1 /* SPDX-License-Identifier: GPL-2.0-or-later */
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 protocol.
8 *
9 * Version: @(#)tcp.h 1.0.2 04/28/93
10 *
11 * Author: Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 */
13 #ifndef _LINUX_TCP_H
14 #define _LINUX_TCP_H
15
16
17 #include <linux/skbuff.h>
18 #include <linux/win_minmax.h>
19 #include <linux/android_kabi.h>
20 #include <net/sock.h>
21 #include <net/inet_connection_sock.h>
22 #include <net/inet_timewait_sock.h>
23 #include <uapi/linux/tcp.h>
24
tcp_hdr(const struct sk_buff * skb)25 static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb)
26 {
27 return (struct tcphdr *)skb_transport_header(skb);
28 }
29
__tcp_hdrlen(const struct tcphdr * th)30 static inline unsigned int __tcp_hdrlen(const struct tcphdr *th)
31 {
32 return th->doff * 4;
33 }
34
tcp_hdrlen(const struct sk_buff * skb)35 static inline unsigned int tcp_hdrlen(const struct sk_buff *skb)
36 {
37 return __tcp_hdrlen(tcp_hdr(skb));
38 }
39
inner_tcp_hdr(const struct sk_buff * skb)40 static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb)
41 {
42 return (struct tcphdr *)skb_inner_transport_header(skb);
43 }
44
inner_tcp_hdrlen(const struct sk_buff * skb)45 static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb)
46 {
47 return inner_tcp_hdr(skb)->doff * 4;
48 }
49
tcp_optlen(const struct sk_buff * skb)50 static inline unsigned int tcp_optlen(const struct sk_buff *skb)
51 {
52 return (tcp_hdr(skb)->doff - 5) * 4;
53 }
54
55 /* TCP Fast Open */
56 #define TCP_FASTOPEN_COOKIE_MIN 4 /* Min Fast Open Cookie size in bytes */
57 #define TCP_FASTOPEN_COOKIE_MAX 16 /* Max Fast Open Cookie size in bytes */
58 #define TCP_FASTOPEN_COOKIE_SIZE 8 /* the size employed by this impl. */
59
60 /* TCP Fast Open Cookie as stored in memory */
61 struct tcp_fastopen_cookie {
62 __le64 val[DIV_ROUND_UP(TCP_FASTOPEN_COOKIE_MAX, sizeof(u64))];
63 s8 len;
64 bool exp; /* In RFC6994 experimental option format */
65 };
66
67 /* This defines a selective acknowledgement block. */
68 struct tcp_sack_block_wire {
69 __be32 start_seq;
70 __be32 end_seq;
71 };
72
73 struct tcp_sack_block {
74 u32 start_seq;
75 u32 end_seq;
76 };
77
78 /*These are used to set the sack_ok field in struct tcp_options_received */
79 #define TCP_SACK_SEEN (1 << 0) /*1 = peer is SACK capable, */
80 #define TCP_DSACK_SEEN (1 << 2) /*1 = DSACK was received from peer*/
81
82 struct tcp_options_received {
83 /* PAWS/RTTM data */
84 int ts_recent_stamp;/* Time we stored ts_recent (for aging) */
85 u32 ts_recent; /* Time stamp to echo next */
86 u32 rcv_tsval; /* Time stamp value */
87 u32 rcv_tsecr; /* Time stamp echo reply */
88 u16 saw_tstamp : 1, /* Saw TIMESTAMP on last packet */
89 tstamp_ok : 1, /* TIMESTAMP seen on SYN packet */
90 dsack : 1, /* D-SACK is scheduled */
91 wscale_ok : 1, /* Wscale seen on SYN packet */
92 sack_ok : 3, /* SACK seen on SYN packet */
93 smc_ok : 1, /* SMC seen on SYN packet */
94 snd_wscale : 4, /* Window scaling received from sender */
95 rcv_wscale : 4; /* Window scaling to send to receiver */
96 u8 saw_unknown:1, /* Received unknown option */
97 unused:7;
98 u8 num_sacks; /* Number of SACK blocks */
99 u16 user_mss; /* mss requested by user in ioctl */
100 u16 mss_clamp; /* Maximal mss, negotiated at connection setup */
101 };
102
tcp_clear_options(struct tcp_options_received * rx_opt)103 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
104 {
105 rx_opt->tstamp_ok = rx_opt->sack_ok = 0;
106 rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
107 #if IS_ENABLED(CONFIG_SMC)
108 rx_opt->smc_ok = 0;
109 #endif
110 }
111
112 /* This is the max number of SACKS that we'll generate and process. It's safe
113 * to increase this, although since:
114 * size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8)
115 * only four options will fit in a standard TCP header */
116 #define TCP_NUM_SACKS 4
117
118 struct tcp_request_sock_ops;
119
120 struct tcp_request_sock {
121 struct inet_request_sock req;
122 const struct tcp_request_sock_ops *af_specific;
123 u64 snt_synack; /* first SYNACK sent time */
124 bool tfo_listener;
125 bool is_mptcp;
126 #if IS_ENABLED(CONFIG_MPTCP)
127 bool drop_req;
128 #endif
129 u32 txhash;
130 u32 rcv_isn;
131 u32 snt_isn;
132 u32 ts_off;
133 u32 last_oow_ack_time; /* last SYNACK */
134 u32 rcv_nxt; /* the ack # by SYNACK. For
135 * FastOpen it's the seq#
136 * after data-in-SYN.
137 */
138 u8 syn_tos;
139 };
140
tcp_rsk(const struct request_sock * req)141 static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
142 {
143 return (struct tcp_request_sock *)req;
144 }
145
146 struct tcp_sock {
147 /* inet_connection_sock has to be the first member of tcp_sock */
148 struct inet_connection_sock inet_conn;
149 u16 tcp_header_len; /* Bytes of tcp header to send */
150 u16 gso_segs; /* Max number of segs per GSO packet */
151
152 /*
153 * Header prediction flags
154 * 0x5?10 << 16 + snd_wnd in net byte order
155 */
156 __be32 pred_flags;
157
158 /*
159 * RFC793 variables by their proper names. This means you can
160 * read the code and the spec side by side (and laugh ...)
161 * See RFC793 and RFC1122. The RFC writes these in capitals.
162 */
163 u64 bytes_received; /* RFC4898 tcpEStatsAppHCThruOctetsReceived
164 * sum(delta(rcv_nxt)), or how many bytes
165 * were acked.
166 */
167 u32 segs_in; /* RFC4898 tcpEStatsPerfSegsIn
168 * total number of segments in.
169 */
170 u32 data_segs_in; /* RFC4898 tcpEStatsPerfDataSegsIn
171 * total number of data segments in.
172 */
173 u32 rcv_nxt; /* What we want to receive next */
174 u32 copied_seq; /* Head of yet unread data */
175 u32 rcv_wup; /* rcv_nxt on last window update sent */
176 u32 snd_nxt; /* Next sequence we send */
177 u32 segs_out; /* RFC4898 tcpEStatsPerfSegsOut
178 * The total number of segments sent.
179 */
180 u32 data_segs_out; /* RFC4898 tcpEStatsPerfDataSegsOut
181 * total number of data segments sent.
182 */
183 u64 bytes_sent; /* RFC4898 tcpEStatsPerfHCDataOctetsOut
184 * total number of data bytes sent.
185 */
186 u64 bytes_acked; /* RFC4898 tcpEStatsAppHCThruOctetsAcked
187 * sum(delta(snd_una)), or how many bytes
188 * were acked.
189 */
190 u32 dsack_dups; /* RFC4898 tcpEStatsStackDSACKDups
191 * total number of DSACK blocks received
192 */
193 u32 snd_una; /* First byte we want an ack for */
194 u32 snd_sml; /* Last byte of the most recently transmitted small packet */
195 u32 rcv_tstamp; /* timestamp of last received ACK (for keepalives) */
196 u32 lsndtime; /* timestamp of last sent data packet (for restart window) */
197 u32 last_oow_ack_time; /* timestamp of last out-of-window ACK */
198 u32 compressed_ack_rcv_nxt;
199
200 u32 tsoffset; /* timestamp offset */
201
202 struct list_head tsq_node; /* anchor in tsq_tasklet.head list */
203 struct list_head tsorted_sent_queue; /* time-sorted sent but un-SACKed skbs */
204
205 u32 snd_wl1; /* Sequence for window update */
206 u32 snd_wnd; /* The window we expect to receive */
207 u32 max_window; /* Maximal window ever seen from peer */
208 u32 mss_cache; /* Cached effective mss, not including SACKS */
209
210 u32 window_clamp; /* Maximal window to advertise */
211 u32 rcv_ssthresh; /* Current window clamp */
212
213 /* Information of the most recently (s)acked skb */
214 struct tcp_rack {
215 u64 mstamp; /* (Re)sent time of the skb */
216 u32 rtt_us; /* Associated RTT */
217 u32 end_seq; /* Ending TCP sequence of the skb */
218 u32 last_delivered; /* tp->delivered at last reo_wnd adj */
219 u8 reo_wnd_steps; /* Allowed reordering window */
220 #define TCP_RACK_RECOVERY_THRESH 16
221 u8 reo_wnd_persist:5, /* No. of recovery since last adj */
222 dsack_seen:1, /* Whether DSACK seen after last adj */
223 advanced:1; /* mstamp advanced since last lost marking */
224 } rack;
225 u16 advmss; /* Advertised MSS */
226 u8 compressed_ack;
227 u8 dup_ack_counter:2,
228 tlp_retrans:1, /* TLP is a retransmission */
229 unused:5;
230 u32 chrono_start; /* Start time in jiffies of a TCP chrono */
231 u32 chrono_stat[3]; /* Time in jiffies for chrono_stat stats */
232 u8 chrono_type:2, /* current chronograph type */
233 rate_app_limited:1, /* rate_{delivered,interval_us} limited? */
234 fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */
235 fastopen_no_cookie:1, /* Allow send/recv SYN+data without a cookie */
236 is_sack_reneg:1, /* in recovery from loss with SACK reneg? */
237 fastopen_client_fail:2; /* reason why fastopen failed */
238 u8 nonagle : 4,/* Disable Nagle algorithm? */
239 thin_lto : 1,/* Use linear timeouts for thin streams */
240 recvmsg_inq : 1,/* Indicate # of bytes in queue upon recvmsg */
241 repair : 1,
242 frto : 1;/* F-RTO (RFC5682) activated in CA_Loss */
243 u8 repair_queue;
244 u8 save_syn:2, /* Save headers of SYN packet */
245 syn_data:1, /* SYN includes data */
246 syn_fastopen:1, /* SYN includes Fast Open option */
247 syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
248 syn_fastopen_ch:1, /* Active TFO re-enabling probe */
249 syn_data_acked:1,/* data in SYN is acked by SYN-ACK */
250 is_cwnd_limited:1;/* forward progress limited by snd_cwnd? */
251 u32 tlp_high_seq; /* snd_nxt at the time of TLP */
252
253 u32 tcp_tx_delay; /* delay (in usec) added to TX packets */
254 u64 tcp_wstamp_ns; /* departure time for next sent data packet */
255 u64 tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */
256
257 /* RTT measurement */
258 u64 tcp_mstamp; /* most recent packet received/sent */
259 u32 srtt_us; /* smoothed round trip time << 3 in usecs */
260 u32 mdev_us; /* medium deviation */
261 u32 mdev_max_us; /* maximal mdev for the last rtt period */
262 u32 rttvar_us; /* smoothed mdev_max */
263 u32 rtt_seq; /* sequence number to update rttvar */
264 struct minmax rtt_min;
265
266 u32 packets_out; /* Packets which are "in flight" */
267 u32 retrans_out; /* Retransmitted packets out */
268 u32 max_packets_out; /* max packets_out in last window */
269 /* GENKSYMS hack to preserve the ABI because of f4ce91ce12a7 ("tcp: fix
270 * tcp_cwnd_validate() to not forget is_cwnd_limited")
271 */
272 #ifndef __GENKSYMS__
273 u32 cwnd_usage_seq; /* right edge of cwnd usage tracking flight */
274 #else
275 u32 max_packets_seq; /* right edge of max_packets_out flight */
276 #endif
277
278 u16 urg_data; /* Saved octet of OOB data and control flags */
279 u8 ecn_flags; /* ECN status bits. */
280 u8 keepalive_probes; /* num of allowed keep alive probes */
281 u32 reordering; /* Packet reordering metric. */
282 u32 reord_seen; /* number of data packet reordering events */
283 u32 snd_up; /* Urgent pointer */
284
285 /*
286 * Options received (usually on last packet, some only on SYN packets).
287 */
288 struct tcp_options_received rx_opt;
289
290 /*
291 * Slow start and congestion control (see also Nagle, and Karn & Partridge)
292 */
293 u32 snd_ssthresh; /* Slow start size threshold */
294 u32 snd_cwnd; /* Sending congestion window */
295 u32 snd_cwnd_cnt; /* Linear increase counter */
296 u32 snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
297 u32 snd_cwnd_used;
298 u32 snd_cwnd_stamp;
299 u32 prior_cwnd; /* cwnd right before starting loss recovery */
300 u32 prr_delivered; /* Number of newly delivered packets to
301 * receiver in Recovery. */
302 u32 prr_out; /* Total number of pkts sent during Recovery. */
303 u32 delivered; /* Total data packets delivered incl. rexmits */
304 u32 delivered_ce; /* Like the above but only ECE marked packets */
305 u32 lost; /* Total data packets lost incl. rexmits */
306 u32 app_limited; /* limited until "delivered" reaches this val */
307 u64 first_tx_mstamp; /* start of window send phase */
308 u64 delivered_mstamp; /* time we reached "delivered" */
309 u32 rate_delivered; /* saved rate sample: packets delivered */
310 u32 rate_interval_us; /* saved rate sample: time elapsed */
311
312 u32 rcv_wnd; /* Current receiver window */
313 u32 write_seq; /* Tail(+1) of data held in tcp send buffer */
314 u32 notsent_lowat; /* TCP_NOTSENT_LOWAT */
315 u32 pushed_seq; /* Last pushed seq, required to talk to windows */
316 u32 lost_out; /* Lost packets */
317 u32 sacked_out; /* SACK'd packets */
318
319 struct hrtimer pacing_timer;
320 struct hrtimer compressed_ack_timer;
321
322 /* from STCP, retrans queue hinting */
323 struct sk_buff* lost_skb_hint;
324 struct sk_buff *retransmit_skb_hint;
325
326 /* OOO segments go in this rbtree. Socket lock must be held. */
327 struct rb_root out_of_order_queue;
328 struct sk_buff *ooo_last_skb; /* cache rb_last(out_of_order_queue) */
329
330 /* SACKs data, these 2 need to be together (see tcp_options_write) */
331 struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
332 struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
333
334 struct tcp_sack_block recv_sack_cache[4];
335
336 struct sk_buff *highest_sack; /* skb just after the highest
337 * skb with SACKed bit set
338 * (validity guaranteed only if
339 * sacked_out > 0)
340 */
341
342 int lost_cnt_hint;
343
344 u32 prior_ssthresh; /* ssthresh saved at recovery start */
345 u32 high_seq; /* snd_nxt at onset of congestion */
346
347 u32 retrans_stamp; /* Timestamp of the last retransmit,
348 * also used in SYN-SENT to remember stamp of
349 * the first SYN. */
350 u32 undo_marker; /* snd_una upon a new recovery episode. */
351 int undo_retrans; /* number of undoable retransmissions. */
352 u64 bytes_retrans; /* RFC4898 tcpEStatsPerfOctetsRetrans
353 * Total data bytes retransmitted
354 */
355 u32 total_retrans; /* Total retransmits for entire connection */
356
357 u32 urg_seq; /* Seq of received urgent pointer */
358 unsigned int keepalive_time; /* time before keep alive takes place */
359 unsigned int keepalive_intvl; /* time interval between keep alive probes */
360
361 int linger2;
362
363
364 /* Sock_ops bpf program related variables */
365 #ifdef CONFIG_BPF
366 u8 bpf_sock_ops_cb_flags; /* Control calling BPF programs
367 * values defined in uapi/linux/tcp.h
368 */
369 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) (TP->bpf_sock_ops_cb_flags & ARG)
370 #else
371 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) 0
372 #endif
373
374 u16 timeout_rehash; /* Timeout-triggered rehash attempts */
375
376 u32 rcv_ooopack; /* Received out-of-order packets, for tcpinfo */
377
378 /* Receiver side RTT estimation */
379 u32 rcv_rtt_last_tsecr;
380 struct {
381 u32 rtt_us;
382 u32 seq;
383 u64 time;
384 } rcv_rtt_est;
385
386 /* Receiver queue space */
387 struct {
388 u32 space;
389 u32 seq;
390 u64 time;
391 } rcvq_space;
392
393 /* TCP-specific MTU probe information. */
394 struct {
395 u32 probe_seq_start;
396 u32 probe_seq_end;
397 } mtu_probe;
398 u32 mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
399 * while socket was owned by user.
400 */
401 #if IS_ENABLED(CONFIG_MPTCP)
402 bool is_mptcp;
403 #endif
404 #if IS_ENABLED(CONFIG_SMC)
405 bool syn_smc; /* SYN includes SMC */
406 #endif
407
408 #ifdef CONFIG_TCP_MD5SIG
409 /* TCP AF-Specific parts; only used by MD5 Signature support so far */
410 const struct tcp_sock_af_ops *af_specific;
411
412 /* TCP MD5 Signature Option information */
413 struct tcp_md5sig_info __rcu *md5sig_info;
414 #endif
415
416 /* TCP fastopen related information */
417 struct tcp_fastopen_request *fastopen_req;
418 /* fastopen_rsk points to request_sock that resulted in this big
419 * socket. Used to retransmit SYNACKs etc.
420 */
421 struct request_sock __rcu *fastopen_rsk;
422 struct saved_syn *saved_syn;
423
424 ANDROID_KABI_RESERVE(1);
425 };
426
427 enum tsq_enum {
428 TSQ_THROTTLED,
429 TSQ_QUEUED,
430 TCP_TSQ_DEFERRED, /* tcp_tasklet_func() found socket was owned */
431 TCP_WRITE_TIMER_DEFERRED, /* tcp_write_timer() found socket was owned */
432 TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */
433 TCP_MTU_REDUCED_DEFERRED, /* tcp_v{4|6}_err() could not call
434 * tcp_v{4|6}_mtu_reduced()
435 */
436 };
437
438 enum tsq_flags {
439 TSQF_THROTTLED = (1UL << TSQ_THROTTLED),
440 TSQF_QUEUED = (1UL << TSQ_QUEUED),
441 TCPF_TSQ_DEFERRED = (1UL << TCP_TSQ_DEFERRED),
442 TCPF_WRITE_TIMER_DEFERRED = (1UL << TCP_WRITE_TIMER_DEFERRED),
443 TCPF_DELACK_TIMER_DEFERRED = (1UL << TCP_DELACK_TIMER_DEFERRED),
444 TCPF_MTU_REDUCED_DEFERRED = (1UL << TCP_MTU_REDUCED_DEFERRED),
445 };
446
tcp_sk(const struct sock * sk)447 static inline struct tcp_sock *tcp_sk(const struct sock *sk)
448 {
449 return (struct tcp_sock *)sk;
450 }
451
452 struct tcp_timewait_sock {
453 struct inet_timewait_sock tw_sk;
454 #define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt
455 #define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt
456 u32 tw_rcv_wnd;
457 u32 tw_ts_offset;
458 u32 tw_ts_recent;
459
460 /* The time we sent the last out-of-window ACK: */
461 u32 tw_last_oow_ack_time;
462
463 int tw_ts_recent_stamp;
464 u32 tw_tx_delay;
465 #ifdef CONFIG_TCP_MD5SIG
466 struct tcp_md5sig_key *tw_md5_key;
467 #endif
468 };
469
tcp_twsk(const struct sock * sk)470 static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
471 {
472 return (struct tcp_timewait_sock *)sk;
473 }
474
tcp_passive_fastopen(const struct sock * sk)475 static inline bool tcp_passive_fastopen(const struct sock *sk)
476 {
477 return sk->sk_state == TCP_SYN_RECV &&
478 rcu_access_pointer(tcp_sk(sk)->fastopen_rsk) != NULL;
479 }
480
fastopen_queue_tune(struct sock * sk,int backlog)481 static inline void fastopen_queue_tune(struct sock *sk, int backlog)
482 {
483 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
484 int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn);
485
486 WRITE_ONCE(queue->fastopenq.max_qlen, min_t(unsigned int, backlog, somaxconn));
487 }
488
tcp_move_syn(struct tcp_sock * tp,struct request_sock * req)489 static inline void tcp_move_syn(struct tcp_sock *tp,
490 struct request_sock *req)
491 {
492 tp->saved_syn = req->saved_syn;
493 req->saved_syn = NULL;
494 }
495
tcp_saved_syn_free(struct tcp_sock * tp)496 static inline void tcp_saved_syn_free(struct tcp_sock *tp)
497 {
498 kfree(tp->saved_syn);
499 tp->saved_syn = NULL;
500 }
501
tcp_saved_syn_len(const struct saved_syn * saved_syn)502 static inline u32 tcp_saved_syn_len(const struct saved_syn *saved_syn)
503 {
504 return saved_syn->mac_hdrlen + saved_syn->network_hdrlen +
505 saved_syn->tcp_hdrlen;
506 }
507
508 struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk,
509 const struct sk_buff *orig_skb);
510
tcp_mss_clamp(const struct tcp_sock * tp,u16 mss)511 static inline u16 tcp_mss_clamp(const struct tcp_sock *tp, u16 mss)
512 {
513 /* We use READ_ONCE() here because socket might not be locked.
514 * This happens for listeners.
515 */
516 u16 user_mss = READ_ONCE(tp->rx_opt.user_mss);
517
518 return (user_mss && user_mss < mss) ? user_mss : mss;
519 }
520
521 int tcp_skb_shift(struct sk_buff *to, struct sk_buff *from, int pcount,
522 int shiftlen);
523
524 void tcp_sock_set_cork(struct sock *sk, bool on);
525 int tcp_sock_set_keepcnt(struct sock *sk, int val);
526 int tcp_sock_set_keepidle_locked(struct sock *sk, int val);
527 int tcp_sock_set_keepidle(struct sock *sk, int val);
528 int tcp_sock_set_keepintvl(struct sock *sk, int val);
529 void tcp_sock_set_nodelay(struct sock *sk);
530 void tcp_sock_set_quickack(struct sock *sk, int val);
531 int tcp_sock_set_syncnt(struct sock *sk, int val);
532 void tcp_sock_set_user_timeout(struct sock *sk, u32 val);
533
534 #endif /* _LINUX_TCP_H */
535