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 <net/sock.h>
20#include <net/inet_connection_sock.h>
21#include <net/inet_timewait_sock.h>
22#include <uapi/linux/tcp.h>
23
24static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb)
25{
26 return (struct tcphdr *)skb_transport_header(skb);
27}
28
29static inline unsigned int __tcp_hdrlen(const struct tcphdr *th)
30{
31 return th->doff * 4;
32}
33
34static inline unsigned int tcp_hdrlen(const struct sk_buff *skb)
35{
36 return __tcp_hdrlen(tcp_hdr(skb));
37}
38
39static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb)
40{
41 return (struct tcphdr *)skb_inner_transport_header(skb);
42}
43
44static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb)
45{
46 return inner_tcp_hdr(skb)->doff * 4;
47}
48
49/**
50 * skb_tcp_all_headers - Returns size of all headers for a TCP packet
51 * @skb: buffer
52 *
53 * Used in TX path, for a packet known to be a TCP one.
54 *
55 * if (skb_is_gso(skb)) {
56 * int hlen = skb_tcp_all_headers(skb);
57 * ...
58 */
59static inline int skb_tcp_all_headers(const struct sk_buff *skb)
60{
61 return skb_transport_offset(skb) + tcp_hdrlen(skb);
62}
63
64/**
65 * skb_inner_tcp_all_headers - Returns size of all headers for an encap TCP packet
66 * @skb: buffer
67 *
68 * Used in TX path, for a packet known to be a TCP one.
69 *
70 * if (skb_is_gso(skb) && skb->encapsulation) {
71 * int hlen = skb_inner_tcp_all_headers(skb);
72 * ...
73 */
74static inline int skb_inner_tcp_all_headers(const struct sk_buff *skb)
75{
76 return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
77}
78
79static inline unsigned int tcp_optlen(const struct sk_buff *skb)
80{
81 return (tcp_hdr(skb)->doff - 5) * 4;
82}
83
84/* TCP Fast Open */
85#define TCP_FASTOPEN_COOKIE_MIN 4 /* Min Fast Open Cookie size in bytes */
86#define TCP_FASTOPEN_COOKIE_MAX 16 /* Max Fast Open Cookie size in bytes */
87#define TCP_FASTOPEN_COOKIE_SIZE 8 /* the size employed by this impl. */
88
89/* TCP Fast Open Cookie as stored in memory */
90struct tcp_fastopen_cookie {
91 __le64 val[DIV_ROUND_UP(TCP_FASTOPEN_COOKIE_MAX, sizeof(u64))];
92 s8 len;
93 bool exp; /* In RFC6994 experimental option format */
94};
95
96/* This defines a selective acknowledgement block. */
97struct tcp_sack_block_wire {
98 __be32 start_seq;
99 __be32 end_seq;
100};
101
102struct tcp_sack_block {
103 u32 start_seq;
104 u32 end_seq;
105};
106
107/*These are used to set the sack_ok field in struct tcp_options_received */
108#define TCP_SACK_SEEN (1 << 0) /*1 = peer is SACK capable, */
109#define TCP_DSACK_SEEN (1 << 2) /*1 = DSACK was received from peer*/
110
111struct tcp_options_received {
112/* PAWS/RTTM data */
113 int ts_recent_stamp;/* Time we stored ts_recent (for aging) */
114 u32 ts_recent; /* Time stamp to echo next */
115 u32 rcv_tsval; /* Time stamp value */
116 u32 rcv_tsecr; /* Time stamp echo reply */
117 u16 saw_tstamp : 1, /* Saw TIMESTAMP on last packet */
118 tstamp_ok : 1, /* TIMESTAMP seen on SYN packet */
119 dsack : 1, /* D-SACK is scheduled */
120 wscale_ok : 1, /* Wscale seen on SYN packet */
121 sack_ok : 3, /* SACK seen on SYN packet */
122 smc_ok : 1, /* SMC seen on SYN packet */
123 snd_wscale : 4, /* Window scaling received from sender */
124 rcv_wscale : 4; /* Window scaling to send to receiver */
125 u8 saw_unknown:1, /* Received unknown option */
126 unused:7;
127 u8 num_sacks; /* Number of SACK blocks */
128 u16 user_mss; /* mss requested by user in ioctl */
129 u16 mss_clamp; /* Maximal mss, negotiated at connection setup */
130};
131
132static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
133{
134 rx_opt->tstamp_ok = rx_opt->sack_ok = 0;
135 rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
136#if IS_ENABLED(CONFIG_SMC)
137 rx_opt->smc_ok = 0;
138#endif
139}
140
141/* This is the max number of SACKS that we'll generate and process. It's safe
142 * to increase this, although since:
143 * size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8)
144 * only four options will fit in a standard TCP header */
145#define TCP_NUM_SACKS 4
146
147struct tcp_request_sock_ops;
148
149struct tcp_request_sock {
150 struct inet_request_sock req;
151 const struct tcp_request_sock_ops *af_specific;
152 u64 snt_synack; /* first SYNACK sent time */
153 bool tfo_listener;
154 bool is_mptcp;
155#if IS_ENABLED(CONFIG_MPTCP)
156 bool drop_req;
157#endif
158 u32 txhash;
159 u32 rcv_isn;
160 u32 snt_isn;
161 u32 ts_off;
162 u32 last_oow_ack_time; /* last SYNACK */
163 u32 rcv_nxt; /* the ack # by SYNACK. For
164 * FastOpen it's the seq#
165 * after data-in-SYN.
166 */
167 u8 syn_tos;
168};
169
170static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
171{
172 return (struct tcp_request_sock *)req;
173}
174
175struct tcp_sock {
176 /* inet_connection_sock has to be the first member of tcp_sock */
177 struct inet_connection_sock inet_conn;
178 u16 tcp_header_len; /* Bytes of tcp header to send */
179 u16 gso_segs; /* Max number of segs per GSO packet */
180
181/*
182 * Header prediction flags
183 * 0x5?10 << 16 + snd_wnd in net byte order
184 */
185 __be32 pred_flags;
186
187/*
188 * RFC793 variables by their proper names. This means you can
189 * read the code and the spec side by side (and laugh ...)
190 * See RFC793 and RFC1122. The RFC writes these in capitals.
191 */
192 u64 bytes_received; /* RFC4898 tcpEStatsAppHCThruOctetsReceived
193 * sum(delta(rcv_nxt)), or how many bytes
194 * were acked.
195 */
196 u32 segs_in; /* RFC4898 tcpEStatsPerfSegsIn
197 * total number of segments in.
198 */
199 u32 data_segs_in; /* RFC4898 tcpEStatsPerfDataSegsIn
200 * total number of data segments in.
201 */
202 u32 rcv_nxt; /* What we want to receive next */
203 u32 copied_seq; /* Head of yet unread data */
204 u32 rcv_wup; /* rcv_nxt on last window update sent */
205 u32 snd_nxt; /* Next sequence we send */
206 u32 segs_out; /* RFC4898 tcpEStatsPerfSegsOut
207 * The total number of segments sent.
208 */
209 u32 data_segs_out; /* RFC4898 tcpEStatsPerfDataSegsOut
210 * total number of data segments sent.
211 */
212 u64 bytes_sent; /* RFC4898 tcpEStatsPerfHCDataOctetsOut
213 * total number of data bytes sent.
214 */
215 u64 bytes_acked; /* RFC4898 tcpEStatsAppHCThruOctetsAcked
216 * sum(delta(snd_una)), or how many bytes
217 * were acked.
218 */
219 u32 dsack_dups; /* RFC4898 tcpEStatsStackDSACKDups
220 * total number of DSACK blocks received
221 */
222 u32 snd_una; /* First byte we want an ack for */
223 u32 snd_sml; /* Last byte of the most recently transmitted small packet */
224 u32 rcv_tstamp; /* timestamp of last received ACK (for keepalives) */
225 u32 lsndtime; /* timestamp of last sent data packet (for restart window) */
226 u32 last_oow_ack_time; /* timestamp of last out-of-window ACK */
227 u32 compressed_ack_rcv_nxt;
228
229 u32 tsoffset; /* timestamp offset */
230
231 struct list_head tsq_node; /* anchor in tsq_tasklet.head list */
232 struct list_head tsorted_sent_queue; /* time-sorted sent but un-SACKed skbs */
233
234 u32 snd_wl1; /* Sequence for window update */
235 u32 snd_wnd; /* The window we expect to receive */
236 u32 max_window; /* Maximal window ever seen from peer */
237 u32 mss_cache; /* Cached effective mss, not including SACKS */
238
239 u32 window_clamp; /* Maximal window to advertise */
240 u32 rcv_ssthresh; /* Current window clamp */
241
242 /* Information of the most recently (s)acked skb */
243 struct tcp_rack {
244 u64 mstamp; /* (Re)sent time of the skb */
245 u32 rtt_us; /* Associated RTT */
246 u32 end_seq; /* Ending TCP sequence of the skb */
247 u32 last_delivered; /* tp->delivered at last reo_wnd adj */
248 u8 reo_wnd_steps; /* Allowed reordering window */
249#define TCP_RACK_RECOVERY_THRESH 16
250 u8 reo_wnd_persist:5, /* No. of recovery since last adj */
251 dsack_seen:1, /* Whether DSACK seen after last adj */
252 advanced:1; /* mstamp advanced since last lost marking */
253 } rack;
254 u16 advmss; /* Advertised MSS */
255 u8 compressed_ack;
256 u8 dup_ack_counter:2,
257 tlp_retrans:1, /* TLP is a retransmission */
258 unused:5;
259 u32 chrono_start; /* Start time in jiffies of a TCP chrono */
260 u32 chrono_stat[3]; /* Time in jiffies for chrono_stat stats */
261 u8 chrono_type:2, /* current chronograph type */
262 rate_app_limited:1, /* rate_{delivered,interval_us} limited? */
263 fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */
264 fastopen_no_cookie:1, /* Allow send/recv SYN+data without a cookie */
265 is_sack_reneg:1, /* in recovery from loss with SACK reneg? */
266 fastopen_client_fail:2; /* reason why fastopen failed */
267 u8 nonagle : 4,/* Disable Nagle algorithm? */
268 thin_lto : 1,/* Use linear timeouts for thin streams */
269 recvmsg_inq : 1,/* Indicate # of bytes in queue upon recvmsg */
270 repair : 1,
271 frto : 1;/* F-RTO (RFC5682) activated in CA_Loss */
272 u8 repair_queue;
273 u8 save_syn:2, /* Save headers of SYN packet */
274 syn_data:1, /* SYN includes data */
275 syn_fastopen:1, /* SYN includes Fast Open option */
276 syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
277 syn_fastopen_ch:1, /* Active TFO re-enabling probe */
278 syn_data_acked:1,/* data in SYN is acked by SYN-ACK */
279 is_cwnd_limited:1;/* forward progress limited by snd_cwnd? */
280 u32 tlp_high_seq; /* snd_nxt at the time of TLP */
281
282 u32 tcp_tx_delay; /* delay (in usec) added to TX packets */
283 u64 tcp_wstamp_ns; /* departure time for next sent data packet */
284 u64 tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */
285
286/* RTT measurement */
287 u64 tcp_mstamp; /* most recent packet received/sent */
288 u32 srtt_us; /* smoothed round trip time << 3 in usecs */
289 u32 mdev_us; /* medium deviation */
290 u32 mdev_max_us; /* maximal mdev for the last rtt period */
291 u32 rttvar_us; /* smoothed mdev_max */
292 u32 rtt_seq; /* sequence number to update rttvar */
293 struct minmax rtt_min;
294
295 u32 packets_out; /* Packets which are "in flight" */
296 u32 retrans_out; /* Retransmitted packets out */
297 u32 max_packets_out; /* max packets_out in last window */
298 u32 max_packets_seq; /* right edge of max_packets_out flight */
299
300 u16 urg_data; /* Saved octet of OOB data and control flags */
301 u8 ecn_flags; /* ECN status bits. */
302 u8 keepalive_probes; /* num of allowed keep alive probes */
303 u32 reordering; /* Packet reordering metric. */
304 u32 reord_seen; /* number of data packet reordering events */
305 u32 snd_up; /* Urgent pointer */
306
307/*
308 * Options received (usually on last packet, some only on SYN packets).
309 */
310 struct tcp_options_received rx_opt;
311
312/*
313 * Slow start and congestion control (see also Nagle, and Karn & Partridge)
314 */
315 u32 snd_ssthresh; /* Slow start size threshold */
316 u32 snd_cwnd; /* Sending congestion window */
317 u32 snd_cwnd_cnt; /* Linear increase counter */
318 u32 snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
319 u32 snd_cwnd_used;
320 u32 snd_cwnd_stamp;
321 u32 prior_cwnd; /* cwnd right before starting loss recovery */
322 u32 prr_delivered; /* Number of newly delivered packets to
323 * receiver in Recovery. */
324 u32 prr_out; /* Total number of pkts sent during Recovery. */
325 u32 delivered; /* Total data packets delivered incl. rexmits */
326 u32 delivered_ce; /* Like the above but only ECE marked packets */
327 u32 lost; /* Total data packets lost incl. rexmits */
328 u32 app_limited; /* limited until "delivered" reaches this val */
329 u64 first_tx_mstamp; /* start of window send phase */
330 u64 delivered_mstamp; /* time we reached "delivered" */
331 u32 rate_delivered; /* saved rate sample: packets delivered */
332 u32 rate_interval_us; /* saved rate sample: time elapsed */
333
334 u32 rcv_wnd; /* Current receiver window */
335 u32 write_seq; /* Tail(+1) of data held in tcp send buffer */
336 u32 notsent_lowat; /* TCP_NOTSENT_LOWAT */
337 u32 pushed_seq; /* Last pushed seq, required to talk to windows */
338 u32 lost_out; /* Lost packets */
339 u32 sacked_out; /* SACK'd packets */
340
341 struct hrtimer pacing_timer;
342 struct hrtimer compressed_ack_timer;
343
344 /* from STCP, retrans queue hinting */
345 struct sk_buff* lost_skb_hint;
346 struct sk_buff *retransmit_skb_hint;
347
348 /* OOO segments go in this rbtree. Socket lock must be held. */
349 struct rb_root out_of_order_queue;
350 struct sk_buff *ooo_last_skb; /* cache rb_last(out_of_order_queue) */
351
352 /* SACKs data, these 2 need to be together (see tcp_options_write) */
353 struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
354 struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
355
356 struct tcp_sack_block recv_sack_cache[4];
357
358 struct sk_buff *highest_sack; /* skb just after the highest
359 * skb with SACKed bit set
360 * (validity guaranteed only if
361 * sacked_out > 0)
362 */
363
364 int lost_cnt_hint;
365
366 u32 prior_ssthresh; /* ssthresh saved at recovery start */
367 u32 high_seq; /* snd_nxt at onset of congestion */
368
369 u32 retrans_stamp; /* Timestamp of the last retransmit,
370 * also used in SYN-SENT to remember stamp of
371 * the first SYN. */
372 u32 undo_marker; /* snd_una upon a new recovery episode. */
373 int undo_retrans; /* number of undoable retransmissions. */
374 u64 bytes_retrans; /* RFC4898 tcpEStatsPerfOctetsRetrans
375 * Total data bytes retransmitted
376 */
377 u32 total_retrans; /* Total retransmits for entire connection */
378
379 u32 urg_seq; /* Seq of received urgent pointer */
380 unsigned int keepalive_time; /* time before keep alive takes place */
381 unsigned int keepalive_intvl; /* time interval between keep alive probes */
382
383 int linger2;
384
385
386/* Sock_ops bpf program related variables */
387#ifdef CONFIG_BPF
388 u8 bpf_sock_ops_cb_flags; /* Control calling BPF programs
389 * values defined in uapi/linux/tcp.h
390 */
391#define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) (TP->bpf_sock_ops_cb_flags & ARG)
392#else
393#define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) 0
394#endif
395
396 u16 timeout_rehash; /* Timeout-triggered rehash attempts */
397
398 u32 rcv_ooopack; /* Received out-of-order packets, for tcpinfo */
399
400/* Receiver side RTT estimation */
401 u32 rcv_rtt_last_tsecr;
402 struct {
403 u32 rtt_us;
404 u32 seq;
405 u64 time;
406 } rcv_rtt_est;
407
408/* Receiver queue space */
409 struct {
410 u32 space;
411 u32 seq;
412 u64 time;
413 } rcvq_space;
414
415/* TCP-specific MTU probe information. */
416 struct {
417 u32 probe_seq_start;
418 u32 probe_seq_end;
419 } mtu_probe;
420 u32 mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
421 * while socket was owned by user.
422 */
423#if IS_ENABLED(CONFIG_MPTCP)
424 bool is_mptcp;
425#endif
426#if IS_ENABLED(CONFIG_SMC)
427 bool (*smc_hs_congested)(const struct sock *sk);
428 bool syn_smc; /* SYN includes SMC */
429#endif
430
431#ifdef CONFIG_TCP_MD5SIG
432/* TCP AF-Specific parts; only used by MD5 Signature support so far */
433 const struct tcp_sock_af_ops *af_specific;
434
435/* TCP MD5 Signature Option information */
436 struct tcp_md5sig_info __rcu *md5sig_info;
437#endif
438
439/* TCP fastopen related information */
440 struct tcp_fastopen_request *fastopen_req;
441 /* fastopen_rsk points to request_sock that resulted in this big
442 * socket. Used to retransmit SYNACKs etc.
443 */
444 struct request_sock __rcu *fastopen_rsk;
445 struct saved_syn *saved_syn;
446};
447
448enum tsq_enum {
449 TSQ_THROTTLED,
450 TSQ_QUEUED,
451 TCP_TSQ_DEFERRED, /* tcp_tasklet_func() found socket was owned */
452 TCP_WRITE_TIMER_DEFERRED, /* tcp_write_timer() found socket was owned */
453 TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */
454 TCP_MTU_REDUCED_DEFERRED, /* tcp_v{4|6}_err() could not call
455 * tcp_v{4|6}_mtu_reduced()
456 */
457};
458
459enum tsq_flags {
460 TSQF_THROTTLED = (1UL << TSQ_THROTTLED),
461 TSQF_QUEUED = (1UL << TSQ_QUEUED),
462 TCPF_TSQ_DEFERRED = (1UL << TCP_TSQ_DEFERRED),
463 TCPF_WRITE_TIMER_DEFERRED = (1UL << TCP_WRITE_TIMER_DEFERRED),
464 TCPF_DELACK_TIMER_DEFERRED = (1UL << TCP_DELACK_TIMER_DEFERRED),
465 TCPF_MTU_REDUCED_DEFERRED = (1UL << TCP_MTU_REDUCED_DEFERRED),
466};
467
468static inline struct tcp_sock *tcp_sk(const struct sock *sk)
469{
470 return (struct tcp_sock *)sk;
471}
472
473struct tcp_timewait_sock {
474 struct inet_timewait_sock tw_sk;
475#define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt
476#define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt
477 u32 tw_rcv_wnd;
478 u32 tw_ts_offset;
479 u32 tw_ts_recent;
480
481 /* The time we sent the last out-of-window ACK: */
482 u32 tw_last_oow_ack_time;
483
484 int tw_ts_recent_stamp;
485 u32 tw_tx_delay;
486#ifdef CONFIG_TCP_MD5SIG
487 struct tcp_md5sig_key *tw_md5_key;
488#endif
489};
490
491static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
492{
493 return (struct tcp_timewait_sock *)sk;
494}
495
496static inline bool tcp_passive_fastopen(const struct sock *sk)
497{
498 return sk->sk_state == TCP_SYN_RECV &&
499 rcu_access_pointer(tcp_sk(sk)->fastopen_rsk) != NULL;
500}
501
502static inline void fastopen_queue_tune(struct sock *sk, int backlog)
503{
504 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
505 int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn);
506
507 queue->fastopenq.max_qlen = min_t(unsigned int, backlog, somaxconn);
508}
509
510static inline void tcp_move_syn(struct tcp_sock *tp,
511 struct request_sock *req)
512{
513 tp->saved_syn = req->saved_syn;
514 req->saved_syn = NULL;
515}
516
517static inline void tcp_saved_syn_free(struct tcp_sock *tp)
518{
519 kfree(tp->saved_syn);
520 tp->saved_syn = NULL;
521}
522
523static inline u32 tcp_saved_syn_len(const struct saved_syn *saved_syn)
524{
525 return saved_syn->mac_hdrlen + saved_syn->network_hdrlen +
526 saved_syn->tcp_hdrlen;
527}
528
529struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk,
530 const struct sk_buff *orig_skb,
531 const struct sk_buff *ack_skb);
532
533static inline u16 tcp_mss_clamp(const struct tcp_sock *tp, u16 mss)
534{
535 /* We use READ_ONCE() here because socket might not be locked.
536 * This happens for listeners.
537 */
538 u16 user_mss = READ_ONCE(tp->rx_opt.user_mss);
539
540 return (user_mss && user_mss < mss) ? user_mss : mss;
541}
542
543int tcp_skb_shift(struct sk_buff *to, struct sk_buff *from, int pcount,
544 int shiftlen);
545
546void __tcp_sock_set_cork(struct sock *sk, bool on);
547void tcp_sock_set_cork(struct sock *sk, bool on);
548int tcp_sock_set_keepcnt(struct sock *sk, int val);
549int tcp_sock_set_keepidle_locked(struct sock *sk, int val);
550int tcp_sock_set_keepidle(struct sock *sk, int val);
551int tcp_sock_set_keepintvl(struct sock *sk, int val);
552void __tcp_sock_set_nodelay(struct sock *sk, bool on);
553void tcp_sock_set_nodelay(struct sock *sk);
554void tcp_sock_set_quickack(struct sock *sk, int val);
555int tcp_sock_set_syncnt(struct sock *sk, int val);
556void tcp_sock_set_user_timeout(struct sock *sk, u32 val);
557
558#endif /* _LINUX_TCP_H */
559

source code of linux/include/linux/tcp.h