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 #ifndef CONFIG_TCP_AO
186 return tcp_rsk(req
)->used_tcp_ao
;
190 #define TCP_RMEM_TO_WIN_SCALE 8
193 /* inet_connection_sock has to be the first member of tcp_sock */
194 struct inet_connection_sock inet_conn
;
195 u16 tcp_header_len
; /* Bytes of tcp header to send */
196 u16 gso_segs
; /* Max number of segs per GSO packet */
199 * Header prediction flags
200 * 0x5?10 << 16 + snd_wnd in net byte order
205 * RFC793 variables by their proper names. This means you can
206 * read the code and the spec side by side (and laugh ...)
207 * See RFC793 and RFC1122. The RFC writes these in capitals.
209 u64 bytes_received
; /* RFC4898 tcpEStatsAppHCThruOctetsReceived
210 * sum(delta(rcv_nxt)), or how many bytes
213 u32 segs_in
; /* RFC4898 tcpEStatsPerfSegsIn
214 * total number of segments in.
216 u32 data_segs_in
; /* RFC4898 tcpEStatsPerfDataSegsIn
217 * total number of data segments in.
219 u32 rcv_nxt
; /* What we want to receive next */
220 u32 copied_seq
; /* Head of yet unread data */
221 u32 rcv_wup
; /* rcv_nxt on last window update sent */
222 u32 snd_nxt
; /* Next sequence we send */
223 u32 segs_out
; /* RFC4898 tcpEStatsPerfSegsOut
224 * The total number of segments sent.
226 u32 data_segs_out
; /* RFC4898 tcpEStatsPerfDataSegsOut
227 * total number of data segments sent.
229 u64 bytes_sent
; /* RFC4898 tcpEStatsPerfHCDataOctetsOut
230 * total number of data bytes sent.
232 u64 bytes_acked
; /* RFC4898 tcpEStatsAppHCThruOctetsAcked
233 * sum(delta(snd_una)), or how many bytes
236 u32 dsack_dups
; /* RFC4898 tcpEStatsStackDSACKDups
237 * total number of DSACK blocks received
239 u32 snd_una
; /* First byte we want an ack for */
240 u32 snd_sml
; /* Last byte of the most recently transmitted small packet */
241 u32 rcv_tstamp
; /* timestamp of last received ACK (for keepalives) */
242 u32 lsndtime
; /* timestamp of last sent data packet (for restart window) */
243 u32 last_oow_ack_time
; /* timestamp of last out-of-window ACK */
244 u32 compressed_ack_rcv_nxt
;
246 u32 tsoffset
; /* timestamp offset */
248 struct list_head tsq_node
; /* anchor in tsq_tasklet.head list */
249 struct list_head tsorted_sent_queue
; /* time-sorted sent but un-SACKed skbs */
251 u32 snd_wl1
; /* Sequence for window update */
252 u32 snd_wnd
; /* The window we expect to receive */
253 u32 max_window
; /* Maximal window ever seen from peer */
254 u32 mss_cache
; /* Cached effective mss, not including SACKS */
256 u32 window_clamp
; /* Maximal window to advertise */
257 u32 rcv_ssthresh
; /* Current window clamp */
258 u8 scaling_ratio
; /* see tcp_win_from_space() */
259 /* Information of the most recently (s)acked skb */
261 u64 mstamp
; /* (Re)sent time of the skb */
262 u32 rtt_us
; /* Associated RTT */
263 u32 end_seq
; /* Ending TCP sequence of the skb */
264 u32 last_delivered
; /* tp->delivered at last reo_wnd adj */
265 u8 reo_wnd_steps
; /* Allowed reordering window */
266 #define TCP_RACK_RECOVERY_THRESH 16
267 u8 reo_wnd_persist
:5, /* No. of recovery since last adj */
268 dsack_seen
:1, /* Whether DSACK seen after last adj */
269 advanced
:1; /* mstamp advanced since last lost marking */
271 u16 advmss
; /* Advertised MSS */
273 u8 dup_ack_counter
:2,
274 tlp_retrans
:1, /* TLP is a retransmission */
275 tcp_usec_ts
:1, /* TSval values in usec */
277 u32 chrono_start
; /* Start time in jiffies of a TCP chrono */
278 u32 chrono_stat
[3]; /* Time in jiffies for chrono_stat stats */
279 u8 chrono_type
:2, /* current chronograph type */
280 rate_app_limited
:1, /* rate_{delivered,interval_us} limited? */
281 fastopen_connect
:1, /* FASTOPEN_CONNECT sockopt */
282 fastopen_no_cookie
:1, /* Allow send/recv SYN+data without a cookie */
283 is_sack_reneg
:1, /* in recovery from loss with SACK reneg? */
284 fastopen_client_fail
:2; /* reason why fastopen failed */
285 u8 nonagle
: 4,/* Disable Nagle algorithm? */
286 thin_lto
: 1,/* Use linear timeouts for thin streams */
287 recvmsg_inq
: 1,/* Indicate # of bytes in queue upon recvmsg */
289 frto
: 1;/* F-RTO (RFC5682) activated in CA_Loss */
291 u8 save_syn
:2, /* Save headers of SYN packet */
292 syn_data
:1, /* SYN includes data */
293 syn_fastopen
:1, /* SYN includes Fast Open option */
294 syn_fastopen_exp
:1,/* SYN includes Fast Open exp. option */
295 syn_fastopen_ch
:1, /* Active TFO re-enabling probe */
296 syn_data_acked
:1,/* data in SYN is acked by SYN-ACK */
297 is_cwnd_limited
:1;/* forward progress limited by snd_cwnd? */
298 u32 tlp_high_seq
; /* snd_nxt at the time of TLP */
300 u32 tcp_tx_delay
; /* delay (in usec) added to TX packets */
301 u64 tcp_wstamp_ns
; /* departure time for next sent data packet */
302 u64 tcp_clock_cache
; /* cache last tcp_clock_ns() (see tcp_mstamp_refresh()) */
304 /* RTT measurement */
305 u64 tcp_mstamp
; /* most recent packet received/sent */
306 u32 srtt_us
; /* smoothed round trip time << 3 in usecs */
307 u32 mdev_us
; /* medium deviation */
308 u32 mdev_max_us
; /* maximal mdev for the last rtt period */
309 u32 rttvar_us
; /* smoothed mdev_max */
310 u32 rtt_seq
; /* sequence number to update rttvar */
311 struct minmax rtt_min
;
313 u32 packets_out
; /* Packets which are "in flight" */
314 u32 retrans_out
; /* Retransmitted packets out */
315 u32 max_packets_out
; /* max packets_out in last window */
316 u32 cwnd_usage_seq
; /* right edge of cwnd usage tracking flight */
318 u16 urg_data
; /* Saved octet of OOB data and control flags */
319 u8 ecn_flags
; /* ECN status bits. */
320 u8 keepalive_probes
; /* num of allowed keep alive probes */
321 u32 reordering
; /* Packet reordering metric. */
322 u32 reord_seen
; /* number of data packet reordering events */
323 u32 snd_up
; /* Urgent pointer */
326 * Options received (usually on last packet, some only on SYN packets).
328 struct tcp_options_received rx_opt
;
331 * Slow start and congestion control (see also Nagle, and Karn & Partridge)
333 u32 snd_ssthresh
; /* Slow start size threshold */
334 u32 snd_cwnd
; /* Sending congestion window */
335 u32 snd_cwnd_cnt
; /* Linear increase counter */
336 u32 snd_cwnd_clamp
; /* Do not allow snd_cwnd to grow above this */
339 u32 prior_cwnd
; /* cwnd right before starting loss recovery */
340 u32 prr_delivered
; /* Number of newly delivered packets to
341 * receiver in Recovery. */
342 u32 prr_out
; /* Total number of pkts sent during Recovery. */
343 u32 delivered
; /* Total data packets delivered incl. rexmits */
344 u32 delivered_ce
; /* Like the above but only ECE marked packets */
345 u32 lost
; /* Total data packets lost incl. rexmits */
346 u32 app_limited
; /* limited until "delivered" reaches this val */
347 u64 first_tx_mstamp
; /* start of window send phase */
348 u64 delivered_mstamp
; /* time we reached "delivered" */
349 u32 rate_delivered
; /* saved rate sample: packets delivered */
350 u32 rate_interval_us
; /* saved rate sample: time elapsed */
352 u32 rcv_wnd
; /* Current receiver window */
353 u32 write_seq
; /* Tail(+1) of data held in tcp send buffer */
354 u32 notsent_lowat
; /* TCP_NOTSENT_LOWAT */
355 u32 pushed_seq
; /* Last pushed seq, required to talk to windows */
356 u32 lost_out
; /* Lost packets */
357 u32 sacked_out
; /* SACK'd packets */
359 struct hrtimer pacing_timer
;
360 struct hrtimer compressed_ack_timer
;
362 /* from STCP, retrans queue hinting */
363 struct sk_buff
* lost_skb_hint
;
364 struct sk_buff
*retransmit_skb_hint
;
366 /* OOO segments go in this rbtree. Socket lock must be held. */
367 struct rb_root out_of_order_queue
;
368 struct sk_buff
*ooo_last_skb
; /* cache rb_last(out_of_order_queue) */
370 /* SACKs data, these 2 need to be together (see tcp_options_write) */
371 struct tcp_sack_block duplicate_sack
[1]; /* D-SACK block */
372 struct tcp_sack_block selective_acks
[4]; /* The SACKS themselves*/
374 struct tcp_sack_block recv_sack_cache
[4];
376 struct sk_buff
*highest_sack
; /* skb just after the highest
377 * skb with SACKed bit set
378 * (validity guaranteed only if
384 u32 prior_ssthresh
; /* ssthresh saved at recovery start */
385 u32 high_seq
; /* snd_nxt at onset of congestion */
387 u32 retrans_stamp
; /* Timestamp of the last retransmit,
388 * also used in SYN-SENT to remember stamp of
390 u32 undo_marker
; /* snd_una upon a new recovery episode. */
391 int undo_retrans
; /* number of undoable retransmissions. */
392 u64 bytes_retrans
; /* RFC4898 tcpEStatsPerfOctetsRetrans
393 * Total data bytes retransmitted
395 u32 total_retrans
; /* Total retransmits for entire connection */
396 u32 rto_stamp
; /* Start time (ms) of last CA_Loss recovery */
397 u16 total_rto
; /* Total number of RTO timeouts, including
398 * SYN/SYN-ACK and recurring timeouts.
400 u16 total_rto_recoveries
; /* Total number of RTO recoveries,
401 * including any unfinished recovery.
403 u32 total_rto_time
; /* ms spent in (completed) RTO recoveries. */
405 u32 urg_seq
; /* Seq of received urgent pointer */
406 unsigned int keepalive_time
; /* time before keep alive takes place */
407 unsigned int keepalive_intvl
; /* time interval between keep alive probes */
412 /* Sock_ops bpf program related variables */
414 u8 bpf_sock_ops_cb_flags
; /* Control calling BPF programs
415 * values defined in uapi/linux/tcp.h
417 u8 bpf_chg_cc_inprogress
:1; /* In the middle of
418 * bpf_setsockopt(TCP_CONGESTION),
419 * it is to avoid the bpf_tcp_cc->init()
420 * to recur itself by calling
421 * bpf_setsockopt(TCP_CONGESTION, "itself").
423 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) (TP->bpf_sock_ops_cb_flags & ARG)
425 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) 0
428 u16 timeout_rehash
; /* Timeout-triggered rehash attempts */
430 u32 rcv_ooopack
; /* Received out-of-order packets, for tcpinfo */
432 /* Receiver side RTT estimation */
433 u32 rcv_rtt_last_tsecr
;
440 /* Receiver queue space */
447 /* TCP-specific MTU probe information. */
452 u32 plb_rehash
; /* PLB-triggered rehash attempts */
453 u32 mtu_info
; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
454 * while socket was owned by user.
456 #if IS_ENABLED(CONFIG_MPTCP)
459 #if IS_ENABLED(CONFIG_SMC)
460 bool (*smc_hs_congested
)(const struct sock
*sk
);
461 bool syn_smc
; /* SYN includes SMC */
464 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
465 /* TCP AF-Specific parts; only used by TCP-AO/MD5 Signature support so far */
466 const struct tcp_sock_af_ops
*af_specific
;
468 #ifdef CONFIG_TCP_MD5SIG
469 /* TCP MD5 Signature Option information */
470 struct tcp_md5sig_info __rcu
*md5sig_info
;
473 struct tcp_ao_info __rcu
*ao_info
;
477 /* TCP fastopen related information */
478 struct tcp_fastopen_request
*fastopen_req
;
479 /* fastopen_rsk points to request_sock that resulted in this big
480 * socket. Used to retransmit SYNACKs etc.
482 struct request_sock __rcu
*fastopen_rsk
;
483 struct saved_syn
*saved_syn
;
489 TCP_TSQ_DEFERRED
, /* tcp_tasklet_func() found socket was owned */
490 TCP_WRITE_TIMER_DEFERRED
, /* tcp_write_timer() found socket was owned */
491 TCP_DELACK_TIMER_DEFERRED
, /* tcp_delack_timer() found socket was owned */
492 TCP_MTU_REDUCED_DEFERRED
, /* tcp_v{4|6}_err() could not call
493 * tcp_v{4|6}_mtu_reduced()
495 TCP_ACK_DEFERRED
, /* TX pure ack is deferred */
499 TSQF_THROTTLED
= BIT(TSQ_THROTTLED
),
500 TSQF_QUEUED
= BIT(TSQ_QUEUED
),
501 TCPF_TSQ_DEFERRED
= BIT(TCP_TSQ_DEFERRED
),
502 TCPF_WRITE_TIMER_DEFERRED
= BIT(TCP_WRITE_TIMER_DEFERRED
),
503 TCPF_DELACK_TIMER_DEFERRED
= BIT(TCP_DELACK_TIMER_DEFERRED
),
504 TCPF_MTU_REDUCED_DEFERRED
= BIT(TCP_MTU_REDUCED_DEFERRED
),
505 TCPF_ACK_DEFERRED
= BIT(TCP_ACK_DEFERRED
),
508 #define tcp_sk(ptr) container_of_const(ptr, struct tcp_sock, inet_conn.icsk_inet.sk)
510 /* Variant of tcp_sk() upgrading a const sock to a read/write tcp socket.
511 * Used in context of (lockless) tcp listeners.
513 #define tcp_sk_rw(ptr) container_of(ptr, struct tcp_sock, inet_conn.icsk_inet.sk)
515 struct tcp_timewait_sock
{
516 struct inet_timewait_sock tw_sk
;
517 #define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt
518 #define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt
523 /* The time we sent the last out-of-window ACK: */
524 u32 tw_last_oow_ack_time
;
526 int tw_ts_recent_stamp
;
528 #ifdef CONFIG_TCP_MD5SIG
529 struct tcp_md5sig_key
*tw_md5_key
;
532 struct tcp_ao_info __rcu
*ao_info
;
536 static inline struct tcp_timewait_sock
*tcp_twsk(const struct sock
*sk
)
538 return (struct tcp_timewait_sock
*)sk
;
541 static inline bool tcp_passive_fastopen(const struct sock
*sk
)
543 return sk
->sk_state
== TCP_SYN_RECV
&&
544 rcu_access_pointer(tcp_sk(sk
)->fastopen_rsk
) != NULL
;
547 static inline void fastopen_queue_tune(struct sock
*sk
, int backlog
)
549 struct request_sock_queue
*queue
= &inet_csk(sk
)->icsk_accept_queue
;
550 int somaxconn
= READ_ONCE(sock_net(sk
)->core
.sysctl_somaxconn
);
552 WRITE_ONCE(queue
->fastopenq
.max_qlen
, min_t(unsigned int, backlog
, somaxconn
));
555 static inline void tcp_move_syn(struct tcp_sock
*tp
,
556 struct request_sock
*req
)
558 tp
->saved_syn
= req
->saved_syn
;
559 req
->saved_syn
= NULL
;
562 static inline void tcp_saved_syn_free(struct tcp_sock
*tp
)
564 kfree(tp
->saved_syn
);
565 tp
->saved_syn
= NULL
;
568 static inline u32
tcp_saved_syn_len(const struct saved_syn
*saved_syn
)
570 return saved_syn
->mac_hdrlen
+ saved_syn
->network_hdrlen
+
571 saved_syn
->tcp_hdrlen
;
574 struct sk_buff
*tcp_get_timestamping_opt_stats(const struct sock
*sk
,
575 const struct sk_buff
*orig_skb
,
576 const struct sk_buff
*ack_skb
);
578 static inline u16
tcp_mss_clamp(const struct tcp_sock
*tp
, u16 mss
)
580 /* We use READ_ONCE() here because socket might not be locked.
581 * This happens for listeners.
583 u16 user_mss
= READ_ONCE(tp
->rx_opt
.user_mss
);
585 return (user_mss
&& user_mss
< mss
) ? user_mss
: mss
;
588 int tcp_skb_shift(struct sk_buff
*to
, struct sk_buff
*from
, int pcount
,
591 void __tcp_sock_set_cork(struct sock
*sk
, bool on
);
592 void tcp_sock_set_cork(struct sock
*sk
, bool on
);
593 int tcp_sock_set_keepcnt(struct sock
*sk
, int val
);
594 int tcp_sock_set_keepidle_locked(struct sock
*sk
, int val
);
595 int tcp_sock_set_keepidle(struct sock
*sk
, int val
);
596 int tcp_sock_set_keepintvl(struct sock
*sk
, int val
);
597 void __tcp_sock_set_nodelay(struct sock
*sk
, bool on
);
598 void tcp_sock_set_nodelay(struct sock
*sk
);
599 void tcp_sock_set_quickack(struct sock
*sk
, int val
);
600 int tcp_sock_set_syncnt(struct sock
*sk
, int val
);
601 int tcp_sock_set_user_timeout(struct sock
*sk
, int val
);
603 static inline bool dst_tcp_usec_ts(const struct dst_entry
*dst
)
605 return dst_feature(dst
, RTAX_FEATURE_TCP_USEC_TS
);
608 #endif /* _LINUX_TCP_H */