489 lines · c
1/* SPDX-License-Identifier: GPL-2.0+ WITH Linux-syscall-note */2/*3 * INET An implementation of the TCP/IP protocol suite for the LINUX4 * operating system. INET is implemented using the BSD Socket5 * 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 04/28/9310 *11 * Author: Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>12 *13 * This program is free software; you can redistribute it and/or14 * modify it under the terms of the GNU General Public License15 * as published by the Free Software Foundation; either version16 * 2 of the License, or (at your option) any later version.17 */18#ifndef _UAPI_LINUX_TCP_H19#define _UAPI_LINUX_TCP_H20 21#include <linux/types.h>22#include <asm/byteorder.h>23#include <linux/socket.h>24 25struct tcphdr {26 __be16 source;27 __be16 dest;28 __be32 seq;29 __be32 ack_seq;30#if defined(__LITTLE_ENDIAN_BITFIELD)31 __u16 res1:4,32 doff:4,33 fin:1,34 syn:1,35 rst:1,36 psh:1,37 ack:1,38 urg:1,39 ece:1,40 cwr:1;41#elif defined(__BIG_ENDIAN_BITFIELD)42 __u16 doff:4,43 res1:4,44 cwr:1,45 ece:1,46 urg:1,47 ack:1,48 psh:1,49 rst:1,50 syn:1,51 fin:1;52#else53#error "Adjust your <asm/byteorder.h> defines"54#endif55 __be16 window;56 __sum16 check;57 __be16 urg_ptr;58};59 60/*61 * The union cast uses a gcc extension to avoid aliasing problems62 * (union is compatible to any of its members)63 * This means this part of the code is -fstrict-aliasing safe now.64 */65union tcp_word_hdr {66 struct tcphdr hdr;67 __be32 words[5];68};69 70#define tcp_flag_word(tp) (((union tcp_word_hdr *)(tp))->words[3])71 72enum {73 TCP_FLAG_CWR = __constant_cpu_to_be32(0x00800000),74 TCP_FLAG_ECE = __constant_cpu_to_be32(0x00400000),75 TCP_FLAG_URG = __constant_cpu_to_be32(0x00200000),76 TCP_FLAG_ACK = __constant_cpu_to_be32(0x00100000),77 TCP_FLAG_PSH = __constant_cpu_to_be32(0x00080000),78 TCP_FLAG_RST = __constant_cpu_to_be32(0x00040000),79 TCP_FLAG_SYN = __constant_cpu_to_be32(0x00020000),80 TCP_FLAG_FIN = __constant_cpu_to_be32(0x00010000),81 TCP_RESERVED_BITS = __constant_cpu_to_be32(0x0F000000),82 TCP_DATA_OFFSET = __constant_cpu_to_be32(0xF0000000)83};84 85/*86 * TCP general constants87 */88#define TCP_MSS_DEFAULT 536U /* IPv4 (RFC1122, RFC2581) */89#define TCP_MSS_DESIRED 1220U /* IPv6 (tunneled), EDNS0 (RFC3226) */90 91/* TCP socket options */92#define TCP_NODELAY 1 /* Turn off Nagle's algorithm. */93#define TCP_MAXSEG 2 /* Limit MSS */94#define TCP_CORK 3 /* Never send partially complete segments */95#define TCP_KEEPIDLE 4 /* Start keeplives after this period */96#define TCP_KEEPINTVL 5 /* Interval between keepalives */97#define TCP_KEEPCNT 6 /* Number of keepalives before death */98#define TCP_SYNCNT 7 /* Number of SYN retransmits */99#define TCP_LINGER2 8 /* Life time of orphaned FIN-WAIT-2 state */100#define TCP_DEFER_ACCEPT 9 /* Wake up listener only when data arrive */101#define TCP_WINDOW_CLAMP 10 /* Bound advertised window */102#define TCP_INFO 11 /* Information about this connection. */103#define TCP_QUICKACK 12 /* Block/reenable quick acks */104#define TCP_CONGESTION 13 /* Congestion control algorithm */105#define TCP_MD5SIG 14 /* TCP MD5 Signature (RFC2385) */106#define TCP_THIN_LINEAR_TIMEOUTS 16 /* Use linear timeouts for thin streams*/107#define TCP_THIN_DUPACK 17 /* Fast retrans. after 1 dupack */108#define TCP_USER_TIMEOUT 18 /* How long for loss retry before timeout */109#define TCP_REPAIR 19 /* TCP sock is under repair right now */110#define TCP_REPAIR_QUEUE 20111#define TCP_QUEUE_SEQ 21112#define TCP_REPAIR_OPTIONS 22113#define TCP_FASTOPEN 23 /* Enable FastOpen on listeners */114#define TCP_TIMESTAMP 24115#define TCP_NOTSENT_LOWAT 25 /* limit number of unsent bytes in write queue */116#define TCP_CC_INFO 26 /* Get Congestion Control (optional) info */117#define TCP_SAVE_SYN 27 /* Record SYN headers for new connections */118#define TCP_SAVED_SYN 28 /* Get SYN headers recorded for connection */119#define TCP_REPAIR_WINDOW 29 /* Get/set window parameters */120#define TCP_FASTOPEN_CONNECT 30 /* Attempt FastOpen with connect */121#define TCP_ULP 31 /* Attach a ULP to a TCP connection */122#define TCP_MD5SIG_EXT 32 /* TCP MD5 Signature with extensions */123#define TCP_FASTOPEN_KEY 33 /* Set the key for Fast Open (cookie) */124#define TCP_FASTOPEN_NO_COOKIE 34 /* Enable TFO without a TFO cookie */125#define TCP_ZEROCOPY_RECEIVE 35126#define TCP_INQ 36 /* Notify bytes available to read as a cmsg on read */127 128#define TCP_CM_INQ TCP_INQ129 130#define TCP_TX_DELAY 37 /* delay outgoing packets by XX usec */131 132#define TCP_AO_ADD_KEY 38 /* Add/Set MKT */133#define TCP_AO_DEL_KEY 39 /* Delete MKT */134#define TCP_AO_INFO 40 /* Set/list TCP-AO per-socket options */135#define TCP_AO_GET_KEYS 41 /* List MKT(s) */136#define TCP_AO_REPAIR 42 /* Get/Set SNEs and ISNs */137 138#define TCP_IS_MPTCP 43 /* Is MPTCP being used? */139 140#define TCP_REPAIR_ON 1141#define TCP_REPAIR_OFF 0142#define TCP_REPAIR_OFF_NO_WP -1 /* Turn off without window probes */143 144struct tcp_repair_opt {145 __u32 opt_code;146 __u32 opt_val;147};148 149struct tcp_repair_window {150 __u32 snd_wl1;151 __u32 snd_wnd;152 __u32 max_window;153 154 __u32 rcv_wnd;155 __u32 rcv_wup;156};157 158enum {159 TCP_NO_QUEUE,160 TCP_RECV_QUEUE,161 TCP_SEND_QUEUE,162 TCP_QUEUES_NR,163};164 165/* why fastopen failed from client perspective */166enum tcp_fastopen_client_fail {167 TFO_STATUS_UNSPEC, /* catch-all */168 TFO_COOKIE_UNAVAILABLE, /* if not in TFO_CLIENT_NO_COOKIE mode */169 TFO_DATA_NOT_ACKED, /* SYN-ACK did not ack SYN data */170 TFO_SYN_RETRANSMITTED, /* SYN-ACK did not ack SYN data after timeout */171};172 173/* for TCP_INFO socket option */174#define TCPI_OPT_TIMESTAMPS 1175#define TCPI_OPT_SACK 2176#define TCPI_OPT_WSCALE 4177#define TCPI_OPT_ECN 8 /* ECN was negociated at TCP session init */178#define TCPI_OPT_ECN_SEEN 16 /* we received at least one packet with ECT */179#define TCPI_OPT_SYN_DATA 32 /* SYN-ACK acked data in SYN sent or rcvd */180#define TCPI_OPT_USEC_TS 64 /* usec timestamps */181 182/*183 * Sender's congestion state indicating normal or abnormal situations184 * in the last round of packets sent. The state is driven by the ACK185 * information and timer events.186 */187enum tcp_ca_state {188 /*189 * Nothing bad has been observed recently.190 * No apparent reordering, packet loss, or ECN marks.191 */192 TCP_CA_Open = 0,193#define TCPF_CA_Open (1<<TCP_CA_Open)194 /*195 * The sender enters disordered state when it has received DUPACKs or196 * SACKs in the last round of packets sent. This could be due to packet197 * loss or reordering but needs further information to confirm packets198 * have been lost.199 */200 TCP_CA_Disorder = 1,201#define TCPF_CA_Disorder (1<<TCP_CA_Disorder)202 /*203 * The sender enters Congestion Window Reduction (CWR) state when it204 * has received ACKs with ECN-ECE marks, or has experienced congestion205 * or packet discard on the sender host (e.g. qdisc).206 */207 TCP_CA_CWR = 2,208#define TCPF_CA_CWR (1<<TCP_CA_CWR)209 /*210 * The sender is in fast recovery and retransmitting lost packets,211 * typically triggered by ACK events.212 */213 TCP_CA_Recovery = 3,214#define TCPF_CA_Recovery (1<<TCP_CA_Recovery)215 /*216 * The sender is in loss recovery triggered by retransmission timeout.217 */218 TCP_CA_Loss = 4219#define TCPF_CA_Loss (1<<TCP_CA_Loss)220};221 222struct tcp_info {223 __u8 tcpi_state;224 __u8 tcpi_ca_state;225 __u8 tcpi_retransmits;226 __u8 tcpi_probes;227 __u8 tcpi_backoff;228 __u8 tcpi_options;229 __u8 tcpi_snd_wscale : 4, tcpi_rcv_wscale : 4;230 __u8 tcpi_delivery_rate_app_limited:1, tcpi_fastopen_client_fail:2;231 232 __u32 tcpi_rto;233 __u32 tcpi_ato;234 __u32 tcpi_snd_mss;235 __u32 tcpi_rcv_mss;236 237 __u32 tcpi_unacked;238 __u32 tcpi_sacked;239 __u32 tcpi_lost;240 __u32 tcpi_retrans;241 __u32 tcpi_fackets;242 243 /* Times. */244 __u32 tcpi_last_data_sent;245 __u32 tcpi_last_ack_sent; /* Not remembered, sorry. */246 __u32 tcpi_last_data_recv;247 __u32 tcpi_last_ack_recv;248 249 /* Metrics. */250 __u32 tcpi_pmtu;251 __u32 tcpi_rcv_ssthresh;252 __u32 tcpi_rtt;253 __u32 tcpi_rttvar;254 __u32 tcpi_snd_ssthresh;255 __u32 tcpi_snd_cwnd;256 __u32 tcpi_advmss;257 __u32 tcpi_reordering;258 259 __u32 tcpi_rcv_rtt;260 __u32 tcpi_rcv_space;261 262 __u32 tcpi_total_retrans;263 264 __u64 tcpi_pacing_rate;265 __u64 tcpi_max_pacing_rate;266 __u64 tcpi_bytes_acked; /* RFC4898 tcpEStatsAppHCThruOctetsAcked */267 __u64 tcpi_bytes_received; /* RFC4898 tcpEStatsAppHCThruOctetsReceived */268 __u32 tcpi_segs_out; /* RFC4898 tcpEStatsPerfSegsOut */269 __u32 tcpi_segs_in; /* RFC4898 tcpEStatsPerfSegsIn */270 271 __u32 tcpi_notsent_bytes;272 __u32 tcpi_min_rtt;273 __u32 tcpi_data_segs_in; /* RFC4898 tcpEStatsDataSegsIn */274 __u32 tcpi_data_segs_out; /* RFC4898 tcpEStatsDataSegsOut */275 276 __u64 tcpi_delivery_rate;277 278 __u64 tcpi_busy_time; /* Time (usec) busy sending data */279 __u64 tcpi_rwnd_limited; /* Time (usec) limited by receive window */280 __u64 tcpi_sndbuf_limited; /* Time (usec) limited by send buffer */281 282 __u32 tcpi_delivered;283 __u32 tcpi_delivered_ce;284 285 __u64 tcpi_bytes_sent; /* RFC4898 tcpEStatsPerfHCDataOctetsOut */286 __u64 tcpi_bytes_retrans; /* RFC4898 tcpEStatsPerfOctetsRetrans */287 __u32 tcpi_dsack_dups; /* RFC4898 tcpEStatsStackDSACKDups */288 __u32 tcpi_reord_seen; /* reordering events seen */289 290 __u32 tcpi_rcv_ooopack; /* Out-of-order packets received */291 292 __u32 tcpi_snd_wnd; /* peer's advertised receive window after293 * scaling (bytes)294 */295 __u32 tcpi_rcv_wnd; /* local advertised receive window after296 * scaling (bytes)297 */298 299 __u32 tcpi_rehash; /* PLB or timeout triggered rehash attempts */300 301 __u16 tcpi_total_rto; /* Total number of RTO timeouts, including302 * SYN/SYN-ACK and recurring timeouts.303 */304 __u16 tcpi_total_rto_recoveries; /* Total number of RTO305 * recoveries, including any306 * unfinished recovery.307 */308 __u32 tcpi_total_rto_time; /* Total time spent in RTO recoveries309 * in milliseconds, including any310 * unfinished recovery.311 */312};313 314/* netlink attributes types for SCM_TIMESTAMPING_OPT_STATS */315enum {316 TCP_NLA_PAD,317 TCP_NLA_BUSY, /* Time (usec) busy sending data */318 TCP_NLA_RWND_LIMITED, /* Time (usec) limited by receive window */319 TCP_NLA_SNDBUF_LIMITED, /* Time (usec) limited by send buffer */320 TCP_NLA_DATA_SEGS_OUT, /* Data pkts sent including retransmission */321 TCP_NLA_TOTAL_RETRANS, /* Data pkts retransmitted */322 TCP_NLA_PACING_RATE, /* Pacing rate in bytes per second */323 TCP_NLA_DELIVERY_RATE, /* Delivery rate in bytes per second */324 TCP_NLA_SND_CWND, /* Sending congestion window */325 TCP_NLA_REORDERING, /* Reordering metric */326 TCP_NLA_MIN_RTT, /* minimum RTT */327 TCP_NLA_RECUR_RETRANS, /* Recurring retransmits for the current pkt */328 TCP_NLA_DELIVERY_RATE_APP_LMT, /* delivery rate application limited ? */329 TCP_NLA_SNDQ_SIZE, /* Data (bytes) pending in send queue */330 TCP_NLA_CA_STATE, /* ca_state of socket */331 TCP_NLA_SND_SSTHRESH, /* Slow start size threshold */332 TCP_NLA_DELIVERED, /* Data pkts delivered incl. out-of-order */333 TCP_NLA_DELIVERED_CE, /* Like above but only ones w/ CE marks */334 TCP_NLA_BYTES_SENT, /* Data bytes sent including retransmission */335 TCP_NLA_BYTES_RETRANS, /* Data bytes retransmitted */336 TCP_NLA_DSACK_DUPS, /* DSACK blocks received */337 TCP_NLA_REORD_SEEN, /* reordering events seen */338 TCP_NLA_SRTT, /* smoothed RTT in usecs */339 TCP_NLA_TIMEOUT_REHASH, /* Timeout-triggered rehash attempts */340 TCP_NLA_BYTES_NOTSENT, /* Bytes in write queue not yet sent */341 TCP_NLA_EDT, /* Earliest departure time (CLOCK_MONOTONIC) */342 TCP_NLA_TTL, /* TTL or hop limit of a packet received */343 TCP_NLA_REHASH, /* PLB and timeout triggered rehash attempts */344};345 346/* for TCP_MD5SIG socket option */347#define TCP_MD5SIG_MAXKEYLEN 80348 349/* tcp_md5sig extension flags for TCP_MD5SIG_EXT */350#define TCP_MD5SIG_FLAG_PREFIX 0x1 /* address prefix length */351#define TCP_MD5SIG_FLAG_IFINDEX 0x2 /* ifindex set */352 353struct tcp_md5sig {354 struct __kernel_sockaddr_storage tcpm_addr; /* address associated */355 __u8 tcpm_flags; /* extension flags */356 __u8 tcpm_prefixlen; /* address prefix */357 __u16 tcpm_keylen; /* key length */358 int tcpm_ifindex; /* device index for scope */359 __u8 tcpm_key[TCP_MD5SIG_MAXKEYLEN]; /* key (binary) */360};361 362/* INET_DIAG_MD5SIG */363struct tcp_diag_md5sig {364 __u8 tcpm_family;365 __u8 tcpm_prefixlen;366 __u16 tcpm_keylen;367 __be32 tcpm_addr[4];368 __u8 tcpm_key[TCP_MD5SIG_MAXKEYLEN];369};370 371#define TCP_AO_MAXKEYLEN 80372 373#define TCP_AO_KEYF_IFINDEX (1 << 0) /* L3 ifindex for VRF */374#define TCP_AO_KEYF_EXCLUDE_OPT (1 << 1) /* "Indicates whether TCP375 * options other than TCP-AO376 * are included in the MAC377 * calculation"378 */379 380struct tcp_ao_add { /* setsockopt(TCP_AO_ADD_KEY) */381 struct __kernel_sockaddr_storage addr; /* peer's address for the key */382 char alg_name[64]; /* crypto hash algorithm to use */383 __s32 ifindex; /* L3 dev index for VRF */384 __u32 set_current :1, /* set key as Current_key at once */385 set_rnext :1, /* request it from peer with RNext_key */386 reserved :30; /* must be 0 */387 __u16 reserved2; /* padding, must be 0 */388 __u8 prefix; /* peer's address prefix */389 __u8 sndid; /* SendID for outgoing segments */390 __u8 rcvid; /* RecvID to match for incoming seg */391 __u8 maclen; /* length of authentication code (hash) */392 __u8 keyflags; /* see TCP_AO_KEYF_ */393 __u8 keylen; /* length of ::key */394 __u8 key[TCP_AO_MAXKEYLEN];395} __attribute__((aligned(8)));396 397struct tcp_ao_del { /* setsockopt(TCP_AO_DEL_KEY) */398 struct __kernel_sockaddr_storage addr; /* peer's address for the key */399 __s32 ifindex; /* L3 dev index for VRF */400 __u32 set_current :1, /* corresponding ::current_key */401 set_rnext :1, /* corresponding ::rnext */402 del_async :1, /* only valid for listen sockets */403 reserved :29; /* must be 0 */404 __u16 reserved2; /* padding, must be 0 */405 __u8 prefix; /* peer's address prefix */406 __u8 sndid; /* SendID for outgoing segments */407 __u8 rcvid; /* RecvID to match for incoming seg */408 __u8 current_key; /* KeyID to set as Current_key */409 __u8 rnext; /* KeyID to set as Rnext_key */410 __u8 keyflags; /* see TCP_AO_KEYF_ */411} __attribute__((aligned(8)));412 413struct tcp_ao_info_opt { /* setsockopt(TCP_AO_INFO), getsockopt(TCP_AO_INFO) */414 /* Here 'in' is for setsockopt(), 'out' is for getsockopt() */415 __u32 set_current :1, /* in/out: corresponding ::current_key */416 set_rnext :1, /* in/out: corresponding ::rnext */417 ao_required :1, /* in/out: don't accept non-AO connects */418 set_counters :1, /* in: set/clear ::pkt_* counters */419 accept_icmps :1, /* in/out: accept incoming ICMPs */420 reserved :27; /* must be 0 */421 __u16 reserved2; /* padding, must be 0 */422 __u8 current_key; /* in/out: KeyID of Current_key */423 __u8 rnext; /* in/out: keyid of RNext_key */424 __u64 pkt_good; /* in/out: verified segments */425 __u64 pkt_bad; /* in/out: failed verification */426 __u64 pkt_key_not_found; /* in/out: could not find a key to verify */427 __u64 pkt_ao_required; /* in/out: segments missing TCP-AO sign */428 __u64 pkt_dropped_icmp; /* in/out: ICMPs that were ignored */429} __attribute__((aligned(8)));430 431struct tcp_ao_getsockopt { /* getsockopt(TCP_AO_GET_KEYS) */432 struct __kernel_sockaddr_storage addr; /* in/out: dump keys for peer433 * with this address/prefix434 */435 char alg_name[64]; /* out: crypto hash algorithm */436 __u8 key[TCP_AO_MAXKEYLEN];437 __u32 nkeys; /* in: size of the userspace buffer438 * @optval, measured in @optlen - the439 * sizeof(struct tcp_ao_getsockopt)440 * out: number of keys that matched441 */442 __u16 is_current :1, /* in: match and dump Current_key,443 * out: the dumped key is Current_key444 */445 446 is_rnext :1, /* in: match and dump RNext_key,447 * out: the dumped key is RNext_key448 */449 get_all :1, /* in: dump all keys */450 reserved :13; /* padding, must be 0 */451 __u8 sndid; /* in/out: dump keys with SendID */452 __u8 rcvid; /* in/out: dump keys with RecvID */453 __u8 prefix; /* in/out: dump keys with address/prefix */454 __u8 maclen; /* out: key's length of authentication455 * code (hash)456 */457 __u8 keyflags; /* in/out: see TCP_AO_KEYF_ */458 __u8 keylen; /* out: length of ::key */459 __s32 ifindex; /* in/out: L3 dev index for VRF */460 __u64 pkt_good; /* out: verified segments */461 __u64 pkt_bad; /* out: segments that failed verification */462} __attribute__((aligned(8)));463 464struct tcp_ao_repair { /* {s,g}etsockopt(TCP_AO_REPAIR) */465 __be32 snt_isn;466 __be32 rcv_isn;467 __u32 snd_sne;468 __u32 rcv_sne;469} __attribute__((aligned(8)));470 471/* setsockopt(fd, IPPROTO_TCP, TCP_ZEROCOPY_RECEIVE, ...) */472 473#define TCP_RECEIVE_ZEROCOPY_FLAG_TLB_CLEAN_HINT 0x1474struct tcp_zerocopy_receive {475 __u64 address; /* in: address of mapping */476 __u32 length; /* in/out: number of bytes to map/mapped */477 __u32 recv_skip_hint; /* out: amount of bytes to skip */478 __u32 inq; /* out: amount of bytes in read queue */479 __s32 err; /* out: socket error */480 __u64 copybuf_address; /* in: copybuf address (small reads) */481 __s32 copybuf_len; /* in/out: copybuf bytes avail/used or error */482 __u32 flags; /* in: flags */483 __u64 msg_control; /* ancillary data */484 __u64 msg_controllen;485 __u32 msg_flags;486 __u32 reserved; /* set to 0 for now */487};488#endif /* _UAPI_LINUX_TCP_H */489