1 /* SPDX-License-Identifier: GPL-2.0-or-later */
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.
7 * Definitions for the TCP protocol.
9 * Version: @(#)tcp.h 1.0.2 04/28/93
11 * Author: Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
17 #include <linux/skbuff.h>
18 #include <linux/win_minmax.h>
20 #include <net/inet_connection_sock.h>
21 #include <net/inet_timewait_sock.h>
22 #include <uapi/linux/tcp.h>
24 static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb)
26 return (struct tcphdr *)skb_transport_header(skb);
29 static inline unsigned int __tcp_hdrlen(const struct tcphdr *th)
34 static inline unsigned int tcp_hdrlen(const struct sk_buff *skb)
36 return __tcp_hdrlen(tcp_hdr(skb));
39 static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb)
41 return (struct tcphdr *)skb_inner_transport_header(skb);
44 static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb)
46 return inner_tcp_hdr(skb)->doff * 4;
50 * skb_tcp_all_headers - Returns size of all headers for a TCP packet
53 * Used in TX path, for a packet known to be a TCP one.
55 * if (skb_is_gso(skb)) {
56 * int hlen = skb_tcp_all_headers(skb);
59 static inline int skb_tcp_all_headers(const struct sk_buff *skb)
61 return skb_transport_offset(skb) + tcp_hdrlen(skb);
65 * skb_inner_tcp_all_headers - Returns size of all headers for an encap TCP packet
68 * Used in TX path, for a packet known to be a TCP one.
70 * if (skb_is_gso(skb) && skb->encapsulation) {
71 * int hlen = skb_inner_tcp_all_headers(skb);
74 static inline int skb_inner_tcp_all_headers(const struct sk_buff *skb)
76 return skb_inner_transport_offset(skb) + inner_tcp_hdrlen(skb);
79 static inline unsigned int tcp_optlen(const struct sk_buff *skb)
81 return (tcp_hdr(skb)->doff - 5) * 4;
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. */
89 /* TCP Fast Open Cookie as stored in memory */
90 struct tcp_fastopen_cookie {
91 __le64 val[DIV_ROUND_UP(TCP_FASTOPEN_COOKIE_MAX, sizeof(u64))];
93 bool exp; /* In RFC6994 experimental option format */
96 /* This defines a selective acknowledgement block. */
97 struct tcp_sack_block_wire {
102 struct tcp_sack_block {
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*/
111 struct tcp_options_received {
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 */
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 */
132 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
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)
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
147 struct tcp_request_sock_ops;
149 struct 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 */
156 #if IS_ENABLED(CONFIG_MPTCP)
163 u32 last_oow_ack_time; /* last SYNACK */
164 u32 rcv_nxt; /* the ack # by SYNACK. For
165 * FastOpen it's the seq#
176 static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
178 return (struct tcp_request_sock *)req;
181 static inline bool tcp_rsk_used_ao(const struct request_sock *req)
183 /* The real length of MAC is saved in the request socket,
184 * signing anything with zero-length makes no sense, so here is
187 #ifndef CONFIG_TCP_AO
190 return tcp_rsk(req)->maclen != 0;
194 #define TCP_RMEM_TO_WIN_SCALE 8
197 /* inet_connection_sock has to be the first member of tcp_sock */
198 struct inet_connection_sock inet_conn;
199 u16 tcp_header_len; /* Bytes of tcp header to send */
200 u16 gso_segs; /* Max number of segs per GSO packet */
203 * Header prediction flags
204 * 0x5?10 << 16 + snd_wnd in net byte order
209 * RFC793 variables by their proper names. This means you can
210 * read the code and the spec side by side (and laugh ...)
211 * See RFC793 and RFC1122. The RFC writes these in capitals.
213 u64 bytes_received; /* RFC4898 tcpEStatsAppHCThruOctetsReceived
214 * sum(delta(rcv_nxt)), or how many bytes
217 u32 segs_in; /* RFC4898 tcpEStatsPerfSegsIn
218 * total number of segments in.
220 u32 data_segs_in; /* RFC4898 tcpEStatsPerfDataSegsIn
221 * total number of data segments in.
223 u32 rcv_nxt; /* What we want to receive next */
224 u32 copied_seq; /* Head of yet unread data */
225 u32 rcv_wup; /* rcv_nxt on last window update sent */
226 u32 snd_nxt; /* Next sequence we send */
227 u32 segs_out; /* RFC4898 tcpEStatsPerfSegsOut
228 * The total number of segments sent.
230 u32 data_segs_out; /* RFC4898 tcpEStatsPerfDataSegsOut
231 * total number of data segments sent.
233 u64 bytes_sent; /* RFC4898 tcpEStatsPerfHCDataOctetsOut
234 * total number of data bytes sent.
236 u64 bytes_acked; /* RFC4898 tcpEStatsAppHCThruOctetsAcked
237 * sum(delta(snd_una)), or how many bytes
240 u32 dsack_dups; /* RFC4898 tcpEStatsStackDSACKDups
241 * total number of DSACK blocks received
243 u32 snd_una; /* First byte we want an ack for */
244 u32 snd_sml; /* Last byte of the most recently transmitted small packet */
245 u32 rcv_tstamp; /* timestamp of last received ACK (for keepalives) */
246 u32 lsndtime; /* timestamp of last sent data packet (for restart window) */
247 u32 last_oow_ack_time; /* timestamp of last out-of-window ACK */
248 u32 compressed_ack_rcv_nxt;
250 u32 tsoffset; /* timestamp offset */
252 struct list_head tsq_node; /* anchor in tsq_tasklet.head list */
253 struct list_head tsorted_sent_queue; /* time-sorted sent but un-SACKed skbs */
255 u32 snd_wl1; /* Sequence for window update */
256 u32 snd_wnd; /* The window we expect to receive */
257 u32 max_window; /* Maximal window ever seen from peer */
258 u32 mss_cache; /* Cached effective mss, not including SACKS */
260 u32 window_clamp; /* Maximal window to advertise */
261 u32 rcv_ssthresh; /* Current window clamp */
262 u8 scaling_ratio; /* see tcp_win_from_space() */
263 /* Information of the most recently (s)acked skb */
265 u64 mstamp; /* (Re)sent time of the skb */
266 u32 rtt_us; /* Associated RTT */
267 u32 end_seq; /* Ending TCP sequence of the skb */
268 u32 last_delivered; /* tp->delivered at last reo_wnd adj */
269 u8 reo_wnd_steps; /* Allowed reordering window */
270 #define TCP_RACK_RECOVERY_THRESH 16
271 u8 reo_wnd_persist:5, /* No. of recovery since last adj */
272 dsack_seen:1, /* Whether DSACK seen after last adj */
273 advanced:1; /* mstamp advanced since last lost marking */
275 u16 advmss; /* Advertised MSS */
277 u8 dup_ack_counter:2,
278 tlp_retrans:1, /* TLP is a retransmission */
279 tcp_usec_ts:1, /* TSval values in usec */
281 u32 chrono_start; /* Start time in jiffies of a TCP chrono */
282 u32 chrono_stat[3]; /* Time in jiffies for chrono_stat stats */
283 u8 chrono_type:2, /* current chronograph type */
284 rate_app_limited:1, /* rate_{delivered,interval_us} limited? */
285 fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */
286 fastopen_no_cookie:1, /* Allow send/recv SYN+data without a cookie */
287 is_sack_reneg:1, /* in recovery from loss with SACK reneg? */
288 fastopen_client_fail:2; /* reason why fastopen failed */
289 u8 nonagle : 4,/* Disable Nagle algorithm? */
290 thin_lto : 1,/* Use linear timeouts for thin streams */
291 recvmsg_inq : 1,/* Indicate # of bytes in queue upon recvmsg */
293 frto : 1;/* F-RTO (RFC5682) activated in CA_Loss */
295 u8 save_syn:2, /* Save headers of SYN packet */
296 syn_data:1, /* SYN includes data */
297 syn_fastopen:1, /* SYN includes Fast Open option */
298 syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
299 syn_fastopen_ch:1, /* Active TFO re-enabling probe */
300 syn_data_acked:1,/* data in SYN is acked by SYN-ACK */
301 is_cwnd_limited:1;/* forward progress limited by snd_cwnd? */
302 u32 tlp_high_seq; /* snd_nxt at the time of TLP */
304 u32 tcp_tx_delay; /* delay (in usec) added to TX packets */
305 u64 tcp_wstamp_ns; /* departure time for next sent data packet */
306 u64 tcp_clock_cache; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */
308 /* RTT measurement */
309 u64 tcp_mstamp; /* most recent packet received/sent */
310 u32 srtt_us; /* smoothed round trip time << 3 in usecs */
311 u32 mdev_us; /* medium deviation */
312 u32 mdev_max_us; /* maximal mdev for the last rtt period */
313 u32 rttvar_us; /* smoothed mdev_max */
314 u32 rtt_seq; /* sequence number to update rttvar */
315 struct minmax rtt_min;
317 u32 packets_out; /* Packets which are "in flight" */
318 u32 retrans_out; /* Retransmitted packets out */
319 u32 max_packets_out; /* max packets_out in last window */
320 u32 cwnd_usage_seq; /* right edge of cwnd usage tracking flight */
322 u16 urg_data; /* Saved octet of OOB data and control flags */
323 u8 ecn_flags; /* ECN status bits. */
324 u8 keepalive_probes; /* num of allowed keep alive probes */
325 u32 reordering; /* Packet reordering metric. */
326 u32 reord_seen; /* number of data packet reordering events */
327 u32 snd_up; /* Urgent pointer */
330 * Options received (usually on last packet, some only on SYN packets).
332 struct tcp_options_received rx_opt;
335 * Slow start and congestion control (see also Nagle, and Karn & Partridge)
337 u32 snd_ssthresh; /* Slow start size threshold */
338 u32 snd_cwnd; /* Sending congestion window */
339 u32 snd_cwnd_cnt; /* Linear increase counter */
340 u32 snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
343 u32 prior_cwnd; /* cwnd right before starting loss recovery */
344 u32 prr_delivered; /* Number of newly delivered packets to
345 * receiver in Recovery. */
346 u32 prr_out; /* Total number of pkts sent during Recovery. */
347 u32 delivered; /* Total data packets delivered incl. rexmits */
348 u32 delivered_ce; /* Like the above but only ECE marked packets */
349 u32 lost; /* Total data packets lost incl. rexmits */
350 u32 app_limited; /* limited until "delivered" reaches this val */
351 u64 first_tx_mstamp; /* start of window send phase */
352 u64 delivered_mstamp; /* time we reached "delivered" */
353 u32 rate_delivered; /* saved rate sample: packets delivered */
354 u32 rate_interval_us; /* saved rate sample: time elapsed */
356 u32 rcv_wnd; /* Current receiver window */
357 u32 write_seq; /* Tail(+1) of data held in tcp send buffer */
358 u32 notsent_lowat; /* TCP_NOTSENT_LOWAT */
359 u32 pushed_seq; /* Last pushed seq, required to talk to windows */
360 u32 lost_out; /* Lost packets */
361 u32 sacked_out; /* SACK'd packets */
363 struct hrtimer pacing_timer;
364 struct hrtimer compressed_ack_timer;
366 /* from STCP, retrans queue hinting */
367 struct sk_buff* lost_skb_hint;
368 struct sk_buff *retransmit_skb_hint;
370 /* OOO segments go in this rbtree. Socket lock must be held. */
371 struct rb_root out_of_order_queue;
372 struct sk_buff *ooo_last_skb; /* cache rb_last(out_of_order_queue) */
374 /* SACKs data, these 2 need to be together (see tcp_options_write) */
375 struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
376 struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
378 struct tcp_sack_block recv_sack_cache[4];
380 struct sk_buff *highest_sack; /* skb just after the highest
381 * skb with SACKed bit set
382 * (validity guaranteed only if
388 u32 prior_ssthresh; /* ssthresh saved at recovery start */
389 u32 high_seq; /* snd_nxt at onset of congestion */
391 u32 retrans_stamp; /* Timestamp of the last retransmit,
392 * also used in SYN-SENT to remember stamp of
394 u32 undo_marker; /* snd_una upon a new recovery episode. */
395 int undo_retrans; /* number of undoable retransmissions. */
396 u64 bytes_retrans; /* RFC4898 tcpEStatsPerfOctetsRetrans
397 * Total data bytes retransmitted
399 u32 total_retrans; /* Total retransmits for entire connection */
400 u32 rto_stamp; /* Start time (ms) of last CA_Loss recovery */
401 u16 total_rto; /* Total number of RTO timeouts, including
402 * SYN/SYN-ACK and recurring timeouts.
404 u16 total_rto_recoveries; /* Total number of RTO recoveries,
405 * including any unfinished recovery.
407 u32 total_rto_time; /* ms spent in (completed) RTO recoveries. */
409 u32 urg_seq; /* Seq of received urgent pointer */
410 unsigned int keepalive_time; /* time before keep alive takes place */
411 unsigned int keepalive_intvl; /* time interval between keep alive probes */
416 /* Sock_ops bpf program related variables */
418 u8 bpf_sock_ops_cb_flags; /* Control calling BPF programs
419 * values defined in uapi/linux/tcp.h
421 u8 bpf_chg_cc_inprogress:1; /* In the middle of
422 * bpf_setsockopt(TCP_CONGESTION),
423 * it is to avoid the bpf_tcp_cc->init()
424 * to recur itself by calling
425 * bpf_setsockopt(TCP_CONGESTION, "itself").
427 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) (TP->bpf_sock_ops_cb_flags & ARG)
429 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) 0
432 u16 timeout_rehash; /* Timeout-triggered rehash attempts */
434 u32 rcv_ooopack; /* Received out-of-order packets, for tcpinfo */
436 /* Receiver side RTT estimation */
437 u32 rcv_rtt_last_tsecr;
444 /* Receiver queue space */
451 /* TCP-specific MTU probe information. */
456 u32 plb_rehash; /* PLB-triggered rehash attempts */
457 u32 mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
458 * while socket was owned by user.
460 #if IS_ENABLED(CONFIG_MPTCP)
463 #if IS_ENABLED(CONFIG_SMC)
464 bool (*smc_hs_congested)(const struct sock *sk);
465 bool syn_smc; /* SYN includes SMC */
468 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
469 /* TCP AF-Specific parts; only used by TCP-AO/MD5 Signature support so far */
470 const struct tcp_sock_af_ops *af_specific;
472 #ifdef CONFIG_TCP_MD5SIG
473 /* TCP MD5 Signature Option information */
474 struct tcp_md5sig_info __rcu *md5sig_info;
477 struct tcp_ao_info __rcu *ao_info;
481 /* TCP fastopen related information */
482 struct tcp_fastopen_request *fastopen_req;
483 /* fastopen_rsk points to request_sock that resulted in this big
484 * socket. Used to retransmit SYNACKs etc.
486 struct request_sock __rcu *fastopen_rsk;
487 struct saved_syn *saved_syn;
493 TCP_TSQ_DEFERRED, /* tcp_tasklet_func() found socket was owned */
494 TCP_WRITE_TIMER_DEFERRED, /* tcp_write_timer() found socket was owned */
495 TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */
496 TCP_MTU_REDUCED_DEFERRED, /* tcp_v{4|6}_err() could not call
497 * tcp_v{4|6}_mtu_reduced()
499 TCP_ACK_DEFERRED, /* TX pure ack is deferred */
503 TSQF_THROTTLED = BIT(TSQ_THROTTLED),
504 TSQF_QUEUED = BIT(TSQ_QUEUED),
505 TCPF_TSQ_DEFERRED = BIT(TCP_TSQ_DEFERRED),
506 TCPF_WRITE_TIMER_DEFERRED = BIT(TCP_WRITE_TIMER_DEFERRED),
507 TCPF_DELACK_TIMER_DEFERRED = BIT(TCP_DELACK_TIMER_DEFERRED),
508 TCPF_MTU_REDUCED_DEFERRED = BIT(TCP_MTU_REDUCED_DEFERRED),
509 TCPF_ACK_DEFERRED = BIT(TCP_ACK_DEFERRED),
512 #define tcp_sk(ptr) container_of_const(ptr, struct tcp_sock, inet_conn.icsk_inet.sk)
514 /* Variant of tcp_sk() upgrading a const sock to a read/write tcp socket.
515 * Used in context of (lockless) tcp listeners.
517 #define tcp_sk_rw(ptr) container_of(ptr, struct tcp_sock, inet_conn.icsk_inet.sk)
519 struct tcp_timewait_sock {
520 struct inet_timewait_sock tw_sk;
521 #define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt
522 #define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt
527 /* The time we sent the last out-of-window ACK: */
528 u32 tw_last_oow_ack_time;
530 int tw_ts_recent_stamp;
532 #ifdef CONFIG_TCP_MD5SIG
533 struct tcp_md5sig_key *tw_md5_key;
536 struct tcp_ao_info __rcu *ao_info;
540 static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
542 return (struct tcp_timewait_sock *)sk;
545 static inline bool tcp_passive_fastopen(const struct sock *sk)
547 return sk->sk_state == TCP_SYN_RECV &&
548 rcu_access_pointer(tcp_sk(sk)->fastopen_rsk) != NULL;
551 static inline void fastopen_queue_tune(struct sock *sk, int backlog)
553 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
554 int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn);
556 WRITE_ONCE(queue->fastopenq.max_qlen, min_t(unsigned int, backlog, somaxconn));
559 static inline void tcp_move_syn(struct tcp_sock *tp,
560 struct request_sock *req)
562 tp->saved_syn = req->saved_syn;
563 req->saved_syn = NULL;
566 static inline void tcp_saved_syn_free(struct tcp_sock *tp)
568 kfree(tp->saved_syn);
569 tp->saved_syn = NULL;
572 static inline u32 tcp_saved_syn_len(const struct saved_syn *saved_syn)
574 return saved_syn->mac_hdrlen + saved_syn->network_hdrlen +
575 saved_syn->tcp_hdrlen;
578 struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk,
579 const struct sk_buff *orig_skb,
580 const struct sk_buff *ack_skb);
582 static inline u16 tcp_mss_clamp(const struct tcp_sock *tp, u16 mss)
584 /* We use READ_ONCE() here because socket might not be locked.
585 * This happens for listeners.
587 u16 user_mss = READ_ONCE(tp->rx_opt.user_mss);
589 return (user_mss && user_mss < mss) ? user_mss : mss;
592 int tcp_skb_shift(struct sk_buff *to, struct sk_buff *from, int pcount,
595 void __tcp_sock_set_cork(struct sock *sk, bool on);
596 void tcp_sock_set_cork(struct sock *sk, bool on);
597 int tcp_sock_set_keepcnt(struct sock *sk, int val);
598 int tcp_sock_set_keepidle_locked(struct sock *sk, int val);
599 int tcp_sock_set_keepidle(struct sock *sk, int val);
600 int tcp_sock_set_keepintvl(struct sock *sk, int val);
601 void __tcp_sock_set_nodelay(struct sock *sk, bool on);
602 void tcp_sock_set_nodelay(struct sock *sk);
603 void tcp_sock_set_quickack(struct sock *sk, int val);
604 int tcp_sock_set_syncnt(struct sock *sk, int val);
605 int tcp_sock_set_user_timeout(struct sock *sk, int val);
607 static inline bool dst_tcp_usec_ts(const struct dst_entry *dst)
609 return dst_feature(dst, RTAX_FEATURE_TCP_USEC_TS);
612 #endif /* _LINUX_TCP_H */