]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - include/net/tcp.h
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: TCP thin linear timeouts
[mirror_ubuntu-jammy-kernel.git] / include / net / 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 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
49 extern struct inet_hashinfo tcp_hashinfo;
50
51 extern struct percpu_counter tcp_orphan_count;
52 extern 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 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
64 #define TCP_MIN_MSS 88U
65
66 /* The least MTU to use for probing */
67 #define TCP_BASE_MSS 512
68
69 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
70 #define TCP_FASTRETRANS_THRESH 3
71
72 /* Maximal reordering. */
73 #define TCP_MAX_REORDERING 127
74
75 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
76 #define TCP_MAX_QUICKACKS 16U
77
78 /* urg_data states */
79 #define TCP_URG_VALID 0x0100
80 #define TCP_URG_NOTYET 0x0200
81 #define TCP_URG_READ 0x0400
82
83 #define TCP_RETR1 3 /*
84 * This is how many retries it does before it
85 * tries to figure out if the gateway is
86 * down. Minimal RFC value is 3; it corresponds
87 * to ~3sec-8min depending on RTO.
88 */
89
90 #define TCP_RETR2 15 /*
91 * This should take at least
92 * 90 minutes to time out.
93 * RFC1122 says that the limit is 100 sec.
94 * 15 is ~13-30min depending on RTO.
95 */
96
97 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
98 * connection: ~180sec is RFC minimum */
99
100 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
101 * connection: ~180sec is RFC minimum */
102
103
104 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
105 * socket. 7 is ~50sec-16min.
106 */
107
108
109 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
110 * state, about 60 seconds */
111 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
112 /* BSD style FIN_WAIT2 deadlock breaker.
113 * It used to be 3min, new value is 60sec,
114 * to combine FIN-WAIT-2 timeout with
115 * TIME-WAIT timer.
116 */
117
118 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
119 #if HZ >= 100
120 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
121 #define TCP_ATO_MIN ((unsigned)(HZ/25))
122 #else
123 #define TCP_DELACK_MIN 4U
124 #define TCP_ATO_MIN 4U
125 #endif
126 #define TCP_RTO_MAX ((unsigned)(120*HZ))
127 #define TCP_RTO_MIN ((unsigned)(HZ/5))
128 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
129
130 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
131 * for local resources.
132 */
133
134 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
135 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
136 #define TCP_KEEPALIVE_INTVL (75*HZ)
137
138 #define MAX_TCP_KEEPIDLE 32767
139 #define MAX_TCP_KEEPINTVL 32767
140 #define MAX_TCP_KEEPCNT 127
141 #define MAX_TCP_SYNCNT 127
142
143 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
144
145 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
146 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
147 * after this time. It should be equal
148 * (or greater than) TCP_TIMEWAIT_LEN
149 * to provide reliability equal to one
150 * provided by timewait state.
151 */
152 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
153 * timestamps. It must be less than
154 * minimal timewait lifetime.
155 */
156 /*
157 * TCP option
158 */
159
160 #define TCPOPT_NOP 1 /* Padding */
161 #define TCPOPT_EOL 0 /* End of options */
162 #define TCPOPT_MSS 2 /* Segment size negotiating */
163 #define TCPOPT_WINDOW 3 /* Window scaling */
164 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
165 #define TCPOPT_SACK 5 /* SACK Block */
166 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
167 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
168 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
169
170 /*
171 * TCP option lengths
172 */
173
174 #define TCPOLEN_MSS 4
175 #define TCPOLEN_WINDOW 3
176 #define TCPOLEN_SACK_PERM 2
177 #define TCPOLEN_TIMESTAMP 10
178 #define TCPOLEN_MD5SIG 18
179 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
180 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
181 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
182 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
183
184 /* But this is what stacks really send out. */
185 #define TCPOLEN_TSTAMP_ALIGNED 12
186 #define TCPOLEN_WSCALE_ALIGNED 4
187 #define TCPOLEN_SACKPERM_ALIGNED 4
188 #define TCPOLEN_SACK_BASE 2
189 #define TCPOLEN_SACK_BASE_ALIGNED 4
190 #define TCPOLEN_SACK_PERBLOCK 8
191 #define TCPOLEN_MD5SIG_ALIGNED 20
192 #define TCPOLEN_MSS_ALIGNED 4
193
194 /* Flags in tp->nonagle */
195 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
196 #define TCP_NAGLE_CORK 2 /* Socket is corked */
197 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
198
199 /* TCP thin-stream limits */
200 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
201
202 extern struct inet_timewait_death_row tcp_death_row;
203
204 /* sysctl variables for tcp */
205 extern int sysctl_tcp_timestamps;
206 extern int sysctl_tcp_window_scaling;
207 extern int sysctl_tcp_sack;
208 extern int sysctl_tcp_fin_timeout;
209 extern int sysctl_tcp_keepalive_time;
210 extern int sysctl_tcp_keepalive_probes;
211 extern int sysctl_tcp_keepalive_intvl;
212 extern int sysctl_tcp_syn_retries;
213 extern int sysctl_tcp_synack_retries;
214 extern int sysctl_tcp_retries1;
215 extern int sysctl_tcp_retries2;
216 extern int sysctl_tcp_orphan_retries;
217 extern int sysctl_tcp_syncookies;
218 extern int sysctl_tcp_retrans_collapse;
219 extern int sysctl_tcp_stdurg;
220 extern int sysctl_tcp_rfc1337;
221 extern int sysctl_tcp_abort_on_overflow;
222 extern int sysctl_tcp_max_orphans;
223 extern int sysctl_tcp_fack;
224 extern int sysctl_tcp_reordering;
225 extern int sysctl_tcp_ecn;
226 extern int sysctl_tcp_dsack;
227 extern int sysctl_tcp_mem[3];
228 extern int sysctl_tcp_wmem[3];
229 extern int sysctl_tcp_rmem[3];
230 extern int sysctl_tcp_app_win;
231 extern int sysctl_tcp_adv_win_scale;
232 extern int sysctl_tcp_tw_reuse;
233 extern int sysctl_tcp_frto;
234 extern int sysctl_tcp_frto_response;
235 extern int sysctl_tcp_low_latency;
236 extern int sysctl_tcp_dma_copybreak;
237 extern int sysctl_tcp_nometrics_save;
238 extern int sysctl_tcp_moderate_rcvbuf;
239 extern int sysctl_tcp_tso_win_divisor;
240 extern int sysctl_tcp_abc;
241 extern int sysctl_tcp_mtu_probing;
242 extern int sysctl_tcp_base_mss;
243 extern int sysctl_tcp_workaround_signed_windows;
244 extern int sysctl_tcp_slow_start_after_idle;
245 extern int sysctl_tcp_max_ssthresh;
246 extern int sysctl_tcp_cookie_size;
247 extern int sysctl_tcp_thin_linear_timeouts;
248
249 extern atomic_t tcp_memory_allocated;
250 extern struct percpu_counter tcp_sockets_allocated;
251 extern int tcp_memory_pressure;
252
253 /*
254 * The next routines deal with comparing 32 bit unsigned ints
255 * and worry about wraparound (automatic with unsigned arithmetic).
256 */
257
258 static inline int before(__u32 seq1, __u32 seq2)
259 {
260 return (__s32)(seq1-seq2) < 0;
261 }
262 #define after(seq2, seq1) before(seq1, seq2)
263
264 /* is s2<=s1<=s3 ? */
265 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
266 {
267 return seq3 - seq2 >= seq1 - seq2;
268 }
269
270 static inline int tcp_too_many_orphans(struct sock *sk, int num)
271 {
272 return (num > sysctl_tcp_max_orphans) ||
273 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
274 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
275 }
276
277 /* syncookies: remember time of last synqueue overflow */
278 static inline void tcp_synq_overflow(struct sock *sk)
279 {
280 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
281 }
282
283 /* syncookies: no recent synqueue overflow on this listening socket? */
284 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
285 {
286 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
287 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
288 }
289
290 extern struct proto tcp_prot;
291
292 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
293 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
294 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
295 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
296
297 extern void tcp_v4_err(struct sk_buff *skb, u32);
298
299 extern void tcp_shutdown (struct sock *sk, int how);
300
301 extern int tcp_v4_rcv(struct sk_buff *skb);
302
303 extern int tcp_v4_remember_stamp(struct sock *sk);
304
305 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
306
307 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
308 struct msghdr *msg, size_t size);
309 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
310
311 extern int tcp_ioctl(struct sock *sk,
312 int cmd,
313 unsigned long arg);
314
315 extern int tcp_rcv_state_process(struct sock *sk,
316 struct sk_buff *skb,
317 struct tcphdr *th,
318 unsigned len);
319
320 extern int tcp_rcv_established(struct sock *sk,
321 struct sk_buff *skb,
322 struct tcphdr *th,
323 unsigned len);
324
325 extern void tcp_rcv_space_adjust(struct sock *sk);
326
327 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
328
329 extern int tcp_twsk_unique(struct sock *sk,
330 struct sock *sktw, void *twp);
331
332 extern void tcp_twsk_destructor(struct sock *sk);
333
334 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
335 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
336
337 static inline void tcp_dec_quickack_mode(struct sock *sk,
338 const unsigned int pkts)
339 {
340 struct inet_connection_sock *icsk = inet_csk(sk);
341
342 if (icsk->icsk_ack.quick) {
343 if (pkts >= icsk->icsk_ack.quick) {
344 icsk->icsk_ack.quick = 0;
345 /* Leaving quickack mode we deflate ATO. */
346 icsk->icsk_ack.ato = TCP_ATO_MIN;
347 } else
348 icsk->icsk_ack.quick -= pkts;
349 }
350 }
351
352 extern void tcp_enter_quickack_mode(struct sock *sk);
353
354 #define TCP_ECN_OK 1
355 #define TCP_ECN_QUEUE_CWR 2
356 #define TCP_ECN_DEMAND_CWR 4
357
358 static __inline__ void
359 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
360 {
361 if (sysctl_tcp_ecn && th->ece && th->cwr)
362 inet_rsk(req)->ecn_ok = 1;
363 }
364
365 enum 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
373 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
374 struct sk_buff *skb,
375 const struct tcphdr *th);
376
377 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
378 struct request_sock *req,
379 struct request_sock **prev);
380 extern int tcp_child_process(struct sock *parent,
381 struct sock *child,
382 struct sk_buff *skb);
383 extern int tcp_use_frto(struct sock *sk);
384 extern void tcp_enter_frto(struct sock *sk);
385 extern void tcp_enter_loss(struct sock *sk, int how);
386 extern void tcp_clear_retrans(struct tcp_sock *tp);
387 extern void tcp_update_metrics(struct sock *sk);
388
389 extern void tcp_close(struct sock *sk,
390 long timeout);
391 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
392
393 extern int tcp_getsockopt(struct sock *sk, int level,
394 int optname,
395 char __user *optval,
396 int __user *optlen);
397 extern int tcp_setsockopt(struct sock *sk, int level,
398 int optname, char __user *optval,
399 unsigned int optlen);
400 extern int compat_tcp_getsockopt(struct sock *sk,
401 int level, int optname,
402 char __user *optval, int __user *optlen);
403 extern int compat_tcp_setsockopt(struct sock *sk,
404 int level, int optname,
405 char __user *optval, unsigned int optlen);
406 extern void tcp_set_keepalive(struct sock *sk, int val);
407 extern void tcp_syn_ack_timeout(struct sock *sk,
408 struct request_sock *req);
409 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
410 struct msghdr *msg,
411 size_t len, int nonblock,
412 int flags, int *addr_len);
413
414 extern void tcp_parse_options(struct sk_buff *skb,
415 struct tcp_options_received *opt_rx,
416 u8 **hvpp,
417 int estab);
418
419 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
420
421 /*
422 * TCP v4 functions exported for the inet6 API
423 */
424
425 extern void tcp_v4_send_check(struct sock *sk, int len,
426 struct sk_buff *skb);
427
428 extern int tcp_v4_conn_request(struct sock *sk,
429 struct sk_buff *skb);
430
431 extern struct sock * tcp_create_openreq_child(struct sock *sk,
432 struct request_sock *req,
433 struct sk_buff *skb);
434
435 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
436 struct sk_buff *skb,
437 struct request_sock *req,
438 struct dst_entry *dst);
439
440 extern int tcp_v4_do_rcv(struct sock *sk,
441 struct sk_buff *skb);
442
443 extern int tcp_v4_connect(struct sock *sk,
444 struct sockaddr *uaddr,
445 int addr_len);
446
447 extern int tcp_connect(struct sock *sk);
448
449 extern struct sk_buff * tcp_make_synack(struct sock *sk,
450 struct dst_entry *dst,
451 struct request_sock *req,
452 struct request_values *rvp);
453
454 extern int tcp_disconnect(struct sock *sk, int flags);
455
456
457 /* From syncookies.c */
458 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
459 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
460 struct ip_options *opt);
461 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
462 __u16 *mss);
463
464 extern __u32 cookie_init_timestamp(struct request_sock *req);
465 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt);
466
467 /* From net/ipv6/syncookies.c */
468 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
469 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
470 __u16 *mss);
471
472 /* tcp_output.c */
473
474 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
475 int nonagle);
476 extern int tcp_may_send_now(struct sock *sk);
477 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
478 extern void tcp_retransmit_timer(struct sock *sk);
479 extern void tcp_xmit_retransmit_queue(struct sock *);
480 extern void tcp_simple_retransmit(struct sock *);
481 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
482 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
483
484 extern void tcp_send_probe0(struct sock *);
485 extern void tcp_send_partial(struct sock *);
486 extern int tcp_write_wakeup(struct sock *);
487 extern void tcp_send_fin(struct sock *sk);
488 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
489 extern int tcp_send_synack(struct sock *);
490 extern void tcp_push_one(struct sock *, unsigned int mss_now);
491 extern void tcp_send_ack(struct sock *sk);
492 extern void tcp_send_delayed_ack(struct sock *sk);
493
494 /* tcp_input.c */
495 extern void tcp_cwnd_application_limited(struct sock *sk);
496
497 /* tcp_timer.c */
498 extern void tcp_init_xmit_timers(struct sock *);
499 static inline void tcp_clear_xmit_timers(struct sock *sk)
500 {
501 inet_csk_clear_xmit_timers(sk);
502 }
503
504 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
505 extern unsigned int tcp_current_mss(struct sock *sk);
506
507 /* Bound MSS / TSO packet size with the half of the window */
508 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
509 {
510 if (tp->max_window && pktsize > (tp->max_window >> 1))
511 return max(tp->max_window >> 1, 68U - tp->tcp_header_len);
512 else
513 return pktsize;
514 }
515
516 /* tcp.c */
517 extern void tcp_get_info(struct sock *, struct tcp_info *);
518
519 /* Read 'sendfile()'-style from a TCP socket */
520 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
521 unsigned int, size_t);
522 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
523 sk_read_actor_t recv_actor);
524
525 extern void tcp_initialize_rcv_mss(struct sock *sk);
526
527 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
528 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
529 extern void tcp_mtup_init(struct sock *sk);
530
531 static inline void tcp_bound_rto(const struct sock *sk)
532 {
533 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
534 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
535 }
536
537 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
538 {
539 return (tp->srtt >> 3) + tp->rttvar;
540 }
541
542 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
543 {
544 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
545 ntohl(TCP_FLAG_ACK) |
546 snd_wnd);
547 }
548
549 static inline void tcp_fast_path_on(struct tcp_sock *tp)
550 {
551 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
552 }
553
554 static inline void tcp_fast_path_check(struct sock *sk)
555 {
556 struct tcp_sock *tp = tcp_sk(sk);
557
558 if (skb_queue_empty(&tp->out_of_order_queue) &&
559 tp->rcv_wnd &&
560 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
561 !tp->urg_data)
562 tcp_fast_path_on(tp);
563 }
564
565 /* Compute the actual rto_min value */
566 static inline u32 tcp_rto_min(struct sock *sk)
567 {
568 struct dst_entry *dst = __sk_dst_get(sk);
569 u32 rto_min = TCP_RTO_MIN;
570
571 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
572 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
573 return rto_min;
574 }
575
576 /* Compute the actual receive window we are currently advertising.
577 * Rcv_nxt can be after the window if our peer push more data
578 * than the offered window.
579 */
580 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
581 {
582 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
583
584 if (win < 0)
585 win = 0;
586 return (u32) win;
587 }
588
589 /* Choose a new window, without checks for shrinking, and without
590 * scaling applied to the result. The caller does these things
591 * if necessary. This is a "raw" window selection.
592 */
593 extern u32 __tcp_select_window(struct sock *sk);
594
595 /* TCP timestamps are only 32-bits, this causes a slight
596 * complication on 64-bit systems since we store a snapshot
597 * of jiffies in the buffer control blocks below. We decided
598 * to use only the low 32-bits of jiffies and hide the ugly
599 * casts with the following macro.
600 */
601 #define tcp_time_stamp ((__u32)(jiffies))
602
603 /* This is what the send packet queuing engine uses to pass
604 * TCP per-packet control information to the transmission
605 * code. We also store the host-order sequence numbers in
606 * here too. This is 36 bytes on 32-bit architectures,
607 * 40 bytes on 64-bit machines, if this grows please adjust
608 * skbuff.h:skbuff->cb[xxx] size appropriately.
609 */
610 struct tcp_skb_cb {
611 union {
612 struct inet_skb_parm h4;
613 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
614 struct inet6_skb_parm h6;
615 #endif
616 } header; /* For incoming frames */
617 __u32 seq; /* Starting sequence number */
618 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
619 __u32 when; /* used to compute rtt's */
620 __u8 flags; /* TCP header flags. */
621
622 /* NOTE: These must match up to the flags byte in a
623 * real TCP header.
624 */
625 #define TCPCB_FLAG_FIN 0x01
626 #define TCPCB_FLAG_SYN 0x02
627 #define TCPCB_FLAG_RST 0x04
628 #define TCPCB_FLAG_PSH 0x08
629 #define TCPCB_FLAG_ACK 0x10
630 #define TCPCB_FLAG_URG 0x20
631 #define TCPCB_FLAG_ECE 0x40
632 #define TCPCB_FLAG_CWR 0x80
633
634 __u8 sacked; /* State flags for SACK/FACK. */
635 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
636 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
637 #define TCPCB_LOST 0x04 /* SKB is lost */
638 #define TCPCB_TAGBITS 0x07 /* All tag bits */
639
640 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
641 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
642
643 __u32 ack_seq; /* Sequence number ACK'd */
644 };
645
646 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
647
648 /* Due to TSO, an SKB can be composed of multiple actual
649 * packets. To keep these tracked properly, we use this.
650 */
651 static inline int tcp_skb_pcount(const struct sk_buff *skb)
652 {
653 return skb_shinfo(skb)->gso_segs;
654 }
655
656 /* This is valid iff tcp_skb_pcount() > 1. */
657 static inline int tcp_skb_mss(const struct sk_buff *skb)
658 {
659 return skb_shinfo(skb)->gso_size;
660 }
661
662 /* Events passed to congestion control interface */
663 enum tcp_ca_event {
664 CA_EVENT_TX_START, /* first transmit when no packets in flight */
665 CA_EVENT_CWND_RESTART, /* congestion window restart */
666 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
667 CA_EVENT_FRTO, /* fast recovery timeout */
668 CA_EVENT_LOSS, /* loss timeout */
669 CA_EVENT_FAST_ACK, /* in sequence ack */
670 CA_EVENT_SLOW_ACK, /* other ack */
671 };
672
673 /*
674 * Interface for adding new TCP congestion control handlers
675 */
676 #define TCP_CA_NAME_MAX 16
677 #define TCP_CA_MAX 128
678 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
679
680 #define TCP_CONG_NON_RESTRICTED 0x1
681 #define TCP_CONG_RTT_STAMP 0x2
682
683 struct tcp_congestion_ops {
684 struct list_head list;
685 unsigned long flags;
686
687 /* initialize private data (optional) */
688 void (*init)(struct sock *sk);
689 /* cleanup private data (optional) */
690 void (*release)(struct sock *sk);
691
692 /* return slow start threshold (required) */
693 u32 (*ssthresh)(struct sock *sk);
694 /* lower bound for congestion window (optional) */
695 u32 (*min_cwnd)(const struct sock *sk);
696 /* do new cwnd calculation (required) */
697 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
698 /* call before changing ca_state (optional) */
699 void (*set_state)(struct sock *sk, u8 new_state);
700 /* call when cwnd event occurs (optional) */
701 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
702 /* new value of cwnd after loss (optional) */
703 u32 (*undo_cwnd)(struct sock *sk);
704 /* hook for packet ack accounting (optional) */
705 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
706 /* get info for inet_diag (optional) */
707 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
708
709 char name[TCP_CA_NAME_MAX];
710 struct module *owner;
711 };
712
713 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
714 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
715
716 extern void tcp_init_congestion_control(struct sock *sk);
717 extern void tcp_cleanup_congestion_control(struct sock *sk);
718 extern int tcp_set_default_congestion_control(const char *name);
719 extern void tcp_get_default_congestion_control(char *name);
720 extern void tcp_get_available_congestion_control(char *buf, size_t len);
721 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
722 extern int tcp_set_allowed_congestion_control(char *allowed);
723 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
724 extern void tcp_slow_start(struct tcp_sock *tp);
725 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
726
727 extern struct tcp_congestion_ops tcp_init_congestion_ops;
728 extern u32 tcp_reno_ssthresh(struct sock *sk);
729 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
730 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
731 extern struct tcp_congestion_ops tcp_reno;
732
733 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
734 {
735 struct inet_connection_sock *icsk = inet_csk(sk);
736
737 if (icsk->icsk_ca_ops->set_state)
738 icsk->icsk_ca_ops->set_state(sk, ca_state);
739 icsk->icsk_ca_state = ca_state;
740 }
741
742 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
743 {
744 const struct inet_connection_sock *icsk = inet_csk(sk);
745
746 if (icsk->icsk_ca_ops->cwnd_event)
747 icsk->icsk_ca_ops->cwnd_event(sk, event);
748 }
749
750 /* These functions determine how the current flow behaves in respect of SACK
751 * handling. SACK is negotiated with the peer, and therefore it can vary
752 * between different flows.
753 *
754 * tcp_is_sack - SACK enabled
755 * tcp_is_reno - No SACK
756 * tcp_is_fack - FACK enabled, implies SACK enabled
757 */
758 static inline int tcp_is_sack(const struct tcp_sock *tp)
759 {
760 return tp->rx_opt.sack_ok;
761 }
762
763 static inline int tcp_is_reno(const struct tcp_sock *tp)
764 {
765 return !tcp_is_sack(tp);
766 }
767
768 static inline int tcp_is_fack(const struct tcp_sock *tp)
769 {
770 return tp->rx_opt.sack_ok & 2;
771 }
772
773 static inline void tcp_enable_fack(struct tcp_sock *tp)
774 {
775 tp->rx_opt.sack_ok |= 2;
776 }
777
778 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
779 {
780 return tp->sacked_out + tp->lost_out;
781 }
782
783 /* This determines how many packets are "in the network" to the best
784 * of our knowledge. In many cases it is conservative, but where
785 * detailed information is available from the receiver (via SACK
786 * blocks etc.) we can make more aggressive calculations.
787 *
788 * Use this for decisions involving congestion control, use just
789 * tp->packets_out to determine if the send queue is empty or not.
790 *
791 * Read this equation as:
792 *
793 * "Packets sent once on transmission queue" MINUS
794 * "Packets left network, but not honestly ACKed yet" PLUS
795 * "Packets fast retransmitted"
796 */
797 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
798 {
799 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
800 }
801
802 #define TCP_INFINITE_SSTHRESH 0x7fffffff
803
804 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
805 {
806 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
807 }
808
809 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
810 * The exception is rate halving phase, when cwnd is decreasing towards
811 * ssthresh.
812 */
813 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
814 {
815 const struct tcp_sock *tp = tcp_sk(sk);
816 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
817 return tp->snd_ssthresh;
818 else
819 return max(tp->snd_ssthresh,
820 ((tp->snd_cwnd >> 1) +
821 (tp->snd_cwnd >> 2)));
822 }
823
824 /* Use define here intentionally to get WARN_ON location shown at the caller */
825 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
826
827 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
828 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
829
830 /* Slow start with delack produces 3 packets of burst, so that
831 * it is safe "de facto". This will be the default - same as
832 * the default reordering threshold - but if reordering increases,
833 * we must be able to allow cwnd to burst at least this much in order
834 * to not pull it back when holes are filled.
835 */
836 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
837 {
838 return tp->reordering;
839 }
840
841 /* Returns end sequence number of the receiver's advertised window */
842 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
843 {
844 return tp->snd_una + tp->snd_wnd;
845 }
846 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
847
848 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
849 const struct sk_buff *skb)
850 {
851 if (skb->len < mss)
852 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
853 }
854
855 static inline void tcp_check_probe_timer(struct sock *sk)
856 {
857 struct tcp_sock *tp = tcp_sk(sk);
858 const struct inet_connection_sock *icsk = inet_csk(sk);
859
860 if (!tp->packets_out && !icsk->icsk_pending)
861 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
862 icsk->icsk_rto, TCP_RTO_MAX);
863 }
864
865 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
866 {
867 tp->snd_wl1 = seq;
868 }
869
870 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
871 {
872 tp->snd_wl1 = seq;
873 }
874
875 /*
876 * Calculate(/check) TCP checksum
877 */
878 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
879 __be32 daddr, __wsum base)
880 {
881 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
882 }
883
884 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
885 {
886 return __skb_checksum_complete(skb);
887 }
888
889 static inline int tcp_checksum_complete(struct sk_buff *skb)
890 {
891 return !skb_csum_unnecessary(skb) &&
892 __tcp_checksum_complete(skb);
893 }
894
895 /* Prequeue for VJ style copy to user, combined with checksumming. */
896
897 static inline void tcp_prequeue_init(struct tcp_sock *tp)
898 {
899 tp->ucopy.task = NULL;
900 tp->ucopy.len = 0;
901 tp->ucopy.memory = 0;
902 skb_queue_head_init(&tp->ucopy.prequeue);
903 #ifdef CONFIG_NET_DMA
904 tp->ucopy.dma_chan = NULL;
905 tp->ucopy.wakeup = 0;
906 tp->ucopy.pinned_list = NULL;
907 tp->ucopy.dma_cookie = 0;
908 #endif
909 }
910
911 /* Packet is added to VJ-style prequeue for processing in process
912 * context, if a reader task is waiting. Apparently, this exciting
913 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
914 * failed somewhere. Latency? Burstiness? Well, at least now we will
915 * see, why it failed. 8)8) --ANK
916 *
917 * NOTE: is this not too big to inline?
918 */
919 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
920 {
921 struct tcp_sock *tp = tcp_sk(sk);
922
923 if (sysctl_tcp_low_latency || !tp->ucopy.task)
924 return 0;
925
926 __skb_queue_tail(&tp->ucopy.prequeue, skb);
927 tp->ucopy.memory += skb->truesize;
928 if (tp->ucopy.memory > sk->sk_rcvbuf) {
929 struct sk_buff *skb1;
930
931 BUG_ON(sock_owned_by_user(sk));
932
933 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
934 sk_backlog_rcv(sk, skb1);
935 NET_INC_STATS_BH(sock_net(sk),
936 LINUX_MIB_TCPPREQUEUEDROPPED);
937 }
938
939 tp->ucopy.memory = 0;
940 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
941 wake_up_interruptible_poll(sk->sk_sleep,
942 POLLIN | POLLRDNORM | POLLRDBAND);
943 if (!inet_csk_ack_scheduled(sk))
944 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
945 (3 * tcp_rto_min(sk)) / 4,
946 TCP_RTO_MAX);
947 }
948 return 1;
949 }
950
951
952 #undef STATE_TRACE
953
954 #ifdef STATE_TRACE
955 static const char *statename[]={
956 "Unused","Established","Syn Sent","Syn Recv",
957 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
958 "Close Wait","Last ACK","Listen","Closing"
959 };
960 #endif
961 extern void tcp_set_state(struct sock *sk, int state);
962
963 extern void tcp_done(struct sock *sk);
964
965 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
966 {
967 rx_opt->dsack = 0;
968 rx_opt->num_sacks = 0;
969 }
970
971 /* Determine a window scaling and initial window to offer. */
972 extern void tcp_select_initial_window(int __space, __u32 mss,
973 __u32 *rcv_wnd, __u32 *window_clamp,
974 int wscale_ok, __u8 *rcv_wscale,
975 __u32 init_rcv_wnd);
976
977 static inline int tcp_win_from_space(int space)
978 {
979 return sysctl_tcp_adv_win_scale<=0 ?
980 (space>>(-sysctl_tcp_adv_win_scale)) :
981 space - (space>>sysctl_tcp_adv_win_scale);
982 }
983
984 /* Note: caller must be prepared to deal with negative returns */
985 static inline int tcp_space(const struct sock *sk)
986 {
987 return tcp_win_from_space(sk->sk_rcvbuf -
988 atomic_read(&sk->sk_rmem_alloc));
989 }
990
991 static inline int tcp_full_space(const struct sock *sk)
992 {
993 return tcp_win_from_space(sk->sk_rcvbuf);
994 }
995
996 static inline void tcp_openreq_init(struct request_sock *req,
997 struct tcp_options_received *rx_opt,
998 struct sk_buff *skb)
999 {
1000 struct inet_request_sock *ireq = inet_rsk(req);
1001
1002 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1003 req->cookie_ts = 0;
1004 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1005 req->mss = rx_opt->mss_clamp;
1006 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1007 ireq->tstamp_ok = rx_opt->tstamp_ok;
1008 ireq->sack_ok = rx_opt->sack_ok;
1009 ireq->snd_wscale = rx_opt->snd_wscale;
1010 ireq->wscale_ok = rx_opt->wscale_ok;
1011 ireq->acked = 0;
1012 ireq->ecn_ok = 0;
1013 ireq->rmt_port = tcp_hdr(skb)->source;
1014 ireq->loc_port = tcp_hdr(skb)->dest;
1015 }
1016
1017 extern void tcp_enter_memory_pressure(struct sock *sk);
1018
1019 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1020 {
1021 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1022 }
1023
1024 static inline int keepalive_time_when(const struct tcp_sock *tp)
1025 {
1026 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1027 }
1028
1029 static inline int keepalive_probes(const struct tcp_sock *tp)
1030 {
1031 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1032 }
1033
1034 static inline int tcp_fin_time(const struct sock *sk)
1035 {
1036 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1037 const int rto = inet_csk(sk)->icsk_rto;
1038
1039 if (fin_timeout < (rto << 2) - (rto >> 1))
1040 fin_timeout = (rto << 2) - (rto >> 1);
1041
1042 return fin_timeout;
1043 }
1044
1045 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1046 int paws_win)
1047 {
1048 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1049 return 1;
1050 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1051 return 1;
1052
1053 return 0;
1054 }
1055
1056 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1057 int rst)
1058 {
1059 if (tcp_paws_check(rx_opt, 0))
1060 return 0;
1061
1062 /* RST segments are not recommended to carry timestamp,
1063 and, if they do, it is recommended to ignore PAWS because
1064 "their cleanup function should take precedence over timestamps."
1065 Certainly, it is mistake. It is necessary to understand the reasons
1066 of this constraint to relax it: if peer reboots, clock may go
1067 out-of-sync and half-open connections will not be reset.
1068 Actually, the problem would be not existing if all
1069 the implementations followed draft about maintaining clock
1070 via reboots. Linux-2.2 DOES NOT!
1071
1072 However, we can relax time bounds for RST segments to MSL.
1073 */
1074 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1075 return 0;
1076 return 1;
1077 }
1078
1079 #define TCP_CHECK_TIMER(sk) do { } while (0)
1080
1081 static inline void tcp_mib_init(struct net *net)
1082 {
1083 /* See RFC 2012 */
1084 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1085 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1086 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1087 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1088 }
1089
1090 /* from STCP */
1091 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1092 {
1093 tp->lost_skb_hint = NULL;
1094 tp->scoreboard_skb_hint = NULL;
1095 }
1096
1097 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1098 {
1099 tcp_clear_retrans_hints_partial(tp);
1100 tp->retransmit_skb_hint = NULL;
1101 }
1102
1103 /* MD5 Signature */
1104 struct crypto_hash;
1105
1106 /* - key database */
1107 struct tcp_md5sig_key {
1108 u8 *key;
1109 u8 keylen;
1110 };
1111
1112 struct tcp4_md5sig_key {
1113 struct tcp_md5sig_key base;
1114 __be32 addr;
1115 };
1116
1117 struct tcp6_md5sig_key {
1118 struct tcp_md5sig_key base;
1119 #if 0
1120 u32 scope_id; /* XXX */
1121 #endif
1122 struct in6_addr addr;
1123 };
1124
1125 /* - sock block */
1126 struct tcp_md5sig_info {
1127 struct tcp4_md5sig_key *keys4;
1128 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1129 struct tcp6_md5sig_key *keys6;
1130 u32 entries6;
1131 u32 alloced6;
1132 #endif
1133 u32 entries4;
1134 u32 alloced4;
1135 };
1136
1137 /* - pseudo header */
1138 struct tcp4_pseudohdr {
1139 __be32 saddr;
1140 __be32 daddr;
1141 __u8 pad;
1142 __u8 protocol;
1143 __be16 len;
1144 };
1145
1146 struct tcp6_pseudohdr {
1147 struct in6_addr saddr;
1148 struct in6_addr daddr;
1149 __be32 len;
1150 __be32 protocol; /* including padding */
1151 };
1152
1153 union tcp_md5sum_block {
1154 struct tcp4_pseudohdr ip4;
1155 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1156 struct tcp6_pseudohdr ip6;
1157 #endif
1158 };
1159
1160 /* - pool: digest algorithm, hash description and scratch buffer */
1161 struct tcp_md5sig_pool {
1162 struct hash_desc md5_desc;
1163 union tcp_md5sum_block md5_blk;
1164 };
1165
1166 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1167
1168 /* - functions */
1169 extern int tcp_v4_md5_hash_skb(char *md5_hash,
1170 struct tcp_md5sig_key *key,
1171 struct sock *sk,
1172 struct request_sock *req,
1173 struct sk_buff *skb);
1174
1175 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1176 struct sock *addr_sk);
1177
1178 extern int tcp_v4_md5_do_add(struct sock *sk,
1179 __be32 addr,
1180 u8 *newkey,
1181 u8 newkeylen);
1182
1183 extern int tcp_v4_md5_do_del(struct sock *sk,
1184 __be32 addr);
1185
1186 #ifdef CONFIG_TCP_MD5SIG
1187 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1188 &(struct tcp_md5sig_key) { \
1189 .key = (twsk)->tw_md5_key, \
1190 .keylen = (twsk)->tw_md5_keylen, \
1191 } : NULL)
1192 #else
1193 #define tcp_twsk_md5_key(twsk) NULL
1194 #endif
1195
1196 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *);
1197 extern void tcp_free_md5sig_pool(void);
1198
1199 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1200 extern void __tcp_put_md5sig_pool(void);
1201 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1202 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1203 unsigned header_len);
1204 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1205 struct tcp_md5sig_key *key);
1206
1207 static inline
1208 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1209 {
1210 int cpu = get_cpu();
1211 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1212 if (!ret)
1213 put_cpu();
1214 return ret;
1215 }
1216
1217 static inline void tcp_put_md5sig_pool(void)
1218 {
1219 __tcp_put_md5sig_pool();
1220 put_cpu();
1221 }
1222
1223 /* write queue abstraction */
1224 static inline void tcp_write_queue_purge(struct sock *sk)
1225 {
1226 struct sk_buff *skb;
1227
1228 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1229 sk_wmem_free_skb(sk, skb);
1230 sk_mem_reclaim(sk);
1231 tcp_clear_all_retrans_hints(tcp_sk(sk));
1232 }
1233
1234 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1235 {
1236 return skb_peek(&sk->sk_write_queue);
1237 }
1238
1239 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1240 {
1241 return skb_peek_tail(&sk->sk_write_queue);
1242 }
1243
1244 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1245 {
1246 return skb_queue_next(&sk->sk_write_queue, skb);
1247 }
1248
1249 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1250 {
1251 return skb_queue_prev(&sk->sk_write_queue, skb);
1252 }
1253
1254 #define tcp_for_write_queue(skb, sk) \
1255 skb_queue_walk(&(sk)->sk_write_queue, skb)
1256
1257 #define tcp_for_write_queue_from(skb, sk) \
1258 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1259
1260 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1261 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1262
1263 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1264 {
1265 return sk->sk_send_head;
1266 }
1267
1268 static inline bool tcp_skb_is_last(const struct sock *sk,
1269 const struct sk_buff *skb)
1270 {
1271 return skb_queue_is_last(&sk->sk_write_queue, skb);
1272 }
1273
1274 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1275 {
1276 if (tcp_skb_is_last(sk, skb))
1277 sk->sk_send_head = NULL;
1278 else
1279 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1280 }
1281
1282 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1283 {
1284 if (sk->sk_send_head == skb_unlinked)
1285 sk->sk_send_head = NULL;
1286 }
1287
1288 static inline void tcp_init_send_head(struct sock *sk)
1289 {
1290 sk->sk_send_head = NULL;
1291 }
1292
1293 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1294 {
1295 __skb_queue_tail(&sk->sk_write_queue, skb);
1296 }
1297
1298 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1299 {
1300 __tcp_add_write_queue_tail(sk, skb);
1301
1302 /* Queue it, remembering where we must start sending. */
1303 if (sk->sk_send_head == NULL) {
1304 sk->sk_send_head = skb;
1305
1306 if (tcp_sk(sk)->highest_sack == NULL)
1307 tcp_sk(sk)->highest_sack = skb;
1308 }
1309 }
1310
1311 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1312 {
1313 __skb_queue_head(&sk->sk_write_queue, skb);
1314 }
1315
1316 /* Insert buff after skb on the write queue of sk. */
1317 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1318 struct sk_buff *buff,
1319 struct sock *sk)
1320 {
1321 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1322 }
1323
1324 /* Insert new before skb on the write queue of sk. */
1325 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1326 struct sk_buff *skb,
1327 struct sock *sk)
1328 {
1329 __skb_queue_before(&sk->sk_write_queue, skb, new);
1330
1331 if (sk->sk_send_head == skb)
1332 sk->sk_send_head = new;
1333 }
1334
1335 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1336 {
1337 __skb_unlink(skb, &sk->sk_write_queue);
1338 }
1339
1340 static inline int tcp_write_queue_empty(struct sock *sk)
1341 {
1342 return skb_queue_empty(&sk->sk_write_queue);
1343 }
1344
1345 static inline void tcp_push_pending_frames(struct sock *sk)
1346 {
1347 if (tcp_send_head(sk)) {
1348 struct tcp_sock *tp = tcp_sk(sk);
1349
1350 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1351 }
1352 }
1353
1354 /* Start sequence of the highest skb with SACKed bit, valid only if
1355 * sacked > 0 or when the caller has ensured validity by itself.
1356 */
1357 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1358 {
1359 if (!tp->sacked_out)
1360 return tp->snd_una;
1361
1362 if (tp->highest_sack == NULL)
1363 return tp->snd_nxt;
1364
1365 return TCP_SKB_CB(tp->highest_sack)->seq;
1366 }
1367
1368 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1369 {
1370 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1371 tcp_write_queue_next(sk, skb);
1372 }
1373
1374 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1375 {
1376 return tcp_sk(sk)->highest_sack;
1377 }
1378
1379 static inline void tcp_highest_sack_reset(struct sock *sk)
1380 {
1381 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1382 }
1383
1384 /* Called when old skb is about to be deleted (to be combined with new skb) */
1385 static inline void tcp_highest_sack_combine(struct sock *sk,
1386 struct sk_buff *old,
1387 struct sk_buff *new)
1388 {
1389 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1390 tcp_sk(sk)->highest_sack = new;
1391 }
1392
1393 /* Determines whether this is a thin stream (which may suffer from
1394 * increased latency). Used to trigger latency-reducing mechanisms.
1395 */
1396 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp)
1397 {
1398 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1399 }
1400
1401 /* /proc */
1402 enum tcp_seq_states {
1403 TCP_SEQ_STATE_LISTENING,
1404 TCP_SEQ_STATE_OPENREQ,
1405 TCP_SEQ_STATE_ESTABLISHED,
1406 TCP_SEQ_STATE_TIME_WAIT,
1407 };
1408
1409 struct tcp_seq_afinfo {
1410 char *name;
1411 sa_family_t family;
1412 struct file_operations seq_fops;
1413 struct seq_operations seq_ops;
1414 };
1415
1416 struct tcp_iter_state {
1417 struct seq_net_private p;
1418 sa_family_t family;
1419 enum tcp_seq_states state;
1420 struct sock *syn_wait_sk;
1421 int bucket, sbucket, num, uid;
1422 };
1423
1424 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1425 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1426
1427 extern struct request_sock_ops tcp_request_sock_ops;
1428 extern struct request_sock_ops tcp6_request_sock_ops;
1429
1430 extern void tcp_v4_destroy_sock(struct sock *sk);
1431
1432 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1433 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1434 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1435 struct sk_buff *skb);
1436 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1437 struct sk_buff *skb);
1438 extern int tcp_gro_complete(struct sk_buff *skb);
1439 extern int tcp4_gro_complete(struct sk_buff *skb);
1440
1441 #ifdef CONFIG_PROC_FS
1442 extern int tcp4_proc_init(void);
1443 extern void tcp4_proc_exit(void);
1444 #endif
1445
1446 /* TCP af-specific functions */
1447 struct tcp_sock_af_ops {
1448 #ifdef CONFIG_TCP_MD5SIG
1449 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1450 struct sock *addr_sk);
1451 int (*calc_md5_hash) (char *location,
1452 struct tcp_md5sig_key *md5,
1453 struct sock *sk,
1454 struct request_sock *req,
1455 struct sk_buff *skb);
1456 int (*md5_add) (struct sock *sk,
1457 struct sock *addr_sk,
1458 u8 *newkey,
1459 u8 len);
1460 int (*md5_parse) (struct sock *sk,
1461 char __user *optval,
1462 int optlen);
1463 #endif
1464 };
1465
1466 struct tcp_request_sock_ops {
1467 #ifdef CONFIG_TCP_MD5SIG
1468 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1469 struct request_sock *req);
1470 int (*calc_md5_hash) (char *location,
1471 struct tcp_md5sig_key *md5,
1472 struct sock *sk,
1473 struct request_sock *req,
1474 struct sk_buff *skb);
1475 #endif
1476 };
1477
1478 /* Using SHA1 for now, define some constants.
1479 */
1480 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1481 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1482 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1483
1484 extern int tcp_cookie_generator(u32 *bakery);
1485
1486 /**
1487 * struct tcp_cookie_values - each socket needs extra space for the
1488 * cookies, together with (optional) space for any SYN data.
1489 *
1490 * A tcp_sock contains a pointer to the current value, and this is
1491 * cloned to the tcp_timewait_sock.
1492 *
1493 * @cookie_pair: variable data from the option exchange.
1494 *
1495 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1496 * indicates default (sysctl_tcp_cookie_size).
1497 * After cookie sent, remembers size of cookie.
1498 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1499 *
1500 * @s_data_desired: user specified tcpct_s_data_desired. When the
1501 * constant payload is specified (@s_data_constant),
1502 * holds its length instead.
1503 * Range 0 to TCP_MSS_DESIRED.
1504 *
1505 * @s_data_payload: constant data that is to be included in the
1506 * payload of SYN or SYNACK segments when the
1507 * cookie option is present.
1508 */
1509 struct tcp_cookie_values {
1510 struct kref kref;
1511 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1512 u8 cookie_pair_size;
1513 u8 cookie_desired;
1514 u16 s_data_desired:11,
1515 s_data_constant:1,
1516 s_data_in:1,
1517 s_data_out:1,
1518 s_data_unused:2;
1519 u8 s_data_payload[0];
1520 };
1521
1522 static inline void tcp_cookie_values_release(struct kref *kref)
1523 {
1524 kfree(container_of(kref, struct tcp_cookie_values, kref));
1525 }
1526
1527 /* The length of constant payload data. Note that s_data_desired is
1528 * overloaded, depending on s_data_constant: either the length of constant
1529 * data (returned here) or the limit on variable data.
1530 */
1531 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1532 {
1533 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1534 ? tp->cookie_values->s_data_desired
1535 : 0;
1536 }
1537
1538 /**
1539 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1540 *
1541 * As tcp_request_sock has already been extended in other places, the
1542 * only remaining method is to pass stack values along as function
1543 * parameters. These parameters are not needed after sending SYNACK.
1544 *
1545 * @cookie_bakery: cryptographic secret and message workspace.
1546 *
1547 * @cookie_plus: bytes in authenticator/cookie option, copied from
1548 * struct tcp_options_received (above).
1549 */
1550 struct tcp_extend_values {
1551 struct request_values rv;
1552 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1553 u8 cookie_plus:6,
1554 cookie_out_never:1,
1555 cookie_in_always:1;
1556 };
1557
1558 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1559 {
1560 return (struct tcp_extend_values *)rvp;
1561 }
1562
1563 extern void tcp_v4_init(void);
1564 extern void tcp_init(void);
1565
1566 #endif /* _TCP_H */
Ubuntu Jammy Linux kernel mirror