1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Definitions for the TCP protocol.
7 *
8 * Version: @(#)tcp.h 1.0.2 04/28/93
9 *
10 * Author: Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 *
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
16 */
17#ifndef _LINUX_TCP_H
18#define _LINUX_TCP_H
19
20
21#include <linux/skbuff.h>
22#include <linux/win_minmax.h>
23#include <net/sock.h>
24#include <net/inet_connection_sock.h>
25#include <net/inet_timewait_sock.h>
26#include <uapi/linux/tcp.h>
27
28static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb)
29{
30 return (struct tcphdr *)skb_transport_header(skb);
31}
32
33static inline unsigned int __tcp_hdrlen(const struct tcphdr *th)
34{
35 return th->doff * 4;
36}
37
38static inline unsigned int tcp_hdrlen(const struct sk_buff *skb)
39{
40 return __tcp_hdrlen(tcp_hdr(skb));
41}
42
43static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb)
44{
45 return (struct tcphdr *)skb_inner_transport_header(skb);
46}
47
48static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb)
49{
50 return inner_tcp_hdr(skb)->doff * 4;
51}
52
53static inline unsigned int tcp_optlen(const struct sk_buff *skb)
54{
55 return (tcp_hdr(skb)->doff - 5) * 4;
56}
57
58/* TCP Fast Open */
59#define TCP_FASTOPEN_COOKIE_MIN 4 /* Min Fast Open Cookie size in bytes */
60#define TCP_FASTOPEN_COOKIE_MAX 16 /* Max Fast Open Cookie size in bytes */
61#define TCP_FASTOPEN_COOKIE_SIZE 8 /* the size employed by this impl. */
62
63/* TCP Fast Open Cookie as stored in memory */
64struct tcp_fastopen_cookie {
65 union {
66 u8 val[TCP_FASTOPEN_COOKIE_MAX];
67#if IS_ENABLED(CONFIG_IPV6)
68 struct in6_addr addr;
69#endif
70 };
71 s8 len;
72 bool exp; /* In RFC6994 experimental option format */
73};
74
75/* This defines a selective acknowledgement block. */
76struct tcp_sack_block_wire {
77 __be32 start_seq;
78 __be32 end_seq;
79};
80
81struct tcp_sack_block {
82 u32 start_seq;
83 u32 end_seq;
84};
85
86/*These are used to set the sack_ok field in struct tcp_options_received */
87#define TCP_SACK_SEEN (1 << 0) /*1 = peer is SACK capable, */
88#define TCP_DSACK_SEEN (1 << 2) /*1 = DSACK was received from peer*/
89
90struct tcp_options_received {
91/* PAWS/RTTM data */
92 int ts_recent_stamp;/* Time we stored ts_recent (for aging) */
93 u32 ts_recent; /* Time stamp to echo next */
94 u32 rcv_tsval; /* Time stamp value */
95 u32 rcv_tsecr; /* Time stamp echo reply */
96 u16 saw_tstamp : 1, /* Saw TIMESTAMP on last packet */
97 tstamp_ok : 1, /* TIMESTAMP seen on SYN packet */
98 dsack : 1, /* D-SACK is scheduled */
99 wscale_ok : 1, /* Wscale seen on SYN packet */
100 sack_ok : 3, /* SACK seen on SYN packet */
101 smc_ok : 1, /* SMC seen on SYN packet */
102 snd_wscale : 4, /* Window scaling received from sender */
103 rcv_wscale : 4; /* Window scaling to send to receiver */
104 u8 num_sacks; /* Number of SACK blocks */
105 u16 user_mss; /* mss requested by user in ioctl */
106 u16 mss_clamp; /* Maximal mss, negotiated at connection setup */
107};
108
109static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
110{
111 rx_opt->tstamp_ok = rx_opt->sack_ok = 0;
112 rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
113#if IS_ENABLED(CONFIG_SMC)
114 rx_opt->smc_ok = 0;
115#endif
116}
117
118/* This is the max number of SACKS that we'll generate and process. It's safe
119 * to increase this, although since:
120 * size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8)
121 * only four options will fit in a standard TCP header */
122#define TCP_NUM_SACKS 4
123
124struct tcp_request_sock_ops;
125
126struct tcp_request_sock {
127 struct inet_request_sock req;
128 const struct tcp_request_sock_ops *af_specific;
129 u64 snt_synack; /* first SYNACK sent time */
130 bool tfo_listener;
131 u32 txhash;
132 u32 rcv_isn;
133 u32 snt_isn;
134 u32 ts_off;
135 u32 last_oow_ack_time; /* last SYNACK */
136 u32 rcv_nxt; /* the ack # by SYNACK. For
137 * FastOpen it's the seq#
138 * after data-in-SYN.
139 */
140};
141
142static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
143{
144 return (struct tcp_request_sock *)req;
145}
146
147struct tcp_sock {
148 /* inet_connection_sock has to be the first member of tcp_sock */
149 struct inet_connection_sock inet_conn;
150 u16 tcp_header_len; /* Bytes of tcp header to send */
151 u16 gso_segs; /* Max number of segs per GSO packet */
152
153/*
154 * Header prediction flags
155 * 0x5?10 << 16 + snd_wnd in net byte order
156 */
157 __be32 pred_flags;
158
159/*
160 * RFC793 variables by their proper names. This means you can
161 * read the code and the spec side by side (and laugh ...)
162 * See RFC793 and RFC1122. The RFC writes these in capitals.
163 */
164 u64 bytes_received; /* RFC4898 tcpEStatsAppHCThruOctetsReceived
165 * sum(delta(rcv_nxt)), or how many bytes
166 * were acked.
167 */
168 u32 segs_in; /* RFC4898 tcpEStatsPerfSegsIn
169 * total number of segments in.
170 */
171 u32 data_segs_in; /* RFC4898 tcpEStatsPerfDataSegsIn
172 * total number of data segments in.
173 */
174 u32 rcv_nxt; /* What we want to receive next */
175 u32 copied_seq; /* Head of yet unread data */
176 u32 rcv_wup; /* rcv_nxt on last window update sent */
177 u32 snd_nxt; /* Next sequence we send */
178 u32 segs_out; /* RFC4898 tcpEStatsPerfSegsOut
179 * The total number of segments sent.
180 */
181 u32 data_segs_out; /* RFC4898 tcpEStatsPerfDataSegsOut
182 * total number of data segments sent.
183 */
184 u64 bytes_sent; /* RFC4898 tcpEStatsPerfHCDataOctetsOut
185 * total number of data bytes sent.
186 */
187 u64 bytes_acked; /* RFC4898 tcpEStatsAppHCThruOctetsAcked
188 * sum(delta(snd_una)), or how many bytes
189 * were acked.
190 */
191 u32 dsack_dups; /* RFC4898 tcpEStatsStackDSACKDups
192 * total number of DSACK blocks received
193 */
194 u32 snd_una; /* First byte we want an ack for */
195 u32 snd_sml; /* Last byte of the most recently transmitted small packet */
196 u32 rcv_tstamp; /* timestamp of last received ACK (for keepalives) */
197 u32 lsndtime; /* timestamp of last sent data packet (for restart window) */
198 u32 last_oow_ack_time; /* timestamp of last out-of-window ACK */
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 u32 chrono_start; /* Start time in jiffies of a TCP chrono */
228 u32 chrono_stat[3]; /* Time in jiffies for chrono_stat stats */
229 u8 chrono_type:2, /* current chronograph type */
230 rate_app_limited:1, /* rate_{delivered,interval_us} limited? */
231 fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */
232 fastopen_no_cookie:1, /* Allow send/recv SYN+data without a cookie */
233 is_sack_reneg:1, /* in recovery from loss with SACK reneg? */
234 unused:2;
235 u8 nonagle : 4,/* Disable Nagle algorithm? */
236 thin_lto : 1,/* Use linear timeouts for thin streams */
237 recvmsg_inq : 1,/* Indicate # of bytes in queue upon recvmsg */
238 repair : 1,
239 frto : 1;/* F-RTO (RFC5682) activated in CA_Loss */
240 u8 repair_queue;
241 u8 syn_data:1, /* SYN includes data */
242 syn_fastopen:1, /* SYN includes Fast Open option */
243 syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
244 syn_fastopen_ch:1, /* Active TFO re-enabling probe */
245 syn_data_acked:1,/* data in SYN is acked by SYN-ACK */
246 save_syn:1, /* Save headers of SYN packet */
247 is_cwnd_limited:1,/* forward progress limited by snd_cwnd? */
248 syn_smc:1; /* SYN includes SMC */
249 u32 tlp_high_seq; /* snd_nxt at the time of TLP retransmit. */
250
251/* RTT measurement */
252 u64 tcp_mstamp; /* most recent packet received/sent */
253 u32 srtt_us; /* smoothed round trip time << 3 in usecs */
254 u32 mdev_us; /* medium deviation */
255 u32 mdev_max_us; /* maximal mdev for the last rtt period */
256 u32 rttvar_us; /* smoothed mdev_max */
257 u32 rtt_seq; /* sequence number to update rttvar */
258 struct minmax rtt_min;
259
260 u32 packets_out; /* Packets which are "in flight" */
261 u32 retrans_out; /* Retransmitted packets out */
262 u32 max_packets_out; /* max packets_out in last window */
263 u32 max_packets_seq; /* right edge of max_packets_out flight */
264
265 u16 urg_data; /* Saved octet of OOB data and control flags */
266 u8 ecn_flags; /* ECN status bits. */
267 u8 keepalive_probes; /* num of allowed keep alive probes */
268 u32 reordering; /* Packet reordering metric. */
269 u32 reord_seen; /* number of data packet reordering events */
270 u32 snd_up; /* Urgent pointer */
271
272/*
273 * Options received (usually on last packet, some only on SYN packets).
274 */
275 struct tcp_options_received rx_opt;
276
277/*
278 * Slow start and congestion control (see also Nagle, and Karn & Partridge)
279 */
280 u32 snd_ssthresh; /* Slow start size threshold */
281 u32 snd_cwnd; /* Sending congestion window */
282 u32 snd_cwnd_cnt; /* Linear increase counter */
283 u32 snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
284 u32 snd_cwnd_used;
285 u32 snd_cwnd_stamp;
286 u32 prior_cwnd; /* cwnd right before starting loss recovery */
287 u32 prr_delivered; /* Number of newly delivered packets to
288 * receiver in Recovery. */
289 u32 prr_out; /* Total number of pkts sent during Recovery. */
290 u32 delivered; /* Total data packets delivered incl. rexmits */
291 u32 delivered_ce; /* Like the above but only ECE marked packets */
292 u32 lost; /* Total data packets lost incl. rexmits */
293 u32 app_limited; /* limited until "delivered" reaches this val */
294 u64 first_tx_mstamp; /* start of window send phase */
295 u64 delivered_mstamp; /* time we reached "delivered" */
296 u32 rate_delivered; /* saved rate sample: packets delivered */
297 u32 rate_interval_us; /* saved rate sample: time elapsed */
298
299 u32 rcv_wnd; /* Current receiver window */
300 u32 write_seq; /* Tail(+1) of data held in tcp send buffer */
301 u32 notsent_lowat; /* TCP_NOTSENT_LOWAT */
302 u32 pushed_seq; /* Last pushed seq, required to talk to windows */
303 u32 lost_out; /* Lost packets */
304 u32 sacked_out; /* SACK'd packets */
305
306 struct hrtimer pacing_timer;
307 struct hrtimer compressed_ack_timer;
308
309 /* from STCP, retrans queue hinting */
310 struct sk_buff* lost_skb_hint;
311 struct sk_buff *retransmit_skb_hint;
312
313 /* OOO segments go in this rbtree. Socket lock must be held. */
314 struct rb_root out_of_order_queue;
315 struct sk_buff *ooo_last_skb; /* cache rb_last(out_of_order_queue) */
316
317 /* SACKs data, these 2 need to be together (see tcp_options_write) */
318 struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
319 struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
320
321 struct tcp_sack_block recv_sack_cache[4];
322
323 struct sk_buff *highest_sack; /* skb just after the highest
324 * skb with SACKed bit set
325 * (validity guaranteed only if
326 * sacked_out > 0)
327 */
328
329 int lost_cnt_hint;
330
331 u32 prior_ssthresh; /* ssthresh saved at recovery start */
332 u32 high_seq; /* snd_nxt at onset of congestion */
333
334 u32 retrans_stamp; /* Timestamp of the last retransmit,
335 * also used in SYN-SENT to remember stamp of
336 * the first SYN. */
337 u32 undo_marker; /* snd_una upon a new recovery episode. */
338 int undo_retrans; /* number of undoable retransmissions. */
339 u64 bytes_retrans; /* RFC4898 tcpEStatsPerfOctetsRetrans
340 * Total data bytes retransmitted
341 */
342 u32 total_retrans; /* Total retransmits for entire connection */
343
344 u32 urg_seq; /* Seq of received urgent pointer */
345 unsigned int keepalive_time; /* time before keep alive takes place */
346 unsigned int keepalive_intvl; /* time interval between keep alive probes */
347
348 int linger2;
349
350
351/* Sock_ops bpf program related variables */
352#ifdef CONFIG_BPF
353 u8 bpf_sock_ops_cb_flags; /* Control calling BPF programs
354 * values defined in uapi/linux/tcp.h
355 */
356#define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) (TP->bpf_sock_ops_cb_flags & ARG)
357#else
358#define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) 0
359#endif
360
361/* Receiver side RTT estimation */
362 u32 rcv_rtt_last_tsecr;
363 struct {
364 u32 rtt_us;
365 u32 seq;
366 u64 time;
367 } rcv_rtt_est;
368
369/* Receiver queue space */
370 struct {
371 u32 space;
372 u32 seq;
373 u64 time;
374 } rcvq_space;
375
376/* TCP-specific MTU probe information. */
377 struct {
378 u32 probe_seq_start;
379 u32 probe_seq_end;
380 } mtu_probe;
381 u32 mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
382 * while socket was owned by user.
383 */
384
385#ifdef CONFIG_TCP_MD5SIG
386/* TCP AF-Specific parts; only used by MD5 Signature support so far */
387 const struct tcp_sock_af_ops *af_specific;
388
389/* TCP MD5 Signature Option information */
390 struct tcp_md5sig_info __rcu *md5sig_info;
391#endif
392
393/* TCP fastopen related information */
394 struct tcp_fastopen_request *fastopen_req;
395 /* fastopen_rsk points to request_sock that resulted in this big
396 * socket. Used to retransmit SYNACKs etc.
397 */
398 struct request_sock *fastopen_rsk;
399 u32 *saved_syn;
400};
401
402enum tsq_enum {
403 TSQ_THROTTLED,
404 TSQ_QUEUED,
405 TCP_TSQ_DEFERRED, /* tcp_tasklet_func() found socket was owned */
406 TCP_WRITE_TIMER_DEFERRED, /* tcp_write_timer() found socket was owned */
407 TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */
408 TCP_MTU_REDUCED_DEFERRED, /* tcp_v{4|6}_err() could not call
409 * tcp_v{4|6}_mtu_reduced()
410 */
411};
412
413enum tsq_flags {
414 TSQF_THROTTLED = (1UL << TSQ_THROTTLED),
415 TSQF_QUEUED = (1UL << TSQ_QUEUED),
416 TCPF_TSQ_DEFERRED = (1UL << TCP_TSQ_DEFERRED),
417 TCPF_WRITE_TIMER_DEFERRED = (1UL << TCP_WRITE_TIMER_DEFERRED),
418 TCPF_DELACK_TIMER_DEFERRED = (1UL << TCP_DELACK_TIMER_DEFERRED),
419 TCPF_MTU_REDUCED_DEFERRED = (1UL << TCP_MTU_REDUCED_DEFERRED),
420};
421
422static inline struct tcp_sock *tcp_sk(const struct sock *sk)
423{
424 return (struct tcp_sock *)sk;
425}
426
427struct tcp_timewait_sock {
428 struct inet_timewait_sock tw_sk;
429#define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt
430#define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt
431 u32 tw_rcv_wnd;
432 u32 tw_ts_offset;
433 u32 tw_ts_recent;
434
435 /* The time we sent the last out-of-window ACK: */
436 u32 tw_last_oow_ack_time;
437
438 int tw_ts_recent_stamp;
439#ifdef CONFIG_TCP_MD5SIG
440 struct tcp_md5sig_key *tw_md5_key;
441#endif
442};
443
444static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
445{
446 return (struct tcp_timewait_sock *)sk;
447}
448
449static inline bool tcp_passive_fastopen(const struct sock *sk)
450{
451 return (sk->sk_state == TCP_SYN_RECV &&
452 tcp_sk(sk)->fastopen_rsk != NULL);
453}
454
455static inline void fastopen_queue_tune(struct sock *sk, int backlog)
456{
457 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
458 int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn);
459
460 queue->fastopenq.max_qlen = min_t(unsigned int, backlog, somaxconn);
461}
462
463static inline void tcp_move_syn(struct tcp_sock *tp,
464 struct request_sock *req)
465{
466 tp->saved_syn = req->saved_syn;
467 req->saved_syn = NULL;
468}
469
470static inline void tcp_saved_syn_free(struct tcp_sock *tp)
471{
472 kfree(tp->saved_syn);
473 tp->saved_syn = NULL;
474}
475
476struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk);
477
478static inline u16 tcp_mss_clamp(const struct tcp_sock *tp, u16 mss)
479{
480 /* We use READ_ONCE() here because socket might not be locked.
481 * This happens for listeners.
482 */
483 u16 user_mss = READ_ONCE(tp->rx_opt.user_mss);
484
485 return (user_mss && user_mss < mss) ? user_mss : mss;
486}
487#endif /* _LINUX_TCP_H */
488