
619void tcp_reset(struct sock *sk, struct sk_buff *skb);
620void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
621void tcp_fin(struct sock *sk);
622
623/* tcp_timer.c */
624void tcp_init_xmit_timers(struct sock *);
625static inline void tcp_clear_xmit_timers(struct sock *sk)
626{
627 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
628 __sock_put(sk);
629
630 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
631 __sock_put(sk);
632
633 inet_csk_clear_xmit_timers(sk);
634}
635
636unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
637unsigned int tcp_current_mss(struct sock *sk);
638u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
639
640/* Bound MSS / TSO packet size with the half of the window */
641static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
642{
643 int cutoff;
644
645 /* When peer uses tiny windows, there is no use in packetizing
646 * to sub-MSS pieces for the sake of SWS or making sure there
647 * are enough packets in the pipe for fast recovery.
648 *
649 * On the other hand, for extremely large MSS devices, handling
650 * smaller than MSS windows in this way does make sense.
651 */
652 if (tp->max_window > TCP_MSS_DEFAULT)
653 cutoff = (tp->max_window >> 1);
654 else
655 cutoff = tp->max_window;
656
657 if (cutoff && pktsize > cutoff)
658 return max_t(int, cutoff, 68U - tp->tcp_header_len);
659 else
660 return pktsize;
661}
662
663/* tcp.c */
664void tcp_get_info(struct sock *, struct tcp_info *);
665
666/* Read 'sendfile()'-style from a TCP socket */
667int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
668 sk_read_actor_t recv_actor);
669
670void tcp_initialize_rcv_mss(struct sock *sk);
671
672int tcp_mtu_to_mss(struct sock *sk, int pmtu);
673int tcp_mss_to_mtu(struct sock *sk, int mss);
674void tcp_mtup_init(struct sock *sk);
675
676static inline void tcp_bound_rto(const struct sock *sk)
677{
678 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
679 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
680}
681
682static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
683{
684 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
685}
686
687static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
688{
689 /* mptcp hooks are only on the slow path */
690 if (sk_is_mptcp((struct sock *)tp))
691 return;
692
693 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
694 ntohl(TCP_FLAG_ACK) |
695 snd_wnd);
696}
697
698static inline void tcp_fast_path_on(struct tcp_sock *tp)
699{
700 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
701}
702
703static inline void tcp_fast_path_check(struct sock *sk)
704{
705 struct tcp_sock *tp = tcp_sk(sk);
706
707 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
708 tp->rcv_wnd &&
709 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
710 !tp->urg_data)
711 tcp_fast_path_on(tp);
712}
713
714/* Compute the actual rto_min value */
715static inline u32 tcp_rto_min(struct sock *sk)
716{
717 const struct dst_entry *dst = __sk_dst_get(sk);
718 u32 rto_min = inet_csk(sk)->icsk_rto_min;
719
720 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
721 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
722 return rto_min;
723}
724
725static inline u32 tcp_rto_min_us(struct sock *sk)
726{
727 return jiffies_to_usecs(tcp_rto_min(sk));
728}
729
730static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
731{
732 return dst_metric_locked(dst, RTAX_CC_ALGO);
733}
734
735/* Minimum RTT in usec. ~0 means not available. */
736static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
737{
738 return minmax_get(&tp->rtt_min);
739}
740
741/* Compute the actual receive window we are currently advertising.
742 * Rcv_nxt can be after the window if our peer push more data
743 * than the offered window.
744 */
745static inline u32 tcp_receive_window(const struct tcp_sock *tp)
746{
747 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
748
749 if (win < 0)
750 win = 0;
751 return (u32) win;
752}
753
754/* Choose a new window, without checks for shrinking, and without
755 * scaling applied to the result. The caller does these things
756 * if necessary. This is a "raw" window selection.
757 */
758u32 __tcp_select_window(struct sock *sk);
759
760void tcp_send_window_probe(struct sock *sk);
761
762/* TCP uses 32bit jiffies to save some space.
763 * Note that this is different from tcp_time_stamp, which
764 * historically has been the same until linux-4.13.
765 */
766#define tcp_jiffies32 ((u32)jiffies)
767
768/*
769 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
770 * It is no longer tied to jiffies, but to 1 ms clock.
771 * Note: double check if you want to use tcp_jiffies32 instead of this.
772 */
773#define TCP_TS_HZ 1000
774
775static inline u64 tcp_clock_ns(void)
776{
777 return ktime_get_ns();
778}
779
780static inline u64 tcp_clock_us(void)
781{
782 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
783}
784
785/* This should only be used in contexts where tp->tcp_mstamp is up to date */
786static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
787{
788 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
789}
790
791/* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
792static inline u32 tcp_ns_to_ts(u64 ns)
793{
794 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
795}
796
797/* Could use tcp_clock_us() / 1000, but this version uses a single divide */
798static inline u32 tcp_time_stamp_raw(void)
799{
800 return tcp_ns_to_ts(tcp_clock_ns());
801}
802
803void tcp_mstamp_refresh(struct tcp_sock *tp);
804
805static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
806{
807 return max_t(s64, t1 - t0, 0);
808}
809
810static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
811{
812 return tcp_ns_to_ts(skb->skb_mstamp_ns);
813}
814
815/* provide the departure time in us unit */
816static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
817{
818 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
819}
820
821
822#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
823
824#define TCPHDR_FIN 0x01
825#define TCPHDR_SYN 0x02
826#define TCPHDR_RST 0x04
827#define TCPHDR_PSH 0x08
828#define TCPHDR_ACK 0x10
829#define TCPHDR_URG 0x20
830#define TCPHDR_ECE 0x40
831#define TCPHDR_CWR 0x80
832
833#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
834
835/* This is what the send packet queuing engine uses to pass
836 * TCP per-packet control information to the transmission code.
837 * We also store the host-order sequence numbers in here too.
838 * This is 44 bytes if IPV6 is enabled.
839 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
840 */
841struct tcp_skb_cb {
842 __u32 seq; /* Starting sequence number */
843 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
844 union {
845 /* Note : tcp_tw_isn is used in input path only
846 * (isn chosen by tcp_timewait_state_process())
847 *
848 * tcp_gso_segs/size are used in write queue only,
849 * cf tcp_skb_pcount()/tcp_skb_mss()
850 */
851 __u32 tcp_tw_isn;
852 struct {
853 u16 tcp_gso_segs;
854 u16 tcp_gso_size;
855 };
856 };
857 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
858
859 __u8 sacked; /* State flags for SACK. */
860#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
861#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
862#define TCPCB_LOST 0x04 /* SKB is lost */
863#define TCPCB_TAGBITS 0x07 /* All tag bits */
864#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
865#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
866#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
867 TCPCB_REPAIRED)
868
869 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
870 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
871 eor:1, /* Is skb MSG_EOR marked? */
872 has_rxtstamp:1, /* SKB has a RX timestamp */
873 unused:5;
874 __u32 ack_seq; /* Sequence number ACK'd */
875 union {
876 struct {
877 /* There is space for up to 24 bytes */
878 __u32 in_flight:30,/* Bytes in flight at transmit */
879 is_app_limited:1, /* cwnd not fully used? */
880 unused:1;
881 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
882 __u32 delivered;
883 /* start of send pipeline phase */
884 u64 first_tx_mstamp;
885 /* when we reached the "delivered" count */
886 u64 delivered_mstamp;
887 } tx; /* only used for outgoing skbs */
888 union {
889 struct inet_skb_parm h4;
890#if IS_ENABLED(CONFIG_IPV6)
891 struct inet6_skb_parm h6;
892#endif
893 } header; /* For incoming skbs */
894 };
895};
896
897#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
898
899extern const struct inet_connection_sock_af_ops ipv4_specific;
900
901#if IS_ENABLED(CONFIG_IPV6)
902/* This is the variant of inet6_iif() that must be used by TCP,
903 * as TCP moves IP6CB into a different location in skb->cb[]
904 */
905static inline int tcp_v6_iif(const struct sk_buff *skb)
906{
907 return TCP_SKB_CB(skb)->header.h6.iif;
908}
909
910static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
911{
912 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
913
914 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
915}
916
917/* TCP_SKB_CB reference means this can not be used from early demux */
918static inline int tcp_v6_sdif(const struct sk_buff *skb)
919{
920#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
921 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
922 return TCP_SKB_CB(skb)->header.h6.iif;
923#endif
924 return 0;
925}
926
927extern const struct inet_connection_sock_af_ops ipv6_specific;
928
929INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
930INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
931INDIRECT_CALLABLE_DECLARE(void tcp_v6_early_demux(struct sk_buff *skb));
932
933#endif
934
935/* TCP_SKB_CB reference means this can not be used from early demux */
936static inline int tcp_v4_sdif(struct sk_buff *skb)
937{
938#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
939 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
940 return TCP_SKB_CB(skb)->header.h4.iif;
941#endif
942 return 0;
943}
944
945/* Due to TSO, an SKB can be composed of multiple actual
946 * packets. To keep these tracked properly, we use this.
947 */
948static inline int tcp_skb_pcount(const struct sk_buff *skb)
949{
950 return TCP_SKB_CB(skb)->tcp_gso_segs;
951}
952
953static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
954{
955 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
956}
957
958static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
959{
960 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
961}
962
963/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
964static inline int tcp_skb_mss(const struct sk_buff *skb)
965{
966 return TCP_SKB_CB(skb)->tcp_gso_size;
967}
968
969static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
970{
971 return likely(!TCP_SKB_CB(skb)->eor);
972}
973
974static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
975 const struct sk_buff *from)
976{
977 return likely(tcp_skb_can_collapse_to(to) &&
978 mptcp_skb_can_collapse(to, from));
979}
980
981/* Events passed to congestion control interface */
982enum tcp_ca_event {
983 CA_EVENT_TX_START, /* first transmit when no packets in flight */
984 CA_EVENT_CWND_RESTART, /* congestion window restart */
985 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
986 CA_EVENT_LOSS, /* loss timeout */
987 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
988 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
989};
990
991/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
992enum tcp_ca_ack_event_flags {
993 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
994 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
995 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
996};
997
998/*
999 * Interface for adding new TCP congestion control handlers
1000 */
1001#define TCP_CA_NAME_MAX 16
1002#define TCP_CA_MAX 128
1003#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
1004
1005#define TCP_CA_UNSPEC 0
1006
1007/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1008#define TCP_CONG_NON_RESTRICTED 0x1
1009/* Requires ECN/ECT set on all packets */
1010#define TCP_CONG_NEEDS_ECN 0x2
1011#define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1012
1013union tcp_cc_info;
1014
1015struct ack_sample {
1016 u32 pkts_acked;
1017 s32 rtt_us;
1018 u32 in_flight;
1019};
1020
1021/* A rate sample measures the number of (original/retransmitted) data
1022 * packets delivered "delivered" over an interval of time "interval_us".
1023 * The tcp_rate.c code fills in the rate sample, and congestion
1024 * control modules that define a cong_control function to run at the end
1025 * of ACK processing can optionally chose to consult this sample when
1026 * setting cwnd and pacing rate.
1027 * A sample is invalid if "delivered" or "interval_us" is negative.
1028 */
1029struct rate_sample {
1030 u64 prior_mstamp; /* starting timestamp for interval */
1031 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1032 s32 delivered; /* number of packets delivered over interval */
1033 long interval_us; /* time for tp->delivered to incr "delivered" */
1034 u32 snd_interval_us; /* snd interval for delivered packets */
1035 u32 rcv_interval_us; /* rcv interval for delivered packets */
1036 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1037 int losses; /* number of packets marked lost upon ACK */
1038 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1039 u32 prior_in_flight; /* in flight before this ACK */
1040 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1041 bool is_retrans; /* is sample from retransmission? */
1042 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1043};
1044
1045struct tcp_congestion_ops {
1046/* fast path fields are put first to fill one cache line */
1047
1048 /* return slow start threshold (required) */
1049 u32 (*ssthresh)(struct sock *sk);
1050
1051 /* do new cwnd calculation (required) */
1052 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1053
1054 /* call before changing ca_state (optional) */
1055 void (*set_state)(struct sock *sk, u8 new_state);
1056
1057 /* call when cwnd event occurs (optional) */
1058 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1059
1060 /* call when ack arrives (optional) */
1061 void (*in_ack_event)(struct sock *sk, u32 flags);
1062
1063 /* hook for packet ack accounting (optional) */
1064 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1065
1066 /* override sysctl_tcp_min_tso_segs */
1067 u32 (*min_tso_segs)(struct sock *sk);
1068
1069 /* call when packets are delivered to update cwnd and pacing rate,
1070 * after all the ca_state processing. (optional)
1071 */
1072 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1073
1074
1075 /* new value of cwnd after loss (required) */
1076 u32 (*undo_cwnd)(struct sock *sk);
1077 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1078 u32 (*sndbuf_expand)(struct sock *sk);
1079
1080/* control/slow paths put last */
1081 /* get info for inet_diag (optional) */
1082 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1083 union tcp_cc_info *info);
1084
1085 char name[TCP_CA_NAME_MAX];
1086 struct module *owner;
1087 struct list_head list;
1088 u32 key;
1089 u32 flags;
1090
1091 /* initialize private data (optional) */
1092 void (*init)(struct sock *sk);
1093 /* cleanup private data (optional) */
1094 void (*release)(struct sock *sk);
1095} ____cacheline_aligned_in_smp;
1096
1097int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1098void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1099
1100void tcp_assign_congestion_control(struct sock *sk);
1101void tcp_init_congestion_control(struct sock *sk);
1102void tcp_cleanup_congestion_control(struct sock *sk);
1103int tcp_set_default_congestion_control(struct net *net, const char *name);
1104void tcp_get_default_congestion_control(struct net *net, char *name);
1105void tcp_get_available_congestion_control(char *buf, size_t len);
1106void tcp_get_allowed_congestion_control(char *buf, size_t len);
1107int tcp_set_allowed_congestion_control(char *allowed);
1108int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1109 bool cap_net_admin);
1110u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1111void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1112
1113u32 tcp_reno_ssthresh(struct sock *sk);
1114u32 tcp_reno_undo_cwnd(struct sock *sk);
1115void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1116extern struct tcp_congestion_ops tcp_reno;
1117
1118struct tcp_congestion_ops *tcp_ca_find(const char *name);
1119struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1120u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1121#ifdef CONFIG_INET
1122char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1123#else
1124static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1125{
1126 return NULL;
1127}
1128#endif
1129
1130static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1131{
1132 const struct inet_connection_sock *icsk = inet_csk(sk);
1133
1134 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1135}
1136
1137static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1138{
1139 struct inet_connection_sock *icsk = inet_csk(sk);
1140
1141 if (icsk->icsk_ca_ops->set_state)
1142 icsk->icsk_ca_ops->set_state(sk, ca_state);
1143 icsk->icsk_ca_state = ca_state;
1144}
1145
1146static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1147{
1148 const struct inet_connection_sock *icsk = inet_csk(sk);
1149
1150 if (icsk->icsk_ca_ops->cwnd_event)
1151 icsk->icsk_ca_ops->cwnd_event(sk, event);
1152}
1153
1154/* From tcp_rate.c */
1155void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1156void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1157 struct rate_sample *rs);
1158void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1159 bool is_sack_reneg, struct rate_sample *rs);
1160void tcp_rate_check_app_limited(struct sock *sk);
1161
1162/* These functions determine how the current flow behaves in respect of SACK
1163 * handling. SACK is negotiated with the peer, and therefore it can vary
1164 * between different flows.
1165 *
1166 * tcp_is_sack - SACK enabled
1167 * tcp_is_reno - No SACK
1168 */
1169static inline int tcp_is_sack(const struct tcp_sock *tp)
1170{
1171 return likely(tp->rx_opt.sack_ok);
1172}
1173
1174static inline bool tcp_is_reno(const struct tcp_sock *tp)
1175{
1176 return !tcp_is_sack(tp);
1177}
1178
1179static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1180{
1181 return tp->sacked_out + tp->lost_out;
1182}
1183
1184/* This determines how many packets are "in the network" to the best
1185 * of our knowledge. In many cases it is conservative, but where
1186 * detailed information is available from the receiver (via SACK
1187 * blocks etc.) we can make more aggressive calculations.
1188 *
1189 * Use this for decisions involving congestion control, use just
1190 * tp->packets_out to determine if the send queue is empty or not.
1191 *
1192 * Read this equation as:
1193 *
1194 * "Packets sent once on transmission queue" MINUS
1195 * "Packets left network, but not honestly ACKed yet" PLUS
1196 * "Packets fast retransmitted"
1197 */
1198static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1199{
1200 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1201}
1202
1203#define TCP_INFINITE_SSTHRESH 0x7fffffff
1204
1205static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1206{
1207 return tp->snd_cwnd < tp->snd_ssthresh;
1208}
1209
1210static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1211{
1212 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1213}
1214
1215static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1216{
1217 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1218 (1 << inet_csk(sk)->icsk_ca_state);
1219}
1220
1221/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1222 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1223 * ssthresh.
1224 */
1225static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1226{
1227 const struct tcp_sock *tp = tcp_sk(sk);
1228
1229 if (tcp_in_cwnd_reduction(sk))
1230 return tp->snd_ssthresh;
1231 else
1232 return max(tp->snd_ssthresh,
1233 ((tp->snd_cwnd >> 1) +
1234 (tp->snd_cwnd >> 2)));
1235}
1236
1237/* Use define here intentionally to get WARN_ON location shown at the caller */
1238#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1239
1240void tcp_enter_cwr(struct sock *sk);
1241__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1242
1243/* The maximum number of MSS of available cwnd for which TSO defers
1244 * sending if not using sysctl_tcp_tso_win_divisor.
1245 */
1246static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1247{
1248 return 3;
1249}
1250
1251/* Returns end sequence number of the receiver's advertised window */
1252static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1253{
1254 return tp->snd_una + tp->snd_wnd;
1255}
1256
1257/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1258 * flexible approach. The RFC suggests cwnd should not be raised unless
1259 * it was fully used previously. And that's exactly what we do in
1260 * congestion avoidance mode. But in slow start we allow cwnd to grow
1261 * as long as the application has used half the cwnd.
1262 * Example :
1263 * cwnd is 10 (IW10), but application sends 9 frames.
1264 * We allow cwnd to reach 18 when all frames are ACKed.
1265 * This check is safe because it's as aggressive as slow start which already
1266 * risks 100% overshoot. The advantage is that we discourage application to
1267 * either send more filler packets or data to artificially blow up the cwnd
1268 * usage, and allow application-limited process to probe bw more aggressively.
1269 */
1270static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1271{
1272 const struct tcp_sock *tp = tcp_sk(sk);
1273
1274 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1275 if (tcp_in_slow_start(tp))
1276 return tp->snd_cwnd < 2 * tp->max_packets_out;
1277
1278 return tp->is_cwnd_limited;
1279}
1280
1281/* BBR congestion control needs pacing.
1282 * Same remark for SO_MAX_PACING_RATE.
1283 * sch_fq packet scheduler is efficiently handling pacing,
1284 * but is not always installed/used.
1285 * Return true if TCP stack should pace packets itself.
1286 */
1287static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1288{
1289 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1290}
1291
1292/* Estimates in how many jiffies next packet for this flow can be sent.
1293 * Scheduling a retransmit timer too early would be silly.
1294 */
1295static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1296{
1297 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1298
1299 return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1300}
1301
1302static inline void tcp_reset_xmit_timer(struct sock *sk,
1303 const int what,
1304 unsigned long when,
1305 const unsigned long max_when)
1306{
1307 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
1308 max_when);
1309}
1310
1311/* Something is really bad, we could not queue an additional packet,
1312 * because qdisc is full or receiver sent a 0 window, or we are paced.
1313 * We do not want to add fuel to the fire, or abort too early,
1314 * so make sure the timer we arm now is at least 200ms in the future,
1315 * regardless of current icsk_rto value (as it could be ~2ms)
1316 */
1317static inline unsigned long tcp_probe0_base(const struct sock *sk)
1318{
1319 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1320}
1321
1322/* Variant of inet_csk_rto_backoff() used for zero window probes */
1323static inline unsigned long tcp_probe0_when(const struct sock *sk,
1324 unsigned long max_when)
1325{
1326 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1327 inet_csk(sk)->icsk_backoff);
1328 u64 when = (u64)tcp_probe0_base(sk) << backoff;
1329
1330 return (unsigned long)min_t(u64, when, max_when);
1331}
1332
1333static inline void tcp_check_probe_timer(struct sock *sk)
1334{
1335 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1336 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1337 tcp_probe0_base(sk), TCP_RTO_MAX);
1338}
1339
1340static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1341{
1342 tp->snd_wl1 = seq;
1343}
1344
1345static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1346{
1347 tp->snd_wl1 = seq;
1348}
1349
1350/*
1351 * Calculate(/check) TCP checksum
1352 */
1353static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1354 __be32 daddr, __wsum base)
1355{
1356 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1357}
1358
1359static inline bool tcp_checksum_complete(struct sk_buff *skb)
1360{
1361 return !skb_csum_unnecessary(skb) &&
1362 __skb_checksum_complete(skb);
1363}
1364
1365bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1366int tcp_filter(struct sock *sk, struct sk_buff *skb);
1367void tcp_set_state(struct sock *sk, int state);
1368void tcp_done(struct sock *sk);
1369int tcp_abort(struct sock *sk, int err);
1370
1371static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1372{
1373 rx_opt->dsack = 0;
1374 rx_opt->num_sacks = 0;
1375}
1376
1377void tcp_cwnd_restart(struct sock *sk, s32 delta);
1378
1379static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1380{
1381 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1382 struct tcp_sock *tp = tcp_sk(sk);
1383 s32 delta;
1384
1385 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1386 ca_ops->cong_control)
1387 return;
1388 delta = tcp_jiffies32 - tp->lsndtime;
1389 if (delta > inet_csk(sk)->icsk_rto)
1390 tcp_cwnd_restart(sk, delta);
1391}
1392
1393/* Determine a window scaling and initial window to offer. */
1394void tcp_select_initial_window(const struct sock *sk, int __space,
1395 __u32 mss, __u32 *rcv_wnd,
1396 __u32 *window_clamp, int wscale_ok,
1397 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1398
1399static inline int tcp_win_from_space(const struct sock *sk, int space)
1400{
1401 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1402
1403 return tcp_adv_win_scale <= 0 ?
1404 (space>>(-tcp_adv_win_scale)) :
1405 space - (space>>tcp_adv_win_scale);
1406}
1407
1408/* Note: caller must be prepared to deal with negative returns */
1409static inline int tcp_space(const struct sock *sk)
1410{
1411 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1412 READ_ONCE(sk->sk_backlog.len) -
1413 atomic_read(&sk->sk_rmem_alloc));
1414}
1415
1416static inline int tcp_full_space(const struct sock *sk)
1417{
1418 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1419}
1420
1421void tcp_cleanup_rbuf(struct sock *sk, int copied);
1422
1423/* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1424 * If 87.5 % (7/8) of the space has been consumed, we want to override
1425 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1426 * len/truesize ratio.
1427 */
1428static inline bool tcp_rmem_pressure(const struct sock *sk)
1429{
1430 int rcvbuf, threshold;
1431
1432 if (tcp_under_memory_pressure(sk))
1433 return true;
1434
1435 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1436 threshold = rcvbuf - (rcvbuf >> 3);
1437
1438 return atomic_read(&sk->sk_rmem_alloc) > threshold;
1439}
1440
1441static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1442{
1443 const struct tcp_sock *tp = tcp_sk(sk);
1444 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1445
1446 if (avail <= 0)
1447 return false;
1448
1449 return (avail >= target) || tcp_rmem_pressure(sk) ||
1450 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1451}
1452
1453extern void tcp_openreq_init_rwin(struct request_sock *req,
1454 const struct sock *sk_listener,
1455 const struct dst_entry *dst);
1456
1457void tcp_enter_memory_pressure(struct sock *sk);
1458void tcp_leave_memory_pressure(struct sock *sk);
1459
1460static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1461{
1462 struct net *net = sock_net((struct sock *)tp);
1463
1464 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1465}
1466
1467static inline int keepalive_time_when(const struct tcp_sock *tp)
1468{
1469 struct net *net = sock_net((struct sock *)tp);
1470
1471 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1472}
1473
1474static inline int keepalive_probes(const struct tcp_sock *tp)
1475{
1476 struct net *net = sock_net((struct sock *)tp);
1477
1478 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1479}
1480
1481static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1482{
1483 const struct inet_connection_sock *icsk = &tp->inet_conn;
1484
1485 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1486 tcp_jiffies32 - tp->rcv_tstamp);
1487}
1488
1489static inline int tcp_fin_time(const struct sock *sk)
1490{
1491 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1492 const int rto = inet_csk(sk)->icsk_rto;
1493
1494 if (fin_timeout < (rto << 2) - (rto >> 1))
1495 fin_timeout = (rto << 2) - (rto >> 1);
1496
1497 return fin_timeout;
1498}
1499
1500static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1501 int paws_win)
1502{
1503 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1504 return true;
1505 if (unlikely(!time_before32(ktime_get_seconds(),
1506 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1507 return true;
1508 /*
1509 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1510 * then following tcp messages have valid values. Ignore 0 value,
1511 * or else 'negative' tsval might forbid us to accept their packets.
1512 */
1513 if (!rx_opt->ts_recent)
1514 return true;
1515 return false;
1516}
1517
1518static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1519 int rst)
1520{
1521 if (tcp_paws_check(rx_opt, 0))
1522 return false;
1523
1524 /* RST segments are not recommended to carry timestamp,
1525 and, if they do, it is recommended to ignore PAWS because
1526 "their cleanup function should take precedence over timestamps."
1527 Certainly, it is mistake. It is necessary to understand the reasons
1528 of this constraint to relax it: if peer reboots, clock may go
1529 out-of-sync and half-open connections will not be reset.
1530 Actually, the problem would be not existing if all
1531 the implementations followed draft about maintaining clock
1532 via reboots. Linux-2.2 DOES NOT!
1533
1534 However, we can relax time bounds for RST segments to MSL.
1535 */
1536 if (rst && !time_before32(ktime_get_seconds(),
1537 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1538 return false;
1539 return true;
1540}
1541
1542bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1543 int mib_idx, u32 *last_oow_ack_time);
1544
1545static inline void tcp_mib_init(struct net *net)
1546{
1547 /* See RFC 2012 */
1548 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1549 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1550 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1551 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1552}
1553
1554/* from STCP */
1555static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1556{
1557 tp->lost_skb_hint = NULL;
1558}
1559
1560static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1561{
1562 tcp_clear_retrans_hints_partial(tp);
1563 tp->retransmit_skb_hint = NULL;
1564}
1565
1566union tcp_md5_addr {
1567 struct in_addr a4;
1568#if IS_ENABLED(CONFIG_IPV6)
1569 struct in6_addr a6;
1570#endif
1571};
1572
1573/* - key database */
1574struct tcp_md5sig_key {
1575 struct hlist_node node;
1576 u8 keylen;
1577 u8 family; /* AF_INET or AF_INET6 */
1578 u8 prefixlen;
1579 union tcp_md5_addr addr;
1580 int l3index; /* set if key added with L3 scope */
1581 u8 key[TCP_MD5SIG_MAXKEYLEN];
1582 struct rcu_head rcu;
1583};
1584
1585/* - sock block */
1586struct tcp_md5sig_info {
1587 struct hlist_head head;
1588 struct rcu_head rcu;
1589};
1590
1591/* - pseudo header */
1592struct tcp4_pseudohdr {
1593 __be32 saddr;
1594 __be32 daddr;
1595 __u8 pad;
1596 __u8 protocol;
1597 __be16 len;
1598};
1599
1600struct tcp6_pseudohdr {
1601 struct in6_addr saddr;
1602 struct in6_addr daddr;
1603 __be32 len;
1604 __be32 protocol; /* including padding */
1605};
1606
1607union tcp_md5sum_block {
1608 struct tcp4_pseudohdr ip4;
1609#if IS_ENABLED(CONFIG_IPV6)
1610 struct tcp6_pseudohdr ip6;
1611#endif
1612};
1613
1614/* - pool: digest algorithm, hash description and scratch buffer */
1615struct tcp_md5sig_pool {
1616 struct ahash_request *md5_req;
1617 void *scratch;
1618};
1619
1620/* - functions */
1621int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1622 const struct sock *sk, const struct sk_buff *skb);
1623int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1624 int family, u8 prefixlen, int l3index,
1625 const u8 *newkey, u8 newkeylen, gfp_t gfp);
1626int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1627 int family, u8 prefixlen, int l3index);
1628struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1629 const struct sock *addr_sk);
1630
1631#ifdef CONFIG_TCP_MD5SIG
1632#include <linux/jump_label.h>
1633extern struct static_key_false tcp_md5_needed;
1634struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1635 const union tcp_md5_addr *addr,
1636 int family);
1637static inline struct tcp_md5sig_key *
1638tcp_md5_do_lookup(const struct sock *sk, int l3index,
1639 const union tcp_md5_addr *addr, int family)
1640{
1641 if (!static_branch_unlikely(&tcp_md5_needed))
1642 return NULL;
1643 return __tcp_md5_do_lookup(sk, l3index, addr, family);
1644}
1645
1646#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1647#else
1648static inline struct tcp_md5sig_key *
1649tcp_md5_do_lookup(const struct sock *sk, int l3index,
1650 const union tcp_md5_addr *addr, int family)
1651{
1652 return NULL;
1653}
1654#define tcp_twsk_md5_key(twsk) NULL
1655#endif
1656
1657bool tcp_alloc_md5sig_pool(void);
1658
1659struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1660static inline void tcp_put_md5sig_pool(void)
1661{
1662 local_bh_enable();
1663}
1664
1665int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1666 unsigned int header_len);
1667int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1668 const struct tcp_md5sig_key *key);
1669
1670/* From tcp_fastopen.c */
1671void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1672 struct tcp_fastopen_cookie *cookie);
1673void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1674 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1675 u16 try_exp);
1676struct tcp_fastopen_request {
1677 /* Fast Open cookie. Size 0 means a cookie request */
1678 struct tcp_fastopen_cookie cookie;
1679 struct msghdr *data; /* data in MSG_FASTOPEN */
1680 size_t size;
1681 int copied; /* queued in tcp_connect() */
1682 struct ubuf_info *uarg;
1683};
1684void tcp_free_fastopen_req(struct tcp_sock *tp);
1685void tcp_fastopen_destroy_cipher(struct sock *sk);
1686void tcp_fastopen_ctx_destroy(struct net *net);
1687int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1688 void *primary_key, void *backup_key);
1689int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1690 u64 *key);
1691void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1692struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1693 struct request_sock *req,
1694 struct tcp_fastopen_cookie *foc,
1695 const struct dst_entry *dst);
1696void tcp_fastopen_init_key_once(struct net *net);
1697bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1698 struct tcp_fastopen_cookie *cookie);
1699bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1700#define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1701#define TCP_FASTOPEN_KEY_MAX 2
1702#define TCP_FASTOPEN_KEY_BUF_LENGTH \
1703 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1704
1705/* Fastopen key context */
1706struct tcp_fastopen_context {
1707 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
1708 int num;
1709 struct rcu_head rcu;
1710};
1711
1712void tcp_fastopen_active_disable(struct sock *sk);
1713bool tcp_fastopen_active_should_disable(struct sock *sk);
1714void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1715void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1716
1717/* Caller needs to wrap with rcu_read_(un)lock() */
1718static inline
1719struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1720{
1721 struct tcp_fastopen_context *ctx;
1722
1723 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1724 if (!ctx)
1725 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1726 return ctx;
1727}
1728
1729static inline
1730bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1731 const struct tcp_fastopen_cookie *orig)
1732{
1733 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1734 orig->len == foc->len &&
1735 !memcmp(orig->val, foc->val, foc->len))
1736 return true;
1737 return false;
1738}
1739
1740static inline
1741int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1742{
1743 return ctx->num;
1744}
1745
1746/* Latencies incurred by various limits for a sender. They are
1747 * chronograph-like stats that are mutually exclusive.
1748 */
1749enum tcp_chrono {
1750 TCP_CHRONO_UNSPEC,
1751 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1752 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1753 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1754 __TCP_CHRONO_MAX,
1755};
1756
1757void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1758void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1759
1760/* This helper is needed, because skb->tcp_tsorted_anchor uses
1761 * the same memory storage than skb->destructor/_skb_refdst
1762 */
1763static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1764{
1765 skb->destructor = NULL;
1766 skb->_skb_refdst = 0UL;
1767}
1768
1769#define tcp_skb_tsorted_save(skb) { \
1770 unsigned long _save = skb->_skb_refdst; \
1771 skb->_skb_refdst = 0UL;
1772
1773#define tcp_skb_tsorted_restore(skb) \
1774 skb->_skb_refdst = _save; \
1775}
1776
1777void tcp_write_queue_purge(struct sock *sk);
1778
1779static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1780{
1781 return skb_rb_first(&sk->tcp_rtx_queue);
1782}
1783
1784static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1785{
1786 return skb_rb_last(&sk->tcp_rtx_queue);
1787}
1788
1789static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1790{
1791 return skb_peek(&sk->sk_write_queue);
1792}
1793
1794static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1795{
1796 return skb_peek_tail(&sk->sk_write_queue);
1797}
1798
1799#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1800 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1801
1802static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1803{
1804 return skb_peek(&sk->sk_write_queue);
1805}
1806
1807static inline bool tcp_skb_is_last(const struct sock *sk,
1808 const struct sk_buff *skb)
1809{
1810 return skb_queue_is_last(&sk->sk_write_queue, skb);
1811}
1812
1813/**
1814 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1815 * @sk: socket
1816 *
1817 * Since the write queue can have a temporary empty skb in it,
1818 * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1819 */
1820static inline bool tcp_write_queue_empty(const struct sock *sk)
1821{
1822 const struct tcp_sock *tp = tcp_sk(sk);
1823
1824 return tp->write_seq == tp->snd_nxt;
1825}
1826
1827static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1828{
1829 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1830}
1831
1832static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1833{
1834 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1835}
1836
1837static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1838{
1839 __skb_queue_tail(&sk->sk_write_queue, skb);
1840
1841 /* Queue it, remembering where we must start sending. */
1842 if (sk->sk_write_queue.next == skb)
1843 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1844}
1845
1846/* Insert new before skb on the write queue of sk. */
1847static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1848 struct sk_buff *skb,
1849 struct sock *sk)
1850{
1851 __skb_queue_before(&sk->sk_write_queue, skb, new);
1852}
1853
1854static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1855{
1856 tcp_skb_tsorted_anchor_cleanup(skb);
1857 __skb_unlink(skb, &sk->sk_write_queue);
1858}
1859
1860void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1861
1862static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1863{
1864 tcp_skb_tsorted_anchor_cleanup(skb);
1865 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1866}
1867
1868static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1869{
1870 list_del(&skb->tcp_tsorted_anchor);
1871 tcp_rtx_queue_unlink(skb, sk);
1872 sk_wmem_free_skb(sk, skb);
1873}
1874
1875static inline void tcp_push_pending_frames(struct sock *sk)
1876{
1877 if (tcp_send_head(sk)) {
1878 struct tcp_sock *tp = tcp_sk(sk);
1879
1880 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1881 }
1882}
1883
1884/* Start sequence of the skb just after the highest skb with SACKed
1885 * bit, valid only if sacked_out > 0 or when the caller has ensured
1886 * validity by itself.
1887 */
1888static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1889{
1890 if (!tp->sacked_out)
1891 return tp->snd_una;
1892
1893 if (tp->highest_sack == NULL)
1894 return tp->snd_nxt;
1895
1896 return TCP_SKB_CB(tp->highest_sack)->seq;
1897}
1898
1899static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1900{
1901 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1902}
1903
1904static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1905{
1906 return tcp_sk(sk)->highest_sack;
1907}
1908
1909static inline void tcp_highest_sack_reset(struct sock *sk)
1910{
1911 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1912}
1913
1914/* Called when old skb is about to be deleted and replaced by new skb */
1915static inline void tcp_highest_sack_replace(struct sock *sk,
1916 struct sk_buff *old,
1917 struct sk_buff *new)
1918{
1919 if (old == tcp_highest_sack(sk))
1920 tcp_sk(sk)->highest_sack = new;
1921}
1922
1923/* This helper checks if socket has IP_TRANSPARENT set */
1924static inline bool inet_sk_transparent(const struct sock *sk)
1925{
1926 switch (sk->sk_state) {
1927 case TCP_TIME_WAIT:
1928 return inet_twsk(sk)->tw_transparent;
1929 case TCP_NEW_SYN_RECV:
1930 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1931 }
1932 return inet_sk(sk)->transparent;
1933}
1934
1935/* Determines whether this is a thin stream (which may suffer from
1936 * increased latency). Used to trigger latency-reducing mechanisms.
1937 */
1938static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1939{
1940 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1941}
1942
1943/* /proc */
1944enum tcp_seq_states {
1945 TCP_SEQ_STATE_LISTENING,
1946 TCP_SEQ_STATE_ESTABLISHED,
1947};
1948
1949void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1950void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1951void tcp_seq_stop(struct seq_file *seq, void *v);
1952
1953struct tcp_seq_afinfo {
1954 sa_family_t family;
1955};
1956
1957struct tcp_iter_state {
1958 struct seq_net_private p;
1959 enum tcp_seq_states state;
1960 struct sock *syn_wait_sk;
1961 struct tcp_seq_afinfo *bpf_seq_afinfo;
1962 int bucket, offset, sbucket, num;
1963 loff_t last_pos;
1964};
1965
1966extern struct request_sock_ops tcp_request_sock_ops;
1967extern struct request_sock_ops tcp6_request_sock_ops;
1968
1969void tcp_v4_destroy_sock(struct sock *sk);
1970
1971struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1972 netdev_features_t features);
1973struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1974INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
1975INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
1976INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
1977INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
1978int tcp_gro_complete(struct sk_buff *skb);
1979
1980void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1981
1982static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1983{
1984 struct net *net = sock_net((struct sock *)tp);
1985 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1986}
1987
1988bool tcp_stream_memory_free(const struct sock *sk, int wake);
1989
1990#ifdef CONFIG_PROC_FS
1991int tcp4_proc_init(void);
1992void tcp4_proc_exit(void);
1993#endif
1994
1995int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1996int tcp_conn_request(struct request_sock_ops *rsk_ops,
1997 const struct tcp_request_sock_ops *af_ops,
1998 struct sock *sk, struct sk_buff *skb);
1999
2000/* TCP af-specific functions */
2001struct tcp_sock_af_ops {
2002#ifdef CONFIG_TCP_MD5SIG
2003 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
2004 const struct sock *addr_sk);
2005 int (*calc_md5_hash)(char *location,
2006 const struct tcp_md5sig_key *md5,
2007 const struct sock *sk,
2008 const struct sk_buff *skb);
2009 int (*md5_parse)(struct sock *sk,
2010 int optname,
2011 sockptr_t optval,
2012 int optlen);
2013#endif
2014};
2015
2016struct tcp_request_sock_ops {
2017 u16 mss_clamp;
2018#ifdef CONFIG_TCP_MD5SIG
2019 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2020 const struct sock *addr_sk);
2021 int (*calc_md5_hash) (char *location,
2022 const struct tcp_md5sig_key *md5,
2023 const struct sock *sk,
2024 const struct sk_buff *skb);
2025#endif
2026#ifdef CONFIG_SYN_COOKIES
2027 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
2028 __u16 *mss);
2029#endif
2030 struct dst_entry *(*route_req)(const struct sock *sk,
2031 struct sk_buff *skb,
2032 struct flowi *fl,
2033 struct request_sock *req);
2034 u32 (*init_seq)(const struct sk_buff *skb);
2035 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2036 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2037 struct flowi *fl, struct request_sock *req,
2038 struct tcp_fastopen_cookie *foc,
2039 enum tcp_synack_type synack_type,
2040 struct sk_buff *syn_skb);
2041};
2042
2043extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2044#if IS_ENABLED(CONFIG_IPV6)
2045extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2046#endif
2047
2048#ifdef CONFIG_SYN_COOKIES
2049static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2050 const struct sock *sk, struct sk_buff *skb,
2051 __u16 *mss)
2052{
2053 tcp_synq_overflow(sk);
2054 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2055 return ops->cookie_init_seq(skb, mss);
2056}
2057#else
2058static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2059 const struct sock *sk, struct sk_buff *skb,
2060 __u16 *mss)
2061{
2062 return 0;
2063}
2064#endif
2065
2066int tcpv4_offload_init(void);
2067
2068void tcp_v4_init(void);
2069void tcp_init(void);
2070
2071/* tcp_recovery.c */
2072void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2073void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2074extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2075 u32 reo_wnd);
2076extern bool tcp_rack_mark_lost(struct sock *sk);
2077extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2078 u64 xmit_time);
2079extern void tcp_rack_reo_timeout(struct sock *sk);
2080extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2081
2082/* At how many usecs into the future should the RTO fire? */
2083static inline s64 tcp_rto_delta_us(const struct sock *sk)
2084{
2085 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2086 u32 rto = inet_csk(sk)->icsk_rto;
2087 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2088
2089 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2090}
2091
2092/*
2093 * Save and compile IPv4 options, return a pointer to it
2094 */
2095static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2096 struct sk_buff *skb)
2097{
2098 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2099 struct ip_options_rcu *dopt = NULL;
2100
2101 if (opt->optlen) {
2102 int opt_size = sizeof(*dopt) + opt->optlen;
2103
2104 dopt = kmalloc(opt_size, GFP_ATOMIC);
2105 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2106 kfree(dopt);
2107 dopt = NULL;
2108 }
2109 }
2110 return dopt;
2111}
2112
2113/* locally generated TCP pure ACKs have skb->truesize == 2
2114 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2115 * This is much faster than dissecting the packet to find out.
2116 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2117 */
2118static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2119{
2120 return skb->truesize == 2;
2121}
2122
2123static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2124{
2125 skb->truesize = 2;
2126}
2127
2128static inline int tcp_inq(struct sock *sk)
2129{
2130 struct tcp_sock *tp = tcp_sk(sk);
2131 int answ;
2132
2133 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2134 answ = 0;
2135 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2136 !tp->urg_data ||
2137 before(tp->urg_seq, tp->copied_seq) ||
2138 !before(tp->urg_seq, tp->rcv_nxt)) {
2139
2140 answ = tp->rcv_nxt - tp->copied_seq;
2141
2142 /* Subtract 1, if FIN was received */
2143 if (answ && sock_flag(sk, SOCK_DONE))
2144 answ--;
2145 } else {
2146 answ = tp->urg_seq - tp->copied_seq;
2147 }
2148
2149 return answ;
2150}
2151
2152int tcp_peek_len(struct socket *sock);
2153
2154static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2155{
2156 u16 segs_in;
2157
2158 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2159 tp->segs_in += segs_in;
2160 if (skb->len > tcp_hdrlen(skb))
2161 tp->data_segs_in += segs_in;
2162}
2163
2164/*
2165 * TCP listen path runs lockless.
2166 * We forced "struct sock" to be const qualified to make sure
2167 * we don't modify one of its field by mistake.
2168 * Here, we increment sk_drops which is an atomic_t, so we can safely
2169 * make sock writable again.
2170 */
2171static inline void tcp_listendrop(const struct sock *sk)
2172{
2173 atomic_inc(&((struct sock *)sk)->sk_drops);
2174 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2175}
2176
2177enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2178
2179/*
2180 * Interface for adding Upper Level Protocols over TCP
2181 */
2182
2183#define TCP_ULP_NAME_MAX 16
2184#define TCP_ULP_MAX 128
2185#define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2186
2187struct tcp_ulp_ops {
2188 struct list_head list;
2189
2190 /* initialize ulp */
2191 int (*init)(struct sock *sk);
2192 /* update ulp */
2193 void (*update)(struct sock *sk, struct proto *p,
2194 void (*write_space)(struct sock *sk));
2195 /* cleanup ulp */
2196 void (*release)(struct sock *sk);
2197 /* diagnostic */
2198 int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2199 size_t (*get_info_size)(const struct sock *sk);
2200 /* clone ulp */
2201 void (*clone)(const struct request_sock *req, struct sock *newsk,
2202 const gfp_t priority);
2203
2204 char name[TCP_ULP_NAME_MAX];
2205 struct module *owner;
2206};
2207int tcp_register_ulp(struct tcp_ulp_ops *type);
2208void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2209int tcp_set_ulp(struct sock *sk, const char *name);
2210void tcp_get_available_ulp(char *buf, size_t len);
2211void tcp_cleanup_ulp(struct sock *sk);
2212void tcp_update_ulp(struct sock *sk, struct proto *p,
2213 void (*write_space)(struct sock *sk));
2214
2215#define MODULE_ALIAS_TCP_ULP(name) \
2216 __MODULE_INFO(alias, alias_userspace, name); \
2217 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2218
2219#ifdef CONFIG_NET_SOCK_MSG
2220struct sk_msg;
2221struct sk_psock;
2222
2223#ifdef CONFIG_BPF_SYSCALL
2224struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
2225int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
2226void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2227#endif /* CONFIG_BPF_SYSCALL */
2228
2229int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2230 int flags);
2231#endif /* CONFIG_NET_SOCK_MSG */
2232
2233#if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2234static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2235{
2236}
2237#endif
2238
2239#ifdef CONFIG_CGROUP_BPF
2240static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2241 struct sk_buff *skb,
2242 unsigned int end_offset)
2243{
2244 skops->skb = skb;
2245 skops->skb_data_end = skb->data + end_offset;
2246}
2247#else
2248static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2249 struct sk_buff *skb,
2250 unsigned int end_offset)
2251{
2252}
2253#endif
2254
2255/* Call BPF_SOCK_OPS program that returns an int. If the return value
2256 * is < 0, then the BPF op failed (for example if the loaded BPF
2257 * program does not support the chosen operation or there is no BPF
2258 * program loaded).
2259 */
2260#ifdef CONFIG_BPF
2261static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2262{
2263 struct bpf_sock_ops_kern sock_ops;
2264 int ret;
2265
2266 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2267 if (sk_fullsock(sk)) {
2268 sock_ops.is_fullsock = 1;
2269 sock_owned_by_me(sk);
2270 }
2271
2272 sock_ops.sk = sk;
2273 sock_ops.op = op;
2274 if (nargs > 0)
2275 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2276
2277 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2278 if (ret == 0)
2279 ret = sock_ops.reply;
2280 else
2281 ret = -1;
2282 return ret;
2283}
2284
2285static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2286{
2287 u32 args[2] = {arg1, arg2};
2288
2289 return tcp_call_bpf(sk, op, 2, args);
2290}
2291
2292static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2293 u32 arg3)
2294{
2295 u32 args[3] = {arg1, arg2, arg3};
2296
2297 return tcp_call_bpf(sk, op, 3, args);
2298}
2299
2300#else
2301static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2302{
2303 return -EPERM;
2304}
2305
2306static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2307{
2308 return -EPERM;
2309}
2310
2311static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2312 u32 arg3)
2313{
2314 return -EPERM;
2315}
2316
2317#endif
2318
2319static inline u32 tcp_timeout_init(struct sock *sk)
2320{
2321 int timeout;
2322
2323 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2324
2325 if (timeout <= 0)
2326 timeout = TCP_TIMEOUT_INIT;
2327 return timeout;
2328}
2329
2330static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2331{
2332 int rwnd;
2333
2334 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2335
2336 if (rwnd < 0)
2337 rwnd = 0;
2338 return rwnd;
2339}
2340
2341static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2342{
2343 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2344}
2345
2346static inline void tcp_bpf_rtt(struct sock *sk)
2347{
2348 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2349 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2350}
2351
2352#if IS_ENABLED(CONFIG_SMC)
2353extern struct static_key_false tcp_have_smc;
2354#endif
2355
2356#if IS_ENABLED(CONFIG_TLS_DEVICE)
2357void clean_acked_data_enable(struct inet_connection_sock *icsk,
2358 void (*cad)(struct sock *sk, u32 ack_seq));
2359void clean_acked_data_disable(struct inet_connection_sock *icsk);
2360void clean_acked_data_flush(void);
2361#endif
2362
2363DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2364static inline void tcp_add_tx_delay(struct sk_buff *skb,
2365 const struct tcp_sock *tp)
2366{
2367 if (static_branch_unlikely(&tcp_tx_delay_enabled))
2368 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2369}
2370
2371/* Compute Earliest Departure Time for some control packets
2372 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2373 */
2374static inline u64 tcp_transmit_time(const struct sock *sk)
2375{
2376 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2377 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2378 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2379
2380 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
2381 }
2382 return 0;
2383}
2384
2385#endif /* _TCP_H */