]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * Definitions for the TCP module. | |
7 | * | |
8 | * Version: @(#)tcp.h 1.0.5 05/23/93 | |
9 | * | |
10 | * Authors: Ross Biro | |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> | |
12 | * | |
13 | * This program is free software; you can redistribute it and/or | |
14 | * modify it under the terms of the GNU General Public License | |
15 | * as published by the Free Software Foundation; either version | |
16 | * 2 of the License, or (at your option) any later version. | |
17 | */ | |
18 | #ifndef _TCP_H | |
19 | #define _TCP_H | |
20 | ||
21 | #define FASTRETRANS_DEBUG 1 | |
22 | ||
23 | #include <linux/list.h> | |
24 | #include <linux/tcp.h> | |
25 | #include <linux/bug.h> | |
26 | #include <linux/slab.h> | |
27 | #include <linux/cache.h> | |
28 | #include <linux/percpu.h> | |
29 | #include <linux/skbuff.h> | |
30 | #include <linux/cryptohash.h> | |
31 | #include <linux/kref.h> | |
32 | #include <linux/ktime.h> | |
33 | ||
34 | #include <net/inet_connection_sock.h> | |
35 | #include <net/inet_timewait_sock.h> | |
36 | #include <net/inet_hashtables.h> | |
37 | #include <net/checksum.h> | |
38 | #include <net/request_sock.h> | |
39 | #include <net/sock.h> | |
40 | #include <net/snmp.h> | |
41 | #include <net/ip.h> | |
42 | #include <net/tcp_states.h> | |
43 | #include <net/inet_ecn.h> | |
44 | #include <net/dst.h> | |
45 | ||
46 | #include <linux/seq_file.h> | |
47 | #include <linux/memcontrol.h> | |
48 | ||
49 | extern struct inet_hashinfo tcp_hashinfo; | |
50 | ||
51 | extern struct percpu_counter tcp_orphan_count; | |
52 | 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 1024 | |
68 | ||
69 | /* probing interval, default to 10 minutes as per RFC4821 */ | |
70 | #define TCP_PROBE_INTERVAL 600 | |
71 | ||
72 | /* Specify interval when tcp mtu probing will stop */ | |
73 | #define TCP_PROBE_THRESHOLD 8 | |
74 | ||
75 | /* After receiving this amount of duplicate ACKs fast retransmit starts. */ | |
76 | #define TCP_FASTRETRANS_THRESH 3 | |
77 | ||
78 | /* Maximal number of ACKs sent quickly to accelerate slow-start. */ | |
79 | #define TCP_MAX_QUICKACKS 16U | |
80 | ||
81 | /* urg_data states */ | |
82 | #define TCP_URG_VALID 0x0100 | |
83 | #define TCP_URG_NOTYET 0x0200 | |
84 | #define TCP_URG_READ 0x0400 | |
85 | ||
86 | #define TCP_RETR1 3 /* | |
87 | * This is how many retries it does before it | |
88 | * tries to figure out if the gateway is | |
89 | * down. Minimal RFC value is 3; it corresponds | |
90 | * to ~3sec-8min depending on RTO. | |
91 | */ | |
92 | ||
93 | #define TCP_RETR2 15 /* | |
94 | * This should take at least | |
95 | * 90 minutes to time out. | |
96 | * RFC1122 says that the limit is 100 sec. | |
97 | * 15 is ~13-30min depending on RTO. | |
98 | */ | |
99 | ||
100 | #define TCP_SYN_RETRIES 6 /* This is how many retries are done | |
101 | * when active opening a connection. | |
102 | * RFC1122 says the minimum retry MUST | |
103 | * be at least 180secs. Nevertheless | |
104 | * this value is corresponding to | |
105 | * 63secs of retransmission with the | |
106 | * current initial RTO. | |
107 | */ | |
108 | ||
109 | #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done | |
110 | * when passive opening a connection. | |
111 | * This is corresponding to 31secs of | |
112 | * retransmission with the current | |
113 | * initial RTO. | |
114 | */ | |
115 | ||
116 | #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT | |
117 | * state, about 60 seconds */ | |
118 | #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN | |
119 | /* BSD style FIN_WAIT2 deadlock breaker. | |
120 | * It used to be 3min, new value is 60sec, | |
121 | * to combine FIN-WAIT-2 timeout with | |
122 | * TIME-WAIT timer. | |
123 | */ | |
124 | ||
125 | #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ | |
126 | #if HZ >= 100 | |
127 | #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ | |
128 | #define TCP_ATO_MIN ((unsigned)(HZ/25)) | |
129 | #else | |
130 | #define TCP_DELACK_MIN 4U | |
131 | #define TCP_ATO_MIN 4U | |
132 | #endif | |
133 | #define TCP_RTO_MAX ((unsigned)(120*HZ)) | |
134 | #define TCP_RTO_MIN ((unsigned)(HZ/5)) | |
135 | #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ | |
136 | #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now | |
137 | * used as a fallback RTO for the | |
138 | * initial data transmission if no | |
139 | * valid RTT sample has been acquired, | |
140 | * most likely due to retrans in 3WHS. | |
141 | */ | |
142 | ||
143 | #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes | |
144 | * for local resources. | |
145 | */ | |
146 | ||
147 | #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ | |
148 | #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ | |
149 | #define TCP_KEEPALIVE_INTVL (75*HZ) | |
150 | ||
151 | #define MAX_TCP_KEEPIDLE 32767 | |
152 | #define MAX_TCP_KEEPINTVL 32767 | |
153 | #define MAX_TCP_KEEPCNT 127 | |
154 | #define MAX_TCP_SYNCNT 127 | |
155 | ||
156 | #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ | |
157 | ||
158 | #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) | |
159 | #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated | |
160 | * after this time. It should be equal | |
161 | * (or greater than) TCP_TIMEWAIT_LEN | |
162 | * to provide reliability equal to one | |
163 | * provided by timewait state. | |
164 | */ | |
165 | #define TCP_PAWS_WINDOW 1 /* Replay window for per-host | |
166 | * timestamps. It must be less than | |
167 | * minimal timewait lifetime. | |
168 | */ | |
169 | /* | |
170 | * TCP option | |
171 | */ | |
172 | ||
173 | #define TCPOPT_NOP 1 /* Padding */ | |
174 | #define TCPOPT_EOL 0 /* End of options */ | |
175 | #define TCPOPT_MSS 2 /* Segment size negotiating */ | |
176 | #define TCPOPT_WINDOW 3 /* Window scaling */ | |
177 | #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ | |
178 | #define TCPOPT_SACK 5 /* SACK Block */ | |
179 | #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ | |
180 | #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ | |
181 | #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ | |
182 | #define TCPOPT_EXP 254 /* Experimental */ | |
183 | /* Magic number to be after the option value for sharing TCP | |
184 | * experimental options. See draft-ietf-tcpm-experimental-options-00.txt | |
185 | */ | |
186 | #define TCPOPT_FASTOPEN_MAGIC 0xF989 | |
187 | ||
188 | /* | |
189 | * TCP option lengths | |
190 | */ | |
191 | ||
192 | #define TCPOLEN_MSS 4 | |
193 | #define TCPOLEN_WINDOW 3 | |
194 | #define TCPOLEN_SACK_PERM 2 | |
195 | #define TCPOLEN_TIMESTAMP 10 | |
196 | #define TCPOLEN_MD5SIG 18 | |
197 | #define TCPOLEN_FASTOPEN_BASE 2 | |
198 | #define TCPOLEN_EXP_FASTOPEN_BASE 4 | |
199 | ||
200 | /* But this is what stacks really send out. */ | |
201 | #define TCPOLEN_TSTAMP_ALIGNED 12 | |
202 | #define TCPOLEN_WSCALE_ALIGNED 4 | |
203 | #define TCPOLEN_SACKPERM_ALIGNED 4 | |
204 | #define TCPOLEN_SACK_BASE 2 | |
205 | #define TCPOLEN_SACK_BASE_ALIGNED 4 | |
206 | #define TCPOLEN_SACK_PERBLOCK 8 | |
207 | #define TCPOLEN_MD5SIG_ALIGNED 20 | |
208 | #define TCPOLEN_MSS_ALIGNED 4 | |
209 | ||
210 | /* Flags in tp->nonagle */ | |
211 | #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ | |
212 | #define TCP_NAGLE_CORK 2 /* Socket is corked */ | |
213 | #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ | |
214 | ||
215 | /* TCP thin-stream limits */ | |
216 | #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ | |
217 | ||
218 | /* TCP initial congestion window as per rfc6928 */ | |
219 | #define TCP_INIT_CWND 10 | |
220 | ||
221 | /* Bit Flags for sysctl_tcp_fastopen */ | |
222 | #define TFO_CLIENT_ENABLE 1 | |
223 | #define TFO_SERVER_ENABLE 2 | |
224 | #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ | |
225 | ||
226 | /* Accept SYN data w/o any cookie option */ | |
227 | #define TFO_SERVER_COOKIE_NOT_REQD 0x200 | |
228 | ||
229 | /* Force enable TFO on all listeners, i.e., not requiring the | |
230 | * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen. | |
231 | */ | |
232 | #define TFO_SERVER_WO_SOCKOPT1 0x400 | |
233 | #define TFO_SERVER_WO_SOCKOPT2 0x800 | |
234 | ||
235 | extern struct inet_timewait_death_row tcp_death_row; | |
236 | ||
237 | /* sysctl variables for tcp */ | |
238 | extern int sysctl_tcp_timestamps; | |
239 | extern int sysctl_tcp_window_scaling; | |
240 | extern int sysctl_tcp_sack; | |
241 | extern int sysctl_tcp_fastopen; | |
242 | extern int sysctl_tcp_retrans_collapse; | |
243 | extern int sysctl_tcp_stdurg; | |
244 | extern int sysctl_tcp_rfc1337; | |
245 | extern int sysctl_tcp_abort_on_overflow; | |
246 | extern int sysctl_tcp_max_orphans; | |
247 | extern int sysctl_tcp_fack; | |
248 | extern int sysctl_tcp_reordering; | |
249 | extern int sysctl_tcp_max_reordering; | |
250 | extern int sysctl_tcp_dsack; | |
251 | extern long sysctl_tcp_mem[3]; | |
252 | extern int sysctl_tcp_wmem[3]; | |
253 | extern int sysctl_tcp_rmem[3]; | |
254 | extern int sysctl_tcp_app_win; | |
255 | extern int sysctl_tcp_adv_win_scale; | |
256 | extern int sysctl_tcp_tw_reuse; | |
257 | extern int sysctl_tcp_frto; | |
258 | extern int sysctl_tcp_low_latency; | |
259 | extern int sysctl_tcp_nometrics_save; | |
260 | extern int sysctl_tcp_moderate_rcvbuf; | |
261 | extern int sysctl_tcp_tso_win_divisor; | |
262 | extern int sysctl_tcp_workaround_signed_windows; | |
263 | extern int sysctl_tcp_slow_start_after_idle; | |
264 | extern int sysctl_tcp_thin_linear_timeouts; | |
265 | extern int sysctl_tcp_thin_dupack; | |
266 | extern int sysctl_tcp_early_retrans; | |
267 | extern int sysctl_tcp_limit_output_bytes; | |
268 | extern int sysctl_tcp_challenge_ack_limit; | |
269 | extern int sysctl_tcp_min_tso_segs; | |
270 | extern int sysctl_tcp_min_rtt_wlen; | |
271 | extern int sysctl_tcp_autocorking; | |
272 | extern int sysctl_tcp_invalid_ratelimit; | |
273 | extern int sysctl_tcp_pacing_ss_ratio; | |
274 | extern int sysctl_tcp_pacing_ca_ratio; | |
275 | ||
276 | extern atomic_long_t tcp_memory_allocated; | |
277 | extern struct percpu_counter tcp_sockets_allocated; | |
278 | extern int tcp_memory_pressure; | |
279 | ||
280 | /* optimized version of sk_under_memory_pressure() for TCP sockets */ | |
281 | static inline bool tcp_under_memory_pressure(const struct sock *sk) | |
282 | { | |
283 | if (mem_cgroup_sockets_enabled && sk->sk_memcg && | |
284 | mem_cgroup_under_socket_pressure(sk->sk_memcg)) | |
285 | return true; | |
286 | ||
287 | return tcp_memory_pressure; | |
288 | } | |
289 | /* | |
290 | * The next routines deal with comparing 32 bit unsigned ints | |
291 | * and worry about wraparound (automatic with unsigned arithmetic). | |
292 | */ | |
293 | ||
294 | static inline bool before(__u32 seq1, __u32 seq2) | |
295 | { | |
296 | return (__s32)(seq1-seq2) < 0; | |
297 | } | |
298 | #define after(seq2, seq1) before(seq1, seq2) | |
299 | ||
300 | /* is s2<=s1<=s3 ? */ | |
301 | static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) | |
302 | { | |
303 | return seq3 - seq2 >= seq1 - seq2; | |
304 | } | |
305 | ||
306 | static 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 | ||
314 | void sk_forced_mem_schedule(struct sock *sk, int size); | |
315 | ||
316 | static inline bool tcp_too_many_orphans(struct sock *sk, int shift) | |
317 | { | |
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 | } | |
326 | return false; | |
327 | } | |
328 | ||
329 | bool tcp_check_oom(struct sock *sk, int shift); | |
330 | ||
331 | ||
332 | extern struct proto tcp_prot; | |
333 | ||
334 | #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) | |
335 | #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((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(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) | |
338 | ||
339 | void tcp_tasklet_init(void); | |
340 | ||
341 | void tcp_v4_err(struct sk_buff *skb, u32); | |
342 | ||
343 | void tcp_shutdown(struct sock *sk, int how); | |
344 | ||
345 | void tcp_v4_early_demux(struct sk_buff *skb); | |
346 | int tcp_v4_rcv(struct sk_buff *skb); | |
347 | ||
348 | int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); | |
349 | int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); | |
350 | int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, | |
351 | int flags); | |
352 | void tcp_release_cb(struct sock *sk); | |
353 | void tcp_wfree(struct sk_buff *skb); | |
354 | void tcp_write_timer_handler(struct sock *sk); | |
355 | void tcp_delack_timer_handler(struct sock *sk); | |
356 | int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); | |
357 | int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); | |
358 | void tcp_rcv_established(struct sock *sk, struct sk_buff *skb, | |
359 | const struct tcphdr *th, unsigned int len); | |
360 | void tcp_rcv_space_adjust(struct sock *sk); | |
361 | int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); | |
362 | void tcp_twsk_destructor(struct sock *sk); | |
363 | ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, | |
364 | struct pipe_inode_info *pipe, size_t len, | |
365 | unsigned int flags); | |
366 | ||
367 | static inline void tcp_dec_quickack_mode(struct sock *sk, | |
368 | const unsigned int pkts) | |
369 | { | |
370 | struct inet_connection_sock *icsk = inet_csk(sk); | |
371 | ||
372 | if (icsk->icsk_ack.quick) { | |
373 | if (pkts >= icsk->icsk_ack.quick) { | |
374 | icsk->icsk_ack.quick = 0; | |
375 | /* Leaving quickack mode we deflate ATO. */ | |
376 | icsk->icsk_ack.ato = TCP_ATO_MIN; | |
377 | } else | |
378 | icsk->icsk_ack.quick -= pkts; | |
379 | } | |
380 | } | |
381 | ||
382 | #define TCP_ECN_OK 1 | |
383 | #define TCP_ECN_QUEUE_CWR 2 | |
384 | #define TCP_ECN_DEMAND_CWR 4 | |
385 | #define TCP_ECN_SEEN 8 | |
386 | ||
387 | enum tcp_tw_status { | |
388 | TCP_TW_SUCCESS = 0, | |
389 | TCP_TW_RST = 1, | |
390 | TCP_TW_ACK = 2, | |
391 | TCP_TW_SYN = 3 | |
392 | }; | |
393 | ||
394 | ||
395 | enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, | |
396 | struct sk_buff *skb, | |
397 | const struct tcphdr *th); | |
398 | struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, | |
399 | struct request_sock *req, bool fastopen); | |
400 | int tcp_child_process(struct sock *parent, struct sock *child, | |
401 | struct sk_buff *skb); | |
402 | void tcp_enter_loss(struct sock *sk); | |
403 | void tcp_clear_retrans(struct tcp_sock *tp); | |
404 | void tcp_update_metrics(struct sock *sk); | |
405 | void tcp_init_metrics(struct sock *sk); | |
406 | void tcp_metrics_init(void); | |
407 | bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, | |
408 | bool paws_check, bool timestamps); | |
409 | bool tcp_remember_stamp(struct sock *sk); | |
410 | bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw); | |
411 | void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst); | |
412 | void tcp_disable_fack(struct tcp_sock *tp); | |
413 | void tcp_close(struct sock *sk, long timeout); | |
414 | void tcp_init_sock(struct sock *sk); | |
415 | unsigned int tcp_poll(struct file *file, struct socket *sock, | |
416 | struct poll_table_struct *wait); | |
417 | int tcp_getsockopt(struct sock *sk, int level, int optname, | |
418 | char __user *optval, int __user *optlen); | |
419 | int tcp_setsockopt(struct sock *sk, int level, int optname, | |
420 | char __user *optval, unsigned int optlen); | |
421 | int compat_tcp_getsockopt(struct sock *sk, int level, int optname, | |
422 | char __user *optval, int __user *optlen); | |
423 | int compat_tcp_setsockopt(struct sock *sk, int level, int optname, | |
424 | char __user *optval, unsigned int optlen); | |
425 | void tcp_set_keepalive(struct sock *sk, int val); | |
426 | void tcp_syn_ack_timeout(const struct request_sock *req); | |
427 | int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, | |
428 | int flags, int *addr_len); | |
429 | void tcp_parse_options(const struct sk_buff *skb, | |
430 | struct tcp_options_received *opt_rx, | |
431 | int estab, struct tcp_fastopen_cookie *foc); | |
432 | const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); | |
433 | ||
434 | /* | |
435 | * TCP v4 functions exported for the inet6 API | |
436 | */ | |
437 | ||
438 | void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); | |
439 | void tcp_v4_mtu_reduced(struct sock *sk); | |
440 | void tcp_req_err(struct sock *sk, u32 seq, bool abort); | |
441 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); | |
442 | struct sock *tcp_create_openreq_child(const struct sock *sk, | |
443 | struct request_sock *req, | |
444 | struct sk_buff *skb); | |
445 | void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); | |
446 | struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, | |
447 | struct request_sock *req, | |
448 | struct dst_entry *dst, | |
449 | struct request_sock *req_unhash, | |
450 | bool *own_req); | |
451 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); | |
452 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); | |
453 | int tcp_connect(struct sock *sk); | |
454 | enum tcp_synack_type { | |
455 | TCP_SYNACK_NORMAL, | |
456 | TCP_SYNACK_FASTOPEN, | |
457 | TCP_SYNACK_COOKIE, | |
458 | }; | |
459 | struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, | |
460 | struct request_sock *req, | |
461 | struct tcp_fastopen_cookie *foc, | |
462 | enum tcp_synack_type synack_type); | |
463 | int tcp_disconnect(struct sock *sk, int flags); | |
464 | ||
465 | void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); | |
466 | int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); | |
467 | void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); | |
468 | ||
469 | /* From syncookies.c */ | |
470 | struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, | |
471 | struct request_sock *req, | |
472 | struct dst_entry *dst); | |
473 | int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, | |
474 | u32 cookie); | |
475 | struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); | |
476 | #ifdef CONFIG_SYN_COOKIES | |
477 | ||
478 | /* Syncookies use a monotonic timer which increments every 60 seconds. | |
479 | * This counter is used both as a hash input and partially encoded into | |
480 | * the cookie value. A cookie is only validated further if the delta | |
481 | * between the current counter value and the encoded one is less than this, | |
482 | * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if | |
483 | * the counter advances immediately after a cookie is generated). | |
484 | */ | |
485 | #define MAX_SYNCOOKIE_AGE 2 | |
486 | #define TCP_SYNCOOKIE_PERIOD (60 * HZ) | |
487 | #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) | |
488 | ||
489 | /* syncookies: remember time of last synqueue overflow | |
490 | * But do not dirty this field too often (once per second is enough) | |
491 | * It is racy as we do not hold a lock, but race is very minor. | |
492 | */ | |
493 | static inline void tcp_synq_overflow(const struct sock *sk) | |
494 | { | |
495 | unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; | |
496 | unsigned long now = jiffies; | |
497 | ||
498 | if (time_after(now, last_overflow + HZ)) | |
499 | tcp_sk(sk)->rx_opt.ts_recent_stamp = now; | |
500 | } | |
501 | ||
502 | /* syncookies: no recent synqueue overflow on this listening socket? */ | |
503 | static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) | |
504 | { | |
505 | unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp; | |
506 | ||
507 | return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID); | |
508 | } | |
509 | ||
510 | static inline u32 tcp_cookie_time(void) | |
511 | { | |
512 | u64 val = get_jiffies_64(); | |
513 | ||
514 | do_div(val, TCP_SYNCOOKIE_PERIOD); | |
515 | return val; | |
516 | } | |
517 | ||
518 | u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, | |
519 | u16 *mssp); | |
520 | __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); | |
521 | __u32 cookie_init_timestamp(struct request_sock *req); | |
522 | bool cookie_timestamp_decode(struct tcp_options_received *opt); | |
523 | bool cookie_ecn_ok(const struct tcp_options_received *opt, | |
524 | const struct net *net, const struct dst_entry *dst); | |
525 | ||
526 | /* From net/ipv6/syncookies.c */ | |
527 | int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, | |
528 | u32 cookie); | |
529 | struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); | |
530 | ||
531 | u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, | |
532 | const struct tcphdr *th, u16 *mssp); | |
533 | __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); | |
534 | #endif | |
535 | /* tcp_output.c */ | |
536 | ||
537 | void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, | |
538 | int nonagle); | |
539 | bool tcp_may_send_now(struct sock *sk); | |
540 | int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); | |
541 | int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); | |
542 | void tcp_retransmit_timer(struct sock *sk); | |
543 | void tcp_xmit_retransmit_queue(struct sock *); | |
544 | void tcp_simple_retransmit(struct sock *); | |
545 | int tcp_trim_head(struct sock *, struct sk_buff *, u32); | |
546 | int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t); | |
547 | ||
548 | void tcp_send_probe0(struct sock *); | |
549 | void tcp_send_partial(struct sock *); | |
550 | int tcp_write_wakeup(struct sock *, int mib); | |
551 | void tcp_send_fin(struct sock *sk); | |
552 | void tcp_send_active_reset(struct sock *sk, gfp_t priority); | |
553 | int tcp_send_synack(struct sock *); | |
554 | void tcp_push_one(struct sock *, unsigned int mss_now); | |
555 | void tcp_send_ack(struct sock *sk); | |
556 | void tcp_send_delayed_ack(struct sock *sk); | |
557 | void tcp_send_loss_probe(struct sock *sk); | |
558 | bool tcp_schedule_loss_probe(struct sock *sk); | |
559 | void tcp_skb_collapse_tstamp(struct sk_buff *skb, | |
560 | const struct sk_buff *next_skb); | |
561 | ||
562 | /* tcp_input.c */ | |
563 | void tcp_resume_early_retransmit(struct sock *sk); | |
564 | void tcp_rearm_rto(struct sock *sk); | |
565 | void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); | |
566 | void tcp_reset(struct sock *sk); | |
567 | void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); | |
568 | void tcp_fin(struct sock *sk); | |
569 | ||
570 | /* tcp_timer.c */ | |
571 | void tcp_init_xmit_timers(struct sock *); | |
572 | static inline void tcp_clear_xmit_timers(struct sock *sk) | |
573 | { | |
574 | inet_csk_clear_xmit_timers(sk); | |
575 | } | |
576 | ||
577 | unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); | |
578 | unsigned int tcp_current_mss(struct sock *sk); | |
579 | ||
580 | /* Bound MSS / TSO packet size with the half of the window */ | |
581 | static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) | |
582 | { | |
583 | int cutoff; | |
584 | ||
585 | /* When peer uses tiny windows, there is no use in packetizing | |
586 | * to sub-MSS pieces for the sake of SWS or making sure there | |
587 | * are enough packets in the pipe for fast recovery. | |
588 | * | |
589 | * On the other hand, for extremely large MSS devices, handling | |
590 | * smaller than MSS windows in this way does make sense. | |
591 | */ | |
592 | if (tp->max_window >= 512) | |
593 | cutoff = (tp->max_window >> 1); | |
594 | else | |
595 | cutoff = tp->max_window; | |
596 | ||
597 | if (cutoff && pktsize > cutoff) | |
598 | return max_t(int, cutoff, 68U - tp->tcp_header_len); | |
599 | else | |
600 | return pktsize; | |
601 | } | |
602 | ||
603 | /* tcp.c */ | |
604 | void tcp_get_info(struct sock *, struct tcp_info *); | |
605 | ||
606 | /* Read 'sendfile()'-style from a TCP socket */ | |
607 | typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, | |
608 | unsigned int, size_t); | |
609 | int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, | |
610 | sk_read_actor_t recv_actor); | |
611 | ||
612 | void tcp_initialize_rcv_mss(struct sock *sk); | |
613 | ||
614 | int tcp_mtu_to_mss(struct sock *sk, int pmtu); | |
615 | int tcp_mss_to_mtu(struct sock *sk, int mss); | |
616 | void tcp_mtup_init(struct sock *sk); | |
617 | void tcp_init_buffer_space(struct sock *sk); | |
618 | ||
619 | static inline void tcp_bound_rto(const struct sock *sk) | |
620 | { | |
621 | if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) | |
622 | inet_csk(sk)->icsk_rto = TCP_RTO_MAX; | |
623 | } | |
624 | ||
625 | static inline u32 __tcp_set_rto(const struct tcp_sock *tp) | |
626 | { | |
627 | return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); | |
628 | } | |
629 | ||
630 | static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) | |
631 | { | |
632 | tp->pred_flags = htonl((tp->tcp_header_len << 26) | | |
633 | ntohl(TCP_FLAG_ACK) | | |
634 | snd_wnd); | |
635 | } | |
636 | ||
637 | static inline void tcp_fast_path_on(struct tcp_sock *tp) | |
638 | { | |
639 | __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); | |
640 | } | |
641 | ||
642 | static inline void tcp_fast_path_check(struct sock *sk) | |
643 | { | |
644 | struct tcp_sock *tp = tcp_sk(sk); | |
645 | ||
646 | if (skb_queue_empty(&tp->out_of_order_queue) && | |
647 | tp->rcv_wnd && | |
648 | atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && | |
649 | !tp->urg_data) | |
650 | tcp_fast_path_on(tp); | |
651 | } | |
652 | ||
653 | /* Compute the actual rto_min value */ | |
654 | static inline u32 tcp_rto_min(struct sock *sk) | |
655 | { | |
656 | const struct dst_entry *dst = __sk_dst_get(sk); | |
657 | u32 rto_min = TCP_RTO_MIN; | |
658 | ||
659 | if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) | |
660 | rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); | |
661 | return rto_min; | |
662 | } | |
663 | ||
664 | static inline u32 tcp_rto_min_us(struct sock *sk) | |
665 | { | |
666 | return jiffies_to_usecs(tcp_rto_min(sk)); | |
667 | } | |
668 | ||
669 | static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) | |
670 | { | |
671 | return dst_metric_locked(dst, RTAX_CC_ALGO); | |
672 | } | |
673 | ||
674 | /* Minimum RTT in usec. ~0 means not available. */ | |
675 | static inline u32 tcp_min_rtt(const struct tcp_sock *tp) | |
676 | { | |
677 | return tp->rtt_min[0].rtt; | |
678 | } | |
679 | ||
680 | /* Compute the actual receive window we are currently advertising. | |
681 | * Rcv_nxt can be after the window if our peer push more data | |
682 | * than the offered window. | |
683 | */ | |
684 | static inline u32 tcp_receive_window(const struct tcp_sock *tp) | |
685 | { | |
686 | s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; | |
687 | ||
688 | if (win < 0) | |
689 | win = 0; | |
690 | return (u32) win; | |
691 | } | |
692 | ||
693 | /* Choose a new window, without checks for shrinking, and without | |
694 | * scaling applied to the result. The caller does these things | |
695 | * if necessary. This is a "raw" window selection. | |
696 | */ | |
697 | u32 __tcp_select_window(struct sock *sk); | |
698 | ||
699 | void tcp_send_window_probe(struct sock *sk); | |
700 | ||
701 | /* TCP timestamps are only 32-bits, this causes a slight | |
702 | * complication on 64-bit systems since we store a snapshot | |
703 | * of jiffies in the buffer control blocks below. We decided | |
704 | * to use only the low 32-bits of jiffies and hide the ugly | |
705 | * casts with the following macro. | |
706 | */ | |
707 | #define tcp_time_stamp ((__u32)(jiffies)) | |
708 | ||
709 | static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) | |
710 | { | |
711 | return skb->skb_mstamp.stamp_jiffies; | |
712 | } | |
713 | ||
714 | ||
715 | #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) | |
716 | ||
717 | #define TCPHDR_FIN 0x01 | |
718 | #define TCPHDR_SYN 0x02 | |
719 | #define TCPHDR_RST 0x04 | |
720 | #define TCPHDR_PSH 0x08 | |
721 | #define TCPHDR_ACK 0x10 | |
722 | #define TCPHDR_URG 0x20 | |
723 | #define TCPHDR_ECE 0x40 | |
724 | #define TCPHDR_CWR 0x80 | |
725 | ||
726 | #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) | |
727 | ||
728 | /* This is what the send packet queuing engine uses to pass | |
729 | * TCP per-packet control information to the transmission code. | |
730 | * We also store the host-order sequence numbers in here too. | |
731 | * This is 44 bytes if IPV6 is enabled. | |
732 | * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. | |
733 | */ | |
734 | struct tcp_skb_cb { | |
735 | __u32 seq; /* Starting sequence number */ | |
736 | __u32 end_seq; /* SEQ + FIN + SYN + datalen */ | |
737 | union { | |
738 | /* Note : tcp_tw_isn is used in input path only | |
739 | * (isn chosen by tcp_timewait_state_process()) | |
740 | * | |
741 | * tcp_gso_segs/size are used in write queue only, | |
742 | * cf tcp_skb_pcount()/tcp_skb_mss() | |
743 | */ | |
744 | __u32 tcp_tw_isn; | |
745 | struct { | |
746 | u16 tcp_gso_segs; | |
747 | u16 tcp_gso_size; | |
748 | }; | |
749 | }; | |
750 | __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ | |
751 | ||
752 | __u8 sacked; /* State flags for SACK/FACK. */ | |
753 | #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ | |
754 | #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ | |
755 | #define TCPCB_LOST 0x04 /* SKB is lost */ | |
756 | #define TCPCB_TAGBITS 0x07 /* All tag bits */ | |
757 | #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */ | |
758 | #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ | |
759 | #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ | |
760 | TCPCB_REPAIRED) | |
761 | ||
762 | __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ | |
763 | __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ | |
764 | eor:1, /* Is skb MSG_EOR marked? */ | |
765 | unused:6; | |
766 | __u32 ack_seq; /* Sequence number ACK'd */ | |
767 | union { | |
768 | struct { | |
769 | /* There is space for up to 20 bytes */ | |
770 | } tx; /* only used for outgoing skbs */ | |
771 | union { | |
772 | struct inet_skb_parm h4; | |
773 | #if IS_ENABLED(CONFIG_IPV6) | |
774 | struct inet6_skb_parm h6; | |
775 | #endif | |
776 | } header; /* For incoming skbs */ | |
777 | }; | |
778 | }; | |
779 | ||
780 | #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) | |
781 | ||
782 | ||
783 | #if IS_ENABLED(CONFIG_IPV6) | |
784 | /* This is the variant of inet6_iif() that must be used by TCP, | |
785 | * as TCP moves IP6CB into a different location in skb->cb[] | |
786 | */ | |
787 | static inline int tcp_v6_iif(const struct sk_buff *skb) | |
788 | { | |
789 | bool l3_slave = skb_l3mdev_slave(TCP_SKB_CB(skb)->header.h6.flags); | |
790 | ||
791 | return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; | |
792 | } | |
793 | #endif | |
794 | ||
795 | /* Due to TSO, an SKB can be composed of multiple actual | |
796 | * packets. To keep these tracked properly, we use this. | |
797 | */ | |
798 | static inline int tcp_skb_pcount(const struct sk_buff *skb) | |
799 | { | |
800 | return TCP_SKB_CB(skb)->tcp_gso_segs; | |
801 | } | |
802 | ||
803 | static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) | |
804 | { | |
805 | TCP_SKB_CB(skb)->tcp_gso_segs = segs; | |
806 | } | |
807 | ||
808 | static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) | |
809 | { | |
810 | TCP_SKB_CB(skb)->tcp_gso_segs += segs; | |
811 | } | |
812 | ||
813 | /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ | |
814 | static inline int tcp_skb_mss(const struct sk_buff *skb) | |
815 | { | |
816 | return TCP_SKB_CB(skb)->tcp_gso_size; | |
817 | } | |
818 | ||
819 | static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) | |
820 | { | |
821 | return likely(!TCP_SKB_CB(skb)->eor); | |
822 | } | |
823 | ||
824 | /* Events passed to congestion control interface */ | |
825 | enum tcp_ca_event { | |
826 | CA_EVENT_TX_START, /* first transmit when no packets in flight */ | |
827 | CA_EVENT_CWND_RESTART, /* congestion window restart */ | |
828 | CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ | |
829 | CA_EVENT_LOSS, /* loss timeout */ | |
830 | CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ | |
831 | CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ | |
832 | CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */ | |
833 | CA_EVENT_NON_DELAYED_ACK, | |
834 | }; | |
835 | ||
836 | /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ | |
837 | enum tcp_ca_ack_event_flags { | |
838 | CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ | |
839 | CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ | |
840 | CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ | |
841 | }; | |
842 | ||
843 | /* | |
844 | * Interface for adding new TCP congestion control handlers | |
845 | */ | |
846 | #define TCP_CA_NAME_MAX 16 | |
847 | #define TCP_CA_MAX 128 | |
848 | #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) | |
849 | ||
850 | #define TCP_CA_UNSPEC 0 | |
851 | ||
852 | /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ | |
853 | #define TCP_CONG_NON_RESTRICTED 0x1 | |
854 | /* Requires ECN/ECT set on all packets */ | |
855 | #define TCP_CONG_NEEDS_ECN 0x2 | |
856 | ||
857 | union tcp_cc_info; | |
858 | ||
859 | struct ack_sample { | |
860 | u32 pkts_acked; | |
861 | s32 rtt_us; | |
862 | }; | |
863 | ||
864 | struct tcp_congestion_ops { | |
865 | struct list_head list; | |
866 | u32 key; | |
867 | u32 flags; | |
868 | ||
869 | /* initialize private data (optional) */ | |
870 | void (*init)(struct sock *sk); | |
871 | /* cleanup private data (optional) */ | |
872 | void (*release)(struct sock *sk); | |
873 | ||
874 | /* return slow start threshold (required) */ | |
875 | u32 (*ssthresh)(struct sock *sk); | |
876 | /* do new cwnd calculation (required) */ | |
877 | void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); | |
878 | /* call before changing ca_state (optional) */ | |
879 | void (*set_state)(struct sock *sk, u8 new_state); | |
880 | /* call when cwnd event occurs (optional) */ | |
881 | void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); | |
882 | /* call when ack arrives (optional) */ | |
883 | void (*in_ack_event)(struct sock *sk, u32 flags); | |
884 | /* new value of cwnd after loss (optional) */ | |
885 | u32 (*undo_cwnd)(struct sock *sk); | |
886 | /* hook for packet ack accounting (optional) */ | |
887 | void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); | |
888 | /* get info for inet_diag (optional) */ | |
889 | size_t (*get_info)(struct sock *sk, u32 ext, int *attr, | |
890 | union tcp_cc_info *info); | |
891 | ||
892 | char name[TCP_CA_NAME_MAX]; | |
893 | struct module *owner; | |
894 | }; | |
895 | ||
896 | int tcp_register_congestion_control(struct tcp_congestion_ops *type); | |
897 | void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); | |
898 | ||
899 | void tcp_assign_congestion_control(struct sock *sk); | |
900 | void tcp_init_congestion_control(struct sock *sk); | |
901 | void tcp_cleanup_congestion_control(struct sock *sk); | |
902 | int tcp_set_default_congestion_control(const char *name); | |
903 | void tcp_get_default_congestion_control(char *name); | |
904 | void tcp_get_available_congestion_control(char *buf, size_t len); | |
905 | void tcp_get_allowed_congestion_control(char *buf, size_t len); | |
906 | int tcp_set_allowed_congestion_control(char *allowed); | |
907 | int tcp_set_congestion_control(struct sock *sk, const char *name); | |
908 | u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); | |
909 | void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); | |
910 | ||
911 | u32 tcp_reno_ssthresh(struct sock *sk); | |
912 | void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); | |
913 | extern struct tcp_congestion_ops tcp_reno; | |
914 | ||
915 | struct tcp_congestion_ops *tcp_ca_find_key(u32 key); | |
916 | u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca); | |
917 | #ifdef CONFIG_INET | |
918 | char *tcp_ca_get_name_by_key(u32 key, char *buffer); | |
919 | #else | |
920 | static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) | |
921 | { | |
922 | return NULL; | |
923 | } | |
924 | #endif | |
925 | ||
926 | static inline bool tcp_ca_needs_ecn(const struct sock *sk) | |
927 | { | |
928 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
929 | ||
930 | return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; | |
931 | } | |
932 | ||
933 | static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) | |
934 | { | |
935 | struct inet_connection_sock *icsk = inet_csk(sk); | |
936 | ||
937 | if (icsk->icsk_ca_ops->set_state) | |
938 | icsk->icsk_ca_ops->set_state(sk, ca_state); | |
939 | icsk->icsk_ca_state = ca_state; | |
940 | } | |
941 | ||
942 | static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) | |
943 | { | |
944 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
945 | ||
946 | if (icsk->icsk_ca_ops->cwnd_event) | |
947 | icsk->icsk_ca_ops->cwnd_event(sk, event); | |
948 | } | |
949 | ||
950 | /* These functions determine how the current flow behaves in respect of SACK | |
951 | * handling. SACK is negotiated with the peer, and therefore it can vary | |
952 | * between different flows. | |
953 | * | |
954 | * tcp_is_sack - SACK enabled | |
955 | * tcp_is_reno - No SACK | |
956 | * tcp_is_fack - FACK enabled, implies SACK enabled | |
957 | */ | |
958 | static inline int tcp_is_sack(const struct tcp_sock *tp) | |
959 | { | |
960 | return tp->rx_opt.sack_ok; | |
961 | } | |
962 | ||
963 | static inline bool tcp_is_reno(const struct tcp_sock *tp) | |
964 | { | |
965 | return !tcp_is_sack(tp); | |
966 | } | |
967 | ||
968 | static inline bool tcp_is_fack(const struct tcp_sock *tp) | |
969 | { | |
970 | return tp->rx_opt.sack_ok & TCP_FACK_ENABLED; | |
971 | } | |
972 | ||
973 | static inline void tcp_enable_fack(struct tcp_sock *tp) | |
974 | { | |
975 | tp->rx_opt.sack_ok |= TCP_FACK_ENABLED; | |
976 | } | |
977 | ||
978 | /* TCP early-retransmit (ER) is similar to but more conservative than | |
979 | * the thin-dupack feature. Enable ER only if thin-dupack is disabled. | |
980 | */ | |
981 | static inline void tcp_enable_early_retrans(struct tcp_sock *tp) | |
982 | { | |
983 | struct net *net = sock_net((struct sock *)tp); | |
984 | ||
985 | tp->do_early_retrans = sysctl_tcp_early_retrans && | |
986 | sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack && | |
987 | net->ipv4.sysctl_tcp_reordering == 3; | |
988 | } | |
989 | ||
990 | static inline void tcp_disable_early_retrans(struct tcp_sock *tp) | |
991 | { | |
992 | tp->do_early_retrans = 0; | |
993 | } | |
994 | ||
995 | static inline unsigned int tcp_left_out(const struct tcp_sock *tp) | |
996 | { | |
997 | return tp->sacked_out + tp->lost_out; | |
998 | } | |
999 | ||
1000 | /* This determines how many packets are "in the network" to the best | |
1001 | * of our knowledge. In many cases it is conservative, but where | |
1002 | * detailed information is available from the receiver (via SACK | |
1003 | * blocks etc.) we can make more aggressive calculations. | |
1004 | * | |
1005 | * Use this for decisions involving congestion control, use just | |
1006 | * tp->packets_out to determine if the send queue is empty or not. | |
1007 | * | |
1008 | * Read this equation as: | |
1009 | * | |
1010 | * "Packets sent once on transmission queue" MINUS | |
1011 | * "Packets left network, but not honestly ACKed yet" PLUS | |
1012 | * "Packets fast retransmitted" | |
1013 | */ | |
1014 | static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) | |
1015 | { | |
1016 | return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; | |
1017 | } | |
1018 | ||
1019 | #define TCP_INFINITE_SSTHRESH 0x7fffffff | |
1020 | ||
1021 | static inline bool tcp_in_slow_start(const struct tcp_sock *tp) | |
1022 | { | |
1023 | return tp->snd_cwnd < tp->snd_ssthresh; | |
1024 | } | |
1025 | ||
1026 | static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) | |
1027 | { | |
1028 | return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; | |
1029 | } | |
1030 | ||
1031 | static inline bool tcp_in_cwnd_reduction(const struct sock *sk) | |
1032 | { | |
1033 | return (TCPF_CA_CWR | TCPF_CA_Recovery) & | |
1034 | (1 << inet_csk(sk)->icsk_ca_state); | |
1035 | } | |
1036 | ||
1037 | /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. | |
1038 | * The exception is cwnd reduction phase, when cwnd is decreasing towards | |
1039 | * ssthresh. | |
1040 | */ | |
1041 | static inline __u32 tcp_current_ssthresh(const struct sock *sk) | |
1042 | { | |
1043 | const struct tcp_sock *tp = tcp_sk(sk); | |
1044 | ||
1045 | if (tcp_in_cwnd_reduction(sk)) | |
1046 | return tp->snd_ssthresh; | |
1047 | else | |
1048 | return max(tp->snd_ssthresh, | |
1049 | ((tp->snd_cwnd >> 1) + | |
1050 | (tp->snd_cwnd >> 2))); | |
1051 | } | |
1052 | ||
1053 | /* Use define here intentionally to get WARN_ON location shown at the caller */ | |
1054 | #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) | |
1055 | ||
1056 | void tcp_enter_cwr(struct sock *sk); | |
1057 | __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); | |
1058 | ||
1059 | /* The maximum number of MSS of available cwnd for which TSO defers | |
1060 | * sending if not using sysctl_tcp_tso_win_divisor. | |
1061 | */ | |
1062 | static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) | |
1063 | { | |
1064 | return 3; | |
1065 | } | |
1066 | ||
1067 | /* Returns end sequence number of the receiver's advertised window */ | |
1068 | static inline u32 tcp_wnd_end(const struct tcp_sock *tp) | |
1069 | { | |
1070 | return tp->snd_una + tp->snd_wnd; | |
1071 | } | |
1072 | ||
1073 | /* We follow the spirit of RFC2861 to validate cwnd but implement a more | |
1074 | * flexible approach. The RFC suggests cwnd should not be raised unless | |
1075 | * it was fully used previously. And that's exactly what we do in | |
1076 | * congestion avoidance mode. But in slow start we allow cwnd to grow | |
1077 | * as long as the application has used half the cwnd. | |
1078 | * Example : | |
1079 | * cwnd is 10 (IW10), but application sends 9 frames. | |
1080 | * We allow cwnd to reach 18 when all frames are ACKed. | |
1081 | * This check is safe because it's as aggressive as slow start which already | |
1082 | * risks 100% overshoot. The advantage is that we discourage application to | |
1083 | * either send more filler packets or data to artificially blow up the cwnd | |
1084 | * usage, and allow application-limited process to probe bw more aggressively. | |
1085 | */ | |
1086 | static inline bool tcp_is_cwnd_limited(const struct sock *sk) | |
1087 | { | |
1088 | const struct tcp_sock *tp = tcp_sk(sk); | |
1089 | ||
1090 | /* If in slow start, ensure cwnd grows to twice what was ACKed. */ | |
1091 | if (tcp_in_slow_start(tp)) | |
1092 | return tp->snd_cwnd < 2 * tp->max_packets_out; | |
1093 | ||
1094 | return tp->is_cwnd_limited; | |
1095 | } | |
1096 | ||
1097 | /* Something is really bad, we could not queue an additional packet, | |
1098 | * because qdisc is full or receiver sent a 0 window. | |
1099 | * We do not want to add fuel to the fire, or abort too early, | |
1100 | * so make sure the timer we arm now is at least 200ms in the future, | |
1101 | * regardless of current icsk_rto value (as it could be ~2ms) | |
1102 | */ | |
1103 | static inline unsigned long tcp_probe0_base(const struct sock *sk) | |
1104 | { | |
1105 | return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); | |
1106 | } | |
1107 | ||
1108 | /* Variant of inet_csk_rto_backoff() used for zero window probes */ | |
1109 | static inline unsigned long tcp_probe0_when(const struct sock *sk, | |
1110 | unsigned long max_when) | |
1111 | { | |
1112 | u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; | |
1113 | ||
1114 | return (unsigned long)min_t(u64, when, max_when); | |
1115 | } | |
1116 | ||
1117 | static inline void tcp_check_probe_timer(struct sock *sk) | |
1118 | { | |
1119 | if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) | |
1120 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0, | |
1121 | tcp_probe0_base(sk), TCP_RTO_MAX); | |
1122 | } | |
1123 | ||
1124 | static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) | |
1125 | { | |
1126 | tp->snd_wl1 = seq; | |
1127 | } | |
1128 | ||
1129 | static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) | |
1130 | { | |
1131 | tp->snd_wl1 = seq; | |
1132 | } | |
1133 | ||
1134 | /* | |
1135 | * Calculate(/check) TCP checksum | |
1136 | */ | |
1137 | static inline __sum16 tcp_v4_check(int len, __be32 saddr, | |
1138 | __be32 daddr, __wsum base) | |
1139 | { | |
1140 | return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); | |
1141 | } | |
1142 | ||
1143 | static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb) | |
1144 | { | |
1145 | return __skb_checksum_complete(skb); | |
1146 | } | |
1147 | ||
1148 | static inline bool tcp_checksum_complete(struct sk_buff *skb) | |
1149 | { | |
1150 | return !skb_csum_unnecessary(skb) && | |
1151 | __tcp_checksum_complete(skb); | |
1152 | } | |
1153 | ||
1154 | /* Prequeue for VJ style copy to user, combined with checksumming. */ | |
1155 | ||
1156 | static inline void tcp_prequeue_init(struct tcp_sock *tp) | |
1157 | { | |
1158 | tp->ucopy.task = NULL; | |
1159 | tp->ucopy.len = 0; | |
1160 | tp->ucopy.memory = 0; | |
1161 | skb_queue_head_init(&tp->ucopy.prequeue); | |
1162 | } | |
1163 | ||
1164 | bool tcp_prequeue(struct sock *sk, struct sk_buff *skb); | |
1165 | ||
1166 | #undef STATE_TRACE | |
1167 | ||
1168 | #ifdef STATE_TRACE | |
1169 | static const char *statename[]={ | |
1170 | "Unused","Established","Syn Sent","Syn Recv", | |
1171 | "Fin Wait 1","Fin Wait 2","Time Wait", "Close", | |
1172 | "Close Wait","Last ACK","Listen","Closing" | |
1173 | }; | |
1174 | #endif | |
1175 | void tcp_set_state(struct sock *sk, int state); | |
1176 | ||
1177 | void tcp_done(struct sock *sk); | |
1178 | ||
1179 | int tcp_abort(struct sock *sk, int err); | |
1180 | ||
1181 | static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) | |
1182 | { | |
1183 | rx_opt->dsack = 0; | |
1184 | rx_opt->num_sacks = 0; | |
1185 | } | |
1186 | ||
1187 | u32 tcp_default_init_rwnd(u32 mss); | |
1188 | void tcp_cwnd_restart(struct sock *sk, s32 delta); | |
1189 | ||
1190 | static inline void tcp_slow_start_after_idle_check(struct sock *sk) | |
1191 | { | |
1192 | struct tcp_sock *tp = tcp_sk(sk); | |
1193 | s32 delta; | |
1194 | ||
1195 | if (!sysctl_tcp_slow_start_after_idle || tp->packets_out) | |
1196 | return; | |
1197 | delta = tcp_time_stamp - tp->lsndtime; | |
1198 | if (delta > inet_csk(sk)->icsk_rto) | |
1199 | tcp_cwnd_restart(sk, delta); | |
1200 | } | |
1201 | ||
1202 | /* Determine a window scaling and initial window to offer. */ | |
1203 | void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd, | |
1204 | __u32 *window_clamp, int wscale_ok, | |
1205 | __u8 *rcv_wscale, __u32 init_rcv_wnd); | |
1206 | ||
1207 | static inline int tcp_win_from_space(int space) | |
1208 | { | |
1209 | return sysctl_tcp_adv_win_scale<=0 ? | |
1210 | (space>>(-sysctl_tcp_adv_win_scale)) : | |
1211 | space - (space>>sysctl_tcp_adv_win_scale); | |
1212 | } | |
1213 | ||
1214 | /* Note: caller must be prepared to deal with negative returns */ | |
1215 | static inline int tcp_space(const struct sock *sk) | |
1216 | { | |
1217 | return tcp_win_from_space(sk->sk_rcvbuf - | |
1218 | atomic_read(&sk->sk_rmem_alloc)); | |
1219 | } | |
1220 | ||
1221 | static inline int tcp_full_space(const struct sock *sk) | |
1222 | { | |
1223 | return tcp_win_from_space(sk->sk_rcvbuf); | |
1224 | } | |
1225 | ||
1226 | extern void tcp_openreq_init_rwin(struct request_sock *req, | |
1227 | const struct sock *sk_listener, | |
1228 | const struct dst_entry *dst); | |
1229 | ||
1230 | void tcp_enter_memory_pressure(struct sock *sk); | |
1231 | ||
1232 | static inline int keepalive_intvl_when(const struct tcp_sock *tp) | |
1233 | { | |
1234 | struct net *net = sock_net((struct sock *)tp); | |
1235 | ||
1236 | return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; | |
1237 | } | |
1238 | ||
1239 | static inline int keepalive_time_when(const struct tcp_sock *tp) | |
1240 | { | |
1241 | struct net *net = sock_net((struct sock *)tp); | |
1242 | ||
1243 | return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; | |
1244 | } | |
1245 | ||
1246 | static inline int keepalive_probes(const struct tcp_sock *tp) | |
1247 | { | |
1248 | struct net *net = sock_net((struct sock *)tp); | |
1249 | ||
1250 | return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; | |
1251 | } | |
1252 | ||
1253 | static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) | |
1254 | { | |
1255 | const struct inet_connection_sock *icsk = &tp->inet_conn; | |
1256 | ||
1257 | return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime, | |
1258 | tcp_time_stamp - tp->rcv_tstamp); | |
1259 | } | |
1260 | ||
1261 | static inline int tcp_fin_time(const struct sock *sk) | |
1262 | { | |
1263 | int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; | |
1264 | const int rto = inet_csk(sk)->icsk_rto; | |
1265 | ||
1266 | if (fin_timeout < (rto << 2) - (rto >> 1)) | |
1267 | fin_timeout = (rto << 2) - (rto >> 1); | |
1268 | ||
1269 | return fin_timeout; | |
1270 | } | |
1271 | ||
1272 | static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, | |
1273 | int paws_win) | |
1274 | { | |
1275 | if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) | |
1276 | return true; | |
1277 | if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)) | |
1278 | return true; | |
1279 | /* | |
1280 | * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, | |
1281 | * then following tcp messages have valid values. Ignore 0 value, | |
1282 | * or else 'negative' tsval might forbid us to accept their packets. | |
1283 | */ | |
1284 | if (!rx_opt->ts_recent) | |
1285 | return true; | |
1286 | return false; | |
1287 | } | |
1288 | ||
1289 | static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, | |
1290 | int rst) | |
1291 | { | |
1292 | if (tcp_paws_check(rx_opt, 0)) | |
1293 | return false; | |
1294 | ||
1295 | /* RST segments are not recommended to carry timestamp, | |
1296 | and, if they do, it is recommended to ignore PAWS because | |
1297 | "their cleanup function should take precedence over timestamps." | |
1298 | Certainly, it is mistake. It is necessary to understand the reasons | |
1299 | of this constraint to relax it: if peer reboots, clock may go | |
1300 | out-of-sync and half-open connections will not be reset. | |
1301 | Actually, the problem would be not existing if all | |
1302 | the implementations followed draft about maintaining clock | |
1303 | via reboots. Linux-2.2 DOES NOT! | |
1304 | ||
1305 | However, we can relax time bounds for RST segments to MSL. | |
1306 | */ | |
1307 | if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) | |
1308 | return false; | |
1309 | return true; | |
1310 | } | |
1311 | ||
1312 | bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, | |
1313 | int mib_idx, u32 *last_oow_ack_time); | |
1314 | ||
1315 | static inline void tcp_mib_init(struct net *net) | |
1316 | { | |
1317 | /* See RFC 2012 */ | |
1318 | TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); | |
1319 | TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); | |
1320 | TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); | |
1321 | TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); | |
1322 | } | |
1323 | ||
1324 | /* from STCP */ | |
1325 | static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) | |
1326 | { | |
1327 | tp->lost_skb_hint = NULL; | |
1328 | } | |
1329 | ||
1330 | static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) | |
1331 | { | |
1332 | tcp_clear_retrans_hints_partial(tp); | |
1333 | tp->retransmit_skb_hint = NULL; | |
1334 | } | |
1335 | ||
1336 | union tcp_md5_addr { | |
1337 | struct in_addr a4; | |
1338 | #if IS_ENABLED(CONFIG_IPV6) | |
1339 | struct in6_addr a6; | |
1340 | #endif | |
1341 | }; | |
1342 | ||
1343 | /* - key database */ | |
1344 | struct tcp_md5sig_key { | |
1345 | struct hlist_node node; | |
1346 | u8 keylen; | |
1347 | u8 family; /* AF_INET or AF_INET6 */ | |
1348 | union tcp_md5_addr addr; | |
1349 | u8 key[TCP_MD5SIG_MAXKEYLEN]; | |
1350 | struct rcu_head rcu; | |
1351 | }; | |
1352 | ||
1353 | /* - sock block */ | |
1354 | struct tcp_md5sig_info { | |
1355 | struct hlist_head head; | |
1356 | struct rcu_head rcu; | |
1357 | }; | |
1358 | ||
1359 | /* - pseudo header */ | |
1360 | struct tcp4_pseudohdr { | |
1361 | __be32 saddr; | |
1362 | __be32 daddr; | |
1363 | __u8 pad; | |
1364 | __u8 protocol; | |
1365 | __be16 len; | |
1366 | }; | |
1367 | ||
1368 | struct tcp6_pseudohdr { | |
1369 | struct in6_addr saddr; | |
1370 | struct in6_addr daddr; | |
1371 | __be32 len; | |
1372 | __be32 protocol; /* including padding */ | |
1373 | }; | |
1374 | ||
1375 | union tcp_md5sum_block { | |
1376 | struct tcp4_pseudohdr ip4; | |
1377 | #if IS_ENABLED(CONFIG_IPV6) | |
1378 | struct tcp6_pseudohdr ip6; | |
1379 | #endif | |
1380 | }; | |
1381 | ||
1382 | /* - pool: digest algorithm, hash description and scratch buffer */ | |
1383 | struct tcp_md5sig_pool { | |
1384 | struct ahash_request *md5_req; | |
1385 | union tcp_md5sum_block md5_blk; | |
1386 | }; | |
1387 | ||
1388 | /* - functions */ | |
1389 | int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, | |
1390 | const struct sock *sk, const struct sk_buff *skb); | |
1391 | int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, | |
1392 | int family, const u8 *newkey, u8 newkeylen, gfp_t gfp); | |
1393 | int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, | |
1394 | int family); | |
1395 | struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, | |
1396 | const struct sock *addr_sk); | |
1397 | ||
1398 | #ifdef CONFIG_TCP_MD5SIG | |
1399 | struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, | |
1400 | const union tcp_md5_addr *addr, | |
1401 | int family); | |
1402 | #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) | |
1403 | #else | |
1404 | static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, | |
1405 | const union tcp_md5_addr *addr, | |
1406 | int family) | |
1407 | { | |
1408 | return NULL; | |
1409 | } | |
1410 | #define tcp_twsk_md5_key(twsk) NULL | |
1411 | #endif | |
1412 | ||
1413 | bool tcp_alloc_md5sig_pool(void); | |
1414 | ||
1415 | struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); | |
1416 | static inline void tcp_put_md5sig_pool(void) | |
1417 | { | |
1418 | local_bh_enable(); | |
1419 | } | |
1420 | ||
1421 | int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *); | |
1422 | int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, | |
1423 | unsigned int header_len); | |
1424 | int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, | |
1425 | const struct tcp_md5sig_key *key); | |
1426 | ||
1427 | /* From tcp_fastopen.c */ | |
1428 | void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, | |
1429 | struct tcp_fastopen_cookie *cookie, int *syn_loss, | |
1430 | unsigned long *last_syn_loss); | |
1431 | void tcp_fastopen_cache_set(struct sock *sk, u16 mss, | |
1432 | struct tcp_fastopen_cookie *cookie, bool syn_lost, | |
1433 | u16 try_exp); | |
1434 | struct tcp_fastopen_request { | |
1435 | /* Fast Open cookie. Size 0 means a cookie request */ | |
1436 | struct tcp_fastopen_cookie cookie; | |
1437 | struct msghdr *data; /* data in MSG_FASTOPEN */ | |
1438 | size_t size; | |
1439 | int copied; /* queued in tcp_connect() */ | |
1440 | }; | |
1441 | void tcp_free_fastopen_req(struct tcp_sock *tp); | |
1442 | ||
1443 | extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx; | |
1444 | int tcp_fastopen_reset_cipher(void *key, unsigned int len); | |
1445 | void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); | |
1446 | struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, | |
1447 | struct request_sock *req, | |
1448 | struct tcp_fastopen_cookie *foc, | |
1449 | struct dst_entry *dst); | |
1450 | void tcp_fastopen_init_key_once(bool publish); | |
1451 | #define TCP_FASTOPEN_KEY_LENGTH 16 | |
1452 | ||
1453 | /* Fastopen key context */ | |
1454 | struct tcp_fastopen_context { | |
1455 | struct crypto_cipher *tfm; | |
1456 | __u8 key[TCP_FASTOPEN_KEY_LENGTH]; | |
1457 | struct rcu_head rcu; | |
1458 | }; | |
1459 | ||
1460 | /* write queue abstraction */ | |
1461 | static inline void tcp_write_queue_purge(struct sock *sk) | |
1462 | { | |
1463 | struct sk_buff *skb; | |
1464 | ||
1465 | while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL) | |
1466 | sk_wmem_free_skb(sk, skb); | |
1467 | sk_mem_reclaim(sk); | |
1468 | tcp_clear_all_retrans_hints(tcp_sk(sk)); | |
1469 | } | |
1470 | ||
1471 | static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) | |
1472 | { | |
1473 | return skb_peek(&sk->sk_write_queue); | |
1474 | } | |
1475 | ||
1476 | static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) | |
1477 | { | |
1478 | return skb_peek_tail(&sk->sk_write_queue); | |
1479 | } | |
1480 | ||
1481 | static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk, | |
1482 | const struct sk_buff *skb) | |
1483 | { | |
1484 | return skb_queue_next(&sk->sk_write_queue, skb); | |
1485 | } | |
1486 | ||
1487 | static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk, | |
1488 | const struct sk_buff *skb) | |
1489 | { | |
1490 | return skb_queue_prev(&sk->sk_write_queue, skb); | |
1491 | } | |
1492 | ||
1493 | #define tcp_for_write_queue(skb, sk) \ | |
1494 | skb_queue_walk(&(sk)->sk_write_queue, skb) | |
1495 | ||
1496 | #define tcp_for_write_queue_from(skb, sk) \ | |
1497 | skb_queue_walk_from(&(sk)->sk_write_queue, skb) | |
1498 | ||
1499 | #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ | |
1500 | skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) | |
1501 | ||
1502 | static inline struct sk_buff *tcp_send_head(const struct sock *sk) | |
1503 | { | |
1504 | return sk->sk_send_head; | |
1505 | } | |
1506 | ||
1507 | static inline bool tcp_skb_is_last(const struct sock *sk, | |
1508 | const struct sk_buff *skb) | |
1509 | { | |
1510 | return skb_queue_is_last(&sk->sk_write_queue, skb); | |
1511 | } | |
1512 | ||
1513 | static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb) | |
1514 | { | |
1515 | if (tcp_skb_is_last(sk, skb)) | |
1516 | sk->sk_send_head = NULL; | |
1517 | else | |
1518 | sk->sk_send_head = tcp_write_queue_next(sk, skb); | |
1519 | } | |
1520 | ||
1521 | static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked) | |
1522 | { | |
1523 | if (sk->sk_send_head == skb_unlinked) | |
1524 | sk->sk_send_head = NULL; | |
1525 | } | |
1526 | ||
1527 | static inline void tcp_init_send_head(struct sock *sk) | |
1528 | { | |
1529 | sk->sk_send_head = NULL; | |
1530 | } | |
1531 | ||
1532 | static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) | |
1533 | { | |
1534 | __skb_queue_tail(&sk->sk_write_queue, skb); | |
1535 | } | |
1536 | ||
1537 | static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) | |
1538 | { | |
1539 | __tcp_add_write_queue_tail(sk, skb); | |
1540 | ||
1541 | /* Queue it, remembering where we must start sending. */ | |
1542 | if (sk->sk_send_head == NULL) { | |
1543 | sk->sk_send_head = skb; | |
1544 | ||
1545 | if (tcp_sk(sk)->highest_sack == NULL) | |
1546 | tcp_sk(sk)->highest_sack = skb; | |
1547 | } | |
1548 | } | |
1549 | ||
1550 | static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb) | |
1551 | { | |
1552 | __skb_queue_head(&sk->sk_write_queue, skb); | |
1553 | } | |
1554 | ||
1555 | /* Insert buff after skb on the write queue of sk. */ | |
1556 | static inline void tcp_insert_write_queue_after(struct sk_buff *skb, | |
1557 | struct sk_buff *buff, | |
1558 | struct sock *sk) | |
1559 | { | |
1560 | __skb_queue_after(&sk->sk_write_queue, skb, buff); | |
1561 | } | |
1562 | ||
1563 | /* Insert new before skb on the write queue of sk. */ | |
1564 | static inline void tcp_insert_write_queue_before(struct sk_buff *new, | |
1565 | struct sk_buff *skb, | |
1566 | struct sock *sk) | |
1567 | { | |
1568 | __skb_queue_before(&sk->sk_write_queue, skb, new); | |
1569 | ||
1570 | if (sk->sk_send_head == skb) | |
1571 | sk->sk_send_head = new; | |
1572 | } | |
1573 | ||
1574 | static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) | |
1575 | { | |
1576 | __skb_unlink(skb, &sk->sk_write_queue); | |
1577 | } | |
1578 | ||
1579 | static inline bool tcp_write_queue_empty(struct sock *sk) | |
1580 | { | |
1581 | return skb_queue_empty(&sk->sk_write_queue); | |
1582 | } | |
1583 | ||
1584 | static inline void tcp_push_pending_frames(struct sock *sk) | |
1585 | { | |
1586 | if (tcp_send_head(sk)) { | |
1587 | struct tcp_sock *tp = tcp_sk(sk); | |
1588 | ||
1589 | __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); | |
1590 | } | |
1591 | } | |
1592 | ||
1593 | /* Start sequence of the skb just after the highest skb with SACKed | |
1594 | * bit, valid only if sacked_out > 0 or when the caller has ensured | |
1595 | * validity by itself. | |
1596 | */ | |
1597 | static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) | |
1598 | { | |
1599 | if (!tp->sacked_out) | |
1600 | return tp->snd_una; | |
1601 | ||
1602 | if (tp->highest_sack == NULL) | |
1603 | return tp->snd_nxt; | |
1604 | ||
1605 | return TCP_SKB_CB(tp->highest_sack)->seq; | |
1606 | } | |
1607 | ||
1608 | static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) | |
1609 | { | |
1610 | tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL : | |
1611 | tcp_write_queue_next(sk, skb); | |
1612 | } | |
1613 | ||
1614 | static inline struct sk_buff *tcp_highest_sack(struct sock *sk) | |
1615 | { | |
1616 | return tcp_sk(sk)->highest_sack; | |
1617 | } | |
1618 | ||
1619 | static inline void tcp_highest_sack_reset(struct sock *sk) | |
1620 | { | |
1621 | tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk); | |
1622 | } | |
1623 | ||
1624 | /* Called when old skb is about to be deleted (to be combined with new skb) */ | |
1625 | static inline void tcp_highest_sack_combine(struct sock *sk, | |
1626 | struct sk_buff *old, | |
1627 | struct sk_buff *new) | |
1628 | { | |
1629 | if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack)) | |
1630 | tcp_sk(sk)->highest_sack = new; | |
1631 | } | |
1632 | ||
1633 | /* This helper checks if socket has IP_TRANSPARENT set */ | |
1634 | static inline bool inet_sk_transparent(const struct sock *sk) | |
1635 | { | |
1636 | switch (sk->sk_state) { | |
1637 | case TCP_TIME_WAIT: | |
1638 | return inet_twsk(sk)->tw_transparent; | |
1639 | case TCP_NEW_SYN_RECV: | |
1640 | return inet_rsk(inet_reqsk(sk))->no_srccheck; | |
1641 | } | |
1642 | return inet_sk(sk)->transparent; | |
1643 | } | |
1644 | ||
1645 | /* Determines whether this is a thin stream (which may suffer from | |
1646 | * increased latency). Used to trigger latency-reducing mechanisms. | |
1647 | */ | |
1648 | static inline bool tcp_stream_is_thin(struct tcp_sock *tp) | |
1649 | { | |
1650 | return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); | |
1651 | } | |
1652 | ||
1653 | /* /proc */ | |
1654 | enum tcp_seq_states { | |
1655 | TCP_SEQ_STATE_LISTENING, | |
1656 | TCP_SEQ_STATE_ESTABLISHED, | |
1657 | }; | |
1658 | ||
1659 | int tcp_seq_open(struct inode *inode, struct file *file); | |
1660 | ||
1661 | struct tcp_seq_afinfo { | |
1662 | char *name; | |
1663 | sa_family_t family; | |
1664 | const struct file_operations *seq_fops; | |
1665 | struct seq_operations seq_ops; | |
1666 | }; | |
1667 | ||
1668 | struct tcp_iter_state { | |
1669 | struct seq_net_private p; | |
1670 | sa_family_t family; | |
1671 | enum tcp_seq_states state; | |
1672 | struct sock *syn_wait_sk; | |
1673 | int bucket, offset, sbucket, num; | |
1674 | loff_t last_pos; | |
1675 | }; | |
1676 | ||
1677 | int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo); | |
1678 | void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo); | |
1679 | ||
1680 | extern struct request_sock_ops tcp_request_sock_ops; | |
1681 | extern struct request_sock_ops tcp6_request_sock_ops; | |
1682 | ||
1683 | void tcp_v4_destroy_sock(struct sock *sk); | |
1684 | ||
1685 | struct sk_buff *tcp_gso_segment(struct sk_buff *skb, | |
1686 | netdev_features_t features); | |
1687 | struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb); | |
1688 | int tcp_gro_complete(struct sk_buff *skb); | |
1689 | ||
1690 | void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); | |
1691 | ||
1692 | static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) | |
1693 | { | |
1694 | struct net *net = sock_net((struct sock *)tp); | |
1695 | return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; | |
1696 | } | |
1697 | ||
1698 | static inline bool tcp_stream_memory_free(const struct sock *sk) | |
1699 | { | |
1700 | const struct tcp_sock *tp = tcp_sk(sk); | |
1701 | u32 notsent_bytes = tp->write_seq - tp->snd_nxt; | |
1702 | ||
1703 | return notsent_bytes < tcp_notsent_lowat(tp); | |
1704 | } | |
1705 | ||
1706 | #ifdef CONFIG_PROC_FS | |
1707 | int tcp4_proc_init(void); | |
1708 | void tcp4_proc_exit(void); | |
1709 | #endif | |
1710 | ||
1711 | int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); | |
1712 | int tcp_conn_request(struct request_sock_ops *rsk_ops, | |
1713 | const struct tcp_request_sock_ops *af_ops, | |
1714 | struct sock *sk, struct sk_buff *skb); | |
1715 | ||
1716 | /* TCP af-specific functions */ | |
1717 | struct tcp_sock_af_ops { | |
1718 | #ifdef CONFIG_TCP_MD5SIG | |
1719 | struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, | |
1720 | const struct sock *addr_sk); | |
1721 | int (*calc_md5_hash)(char *location, | |
1722 | const struct tcp_md5sig_key *md5, | |
1723 | const struct sock *sk, | |
1724 | const struct sk_buff *skb); | |
1725 | int (*md5_parse)(struct sock *sk, | |
1726 | char __user *optval, | |
1727 | int optlen); | |
1728 | #endif | |
1729 | }; | |
1730 | ||
1731 | struct tcp_request_sock_ops { | |
1732 | u16 mss_clamp; | |
1733 | #ifdef CONFIG_TCP_MD5SIG | |
1734 | struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, | |
1735 | const struct sock *addr_sk); | |
1736 | int (*calc_md5_hash) (char *location, | |
1737 | const struct tcp_md5sig_key *md5, | |
1738 | const struct sock *sk, | |
1739 | const struct sk_buff *skb); | |
1740 | #endif | |
1741 | void (*init_req)(struct request_sock *req, | |
1742 | const struct sock *sk_listener, | |
1743 | struct sk_buff *skb); | |
1744 | #ifdef CONFIG_SYN_COOKIES | |
1745 | __u32 (*cookie_init_seq)(const struct sk_buff *skb, | |
1746 | __u16 *mss); | |
1747 | #endif | |
1748 | struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, | |
1749 | const struct request_sock *req, | |
1750 | bool *strict); | |
1751 | __u32 (*init_seq)(const struct sk_buff *skb); | |
1752 | int (*send_synack)(const struct sock *sk, struct dst_entry *dst, | |
1753 | struct flowi *fl, struct request_sock *req, | |
1754 | struct tcp_fastopen_cookie *foc, | |
1755 | enum tcp_synack_type synack_type); | |
1756 | }; | |
1757 | ||
1758 | #ifdef CONFIG_SYN_COOKIES | |
1759 | static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, | |
1760 | const struct sock *sk, struct sk_buff *skb, | |
1761 | __u16 *mss) | |
1762 | { | |
1763 | tcp_synq_overflow(sk); | |
1764 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); | |
1765 | return ops->cookie_init_seq(skb, mss); | |
1766 | } | |
1767 | #else | |
1768 | static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, | |
1769 | const struct sock *sk, struct sk_buff *skb, | |
1770 | __u16 *mss) | |
1771 | { | |
1772 | return 0; | |
1773 | } | |
1774 | #endif | |
1775 | ||
1776 | int tcpv4_offload_init(void); | |
1777 | ||
1778 | void tcp_v4_init(void); | |
1779 | void tcp_init(void); | |
1780 | ||
1781 | /* tcp_recovery.c */ | |
1782 | ||
1783 | /* Flags to enable various loss recovery features. See below */ | |
1784 | extern int sysctl_tcp_recovery; | |
1785 | ||
1786 | /* Use TCP RACK to detect (some) tail and retransmit losses */ | |
1787 | #define TCP_RACK_LOST_RETRANS 0x1 | |
1788 | ||
1789 | extern int tcp_rack_mark_lost(struct sock *sk); | |
1790 | ||
1791 | extern void tcp_rack_advance(struct tcp_sock *tp, | |
1792 | const struct skb_mstamp *xmit_time, u8 sacked); | |
1793 | ||
1794 | /* | |
1795 | * Save and compile IPv4 options, return a pointer to it | |
1796 | */ | |
1797 | static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) | |
1798 | { | |
1799 | const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; | |
1800 | struct ip_options_rcu *dopt = NULL; | |
1801 | ||
1802 | if (opt->optlen) { | |
1803 | int opt_size = sizeof(*dopt) + opt->optlen; | |
1804 | ||
1805 | dopt = kmalloc(opt_size, GFP_ATOMIC); | |
1806 | if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) { | |
1807 | kfree(dopt); | |
1808 | dopt = NULL; | |
1809 | } | |
1810 | } | |
1811 | return dopt; | |
1812 | } | |
1813 | ||
1814 | /* locally generated TCP pure ACKs have skb->truesize == 2 | |
1815 | * (check tcp_send_ack() in net/ipv4/tcp_output.c ) | |
1816 | * This is much faster than dissecting the packet to find out. | |
1817 | * (Think of GRE encapsulations, IPv4, IPv6, ...) | |
1818 | */ | |
1819 | static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) | |
1820 | { | |
1821 | return skb->truesize == 2; | |
1822 | } | |
1823 | ||
1824 | static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) | |
1825 | { | |
1826 | skb->truesize = 2; | |
1827 | } | |
1828 | ||
1829 | static inline int tcp_inq(struct sock *sk) | |
1830 | { | |
1831 | struct tcp_sock *tp = tcp_sk(sk); | |
1832 | int answ; | |
1833 | ||
1834 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { | |
1835 | answ = 0; | |
1836 | } else if (sock_flag(sk, SOCK_URGINLINE) || | |
1837 | !tp->urg_data || | |
1838 | before(tp->urg_seq, tp->copied_seq) || | |
1839 | !before(tp->urg_seq, tp->rcv_nxt)) { | |
1840 | ||
1841 | answ = tp->rcv_nxt - tp->copied_seq; | |
1842 | ||
1843 | /* Subtract 1, if FIN was received */ | |
1844 | if (answ && sock_flag(sk, SOCK_DONE)) | |
1845 | answ--; | |
1846 | } else { | |
1847 | answ = tp->urg_seq - tp->copied_seq; | |
1848 | } | |
1849 | ||
1850 | return answ; | |
1851 | } | |
1852 | ||
1853 | static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) | |
1854 | { | |
1855 | u16 segs_in; | |
1856 | ||
1857 | segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); | |
1858 | tp->segs_in += segs_in; | |
1859 | if (skb->len > tcp_hdrlen(skb)) | |
1860 | tp->data_segs_in += segs_in; | |
1861 | } | |
1862 | ||
1863 | /* | |
1864 | * TCP listen path runs lockless. | |
1865 | * We forced "struct sock" to be const qualified to make sure | |
1866 | * we don't modify one of its field by mistake. | |
1867 | * Here, we increment sk_drops which is an atomic_t, so we can safely | |
1868 | * make sock writable again. | |
1869 | */ | |
1870 | static inline void tcp_listendrop(const struct sock *sk) | |
1871 | { | |
1872 | atomic_inc(&((struct sock *)sk)->sk_drops); | |
1873 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); | |
1874 | } | |
1875 | ||
1876 | #endif /* _TCP_H */ |