]>
Commit | Line | Data |
---|---|---|
2874c5fd | 1 | /* SPDX-License-Identifier: GPL-2.0-or-later */ |
1da177e4 LT |
2 | /* |
3 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
4 | * operating system. INET is implemented using the BSD Socket | |
5 | * interface as the means of communication with the user level. | |
6 | * | |
7 | * Definitions for the TCP module. | |
8 | * | |
9 | * Version: @(#)tcp.h 1.0.5 05/23/93 | |
10 | * | |
02c30a84 | 11 | * Authors: Ross Biro |
1da177e4 | 12 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
1da177e4 LT |
13 | */ |
14 | #ifndef _TCP_H | |
15 | #define _TCP_H | |
16 | ||
1da177e4 LT |
17 | #define FASTRETRANS_DEBUG 1 |
18 | ||
1da177e4 LT |
19 | #include <linux/list.h> |
20 | #include <linux/tcp.h> | |
187f1882 | 21 | #include <linux/bug.h> |
1da177e4 LT |
22 | #include <linux/slab.h> |
23 | #include <linux/cache.h> | |
24 | #include <linux/percpu.h> | |
fb286bb2 | 25 | #include <linux/skbuff.h> |
c6aefafb | 26 | #include <linux/cryptohash.h> |
435cf559 | 27 | #include <linux/kref.h> |
740b0f18 | 28 | #include <linux/ktime.h> |
3f421baa ACM |
29 | |
30 | #include <net/inet_connection_sock.h> | |
295ff7ed | 31 | #include <net/inet_timewait_sock.h> |
77d8bf9c | 32 | #include <net/inet_hashtables.h> |
1da177e4 | 33 | #include <net/checksum.h> |
2e6599cb | 34 | #include <net/request_sock.h> |
40a1227e | 35 | #include <net/sock_reuseport.h> |
1da177e4 LT |
36 | #include <net/sock.h> |
37 | #include <net/snmp.h> | |
38 | #include <net/ip.h> | |
c752f073 | 39 | #include <net/tcp_states.h> |
bdf1ee5d | 40 | #include <net/inet_ecn.h> |
0c266898 | 41 | #include <net/dst.h> |
c752f073 | 42 | |
1da177e4 | 43 | #include <linux/seq_file.h> |
180d8cd9 | 44 | #include <linux/memcontrol.h> |
40304b2a | 45 | #include <linux/bpf-cgroup.h> |
438ac880 | 46 | #include <linux/siphash.h> |
40304b2a | 47 | |
6e04e021 | 48 | extern struct inet_hashinfo tcp_hashinfo; |
1da177e4 | 49 | |
dd24c001 | 50 | extern struct percpu_counter tcp_orphan_count; |
5c9f3023 | 51 | void tcp_time_wait(struct sock *sk, int state, int timeo); |
1da177e4 | 52 | |
bee6548e | 53 | #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER) |
33ad798c | 54 | #define MAX_TCP_OPTION_SPACE 40 |
3b4929f6 ED |
55 | #define TCP_MIN_SND_MSS 48 |
56 | #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE) | |
1da177e4 | 57 | |
105970f6 | 58 | /* |
1da177e4 | 59 | * Never offer a window over 32767 without using window scaling. Some |
105970f6 | 60 | * poor stacks do signed 16bit maths! |
1da177e4 LT |
61 | */ |
62 | #define MAX_TCP_WINDOW 32767U | |
63 | ||
64 | /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ | |
65 | #define TCP_MIN_MSS 88U | |
66 | ||
1555e6fd | 67 | /* The initial MTU to use for probing */ |
dcd8fb85 | 68 | #define TCP_BASE_MSS 1024 |
5d424d5a | 69 | |
05cbc0db FD |
70 | /* probing interval, default to 10 minutes as per RFC4821 */ |
71 | #define TCP_PROBE_INTERVAL 600 | |
72 | ||
6b58e0a5 FD |
73 | /* Specify interval when tcp mtu probing will stop */ |
74 | #define TCP_PROBE_THRESHOLD 8 | |
75 | ||
1da177e4 LT |
76 | /* After receiving this amount of duplicate ACKs fast retransmit starts. */ |
77 | #define TCP_FASTRETRANS_THRESH 3 | |
78 | ||
1da177e4 LT |
79 | /* Maximal number of ACKs sent quickly to accelerate slow-start. */ |
80 | #define TCP_MAX_QUICKACKS 16U | |
81 | ||
589c49cb GF |
82 | /* Maximal number of window scale according to RFC1323 */ |
83 | #define TCP_MAX_WSCALE 14U | |
84 | ||
1da177e4 LT |
85 | /* urg_data states */ |
86 | #define TCP_URG_VALID 0x0100 | |
87 | #define TCP_URG_NOTYET 0x0200 | |
88 | #define TCP_URG_READ 0x0400 | |
89 | ||
90 | #define TCP_RETR1 3 /* | |
91 | * This is how many retries it does before it | |
92 | * tries to figure out if the gateway is | |
93 | * down. Minimal RFC value is 3; it corresponds | |
94 | * to ~3sec-8min depending on RTO. | |
95 | */ | |
96 | ||
97 | #define TCP_RETR2 15 /* | |
98 | * This should take at least | |
99 | * 90 minutes to time out. | |
100 | * RFC1122 says that the limit is 100 sec. | |
101 | * 15 is ~13-30min depending on RTO. | |
102 | */ | |
103 | ||
6c9ff979 AB |
104 | #define TCP_SYN_RETRIES 6 /* This is how many retries are done |
105 | * when active opening a connection. | |
106 | * RFC1122 says the minimum retry MUST | |
107 | * be at least 180secs. Nevertheless | |
108 | * this value is corresponding to | |
109 | * 63secs of retransmission with the | |
110 | * current initial RTO. | |
111 | */ | |
1da177e4 | 112 | |
6c9ff979 AB |
113 | #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done |
114 | * when passive opening a connection. | |
115 | * This is corresponding to 31secs of | |
116 | * retransmission with the current | |
117 | * initial RTO. | |
118 | */ | |
1da177e4 | 119 | |
1da177e4 LT |
120 | #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT |
121 | * state, about 60 seconds */ | |
122 | #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN | |
123 | /* BSD style FIN_WAIT2 deadlock breaker. | |
124 | * It used to be 3min, new value is 60sec, | |
125 | * to combine FIN-WAIT-2 timeout with | |
126 | * TIME-WAIT timer. | |
127 | */ | |
128 | ||
129 | #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ | |
130 | #if HZ >= 100 | |
131 | #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ | |
132 | #define TCP_ATO_MIN ((unsigned)(HZ/25)) | |
133 | #else | |
134 | #define TCP_DELACK_MIN 4U | |
135 | #define TCP_ATO_MIN 4U | |
136 | #endif | |
137 | #define TCP_RTO_MAX ((unsigned)(120*HZ)) | |
138 | #define TCP_RTO_MIN ((unsigned)(HZ/5)) | |
bb4d991a | 139 | #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */ |
fd4f2cea | 140 | #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */ |
9ad7c049 JC |
141 | #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now |
142 | * used as a fallback RTO for the | |
143 | * initial data transmission if no | |
144 | * valid RTT sample has been acquired, | |
145 | * most likely due to retrans in 3WHS. | |
146 | */ | |
1da177e4 LT |
147 | |
148 | #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes | |
149 | * for local resources. | |
150 | */ | |
1da177e4 LT |
151 | #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ |
152 | #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ | |
153 | #define TCP_KEEPALIVE_INTVL (75*HZ) | |
154 | ||
155 | #define MAX_TCP_KEEPIDLE 32767 | |
156 | #define MAX_TCP_KEEPINTVL 32767 | |
157 | #define MAX_TCP_KEEPCNT 127 | |
158 | #define MAX_TCP_SYNCNT 127 | |
159 | ||
160 | #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ | |
1da177e4 LT |
161 | |
162 | #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) | |
163 | #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated | |
164 | * after this time. It should be equal | |
165 | * (or greater than) TCP_TIMEWAIT_LEN | |
166 | * to provide reliability equal to one | |
167 | * provided by timewait state. | |
168 | */ | |
169 | #define TCP_PAWS_WINDOW 1 /* Replay window for per-host | |
170 | * timestamps. It must be less than | |
171 | * minimal timewait lifetime. | |
172 | */ | |
1da177e4 LT |
173 | /* |
174 | * TCP option | |
175 | */ | |
105970f6 | 176 | |
1da177e4 LT |
177 | #define TCPOPT_NOP 1 /* Padding */ |
178 | #define TCPOPT_EOL 0 /* End of options */ | |
179 | #define TCPOPT_MSS 2 /* Segment size negotiating */ | |
180 | #define TCPOPT_WINDOW 3 /* Window scaling */ | |
181 | #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ | |
182 | #define TCPOPT_SACK 5 /* SACK Block */ | |
183 | #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ | |
cfb6eeb4 | 184 | #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */ |
7f9b838b | 185 | #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */ |
2100c8d2 YC |
186 | #define TCPOPT_EXP 254 /* Experimental */ |
187 | /* Magic number to be after the option value for sharing TCP | |
188 | * experimental options. See draft-ietf-tcpm-experimental-options-00.txt | |
189 | */ | |
190 | #define TCPOPT_FASTOPEN_MAGIC 0xF989 | |
60e2a778 | 191 | #define TCPOPT_SMC_MAGIC 0xE2D4C3D9 |
1da177e4 LT |
192 | |
193 | /* | |
194 | * TCP option lengths | |
195 | */ | |
196 | ||
197 | #define TCPOLEN_MSS 4 | |
198 | #define TCPOLEN_WINDOW 3 | |
199 | #define TCPOLEN_SACK_PERM 2 | |
200 | #define TCPOLEN_TIMESTAMP 10 | |
cfb6eeb4 | 201 | #define TCPOLEN_MD5SIG 18 |
7f9b838b | 202 | #define TCPOLEN_FASTOPEN_BASE 2 |
2100c8d2 | 203 | #define TCPOLEN_EXP_FASTOPEN_BASE 4 |
60e2a778 | 204 | #define TCPOLEN_EXP_SMC_BASE 6 |
1da177e4 LT |
205 | |
206 | /* But this is what stacks really send out. */ | |
207 | #define TCPOLEN_TSTAMP_ALIGNED 12 | |
208 | #define TCPOLEN_WSCALE_ALIGNED 4 | |
209 | #define TCPOLEN_SACKPERM_ALIGNED 4 | |
210 | #define TCPOLEN_SACK_BASE 2 | |
211 | #define TCPOLEN_SACK_BASE_ALIGNED 4 | |
212 | #define TCPOLEN_SACK_PERBLOCK 8 | |
cfb6eeb4 | 213 | #define TCPOLEN_MD5SIG_ALIGNED 20 |
33ad798c | 214 | #define TCPOLEN_MSS_ALIGNED 4 |
60e2a778 | 215 | #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8 |
1da177e4 | 216 | |
1da177e4 LT |
217 | /* Flags in tp->nonagle */ |
218 | #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ | |
219 | #define TCP_NAGLE_CORK 2 /* Socket is corked */ | |
caa20d9a | 220 | #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */ |
1da177e4 | 221 | |
36e31b0a AP |
222 | /* TCP thin-stream limits */ |
223 | #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */ | |
224 | ||
21603fc4 | 225 | /* TCP initial congestion window as per rfc6928 */ |
442b9635 DM |
226 | #define TCP_INIT_CWND 10 |
227 | ||
cf60af03 YC |
228 | /* Bit Flags for sysctl_tcp_fastopen */ |
229 | #define TFO_CLIENT_ENABLE 1 | |
10467163 | 230 | #define TFO_SERVER_ENABLE 2 |
67da22d2 | 231 | #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */ |
cf60af03 | 232 | |
10467163 JC |
233 | /* Accept SYN data w/o any cookie option */ |
234 | #define TFO_SERVER_COOKIE_NOT_REQD 0x200 | |
235 | ||
236 | /* Force enable TFO on all listeners, i.e., not requiring the | |
cebc5cba | 237 | * TCP_FASTOPEN socket option. |
10467163 JC |
238 | */ |
239 | #define TFO_SERVER_WO_SOCKOPT1 0x400 | |
10467163 | 240 | |
295ff7ed | 241 | |
1da177e4 | 242 | /* sysctl variables for tcp */ |
1da177e4 | 243 | extern int sysctl_tcp_max_orphans; |
a4fe34bf | 244 | extern long sysctl_tcp_mem[3]; |
e20223f1 | 245 | |
a0370b3f | 246 | #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */ |
1f255691 | 247 | #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */ |
20b654df | 248 | #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */ |
a0370b3f | 249 | |
8d987e5c | 250 | extern atomic_long_t tcp_memory_allocated; |
1748376b | 251 | extern struct percpu_counter tcp_sockets_allocated; |
06044751 | 252 | extern unsigned long tcp_memory_pressure; |
1da177e4 | 253 | |
b8da51eb ED |
254 | /* optimized version of sk_under_memory_pressure() for TCP sockets */ |
255 | static inline bool tcp_under_memory_pressure(const struct sock *sk) | |
256 | { | |
baac50bb JW |
257 | if (mem_cgroup_sockets_enabled && sk->sk_memcg && |
258 | mem_cgroup_under_socket_pressure(sk->sk_memcg)) | |
e805605c | 259 | return true; |
b8da51eb | 260 | |
1f142c17 | 261 | return READ_ONCE(tcp_memory_pressure); |
b8da51eb | 262 | } |
1da177e4 LT |
263 | /* |
264 | * The next routines deal with comparing 32 bit unsigned ints | |
265 | * and worry about wraparound (automatic with unsigned arithmetic). | |
266 | */ | |
267 | ||
a2a385d6 | 268 | static inline bool before(__u32 seq1, __u32 seq2) |
1da177e4 | 269 | { |
0d630cc0 | 270 | return (__s32)(seq1-seq2) < 0; |
1da177e4 | 271 | } |
9a036b9c | 272 | #define after(seq2, seq1) before(seq1, seq2) |
1da177e4 LT |
273 | |
274 | /* is s2<=s1<=s3 ? */ | |
a2a385d6 | 275 | static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3) |
1da177e4 LT |
276 | { |
277 | return seq3 - seq2 >= seq1 - seq2; | |
278 | } | |
279 | ||
efcdbf24 AS |
280 | static inline bool tcp_out_of_memory(struct sock *sk) |
281 | { | |
282 | if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF && | |
283 | sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2)) | |
284 | return true; | |
285 | return false; | |
286 | } | |
287 | ||
a6c5ea4c ED |
288 | void sk_forced_mem_schedule(struct sock *sk, int size); |
289 | ||
ad1af0fe | 290 | static inline bool tcp_too_many_orphans(struct sock *sk, int shift) |
e4fd5da3 | 291 | { |
ad1af0fe DM |
292 | struct percpu_counter *ocp = sk->sk_prot->orphan_count; |
293 | int orphans = percpu_counter_read_positive(ocp); | |
294 | ||
295 | if (orphans << shift > sysctl_tcp_max_orphans) { | |
296 | orphans = percpu_counter_sum_positive(ocp); | |
297 | if (orphans << shift > sysctl_tcp_max_orphans) | |
298 | return true; | |
299 | } | |
ad1af0fe | 300 | return false; |
e4fd5da3 | 301 | } |
1da177e4 | 302 | |
5c9f3023 | 303 | bool tcp_check_oom(struct sock *sk, int shift); |
efcdbf24 | 304 | |
a0f82f64 | 305 | |
1da177e4 LT |
306 | extern struct proto tcp_prot; |
307 | ||
57ef42d5 | 308 | #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field) |
13415e46 | 309 | #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field) |
57ef42d5 | 310 | #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field) |
aa2ea058 | 311 | #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val) |
1da177e4 | 312 | |
5c9f3023 JP |
313 | void tcp_tasklet_init(void); |
314 | ||
32bbd879 | 315 | int tcp_v4_err(struct sk_buff *skb, u32); |
5c9f3023 JP |
316 | |
317 | void tcp_shutdown(struct sock *sk, int how); | |
318 | ||
7487449c | 319 | int tcp_v4_early_demux(struct sk_buff *skb); |
5c9f3023 JP |
320 | int tcp_v4_rcv(struct sk_buff *skb); |
321 | ||
322 | int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); | |
1b784140 | 323 | int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size); |
306b13eb | 324 | int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size); |
5c9f3023 JP |
325 | int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size, |
326 | int flags); | |
306b13eb TH |
327 | int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset, |
328 | size_t size, int flags); | |
e3b5616a DW |
329 | ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset, |
330 | size_t size, int flags); | |
5c9f3023 JP |
331 | void tcp_release_cb(struct sock *sk); |
332 | void tcp_wfree(struct sk_buff *skb); | |
333 | void tcp_write_timer_handler(struct sock *sk); | |
334 | void tcp_delack_timer_handler(struct sock *sk); | |
335 | int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg); | |
72ab4a86 | 336 | int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb); |
3d97d88e | 337 | void tcp_rcv_established(struct sock *sk, struct sk_buff *skb); |
5c9f3023 | 338 | void tcp_rcv_space_adjust(struct sock *sk); |
5c9f3023 JP |
339 | int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp); |
340 | void tcp_twsk_destructor(struct sock *sk); | |
341 | ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos, | |
342 | struct pipe_inode_info *pipe, size_t len, | |
343 | unsigned int flags); | |
9c55e01c | 344 | |
a0496ef2 | 345 | void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks); |
463c84b9 ACM |
346 | static inline void tcp_dec_quickack_mode(struct sock *sk, |
347 | const unsigned int pkts) | |
1da177e4 | 348 | { |
463c84b9 | 349 | struct inet_connection_sock *icsk = inet_csk(sk); |
fc6415bc | 350 | |
463c84b9 ACM |
351 | if (icsk->icsk_ack.quick) { |
352 | if (pkts >= icsk->icsk_ack.quick) { | |
353 | icsk->icsk_ack.quick = 0; | |
fc6415bc | 354 | /* Leaving quickack mode we deflate ATO. */ |
463c84b9 | 355 | icsk->icsk_ack.ato = TCP_ATO_MIN; |
fc6415bc | 356 | } else |
463c84b9 | 357 | icsk->icsk_ack.quick -= pkts; |
1da177e4 LT |
358 | } |
359 | } | |
360 | ||
bdf1ee5d IJ |
361 | #define TCP_ECN_OK 1 |
362 | #define TCP_ECN_QUEUE_CWR 2 | |
363 | #define TCP_ECN_DEMAND_CWR 4 | |
7a269ffa | 364 | #define TCP_ECN_SEEN 8 |
bdf1ee5d | 365 | |
fd2c3ef7 | 366 | enum tcp_tw_status { |
1da177e4 LT |
367 | TCP_TW_SUCCESS = 0, |
368 | TCP_TW_RST = 1, | |
369 | TCP_TW_ACK = 2, | |
370 | TCP_TW_SYN = 3 | |
371 | }; | |
372 | ||
373 | ||
5c9f3023 JP |
374 | enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, |
375 | struct sk_buff *skb, | |
376 | const struct tcphdr *th); | |
377 | struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, | |
e0f9759f ED |
378 | struct request_sock *req, bool fastopen, |
379 | bool *lost_race); | |
5c9f3023 JP |
380 | int tcp_child_process(struct sock *parent, struct sock *child, |
381 | struct sk_buff *skb); | |
5ae344c9 | 382 | void tcp_enter_loss(struct sock *sk); |
57dde7f7 | 383 | void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag); |
5c9f3023 JP |
384 | void tcp_clear_retrans(struct tcp_sock *tp); |
385 | void tcp_update_metrics(struct sock *sk); | |
386 | void tcp_init_metrics(struct sock *sk); | |
387 | void tcp_metrics_init(void); | |
d82bae12 | 388 | bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst); |
5c9f3023 JP |
389 | void tcp_close(struct sock *sk, long timeout); |
390 | void tcp_init_sock(struct sock *sk); | |
27204aaa | 391 | void tcp_init_transfer(struct sock *sk, int bpf_op); |
a11e1d43 LT |
392 | __poll_t tcp_poll(struct file *file, struct socket *sock, |
393 | struct poll_table_struct *wait); | |
5c9f3023 JP |
394 | int tcp_getsockopt(struct sock *sk, int level, int optname, |
395 | char __user *optval, int __user *optlen); | |
396 | int tcp_setsockopt(struct sock *sk, int level, int optname, | |
397 | char __user *optval, unsigned int optlen); | |
398 | int compat_tcp_getsockopt(struct sock *sk, int level, int optname, | |
53d3176b | 399 | char __user *optval, int __user *optlen); |
5c9f3023 | 400 | int compat_tcp_setsockopt(struct sock *sk, int level, int optname, |
53d3176b | 401 | char __user *optval, unsigned int optlen); |
5c9f3023 | 402 | void tcp_set_keepalive(struct sock *sk, int val); |
42cb80a2 | 403 | void tcp_syn_ack_timeout(const struct request_sock *req); |
1b784140 YX |
404 | int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock, |
405 | int flags, int *addr_len); | |
d1361840 | 406 | int tcp_set_rcvlowat(struct sock *sk, int val); |
03f45c88 | 407 | void tcp_data_ready(struct sock *sk); |
340a6f3d | 408 | #ifdef CONFIG_MMU |
93ab6cc6 ED |
409 | int tcp_mmap(struct file *file, struct socket *sock, |
410 | struct vm_area_struct *vma); | |
340a6f3d | 411 | #endif |
eed29f17 | 412 | void tcp_parse_options(const struct net *net, const struct sk_buff *skb, |
5c9f3023 JP |
413 | struct tcp_options_received *opt_rx, |
414 | int estab, struct tcp_fastopen_cookie *foc); | |
415 | const u8 *tcp_parse_md5sig_option(const struct tcphdr *th); | |
7d5d5525 | 416 | |
9349d600 PP |
417 | /* |
418 | * BPF SKB-less helpers | |
419 | */ | |
420 | u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph, | |
421 | struct tcphdr *th, u32 *cookie); | |
422 | u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, | |
423 | struct tcphdr *th, u32 *cookie); | |
424 | u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops, | |
425 | const struct tcp_request_sock_ops *af_ops, | |
426 | struct sock *sk, struct tcphdr *th); | |
1da177e4 LT |
427 | /* |
428 | * TCP v4 functions exported for the inet6 API | |
429 | */ | |
430 | ||
5c9f3023 | 431 | void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb); |
4fab9071 | 432 | void tcp_v4_mtu_reduced(struct sock *sk); |
9cf74903 | 433 | void tcp_req_err(struct sock *sk, u32 seq, bool abort); |
5c9f3023 | 434 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb); |
c28c6f04 | 435 | struct sock *tcp_create_openreq_child(const struct sock *sk, |
5c9f3023 JP |
436 | struct request_sock *req, |
437 | struct sk_buff *skb); | |
81164413 | 438 | void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst); |
0c27171e | 439 | struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, |
5c9f3023 | 440 | struct request_sock *req, |
5e0724d0 ED |
441 | struct dst_entry *dst, |
442 | struct request_sock *req_unhash, | |
443 | bool *own_req); | |
5c9f3023 JP |
444 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb); |
445 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len); | |
446 | int tcp_connect(struct sock *sk); | |
b3d05147 ED |
447 | enum tcp_synack_type { |
448 | TCP_SYNACK_NORMAL, | |
449 | TCP_SYNACK_FASTOPEN, | |
450 | TCP_SYNACK_COOKIE, | |
451 | }; | |
5d062de7 | 452 | struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst, |
5c9f3023 | 453 | struct request_sock *req, |
ca6fb065 | 454 | struct tcp_fastopen_cookie *foc, |
b3d05147 | 455 | enum tcp_synack_type synack_type); |
5c9f3023 | 456 | int tcp_disconnect(struct sock *sk, int flags); |
1da177e4 | 457 | |
370816ae | 458 | void tcp_finish_connect(struct sock *sk, struct sk_buff *skb); |
292e8d8c | 459 | int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size); |
63d02d15 | 460 | void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb); |
1da177e4 | 461 | |
1da177e4 | 462 | /* From syncookies.c */ |
b80c0e78 ED |
463 | struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb, |
464 | struct request_sock *req, | |
84b114b9 | 465 | struct dst_entry *dst, u32 tsoff); |
5c9f3023 JP |
466 | int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th, |
467 | u32 cookie); | |
461b74c3 | 468 | struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb); |
e05c82d3 | 469 | #ifdef CONFIG_SYN_COOKIES |
8c27bd75 | 470 | |
63262315 | 471 | /* Syncookies use a monotonic timer which increments every 60 seconds. |
8c27bd75 FW |
472 | * This counter is used both as a hash input and partially encoded into |
473 | * the cookie value. A cookie is only validated further if the delta | |
474 | * between the current counter value and the encoded one is less than this, | |
63262315 | 475 | * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if |
8c27bd75 FW |
476 | * the counter advances immediately after a cookie is generated). |
477 | */ | |
264ea103 ED |
478 | #define MAX_SYNCOOKIE_AGE 2 |
479 | #define TCP_SYNCOOKIE_PERIOD (60 * HZ) | |
480 | #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD) | |
481 | ||
482 | /* syncookies: remember time of last synqueue overflow | |
483 | * But do not dirty this field too often (once per second is enough) | |
3f684b4b | 484 | * It is racy as we do not hold a lock, but race is very minor. |
264ea103 | 485 | */ |
3f684b4b | 486 | static inline void tcp_synq_overflow(const struct sock *sk) |
264ea103 | 487 | { |
40a1227e | 488 | unsigned int last_overflow; |
cca9bab1 | 489 | unsigned int now = jiffies; |
264ea103 | 490 | |
40a1227e MKL |
491 | if (sk->sk_reuseport) { |
492 | struct sock_reuseport *reuse; | |
493 | ||
494 | reuse = rcu_dereference(sk->sk_reuseport_cb); | |
495 | if (likely(reuse)) { | |
496 | last_overflow = READ_ONCE(reuse->synq_overflow_ts); | |
e8d61985 GN |
497 | if (!time_between32(now, last_overflow, |
498 | last_overflow + HZ)) | |
40a1227e MKL |
499 | WRITE_ONCE(reuse->synq_overflow_ts, now); |
500 | return; | |
501 | } | |
502 | } | |
503 | ||
7ae651a3 | 504 | last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); |
e8d61985 | 505 | if (!time_between32(now, last_overflow, last_overflow + HZ)) |
7ae651a3 | 506 | WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now); |
264ea103 ED |
507 | } |
508 | ||
509 | /* syncookies: no recent synqueue overflow on this listening socket? */ | |
510 | static inline bool tcp_synq_no_recent_overflow(const struct sock *sk) | |
511 | { | |
40a1227e | 512 | unsigned int last_overflow; |
cca9bab1 | 513 | unsigned int now = jiffies; |
264ea103 | 514 | |
40a1227e MKL |
515 | if (sk->sk_reuseport) { |
516 | struct sock_reuseport *reuse; | |
517 | ||
518 | reuse = rcu_dereference(sk->sk_reuseport_cb); | |
519 | if (likely(reuse)) { | |
520 | last_overflow = READ_ONCE(reuse->synq_overflow_ts); | |
01c1c4b7 GN |
521 | return !time_between32(now, last_overflow - HZ, |
522 | last_overflow + | |
523 | TCP_SYNCOOKIE_VALID); | |
40a1227e MKL |
524 | } |
525 | } | |
526 | ||
7ae651a3 | 527 | last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp); |
01c1c4b7 GN |
528 | |
529 | /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID, | |
530 | * then we're under synflood. However, we have to use | |
531 | * 'last_overflow - HZ' as lower bound. That's because a concurrent | |
532 | * tcp_synq_overflow() could update .ts_recent_stamp after we read | |
533 | * jiffies but before we store .ts_recent_stamp into last_overflow, | |
534 | * which could lead to rejecting a valid syncookie. | |
535 | */ | |
536 | return !time_between32(now, last_overflow - HZ, | |
537 | last_overflow + TCP_SYNCOOKIE_VALID); | |
264ea103 | 538 | } |
8c27bd75 FW |
539 | |
540 | static inline u32 tcp_cookie_time(void) | |
541 | { | |
63262315 ED |
542 | u64 val = get_jiffies_64(); |
543 | ||
264ea103 | 544 | do_div(val, TCP_SYNCOOKIE_PERIOD); |
63262315 | 545 | return val; |
8c27bd75 FW |
546 | } |
547 | ||
5c9f3023 JP |
548 | u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th, |
549 | u16 *mssp); | |
3f684b4b | 550 | __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss); |
9a568de4 | 551 | u64 cookie_init_timestamp(struct request_sock *req); |
f9301034 ED |
552 | bool cookie_timestamp_decode(const struct net *net, |
553 | struct tcp_options_received *opt); | |
f1673381 | 554 | bool cookie_ecn_ok(const struct tcp_options_received *opt, |
f7b3bec6 | 555 | const struct net *net, const struct dst_entry *dst); |
4dfc2817 | 556 | |
c6aefafb | 557 | /* From net/ipv6/syncookies.c */ |
5c9f3023 JP |
558 | int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th, |
559 | u32 cookie); | |
560 | struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb); | |
f1673381 | 561 | |
5c9f3023 JP |
562 | u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph, |
563 | const struct tcphdr *th, u16 *mssp); | |
3f684b4b | 564 | __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss); |
e05c82d3 | 565 | #endif |
1da177e4 LT |
566 | /* tcp_output.c */ |
567 | ||
5c9f3023 JP |
568 | void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss, |
569 | int nonagle); | |
10d3be56 ED |
570 | int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); |
571 | int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs); | |
5c9f3023 JP |
572 | void tcp_retransmit_timer(struct sock *sk); |
573 | void tcp_xmit_retransmit_queue(struct sock *); | |
574 | void tcp_simple_retransmit(struct sock *); | |
57dde7f7 | 575 | void tcp_enter_recovery(struct sock *sk, bool ece_ack); |
5c9f3023 | 576 | int tcp_trim_head(struct sock *, struct sk_buff *, u32); |
75c119af ED |
577 | enum tcp_queue { |
578 | TCP_FRAG_IN_WRITE_QUEUE, | |
579 | TCP_FRAG_IN_RTX_QUEUE, | |
580 | }; | |
581 | int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue, | |
582 | struct sk_buff *skb, u32 len, | |
583 | unsigned int mss_now, gfp_t gfp); | |
5c9f3023 JP |
584 | |
585 | void tcp_send_probe0(struct sock *); | |
586 | void tcp_send_partial(struct sock *); | |
e520af48 | 587 | int tcp_write_wakeup(struct sock *, int mib); |
5c9f3023 JP |
588 | void tcp_send_fin(struct sock *sk); |
589 | void tcp_send_active_reset(struct sock *sk, gfp_t priority); | |
590 | int tcp_send_synack(struct sock *); | |
5c9f3023 | 591 | void tcp_push_one(struct sock *, unsigned int mss_now); |
27cde44a | 592 | void __tcp_send_ack(struct sock *sk, u32 rcv_nxt); |
5c9f3023 JP |
593 | void tcp_send_ack(struct sock *sk); |
594 | void tcp_send_delayed_ack(struct sock *sk); | |
595 | void tcp_send_loss_probe(struct sock *sk); | |
ed66dfaf | 596 | bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto); |
cfea5a68 MKL |
597 | void tcp_skb_collapse_tstamp(struct sk_buff *skb, |
598 | const struct sk_buff *next_skb); | |
1da177e4 | 599 | |
a762a980 | 600 | /* tcp_input.c */ |
5c9f3023 | 601 | void tcp_rearm_rto(struct sock *sk); |
0f1c28ae | 602 | void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req); |
5c9f3023 | 603 | void tcp_reset(struct sock *sk); |
4f41b1c5 | 604 | void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb); |
e3e17b77 | 605 | void tcp_fin(struct sock *sk); |
a762a980 | 606 | |
1da177e4 | 607 | /* tcp_timer.c */ |
5c9f3023 | 608 | void tcp_init_xmit_timers(struct sock *); |
463c84b9 ACM |
609 | static inline void tcp_clear_xmit_timers(struct sock *sk) |
610 | { | |
73a6bab5 | 611 | if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1) |
cf0dd203 | 612 | __sock_put(sk); |
73a6bab5 | 613 | |
5d9f4262 ED |
614 | if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1) |
615 | __sock_put(sk); | |
616 | ||
463c84b9 ACM |
617 | inet_csk_clear_xmit_timers(sk); |
618 | } | |
1da177e4 | 619 | |
5c9f3023 JP |
620 | unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); |
621 | unsigned int tcp_current_mss(struct sock *sk); | |
94695e5e | 622 | u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when); |
0c54b85f IJ |
623 | |
624 | /* Bound MSS / TSO packet size with the half of the window */ | |
625 | static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize) | |
626 | { | |
01f83d69 AK |
627 | int cutoff; |
628 | ||
629 | /* When peer uses tiny windows, there is no use in packetizing | |
630 | * to sub-MSS pieces for the sake of SWS or making sure there | |
631 | * are enough packets in the pipe for fast recovery. | |
632 | * | |
633 | * On the other hand, for extremely large MSS devices, handling | |
634 | * smaller than MSS windows in this way does make sense. | |
635 | */ | |
2631b79f | 636 | if (tp->max_window > TCP_MSS_DEFAULT) |
01f83d69 AK |
637 | cutoff = (tp->max_window >> 1); |
638 | else | |
639 | cutoff = tp->max_window; | |
640 | ||
641 | if (cutoff && pktsize > cutoff) | |
642 | return max_t(int, cutoff, 68U - tp->tcp_header_len); | |
0c54b85f IJ |
643 | else |
644 | return pktsize; | |
645 | } | |
1da177e4 | 646 | |
17b085ea | 647 | /* tcp.c */ |
0df48c26 | 648 | void tcp_get_info(struct sock *, struct tcp_info *); |
1da177e4 LT |
649 | |
650 | /* Read 'sendfile()'-style from a TCP socket */ | |
5c9f3023 JP |
651 | int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, |
652 | sk_read_actor_t recv_actor); | |
1da177e4 | 653 | |
5c9f3023 | 654 | void tcp_initialize_rcv_mss(struct sock *sk); |
1da177e4 | 655 | |
5c9f3023 JP |
656 | int tcp_mtu_to_mss(struct sock *sk, int pmtu); |
657 | int tcp_mss_to_mtu(struct sock *sk, int mss); | |
658 | void tcp_mtup_init(struct sock *sk); | |
659 | void tcp_init_buffer_space(struct sock *sk); | |
5d424d5a | 660 | |
f1ecd5d9 DL |
661 | static inline void tcp_bound_rto(const struct sock *sk) |
662 | { | |
663 | if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX) | |
664 | inet_csk(sk)->icsk_rto = TCP_RTO_MAX; | |
665 | } | |
666 | ||
667 | static inline u32 __tcp_set_rto(const struct tcp_sock *tp) | |
668 | { | |
740b0f18 | 669 | return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us); |
f1ecd5d9 DL |
670 | } |
671 | ||
31770e34 FW |
672 | static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) |
673 | { | |
674 | tp->pred_flags = htonl((tp->tcp_header_len << 26) | | |
675 | ntohl(TCP_FLAG_ACK) | | |
676 | snd_wnd); | |
677 | } | |
678 | ||
679 | static inline void tcp_fast_path_on(struct tcp_sock *tp) | |
680 | { | |
681 | __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); | |
682 | } | |
683 | ||
684 | static inline void tcp_fast_path_check(struct sock *sk) | |
685 | { | |
686 | struct tcp_sock *tp = tcp_sk(sk); | |
687 | ||
688 | if (RB_EMPTY_ROOT(&tp->out_of_order_queue) && | |
689 | tp->rcv_wnd && | |
690 | atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && | |
691 | !tp->urg_data) | |
692 | tcp_fast_path_on(tp); | |
693 | } | |
694 | ||
0c266898 SS |
695 | /* Compute the actual rto_min value */ |
696 | static inline u32 tcp_rto_min(struct sock *sk) | |
697 | { | |
cf533ea5 | 698 | const struct dst_entry *dst = __sk_dst_get(sk); |
0c266898 SS |
699 | u32 rto_min = TCP_RTO_MIN; |
700 | ||
701 | if (dst && dst_metric_locked(dst, RTAX_RTO_MIN)) | |
702 | rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN); | |
703 | return rto_min; | |
704 | } | |
705 | ||
740b0f18 ED |
706 | static inline u32 tcp_rto_min_us(struct sock *sk) |
707 | { | |
708 | return jiffies_to_usecs(tcp_rto_min(sk)); | |
709 | } | |
710 | ||
81164413 DB |
711 | static inline bool tcp_ca_dst_locked(const struct dst_entry *dst) |
712 | { | |
713 | return dst_metric_locked(dst, RTAX_CC_ALGO); | |
714 | } | |
715 | ||
f6722583 YC |
716 | /* Minimum RTT in usec. ~0 means not available. */ |
717 | static inline u32 tcp_min_rtt(const struct tcp_sock *tp) | |
718 | { | |
64033892 | 719 | return minmax_get(&tp->rtt_min); |
f6722583 YC |
720 | } |
721 | ||
1da177e4 LT |
722 | /* Compute the actual receive window we are currently advertising. |
723 | * Rcv_nxt can be after the window if our peer push more data | |
724 | * than the offered window. | |
725 | */ | |
40efc6fa | 726 | static inline u32 tcp_receive_window(const struct tcp_sock *tp) |
1da177e4 LT |
727 | { |
728 | s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; | |
729 | ||
730 | if (win < 0) | |
731 | win = 0; | |
732 | return (u32) win; | |
733 | } | |
734 | ||
735 | /* Choose a new window, without checks for shrinking, and without | |
736 | * scaling applied to the result. The caller does these things | |
737 | * if necessary. This is a "raw" window selection. | |
738 | */ | |
5c9f3023 | 739 | u32 __tcp_select_window(struct sock *sk); |
1da177e4 | 740 | |
ee995283 PE |
741 | void tcp_send_window_probe(struct sock *sk); |
742 | ||
ec66eda8 ED |
743 | /* TCP uses 32bit jiffies to save some space. |
744 | * Note that this is different from tcp_time_stamp, which | |
745 | * historically has been the same until linux-4.13. | |
746 | */ | |
747 | #define tcp_jiffies32 ((u32)jiffies) | |
748 | ||
9a568de4 ED |
749 | /* |
750 | * Deliver a 32bit value for TCP timestamp option (RFC 7323) | |
751 | * It is no longer tied to jiffies, but to 1 ms clock. | |
752 | * Note: double check if you want to use tcp_jiffies32 instead of this. | |
753 | */ | |
754 | #define TCP_TS_HZ 1000 | |
755 | ||
756 | static inline u64 tcp_clock_ns(void) | |
757 | { | |
fb420d5d | 758 | return ktime_get_ns(); |
9a568de4 ED |
759 | } |
760 | ||
761 | static inline u64 tcp_clock_us(void) | |
762 | { | |
763 | return div_u64(tcp_clock_ns(), NSEC_PER_USEC); | |
764 | } | |
765 | ||
766 | /* This should only be used in contexts where tp->tcp_mstamp is up to date */ | |
767 | static inline u32 tcp_time_stamp(const struct tcp_sock *tp) | |
768 | { | |
769 | return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ); | |
770 | } | |
771 | ||
772 | /* Could use tcp_clock_us() / 1000, but this version uses a single divide */ | |
773 | static inline u32 tcp_time_stamp_raw(void) | |
774 | { | |
775 | return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ); | |
776 | } | |
777 | ||
9799ccb0 | 778 | void tcp_mstamp_refresh(struct tcp_sock *tp); |
9a568de4 ED |
779 | |
780 | static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0) | |
781 | { | |
782 | return max_t(s64, t1 - t0, 0); | |
783 | } | |
1da177e4 | 784 | |
7faee5c0 ED |
785 | static inline u32 tcp_skb_timestamp(const struct sk_buff *skb) |
786 | { | |
d3edd06e | 787 | return div_u64(skb->skb_mstamp_ns, NSEC_PER_SEC / TCP_TS_HZ); |
7faee5c0 ED |
788 | } |
789 | ||
2fd66ffb ED |
790 | /* provide the departure time in us unit */ |
791 | static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb) | |
792 | { | |
d3edd06e | 793 | return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC); |
2fd66ffb ED |
794 | } |
795 | ||
7faee5c0 | 796 | |
a3433f35 CG |
797 | #define tcp_flag_byte(th) (((u_int8_t *)th)[13]) |
798 | ||
799 | #define TCPHDR_FIN 0x01 | |
800 | #define TCPHDR_SYN 0x02 | |
801 | #define TCPHDR_RST 0x04 | |
802 | #define TCPHDR_PSH 0x08 | |
803 | #define TCPHDR_ACK 0x10 | |
804 | #define TCPHDR_URG 0x20 | |
805 | #define TCPHDR_ECE 0x40 | |
806 | #define TCPHDR_CWR 0x80 | |
807 | ||
49213555 DB |
808 | #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR) |
809 | ||
caa20d9a | 810 | /* This is what the send packet queuing engine uses to pass |
f86586fa ED |
811 | * TCP per-packet control information to the transmission code. |
812 | * We also store the host-order sequence numbers in here too. | |
813 | * This is 44 bytes if IPV6 is enabled. | |
814 | * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately. | |
1da177e4 LT |
815 | */ |
816 | struct tcp_skb_cb { | |
1da177e4 LT |
817 | __u32 seq; /* Starting sequence number */ |
818 | __u32 end_seq; /* SEQ + FIN + SYN + datalen */ | |
cd7d8498 ED |
819 | union { |
820 | /* Note : tcp_tw_isn is used in input path only | |
821 | * (isn chosen by tcp_timewait_state_process()) | |
822 | * | |
f69ad292 ED |
823 | * tcp_gso_segs/size are used in write queue only, |
824 | * cf tcp_skb_pcount()/tcp_skb_mss() | |
cd7d8498 ED |
825 | */ |
826 | __u32 tcp_tw_isn; | |
f69ad292 ED |
827 | struct { |
828 | u16 tcp_gso_segs; | |
829 | u16 tcp_gso_size; | |
830 | }; | |
cd7d8498 | 831 | }; |
4de075e0 | 832 | __u8 tcp_flags; /* TCP header flags. (tcp[13]) */ |
f4f9f6e7 | 833 | |
713bafea | 834 | __u8 sacked; /* State flags for SACK. */ |
1da177e4 LT |
835 | #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ |
836 | #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ | |
837 | #define TCPCB_LOST 0x04 /* SKB is lost */ | |
838 | #define TCPCB_TAGBITS 0x07 /* All tag bits */ | |
d3edd06e | 839 | #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */ |
1da177e4 | 840 | #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ |
9d186cac AV |
841 | #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \ |
842 | TCPCB_REPAIRED) | |
1da177e4 | 843 | |
f4f9f6e7 | 844 | __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */ |
6b084928 | 845 | __u8 txstamp_ack:1, /* Record TX timestamp for ack? */ |
c134ecb8 | 846 | eor:1, /* Is skb MSG_EOR marked? */ |
98aaa913 MM |
847 | has_rxtstamp:1, /* SKB has a RX timestamp */ |
848 | unused:5; | |
1da177e4 | 849 | __u32 ack_seq; /* Sequence number ACK'd */ |
971f10ec | 850 | union { |
b75803d5 | 851 | struct { |
b9f64820 | 852 | /* There is space for up to 24 bytes */ |
d7722e85 SHY |
853 | __u32 in_flight:30,/* Bytes in flight at transmit */ |
854 | is_app_limited:1, /* cwnd not fully used? */ | |
855 | unused:1; | |
b9f64820 YC |
856 | /* pkts S/ACKed so far upon tx of skb, incl retrans: */ |
857 | __u32 delivered; | |
858 | /* start of send pipeline phase */ | |
9a568de4 | 859 | u64 first_tx_mstamp; |
b9f64820 | 860 | /* when we reached the "delivered" count */ |
9a568de4 | 861 | u64 delivered_mstamp; |
b75803d5 LB |
862 | } tx; /* only used for outgoing skbs */ |
863 | union { | |
864 | struct inet_skb_parm h4; | |
971f10ec | 865 | #if IS_ENABLED(CONFIG_IPV6) |
b75803d5 | 866 | struct inet6_skb_parm h6; |
971f10ec | 867 | #endif |
b75803d5 | 868 | } header; /* For incoming skbs */ |
34f79502 | 869 | struct { |
34f79502 | 870 | __u32 flags; |
e5cd3abc | 871 | struct sock *sk_redir; |
8108a775 | 872 | void *data_end; |
34f79502 | 873 | } bpf; |
b75803d5 | 874 | }; |
1da177e4 LT |
875 | }; |
876 | ||
877 | #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) | |
878 | ||
0ea488ff JF |
879 | static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb) |
880 | { | |
881 | TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb); | |
882 | } | |
870c3151 | 883 | |
604326b4 DB |
884 | static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb) |
885 | { | |
886 | return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS; | |
887 | } | |
888 | ||
889 | static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb) | |
890 | { | |
891 | return TCP_SKB_CB(skb)->bpf.sk_redir; | |
892 | } | |
893 | ||
894 | static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb) | |
895 | { | |
896 | TCP_SKB_CB(skb)->bpf.sk_redir = NULL; | |
897 | } | |
898 | ||
815afe17 | 899 | #if IS_ENABLED(CONFIG_IPV6) |
870c3151 ED |
900 | /* This is the variant of inet6_iif() that must be used by TCP, |
901 | * as TCP moves IP6CB into a different location in skb->cb[] | |
902 | */ | |
903 | static inline int tcp_v6_iif(const struct sk_buff *skb) | |
24b711ed DA |
904 | { |
905 | return TCP_SKB_CB(skb)->header.h6.iif; | |
906 | } | |
907 | ||
908 | static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb) | |
870c3151 | 909 | { |
a04a480d | 910 | bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); |
74b20582 DA |
911 | |
912 | return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif; | |
870c3151 | 913 | } |
4297a0ef DA |
914 | |
915 | /* TCP_SKB_CB reference means this can not be used from early demux */ | |
916 | static inline int tcp_v6_sdif(const struct sk_buff *skb) | |
917 | { | |
918 | #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) | |
919 | if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags)) | |
920 | return TCP_SKB_CB(skb)->header.h6.iif; | |
921 | #endif | |
922 | return 0; | |
923 | } | |
815afe17 | 924 | #endif |
870c3151 | 925 | |
a04a480d DA |
926 | static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb) |
927 | { | |
928 | #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) | |
929 | if (!net->ipv4.sysctl_tcp_l3mdev_accept && | |
b4d1605a | 930 | skb && ipv4_l3mdev_skb(IPCB(skb)->flags)) |
a04a480d DA |
931 | return true; |
932 | #endif | |
933 | return false; | |
934 | } | |
935 | ||
3fa6f616 DA |
936 | /* TCP_SKB_CB reference means this can not be used from early demux */ |
937 | static inline int tcp_v4_sdif(struct sk_buff *skb) | |
938 | { | |
939 | #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) | |
940 | if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags)) | |
941 | return TCP_SKB_CB(skb)->header.h4.iif; | |
942 | #endif | |
943 | return 0; | |
944 | } | |
945 | ||
1da177e4 LT |
946 | /* Due to TSO, an SKB can be composed of multiple actual |
947 | * packets. To keep these tracked properly, we use this. | |
bd14b1b2 | 948 | */ |
1da177e4 | 949 | static inline int tcp_skb_pcount(const struct sk_buff *skb) |
bd14b1b2 | 950 | { |
cd7d8498 ED |
951 | return TCP_SKB_CB(skb)->tcp_gso_segs; |
952 | } | |
bd14b1b2 | 953 | |
cd7d8498 ED |
954 | static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs) |
955 | { | |
956 | TCP_SKB_CB(skb)->tcp_gso_segs = segs; | |
bd14b1b2 ED |
957 | } |
958 | ||
cd7d8498 | 959 | static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs) |
1da177e4 | 960 | { |
cd7d8498 | 961 | TCP_SKB_CB(skb)->tcp_gso_segs += segs; |
1da177e4 LT |
962 | } |
963 | ||
f69ad292 | 964 | /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */ |
1da177e4 LT |
965 | static inline int tcp_skb_mss(const struct sk_buff *skb) |
966 | { | |
f69ad292 | 967 | return TCP_SKB_CB(skb)->tcp_gso_size; |
1da177e4 LT |
968 | } |
969 | ||
c134ecb8 MKL |
970 | static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb) |
971 | { | |
972 | return likely(!TCP_SKB_CB(skb)->eor); | |
973 | } | |
974 | ||
317a76f9 SH |
975 | /* Events passed to congestion control interface */ |
976 | enum tcp_ca_event { | |
977 | CA_EVENT_TX_START, /* first transmit when no packets in flight */ | |
978 | CA_EVENT_CWND_RESTART, /* congestion window restart */ | |
979 | CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ | |
317a76f9 | 980 | CA_EVENT_LOSS, /* loss timeout */ |
9890092e FW |
981 | CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */ |
982 | CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */ | |
7354c8c3 FW |
983 | }; |
984 | ||
9890092e | 985 | /* Information about inbound ACK, passed to cong_ops->in_ack_event() */ |
7354c8c3 | 986 | enum tcp_ca_ack_event_flags { |
c1d2b4c3 FW |
987 | CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */ |
988 | CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */ | |
989 | CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */ | |
317a76f9 SH |
990 | }; |
991 | ||
992 | /* | |
993 | * Interface for adding new TCP congestion control handlers | |
994 | */ | |
995 | #define TCP_CA_NAME_MAX 16 | |
3ff825b2 SH |
996 | #define TCP_CA_MAX 128 |
997 | #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX) | |
998 | ||
c5c6a8ab DB |
999 | #define TCP_CA_UNSPEC 0 |
1000 | ||
30e502a3 | 1001 | /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */ |
164891aa | 1002 | #define TCP_CONG_NON_RESTRICTED 0x1 |
30e502a3 DB |
1003 | /* Requires ECN/ECT set on all packets */ |
1004 | #define TCP_CONG_NEEDS_ECN 0x2 | |
164891aa | 1005 | |
64f40ff5 ED |
1006 | union tcp_cc_info; |
1007 | ||
756ee172 LB |
1008 | struct ack_sample { |
1009 | u32 pkts_acked; | |
1010 | s32 rtt_us; | |
6f094b9e | 1011 | u32 in_flight; |
756ee172 LB |
1012 | }; |
1013 | ||
b9f64820 YC |
1014 | /* A rate sample measures the number of (original/retransmitted) data |
1015 | * packets delivered "delivered" over an interval of time "interval_us". | |
1016 | * The tcp_rate.c code fills in the rate sample, and congestion | |
1017 | * control modules that define a cong_control function to run at the end | |
1018 | * of ACK processing can optionally chose to consult this sample when | |
1019 | * setting cwnd and pacing rate. | |
1020 | * A sample is invalid if "delivered" or "interval_us" is negative. | |
1021 | */ | |
1022 | struct rate_sample { | |
9a568de4 | 1023 | u64 prior_mstamp; /* starting timestamp for interval */ |
b9f64820 YC |
1024 | u32 prior_delivered; /* tp->delivered at "prior_mstamp" */ |
1025 | s32 delivered; /* number of packets delivered over interval */ | |
1026 | long interval_us; /* time for tp->delivered to incr "delivered" */ | |
4929c942 DR |
1027 | u32 snd_interval_us; /* snd interval for delivered packets */ |
1028 | u32 rcv_interval_us; /* rcv interval for delivered packets */ | |
b9f64820 YC |
1029 | long rtt_us; /* RTT of last (S)ACKed packet (or -1) */ |
1030 | int losses; /* number of packets marked lost upon ACK */ | |
1031 | u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */ | |
1032 | u32 prior_in_flight; /* in flight before this ACK */ | |
d7722e85 | 1033 | bool is_app_limited; /* is sample from packet with bubble in pipe? */ |
b9f64820 | 1034 | bool is_retrans; /* is sample from retransmission? */ |
e4286603 | 1035 | bool is_ack_delayed; /* is this (likely) a delayed ACK? */ |
b9f64820 YC |
1036 | }; |
1037 | ||
317a76f9 SH |
1038 | struct tcp_congestion_ops { |
1039 | struct list_head list; | |
c5c6a8ab DB |
1040 | u32 key; |
1041 | u32 flags; | |
317a76f9 SH |
1042 | |
1043 | /* initialize private data (optional) */ | |
6687e988 | 1044 | void (*init)(struct sock *sk); |
317a76f9 | 1045 | /* cleanup private data (optional) */ |
6687e988 | 1046 | void (*release)(struct sock *sk); |
317a76f9 SH |
1047 | |
1048 | /* return slow start threshold (required) */ | |
6687e988 | 1049 | u32 (*ssthresh)(struct sock *sk); |
317a76f9 | 1050 | /* do new cwnd calculation (required) */ |
24901551 | 1051 | void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked); |
317a76f9 | 1052 | /* call before changing ca_state (optional) */ |
6687e988 | 1053 | void (*set_state)(struct sock *sk, u8 new_state); |
317a76f9 | 1054 | /* call when cwnd event occurs (optional) */ |
6687e988 | 1055 | void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev); |
7354c8c3 FW |
1056 | /* call when ack arrives (optional) */ |
1057 | void (*in_ack_event)(struct sock *sk, u32 flags); | |
1e0ce2a1 | 1058 | /* new value of cwnd after loss (required) */ |
6687e988 | 1059 | u32 (*undo_cwnd)(struct sock *sk); |
317a76f9 | 1060 | /* hook for packet ack accounting (optional) */ |
756ee172 | 1061 | void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample); |
dcb8c9b4 ED |
1062 | /* override sysctl_tcp_min_tso_segs */ |
1063 | u32 (*min_tso_segs)(struct sock *sk); | |
77bfc174 YC |
1064 | /* returns the multiplier used in tcp_sndbuf_expand (optional) */ |
1065 | u32 (*sndbuf_expand)(struct sock *sk); | |
c0402760 YC |
1066 | /* call when packets are delivered to update cwnd and pacing rate, |
1067 | * after all the ca_state processing. (optional) | |
1068 | */ | |
1069 | void (*cong_control)(struct sock *sk, const struct rate_sample *rs); | |
73c1f4a0 | 1070 | /* get info for inet_diag (optional) */ |
64f40ff5 ED |
1071 | size_t (*get_info)(struct sock *sk, u32 ext, int *attr, |
1072 | union tcp_cc_info *info); | |
317a76f9 SH |
1073 | |
1074 | char name[TCP_CA_NAME_MAX]; | |
1075 | struct module *owner; | |
1076 | }; | |
1077 | ||
5c9f3023 JP |
1078 | int tcp_register_congestion_control(struct tcp_congestion_ops *type); |
1079 | void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); | |
317a76f9 | 1080 | |
55d8694f | 1081 | void tcp_assign_congestion_control(struct sock *sk); |
5c9f3023 JP |
1082 | void tcp_init_congestion_control(struct sock *sk); |
1083 | void tcp_cleanup_congestion_control(struct sock *sk); | |
6670e152 SH |
1084 | int tcp_set_default_congestion_control(struct net *net, const char *name); |
1085 | void tcp_get_default_congestion_control(struct net *net, char *name); | |
5c9f3023 JP |
1086 | void tcp_get_available_congestion_control(char *buf, size_t len); |
1087 | void tcp_get_allowed_congestion_control(char *buf, size_t len); | |
1088 | int tcp_set_allowed_congestion_control(char *allowed); | |
8d650cde ED |
1089 | int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, |
1090 | bool reinit, bool cap_net_admin); | |
e73ebb08 NC |
1091 | u32 tcp_slow_start(struct tcp_sock *tp, u32 acked); |
1092 | void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked); | |
317a76f9 | 1093 | |
5c9f3023 | 1094 | u32 tcp_reno_ssthresh(struct sock *sk); |
e9799183 | 1095 | u32 tcp_reno_undo_cwnd(struct sock *sk); |
24901551 | 1096 | void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked); |
a8acfbac | 1097 | extern struct tcp_congestion_ops tcp_reno; |
317a76f9 | 1098 | |
c5c6a8ab | 1099 | struct tcp_congestion_ops *tcp_ca_find_key(u32 key); |
6670e152 | 1100 | u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca); |
ea697639 | 1101 | #ifdef CONFIG_INET |
c5c6a8ab | 1102 | char *tcp_ca_get_name_by_key(u32 key, char *buffer); |
ea697639 DB |
1103 | #else |
1104 | static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer) | |
1105 | { | |
1106 | return NULL; | |
1107 | } | |
1108 | #endif | |
c5c6a8ab | 1109 | |
30e502a3 DB |
1110 | static inline bool tcp_ca_needs_ecn(const struct sock *sk) |
1111 | { | |
1112 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
1113 | ||
1114 | return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN; | |
1115 | } | |
1116 | ||
6687e988 | 1117 | static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state) |
317a76f9 | 1118 | { |
6687e988 ACM |
1119 | struct inet_connection_sock *icsk = inet_csk(sk); |
1120 | ||
1121 | if (icsk->icsk_ca_ops->set_state) | |
1122 | icsk->icsk_ca_ops->set_state(sk, ca_state); | |
1123 | icsk->icsk_ca_state = ca_state; | |
317a76f9 SH |
1124 | } |
1125 | ||
6687e988 | 1126 | static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event) |
317a76f9 | 1127 | { |
6687e988 ACM |
1128 | const struct inet_connection_sock *icsk = inet_csk(sk); |
1129 | ||
1130 | if (icsk->icsk_ca_ops->cwnd_event) | |
1131 | icsk->icsk_ca_ops->cwnd_event(sk, event); | |
317a76f9 SH |
1132 | } |
1133 | ||
b9f64820 YC |
1134 | /* From tcp_rate.c */ |
1135 | void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb); | |
1136 | void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb, | |
1137 | struct rate_sample *rs); | |
1138 | void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost, | |
d4761754 | 1139 | bool is_sack_reneg, struct rate_sample *rs); |
d7722e85 | 1140 | void tcp_rate_check_app_limited(struct sock *sk); |
b9f64820 | 1141 | |
e60402d0 IJ |
1142 | /* These functions determine how the current flow behaves in respect of SACK |
1143 | * handling. SACK is negotiated with the peer, and therefore it can vary | |
1144 | * between different flows. | |
1145 | * | |
1146 | * tcp_is_sack - SACK enabled | |
1147 | * tcp_is_reno - No SACK | |
e60402d0 IJ |
1148 | */ |
1149 | static inline int tcp_is_sack(const struct tcp_sock *tp) | |
1150 | { | |
ebeef4bc | 1151 | return likely(tp->rx_opt.sack_ok); |
e60402d0 IJ |
1152 | } |
1153 | ||
a2a385d6 | 1154 | static inline bool tcp_is_reno(const struct tcp_sock *tp) |
e60402d0 IJ |
1155 | { |
1156 | return !tcp_is_sack(tp); | |
1157 | } | |
1158 | ||
83ae4088 IJ |
1159 | static inline unsigned int tcp_left_out(const struct tcp_sock *tp) |
1160 | { | |
1161 | return tp->sacked_out + tp->lost_out; | |
1162 | } | |
1163 | ||
1da177e4 LT |
1164 | /* This determines how many packets are "in the network" to the best |
1165 | * of our knowledge. In many cases it is conservative, but where | |
1166 | * detailed information is available from the receiver (via SACK | |
1167 | * blocks etc.) we can make more aggressive calculations. | |
1168 | * | |
1169 | * Use this for decisions involving congestion control, use just | |
1170 | * tp->packets_out to determine if the send queue is empty or not. | |
1171 | * | |
1172 | * Read this equation as: | |
1173 | * | |
1174 | * "Packets sent once on transmission queue" MINUS | |
1175 | * "Packets left network, but not honestly ACKed yet" PLUS | |
1176 | * "Packets fast retransmitted" | |
1177 | */ | |
40efc6fa | 1178 | static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) |
1da177e4 | 1179 | { |
83ae4088 | 1180 | return tp->packets_out - tcp_left_out(tp) + tp->retrans_out; |
1da177e4 LT |
1181 | } |
1182 | ||
0b6a05c1 IJ |
1183 | #define TCP_INFINITE_SSTHRESH 0x7fffffff |
1184 | ||
071d5080 YC |
1185 | static inline bool tcp_in_slow_start(const struct tcp_sock *tp) |
1186 | { | |
76174004 | 1187 | return tp->snd_cwnd < tp->snd_ssthresh; |
071d5080 YC |
1188 | } |
1189 | ||
0b6a05c1 IJ |
1190 | static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp) |
1191 | { | |
1192 | return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH; | |
1193 | } | |
1194 | ||
684bad11 YC |
1195 | static inline bool tcp_in_cwnd_reduction(const struct sock *sk) |
1196 | { | |
1197 | return (TCPF_CA_CWR | TCPF_CA_Recovery) & | |
1198 | (1 << inet_csk(sk)->icsk_ca_state); | |
1199 | } | |
1200 | ||
1da177e4 | 1201 | /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. |
684bad11 | 1202 | * The exception is cwnd reduction phase, when cwnd is decreasing towards |
1da177e4 LT |
1203 | * ssthresh. |
1204 | */ | |
6687e988 | 1205 | static inline __u32 tcp_current_ssthresh(const struct sock *sk) |
1da177e4 | 1206 | { |
6687e988 | 1207 | const struct tcp_sock *tp = tcp_sk(sk); |
cf533ea5 | 1208 | |
684bad11 | 1209 | if (tcp_in_cwnd_reduction(sk)) |
1da177e4 LT |
1210 | return tp->snd_ssthresh; |
1211 | else | |
1212 | return max(tp->snd_ssthresh, | |
1213 | ((tp->snd_cwnd >> 1) + | |
1214 | (tp->snd_cwnd >> 2))); | |
1215 | } | |
1216 | ||
b9c4595b IJ |
1217 | /* Use define here intentionally to get WARN_ON location shown at the caller */ |
1218 | #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out) | |
1da177e4 | 1219 | |
5ee2c941 | 1220 | void tcp_enter_cwr(struct sock *sk); |
5c9f3023 | 1221 | __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst); |
1da177e4 | 1222 | |
6b5a5c0d NC |
1223 | /* The maximum number of MSS of available cwnd for which TSO defers |
1224 | * sending if not using sysctl_tcp_tso_win_divisor. | |
1225 | */ | |
1226 | static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp) | |
1227 | { | |
1228 | return 3; | |
1229 | } | |
1230 | ||
90840def IJ |
1231 | /* Returns end sequence number of the receiver's advertised window */ |
1232 | static inline u32 tcp_wnd_end(const struct tcp_sock *tp) | |
1233 | { | |
1234 | return tp->snd_una + tp->snd_wnd; | |
1235 | } | |
e114a710 ED |
1236 | |
1237 | /* We follow the spirit of RFC2861 to validate cwnd but implement a more | |
1238 | * flexible approach. The RFC suggests cwnd should not be raised unless | |
ca8a2263 NC |
1239 | * it was fully used previously. And that's exactly what we do in |
1240 | * congestion avoidance mode. But in slow start we allow cwnd to grow | |
1241 | * as long as the application has used half the cwnd. | |
e114a710 ED |
1242 | * Example : |
1243 | * cwnd is 10 (IW10), but application sends 9 frames. | |
1244 | * We allow cwnd to reach 18 when all frames are ACKed. | |
1245 | * This check is safe because it's as aggressive as slow start which already | |
1246 | * risks 100% overshoot. The advantage is that we discourage application to | |
1247 | * either send more filler packets or data to artificially blow up the cwnd | |
1248 | * usage, and allow application-limited process to probe bw more aggressively. | |
e114a710 | 1249 | */ |
24901551 | 1250 | static inline bool tcp_is_cwnd_limited(const struct sock *sk) |
e114a710 ED |
1251 | { |
1252 | const struct tcp_sock *tp = tcp_sk(sk); | |
1253 | ||
ca8a2263 | 1254 | /* If in slow start, ensure cwnd grows to twice what was ACKed. */ |
071d5080 | 1255 | if (tcp_in_slow_start(tp)) |
ca8a2263 NC |
1256 | return tp->snd_cwnd < 2 * tp->max_packets_out; |
1257 | ||
1258 | return tp->is_cwnd_limited; | |
e114a710 | 1259 | } |
f4805ede | 1260 | |
cadefe5f ED |
1261 | /* BBR congestion control needs pacing. |
1262 | * Same remark for SO_MAX_PACING_RATE. | |
1263 | * sch_fq packet scheduler is efficiently handling pacing, | |
1264 | * but is not always installed/used. | |
1265 | * Return true if TCP stack should pace packets itself. | |
1266 | */ | |
1267 | static inline bool tcp_needs_internal_pacing(const struct sock *sk) | |
1268 | { | |
1269 | return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED; | |
1270 | } | |
1271 | ||
3f80e08f ED |
1272 | /* Return in jiffies the delay before one skb is sent. |
1273 | * If @skb is NULL, we look at EDT for next packet being sent on the socket. | |
1274 | */ | |
1275 | static inline unsigned long tcp_pacing_delay(const struct sock *sk, | |
1276 | const struct sk_buff *skb) | |
1277 | { | |
1278 | s64 pacing_delay = skb ? skb->tstamp : tcp_sk(sk)->tcp_wstamp_ns; | |
1279 | ||
1280 | pacing_delay -= tcp_sk(sk)->tcp_clock_cache; | |
1281 | ||
1282 | return pacing_delay > 0 ? nsecs_to_jiffies(pacing_delay) : 0; | |
1283 | } | |
1284 | ||
1285 | static inline void tcp_reset_xmit_timer(struct sock *sk, | |
1286 | const int what, | |
1287 | unsigned long when, | |
1288 | const unsigned long max_when, | |
1289 | const struct sk_buff *skb) | |
1290 | { | |
1291 | inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk, skb), | |
1292 | max_when); | |
1293 | } | |
1294 | ||
21c8fe99 | 1295 | /* Something is really bad, we could not queue an additional packet, |
3f80e08f | 1296 | * because qdisc is full or receiver sent a 0 window, or we are paced. |
21c8fe99 ED |
1297 | * We do not want to add fuel to the fire, or abort too early, |
1298 | * so make sure the timer we arm now is at least 200ms in the future, | |
1299 | * regardless of current icsk_rto value (as it could be ~2ms) | |
1300 | */ | |
1301 | static inline unsigned long tcp_probe0_base(const struct sock *sk) | |
1da177e4 | 1302 | { |
21c8fe99 ED |
1303 | return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN); |
1304 | } | |
9e412ba7 | 1305 | |
21c8fe99 ED |
1306 | /* Variant of inet_csk_rto_backoff() used for zero window probes */ |
1307 | static inline unsigned long tcp_probe0_when(const struct sock *sk, | |
1308 | unsigned long max_when) | |
1309 | { | |
1310 | u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff; | |
1311 | ||
1312 | return (unsigned long)min_t(u64, when, max_when); | |
1313 | } | |
1314 | ||
1315 | static inline void tcp_check_probe_timer(struct sock *sk) | |
1316 | { | |
1317 | if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending) | |
3f80e08f ED |
1318 | tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0, |
1319 | tcp_probe0_base(sk), TCP_RTO_MAX, | |
1320 | NULL); | |
1da177e4 LT |
1321 | } |
1322 | ||
ee7537b6 | 1323 | static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq) |
1da177e4 LT |
1324 | { |
1325 | tp->snd_wl1 = seq; | |
1326 | } | |
1327 | ||
ee7537b6 | 1328 | static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq) |
1da177e4 LT |
1329 | { |
1330 | tp->snd_wl1 = seq; | |
1331 | } | |
1332 | ||
1da177e4 LT |
1333 | /* |
1334 | * Calculate(/check) TCP checksum | |
1335 | */ | |
ba7808ea FD |
1336 | static inline __sum16 tcp_v4_check(int len, __be32 saddr, |
1337 | __be32 daddr, __wsum base) | |
1da177e4 | 1338 | { |
0b13c9bb | 1339 | return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base); |
1da177e4 LT |
1340 | } |
1341 | ||
a2a385d6 | 1342 | static inline bool tcp_checksum_complete(struct sk_buff *skb) |
1da177e4 | 1343 | { |
60476372 | 1344 | return !skb_csum_unnecessary(skb) && |
6ab6dfa6 | 1345 | __skb_checksum_complete(skb); |
1da177e4 LT |
1346 | } |
1347 | ||
c9c33212 | 1348 | bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb); |
ac6e7800 | 1349 | int tcp_filter(struct sock *sk, struct sk_buff *skb); |
5c9f3023 | 1350 | void tcp_set_state(struct sock *sk, int state); |
5c9f3023 | 1351 | void tcp_done(struct sock *sk); |
c1e64e29 LC |
1352 | int tcp_abort(struct sock *sk, int err); |
1353 | ||
40efc6fa | 1354 | static inline void tcp_sack_reset(struct tcp_options_received *rx_opt) |
1da177e4 LT |
1355 | { |
1356 | rx_opt->dsack = 0; | |
1da177e4 LT |
1357 | rx_opt->num_sacks = 0; |
1358 | } | |
1359 | ||
5c9f3023 | 1360 | u32 tcp_default_init_rwnd(u32 mss); |
6f021c62 ED |
1361 | void tcp_cwnd_restart(struct sock *sk, s32 delta); |
1362 | ||
1363 | static inline void tcp_slow_start_after_idle_check(struct sock *sk) | |
1364 | { | |
1b1fc3fd | 1365 | const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops; |
6f021c62 ED |
1366 | struct tcp_sock *tp = tcp_sk(sk); |
1367 | s32 delta; | |
1368 | ||
b510f0d2 | 1369 | if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out || |
1b1fc3fd | 1370 | ca_ops->cong_control) |
6f021c62 | 1371 | return; |
d635fbe2 | 1372 | delta = tcp_jiffies32 - tp->lsndtime; |
6f021c62 ED |
1373 | if (delta > inet_csk(sk)->icsk_rto) |
1374 | tcp_cwnd_restart(sk, delta); | |
1375 | } | |
85f16525 | 1376 | |
1da177e4 | 1377 | /* Determine a window scaling and initial window to offer. */ |
ceef9ab6 ED |
1378 | void tcp_select_initial_window(const struct sock *sk, int __space, |
1379 | __u32 mss, __u32 *rcv_wnd, | |
5c9f3023 JP |
1380 | __u32 *window_clamp, int wscale_ok, |
1381 | __u8 *rcv_wscale, __u32 init_rcv_wnd); | |
1da177e4 | 1382 | |
94f0893e | 1383 | static inline int tcp_win_from_space(const struct sock *sk, int space) |
1da177e4 | 1384 | { |
94f0893e | 1385 | int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale; |
c4836742 GF |
1386 | |
1387 | return tcp_adv_win_scale <= 0 ? | |
1388 | (space>>(-tcp_adv_win_scale)) : | |
1389 | space - (space>>tcp_adv_win_scale); | |
1da177e4 LT |
1390 | } |
1391 | ||
105970f6 | 1392 | /* Note: caller must be prepared to deal with negative returns */ |
1da177e4 LT |
1393 | static inline int tcp_space(const struct sock *sk) |
1394 | { | |
ebb3b78d | 1395 | return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) - |
70c26558 | 1396 | READ_ONCE(sk->sk_backlog.len) - |
1da177e4 | 1397 | atomic_read(&sk->sk_rmem_alloc)); |
105970f6 | 1398 | } |
1da177e4 LT |
1399 | |
1400 | static inline int tcp_full_space(const struct sock *sk) | |
1401 | { | |
ebb3b78d | 1402 | return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf)); |
1da177e4 LT |
1403 | } |
1404 | ||
311224d0 ED |
1405 | /* We provision sk_rcvbuf around 200% of sk_rcvlowat. |
1406 | * If 87.5 % (7/8) of the space has been consumed, we want to override | |
1407 | * SO_RCVLOWAT constraint, since we are receiving skbs with too small | |
1408 | * len/truesize ratio. | |
1409 | */ | |
1410 | static inline bool tcp_rmem_pressure(const struct sock *sk) | |
1411 | { | |
87f73c94 ED |
1412 | int rcvbuf, threshold; |
1413 | ||
1414 | if (tcp_under_memory_pressure(sk)) | |
1415 | return true; | |
1416 | ||
1417 | rcvbuf = READ_ONCE(sk->sk_rcvbuf); | |
1418 | threshold = rcvbuf - (rcvbuf >> 3); | |
311224d0 ED |
1419 | |
1420 | return atomic_read(&sk->sk_rmem_alloc) > threshold; | |
1421 | } | |
1422 | ||
843f4a55 | 1423 | extern void tcp_openreq_init_rwin(struct request_sock *req, |
b1964b5f ED |
1424 | const struct sock *sk_listener, |
1425 | const struct dst_entry *dst); | |
843f4a55 | 1426 | |
5c9f3023 | 1427 | void tcp_enter_memory_pressure(struct sock *sk); |
06044751 | 1428 | void tcp_leave_memory_pressure(struct sock *sk); |
1da177e4 | 1429 | |
1da177e4 LT |
1430 | static inline int keepalive_intvl_when(const struct tcp_sock *tp) |
1431 | { | |
b840d15d NB |
1432 | struct net *net = sock_net((struct sock *)tp); |
1433 | ||
1434 | return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl; | |
1da177e4 LT |
1435 | } |
1436 | ||
1437 | static inline int keepalive_time_when(const struct tcp_sock *tp) | |
1438 | { | |
13b287e8 NB |
1439 | struct net *net = sock_net((struct sock *)tp); |
1440 | ||
1441 | return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time; | |
1da177e4 LT |
1442 | } |
1443 | ||
df19a626 ED |
1444 | static inline int keepalive_probes(const struct tcp_sock *tp) |
1445 | { | |
9bd6861b NB |
1446 | struct net *net = sock_net((struct sock *)tp); |
1447 | ||
1448 | return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes; | |
df19a626 ED |
1449 | } |
1450 | ||
6c37e5de FL |
1451 | static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp) |
1452 | { | |
1453 | const struct inet_connection_sock *icsk = &tp->inet_conn; | |
1454 | ||
70eabf0e ED |
1455 | return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime, |
1456 | tcp_jiffies32 - tp->rcv_tstamp); | |
6c37e5de FL |
1457 | } |
1458 | ||
463c84b9 | 1459 | static inline int tcp_fin_time(const struct sock *sk) |
1da177e4 | 1460 | { |
1e579caa | 1461 | int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout; |
463c84b9 | 1462 | const int rto = inet_csk(sk)->icsk_rto; |
1da177e4 | 1463 | |
463c84b9 ACM |
1464 | if (fin_timeout < (rto << 2) - (rto >> 1)) |
1465 | fin_timeout = (rto << 2) - (rto >> 1); | |
1da177e4 LT |
1466 | |
1467 | return fin_timeout; | |
1468 | } | |
1469 | ||
a2a385d6 ED |
1470 | static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt, |
1471 | int paws_win) | |
1da177e4 | 1472 | { |
c887e6d2 | 1473 | if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win) |
a2a385d6 | 1474 | return true; |
cca9bab1 AB |
1475 | if (unlikely(!time_before32(ktime_get_seconds(), |
1476 | rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))) | |
a2a385d6 | 1477 | return true; |
bc2ce894 ED |
1478 | /* |
1479 | * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0, | |
1480 | * then following tcp messages have valid values. Ignore 0 value, | |
1481 | * or else 'negative' tsval might forbid us to accept their packets. | |
1482 | */ | |
1483 | if (!rx_opt->ts_recent) | |
a2a385d6 ED |
1484 | return true; |
1485 | return false; | |
c887e6d2 IJ |
1486 | } |
1487 | ||
a2a385d6 ED |
1488 | static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt, |
1489 | int rst) | |
c887e6d2 IJ |
1490 | { |
1491 | if (tcp_paws_check(rx_opt, 0)) | |
a2a385d6 | 1492 | return false; |
1da177e4 LT |
1493 | |
1494 | /* RST segments are not recommended to carry timestamp, | |
1495 | and, if they do, it is recommended to ignore PAWS because | |
1496 | "their cleanup function should take precedence over timestamps." | |
1497 | Certainly, it is mistake. It is necessary to understand the reasons | |
1498 | of this constraint to relax it: if peer reboots, clock may go | |
1499 | out-of-sync and half-open connections will not be reset. | |
1500 | Actually, the problem would be not existing if all | |
1501 | the implementations followed draft about maintaining clock | |
1502 | via reboots. Linux-2.2 DOES NOT! | |
1503 | ||
1504 | However, we can relax time bounds for RST segments to MSL. | |
1505 | */ | |
cca9bab1 AB |
1506 | if (rst && !time_before32(ktime_get_seconds(), |
1507 | rx_opt->ts_recent_stamp + TCP_PAWS_MSL)) | |
a2a385d6 ED |
1508 | return false; |
1509 | return true; | |
1da177e4 LT |
1510 | } |
1511 | ||
7970ddc8 ED |
1512 | bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb, |
1513 | int mib_idx, u32 *last_oow_ack_time); | |
032ee423 | 1514 | |
a9c19329 | 1515 | static inline void tcp_mib_init(struct net *net) |
1da177e4 LT |
1516 | { |
1517 | /* See RFC 2012 */ | |
6aef70a8 ED |
1518 | TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1); |
1519 | TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); | |
1520 | TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); | |
1521 | TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1); | |
1da177e4 LT |
1522 | } |
1523 | ||
5af4ec23 | 1524 | /* from STCP */ |
ef9da47c | 1525 | static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp) |
0800f170 | 1526 | { |
6a438bbe | 1527 | tp->lost_skb_hint = NULL; |
ef9da47c IJ |
1528 | } |
1529 | ||
1530 | static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp) | |
1531 | { | |
1532 | tcp_clear_retrans_hints_partial(tp); | |
6a438bbe | 1533 | tp->retransmit_skb_hint = NULL; |
b7689205 IJ |
1534 | } |
1535 | ||
a915da9b ED |
1536 | union tcp_md5_addr { |
1537 | struct in_addr a4; | |
1538 | #if IS_ENABLED(CONFIG_IPV6) | |
1539 | struct in6_addr a6; | |
1540 | #endif | |
1541 | }; | |
1542 | ||
cfb6eeb4 YH |
1543 | /* - key database */ |
1544 | struct tcp_md5sig_key { | |
a915da9b | 1545 | struct hlist_node node; |
cfb6eeb4 | 1546 | u8 keylen; |
a915da9b ED |
1547 | u8 family; /* AF_INET or AF_INET6 */ |
1548 | union tcp_md5_addr addr; | |
6797318e | 1549 | u8 prefixlen; |
a915da9b ED |
1550 | u8 key[TCP_MD5SIG_MAXKEYLEN]; |
1551 | struct rcu_head rcu; | |
cfb6eeb4 YH |
1552 | }; |
1553 | ||
1554 | /* - sock block */ | |
1555 | struct tcp_md5sig_info { | |
a915da9b | 1556 | struct hlist_head head; |
a8afca03 | 1557 | struct rcu_head rcu; |
cfb6eeb4 YH |
1558 | }; |
1559 | ||
1560 | /* - pseudo header */ | |
1561 | struct tcp4_pseudohdr { | |
1562 | __be32 saddr; | |
1563 | __be32 daddr; | |
1564 | __u8 pad; | |
1565 | __u8 protocol; | |
1566 | __be16 len; | |
1567 | }; | |
1568 | ||
1569 | struct tcp6_pseudohdr { | |
1570 | struct in6_addr saddr; | |
1571 | struct in6_addr daddr; | |
1572 | __be32 len; | |
1573 | __be32 protocol; /* including padding */ | |
1574 | }; | |
1575 | ||
1576 | union tcp_md5sum_block { | |
1577 | struct tcp4_pseudohdr ip4; | |
dfd56b8b | 1578 | #if IS_ENABLED(CONFIG_IPV6) |
cfb6eeb4 YH |
1579 | struct tcp6_pseudohdr ip6; |
1580 | #endif | |
1581 | }; | |
1582 | ||
1583 | /* - pool: digest algorithm, hash description and scratch buffer */ | |
1584 | struct tcp_md5sig_pool { | |
cf80e0e4 | 1585 | struct ahash_request *md5_req; |
19689e38 | 1586 | void *scratch; |
cfb6eeb4 YH |
1587 | }; |
1588 | ||
cfb6eeb4 | 1589 | /* - functions */ |
39f8e58e ED |
1590 | int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, |
1591 | const struct sock *sk, const struct sk_buff *skb); | |
5c9f3023 | 1592 | int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, |
6797318e ID |
1593 | int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, |
1594 | gfp_t gfp); | |
5c9f3023 | 1595 | int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, |
6797318e | 1596 | int family, u8 prefixlen); |
b83e3deb | 1597 | struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, |
fd3a154a | 1598 | const struct sock *addr_sk); |
cfb6eeb4 | 1599 | |
9501f972 | 1600 | #ifdef CONFIG_TCP_MD5SIG |
6015c71e | 1601 | #include <linux/jump_label.h> |
921f9a0f | 1602 | extern struct static_key_false tcp_md5_needed; |
6015c71e ED |
1603 | struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, |
1604 | const union tcp_md5_addr *addr, | |
1605 | int family); | |
1606 | static inline struct tcp_md5sig_key * | |
1607 | tcp_md5_do_lookup(const struct sock *sk, | |
1608 | const union tcp_md5_addr *addr, | |
1609 | int family) | |
1610 | { | |
921f9a0f | 1611 | if (!static_branch_unlikely(&tcp_md5_needed)) |
6015c71e ED |
1612 | return NULL; |
1613 | return __tcp_md5_do_lookup(sk, addr, family); | |
1614 | } | |
1615 | ||
a915da9b | 1616 | #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key) |
9501f972 | 1617 | #else |
b83e3deb | 1618 | static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, |
a915da9b ED |
1619 | const union tcp_md5_addr *addr, |
1620 | int family) | |
1621 | { | |
1622 | return NULL; | |
1623 | } | |
9501f972 YH |
1624 | #define tcp_twsk_md5_key(twsk) NULL |
1625 | #endif | |
1626 | ||
5c9f3023 | 1627 | bool tcp_alloc_md5sig_pool(void); |
cfb6eeb4 | 1628 | |
5c9f3023 | 1629 | struct tcp_md5sig_pool *tcp_get_md5sig_pool(void); |
71cea17e ED |
1630 | static inline void tcp_put_md5sig_pool(void) |
1631 | { | |
1632 | local_bh_enable(); | |
1633 | } | |
35790c04 | 1634 | |
5c9f3023 JP |
1635 | int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *, |
1636 | unsigned int header_len); | |
1637 | int tcp_md5_hash_key(struct tcp_md5sig_pool *hp, | |
1638 | const struct tcp_md5sig_key *key); | |
cfb6eeb4 | 1639 | |
10467163 | 1640 | /* From tcp_fastopen.c */ |
5c9f3023 | 1641 | void tcp_fastopen_cache_get(struct sock *sk, u16 *mss, |
7268586b | 1642 | struct tcp_fastopen_cookie *cookie); |
5c9f3023 | 1643 | void tcp_fastopen_cache_set(struct sock *sk, u16 mss, |
2646c831 DL |
1644 | struct tcp_fastopen_cookie *cookie, bool syn_lost, |
1645 | u16 try_exp); | |
783237e8 YC |
1646 | struct tcp_fastopen_request { |
1647 | /* Fast Open cookie. Size 0 means a cookie request */ | |
1648 | struct tcp_fastopen_cookie cookie; | |
1649 | struct msghdr *data; /* data in MSG_FASTOPEN */ | |
f5ddcbbb ED |
1650 | size_t size; |
1651 | int copied; /* queued in tcp_connect() */ | |
f859a448 | 1652 | struct ubuf_info *uarg; |
783237e8 | 1653 | }; |
783237e8 | 1654 | void tcp_free_fastopen_req(struct tcp_sock *tp); |
1fba70e5 | 1655 | void tcp_fastopen_destroy_cipher(struct sock *sk); |
43713848 | 1656 | void tcp_fastopen_ctx_destroy(struct net *net); |
1fba70e5 | 1657 | int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk, |
438ac880 | 1658 | void *primary_key, void *backup_key); |
a674d84e JB |
1659 | int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk, |
1660 | u64 *key); | |
61d2bcae | 1661 | void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb); |
7c85af88 ED |
1662 | struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb, |
1663 | struct request_sock *req, | |
71c02379 CP |
1664 | struct tcp_fastopen_cookie *foc, |
1665 | const struct dst_entry *dst); | |
43713848 | 1666 | void tcp_fastopen_init_key_once(struct net *net); |
065263f4 WW |
1667 | bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss, |
1668 | struct tcp_fastopen_cookie *cookie); | |
19f6d3f3 | 1669 | bool tcp_fastopen_defer_connect(struct sock *sk, int *err); |
438ac880 | 1670 | #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t) |
9092a76d JB |
1671 | #define TCP_FASTOPEN_KEY_MAX 2 |
1672 | #define TCP_FASTOPEN_KEY_BUF_LENGTH \ | |
1673 | (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX) | |
10467163 JC |
1674 | |
1675 | /* Fastopen key context */ | |
1676 | struct tcp_fastopen_context { | |
438ac880 | 1677 | siphash_key_t key[TCP_FASTOPEN_KEY_MAX]; |
c681edae AB |
1678 | int num; |
1679 | struct rcu_head rcu; | |
10467163 JC |
1680 | }; |
1681 | ||
cf1ef3f0 | 1682 | extern unsigned int sysctl_tcp_fastopen_blackhole_timeout; |
46c2fa39 | 1683 | void tcp_fastopen_active_disable(struct sock *sk); |
cf1ef3f0 WW |
1684 | bool tcp_fastopen_active_should_disable(struct sock *sk); |
1685 | void tcp_fastopen_active_disable_ofo_check(struct sock *sk); | |
7268586b | 1686 | void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired); |
cf1ef3f0 | 1687 | |
9092a76d JB |
1688 | /* Caller needs to wrap with rcu_read_(un)lock() */ |
1689 | static inline | |
1690 | struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk) | |
1691 | { | |
1692 | struct tcp_fastopen_context *ctx; | |
1693 | ||
1694 | ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx); | |
1695 | if (!ctx) | |
1696 | ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx); | |
1697 | return ctx; | |
1698 | } | |
1699 | ||
1700 | static inline | |
1701 | bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, | |
1702 | const struct tcp_fastopen_cookie *orig) | |
1703 | { | |
1704 | if (orig->len == TCP_FASTOPEN_COOKIE_SIZE && | |
1705 | orig->len == foc->len && | |
1706 | !memcmp(orig->val, foc->val, foc->len)) | |
1707 | return true; | |
1708 | return false; | |
1709 | } | |
1710 | ||
1711 | static inline | |
1712 | int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx) | |
1713 | { | |
c681edae | 1714 | return ctx->num; |
9092a76d JB |
1715 | } |
1716 | ||
05b055e8 FY |
1717 | /* Latencies incurred by various limits for a sender. They are |
1718 | * chronograph-like stats that are mutually exclusive. | |
1719 | */ | |
1720 | enum tcp_chrono { | |
1721 | TCP_CHRONO_UNSPEC, | |
1722 | TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */ | |
1723 | TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */ | |
1724 | TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */ | |
1725 | __TCP_CHRONO_MAX, | |
1726 | }; | |
1727 | ||
1728 | void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type); | |
1729 | void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type); | |
1730 | ||
e2080072 ED |
1731 | /* This helper is needed, because skb->tcp_tsorted_anchor uses |
1732 | * the same memory storage than skb->destructor/_skb_refdst | |
1733 | */ | |
1734 | static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb) | |
1735 | { | |
1736 | skb->destructor = NULL; | |
1737 | skb->_skb_refdst = 0UL; | |
1738 | } | |
1739 | ||
1740 | #define tcp_skb_tsorted_save(skb) { \ | |
1741 | unsigned long _save = skb->_skb_refdst; \ | |
1742 | skb->_skb_refdst = 0UL; | |
1743 | ||
1744 | #define tcp_skb_tsorted_restore(skb) \ | |
1745 | skb->_skb_refdst = _save; \ | |
1746 | } | |
1747 | ||
ac3f09ba | 1748 | void tcp_write_queue_purge(struct sock *sk); |
fe067e8a | 1749 | |
75c119af ED |
1750 | static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk) |
1751 | { | |
1752 | return skb_rb_first(&sk->tcp_rtx_queue); | |
1753 | } | |
1754 | ||
b617158d ED |
1755 | static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk) |
1756 | { | |
1757 | return skb_rb_last(&sk->tcp_rtx_queue); | |
1758 | } | |
1759 | ||
cf533ea5 | 1760 | static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk) |
fe067e8a | 1761 | { |
cd07a8ea | 1762 | return skb_peek(&sk->sk_write_queue); |
fe067e8a DM |
1763 | } |
1764 | ||
cf533ea5 | 1765 | static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk) |
fe067e8a | 1766 | { |
cd07a8ea | 1767 | return skb_peek_tail(&sk->sk_write_queue); |
fe067e8a DM |
1768 | } |
1769 | ||
234b6860 | 1770 | #define tcp_for_write_queue_from_safe(skb, tmp, sk) \ |
cd07a8ea | 1771 | skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp) |
234b6860 | 1772 | |
cf533ea5 | 1773 | static inline struct sk_buff *tcp_send_head(const struct sock *sk) |
fe067e8a | 1774 | { |
75c119af | 1775 | return skb_peek(&sk->sk_write_queue); |
fe067e8a DM |
1776 | } |
1777 | ||
cd07a8ea DM |
1778 | static inline bool tcp_skb_is_last(const struct sock *sk, |
1779 | const struct sk_buff *skb) | |
1780 | { | |
1781 | return skb_queue_is_last(&sk->sk_write_queue, skb); | |
1782 | } | |
1783 | ||
75c119af | 1784 | static inline bool tcp_write_queue_empty(const struct sock *sk) |
fe067e8a | 1785 | { |
75c119af ED |
1786 | return skb_queue_empty(&sk->sk_write_queue); |
1787 | } | |
1788 | ||
1789 | static inline bool tcp_rtx_queue_empty(const struct sock *sk) | |
1790 | { | |
1791 | return RB_EMPTY_ROOT(&sk->tcp_rtx_queue); | |
1792 | } | |
1793 | ||
1794 | static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk) | |
1795 | { | |
1796 | return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk); | |
fe067e8a DM |
1797 | } |
1798 | ||
fe067e8a DM |
1799 | static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb) |
1800 | { | |
a43e052b | 1801 | __skb_queue_tail(&sk->sk_write_queue, skb); |
fe067e8a DM |
1802 | |
1803 | /* Queue it, remembering where we must start sending. */ | |
50895b9d | 1804 | if (sk->sk_write_queue.next == skb) |
0f87230d | 1805 | tcp_chrono_start(sk, TCP_CHRONO_BUSY); |
fe067e8a DM |
1806 | } |
1807 | ||
43f59c89 | 1808 | /* Insert new before skb on the write queue of sk. */ |
fe067e8a DM |
1809 | static inline void tcp_insert_write_queue_before(struct sk_buff *new, |
1810 | struct sk_buff *skb, | |
1811 | struct sock *sk) | |
1812 | { | |
43f59c89 | 1813 | __skb_queue_before(&sk->sk_write_queue, skb, new); |
fe067e8a DM |
1814 | } |
1815 | ||
1816 | static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk) | |
1817 | { | |
4a269818 | 1818 | tcp_skb_tsorted_anchor_cleanup(skb); |
fe067e8a DM |
1819 | __skb_unlink(skb, &sk->sk_write_queue); |
1820 | } | |
1821 | ||
75c119af ED |
1822 | void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb); |
1823 | ||
1824 | static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk) | |
fe067e8a | 1825 | { |
75c119af ED |
1826 | tcp_skb_tsorted_anchor_cleanup(skb); |
1827 | rb_erase(&skb->rbnode, &sk->tcp_rtx_queue); | |
1828 | } | |
1829 | ||
1830 | static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk) | |
1831 | { | |
1832 | list_del(&skb->tcp_tsorted_anchor); | |
1833 | tcp_rtx_queue_unlink(skb, sk); | |
1834 | sk_wmem_free_skb(sk, skb); | |
fe067e8a DM |
1835 | } |
1836 | ||
12d50c46 KK |
1837 | static inline void tcp_push_pending_frames(struct sock *sk) |
1838 | { | |
1839 | if (tcp_send_head(sk)) { | |
1840 | struct tcp_sock *tp = tcp_sk(sk); | |
1841 | ||
1842 | __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle); | |
1843 | } | |
1844 | } | |
1845 | ||
ecb97192 NC |
1846 | /* Start sequence of the skb just after the highest skb with SACKed |
1847 | * bit, valid only if sacked_out > 0 or when the caller has ensured | |
1848 | * validity by itself. | |
a47e5a98 IJ |
1849 | */ |
1850 | static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp) | |
1851 | { | |
1852 | if (!tp->sacked_out) | |
1853 | return tp->snd_una; | |
6859d494 IJ |
1854 | |
1855 | if (tp->highest_sack == NULL) | |
1856 | return tp->snd_nxt; | |
1857 | ||
a47e5a98 IJ |
1858 | return TCP_SKB_CB(tp->highest_sack)->seq; |
1859 | } | |
1860 | ||
6859d494 IJ |
1861 | static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb) |
1862 | { | |
50895b9d | 1863 | tcp_sk(sk)->highest_sack = skb_rb_next(skb); |
6859d494 IJ |
1864 | } |
1865 | ||
1866 | static inline struct sk_buff *tcp_highest_sack(struct sock *sk) | |
1867 | { | |
1868 | return tcp_sk(sk)->highest_sack; | |
1869 | } | |
1870 | ||
1871 | static inline void tcp_highest_sack_reset(struct sock *sk) | |
1872 | { | |
50895b9d | 1873 | tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk); |
6859d494 IJ |
1874 | } |
1875 | ||
2b7cda9c ED |
1876 | /* Called when old skb is about to be deleted and replaced by new skb */ |
1877 | static inline void tcp_highest_sack_replace(struct sock *sk, | |
6859d494 IJ |
1878 | struct sk_buff *old, |
1879 | struct sk_buff *new) | |
1880 | { | |
2b7cda9c | 1881 | if (old == tcp_highest_sack(sk)) |
6859d494 IJ |
1882 | tcp_sk(sk)->highest_sack = new; |
1883 | } | |
1884 | ||
b1f0a0e9 FW |
1885 | /* This helper checks if socket has IP_TRANSPARENT set */ |
1886 | static inline bool inet_sk_transparent(const struct sock *sk) | |
1887 | { | |
1888 | switch (sk->sk_state) { | |
1889 | case TCP_TIME_WAIT: | |
1890 | return inet_twsk(sk)->tw_transparent; | |
1891 | case TCP_NEW_SYN_RECV: | |
1892 | return inet_rsk(inet_reqsk(sk))->no_srccheck; | |
1893 | } | |
1894 | return inet_sk(sk)->transparent; | |
1895 | } | |
1896 | ||
5aa4b32f AP |
1897 | /* Determines whether this is a thin stream (which may suffer from |
1898 | * increased latency). Used to trigger latency-reducing mechanisms. | |
1899 | */ | |
a2a385d6 | 1900 | static inline bool tcp_stream_is_thin(struct tcp_sock *tp) |
5aa4b32f AP |
1901 | { |
1902 | return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp); | |
1903 | } | |
1904 | ||
1da177e4 LT |
1905 | /* /proc */ |
1906 | enum tcp_seq_states { | |
1907 | TCP_SEQ_STATE_LISTENING, | |
1da177e4 | 1908 | TCP_SEQ_STATE_ESTABLISHED, |
1da177e4 LT |
1909 | }; |
1910 | ||
37d849bb CH |
1911 | void *tcp_seq_start(struct seq_file *seq, loff_t *pos); |
1912 | void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos); | |
1913 | void tcp_seq_stop(struct seq_file *seq, void *v); | |
73cb88ec | 1914 | |
1da177e4 | 1915 | struct tcp_seq_afinfo { |
73cb88ec | 1916 | sa_family_t family; |
1da177e4 LT |
1917 | }; |
1918 | ||
1919 | struct tcp_iter_state { | |
a4146b1b | 1920 | struct seq_net_private p; |
1da177e4 LT |
1921 | enum tcp_seq_states state; |
1922 | struct sock *syn_wait_sk; | |
a7cb5a49 | 1923 | int bucket, offset, sbucket, num; |
a8b690f9 | 1924 | loff_t last_pos; |
1da177e4 LT |
1925 | }; |
1926 | ||
20380731 | 1927 | extern struct request_sock_ops tcp_request_sock_ops; |
c6aefafb | 1928 | extern struct request_sock_ops tcp6_request_sock_ops; |
20380731 | 1929 | |
5c9f3023 | 1930 | void tcp_v4_destroy_sock(struct sock *sk); |
20380731 | 1931 | |
28be6e07 | 1932 | struct sk_buff *tcp_gso_segment(struct sk_buff *skb, |
5c9f3023 | 1933 | netdev_features_t features); |
d4546c25 | 1934 | struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb); |
5c9f3023 | 1935 | int tcp_gro_complete(struct sk_buff *skb); |
28850dc7 | 1936 | |
5c9f3023 | 1937 | void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr); |
f4c50d99 | 1938 | |
c9bee3b7 ED |
1939 | static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp) |
1940 | { | |
4979f2d9 NB |
1941 | struct net *net = sock_net((struct sock *)tp); |
1942 | return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat; | |
c9bee3b7 ED |
1943 | } |
1944 | ||
a74f0fa0 ED |
1945 | /* @wake is one when sk_stream_write_space() calls us. |
1946 | * This sends EPOLLOUT only if notsent_bytes is half the limit. | |
1947 | * This mimics the strategy used in sock_def_write_space(). | |
1948 | */ | |
1949 | static inline bool tcp_stream_memory_free(const struct sock *sk, int wake) | |
c9bee3b7 ED |
1950 | { |
1951 | const struct tcp_sock *tp = tcp_sk(sk); | |
e0d694d6 ED |
1952 | u32 notsent_bytes = READ_ONCE(tp->write_seq) - |
1953 | READ_ONCE(tp->snd_nxt); | |
c9bee3b7 | 1954 | |
a74f0fa0 | 1955 | return (notsent_bytes << wake) < tcp_notsent_lowat(tp); |
c9bee3b7 ED |
1956 | } |
1957 | ||
20380731 | 1958 | #ifdef CONFIG_PROC_FS |
5c9f3023 JP |
1959 | int tcp4_proc_init(void); |
1960 | void tcp4_proc_exit(void); | |
20380731 ACM |
1961 | #endif |
1962 | ||
ea3bea3a | 1963 | int tcp_rtx_synack(const struct sock *sk, struct request_sock *req); |
1fb6f159 OP |
1964 | int tcp_conn_request(struct request_sock_ops *rsk_ops, |
1965 | const struct tcp_request_sock_ops *af_ops, | |
1966 | struct sock *sk, struct sk_buff *skb); | |
5db92c99 | 1967 | |
cfb6eeb4 YH |
1968 | /* TCP af-specific functions */ |
1969 | struct tcp_sock_af_ops { | |
1970 | #ifdef CONFIG_TCP_MD5SIG | |
b83e3deb | 1971 | struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk, |
fd3a154a | 1972 | const struct sock *addr_sk); |
39f8e58e ED |
1973 | int (*calc_md5_hash)(char *location, |
1974 | const struct tcp_md5sig_key *md5, | |
1975 | const struct sock *sk, | |
1976 | const struct sk_buff *skb); | |
1977 | int (*md5_parse)(struct sock *sk, | |
8917a777 | 1978 | int optname, |
39f8e58e ED |
1979 | char __user *optval, |
1980 | int optlen); | |
cfb6eeb4 YH |
1981 | #endif |
1982 | }; | |
1983 | ||
1984 | struct tcp_request_sock_ops { | |
2aec4a29 | 1985 | u16 mss_clamp; |
cfb6eeb4 | 1986 | #ifdef CONFIG_TCP_MD5SIG |
b83e3deb | 1987 | struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk, |
fd3a154a | 1988 | const struct sock *addr_sk); |
39f8e58e ED |
1989 | int (*calc_md5_hash) (char *location, |
1990 | const struct tcp_md5sig_key *md5, | |
1991 | const struct sock *sk, | |
1992 | const struct sk_buff *skb); | |
cfb6eeb4 | 1993 | #endif |
b40cf18e ED |
1994 | void (*init_req)(struct request_sock *req, |
1995 | const struct sock *sk_listener, | |
16bea70a | 1996 | struct sk_buff *skb); |
fb7b37a7 | 1997 | #ifdef CONFIG_SYN_COOKIES |
3f684b4b | 1998 | __u32 (*cookie_init_seq)(const struct sk_buff *skb, |
fb7b37a7 OP |
1999 | __u16 *mss); |
2000 | #endif | |
f964629e | 2001 | struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl, |
4396e461 | 2002 | const struct request_sock *req); |
84b114b9 | 2003 | u32 (*init_seq)(const struct sk_buff *skb); |
5d2ed052 | 2004 | u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb); |
0f935dbe | 2005 | int (*send_synack)(const struct sock *sk, struct dst_entry *dst, |
d6274bd8 | 2006 | struct flowi *fl, struct request_sock *req, |
dc6ef6be | 2007 | struct tcp_fastopen_cookie *foc, |
b3d05147 | 2008 | enum tcp_synack_type synack_type); |
cfb6eeb4 YH |
2009 | }; |
2010 | ||
fb7b37a7 OP |
2011 | #ifdef CONFIG_SYN_COOKIES |
2012 | static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, | |
3f684b4b | 2013 | const struct sock *sk, struct sk_buff *skb, |
fb7b37a7 OP |
2014 | __u16 *mss) |
2015 | { | |
3f684b4b | 2016 | tcp_synq_overflow(sk); |
02a1d6e7 | 2017 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT); |
3f684b4b | 2018 | return ops->cookie_init_seq(skb, mss); |
fb7b37a7 OP |
2019 | } |
2020 | #else | |
2021 | static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops, | |
3f684b4b | 2022 | const struct sock *sk, struct sk_buff *skb, |
fb7b37a7 OP |
2023 | __u16 *mss) |
2024 | { | |
2025 | return 0; | |
2026 | } | |
2027 | #endif | |
2028 | ||
5c9f3023 | 2029 | int tcpv4_offload_init(void); |
28850dc7 | 2030 | |
5c9f3023 JP |
2031 | void tcp_v4_init(void); |
2032 | void tcp_init(void); | |
20380731 | 2033 | |
659a8ad5 | 2034 | /* tcp_recovery.c */ |
d716bfdb | 2035 | void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb); |
6ac06ecd | 2036 | void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced); |
b8fef65a YC |
2037 | extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb, |
2038 | u32 reo_wnd); | |
0c8ee050 | 2039 | extern bool tcp_rack_mark_lost(struct sock *sk); |
1d0833df | 2040 | extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq, |
9a568de4 | 2041 | u64 xmit_time); |
57dde7f7 | 2042 | extern void tcp_rack_reo_timeout(struct sock *sk); |
1f255691 | 2043 | extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs); |
659a8ad5 | 2044 | |
e1a10ef7 NC |
2045 | /* At how many usecs into the future should the RTO fire? */ |
2046 | static inline s64 tcp_rto_delta_us(const struct sock *sk) | |
2047 | { | |
75c119af | 2048 | const struct sk_buff *skb = tcp_rtx_queue_head(sk); |
e1a10ef7 | 2049 | u32 rto = inet_csk(sk)->icsk_rto; |
2fd66ffb | 2050 | u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto); |
e1a10ef7 NC |
2051 | |
2052 | return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp; | |
2053 | } | |
2054 | ||
e25f866f CW |
2055 | /* |
2056 | * Save and compile IPv4 options, return a pointer to it | |
2057 | */ | |
91ed1e66 PA |
2058 | static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net, |
2059 | struct sk_buff *skb) | |
e25f866f CW |
2060 | { |
2061 | const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt; | |
2062 | struct ip_options_rcu *dopt = NULL; | |
2063 | ||
461b74c3 | 2064 | if (opt->optlen) { |
e25f866f CW |
2065 | int opt_size = sizeof(*dopt) + opt->optlen; |
2066 | ||
2067 | dopt = kmalloc(opt_size, GFP_ATOMIC); | |
91ed1e66 | 2068 | if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) { |
e25f866f CW |
2069 | kfree(dopt); |
2070 | dopt = NULL; | |
2071 | } | |
2072 | } | |
2073 | return dopt; | |
2074 | } | |
2075 | ||
98781965 ED |
2076 | /* locally generated TCP pure ACKs have skb->truesize == 2 |
2077 | * (check tcp_send_ack() in net/ipv4/tcp_output.c ) | |
2078 | * This is much faster than dissecting the packet to find out. | |
2079 | * (Think of GRE encapsulations, IPv4, IPv6, ...) | |
2080 | */ | |
2081 | static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb) | |
2082 | { | |
2083 | return skb->truesize == 2; | |
2084 | } | |
2085 | ||
2086 | static inline void skb_set_tcp_pure_ack(struct sk_buff *skb) | |
2087 | { | |
2088 | skb->truesize = 2; | |
2089 | } | |
2090 | ||
473bd239 TH |
2091 | static inline int tcp_inq(struct sock *sk) |
2092 | { | |
2093 | struct tcp_sock *tp = tcp_sk(sk); | |
2094 | int answ; | |
2095 | ||
2096 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { | |
2097 | answ = 0; | |
2098 | } else if (sock_flag(sk, SOCK_URGINLINE) || | |
2099 | !tp->urg_data || | |
2100 | before(tp->urg_seq, tp->copied_seq) || | |
2101 | !before(tp->urg_seq, tp->rcv_nxt)) { | |
2102 | ||
2103 | answ = tp->rcv_nxt - tp->copied_seq; | |
2104 | ||
2105 | /* Subtract 1, if FIN was received */ | |
2106 | if (answ && sock_flag(sk, SOCK_DONE)) | |
2107 | answ--; | |
2108 | } else { | |
2109 | answ = tp->urg_seq - tp->copied_seq; | |
2110 | } | |
2111 | ||
2112 | return answ; | |
2113 | } | |
2114 | ||
32035585 TH |
2115 | int tcp_peek_len(struct socket *sock); |
2116 | ||
a44d6eac MKL |
2117 | static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb) |
2118 | { | |
2119 | u16 segs_in; | |
2120 | ||
2121 | segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs); | |
2122 | tp->segs_in += segs_in; | |
2123 | if (skb->len > tcp_hdrlen(skb)) | |
2124 | tp->data_segs_in += segs_in; | |
2125 | } | |
2126 | ||
9caad864 ED |
2127 | /* |
2128 | * TCP listen path runs lockless. | |
2129 | * We forced "struct sock" to be const qualified to make sure | |
2130 | * we don't modify one of its field by mistake. | |
2131 | * Here, we increment sk_drops which is an atomic_t, so we can safely | |
2132 | * make sock writable again. | |
2133 | */ | |
2134 | static inline void tcp_listendrop(const struct sock *sk) | |
2135 | { | |
2136 | atomic_inc(&((struct sock *)sk)->sk_drops); | |
02a1d6e7 | 2137 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS); |
9caad864 ED |
2138 | } |
2139 | ||
218af599 ED |
2140 | enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer); |
2141 | ||
734942cc DW |
2142 | /* |
2143 | * Interface for adding Upper Level Protocols over TCP | |
2144 | */ | |
2145 | ||
2146 | #define TCP_ULP_NAME_MAX 16 | |
2147 | #define TCP_ULP_MAX 128 | |
2148 | #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX) | |
2149 | ||
2150 | struct tcp_ulp_ops { | |
2151 | struct list_head list; | |
2152 | ||
2153 | /* initialize ulp */ | |
2154 | int (*init)(struct sock *sk); | |
95fa1454 | 2155 | /* update ulp */ |
35df9c9b JF |
2156 | void (*update)(struct sock *sk, struct proto *p, |
2157 | void (*write_space)(struct sock *sk)); | |
734942cc DW |
2158 | /* cleanup ulp */ |
2159 | void (*release)(struct sock *sk); | |
61723b39 DC |
2160 | /* diagnostic */ |
2161 | int (*get_info)(const struct sock *sk, struct sk_buff *skb); | |
2162 | size_t (*get_info_size)(const struct sock *sk); | |
734942cc DW |
2163 | |
2164 | char name[TCP_ULP_NAME_MAX]; | |
2165 | struct module *owner; | |
2166 | }; | |
2167 | int tcp_register_ulp(struct tcp_ulp_ops *type); | |
2168 | void tcp_unregister_ulp(struct tcp_ulp_ops *type); | |
2169 | int tcp_set_ulp(struct sock *sk, const char *name); | |
2170 | void tcp_get_available_ulp(char *buf, size_t len); | |
2171 | void tcp_cleanup_ulp(struct sock *sk); | |
35df9c9b JF |
2172 | void tcp_update_ulp(struct sock *sk, struct proto *p, |
2173 | void (*write_space)(struct sock *sk)); | |
734942cc | 2174 | |
037b0b86 DB |
2175 | #define MODULE_ALIAS_TCP_ULP(name) \ |
2176 | __MODULE_INFO(alias, alias_userspace, name); \ | |
2177 | __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name) | |
2178 | ||
604326b4 DB |
2179 | struct sk_msg; |
2180 | struct sk_psock; | |
2181 | ||
2182 | int tcp_bpf_init(struct sock *sk); | |
2183 | void tcp_bpf_reinit(struct sock *sk); | |
2184 | int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes, | |
2185 | int flags); | |
2186 | int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, | |
2187 | int nonblock, int flags, int *addr_len); | |
2188 | int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock, | |
02c558b2 | 2189 | struct msghdr *msg, int len, int flags); |
604326b4 | 2190 | |
40304b2a LB |
2191 | /* Call BPF_SOCK_OPS program that returns an int. If the return value |
2192 | * is < 0, then the BPF op failed (for example if the loaded BPF | |
2193 | * program does not support the chosen operation or there is no BPF | |
2194 | * program loaded). | |
2195 | */ | |
2196 | #ifdef CONFIG_BPF | |
de525be2 | 2197 | static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) |
40304b2a LB |
2198 | { |
2199 | struct bpf_sock_ops_kern sock_ops; | |
2200 | int ret; | |
2201 | ||
b73042b8 | 2202 | memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); |
f19397a5 LB |
2203 | if (sk_fullsock(sk)) { |
2204 | sock_ops.is_fullsock = 1; | |
40304b2a | 2205 | sock_owned_by_me(sk); |
f19397a5 | 2206 | } |
40304b2a | 2207 | |
40304b2a LB |
2208 | sock_ops.sk = sk; |
2209 | sock_ops.op = op; | |
de525be2 LB |
2210 | if (nargs > 0) |
2211 | memcpy(sock_ops.args, args, nargs * sizeof(*args)); | |
40304b2a LB |
2212 | |
2213 | ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops); | |
2214 | if (ret == 0) | |
2215 | ret = sock_ops.reply; | |
2216 | else | |
2217 | ret = -1; | |
2218 | return ret; | |
2219 | } | |
de525be2 LB |
2220 | |
2221 | static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) | |
2222 | { | |
2223 | u32 args[2] = {arg1, arg2}; | |
2224 | ||
2225 | return tcp_call_bpf(sk, op, 2, args); | |
2226 | } | |
2227 | ||
2228 | static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, | |
2229 | u32 arg3) | |
2230 | { | |
2231 | u32 args[3] = {arg1, arg2, arg3}; | |
2232 | ||
2233 | return tcp_call_bpf(sk, op, 3, args); | |
2234 | } | |
2235 | ||
40304b2a | 2236 | #else |
de525be2 | 2237 | static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args) |
40304b2a LB |
2238 | { |
2239 | return -EPERM; | |
2240 | } | |
de525be2 LB |
2241 | |
2242 | static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2) | |
2243 | { | |
2244 | return -EPERM; | |
2245 | } | |
2246 | ||
2247 | static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, | |
2248 | u32 arg3) | |
2249 | { | |
2250 | return -EPERM; | |
2251 | } | |
2252 | ||
40304b2a LB |
2253 | #endif |
2254 | ||
8550f328 LB |
2255 | static inline u32 tcp_timeout_init(struct sock *sk) |
2256 | { | |
2257 | int timeout; | |
2258 | ||
de525be2 | 2259 | timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL); |
8550f328 LB |
2260 | |
2261 | if (timeout <= 0) | |
2262 | timeout = TCP_TIMEOUT_INIT; | |
2263 | return timeout; | |
2264 | } | |
2265 | ||
13d3b1eb LB |
2266 | static inline u32 tcp_rwnd_init_bpf(struct sock *sk) |
2267 | { | |
2268 | int rwnd; | |
2269 | ||
de525be2 | 2270 | rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL); |
13d3b1eb LB |
2271 | |
2272 | if (rwnd < 0) | |
2273 | rwnd = 0; | |
2274 | return rwnd; | |
2275 | } | |
91b5b21c LB |
2276 | |
2277 | static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk) | |
2278 | { | |
de525be2 | 2279 | return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1); |
91b5b21c | 2280 | } |
60e2a778 | 2281 | |
23729ff2 SF |
2282 | static inline void tcp_bpf_rtt(struct sock *sk) |
2283 | { | |
bef8e263 | 2284 | if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG)) |
23729ff2 SF |
2285 | tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL); |
2286 | } | |
2287 | ||
60e2a778 UB |
2288 | #if IS_ENABLED(CONFIG_SMC) |
2289 | extern struct static_key_false tcp_have_smc; | |
2290 | #endif | |
6dac1523 IL |
2291 | |
2292 | #if IS_ENABLED(CONFIG_TLS_DEVICE) | |
2293 | void clean_acked_data_enable(struct inet_connection_sock *icsk, | |
2294 | void (*cad)(struct sock *sk, u32 ack_seq)); | |
2295 | void clean_acked_data_disable(struct inet_connection_sock *icsk); | |
494bc1d2 | 2296 | void clean_acked_data_flush(void); |
6dac1523 IL |
2297 | #endif |
2298 | ||
a842fe14 ED |
2299 | DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled); |
2300 | static inline void tcp_add_tx_delay(struct sk_buff *skb, | |
2301 | const struct tcp_sock *tp) | |
2302 | { | |
2303 | if (static_branch_unlikely(&tcp_tx_delay_enabled)) | |
2304 | skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC; | |
2305 | } | |
2306 | ||
d6fb396c ED |
2307 | /* Compute Earliest Departure Time for some control packets |
2308 | * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets. | |
2309 | */ | |
2310 | static inline u64 tcp_transmit_time(const struct sock *sk) | |
a842fe14 ED |
2311 | { |
2312 | if (static_branch_unlikely(&tcp_tx_delay_enabled)) { | |
2313 | u32 delay = (sk->sk_state == TCP_TIME_WAIT) ? | |
2314 | tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay; | |
2315 | ||
d6fb396c | 2316 | return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC; |
a842fe14 | 2317 | } |
d6fb396c | 2318 | return 0; |
a842fe14 ED |
2319 | } |
2320 | ||
1da177e4 | 2321 | #endif /* _TCP_H */ |