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