]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - include/net/tcp.h
qlge: Replace create_singlethread_workqueue with alloc_ordered_workqueue
[mirror_ubuntu-bionic-kernel.git] / include / net / tcp.h
CommitLineData
1da177e4
LT
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 *
02c30a84 10 * Authors: Ross Biro
1da177e4
LT
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
1da177e4
LT
21#define FASTRETRANS_DEBUG 1
22
1da177e4
LT
23#include <linux/list.h>
24#include <linux/tcp.h>
187f1882 25#include <linux/bug.h>
1da177e4
LT
26#include <linux/slab.h>
27#include <linux/cache.h>
28#include <linux/percpu.h>
fb286bb2 29#include <linux/skbuff.h>
c6aefafb 30#include <linux/cryptohash.h>
435cf559 31#include <linux/kref.h>
740b0f18 32#include <linux/ktime.h>
3f421baa
ACM
33
34#include <net/inet_connection_sock.h>
295ff7ed 35#include <net/inet_timewait_sock.h>
77d8bf9c 36#include <net/inet_hashtables.h>
1da177e4 37#include <net/checksum.h>
2e6599cb 38#include <net/request_sock.h>
1da177e4
LT
39#include <net/sock.h>
40#include <net/snmp.h>
41#include <net/ip.h>
c752f073 42#include <net/tcp_states.h>
bdf1ee5d 43#include <net/inet_ecn.h>
0c266898 44#include <net/dst.h>
c752f073 45
1da177e4 46#include <linux/seq_file.h>
180d8cd9 47#include <linux/memcontrol.h>
1da177e4 48
6e04e021 49extern struct inet_hashinfo tcp_hashinfo;
1da177e4 50
dd24c001 51extern struct percpu_counter tcp_orphan_count;
5c9f3023 52void tcp_time_wait(struct sock *sk, int state, int timeo);
1da177e4 53
1da177e4 54#define MAX_TCP_HEADER (128 + MAX_HEADER)
33ad798c 55#define MAX_TCP_OPTION_SPACE 40
1da177e4 56
105970f6 57/*
1da177e4 58 * Never offer a window over 32767 without using window scaling. Some
105970f6 59 * poor stacks do signed 16bit maths!
1da177e4
LT
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
5d424d5a 66/* The least MTU to use for probing */
dcd8fb85 67#define TCP_BASE_MSS 1024
5d424d5a 68
05cbc0db
FD
69/* probing interval, default to 10 minutes as per RFC4821 */
70#define TCP_PROBE_INTERVAL 600
71
6b58e0a5
FD
72/* Specify interval when tcp mtu probing will stop */
73#define TCP_PROBE_THRESHOLD 8
74
1da177e4
LT
75/* After receiving this amount of duplicate ACKs fast retransmit starts. */
76#define TCP_FASTRETRANS_THRESH 3
77
1da177e4
LT
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
6c9ff979
AB
100#define TCP_SYN_RETRIES 6 /* This is how many retries are done
101 * when active opening a connection.
102 * RFC1122 says the minimum retry MUST
103 * be at least 180secs. Nevertheless
104 * this value is corresponding to
105 * 63secs of retransmission with the
106 * current initial RTO.
107 */
1da177e4 108
6c9ff979
AB
109#define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
110 * when passive opening a connection.
111 * This is corresponding to 31secs of
112 * retransmission with the current
113 * initial RTO.
114 */
1da177e4 115
1da177e4
LT
116#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
117 * state, about 60 seconds */
118#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
119 /* BSD style FIN_WAIT2 deadlock breaker.
120 * It used to be 3min, new value is 60sec,
121 * to combine FIN-WAIT-2 timeout with
122 * TIME-WAIT timer.
123 */
124
125#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
126#if HZ >= 100
127#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
128#define TCP_ATO_MIN ((unsigned)(HZ/25))
129#else
130#define TCP_DELACK_MIN 4U
131#define TCP_ATO_MIN 4U
132#endif
133#define TCP_RTO_MAX ((unsigned)(120*HZ))
134#define TCP_RTO_MIN ((unsigned)(HZ/5))
fd4f2cea 135#define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
9ad7c049
JC
136#define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
137 * used as a fallback RTO for the
138 * initial data transmission if no
139 * valid RTT sample has been acquired,
140 * most likely due to retrans in 3WHS.
141 */
1da177e4
LT
142
143#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
144 * for local resources.
145 */
146
147#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
148#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
149#define TCP_KEEPALIVE_INTVL (75*HZ)
150
151#define MAX_TCP_KEEPIDLE 32767
152#define MAX_TCP_KEEPINTVL 32767
153#define MAX_TCP_KEEPCNT 127
154#define MAX_TCP_SYNCNT 127
155
156#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
1da177e4
LT
157
158#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
159#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
160 * after this time. It should be equal
161 * (or greater than) TCP_TIMEWAIT_LEN
162 * to provide reliability equal to one
163 * provided by timewait state.
164 */
165#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
166 * timestamps. It must be less than
167 * minimal timewait lifetime.
168 */
1da177e4
LT
169/*
170 * TCP option
171 */
105970f6 172
1da177e4
LT
173#define TCPOPT_NOP 1 /* Padding */
174#define TCPOPT_EOL 0 /* End of options */
175#define TCPOPT_MSS 2 /* Segment size negotiating */
176#define TCPOPT_WINDOW 3 /* Window scaling */
177#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
178#define TCPOPT_SACK 5 /* SACK Block */
179#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
cfb6eeb4 180#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
7f9b838b 181#define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
2100c8d2
YC
182#define TCPOPT_EXP 254 /* Experimental */
183/* Magic number to be after the option value for sharing TCP
184 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
185 */
186#define TCPOPT_FASTOPEN_MAGIC 0xF989
1da177e4
LT
187
188/*
189 * TCP option lengths
190 */
191
192#define TCPOLEN_MSS 4
193#define TCPOLEN_WINDOW 3
194#define TCPOLEN_SACK_PERM 2
195#define TCPOLEN_TIMESTAMP 10
cfb6eeb4 196#define TCPOLEN_MD5SIG 18
7f9b838b 197#define TCPOLEN_FASTOPEN_BASE 2
2100c8d2 198#define TCPOLEN_EXP_FASTOPEN_BASE 4
1da177e4
LT
199
200/* But this is what stacks really send out. */
201#define TCPOLEN_TSTAMP_ALIGNED 12
202#define TCPOLEN_WSCALE_ALIGNED 4
203#define TCPOLEN_SACKPERM_ALIGNED 4
204#define TCPOLEN_SACK_BASE 2
205#define TCPOLEN_SACK_BASE_ALIGNED 4
206#define TCPOLEN_SACK_PERBLOCK 8
cfb6eeb4 207#define TCPOLEN_MD5SIG_ALIGNED 20
33ad798c 208#define TCPOLEN_MSS_ALIGNED 4
1da177e4 209
1da177e4
LT
210/* Flags in tp->nonagle */
211#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
212#define TCP_NAGLE_CORK 2 /* Socket is corked */
caa20d9a 213#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
1da177e4 214
36e31b0a
AP
215/* TCP thin-stream limits */
216#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
217
21603fc4 218/* TCP initial congestion window as per rfc6928 */
442b9635
DM
219#define TCP_INIT_CWND 10
220
cf60af03
YC
221/* Bit Flags for sysctl_tcp_fastopen */
222#define TFO_CLIENT_ENABLE 1
10467163 223#define TFO_SERVER_ENABLE 2
67da22d2 224#define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
cf60af03 225
10467163
JC
226/* Accept SYN data w/o any cookie option */
227#define TFO_SERVER_COOKIE_NOT_REQD 0x200
228
229/* Force enable TFO on all listeners, i.e., not requiring the
230 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
231 */
232#define TFO_SERVER_WO_SOCKOPT1 0x400
233#define TFO_SERVER_WO_SOCKOPT2 0x800
10467163 234
295ff7ed
ACM
235extern struct inet_timewait_death_row tcp_death_row;
236
1da177e4 237/* sysctl variables for tcp */
1da177e4
LT
238extern int sysctl_tcp_timestamps;
239extern int sysctl_tcp_window_scaling;
240extern int sysctl_tcp_sack;
2100c8d2 241extern int sysctl_tcp_fastopen;
1da177e4
LT
242extern int sysctl_tcp_retrans_collapse;
243extern int sysctl_tcp_stdurg;
244extern int sysctl_tcp_rfc1337;
245extern int sysctl_tcp_abort_on_overflow;
246extern int sysctl_tcp_max_orphans;
1da177e4
LT
247extern int sysctl_tcp_fack;
248extern int sysctl_tcp_reordering;
dca145ff 249extern int sysctl_tcp_max_reordering;
1da177e4 250extern int sysctl_tcp_dsack;
a4fe34bf 251extern long sysctl_tcp_mem[3];
1da177e4
LT
252extern int sysctl_tcp_wmem[3];
253extern int sysctl_tcp_rmem[3];
254extern int sysctl_tcp_app_win;
255extern int sysctl_tcp_adv_win_scale;
256extern int sysctl_tcp_tw_reuse;
257extern int sysctl_tcp_frto;
258extern int sysctl_tcp_low_latency;
1da177e4 259extern int sysctl_tcp_nometrics_save;
1da177e4
LT
260extern int sysctl_tcp_moderate_rcvbuf;
261extern int sysctl_tcp_tso_win_divisor;
15d99e02 262extern int sysctl_tcp_workaround_signed_windows;
35089bb2 263extern int sysctl_tcp_slow_start_after_idle;
36e31b0a 264extern int sysctl_tcp_thin_linear_timeouts;
7e380175 265extern int sysctl_tcp_thin_dupack;
eed530b6 266extern int sysctl_tcp_early_retrans;
46d3ceab 267extern int sysctl_tcp_limit_output_bytes;
282f23c6 268extern int sysctl_tcp_challenge_ack_limit;
95bd09eb 269extern int sysctl_tcp_min_tso_segs;
f6722583 270extern int sysctl_tcp_min_rtt_wlen;
f54b3111 271extern int sysctl_tcp_autocorking;
032ee423 272extern int sysctl_tcp_invalid_ratelimit;
43e122b0
ED
273extern int sysctl_tcp_pacing_ss_ratio;
274extern int sysctl_tcp_pacing_ca_ratio;
1da177e4 275
8d987e5c 276extern atomic_long_t tcp_memory_allocated;
1748376b 277extern struct percpu_counter tcp_sockets_allocated;
1da177e4
LT
278extern int tcp_memory_pressure;
279
b8da51eb
ED
280/* optimized version of sk_under_memory_pressure() for TCP sockets */
281static inline bool tcp_under_memory_pressure(const struct sock *sk)
282{
baac50bb
JW
283 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
284 mem_cgroup_under_socket_pressure(sk->sk_memcg))
e805605c 285 return true;
b8da51eb
ED
286
287 return tcp_memory_pressure;
288}
1da177e4
LT
289/*
290 * The next routines deal with comparing 32 bit unsigned ints
291 * and worry about wraparound (automatic with unsigned arithmetic).
292 */
293
a2a385d6 294static inline bool before(__u32 seq1, __u32 seq2)
1da177e4 295{
0d630cc0 296 return (__s32)(seq1-seq2) < 0;
1da177e4 297}
9a036b9c 298#define after(seq2, seq1) before(seq1, seq2)
1da177e4
LT
299
300/* is s2<=s1<=s3 ? */
a2a385d6 301static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
1da177e4
LT
302{
303 return seq3 - seq2 >= seq1 - seq2;
304}
305
efcdbf24
AS
306static inline bool tcp_out_of_memory(struct sock *sk)
307{
308 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
309 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
310 return true;
311 return false;
312}
313
a6c5ea4c
ED
314void sk_forced_mem_schedule(struct sock *sk, int size);
315
ad1af0fe 316static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
e4fd5da3 317{
ad1af0fe
DM
318 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
319 int orphans = percpu_counter_read_positive(ocp);
320
321 if (orphans << shift > sysctl_tcp_max_orphans) {
322 orphans = percpu_counter_sum_positive(ocp);
323 if (orphans << shift > sysctl_tcp_max_orphans)
324 return true;
325 }
ad1af0fe 326 return false;
e4fd5da3 327}
1da177e4 328
5c9f3023 329bool tcp_check_oom(struct sock *sk, int shift);
efcdbf24 330
a0f82f64 331
1da177e4
LT
332extern struct proto tcp_prot;
333
57ef42d5
PE
334#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
335#define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
336#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
337#define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
aa2ea058 338#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
1da177e4 339
5c9f3023
JP
340void tcp_tasklet_init(void);
341
342void tcp_v4_err(struct sk_buff *skb, u32);
343
344void tcp_shutdown(struct sock *sk, int how);
345
346void tcp_v4_early_demux(struct sk_buff *skb);
347int tcp_v4_rcv(struct sk_buff *skb);
348
349int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
1b784140 350int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
5c9f3023
JP
351int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
352 int flags);
353void tcp_release_cb(struct sock *sk);
354void tcp_wfree(struct sk_buff *skb);
355void tcp_write_timer_handler(struct sock *sk);
356void tcp_delack_timer_handler(struct sock *sk);
357int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
72ab4a86 358int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
5c9f3023
JP
359void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
360 const struct tcphdr *th, unsigned int len);
361void tcp_rcv_space_adjust(struct sock *sk);
5c9f3023
JP
362int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
363void tcp_twsk_destructor(struct sock *sk);
364ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
365 struct pipe_inode_info *pipe, size_t len,
366 unsigned int flags);
9c55e01c 367
463c84b9
ACM
368static inline void tcp_dec_quickack_mode(struct sock *sk,
369 const unsigned int pkts)
1da177e4 370{
463c84b9 371 struct inet_connection_sock *icsk = inet_csk(sk);
fc6415bc 372
463c84b9
ACM
373 if (icsk->icsk_ack.quick) {
374 if (pkts >= icsk->icsk_ack.quick) {
375 icsk->icsk_ack.quick = 0;
fc6415bc 376 /* Leaving quickack mode we deflate ATO. */
463c84b9 377 icsk->icsk_ack.ato = TCP_ATO_MIN;
fc6415bc 378 } else
463c84b9 379 icsk->icsk_ack.quick -= pkts;
1da177e4
LT
380 }
381}
382
bdf1ee5d
IJ
383#define TCP_ECN_OK 1
384#define TCP_ECN_QUEUE_CWR 2
385#define TCP_ECN_DEMAND_CWR 4
7a269ffa 386#define TCP_ECN_SEEN 8
bdf1ee5d 387
fd2c3ef7 388enum tcp_tw_status {
1da177e4
LT
389 TCP_TW_SUCCESS = 0,
390 TCP_TW_RST = 1,
391 TCP_TW_ACK = 2,
392 TCP_TW_SYN = 3
393};
394
395
5c9f3023
JP
396enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
397 struct sk_buff *skb,
398 const struct tcphdr *th);
399struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
52452c54 400 struct request_sock *req, bool fastopen);
5c9f3023
JP
401int tcp_child_process(struct sock *parent, struct sock *child,
402 struct sk_buff *skb);
5ae344c9 403void tcp_enter_loss(struct sock *sk);
5c9f3023
JP
404void tcp_clear_retrans(struct tcp_sock *tp);
405void tcp_update_metrics(struct sock *sk);
406void tcp_init_metrics(struct sock *sk);
407void tcp_metrics_init(void);
408bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
a26552af 409 bool paws_check, bool timestamps);
5c9f3023
JP
410bool tcp_remember_stamp(struct sock *sk);
411bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
412void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
413void tcp_disable_fack(struct tcp_sock *tp);
414void tcp_close(struct sock *sk, long timeout);
415void tcp_init_sock(struct sock *sk);
416unsigned int tcp_poll(struct file *file, struct socket *sock,
417 struct poll_table_struct *wait);
418int tcp_getsockopt(struct sock *sk, int level, int optname,
419 char __user *optval, int __user *optlen);
420int tcp_setsockopt(struct sock *sk, int level, int optname,
421 char __user *optval, unsigned int optlen);
422int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
53d3176b 423 char __user *optval, int __user *optlen);
5c9f3023 424int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
53d3176b 425 char __user *optval, unsigned int optlen);
5c9f3023 426void tcp_set_keepalive(struct sock *sk, int val);
42cb80a2 427void tcp_syn_ack_timeout(const struct request_sock *req);
1b784140
YX
428int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
429 int flags, int *addr_len);
5c9f3023
JP
430void tcp_parse_options(const struct sk_buff *skb,
431 struct tcp_options_received *opt_rx,
432 int estab, struct tcp_fastopen_cookie *foc);
433const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
7d5d5525 434
1da177e4
LT
435/*
436 * TCP v4 functions exported for the inet6 API
437 */
438
5c9f3023 439void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
4fab9071 440void tcp_v4_mtu_reduced(struct sock *sk);
9cf74903 441void tcp_req_err(struct sock *sk, u32 seq, bool abort);
5c9f3023 442int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
c28c6f04 443struct sock *tcp_create_openreq_child(const struct sock *sk,
5c9f3023
JP
444 struct request_sock *req,
445 struct sk_buff *skb);
81164413 446void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
0c27171e 447struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
5c9f3023 448 struct request_sock *req,
5e0724d0
ED
449 struct dst_entry *dst,
450 struct request_sock *req_unhash,
451 bool *own_req);
5c9f3023
JP
452int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
453int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
454int tcp_connect(struct sock *sk);
5d062de7 455struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
5c9f3023 456 struct request_sock *req,
ca6fb065
ED
457 struct tcp_fastopen_cookie *foc,
458 bool attach_req);
5c9f3023 459int tcp_disconnect(struct sock *sk, int flags);
1da177e4 460
370816ae 461void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
292e8d8c 462int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
63d02d15 463void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
1da177e4 464
1da177e4 465/* From syncookies.c */
b80c0e78
ED
466struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
467 struct request_sock *req,
468 struct dst_entry *dst);
5c9f3023
JP
469int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
470 u32 cookie);
461b74c3 471struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
e05c82d3 472#ifdef CONFIG_SYN_COOKIES
8c27bd75 473
63262315 474/* Syncookies use a monotonic timer which increments every 60 seconds.
8c27bd75
FW
475 * This counter is used both as a hash input and partially encoded into
476 * the cookie value. A cookie is only validated further if the delta
477 * between the current counter value and the encoded one is less than this,
63262315 478 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
8c27bd75
FW
479 * the counter advances immediately after a cookie is generated).
480 */
264ea103
ED
481#define MAX_SYNCOOKIE_AGE 2
482#define TCP_SYNCOOKIE_PERIOD (60 * HZ)
483#define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
484
485/* syncookies: remember time of last synqueue overflow
486 * But do not dirty this field too often (once per second is enough)
3f684b4b 487 * It is racy as we do not hold a lock, but race is very minor.
264ea103 488 */
3f684b4b 489static inline void tcp_synq_overflow(const struct sock *sk)
264ea103
ED
490{
491 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
492 unsigned long now = jiffies;
493
494 if (time_after(now, last_overflow + HZ))
495 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
496}
497
498/* syncookies: no recent synqueue overflow on this listening socket? */
499static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
500{
501 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
502
503 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
504}
8c27bd75
FW
505
506static inline u32 tcp_cookie_time(void)
507{
63262315
ED
508 u64 val = get_jiffies_64();
509
264ea103 510 do_div(val, TCP_SYNCOOKIE_PERIOD);
63262315 511 return val;
8c27bd75
FW
512}
513
5c9f3023
JP
514u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
515 u16 *mssp);
3f684b4b 516__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
5c9f3023 517__u32 cookie_init_timestamp(struct request_sock *req);
f1673381
FW
518bool cookie_timestamp_decode(struct tcp_options_received *opt);
519bool cookie_ecn_ok(const struct tcp_options_received *opt,
f7b3bec6 520 const struct net *net, const struct dst_entry *dst);
4dfc2817 521
c6aefafb 522/* From net/ipv6/syncookies.c */
5c9f3023
JP
523int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
524 u32 cookie);
525struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
f1673381 526
5c9f3023
JP
527u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
528 const struct tcphdr *th, u16 *mssp);
3f684b4b 529__u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
e05c82d3 530#endif
1da177e4
LT
531/* tcp_output.c */
532
5c9f3023
JP
533void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
534 int nonagle);
535bool tcp_may_send_now(struct sock *sk);
536int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
537int tcp_retransmit_skb(struct sock *, struct sk_buff *);
538void tcp_retransmit_timer(struct sock *sk);
539void tcp_xmit_retransmit_queue(struct sock *);
540void tcp_simple_retransmit(struct sock *);
541int tcp_trim_head(struct sock *, struct sk_buff *, u32);
6cc55e09 542int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
5c9f3023
JP
543
544void tcp_send_probe0(struct sock *);
545void tcp_send_partial(struct sock *);
e520af48 546int tcp_write_wakeup(struct sock *, int mib);
5c9f3023
JP
547void tcp_send_fin(struct sock *sk);
548void tcp_send_active_reset(struct sock *sk, gfp_t priority);
549int tcp_send_synack(struct sock *);
5c9f3023
JP
550void tcp_push_one(struct sock *, unsigned int mss_now);
551void tcp_send_ack(struct sock *sk);
552void tcp_send_delayed_ack(struct sock *sk);
553void tcp_send_loss_probe(struct sock *sk);
554bool tcp_schedule_loss_probe(struct sock *sk);
1da177e4 555
a762a980 556/* tcp_input.c */
5c9f3023
JP
557void tcp_resume_early_retransmit(struct sock *sk);
558void tcp_rearm_rto(struct sock *sk);
0f1c28ae 559void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
5c9f3023 560void tcp_reset(struct sock *sk);
4f41b1c5 561void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
e3e17b77 562void tcp_fin(struct sock *sk);
a762a980 563
1da177e4 564/* tcp_timer.c */
5c9f3023 565void tcp_init_xmit_timers(struct sock *);
463c84b9
ACM
566static inline void tcp_clear_xmit_timers(struct sock *sk)
567{
568 inet_csk_clear_xmit_timers(sk);
569}
1da177e4 570
5c9f3023
JP
571unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
572unsigned int tcp_current_mss(struct sock *sk);
0c54b85f
IJ
573
574/* Bound MSS / TSO packet size with the half of the window */
575static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
576{
01f83d69
AK
577 int cutoff;
578
579 /* When peer uses tiny windows, there is no use in packetizing
580 * to sub-MSS pieces for the sake of SWS or making sure there
581 * are enough packets in the pipe for fast recovery.
582 *
583 * On the other hand, for extremely large MSS devices, handling
584 * smaller than MSS windows in this way does make sense.
585 */
586 if (tp->max_window >= 512)
587 cutoff = (tp->max_window >> 1);
588 else
589 cutoff = tp->max_window;
590
591 if (cutoff && pktsize > cutoff)
592 return max_t(int, cutoff, 68U - tp->tcp_header_len);
0c54b85f
IJ
593 else
594 return pktsize;
595}
1da177e4 596
17b085ea 597/* tcp.c */
0df48c26 598void tcp_get_info(struct sock *, struct tcp_info *);
1da177e4
LT
599
600/* Read 'sendfile()'-style from a TCP socket */
601typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
602 unsigned int, size_t);
5c9f3023
JP
603int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
604 sk_read_actor_t recv_actor);
1da177e4 605
5c9f3023 606void tcp_initialize_rcv_mss(struct sock *sk);
1da177e4 607
5c9f3023
JP
608int tcp_mtu_to_mss(struct sock *sk, int pmtu);
609int tcp_mss_to_mtu(struct sock *sk, int mss);
610void tcp_mtup_init(struct sock *sk);
611void tcp_init_buffer_space(struct sock *sk);
5d424d5a 612
f1ecd5d9
DL
613static inline void tcp_bound_rto(const struct sock *sk)
614{
615 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
616 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
617}
618
619static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
620{
740b0f18 621 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
f1ecd5d9
DL
622}
623
40efc6fa 624static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
1da177e4
LT
625{
626 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
627 ntohl(TCP_FLAG_ACK) |
628 snd_wnd);
629}
630
40efc6fa 631static inline void tcp_fast_path_on(struct tcp_sock *tp)
1da177e4
LT
632{
633 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
634}
635
9e412ba7 636static inline void tcp_fast_path_check(struct sock *sk)
1da177e4 637{
9e412ba7
IJ
638 struct tcp_sock *tp = tcp_sk(sk);
639
b03efcfb 640 if (skb_queue_empty(&tp->out_of_order_queue) &&
1da177e4
LT
641 tp->rcv_wnd &&
642 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
643 !tp->urg_data)
644 tcp_fast_path_on(tp);
645}
646
0c266898
SS
647/* Compute the actual rto_min value */
648static inline u32 tcp_rto_min(struct sock *sk)
649{
cf533ea5 650 const struct dst_entry *dst = __sk_dst_get(sk);
0c266898
SS
651 u32 rto_min = TCP_RTO_MIN;
652
653 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
654 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
655 return rto_min;
656}
657
740b0f18
ED
658static inline u32 tcp_rto_min_us(struct sock *sk)
659{
660 return jiffies_to_usecs(tcp_rto_min(sk));
661}
662
81164413
DB
663static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
664{
665 return dst_metric_locked(dst, RTAX_CC_ALGO);
666}
667
f6722583
YC
668/* Minimum RTT in usec. ~0 means not available. */
669static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
670{
671 return tp->rtt_min[0].rtt;
672}
673
1da177e4
LT
674/* Compute the actual receive window we are currently advertising.
675 * Rcv_nxt can be after the window if our peer push more data
676 * than the offered window.
677 */
40efc6fa 678static inline u32 tcp_receive_window(const struct tcp_sock *tp)
1da177e4
LT
679{
680 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
681
682 if (win < 0)
683 win = 0;
684 return (u32) win;
685}
686
687/* Choose a new window, without checks for shrinking, and without
688 * scaling applied to the result. The caller does these things
689 * if necessary. This is a "raw" window selection.
690 */
5c9f3023 691u32 __tcp_select_window(struct sock *sk);
1da177e4 692
ee995283
PE
693void tcp_send_window_probe(struct sock *sk);
694
1da177e4
LT
695/* TCP timestamps are only 32-bits, this causes a slight
696 * complication on 64-bit systems since we store a snapshot
31f34269
SH
697 * of jiffies in the buffer control blocks below. We decided
698 * to use only the low 32-bits of jiffies and hide the ugly
1da177e4
LT
699 * casts with the following macro.
700 */
701#define tcp_time_stamp ((__u32)(jiffies))
702
7faee5c0
ED
703static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
704{
705 return skb->skb_mstamp.stamp_jiffies;
706}
707
708
a3433f35
CG
709#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
710
711#define TCPHDR_FIN 0x01
712#define TCPHDR_SYN 0x02
713#define TCPHDR_RST 0x04
714#define TCPHDR_PSH 0x08
715#define TCPHDR_ACK 0x10
716#define TCPHDR_URG 0x20
717#define TCPHDR_ECE 0x40
718#define TCPHDR_CWR 0x80
719
49213555
DB
720#define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
721
caa20d9a 722/* This is what the send packet queuing engine uses to pass
f86586fa
ED
723 * TCP per-packet control information to the transmission code.
724 * We also store the host-order sequence numbers in here too.
725 * This is 44 bytes if IPV6 is enabled.
726 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
1da177e4
LT
727 */
728struct tcp_skb_cb {
1da177e4
LT
729 __u32 seq; /* Starting sequence number */
730 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
cd7d8498
ED
731 union {
732 /* Note : tcp_tw_isn is used in input path only
733 * (isn chosen by tcp_timewait_state_process())
734 *
f69ad292
ED
735 * tcp_gso_segs/size are used in write queue only,
736 * cf tcp_skb_pcount()/tcp_skb_mss()
cd7d8498
ED
737 */
738 __u32 tcp_tw_isn;
f69ad292
ED
739 struct {
740 u16 tcp_gso_segs;
741 u16 tcp_gso_size;
742 };
cd7d8498 743 };
4de075e0 744 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
f4f9f6e7 745
1da177e4
LT
746 __u8 sacked; /* State flags for SACK/FACK. */
747#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
748#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
749#define TCPCB_LOST 0x04 /* SKB is lost */
750#define TCPCB_TAGBITS 0x07 /* All tag bits */
9d186cac 751#define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
1da177e4 752#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
9d186cac
AV
753#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
754 TCPCB_REPAIRED)
1da177e4 755
f4f9f6e7 756 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
6b084928
SHY
757 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
758 unused:7;
1da177e4 759 __u32 ack_seq; /* Sequence number ACK'd */
971f10ec
ED
760 union {
761 struct inet_skb_parm h4;
762#if IS_ENABLED(CONFIG_IPV6)
763 struct inet6_skb_parm h6;
764#endif
765 } header; /* For incoming frames */
1da177e4
LT
766};
767
768#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
769
870c3151 770
815afe17 771#if IS_ENABLED(CONFIG_IPV6)
870c3151
ED
772/* This is the variant of inet6_iif() that must be used by TCP,
773 * as TCP moves IP6CB into a different location in skb->cb[]
774 */
775static inline int tcp_v6_iif(const struct sk_buff *skb)
776{
777 return TCP_SKB_CB(skb)->header.h6.iif;
778}
815afe17 779#endif
870c3151 780
1da177e4
LT
781/* Due to TSO, an SKB can be composed of multiple actual
782 * packets. To keep these tracked properly, we use this.
bd14b1b2 783 */
1da177e4 784static inline int tcp_skb_pcount(const struct sk_buff *skb)
bd14b1b2 785{
cd7d8498
ED
786 return TCP_SKB_CB(skb)->tcp_gso_segs;
787}
bd14b1b2 788
cd7d8498
ED
789static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
790{
791 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
bd14b1b2
ED
792}
793
cd7d8498 794static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
1da177e4 795{
cd7d8498 796 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
1da177e4
LT
797}
798
f69ad292 799/* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
1da177e4
LT
800static inline int tcp_skb_mss(const struct sk_buff *skb)
801{
f69ad292 802 return TCP_SKB_CB(skb)->tcp_gso_size;
1da177e4
LT
803}
804
317a76f9
SH
805/* Events passed to congestion control interface */
806enum tcp_ca_event {
807 CA_EVENT_TX_START, /* first transmit when no packets in flight */
808 CA_EVENT_CWND_RESTART, /* congestion window restart */
809 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
317a76f9 810 CA_EVENT_LOSS, /* loss timeout */
9890092e
FW
811 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
812 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
813 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
814 CA_EVENT_NON_DELAYED_ACK,
7354c8c3
FW
815};
816
9890092e 817/* Information about inbound ACK, passed to cong_ops->in_ack_event() */
7354c8c3 818enum tcp_ca_ack_event_flags {
9890092e
FW
819 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
820 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
821 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
317a76f9
SH
822};
823
824/*
825 * Interface for adding new TCP congestion control handlers
826 */
827#define TCP_CA_NAME_MAX 16
3ff825b2
SH
828#define TCP_CA_MAX 128
829#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
830
c5c6a8ab
DB
831#define TCP_CA_UNSPEC 0
832
30e502a3 833/* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
164891aa 834#define TCP_CONG_NON_RESTRICTED 0x1
30e502a3
DB
835/* Requires ECN/ECT set on all packets */
836#define TCP_CONG_NEEDS_ECN 0x2
164891aa 837
64f40ff5
ED
838union tcp_cc_info;
839
317a76f9
SH
840struct tcp_congestion_ops {
841 struct list_head list;
c5c6a8ab
DB
842 u32 key;
843 u32 flags;
317a76f9
SH
844
845 /* initialize private data (optional) */
6687e988 846 void (*init)(struct sock *sk);
317a76f9 847 /* cleanup private data (optional) */
6687e988 848 void (*release)(struct sock *sk);
317a76f9
SH
849
850 /* return slow start threshold (required) */
6687e988 851 u32 (*ssthresh)(struct sock *sk);
317a76f9 852 /* do new cwnd calculation (required) */
24901551 853 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
317a76f9 854 /* call before changing ca_state (optional) */
6687e988 855 void (*set_state)(struct sock *sk, u8 new_state);
317a76f9 856 /* call when cwnd event occurs (optional) */
6687e988 857 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
7354c8c3
FW
858 /* call when ack arrives (optional) */
859 void (*in_ack_event)(struct sock *sk, u32 flags);
317a76f9 860 /* new value of cwnd after loss (optional) */
6687e988 861 u32 (*undo_cwnd)(struct sock *sk);
317a76f9 862 /* hook for packet ack accounting (optional) */
30cfd0ba 863 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
73c1f4a0 864 /* get info for inet_diag (optional) */
64f40ff5
ED
865 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
866 union tcp_cc_info *info);
317a76f9
SH
867
868 char name[TCP_CA_NAME_MAX];
869 struct module *owner;
870};
871
5c9f3023
JP
872int tcp_register_congestion_control(struct tcp_congestion_ops *type);
873void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
317a76f9 874
55d8694f 875void tcp_assign_congestion_control(struct sock *sk);
5c9f3023
JP
876void tcp_init_congestion_control(struct sock *sk);
877void tcp_cleanup_congestion_control(struct sock *sk);
878int tcp_set_default_congestion_control(const char *name);
879void tcp_get_default_congestion_control(char *name);
880void tcp_get_available_congestion_control(char *buf, size_t len);
881void tcp_get_allowed_congestion_control(char *buf, size_t len);
882int tcp_set_allowed_congestion_control(char *allowed);
883int tcp_set_congestion_control(struct sock *sk, const char *name);
e73ebb08
NC
884u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
885void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
317a76f9 886
5c9f3023 887u32 tcp_reno_ssthresh(struct sock *sk);
24901551 888void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
a8acfbac 889extern struct tcp_congestion_ops tcp_reno;
317a76f9 890
c5c6a8ab 891struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
c3a8d947 892u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
ea697639 893#ifdef CONFIG_INET
c5c6a8ab 894char *tcp_ca_get_name_by_key(u32 key, char *buffer);
ea697639
DB
895#else
896static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
897{
898 return NULL;
899}
900#endif
c5c6a8ab 901
30e502a3
DB
902static inline bool tcp_ca_needs_ecn(const struct sock *sk)
903{
904 const struct inet_connection_sock *icsk = inet_csk(sk);
905
906 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
907}
908
6687e988 909static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
317a76f9 910{
6687e988
ACM
911 struct inet_connection_sock *icsk = inet_csk(sk);
912
913 if (icsk->icsk_ca_ops->set_state)
914 icsk->icsk_ca_ops->set_state(sk, ca_state);
915 icsk->icsk_ca_state = ca_state;
317a76f9
SH
916}
917
6687e988 918static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
317a76f9 919{
6687e988
ACM
920 const struct inet_connection_sock *icsk = inet_csk(sk);
921
922 if (icsk->icsk_ca_ops->cwnd_event)
923 icsk->icsk_ca_ops->cwnd_event(sk, event);
317a76f9
SH
924}
925
e60402d0
IJ
926/* These functions determine how the current flow behaves in respect of SACK
927 * handling. SACK is negotiated with the peer, and therefore it can vary
928 * between different flows.
929 *
930 * tcp_is_sack - SACK enabled
931 * tcp_is_reno - No SACK
932 * tcp_is_fack - FACK enabled, implies SACK enabled
933 */
934static inline int tcp_is_sack(const struct tcp_sock *tp)
935{
936 return tp->rx_opt.sack_ok;
937}
938
a2a385d6 939static inline bool tcp_is_reno(const struct tcp_sock *tp)
e60402d0
IJ
940{
941 return !tcp_is_sack(tp);
942}
943
a2a385d6 944static inline bool tcp_is_fack(const struct tcp_sock *tp)
e60402d0 945{
ab56222a 946 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
e60402d0
IJ
947}
948
949static inline void tcp_enable_fack(struct tcp_sock *tp)
950{
ab56222a 951 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
e60402d0
IJ
952}
953
eed530b6
YC
954/* TCP early-retransmit (ER) is similar to but more conservative than
955 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
956 */
957static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
958{
1043e25f
NB
959 struct net *net = sock_net((struct sock *)tp);
960
eed530b6 961 tp->do_early_retrans = sysctl_tcp_early_retrans &&
6ba8a3b1 962 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
1043e25f 963 net->ipv4.sysctl_tcp_reordering == 3;
eed530b6
YC
964}
965
966static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
967{
968 tp->do_early_retrans = 0;
969}
970
83ae4088
IJ
971static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
972{
973 return tp->sacked_out + tp->lost_out;
974}
975
1da177e4
LT
976/* This determines how many packets are "in the network" to the best
977 * of our knowledge. In many cases it is conservative, but where
978 * detailed information is available from the receiver (via SACK
979 * blocks etc.) we can make more aggressive calculations.
980 *
981 * Use this for decisions involving congestion control, use just
982 * tp->packets_out to determine if the send queue is empty or not.
983 *
984 * Read this equation as:
985 *
986 * "Packets sent once on transmission queue" MINUS
987 * "Packets left network, but not honestly ACKed yet" PLUS
988 * "Packets fast retransmitted"
989 */
40efc6fa 990static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1da177e4 991{
83ae4088 992 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1da177e4
LT
993}
994
0b6a05c1
IJ
995#define TCP_INFINITE_SSTHRESH 0x7fffffff
996
071d5080
YC
997static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
998{
76174004 999 return tp->snd_cwnd < tp->snd_ssthresh;
071d5080
YC
1000}
1001
0b6a05c1
IJ
1002static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1003{
1004 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1005}
1006
684bad11
YC
1007static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1008{
1009 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1010 (1 << inet_csk(sk)->icsk_ca_state);
1011}
1012
1da177e4 1013/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
684bad11 1014 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1da177e4
LT
1015 * ssthresh.
1016 */
6687e988 1017static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1da177e4 1018{
6687e988 1019 const struct tcp_sock *tp = tcp_sk(sk);
cf533ea5 1020
684bad11 1021 if (tcp_in_cwnd_reduction(sk))
1da177e4
LT
1022 return tp->snd_ssthresh;
1023 else
1024 return max(tp->snd_ssthresh,
1025 ((tp->snd_cwnd >> 1) +
1026 (tp->snd_cwnd >> 2)));
1027}
1028
b9c4595b
IJ
1029/* Use define here intentionally to get WARN_ON location shown at the caller */
1030#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1da177e4 1031
5ee2c941 1032void tcp_enter_cwr(struct sock *sk);
5c9f3023 1033__u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1da177e4 1034
6b5a5c0d
NC
1035/* The maximum number of MSS of available cwnd for which TSO defers
1036 * sending if not using sysctl_tcp_tso_win_divisor.
1037 */
1038static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1039{
1040 return 3;
1041}
1042
90840def
IJ
1043/* Returns end sequence number of the receiver's advertised window */
1044static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1045{
1046 return tp->snd_una + tp->snd_wnd;
1047}
e114a710
ED
1048
1049/* We follow the spirit of RFC2861 to validate cwnd but implement a more
1050 * flexible approach. The RFC suggests cwnd should not be raised unless
ca8a2263
NC
1051 * it was fully used previously. And that's exactly what we do in
1052 * congestion avoidance mode. But in slow start we allow cwnd to grow
1053 * as long as the application has used half the cwnd.
e114a710
ED
1054 * Example :
1055 * cwnd is 10 (IW10), but application sends 9 frames.
1056 * We allow cwnd to reach 18 when all frames are ACKed.
1057 * This check is safe because it's as aggressive as slow start which already
1058 * risks 100% overshoot. The advantage is that we discourage application to
1059 * either send more filler packets or data to artificially blow up the cwnd
1060 * usage, and allow application-limited process to probe bw more aggressively.
e114a710 1061 */
24901551 1062static inline bool tcp_is_cwnd_limited(const struct sock *sk)
e114a710
ED
1063{
1064 const struct tcp_sock *tp = tcp_sk(sk);
1065
ca8a2263 1066 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
071d5080 1067 if (tcp_in_slow_start(tp))
ca8a2263
NC
1068 return tp->snd_cwnd < 2 * tp->max_packets_out;
1069
1070 return tp->is_cwnd_limited;
e114a710 1071}
f4805ede 1072
21c8fe99
ED
1073/* Something is really bad, we could not queue an additional packet,
1074 * because qdisc is full or receiver sent a 0 window.
1075 * We do not want to add fuel to the fire, or abort too early,
1076 * so make sure the timer we arm now is at least 200ms in the future,
1077 * regardless of current icsk_rto value (as it could be ~2ms)
1078 */
1079static inline unsigned long tcp_probe0_base(const struct sock *sk)
1da177e4 1080{
21c8fe99
ED
1081 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1082}
9e412ba7 1083
21c8fe99
ED
1084/* Variant of inet_csk_rto_backoff() used for zero window probes */
1085static inline unsigned long tcp_probe0_when(const struct sock *sk,
1086 unsigned long max_when)
1087{
1088 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1089
1090 return (unsigned long)min_t(u64, when, max_when);
1091}
1092
1093static inline void tcp_check_probe_timer(struct sock *sk)
1094{
1095 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
3f421baa 1096 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
21c8fe99 1097 tcp_probe0_base(sk), TCP_RTO_MAX);
1da177e4
LT
1098}
1099
ee7537b6 1100static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1da177e4
LT
1101{
1102 tp->snd_wl1 = seq;
1103}
1104
ee7537b6 1105static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1da177e4
LT
1106{
1107 tp->snd_wl1 = seq;
1108}
1109
1da177e4
LT
1110/*
1111 * Calculate(/check) TCP checksum
1112 */
ba7808ea
FD
1113static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1114 __be32 daddr, __wsum base)
1da177e4
LT
1115{
1116 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1117}
1118
b51655b9 1119static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1da177e4 1120{
fb286bb2 1121 return __skb_checksum_complete(skb);
1da177e4
LT
1122}
1123
a2a385d6 1124static inline bool tcp_checksum_complete(struct sk_buff *skb)
1da177e4 1125{
60476372 1126 return !skb_csum_unnecessary(skb) &&
1da177e4
LT
1127 __tcp_checksum_complete(skb);
1128}
1129
1130/* Prequeue for VJ style copy to user, combined with checksumming. */
1131
40efc6fa 1132static inline void tcp_prequeue_init(struct tcp_sock *tp)
1da177e4
LT
1133{
1134 tp->ucopy.task = NULL;
1135 tp->ucopy.len = 0;
1136 tp->ucopy.memory = 0;
1137 skb_queue_head_init(&tp->ucopy.prequeue);
1138}
1139
5c9f3023 1140bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1da177e4
LT
1141
1142#undef STATE_TRACE
1143
1144#ifdef STATE_TRACE
1145static const char *statename[]={
1146 "Unused","Established","Syn Sent","Syn Recv",
1147 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1148 "Close Wait","Last ACK","Listen","Closing"
1149};
1150#endif
5c9f3023 1151void tcp_set_state(struct sock *sk, int state);
1da177e4 1152
5c9f3023 1153void tcp_done(struct sock *sk);
1da177e4 1154
c1e64e29
LC
1155int tcp_abort(struct sock *sk, int err);
1156
40efc6fa 1157static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1da177e4
LT
1158{
1159 rx_opt->dsack = 0;
1da177e4
LT
1160 rx_opt->num_sacks = 0;
1161}
1162
5c9f3023 1163u32 tcp_default_init_rwnd(u32 mss);
6f021c62
ED
1164void tcp_cwnd_restart(struct sock *sk, s32 delta);
1165
1166static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1167{
1168 struct tcp_sock *tp = tcp_sk(sk);
1169 s32 delta;
1170
1171 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1172 return;
1173 delta = tcp_time_stamp - tp->lsndtime;
1174 if (delta > inet_csk(sk)->icsk_rto)
1175 tcp_cwnd_restart(sk, delta);
1176}
85f16525 1177
1da177e4 1178/* Determine a window scaling and initial window to offer. */
5c9f3023
JP
1179void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1180 __u32 *window_clamp, int wscale_ok,
1181 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1da177e4
LT
1182
1183static inline int tcp_win_from_space(int space)
1184{
1185 return sysctl_tcp_adv_win_scale<=0 ?
1186 (space>>(-sysctl_tcp_adv_win_scale)) :
1187 space - (space>>sysctl_tcp_adv_win_scale);
1188}
1189
105970f6 1190/* Note: caller must be prepared to deal with negative returns */
1da177e4
LT
1191static inline int tcp_space(const struct sock *sk)
1192{
1193 return tcp_win_from_space(sk->sk_rcvbuf -
1194 atomic_read(&sk->sk_rmem_alloc));
105970f6 1195}
1da177e4
LT
1196
1197static inline int tcp_full_space(const struct sock *sk)
1198{
105970f6 1199 return tcp_win_from_space(sk->sk_rcvbuf);
1da177e4
LT
1200}
1201
843f4a55 1202extern void tcp_openreq_init_rwin(struct request_sock *req,
b1964b5f
ED
1203 const struct sock *sk_listener,
1204 const struct dst_entry *dst);
843f4a55 1205
5c9f3023 1206void tcp_enter_memory_pressure(struct sock *sk);
1da177e4 1207
1da177e4
LT
1208static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1209{
b840d15d
NB
1210 struct net *net = sock_net((struct sock *)tp);
1211
1212 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1da177e4
LT
1213}
1214
1215static inline int keepalive_time_when(const struct tcp_sock *tp)
1216{
13b287e8
NB
1217 struct net *net = sock_net((struct sock *)tp);
1218
1219 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1da177e4
LT
1220}
1221
df19a626
ED
1222static inline int keepalive_probes(const struct tcp_sock *tp)
1223{
9bd6861b
NB
1224 struct net *net = sock_net((struct sock *)tp);
1225
1226 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
df19a626
ED
1227}
1228
6c37e5de
FL
1229static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1230{
1231 const struct inet_connection_sock *icsk = &tp->inet_conn;
1232
1233 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1234 tcp_time_stamp - tp->rcv_tstamp);
1235}
1236
463c84b9 1237static inline int tcp_fin_time(const struct sock *sk)
1da177e4 1238{
1e579caa 1239 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
463c84b9 1240 const int rto = inet_csk(sk)->icsk_rto;
1da177e4 1241
463c84b9
ACM
1242 if (fin_timeout < (rto << 2) - (rto >> 1))
1243 fin_timeout = (rto << 2) - (rto >> 1);
1da177e4
LT
1244
1245 return fin_timeout;
1246}
1247
a2a385d6
ED
1248static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1249 int paws_win)
1da177e4 1250{
c887e6d2 1251 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
a2a385d6 1252 return true;
c887e6d2 1253 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
a2a385d6 1254 return true;
bc2ce894
ED
1255 /*
1256 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1257 * then following tcp messages have valid values. Ignore 0 value,
1258 * or else 'negative' tsval might forbid us to accept their packets.
1259 */
1260 if (!rx_opt->ts_recent)
a2a385d6
ED
1261 return true;
1262 return false;
c887e6d2
IJ
1263}
1264
a2a385d6
ED
1265static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1266 int rst)
c887e6d2
IJ
1267{
1268 if (tcp_paws_check(rx_opt, 0))
a2a385d6 1269 return false;
1da177e4
LT
1270
1271 /* RST segments are not recommended to carry timestamp,
1272 and, if they do, it is recommended to ignore PAWS because
1273 "their cleanup function should take precedence over timestamps."
1274 Certainly, it is mistake. It is necessary to understand the reasons
1275 of this constraint to relax it: if peer reboots, clock may go
1276 out-of-sync and half-open connections will not be reset.
1277 Actually, the problem would be not existing if all
1278 the implementations followed draft about maintaining clock
1279 via reboots. Linux-2.2 DOES NOT!
1280
1281 However, we can relax time bounds for RST segments to MSL.
1282 */
9d729f72 1283 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
a2a385d6
ED
1284 return false;
1285 return true;
1da177e4
LT
1286}
1287
7970ddc8
ED
1288bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1289 int mib_idx, u32 *last_oow_ack_time);
032ee423 1290
a9c19329 1291static inline void tcp_mib_init(struct net *net)
1da177e4
LT
1292{
1293 /* See RFC 2012 */
cf1100a7
PE
1294 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1295 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1296 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1297 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1da177e4
LT
1298}
1299
5af4ec23 1300/* from STCP */
ef9da47c 1301static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
0800f170 1302{
6a438bbe 1303 tp->lost_skb_hint = NULL;
ef9da47c
IJ
1304}
1305
1306static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1307{
1308 tcp_clear_retrans_hints_partial(tp);
6a438bbe 1309 tp->retransmit_skb_hint = NULL;
b7689205
IJ
1310}
1311
a915da9b
ED
1312union tcp_md5_addr {
1313 struct in_addr a4;
1314#if IS_ENABLED(CONFIG_IPV6)
1315 struct in6_addr a6;
1316#endif
1317};
1318
cfb6eeb4
YH
1319/* - key database */
1320struct tcp_md5sig_key {
a915da9b 1321 struct hlist_node node;
cfb6eeb4 1322 u8 keylen;
a915da9b
ED
1323 u8 family; /* AF_INET or AF_INET6 */
1324 union tcp_md5_addr addr;
1325 u8 key[TCP_MD5SIG_MAXKEYLEN];
1326 struct rcu_head rcu;
cfb6eeb4
YH
1327};
1328
1329/* - sock block */
1330struct tcp_md5sig_info {
a915da9b 1331 struct hlist_head head;
a8afca03 1332 struct rcu_head rcu;
cfb6eeb4
YH
1333};
1334
1335/* - pseudo header */
1336struct tcp4_pseudohdr {
1337 __be32 saddr;
1338 __be32 daddr;
1339 __u8 pad;
1340 __u8 protocol;
1341 __be16 len;
1342};
1343
1344struct tcp6_pseudohdr {
1345 struct in6_addr saddr;
1346 struct in6_addr daddr;
1347 __be32 len;
1348 __be32 protocol; /* including padding */
1349};
1350
1351union tcp_md5sum_block {
1352 struct tcp4_pseudohdr ip4;
dfd56b8b 1353#if IS_ENABLED(CONFIG_IPV6)
cfb6eeb4
YH
1354 struct tcp6_pseudohdr ip6;
1355#endif
1356};
1357
1358/* - pool: digest algorithm, hash description and scratch buffer */
1359struct tcp_md5sig_pool {
cf80e0e4 1360 struct ahash_request *md5_req;
cfb6eeb4
YH
1361 union tcp_md5sum_block md5_blk;
1362};
1363
cfb6eeb4 1364/* - functions */
39f8e58e
ED
1365int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1366 const struct sock *sk, const struct sk_buff *skb);
5c9f3023
JP
1367int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1368 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1369int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1370 int family);
b83e3deb 1371struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
fd3a154a 1372 const struct sock *addr_sk);
cfb6eeb4 1373
9501f972 1374#ifdef CONFIG_TCP_MD5SIG
b83e3deb 1375struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
5c9f3023
JP
1376 const union tcp_md5_addr *addr,
1377 int family);
a915da9b 1378#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
9501f972 1379#else
b83e3deb 1380static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
a915da9b
ED
1381 const union tcp_md5_addr *addr,
1382 int family)
1383{
1384 return NULL;
1385}
9501f972
YH
1386#define tcp_twsk_md5_key(twsk) NULL
1387#endif
1388
5c9f3023 1389bool tcp_alloc_md5sig_pool(void);
cfb6eeb4 1390
5c9f3023 1391struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
71cea17e
ED
1392static inline void tcp_put_md5sig_pool(void)
1393{
1394 local_bh_enable();
1395}
35790c04 1396
5c9f3023
JP
1397int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1398int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1399 unsigned int header_len);
1400int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1401 const struct tcp_md5sig_key *key);
cfb6eeb4 1402
10467163 1403/* From tcp_fastopen.c */
5c9f3023
JP
1404void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1405 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1406 unsigned long *last_syn_loss);
1407void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
2646c831
DL
1408 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1409 u16 try_exp);
783237e8
YC
1410struct tcp_fastopen_request {
1411 /* Fast Open cookie. Size 0 means a cookie request */
1412 struct tcp_fastopen_cookie cookie;
1413 struct msghdr *data; /* data in MSG_FASTOPEN */
f5ddcbbb
ED
1414 size_t size;
1415 int copied; /* queued in tcp_connect() */
783237e8 1416};
783237e8
YC
1417void tcp_free_fastopen_req(struct tcp_sock *tp);
1418
10467163
JC
1419extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1420int tcp_fastopen_reset_cipher(void *key, unsigned int len);
61d2bcae 1421void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
7c85af88
ED
1422struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1423 struct request_sock *req,
1424 struct tcp_fastopen_cookie *foc,
1425 struct dst_entry *dst);
222e83d2 1426void tcp_fastopen_init_key_once(bool publish);
10467163
JC
1427#define TCP_FASTOPEN_KEY_LENGTH 16
1428
1429/* Fastopen key context */
1430struct tcp_fastopen_context {
7ae8639c
ED
1431 struct crypto_cipher *tfm;
1432 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1433 struct rcu_head rcu;
10467163
JC
1434};
1435
fe067e8a
DM
1436/* write queue abstraction */
1437static inline void tcp_write_queue_purge(struct sock *sk)
1438{
1439 struct sk_buff *skb;
1440
1441 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
3ab224be
HA
1442 sk_wmem_free_skb(sk, skb);
1443 sk_mem_reclaim(sk);
8818a9d8 1444 tcp_clear_all_retrans_hints(tcp_sk(sk));
fe067e8a
DM
1445}
1446
cf533ea5 1447static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
fe067e8a 1448{
cd07a8ea 1449 return skb_peek(&sk->sk_write_queue);
fe067e8a
DM
1450}
1451
cf533ea5 1452static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
fe067e8a 1453{
cd07a8ea 1454 return skb_peek_tail(&sk->sk_write_queue);
fe067e8a
DM
1455}
1456
cf533ea5
ED
1457static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1458 const struct sk_buff *skb)
fe067e8a 1459{
cd07a8ea 1460 return skb_queue_next(&sk->sk_write_queue, skb);
fe067e8a
DM
1461}
1462
cf533ea5
ED
1463static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1464 const struct sk_buff *skb)
832d11c5
IJ
1465{
1466 return skb_queue_prev(&sk->sk_write_queue, skb);
1467}
1468
fe067e8a 1469#define tcp_for_write_queue(skb, sk) \
cd07a8ea 1470 skb_queue_walk(&(sk)->sk_write_queue, skb)
fe067e8a
DM
1471
1472#define tcp_for_write_queue_from(skb, sk) \
cd07a8ea 1473 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
fe067e8a 1474
234b6860 1475#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
cd07a8ea 1476 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
234b6860 1477
cf533ea5 1478static inline struct sk_buff *tcp_send_head(const struct sock *sk)
fe067e8a
DM
1479{
1480 return sk->sk_send_head;
1481}
1482
cd07a8ea
DM
1483static inline bool tcp_skb_is_last(const struct sock *sk,
1484 const struct sk_buff *skb)
1485{
1486 return skb_queue_is_last(&sk->sk_write_queue, skb);
1487}
1488
cf533ea5 1489static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
fe067e8a 1490{
cd07a8ea 1491 if (tcp_skb_is_last(sk, skb))
fe067e8a 1492 sk->sk_send_head = NULL;
cd07a8ea
DM
1493 else
1494 sk->sk_send_head = tcp_write_queue_next(sk, skb);
fe067e8a
DM
1495}
1496
1497static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1498{
1499 if (sk->sk_send_head == skb_unlinked)
1500 sk->sk_send_head = NULL;
1501}
1502
1503static inline void tcp_init_send_head(struct sock *sk)
1504{
1505 sk->sk_send_head = NULL;
1506}
1507
1508static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1509{
1510 __skb_queue_tail(&sk->sk_write_queue, skb);
1511}
1512
1513static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1514{
1515 __tcp_add_write_queue_tail(sk, skb);
1516
1517 /* Queue it, remembering where we must start sending. */
6859d494 1518 if (sk->sk_send_head == NULL) {
fe067e8a 1519 sk->sk_send_head = skb;
6859d494
IJ
1520
1521 if (tcp_sk(sk)->highest_sack == NULL)
1522 tcp_sk(sk)->highest_sack = skb;
1523 }
fe067e8a
DM
1524}
1525
1526static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1527{
1528 __skb_queue_head(&sk->sk_write_queue, skb);
1529}
1530
1531/* Insert buff after skb on the write queue of sk. */
1532static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1533 struct sk_buff *buff,
1534 struct sock *sk)
1535{
7de6c033 1536 __skb_queue_after(&sk->sk_write_queue, skb, buff);
fe067e8a
DM
1537}
1538
43f59c89 1539/* Insert new before skb on the write queue of sk. */
fe067e8a
DM
1540static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1541 struct sk_buff *skb,
1542 struct sock *sk)
1543{
43f59c89 1544 __skb_queue_before(&sk->sk_write_queue, skb, new);