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/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Definitions for the TCP module.
7 *
8 * Version: @(#)tcp.h 1.0.5 05/23/93
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
17 */
18#ifndef _TCP_H
19#define _TCP_H
20
21#define FASTRETRANS_DEBUG 1
22
23#include <linux/list.h>
24#include <linux/tcp.h>
25#include <linux/bug.h>
26#include <linux/slab.h>
27#include <linux/cache.h>
28#include <linux/percpu.h>
29#include <linux/skbuff.h>
30#include <linux/cryptohash.h>
31#include <linux/kref.h>
32#include <linux/ktime.h>
33
34#include <net/inet_connection_sock.h>
35#include <net/inet_timewait_sock.h>
36#include <net/inet_hashtables.h>
37#include <net/checksum.h>
38#include <net/request_sock.h>
39#include <net/sock.h>
40#include <net/snmp.h>
41#include <net/ip.h>
42#include <net/tcp_states.h>
43#include <net/inet_ecn.h>
44#include <net/dst.h>
45
46#include <linux/seq_file.h>
47#include <linux/memcontrol.h>
48#include <linux/bpf-cgroup.h>
49
50extern struct inet_hashinfo tcp_hashinfo;
51
52extern struct percpu_counter tcp_orphan_count;
53void tcp_time_wait(struct sock *sk, int state, int timeo);
54
55#define MAX_TCP_HEADER (128 + MAX_HEADER)
56#define MAX_TCP_OPTION_SPACE 40
57
58/*
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
61 */
62#define MAX_TCP_WINDOW 32767U
63
64/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65#define TCP_MIN_MSS 88U
66
67/* The least MTU to use for probing */
68#define TCP_BASE_MSS 1024
69
70/* probing interval, default to 10 minutes as per RFC4821 */
71#define TCP_PROBE_INTERVAL 600
72
73/* Specify interval when tcp mtu probing will stop */
74#define TCP_PROBE_THRESHOLD 8
75
76/* After receiving this amount of duplicate ACKs fast retransmit starts. */
77#define TCP_FASTRETRANS_THRESH 3
78
79/* Maximal number of ACKs sent quickly to accelerate slow-start. */
80#define TCP_MAX_QUICKACKS 16U
81
82/* Maximal number of window scale according to RFC1323 */
83#define TCP_MAX_WSCALE 14U
84
85/* urg_data states */
86#define TCP_URG_VALID 0x0100
87#define TCP_URG_NOTYET 0x0200
88#define TCP_URG_READ 0x0400
89
90#define TCP_RETR1 3 /*
91 * This is how many retries it does before it
92 * tries to figure out if the gateway is
93 * down. Minimal RFC value is 3; it corresponds
94 * to ~3sec-8min depending on RTO.
95 */
96
97#define TCP_RETR2 15 /*
98 * This should take at least
99 * 90 minutes to time out.
100 * RFC1122 says that the limit is 100 sec.
101 * 15 is ~13-30min depending on RTO.
102 */
103
104#define TCP_SYN_RETRIES 6 /* This is how many retries are done
105 * when active opening a connection.
106 * RFC1122 says the minimum retry MUST
107 * be at least 180secs. Nevertheless
108 * this value is corresponding to
109 * 63secs of retransmission with the
110 * current initial RTO.
111 */
112
113#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
114 * when passive opening a connection.
115 * This is corresponding to 31secs of
116 * retransmission with the current
117 * initial RTO.
118 */
119
120#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
121 * state, about 60 seconds */
122#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
123 /* BSD style FIN_WAIT2 deadlock breaker.
124 * It used to be 3min, new value is 60sec,
125 * to combine FIN-WAIT-2 timeout with
126 * TIME-WAIT timer.
127 */
128
129#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
130#if HZ >= 100
131#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
132#define TCP_ATO_MIN ((unsigned)(HZ/25))
133#else
134#define TCP_DELACK_MIN 4U
135#define TCP_ATO_MIN 4U
136#endif
137#define TCP_RTO_MAX ((unsigned)(120*HZ))
138#define TCP_RTO_MIN ((unsigned)(HZ/5))
139#define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
140#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
141#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
142 * used as a fallback RTO for the
143 * initial data transmission if no
144 * valid RTT sample has been acquired,
145 * most likely due to retrans in 3WHS.
146 */
147
148#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
149 * for local resources.
150 */
151#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
152#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
153#define TCP_KEEPALIVE_INTVL (75*HZ)
154
155#define MAX_TCP_KEEPIDLE 32767
156#define MAX_TCP_KEEPINTVL 32767
157#define MAX_TCP_KEEPCNT 127
158#define MAX_TCP_SYNCNT 127
159
160#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
161
162#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
163#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
164 * after this time. It should be equal
165 * (or greater than) TCP_TIMEWAIT_LEN
166 * to provide reliability equal to one
167 * provided by timewait state.
168 */
169#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
170 * timestamps. It must be less than
171 * minimal timewait lifetime.
172 */
173/*
174 * TCP option
175 */
176
177#define TCPOPT_NOP 1 /* Padding */
178#define TCPOPT_EOL 0 /* End of options */
179#define TCPOPT_MSS 2 /* Segment size negotiating */
180#define TCPOPT_WINDOW 3 /* Window scaling */
181#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
182#define TCPOPT_SACK 5 /* SACK Block */
183#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
184#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
185#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
186#define TCPOPT_EXP 254 /* Experimental */
187/* Magic number to be after the option value for sharing TCP
188 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
189 */
190#define TCPOPT_FASTOPEN_MAGIC 0xF989
191#define TCPOPT_SMC_MAGIC 0xE2D4C3D9
192
193/*
194 * TCP option lengths
195 */
196
197#define TCPOLEN_MSS 4
198#define TCPOLEN_WINDOW 3
199#define TCPOLEN_SACK_PERM 2
200#define TCPOLEN_TIMESTAMP 10
201#define TCPOLEN_MD5SIG 18
202#define TCPOLEN_FASTOPEN_BASE 2
203#define TCPOLEN_EXP_FASTOPEN_BASE 4
204#define TCPOLEN_EXP_SMC_BASE 6
205
206/* But this is what stacks really send out. */
207#define TCPOLEN_TSTAMP_ALIGNED 12
208#define TCPOLEN_WSCALE_ALIGNED 4
209#define TCPOLEN_SACKPERM_ALIGNED 4
210#define TCPOLEN_SACK_BASE 2
211#define TCPOLEN_SACK_BASE_ALIGNED 4
212#define TCPOLEN_SACK_PERBLOCK 8
213#define TCPOLEN_MD5SIG_ALIGNED 20
214#define TCPOLEN_MSS_ALIGNED 4
215#define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
216
217/* Flags in tp->nonagle */
218#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
219#define TCP_NAGLE_CORK 2 /* Socket is corked */
220#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
221
222/* TCP thin-stream limits */
223#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
224
225/* TCP initial congestion window as per rfc6928 */
226#define TCP_INIT_CWND 10
227
228/* Bit Flags for sysctl_tcp_fastopen */
229#define TFO_CLIENT_ENABLE 1
230#define TFO_SERVER_ENABLE 2
231#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
232
233/* Accept SYN data w/o any cookie option */
234#define TFO_SERVER_COOKIE_NOT_REQD 0x200
235
236/* Force enable TFO on all listeners, i.e., not requiring the
237 * TCP_FASTOPEN socket option.
238 */
239#define TFO_SERVER_WO_SOCKOPT1 0x400
240
241
242/* sysctl variables for tcp */
243extern int sysctl_tcp_max_orphans;
244extern long sysctl_tcp_mem[3];
245
246#define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
247#define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
248
249extern atomic_long_t tcp_memory_allocated;
250extern struct percpu_counter tcp_sockets_allocated;
251extern unsigned long tcp_memory_pressure;
252
253/* optimized version of sk_under_memory_pressure() for TCP sockets */
254static inline bool tcp_under_memory_pressure(const struct sock *sk)
255{
256 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
257 mem_cgroup_under_socket_pressure(sk->sk_memcg))
258 return true;
259
260 return tcp_memory_pressure;
261}
262/*
263 * The next routines deal with comparing 32 bit unsigned ints
264 * and worry about wraparound (automatic with unsigned arithmetic).
265 */
266
267static inline bool before(__u32 seq1, __u32 seq2)
268{
269 return (__s32)(seq1-seq2) < 0;
270}
271#define after(seq2, seq1) before(seq1, seq2)
272
273/* is s2<=s1<=s3 ? */
274static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
275{
276 return seq3 - seq2 >= seq1 - seq2;
277}
278
279static inline bool tcp_out_of_memory(struct sock *sk)
280{
281 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
282 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
283 return true;
284 return false;
285}
286
287void sk_forced_mem_schedule(struct sock *sk, int size);
288
289static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
290{
291 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
292 int orphans = percpu_counter_read_positive(ocp);
293
294 if (orphans << shift > sysctl_tcp_max_orphans) {
295 orphans = percpu_counter_sum_positive(ocp);
296 if (orphans << shift > sysctl_tcp_max_orphans)
297 return true;
298 }
299 return false;
300}
301
302bool tcp_check_oom(struct sock *sk, int shift);
303
304
305extern struct proto tcp_prot;
306
307#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
308#define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
309#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
310#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
311
312void tcp_tasklet_init(void);
313
314void tcp_v4_err(struct sk_buff *skb, u32);
315
316void tcp_shutdown(struct sock *sk, int how);
317
318int tcp_v4_early_demux(struct sk_buff *skb);
319int tcp_v4_rcv(struct sk_buff *skb);
320
321int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
322int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
323int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
324int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
325 int flags);
326int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
327 size_t size, int flags);
328ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
329 size_t size, int flags);
330void tcp_release_cb(struct sock *sk);
331void tcp_wfree(struct sk_buff *skb);
332void tcp_write_timer_handler(struct sock *sk);
333void tcp_delack_timer_handler(struct sock *sk);
334int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
335int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
336void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
337 const struct tcphdr *th);
338void tcp_rcv_space_adjust(struct sock *sk);
339int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
340void tcp_twsk_destructor(struct sock *sk);
341ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
342 struct pipe_inode_info *pipe, size_t len,
343 unsigned int flags);
344
345static inline void tcp_dec_quickack_mode(struct sock *sk,
346 const unsigned int pkts)
347{
348 struct inet_connection_sock *icsk = inet_csk(sk);
349
350 if (icsk->icsk_ack.quick) {
351 if (pkts >= icsk->icsk_ack.quick) {
352 icsk->icsk_ack.quick = 0;
353 /* Leaving quickack mode we deflate ATO. */
354 icsk->icsk_ack.ato = TCP_ATO_MIN;
355 } else
356 icsk->icsk_ack.quick -= pkts;
357 }
358}
359
360#define TCP_ECN_OK 1
361#define TCP_ECN_QUEUE_CWR 2
362#define TCP_ECN_DEMAND_CWR 4
363#define TCP_ECN_SEEN 8
364
365enum tcp_tw_status {
366 TCP_TW_SUCCESS = 0,
367 TCP_TW_RST = 1,
368 TCP_TW_ACK = 2,
369 TCP_TW_SYN = 3
370};
371
372
373enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
374 struct sk_buff *skb,
375 const struct tcphdr *th);
376struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
377 struct request_sock *req, bool fastopen,
378 bool *lost_race);
379int tcp_child_process(struct sock *parent, struct sock *child,
380 struct sk_buff *skb);
381void tcp_enter_loss(struct sock *sk);
382void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
383void tcp_clear_retrans(struct tcp_sock *tp);
384void tcp_update_metrics(struct sock *sk);
385void tcp_init_metrics(struct sock *sk);
386void tcp_metrics_init(void);
387bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
388void tcp_close(struct sock *sk, long timeout);
389void tcp_init_sock(struct sock *sk);
390void tcp_init_transfer(struct sock *sk, int bpf_op);
391__poll_t tcp_poll(struct file *file, struct socket *sock,
392 struct poll_table_struct *wait);
393int tcp_getsockopt(struct sock *sk, int level, int optname,
394 char __user *optval, int __user *optlen);
395int tcp_setsockopt(struct sock *sk, int level, int optname,
396 char __user *optval, unsigned int optlen);
397int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
398 char __user *optval, int __user *optlen);
399int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
400 char __user *optval, unsigned int optlen);
401void tcp_set_keepalive(struct sock *sk, int val);
402void tcp_syn_ack_timeout(const struct request_sock *req);
403int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
404 int flags, int *addr_len);
405void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
406 struct tcp_options_received *opt_rx,
407 int estab, struct tcp_fastopen_cookie *foc);
408const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
409
410/*
411 * TCP v4 functions exported for the inet6 API
412 */
413
414void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
415void tcp_v4_mtu_reduced(struct sock *sk);
416void tcp_req_err(struct sock *sk, u32 seq, bool abort);
417int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
418struct sock *tcp_create_openreq_child(const struct sock *sk,
419 struct request_sock *req,
420 struct sk_buff *skb);
421void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
422struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
423 struct request_sock *req,
424 struct dst_entry *dst,
425 struct request_sock *req_unhash,
426 bool *own_req);
427int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
428int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
429int tcp_connect(struct sock *sk);
430enum tcp_synack_type {
431 TCP_SYNACK_NORMAL,
432 TCP_SYNACK_FASTOPEN,
433 TCP_SYNACK_COOKIE,
434};
435struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
436 struct request_sock *req,
437 struct tcp_fastopen_cookie *foc,
438 enum tcp_synack_type synack_type);
439int tcp_disconnect(struct sock *sk, int flags);
440
441void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
442int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
443void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
444
445/* From syncookies.c */
446struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
447 struct request_sock *req,
448 struct dst_entry *dst, u32 tsoff);
449int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
450 u32 cookie);
451struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
452#ifdef CONFIG_SYN_COOKIES
453
454/* Syncookies use a monotonic timer which increments every 60 seconds.
455 * This counter is used both as a hash input and partially encoded into
456 * the cookie value. A cookie is only validated further if the delta
457 * between the current counter value and the encoded one is less than this,
458 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
459 * the counter advances immediately after a cookie is generated).
460 */
461#define MAX_SYNCOOKIE_AGE 2
462#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
463#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
464
465/* syncookies: remember time of last synqueue overflow
466 * But do not dirty this field too often (once per second is enough)
467 * It is racy as we do not hold a lock, but race is very minor.
468 */
469static inline void tcp_synq_overflow(const struct sock *sk)
470{
471 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
472 unsigned long now = jiffies;
473
474 if (time_after(now, last_overflow + HZ))
475 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
476}
477
478/* syncookies: no recent synqueue overflow on this listening socket? */
479static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
480{
481 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
482
483 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
484}
485
486static inline u32 tcp_cookie_time(void)
487{
488 u64 val = get_jiffies_64();
489
490 do_div(val, TCP_SYNCOOKIE_PERIOD);
491 return val;
492}
493
494u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
495 u16 *mssp);
496__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
497u64 cookie_init_timestamp(struct request_sock *req);
498bool cookie_timestamp_decode(const struct net *net,
499 struct tcp_options_received *opt);
500bool cookie_ecn_ok(const struct tcp_options_received *opt,
501 const struct net *net, const struct dst_entry *dst);
502
503/* From net/ipv6/syncookies.c */
504int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
505 u32 cookie);
506struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
507
508u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
509 const struct tcphdr *th, u16 *mssp);
510__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
511#endif
512/* tcp_output.c */
513
514void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
515 int nonagle);
516int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
517int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
518void tcp_retransmit_timer(struct sock *sk);
519void tcp_xmit_retransmit_queue(struct sock *);
520void tcp_simple_retransmit(struct sock *);
521void tcp_enter_recovery(struct sock *sk, bool ece_ack);
522int tcp_trim_head(struct sock *, struct sk_buff *, u32);
523enum tcp_queue {
524 TCP_FRAG_IN_WRITE_QUEUE,
525 TCP_FRAG_IN_RTX_QUEUE,
526};
527int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
528 struct sk_buff *skb, u32 len,
529 unsigned int mss_now, gfp_t gfp);
530
531void tcp_send_probe0(struct sock *);
532void tcp_send_partial(struct sock *);
533int tcp_write_wakeup(struct sock *, int mib);
534void tcp_send_fin(struct sock *sk);
535void tcp_send_active_reset(struct sock *sk, gfp_t priority);
536int tcp_send_synack(struct sock *);
537void tcp_push_one(struct sock *, unsigned int mss_now);
538void tcp_send_ack(struct sock *sk);
539void tcp_send_delayed_ack(struct sock *sk);
540void tcp_send_loss_probe(struct sock *sk);
541bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
542void tcp_skb_collapse_tstamp(struct sk_buff *skb,
543 const struct sk_buff *next_skb);
544
545/* tcp_input.c */
546void tcp_rearm_rto(struct sock *sk);
547void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
548void tcp_reset(struct sock *sk);
549void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
550void tcp_fin(struct sock *sk);
551
552/* tcp_timer.c */
553void tcp_init_xmit_timers(struct sock *);
554static inline void tcp_clear_xmit_timers(struct sock *sk)
555{
556 hrtimer_cancel(&tcp_sk(sk)->pacing_timer);
557 inet_csk_clear_xmit_timers(sk);
558}
559
560unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
561unsigned int tcp_current_mss(struct sock *sk);
562
563/* Bound MSS / TSO packet size with the half of the window */
564static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
565{
566 int cutoff;
567
568 /* When peer uses tiny windows, there is no use in packetizing
569 * to sub-MSS pieces for the sake of SWS or making sure there
570 * are enough packets in the pipe for fast recovery.
571 *
572 * On the other hand, for extremely large MSS devices, handling
573 * smaller than MSS windows in this way does make sense.
574 */
575 if (tp->max_window > TCP_MSS_DEFAULT)
576 cutoff = (tp->max_window >> 1);
577 else
578 cutoff = tp->max_window;
579
580 if (cutoff && pktsize > cutoff)
581 return max_t(int, cutoff, 68U - tp->tcp_header_len);
582 else
583 return pktsize;
584}
585
586/* tcp.c */
587void tcp_get_info(struct sock *, struct tcp_info *);
588
589/* Read 'sendfile()'-style from a TCP socket */
590int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
591 sk_read_actor_t recv_actor);
592
593void tcp_initialize_rcv_mss(struct sock *sk);
594
595int tcp_mtu_to_mss(struct sock *sk, int pmtu);
596int tcp_mss_to_mtu(struct sock *sk, int mss);
597void tcp_mtup_init(struct sock *sk);
598void tcp_init_buffer_space(struct sock *sk);
599
600static inline void tcp_bound_rto(const struct sock *sk)
601{
602 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
603 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
604}
605
606static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
607{
608 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
609}
610
611static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
612{
613 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
614 ntohl(TCP_FLAG_ACK) |
615 snd_wnd);
616}
617
618static inline void tcp_fast_path_on(struct tcp_sock *tp)
619{
620 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
621}
622
623static inline void tcp_fast_path_check(struct sock *sk)
624{
625 struct tcp_sock *tp = tcp_sk(sk);
626
627 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
628 tp->rcv_wnd &&
629 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
630 !tp->urg_data)
631 tcp_fast_path_on(tp);
632}
633
634/* Compute the actual rto_min value */
635static inline u32 tcp_rto_min(struct sock *sk)
636{
637 const struct dst_entry *dst = __sk_dst_get(sk);
638 u32 rto_min = TCP_RTO_MIN;
639
640 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
641 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
642 return rto_min;
643}
644
645static inline u32 tcp_rto_min_us(struct sock *sk)
646{
647 return jiffies_to_usecs(tcp_rto_min(sk));
648}
649
650static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
651{
652 return dst_metric_locked(dst, RTAX_CC_ALGO);
653}
654
655/* Minimum RTT in usec. ~0 means not available. */
656static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
657{
658 return minmax_get(&tp->rtt_min);
659}
660
661/* Compute the actual receive window we are currently advertising.
662 * Rcv_nxt can be after the window if our peer push more data
663 * than the offered window.
664 */
665static inline u32 tcp_receive_window(const struct tcp_sock *tp)
666{
667 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
668
669 if (win < 0)
670 win = 0;
671 return (u32) win;
672}
673
674/* Choose a new window, without checks for shrinking, and without
675 * scaling applied to the result. The caller does these things
676 * if necessary. This is a "raw" window selection.
677 */
678u32 __tcp_select_window(struct sock *sk);
679
680void tcp_send_window_probe(struct sock *sk);
681
682/* TCP uses 32bit jiffies to save some space.
683 * Note that this is different from tcp_time_stamp, which
684 * historically has been the same until linux-4.13.
685 */
686#define tcp_jiffies32 ((u32)jiffies)
687
688/*
689 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
690 * It is no longer tied to jiffies, but to 1 ms clock.
691 * Note: double check if you want to use tcp_jiffies32 instead of this.
692 */
693#define TCP_TS_HZ 1000
694
695static inline u64 tcp_clock_ns(void)
696{
697 return local_clock();
698}
699
700static inline u64 tcp_clock_us(void)
701{
702 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
703}
704
705/* This should only be used in contexts where tp->tcp_mstamp is up to date */
706static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
707{
708 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
709}
710
711/* Could use tcp_clock_us() / 1000, but this version uses a single divide */
712static inline u32 tcp_time_stamp_raw(void)
713{
714 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
715}
716
717
718/* Refresh 1us clock of a TCP socket,
719 * ensuring monotically increasing values.
720 */
721static inline void tcp_mstamp_refresh(struct tcp_sock *tp)
722{
723 u64 val = tcp_clock_us();
724
725 if (val > tp->tcp_mstamp)
726 tp->tcp_mstamp = val;
727}
728
729static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
730{
731 return max_t(s64, t1 - t0, 0);
732}
733
734static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
735{
736 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ);
737}
738
739
740#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
741
742#define TCPHDR_FIN 0x01
743#define TCPHDR_SYN 0x02
744#define TCPHDR_RST 0x04
745#define TCPHDR_PSH 0x08
746#define TCPHDR_ACK 0x10
747#define TCPHDR_URG 0x20
748#define TCPHDR_ECE 0x40
749#define TCPHDR_CWR 0x80
750
751#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
752
753/* This is what the send packet queuing engine uses to pass
754 * TCP per-packet control information to the transmission code.
755 * We also store the host-order sequence numbers in here too.
756 * This is 44 bytes if IPV6 is enabled.
757 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
758 */
759struct tcp_skb_cb {
760 __u32 seq; /* Starting sequence number */
761 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
762 union {
763 /* Note : tcp_tw_isn is used in input path only
764 * (isn chosen by tcp_timewait_state_process())
765 *
766 * tcp_gso_segs/size are used in write queue only,
767 * cf tcp_skb_pcount()/tcp_skb_mss()
768 */
769 __u32 tcp_tw_isn;
770 struct {
771 u16 tcp_gso_segs;
772 u16 tcp_gso_size;
773 };
774 };
775 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
776
777 __u8 sacked; /* State flags for SACK. */
778#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
779#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
780#define TCPCB_LOST 0x04 /* SKB is lost */
781#define TCPCB_TAGBITS 0x07 /* All tag bits */
782#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
783#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
784#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
785 TCPCB_REPAIRED)
786
787 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
788 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
789 eor:1, /* Is skb MSG_EOR marked? */
790 has_rxtstamp:1, /* SKB has a RX timestamp */
791 unused:5;
792 __u32 ack_seq; /* Sequence number ACK'd */
793 union {
794 struct {
795 /* There is space for up to 24 bytes */
796 __u32 in_flight:30,/* Bytes in flight at transmit */
797 is_app_limited:1, /* cwnd not fully used? */
798 unused:1;
799 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
800 __u32 delivered;
801 /* start of send pipeline phase */
802 u64 first_tx_mstamp;
803 /* when we reached the "delivered" count */
804 u64 delivered_mstamp;
805 } tx; /* only used for outgoing skbs */
806 union {
807 struct inet_skb_parm h4;
808#if IS_ENABLED(CONFIG_IPV6)
809 struct inet6_skb_parm h6;
810#endif
811 } header; /* For incoming skbs */
812 struct {
813 __u32 key;
814 __u32 flags;
815 struct bpf_map *map;
816 void *data_end;
817 } bpf;
818 };
819};
820
821#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
822
823
824#if IS_ENABLED(CONFIG_IPV6)
825/* This is the variant of inet6_iif() that must be used by TCP,
826 * as TCP moves IP6CB into a different location in skb->cb[]
827 */
828static inline int tcp_v6_iif(const struct sk_buff *skb)
829{
830 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
831
832 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
833}
834
835/* TCP_SKB_CB reference means this can not be used from early demux */
836static inline int tcp_v6_sdif(const struct sk_buff *skb)
837{
838#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
839 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
840 return TCP_SKB_CB(skb)->header.h6.iif;
841#endif
842 return 0;
843}
844#endif
845
846static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
847{
848#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
849 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
850 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
851 return true;
852#endif
853 return false;
854}
855
856/* TCP_SKB_CB reference means this can not be used from early demux */
857static inline int tcp_v4_sdif(struct sk_buff *skb)
858{
859#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
860 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
861 return TCP_SKB_CB(skb)->header.h4.iif;
862#endif
863 return 0;
864}
865
866/* Due to TSO, an SKB can be composed of multiple actual
867 * packets. To keep these tracked properly, we use this.
868 */
869static inline int tcp_skb_pcount(const struct sk_buff *skb)
870{
871 return TCP_SKB_CB(skb)->tcp_gso_segs;
872}
873
874static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
875{
876 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
877}
878
879static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
880{
881 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
882}
883
884/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
885static inline int tcp_skb_mss(const struct sk_buff *skb)
886{
887 return TCP_SKB_CB(skb)->tcp_gso_size;
888}
889
890static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
891{
892 return likely(!TCP_SKB_CB(skb)->eor);
893}
894
895/* Events passed to congestion control interface */
896enum tcp_ca_event {
897 CA_EVENT_TX_START, /* first transmit when no packets in flight */
898 CA_EVENT_CWND_RESTART, /* congestion window restart */
899 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
900 CA_EVENT_LOSS, /* loss timeout */
901 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
902 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
903 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
904 CA_EVENT_NON_DELAYED_ACK,
905};
906
907/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
908enum tcp_ca_ack_event_flags {
909 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
910 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
911 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
912};
913
914/*
915 * Interface for adding new TCP congestion control handlers
916 */
917#define TCP_CA_NAME_MAX 16
918#define TCP_CA_MAX 128
919#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
920
921#define TCP_CA_UNSPEC 0
922
923/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
924#define TCP_CONG_NON_RESTRICTED 0x1
925/* Requires ECN/ECT set on all packets */
926#define TCP_CONG_NEEDS_ECN 0x2
927
928union tcp_cc_info;
929
930struct ack_sample {
931 u32 pkts_acked;
932 s32 rtt_us;
933 u32 in_flight;
934};
935
936/* A rate sample measures the number of (original/retransmitted) data
937 * packets delivered "delivered" over an interval of time "interval_us".
938 * The tcp_rate.c code fills in the rate sample, and congestion
939 * control modules that define a cong_control function to run at the end
940 * of ACK processing can optionally chose to consult this sample when
941 * setting cwnd and pacing rate.
942 * A sample is invalid if "delivered" or "interval_us" is negative.
943 */
944struct rate_sample {
945 u64 prior_mstamp; /* starting timestamp for interval */
946 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
947 s32 delivered; /* number of packets delivered over interval */
948 long interval_us; /* time for tp->delivered to incr "delivered" */
949 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
950 int losses; /* number of packets marked lost upon ACK */
951 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
952 u32 prior_in_flight; /* in flight before this ACK */
953 bool is_app_limited; /* is sample from packet with bubble in pipe? */
954 bool is_retrans; /* is sample from retransmission? */
955 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
956};
957
958struct tcp_congestion_ops {
959 struct list_head list;
960 u32 key;
961 u32 flags;
962
963 /* initialize private data (optional) */
964 void (*init)(struct sock *sk);
965 /* cleanup private data (optional) */
966 void (*release)(struct sock *sk);
967
968 /* return slow start threshold (required) */
969 u32 (*ssthresh)(struct sock *sk);
970 /* do new cwnd calculation (required) */
971 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
972 /* call before changing ca_state (optional) */
973 void (*set_state)(struct sock *sk, u8 new_state);
974 /* call when cwnd event occurs (optional) */
975 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
976 /* call when ack arrives (optional) */
977 void (*in_ack_event)(struct sock *sk, u32 flags);
978 /* new value of cwnd after loss (required) */
979 u32 (*undo_cwnd)(struct sock *sk);
980 /* hook for packet ack accounting (optional) */
981 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
982 /* override sysctl_tcp_min_tso_segs */
983 u32 (*min_tso_segs)(struct sock *sk);
984 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
985 u32 (*sndbuf_expand)(struct sock *sk);
986 /* call when packets are delivered to update cwnd and pacing rate,
987 * after all the ca_state processing. (optional)
988 */
989 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
990 /* get info for inet_diag (optional) */
991 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
992 union tcp_cc_info *info);
993
994 char name[TCP_CA_NAME_MAX];
995 struct module *owner;
996};
997
998int tcp_register_congestion_control(struct tcp_congestion_ops *type);
999void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1000
1001void tcp_assign_congestion_control(struct sock *sk);
1002void tcp_init_congestion_control(struct sock *sk);
1003void tcp_cleanup_congestion_control(struct sock *sk);
1004int tcp_set_default_congestion_control(struct net *net, const char *name);
1005void tcp_get_default_congestion_control(struct net *net, char *name);
1006void tcp_get_available_congestion_control(char *buf, size_t len);
1007void tcp_get_allowed_congestion_control(char *buf, size_t len);
1008int tcp_set_allowed_congestion_control(char *allowed);
1009int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit);
1010u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1011void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1012
1013u32 tcp_reno_ssthresh(struct sock *sk);
1014u32 tcp_reno_undo_cwnd(struct sock *sk);
1015void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1016extern struct tcp_congestion_ops tcp_reno;
1017
1018struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1019u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1020#ifdef CONFIG_INET
1021char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1022#else
1023static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1024{
1025 return NULL;
1026}
1027#endif
1028
1029static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1030{
1031 const struct inet_connection_sock *icsk = inet_csk(sk);
1032
1033 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1034}
1035
1036static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1037{
1038 struct inet_connection_sock *icsk = inet_csk(sk);
1039
1040 if (icsk->icsk_ca_ops->set_state)
1041 icsk->icsk_ca_ops->set_state(sk, ca_state);
1042 icsk->icsk_ca_state = ca_state;
1043}
1044
1045static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1046{
1047 const struct inet_connection_sock *icsk = inet_csk(sk);
1048
1049 if (icsk->icsk_ca_ops->cwnd_event)
1050 icsk->icsk_ca_ops->cwnd_event(sk, event);
1051}
1052
1053/* From tcp_rate.c */
1054void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1055void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1056 struct rate_sample *rs);
1057void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1058 bool is_sack_reneg, struct rate_sample *rs);
1059void tcp_rate_check_app_limited(struct sock *sk);
1060
1061/* These functions determine how the current flow behaves in respect of SACK
1062 * handling. SACK is negotiated with the peer, and therefore it can vary
1063 * between different flows.
1064 *
1065 * tcp_is_sack - SACK enabled
1066 * tcp_is_reno - No SACK
1067 */
1068static inline int tcp_is_sack(const struct tcp_sock *tp)
1069{
1070 return tp->rx_opt.sack_ok;
1071}
1072
1073static inline bool tcp_is_reno(const struct tcp_sock *tp)
1074{
1075 return !tcp_is_sack(tp);
1076}
1077
1078static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1079{
1080 return tp->sacked_out + tp->lost_out;
1081}
1082
1083/* This determines how many packets are "in the network" to the best
1084 * of our knowledge. In many cases it is conservative, but where
1085 * detailed information is available from the receiver (via SACK
1086 * blocks etc.) we can make more aggressive calculations.
1087 *
1088 * Use this for decisions involving congestion control, use just
1089 * tp->packets_out to determine if the send queue is empty or not.
1090 *
1091 * Read this equation as:
1092 *
1093 * "Packets sent once on transmission queue" MINUS
1094 * "Packets left network, but not honestly ACKed yet" PLUS
1095 * "Packets fast retransmitted"
1096 */
1097static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1098{
1099 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1100}
1101
1102#define TCP_INFINITE_SSTHRESH 0x7fffffff
1103
1104static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1105{
1106 return tp->snd_cwnd < tp->snd_ssthresh;
1107}
1108
1109static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1110{
1111 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1112}
1113
1114static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1115{
1116 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1117 (1 << inet_csk(sk)->icsk_ca_state);
1118}
1119
1120/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1121 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1122 * ssthresh.
1123 */
1124static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1125{
1126 const struct tcp_sock *tp = tcp_sk(sk);
1127
1128 if (tcp_in_cwnd_reduction(sk))
1129 return tp->snd_ssthresh;
1130 else
1131 return max(tp->snd_ssthresh,
1132 ((tp->snd_cwnd >> 1) +
1133 (tp->snd_cwnd >> 2)));
1134}
1135
1136/* Use define here intentionally to get WARN_ON location shown at the caller */
1137#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1138
1139void tcp_enter_cwr(struct sock *sk);
1140__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1141
1142/* The maximum number of MSS of available cwnd for which TSO defers
1143 * sending if not using sysctl_tcp_tso_win_divisor.
1144 */
1145static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1146{
1147 return 3;
1148}
1149
1150/* Returns end sequence number of the receiver's advertised window */
1151static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1152{
1153 return tp->snd_una + tp->snd_wnd;
1154}
1155
1156/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1157 * flexible approach. The RFC suggests cwnd should not be raised unless
1158 * it was fully used previously. And that's exactly what we do in
1159 * congestion avoidance mode. But in slow start we allow cwnd to grow
1160 * as long as the application has used half the cwnd.
1161 * Example :
1162 * cwnd is 10 (IW10), but application sends 9 frames.
1163 * We allow cwnd to reach 18 when all frames are ACKed.
1164 * This check is safe because it's as aggressive as slow start which already
1165 * risks 100% overshoot. The advantage is that we discourage application to
1166 * either send more filler packets or data to artificially blow up the cwnd
1167 * usage, and allow application-limited process to probe bw more aggressively.
1168 */
1169static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1170{
1171 const struct tcp_sock *tp = tcp_sk(sk);
1172
1173 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1174 if (tcp_in_slow_start(tp))
1175 return tp->snd_cwnd < 2 * tp->max_packets_out;
1176
1177 return tp->is_cwnd_limited;
1178}
1179
1180/* Something is really bad, we could not queue an additional packet,
1181 * because qdisc is full or receiver sent a 0 window.
1182 * We do not want to add fuel to the fire, or abort too early,
1183 * so make sure the timer we arm now is at least 200ms in the future,
1184 * regardless of current icsk_rto value (as it could be ~2ms)
1185 */
1186static inline unsigned long tcp_probe0_base(const struct sock *sk)
1187{
1188 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1189}
1190
1191/* Variant of inet_csk_rto_backoff() used for zero window probes */
1192static inline unsigned long tcp_probe0_when(const struct sock *sk,
1193 unsigned long max_when)
1194{
1195 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1196
1197 return (unsigned long)min_t(u64, when, max_when);
1198}
1199
1200static inline void tcp_check_probe_timer(struct sock *sk)
1201{
1202 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1203 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1204 tcp_probe0_base(sk), TCP_RTO_MAX);
1205}
1206
1207static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1208{
1209 tp->snd_wl1 = seq;
1210}
1211
1212static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1213{
1214 tp->snd_wl1 = seq;
1215}
1216
1217/*
1218 * Calculate(/check) TCP checksum
1219 */
1220static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1221 __be32 daddr, __wsum base)
1222{
1223 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1224}
1225
1226static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1227{
1228 return __skb_checksum_complete(skb);
1229}
1230
1231static inline bool tcp_checksum_complete(struct sk_buff *skb)
1232{
1233 return !skb_csum_unnecessary(skb) &&
1234 __tcp_checksum_complete(skb);
1235}
1236
1237bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1238int tcp_filter(struct sock *sk, struct sk_buff *skb);
1239
1240#undef STATE_TRACE
1241
1242#ifdef STATE_TRACE
1243static const char *statename[]={
1244 "Unused","Established","Syn Sent","Syn Recv",
1245 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1246 "Close Wait","Last ACK","Listen","Closing"
1247};
1248#endif
1249void tcp_set_state(struct sock *sk, int state);
1250
1251void tcp_done(struct sock *sk);
1252
1253int tcp_abort(struct sock *sk, int err);
1254
1255static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1256{
1257 rx_opt->dsack = 0;
1258 rx_opt->num_sacks = 0;
1259}
1260
1261u32 tcp_default_init_rwnd(u32 mss);
1262void tcp_cwnd_restart(struct sock *sk, s32 delta);
1263
1264static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1265{
1266 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1267 struct tcp_sock *tp = tcp_sk(sk);
1268 s32 delta;
1269
1270 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1271 ca_ops->cong_control)
1272 return;
1273 delta = tcp_jiffies32 - tp->lsndtime;
1274 if (delta > inet_csk(sk)->icsk_rto)
1275 tcp_cwnd_restart(sk, delta);
1276}
1277
1278/* Determine a window scaling and initial window to offer. */
1279void tcp_select_initial_window(const struct sock *sk, int __space,
1280 __u32 mss, __u32 *rcv_wnd,
1281 __u32 *window_clamp, int wscale_ok,
1282 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1283
1284static inline int tcp_win_from_space(const struct sock *sk, int space)
1285{
1286 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1287
1288 return tcp_adv_win_scale <= 0 ?
1289 (space>>(-tcp_adv_win_scale)) :
1290 space - (space>>tcp_adv_win_scale);
1291}
1292
1293/* Note: caller must be prepared to deal with negative returns */
1294static inline int tcp_space(const struct sock *sk)
1295{
1296 return tcp_win_from_space(sk, sk->sk_rcvbuf -
1297 atomic_read(&sk->sk_rmem_alloc));
1298}
1299
1300static inline int tcp_full_space(const struct sock *sk)
1301{
1302 return tcp_win_from_space(sk, sk->sk_rcvbuf);
1303}
1304
1305extern void tcp_openreq_init_rwin(struct request_sock *req,
1306 const struct sock *sk_listener,
1307 const struct dst_entry *dst);
1308
1309void tcp_enter_memory_pressure(struct sock *sk);
1310void tcp_leave_memory_pressure(struct sock *sk);
1311
1312static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1313{
1314 struct net *net = sock_net((struct sock *)tp);
1315
1316 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1317}
1318
1319static inline int keepalive_time_when(const struct tcp_sock *tp)
1320{
1321 struct net *net = sock_net((struct sock *)tp);
1322
1323 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1324}
1325
1326static inline int keepalive_probes(const struct tcp_sock *tp)
1327{
1328 struct net *net = sock_net((struct sock *)tp);
1329
1330 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1331}
1332
1333static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1334{
1335 const struct inet_connection_sock *icsk = &tp->inet_conn;
1336
1337 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1338 tcp_jiffies32 - tp->rcv_tstamp);
1339}
1340
1341static inline int tcp_fin_time(const struct sock *sk)
1342{
1343 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1344 const int rto = inet_csk(sk)->icsk_rto;
1345
1346 if (fin_timeout < (rto << 2) - (rto >> 1))
1347 fin_timeout = (rto << 2) - (rto >> 1);
1348
1349 return fin_timeout;
1350}
1351
1352static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1353 int paws_win)
1354{
1355 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1356 return true;
1357 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1358 return true;
1359 /*
1360 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1361 * then following tcp messages have valid values. Ignore 0 value,
1362 * or else 'negative' tsval might forbid us to accept their packets.
1363 */
1364 if (!rx_opt->ts_recent)
1365 return true;
1366 return false;
1367}
1368
1369static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1370 int rst)
1371{
1372 if (tcp_paws_check(rx_opt, 0))
1373 return false;
1374
1375 /* RST segments are not recommended to carry timestamp,
1376 and, if they do, it is recommended to ignore PAWS because
1377 "their cleanup function should take precedence over timestamps."
1378 Certainly, it is mistake. It is necessary to understand the reasons
1379 of this constraint to relax it: if peer reboots, clock may go
1380 out-of-sync and half-open connections will not be reset.
1381 Actually, the problem would be not existing if all
1382 the implementations followed draft about maintaining clock
1383 via reboots. Linux-2.2 DOES NOT!
1384
1385 However, we can relax time bounds for RST segments to MSL.
1386 */
1387 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1388 return false;
1389 return true;
1390}
1391
1392bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1393 int mib_idx, u32 *last_oow_ack_time);
1394
1395static inline void tcp_mib_init(struct net *net)
1396{
1397 /* See RFC 2012 */
1398 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1399 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1400 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1401 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1402}
1403
1404/* from STCP */
1405static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1406{
1407 tp->lost_skb_hint = NULL;
1408}
1409
1410static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1411{
1412 tcp_clear_retrans_hints_partial(tp);
1413 tp->retransmit_skb_hint = NULL;
1414}
1415
1416union tcp_md5_addr {
1417 struct in_addr a4;
1418#if IS_ENABLED(CONFIG_IPV6)
1419 struct in6_addr a6;
1420#endif
1421};
1422
1423/* - key database */
1424struct tcp_md5sig_key {
1425 struct hlist_node node;
1426 u8 keylen;
1427 u8 family; /* AF_INET or AF_INET6 */
1428 union tcp_md5_addr addr;
1429 u8 prefixlen;
1430 u8 key[TCP_MD5SIG_MAXKEYLEN];
1431 struct rcu_head rcu;
1432};
1433
1434/* - sock block */
1435struct tcp_md5sig_info {
1436 struct hlist_head head;
1437 struct rcu_head rcu;
1438};
1439
1440/* - pseudo header */
1441struct tcp4_pseudohdr {
1442 __be32 saddr;
1443 __be32 daddr;
1444 __u8 pad;
1445 __u8 protocol;
1446 __be16 len;
1447};
1448
1449struct tcp6_pseudohdr {
1450 struct in6_addr saddr;
1451 struct in6_addr daddr;
1452 __be32 len;
1453 __be32 protocol; /* including padding */
1454};
1455
1456union tcp_md5sum_block {
1457 struct tcp4_pseudohdr ip4;
1458#if IS_ENABLED(CONFIG_IPV6)
1459 struct tcp6_pseudohdr ip6;
1460#endif
1461};
1462
1463/* - pool: digest algorithm, hash description and scratch buffer */
1464struct tcp_md5sig_pool {
1465 struct ahash_request *md5_req;
1466 void *scratch;
1467};
1468
1469/* - functions */
1470int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1471 const struct sock *sk, const struct sk_buff *skb);
1472int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1473 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1474 gfp_t gfp);
1475int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1476 int family, u8 prefixlen);
1477struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1478 const struct sock *addr_sk);
1479
1480#ifdef CONFIG_TCP_MD5SIG
1481struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1482 const union tcp_md5_addr *addr,
1483 int family);
1484#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1485#else
1486static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1487 const union tcp_md5_addr *addr,
1488 int family)
1489{
1490 return NULL;
1491}
1492#define tcp_twsk_md5_key(twsk) NULL
1493#endif
1494
1495bool tcp_alloc_md5sig_pool(void);
1496
1497struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1498static inline void tcp_put_md5sig_pool(void)
1499{
1500 local_bh_enable();
1501}
1502
1503int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1504 unsigned int header_len);
1505int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1506 const struct tcp_md5sig_key *key);
1507
1508/* From tcp_fastopen.c */
1509void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1510 struct tcp_fastopen_cookie *cookie);
1511void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1512 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1513 u16 try_exp);
1514struct tcp_fastopen_request {
1515 /* Fast Open cookie. Size 0 means a cookie request */
1516 struct tcp_fastopen_cookie cookie;
1517 struct msghdr *data; /* data in MSG_FASTOPEN */
1518 size_t size;
1519 int copied; /* queued in tcp_connect() */
1520};
1521void tcp_free_fastopen_req(struct tcp_sock *tp);
1522void tcp_fastopen_destroy_cipher(struct sock *sk);
1523void tcp_fastopen_ctx_destroy(struct net *net);
1524int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1525 void *key, unsigned int len);
1526void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1527struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1528 struct request_sock *req,
1529 struct tcp_fastopen_cookie *foc,
1530 const struct dst_entry *dst);
1531void tcp_fastopen_init_key_once(struct net *net);
1532bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1533 struct tcp_fastopen_cookie *cookie);
1534bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1535#define TCP_FASTOPEN_KEY_LENGTH 16
1536
1537/* Fastopen key context */
1538struct tcp_fastopen_context {
1539 struct crypto_cipher *tfm;
1540 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1541 struct rcu_head rcu;
1542};
1543
1544extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1545void tcp_fastopen_active_disable(struct sock *sk);
1546bool tcp_fastopen_active_should_disable(struct sock *sk);
1547void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1548void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1549
1550/* Latencies incurred by various limits for a sender. They are
1551 * chronograph-like stats that are mutually exclusive.
1552 */
1553enum tcp_chrono {
1554 TCP_CHRONO_UNSPEC,
1555 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1556 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1557 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1558 __TCP_CHRONO_MAX,
1559};
1560
1561void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1562void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1563
1564/* This helper is needed, because skb->tcp_tsorted_anchor uses
1565 * the same memory storage than skb->destructor/_skb_refdst
1566 */
1567static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1568{
1569 skb->destructor = NULL;
1570 skb->_skb_refdst = 0UL;
1571}
1572
1573#define tcp_skb_tsorted_save(skb) { \
1574 unsigned long _save = skb->_skb_refdst; \
1575 skb->_skb_refdst = 0UL;
1576
1577#define tcp_skb_tsorted_restore(skb) \
1578 skb->_skb_refdst = _save; \
1579}
1580
1581void tcp_write_queue_purge(struct sock *sk);
1582
1583static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1584{
1585 return skb_rb_first(&sk->tcp_rtx_queue);
1586}
1587
1588static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1589{
1590 return skb_peek(&sk->sk_write_queue);
1591}
1592
1593static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1594{
1595 return skb_peek_tail(&sk->sk_write_queue);
1596}
1597
1598#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1599 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1600
1601static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1602{
1603 return skb_peek(&sk->sk_write_queue);
1604}
1605
1606static inline bool tcp_skb_is_last(const struct sock *sk,
1607 const struct sk_buff *skb)
1608{
1609 return skb_queue_is_last(&sk->sk_write_queue, skb);
1610}
1611
1612static inline bool tcp_write_queue_empty(const struct sock *sk)
1613{
1614 return skb_queue_empty(&sk->sk_write_queue);
1615}
1616
1617static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1618{
1619 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1620}
1621
1622static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1623{
1624 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1625}
1626
1627static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1628{
1629 if (tcp_write_queue_empty(sk))
1630 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1631}
1632
1633static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1634{
1635 __skb_queue_tail(&sk->sk_write_queue, skb);
1636}
1637
1638static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1639{
1640 __tcp_add_write_queue_tail(sk, skb);
1641
1642 /* Queue it, remembering where we must start sending. */
1643 if (sk->sk_write_queue.next == skb)
1644 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1645}
1646
1647/* Insert new before skb on the write queue of sk. */
1648static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1649 struct sk_buff *skb,
1650 struct sock *sk)
1651{
1652 __skb_queue_before(&sk->sk_write_queue, skb, new);
1653}
1654
1655static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1656{
1657 tcp_skb_tsorted_anchor_cleanup(skb);
1658 __skb_unlink(skb, &sk->sk_write_queue);
1659}
1660
1661void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1662
1663static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1664{
1665 tcp_skb_tsorted_anchor_cleanup(skb);
1666 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1667}
1668
1669static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1670{
1671 list_del(&skb->tcp_tsorted_anchor);
1672 tcp_rtx_queue_unlink(skb, sk);
1673 sk_wmem_free_skb(sk, skb);
1674}
1675
1676static inline void tcp_push_pending_frames(struct sock *sk)
1677{
1678 if (tcp_send_head(sk)) {
1679 struct tcp_sock *tp = tcp_sk(sk);
1680
1681 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1682 }
1683}
1684
1685/* Start sequence of the skb just after the highest skb with SACKed
1686 * bit, valid only if sacked_out > 0 or when the caller has ensured
1687 * validity by itself.
1688 */
1689static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1690{
1691 if (!tp->sacked_out)
1692 return tp->snd_una;
1693
1694 if (tp->highest_sack == NULL)
1695 return tp->snd_nxt;
1696
1697 return TCP_SKB_CB(tp->highest_sack)->seq;
1698}
1699
1700static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1701{
1702 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1703}
1704
1705static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1706{
1707 return tcp_sk(sk)->highest_sack;
1708}
1709
1710static inline void tcp_highest_sack_reset(struct sock *sk)
1711{
1712 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1713}
1714
1715/* Called when old skb is about to be deleted and replaced by new skb */
1716static inline void tcp_highest_sack_replace(struct sock *sk,
1717 struct sk_buff *old,
1718 struct sk_buff *new)
1719{
1720 if (old == tcp_highest_sack(sk))
1721 tcp_sk(sk)->highest_sack = new;
1722}
1723
1724/* This helper checks if socket has IP_TRANSPARENT set */
1725static inline bool inet_sk_transparent(const struct sock *sk)
1726{
1727 switch (sk->sk_state) {
1728 case TCP_TIME_WAIT:
1729 return inet_twsk(sk)->tw_transparent;
1730 case TCP_NEW_SYN_RECV:
1731 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1732 }
1733 return inet_sk(sk)->transparent;
1734}
1735
1736/* Determines whether this is a thin stream (which may suffer from
1737 * increased latency). Used to trigger latency-reducing mechanisms.
1738 */
1739static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1740{
1741 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1742}
1743
1744/* /proc */
1745enum tcp_seq_states {
1746 TCP_SEQ_STATE_LISTENING,
1747 TCP_SEQ_STATE_ESTABLISHED,
1748};
1749
1750int tcp_seq_open(struct inode *inode, struct file *file);
1751
1752struct tcp_seq_afinfo {
1753 char *name;
1754 sa_family_t family;
1755 const struct file_operations *seq_fops;
1756 struct seq_operations seq_ops;
1757};
1758
1759struct tcp_iter_state {
1760 struct seq_net_private p;
1761 sa_family_t family;
1762 enum tcp_seq_states state;
1763 struct sock *syn_wait_sk;
1764 int bucket, offset, sbucket, num;
1765 loff_t last_pos;
1766};
1767
1768int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1769void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1770
1771extern struct request_sock_ops tcp_request_sock_ops;
1772extern struct request_sock_ops tcp6_request_sock_ops;
1773
1774void tcp_v4_destroy_sock(struct sock *sk);
1775
1776struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1777 netdev_features_t features);
1778struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1779int tcp_gro_complete(struct sk_buff *skb);
1780
1781void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1782
1783static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1784{
1785 struct net *net = sock_net((struct sock *)tp);
1786 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1787}
1788
1789static inline bool tcp_stream_memory_free(const struct sock *sk)
1790{
1791 const struct tcp_sock *tp = tcp_sk(sk);
1792 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1793
1794 return notsent_bytes < tcp_notsent_lowat(tp);
1795}
1796
1797#ifdef CONFIG_PROC_FS
1798int tcp4_proc_init(void);
1799void tcp4_proc_exit(void);
1800#endif
1801
1802int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1803int tcp_conn_request(struct request_sock_ops *rsk_ops,
1804 const struct tcp_request_sock_ops *af_ops,
1805 struct sock *sk, struct sk_buff *skb);
1806
1807/* TCP af-specific functions */
1808struct tcp_sock_af_ops {
1809#ifdef CONFIG_TCP_MD5SIG
1810 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1811 const struct sock *addr_sk);
1812 int (*calc_md5_hash)(char *location,
1813 const struct tcp_md5sig_key *md5,
1814 const struct sock *sk,
1815 const struct sk_buff *skb);
1816 int (*md5_parse)(struct sock *sk,
1817 int optname,
1818 char __user *optval,
1819 int optlen);
1820#endif
1821};
1822
1823struct tcp_request_sock_ops {
1824 u16 mss_clamp;
1825#ifdef CONFIG_TCP_MD5SIG
1826 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1827 const struct sock *addr_sk);
1828 int (*calc_md5_hash) (char *location,
1829 const struct tcp_md5sig_key *md5,
1830 const struct sock *sk,
1831 const struct sk_buff *skb);
1832#endif
1833 void (*init_req)(struct request_sock *req,
1834 const struct sock *sk_listener,
1835 struct sk_buff *skb);
1836#ifdef CONFIG_SYN_COOKIES
1837 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1838 __u16 *mss);
1839#endif
1840 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1841 const struct request_sock *req);
1842 u32 (*init_seq)(const struct sk_buff *skb);
1843 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1844 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1845 struct flowi *fl, struct request_sock *req,
1846 struct tcp_fastopen_cookie *foc,
1847 enum tcp_synack_type synack_type);
1848};
1849
1850#ifdef CONFIG_SYN_COOKIES
1851static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1852 const struct sock *sk, struct sk_buff *skb,
1853 __u16 *mss)
1854{
1855 tcp_synq_overflow(sk);
1856 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1857 return ops->cookie_init_seq(skb, mss);
1858}
1859#else
1860static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1861 const struct sock *sk, struct sk_buff *skb,
1862 __u16 *mss)
1863{
1864 return 0;
1865}
1866#endif
1867
1868int tcpv4_offload_init(void);
1869
1870void tcp_v4_init(void);
1871void tcp_init(void);
1872
1873/* tcp_recovery.c */
1874extern void tcp_rack_mark_lost(struct sock *sk);
1875extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1876 u64 xmit_time);
1877extern void tcp_rack_reo_timeout(struct sock *sk);
1878extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
1879
1880/* At how many usecs into the future should the RTO fire? */
1881static inline s64 tcp_rto_delta_us(const struct sock *sk)
1882{
1883 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
1884 u32 rto = inet_csk(sk)->icsk_rto;
1885 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto);
1886
1887 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
1888}
1889
1890/*
1891 * Save and compile IPv4 options, return a pointer to it
1892 */
1893static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
1894 struct sk_buff *skb)
1895{
1896 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1897 struct ip_options_rcu *dopt = NULL;
1898
1899 if (opt->optlen) {
1900 int opt_size = sizeof(*dopt) + opt->optlen;
1901
1902 dopt = kmalloc(opt_size, GFP_ATOMIC);
1903 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
1904 kfree(dopt);
1905 dopt = NULL;
1906 }
1907 }
1908 return dopt;
1909}
1910
1911/* locally generated TCP pure ACKs have skb->truesize == 2
1912 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1913 * This is much faster than dissecting the packet to find out.
1914 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1915 */
1916static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1917{
1918 return skb->truesize == 2;
1919}
1920
1921static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1922{
1923 skb->truesize = 2;
1924}
1925
1926static inline int tcp_inq(struct sock *sk)
1927{
1928 struct tcp_sock *tp = tcp_sk(sk);
1929 int answ;
1930
1931 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1932 answ = 0;
1933 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1934 !tp->urg_data ||
1935 before(tp->urg_seq, tp->copied_seq) ||
1936 !before(tp->urg_seq, tp->rcv_nxt)) {
1937
1938 answ = tp->rcv_nxt - tp->copied_seq;
1939
1940 /* Subtract 1, if FIN was received */
1941 if (answ && sock_flag(sk, SOCK_DONE))
1942 answ--;
1943 } else {
1944 answ = tp->urg_seq - tp->copied_seq;
1945 }
1946
1947 return answ;
1948}
1949
1950int tcp_peek_len(struct socket *sock);
1951
1952static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1953{
1954 u16 segs_in;
1955
1956 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1957 tp->segs_in += segs_in;
1958 if (skb->len > tcp_hdrlen(skb))
1959 tp->data_segs_in += segs_in;
1960}
1961
1962/*
1963 * TCP listen path runs lockless.
1964 * We forced "struct sock" to be const qualified to make sure
1965 * we don't modify one of its field by mistake.
1966 * Here, we increment sk_drops which is an atomic_t, so we can safely
1967 * make sock writable again.
1968 */
1969static inline void tcp_listendrop(const struct sock *sk)
1970{
1971 atomic_inc(&((struct sock *)sk)->sk_drops);
1972 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
1973}
1974
1975enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
1976
1977/*
1978 * Interface for adding Upper Level Protocols over TCP
1979 */
1980
1981#define TCP_ULP_NAME_MAX 16
1982#define TCP_ULP_MAX 128
1983#define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
1984
1985enum {
1986 TCP_ULP_TLS,
1987 TCP_ULP_BPF,
1988};
1989
1990struct tcp_ulp_ops {
1991 struct list_head list;
1992
1993 /* initialize ulp */
1994 int (*init)(struct sock *sk);
1995 /* cleanup ulp */
1996 void (*release)(struct sock *sk);
1997
1998 int uid;
1999 char name[TCP_ULP_NAME_MAX];
2000 bool user_visible;
2001 struct module *owner;
2002};
2003int tcp_register_ulp(struct tcp_ulp_ops *type);
2004void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2005int tcp_set_ulp(struct sock *sk, const char *name);
2006int tcp_set_ulp_id(struct sock *sk, const int ulp);
2007void tcp_get_available_ulp(char *buf, size_t len);
2008void tcp_cleanup_ulp(struct sock *sk);
2009
2010/* Call BPF_SOCK_OPS program that returns an int. If the return value
2011 * is < 0, then the BPF op failed (for example if the loaded BPF
2012 * program does not support the chosen operation or there is no BPF
2013 * program loaded).
2014 */
2015#ifdef CONFIG_BPF
2016static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2017{
2018 struct bpf_sock_ops_kern sock_ops;
2019 int ret;
2020
2021 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2022 if (sk_fullsock(sk)) {
2023 sock_ops.is_fullsock = 1;
2024 sock_owned_by_me(sk);
2025 }
2026
2027 sock_ops.sk = sk;
2028 sock_ops.op = op;
2029 if (nargs > 0)
2030 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2031
2032 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2033 if (ret == 0)
2034 ret = sock_ops.reply;
2035 else
2036 ret = -1;
2037 return ret;
2038}
2039
2040static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2041{
2042 u32 args[2] = {arg1, arg2};
2043
2044 return tcp_call_bpf(sk, op, 2, args);
2045}
2046
2047static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2048 u32 arg3)
2049{
2050 u32 args[3] = {arg1, arg2, arg3};
2051
2052 return tcp_call_bpf(sk, op, 3, args);
2053}
2054
2055#else
2056static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2057{
2058 return -EPERM;
2059}
2060
2061static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2062{
2063 return -EPERM;
2064}
2065
2066static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2067 u32 arg3)
2068{
2069 return -EPERM;
2070}
2071
2072#endif
2073
2074static inline u32 tcp_timeout_init(struct sock *sk)
2075{
2076 int timeout;
2077
2078 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2079
2080 if (timeout <= 0)
2081 timeout = TCP_TIMEOUT_INIT;
2082 return timeout;
2083}
2084
2085static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2086{
2087 int rwnd;
2088
2089 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2090
2091 if (rwnd < 0)
2092 rwnd = 0;
2093 return rwnd;
2094}
2095
2096static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2097{
2098 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2099}
2100
2101#if IS_ENABLED(CONFIG_SMC)
2102extern struct static_key_false tcp_have_smc;
2103#endif
2104#endif /* _TCP_H */