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.
6 * Definitions for the TCP protocol.
8 * Version: @(#)tcp.h 1.0.2 04/28/93
10 * Author: Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
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.
21 #include <linux/skbuff.h>
22 #include <linux/win_minmax.h>
24 #include <net/inet_connection_sock.h>
25 #include <net/inet_timewait_sock.h>
26 #include <uapi/linux/tcp.h>
28 static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb)
30 return (struct tcphdr *)skb_transport_header(skb);
33 static inline unsigned int __tcp_hdrlen(const struct tcphdr *th)
38 static inline unsigned int tcp_hdrlen(const struct sk_buff *skb)
40 return __tcp_hdrlen(tcp_hdr(skb));
43 static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb)
45 return (struct tcphdr *)skb_inner_transport_header(skb);
48 static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb)
50 return inner_tcp_hdr(skb)->doff * 4;
53 static inline unsigned int tcp_optlen(const struct sk_buff *skb)
55 return (tcp_hdr(skb)->doff - 5) * 4;
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. */
63 /* TCP Fast Open Cookie as stored in memory */
64 struct tcp_fastopen_cookie {
66 u8 val[TCP_FASTOPEN_COOKIE_MAX];
67 #if IS_ENABLED(CONFIG_IPV6)
72 bool exp; /* In RFC6994 experimental option format */
75 /* This defines a selective acknowledgement block. */
76 struct tcp_sack_block_wire {
81 struct tcp_sack_block {
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_FACK_ENABLED (1 << 1) /*1 = FACK is enabled locally*/
89 #define TCP_DSACK_SEEN (1 << 2) /*1 = DSACK was received from peer*/
91 struct tcp_options_received {
93 long ts_recent_stamp;/* Time we stored ts_recent (for aging) */
94 u32 ts_recent; /* Time stamp to echo next */
95 u32 rcv_tsval; /* Time stamp value */
96 u32 rcv_tsecr; /* Time stamp echo reply */
97 u16 saw_tstamp : 1, /* Saw TIMESTAMP on last packet */
98 tstamp_ok : 1, /* TIMESTAMP seen on SYN packet */
99 dsack : 1, /* D-SACK is scheduled */
100 wscale_ok : 1, /* Wscale seen on SYN packet */
101 sack_ok : 4, /* SACK 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 */
109 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
111 rx_opt->tstamp_ok = rx_opt->sack_ok = 0;
112 rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
115 /* This is the max number of SACKS that we'll generate and process. It's safe
116 * to increase this, although since:
117 * size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8)
118 * only four options will fit in a standard TCP header */
119 #define TCP_NUM_SACKS 4
121 struct tcp_request_sock_ops;
123 struct tcp_request_sock {
124 struct inet_request_sock req;
125 const struct tcp_request_sock_ops *af_specific;
126 struct skb_mstamp snt_synack; /* first SYNACK sent time */
132 u32 last_oow_ack_time; /* last SYNACK */
133 u32 rcv_nxt; /* the ack # by SYNACK. For
134 * FastOpen it's the seq#
139 static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
141 return (struct tcp_request_sock *)req;
145 /* inet_connection_sock has to be the first member of tcp_sock */
146 struct inet_connection_sock inet_conn;
147 u16 tcp_header_len; /* Bytes of tcp header to send */
148 u16 gso_segs; /* Max number of segs per GSO packet */
151 * Header prediction flags
152 * 0x5?10 << 16 + snd_wnd in net byte order
157 * RFC793 variables by their proper names. This means you can
158 * read the code and the spec side by side (and laugh ...)
159 * See RFC793 and RFC1122. The RFC writes these in capitals.
161 u64 bytes_received; /* RFC4898 tcpEStatsAppHCThruOctetsReceived
162 * sum(delta(rcv_nxt)), or how many bytes
165 u32 segs_in; /* RFC4898 tcpEStatsPerfSegsIn
166 * total number of segments in.
168 u32 data_segs_in; /* RFC4898 tcpEStatsPerfDataSegsIn
169 * total number of data segments in.
171 u32 rcv_nxt; /* What we want to receive next */
172 u32 copied_seq; /* Head of yet unread data */
173 u32 rcv_wup; /* rcv_nxt on last window update sent */
174 u32 snd_nxt; /* Next sequence we send */
175 u32 segs_out; /* RFC4898 tcpEStatsPerfSegsOut
176 * The total number of segments sent.
178 u32 data_segs_out; /* RFC4898 tcpEStatsPerfDataSegsOut
179 * total number of data segments sent.
181 u64 bytes_acked; /* RFC4898 tcpEStatsAppHCThruOctetsAcked
182 * sum(delta(snd_una)), or how many bytes
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 */
191 u32 tsoffset; /* timestamp offset */
193 struct list_head tsq_node; /* anchor in tsq_tasklet.head list */
195 /* Data for direct copy to user */
197 struct sk_buff_head prequeue;
198 struct task_struct *task;
204 u32 snd_wl1; /* Sequence for window update */
205 u32 snd_wnd; /* The window we expect to receive */
206 u32 max_window; /* Maximal window ever seen from peer */
207 u32 mss_cache; /* Cached effective mss, not including SACKS */
209 u32 window_clamp; /* Maximal window to advertise */
210 u32 rcv_ssthresh; /* Current window clamp */
212 /* Information of the most recently (s)acked skb */
214 struct skb_mstamp mstamp; /* (Re)sent time of the skb */
215 u8 advanced; /* mstamp advanced since last lost marking */
216 u8 reord; /* reordering detected */
218 u16 advmss; /* Advertised MSS */
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? */
224 u8 nonagle : 4,/* Disable Nagle algorithm? */
225 thin_lto : 1,/* Use linear timeouts for thin streams */
226 thin_dupack : 1,/* Fast retransmit on first dupack */
228 frto : 1;/* F-RTO (RFC5682) activated in CA_Loss */
230 u8 do_early_retrans:1,/* Enable RFC5827 early-retransmit */
231 syn_data:1, /* SYN includes data */
232 syn_fastopen:1, /* SYN includes Fast Open option */
233 syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
234 syn_data_acked:1,/* data in SYN is acked by SYN-ACK */
235 save_syn:1, /* Save headers of SYN packet */
236 is_cwnd_limited:1;/* forward progress limited by snd_cwnd? */
237 u32 tlp_high_seq; /* snd_nxt at the time of TLP retransmit. */
239 /* RTT measurement */
240 u32 srtt_us; /* smoothed round trip time << 3 in usecs */
241 u32 mdev_us; /* medium deviation */
242 u32 mdev_max_us; /* maximal mdev for the last rtt period */
243 u32 rttvar_us; /* smoothed mdev_max */
244 u32 rtt_seq; /* sequence number to update rttvar */
245 struct minmax rtt_min;
247 u32 packets_out; /* Packets which are "in flight" */
248 u32 retrans_out; /* Retransmitted packets out */
249 u32 max_packets_out; /* max packets_out in last window */
250 u32 max_packets_seq; /* right edge of max_packets_out flight */
252 u16 urg_data; /* Saved octet of OOB data and control flags */
253 u8 ecn_flags; /* ECN status bits. */
254 u8 keepalive_probes; /* num of allowed keep alive probes */
255 u32 reordering; /* Packet reordering metric. */
256 u32 snd_up; /* Urgent pointer */
259 * Options received (usually on last packet, some only on SYN packets).
261 struct tcp_options_received rx_opt;
264 * Slow start and congestion control (see also Nagle, and Karn & Partridge)
266 u32 snd_ssthresh; /* Slow start size threshold */
267 u32 snd_cwnd; /* Sending congestion window */
268 u32 snd_cwnd_cnt; /* Linear increase counter */
269 u32 snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
272 u32 prior_cwnd; /* Congestion window at start of Recovery. */
273 u32 prr_delivered; /* Number of newly delivered packets to
274 * receiver in Recovery. */
275 u32 prr_out; /* Total number of pkts sent during Recovery. */
276 u32 delivered; /* Total data packets delivered incl. rexmits */
277 u32 lost; /* Total data packets lost incl. rexmits */
278 u32 app_limited; /* limited until "delivered" reaches this val */
279 struct skb_mstamp first_tx_mstamp; /* start of window send phase */
280 struct skb_mstamp delivered_mstamp; /* time we reached "delivered" */
281 u32 rate_delivered; /* saved rate sample: packets delivered */
282 u32 rate_interval_us; /* saved rate sample: time elapsed */
284 u32 rcv_wnd; /* Current receiver window */
285 u32 write_seq; /* Tail(+1) of data held in tcp send buffer */
286 u32 notsent_lowat; /* TCP_NOTSENT_LOWAT */
287 u32 pushed_seq; /* Last pushed seq, required to talk to windows */
288 u32 lost_out; /* Lost packets */
289 u32 sacked_out; /* SACK'd packets */
290 u32 fackets_out; /* FACK'd packets */
292 /* from STCP, retrans queue hinting */
293 struct sk_buff* lost_skb_hint;
294 struct sk_buff *retransmit_skb_hint;
296 /* OOO segments go in this rbtree. Socket lock must be held. */
297 struct rb_root out_of_order_queue;
298 struct sk_buff *ooo_last_skb; /* cache rb_last(out_of_order_queue) */
300 /* SACKs data, these 2 need to be together (see tcp_options_write) */
301 struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
302 struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
304 struct tcp_sack_block recv_sack_cache[4];
306 struct sk_buff *highest_sack; /* skb just after the highest
307 * skb with SACKed bit set
308 * (validity guaranteed only if
313 u32 retransmit_high; /* L-bits may be on up to this seqno */
315 u32 prior_ssthresh; /* ssthresh saved at recovery start */
316 u32 high_seq; /* snd_nxt at onset of congestion */
318 u32 retrans_stamp; /* Timestamp of the last retransmit,
319 * also used in SYN-SENT to remember stamp of
321 u32 undo_marker; /* snd_una upon a new recovery episode. */
322 int undo_retrans; /* number of undoable retransmissions. */
323 u32 total_retrans; /* Total retransmits for entire connection */
325 u32 urg_seq; /* Seq of received urgent pointer */
326 unsigned int keepalive_time; /* time before keep alive takes place */
327 unsigned int keepalive_intvl; /* time interval between keep alive probes */
331 /* Receiver side RTT estimation */
338 /* Receiver queue space */
345 /* TCP-specific MTU probe information. */
350 u32 mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
351 * while socket was owned by user.
354 #ifdef CONFIG_TCP_MD5SIG
355 /* TCP AF-Specific parts; only used by MD5 Signature support so far */
356 const struct tcp_sock_af_ops *af_specific;
358 /* TCP MD5 Signature Option information */
359 struct tcp_md5sig_info __rcu *md5sig_info;
362 /* TCP fastopen related information */
363 struct tcp_fastopen_request *fastopen_req;
364 /* fastopen_rsk points to request_sock that resulted in this big
365 * socket. Used to retransmit SYNACKs etc.
367 struct request_sock *fastopen_rsk;
374 TCP_TSQ_DEFERRED, /* tcp_tasklet_func() found socket was owned */
375 TCP_WRITE_TIMER_DEFERRED, /* tcp_write_timer() found socket was owned */
376 TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */
377 TCP_MTU_REDUCED_DEFERRED, /* tcp_v{4|6}_err() could not call
378 * tcp_v{4|6}_mtu_reduced()
383 TSQF_THROTTLED = (1UL << TSQ_THROTTLED),
384 TSQF_QUEUED = (1UL << TSQ_QUEUED),
385 TCPF_TSQ_DEFERRED = (1UL << TCP_TSQ_DEFERRED),
386 TCPF_WRITE_TIMER_DEFERRED = (1UL << TCP_WRITE_TIMER_DEFERRED),
387 TCPF_DELACK_TIMER_DEFERRED = (1UL << TCP_DELACK_TIMER_DEFERRED),
388 TCPF_MTU_REDUCED_DEFERRED = (1UL << TCP_MTU_REDUCED_DEFERRED),
391 static inline struct tcp_sock *tcp_sk(const struct sock *sk)
393 return (struct tcp_sock *)sk;
396 struct tcp_timewait_sock {
397 struct inet_timewait_sock tw_sk;
398 #define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt
399 #define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt
404 /* The time we sent the last out-of-window ACK: */
405 u32 tw_last_oow_ack_time;
407 long tw_ts_recent_stamp;
408 #ifdef CONFIG_TCP_MD5SIG
409 struct tcp_md5sig_key *tw_md5_key;
413 static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
415 return (struct tcp_timewait_sock *)sk;
418 static inline bool tcp_passive_fastopen(const struct sock *sk)
420 return (sk->sk_state == TCP_SYN_RECV &&
421 tcp_sk(sk)->fastopen_rsk != NULL);
424 static inline void fastopen_queue_tune(struct sock *sk, int backlog)
426 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
427 int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn);
429 queue->fastopenq.max_qlen = min_t(unsigned int, backlog, somaxconn);
432 static inline void tcp_move_syn(struct tcp_sock *tp,
433 struct request_sock *req)
435 tp->saved_syn = req->saved_syn;
436 req->saved_syn = NULL;
439 static inline void tcp_saved_syn_free(struct tcp_sock *tp)
441 kfree(tp->saved_syn);
442 tp->saved_syn = NULL;
445 struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk);
447 #endif /* _LINUX_TCP_H */