include/linux/tcp.h

   1 /*
   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.
   5  *
   6  *              Definitions for the TCP protocol.
   7  *
   8  * Version:     @(#)tcp.h       1.0.2   04/28/93
   9  *
  10  * Author:      Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11  *
  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.
  16  */
  17 #ifndef _LINUX_TCP_H
  18 #define _LINUX_TCP_H
  19
  20
  21 #include <linux/skbuff.h>
  22 #include <linux/win_minmax.h>
  23 #include <net/sock.h>
  24 #include <net/inet_connection_sock.h>
  25 #include <net/inet_timewait_sock.h>
  26 #include <uapi/linux/tcp.h>
  27
  28 static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb)
  29 {
  30         return (struct tcphdr *)skb_transport_header(skb);
  31 }
  32
  33 static inline unsigned int __tcp_hdrlen(const struct tcphdr *th)
  34 {
  35         return th->doff * 4;
  36 }
  37
  38 static inline unsigned int tcp_hdrlen(const struct sk_buff *skb)
  39 {
  40         return __tcp_hdrlen(tcp_hdr(skb));
  41 }
  42
  43 static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb)
  44 {
  45         return (struct tcphdr *)skb_inner_transport_header(skb);
  46 }
  47
  48 static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb)
  49 {
  50         return inner_tcp_hdr(skb)->doff * 4;
  51 }
  52
  53 static inline unsigned int tcp_optlen(const struct sk_buff *skb)
  54 {
  55         return (tcp_hdr(skb)->doff - 5) * 4;
  56 }
  57
  58 /* TCP Fast Open */
  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. */
  62
  63 /* TCP Fast Open Cookie as stored in memory */
  64 struct tcp_fastopen_cookie {
  65         union {
  66                 u8      val[TCP_FASTOPEN_COOKIE_MAX];
  67 #if IS_ENABLED(CONFIG_IPV6)
  68                 struct in6_addr addr;
  69 #endif
  70         };
  71         s8      len;
  72         bool    exp;    /* In RFC6994 experimental option format */
  73 };
  74
  75 /* This defines a selective acknowledgement block. */
  76 struct tcp_sack_block_wire {
  77         __be32  start_seq;
  78         __be32  end_seq;
  79 };
  80
  81 struct tcp_sack_block {
  82         u32     start_seq;
  83         u32     end_seq;
  84 };
  85
  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_DSACK_SEEN    (1 << 2)   /*1 = DSACK was received from peer*/
  89
  90 struct tcp_options_received {
  91 /*      PAWS/RTTM data  */
  92         long    ts_recent_stamp;/* Time we stored ts_recent (for aging) */
  93         u32     ts_recent;      /* Time stamp to echo next              */
  94         u32     rcv_tsval;      /* Time stamp value                     */
  95         u32     rcv_tsecr;      /* Time stamp echo reply                */
  96         u16     saw_tstamp : 1, /* Saw TIMESTAMP on last packet         */
  97                 tstamp_ok : 1,  /* TIMESTAMP seen on SYN packet         */
  98                 dsack : 1,      /* D-SACK is scheduled                  */
  99                 wscale_ok : 1,  /* Wscale seen on SYN packet            */
 100                 sack_ok : 3,    /* SACK seen on SYN packet              */
 101                 smc_ok : 1,     /* SMC 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 */
 107 };
 108
 109 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
 110 {
 111         rx_opt->tstamp_ok = rx_opt->sack_ok = 0;
 112         rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
 113 #if IS_ENABLED(CONFIG_SMC)
 114         rx_opt->smc_ok = 0;
 115 #endif
 116 }
 117
 118 /* This is the max number of SACKS that we'll generate and process. It's safe
 119  * to increase this, although since:
 120  *   size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8)
 121  * only four options will fit in a standard TCP header */
 122 #define TCP_NUM_SACKS 4
 123
 124 struct tcp_request_sock_ops;
 125
 126 struct tcp_request_sock {
 127         struct inet_request_sock        req;
 128         const struct tcp_request_sock_ops *af_specific;
 129         u64                             snt_synack; /* first SYNACK sent time */
 130         bool                            tfo_listener;
 131         u32                             txhash;
 132         u32                             rcv_isn;
 133         u32                             snt_isn;
 134         u32                             ts_off;
 135         u32                             last_oow_ack_time; /* last SYNACK */
 136         u32                             rcv_nxt; /* the ack # by SYNACK. For
 137                                                   * FastOpen it's the seq#
 138                                                   * after data-in-SYN.
 139                                                   */
 140 };
 141
 142 static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
 143 {
 144         return (struct tcp_request_sock *)req;
 145 }
 146
 147 struct tcp_sock {
 148         /* inet_connection_sock has to be the first member of tcp_sock */
 149         struct inet_connection_sock     inet_conn;
 150         u16     tcp_header_len; /* Bytes of tcp header to send          */
 151         u16     gso_segs;       /* Max number of segs per GSO packet    */
 152
 153 /*
 154  *      Header prediction flags
 155  *      0x5?10 << 16 + snd_wnd in net byte order
 156  */
 157         __be32  pred_flags;
 158
 159 /*
 160  *      RFC793 variables by their proper names. This means you can
 161  *      read the code and the spec side by side (and laugh ...)
 162  *      See RFC793 and RFC1122. The RFC writes these in capitals.
 163  */
 164         u64     bytes_received; /* RFC4898 tcpEStatsAppHCThruOctetsReceived
 165                                  * sum(delta(rcv_nxt)), or how many bytes
 166                                  * were acked.
 167                                  */
 168         u32     segs_in;        /* RFC4898 tcpEStatsPerfSegsIn
 169                                  * total number of segments in.
 170                                  */
 171         u32     data_segs_in;   /* RFC4898 tcpEStatsPerfDataSegsIn
 172                                  * total number of data segments in.
 173                                  */
 174         u32     rcv_nxt;        /* What we want to receive next         */
 175         u32     copied_seq;     /* Head of yet unread data              */
 176         u32     rcv_wup;        /* rcv_nxt on last window update sent   */
 177         u32     snd_nxt;        /* Next sequence we send                */
 178         u32     segs_out;       /* RFC4898 tcpEStatsPerfSegsOut
 179                                  * The total number of segments sent.
 180                                  */
 181         u32     data_segs_out;  /* RFC4898 tcpEStatsPerfDataSegsOut
 182                                  * total number of data segments sent.
 183                                  */
 184         u64     bytes_acked;    /* RFC4898 tcpEStatsAppHCThruOctetsAcked
 185                                  * sum(delta(snd_una)), or how many bytes
 186                                  * were acked.
 187                                  */
 188         u32     snd_una;        /* First byte we want an ack for        */
 189         u32     snd_sml;        /* Last byte of the most recently transmitted small packet */
 190         u32     rcv_tstamp;     /* timestamp of last received ACK (for keepalives) */
 191         u32     lsndtime;       /* timestamp of last sent data packet (for restart window) */
 192         u32     last_oow_ack_time;  /* timestamp of last out-of-window ACK */
 193
 194         u32     tsoffset;       /* timestamp offset */
 195
 196         struct list_head tsq_node; /* anchor in tsq_tasklet.head list */
 197         struct list_head tsorted_sent_queue; /* time-sorted sent but un-SACKed skbs */
 198
 199         u32     snd_wl1;        /* Sequence for window update           */
 200         u32     snd_wnd;        /* The window we expect to receive      */
 201         u32     max_window;     /* Maximal window ever seen from peer   */
 202         u32     mss_cache;      /* Cached effective mss, not including SACKS */
 203
 204         u32     window_clamp;   /* Maximal window to advertise          */
 205         u32     rcv_ssthresh;   /* Current window clamp                 */
 206
 207         /* Information of the most recently (s)acked skb */
 208         struct tcp_rack {
 209                 u64 mstamp; /* (Re)sent time of the skb */
 210                 u32 rtt_us;  /* Associated RTT */
 211                 u32 end_seq; /* Ending TCP sequence of the skb */
 212                 u32 last_delivered; /* tp->delivered at last reo_wnd adj */
 213                 u8 reo_wnd_steps;   /* Allowed reordering window */
 214 #define TCP_RACK_RECOVERY_THRESH 16
 215                 u8 reo_wnd_persist:5, /* No. of recovery since last adj */
 216                    dsack_seen:1, /* Whether DSACK seen after last adj */
 217                    advanced:1,   /* mstamp advanced since last lost marking */
 218                    reord:1;      /* reordering detected */
 219         } rack;
 220         u16     advmss;         /* Advertised MSS                       */
 221         u32     chrono_start;   /* Start time in jiffies of a TCP chrono */
 222         u32     chrono_stat[3]; /* Time in jiffies for chrono_stat stats */
 223         u8      chrono_type:2,  /* current chronograph type */
 224                 rate_app_limited:1,  /* rate_{delivered,interval_us} limited? */
 225                 fastopen_connect:1, /* FASTOPEN_CONNECT sockopt */
 226                 fastopen_no_cookie:1, /* Allow send/recv SYN+data without a cookie */
 227                 is_sack_reneg:1,    /* in recovery from loss with SACK reneg? */
 228                 unused:2;
 229         u8      nonagle     : 4,/* Disable Nagle algorithm?             */
 230                 thin_lto    : 1,/* Use linear timeouts for thin streams */
 231                 unused1     : 1,
 232                 repair      : 1,
 233                 frto        : 1;/* F-RTO (RFC5682) activated in CA_Loss */
 234         u8      repair_queue;
 235         u8      syn_data:1,     /* SYN includes data */
 236                 syn_fastopen:1, /* SYN includes Fast Open option */
 237                 syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
 238                 syn_fastopen_ch:1, /* Active TFO re-enabling probe */
 239                 syn_data_acked:1,/* data in SYN is acked by SYN-ACK */
 240                 save_syn:1,     /* Save headers of SYN packet */
 241                 is_cwnd_limited:1,/* forward progress limited by snd_cwnd? */
 242                 syn_smc:1;      /* SYN includes SMC */
 243         u32     tlp_high_seq;   /* snd_nxt at the time of TLP retransmit. */
 244
 245 /* RTT measurement */
 246         u64     tcp_mstamp;     /* most recent packet received/sent */
 247         u32     srtt_us;        /* smoothed round trip time << 3 in usecs */
 248         u32     mdev_us;        /* medium deviation                     */
 249         u32     mdev_max_us;    /* maximal mdev for the last rtt period */
 250         u32     rttvar_us;      /* smoothed mdev_max                    */
 251         u32     rtt_seq;        /* sequence number to update rttvar     */
 252         struct  minmax rtt_min;
 253
 254         u32     packets_out;    /* Packets which are "in flight"        */
 255         u32     retrans_out;    /* Retransmitted packets out            */
 256         u32     max_packets_out;  /* max packets_out in last window */
 257         u32     max_packets_seq;  /* right edge of max_packets_out flight */
 258
 259         u16     urg_data;       /* Saved octet of OOB data and control flags */
 260         u8      ecn_flags;      /* ECN status bits.                     */
 261         u8      keepalive_probes; /* num of allowed keep alive probes   */
 262         u32     reordering;     /* Packet reordering metric.            */
 263         u32     snd_up;         /* Urgent pointer               */
 264
 265 /*
 266  *      Options received (usually on last packet, some only on SYN packets).
 267  */
 268         struct tcp_options_received rx_opt;
 269
 270 /*
 271  *      Slow start and congestion control (see also Nagle, and Karn & Partridge)
 272  */
 273         u32     snd_ssthresh;   /* Slow start size threshold            */
 274         u32     snd_cwnd;       /* Sending congestion window            */
 275         u32     snd_cwnd_cnt;   /* Linear increase counter              */
 276         u32     snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
 277         u32     snd_cwnd_used;
 278         u32     snd_cwnd_stamp;
 279         u32     prior_cwnd;     /* cwnd right before starting loss recovery */
 280         u32     prr_delivered;  /* Number of newly delivered packets to
 281                                  * receiver in Recovery. */
 282         u32     prr_out;        /* Total number of pkts sent during Recovery. */
 283         u32     delivered;      /* Total data packets delivered incl. rexmits */
 284         u32     lost;           /* Total data packets lost incl. rexmits */
 285         u32     app_limited;    /* limited until "delivered" reaches this val */
 286         u64     first_tx_mstamp;  /* start of window send phase */
 287         u64     delivered_mstamp; /* time we reached "delivered" */
 288         u32     rate_delivered;    /* saved rate sample: packets delivered */
 289         u32     rate_interval_us;  /* saved rate sample: time elapsed */
 290
 291         u32     rcv_wnd;        /* Current receiver window              */
 292         u32     write_seq;      /* Tail(+1) of data held in tcp send buffer */
 293         u32     notsent_lowat;  /* TCP_NOTSENT_LOWAT */
 294         u32     pushed_seq;     /* Last pushed seq, required to talk to windows */
 295         u32     lost_out;       /* Lost packets                 */
 296         u32     sacked_out;     /* SACK'd packets                       */
 297
 298         struct hrtimer  pacing_timer;
 299
 300         /* from STCP, retrans queue hinting */
 301         struct sk_buff* lost_skb_hint;
 302         struct sk_buff *retransmit_skb_hint;
 303
 304         /* OOO segments go in this rbtree. Socket lock must be held. */
 305         struct rb_root  out_of_order_queue;
 306         struct sk_buff  *ooo_last_skb; /* cache rb_last(out_of_order_queue) */
 307
 308         /* SACKs data, these 2 need to be together (see tcp_options_write) */
 309         struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
 310         struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
 311
 312         struct tcp_sack_block recv_sack_cache[4];
 313
 314         struct sk_buff *highest_sack;   /* skb just after the highest
 315                                          * skb with SACKed bit set
 316                                          * (validity guaranteed only if
 317                                          * sacked_out > 0)
 318                                          */
 319
 320         int     lost_cnt_hint;
 321
 322         u32     prior_ssthresh; /* ssthresh saved at recovery start     */
 323         u32     high_seq;       /* snd_nxt at onset of congestion       */
 324
 325         u32     retrans_stamp;  /* Timestamp of the last retransmit,
 326                                  * also used in SYN-SENT to remember stamp of
 327                                  * the first SYN. */
 328         u32     undo_marker;    /* snd_una upon a new recovery episode. */
 329         int     undo_retrans;   /* number of undoable retransmissions. */
 330         u32     total_retrans;  /* Total retransmits for entire connection */
 331
 332         u32     urg_seq;        /* Seq of received urgent pointer */
 333         unsigned int            keepalive_time;   /* time before keep alive takes place */
 334         unsigned int            keepalive_intvl;  /* time interval between keep alive probes */
 335
 336         int                     linger2;
 337
 338
 339 /* Sock_ops bpf program related variables */
 340 #ifdef CONFIG_BPF
 341         u8      bpf_sock_ops_cb_flags;  /* Control calling BPF programs
 342                                          * values defined in uapi/linux/tcp.h
 343                                          */
 344 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) (TP->bpf_sock_ops_cb_flags & ARG)
 345 #else
 346 #define BPF_SOCK_OPS_TEST_FLAG(TP, ARG) 0
 347 #endif
 348
 349 /* Receiver side RTT estimation */
 350         struct {
 351                 u32     rtt_us;
 352                 u32     seq;
 353                 u64     time;
 354         } rcv_rtt_est;
 355
 356 /* Receiver queue space */
 357         struct {
 358                 u32     space;
 359                 u32     seq;
 360                 u64     time;
 361         } rcvq_space;
 362
 363 /* TCP-specific MTU probe information. */
 364         struct {
 365                 u32               probe_seq_start;
 366                 u32               probe_seq_end;
 367         } mtu_probe;
 368         u32     mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
 369                            * while socket was owned by user.
 370                            */
 371
 372 #ifdef CONFIG_TCP_MD5SIG
 373 /* TCP AF-Specific parts; only used by MD5 Signature support so far */
 374         const struct tcp_sock_af_ops    *af_specific;
 375
 376 /* TCP MD5 Signature Option information */
 377         struct tcp_md5sig_info  __rcu *md5sig_info;
 378 #endif
 379
 380 /* TCP fastopen related information */
 381         struct tcp_fastopen_request *fastopen_req;
 382         /* fastopen_rsk points to request_sock that resulted in this big
 383          * socket. Used to retransmit SYNACKs etc.
 384          */
 385         struct request_sock *fastopen_rsk;
 386         u32     *saved_syn;
 387 };
 388
 389 enum tsq_enum {
 390         TSQ_THROTTLED,
 391         TSQ_QUEUED,
 392         TCP_TSQ_DEFERRED,          /* tcp_tasklet_func() found socket was owned */
 393         TCP_WRITE_TIMER_DEFERRED,  /* tcp_write_timer() found socket was owned */
 394         TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */
 395         TCP_MTU_REDUCED_DEFERRED,  /* tcp_v{4|6}_err() could not call
 396                                     * tcp_v{4|6}_mtu_reduced()
 397                                     */
 398 };
 399
 400 enum tsq_flags {
 401         TSQF_THROTTLED                  = (1UL << TSQ_THROTTLED),
 402         TSQF_QUEUED                     = (1UL << TSQ_QUEUED),
 403         TCPF_TSQ_DEFERRED               = (1UL << TCP_TSQ_DEFERRED),
 404         TCPF_WRITE_TIMER_DEFERRED       = (1UL << TCP_WRITE_TIMER_DEFERRED),
 405         TCPF_DELACK_TIMER_DEFERRED      = (1UL << TCP_DELACK_TIMER_DEFERRED),
 406         TCPF_MTU_REDUCED_DEFERRED       = (1UL << TCP_MTU_REDUCED_DEFERRED),
 407 };
 408
 409 static inline struct tcp_sock *tcp_sk(const struct sock *sk)
 410 {
 411         return (struct tcp_sock *)sk;
 412 }
 413
 414 struct tcp_timewait_sock {
 415         struct inet_timewait_sock tw_sk;
 416 #define tw_rcv_nxt tw_sk.__tw_common.skc_tw_rcv_nxt
 417 #define tw_snd_nxt tw_sk.__tw_common.skc_tw_snd_nxt
 418         u32                       tw_rcv_wnd;
 419         u32                       tw_ts_offset;
 420         u32                       tw_ts_recent;
 421
 422         /* The time we sent the last out-of-window ACK: */
 423         u32                       tw_last_oow_ack_time;
 424
 425         long                      tw_ts_recent_stamp;
 426 #ifdef CONFIG_TCP_MD5SIG
 427         struct tcp_md5sig_key     *tw_md5_key;
 428 #endif
 429 };
 430
 431 static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
 432 {
 433         return (struct tcp_timewait_sock *)sk;
 434 }
 435
 436 static inline bool tcp_passive_fastopen(const struct sock *sk)
 437 {
 438         return (sk->sk_state == TCP_SYN_RECV &&
 439                 tcp_sk(sk)->fastopen_rsk != NULL);
 440 }
 441
 442 static inline void fastopen_queue_tune(struct sock *sk, int backlog)
 443 {
 444         struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
 445         int somaxconn = READ_ONCE(sock_net(sk)->core.sysctl_somaxconn);
 446
 447         queue->fastopenq.max_qlen = min_t(unsigned int, backlog, somaxconn);
 448 }
 449
 450 static inline void tcp_move_syn(struct tcp_sock *tp,
 451                                 struct request_sock *req)
 452 {
 453         tp->saved_syn = req->saved_syn;
 454         req->saved_syn = NULL;
 455 }
 456
 457 static inline void tcp_saved_syn_free(struct tcp_sock *tp)
 458 {
 459         kfree(tp->saved_syn);
 460         tp->saved_syn = NULL;
 461 }
 462
 463 struct sk_buff *tcp_get_timestamping_opt_stats(const struct sock *sk);
 464
 465 static inline u16 tcp_mss_clamp(const struct tcp_sock *tp, u16 mss)
 466 {
 467         /* We use READ_ONCE() here because socket might not be locked.
 468          * This happens for listeners.
 469          */
 470         u16 user_mss = READ_ONCE(tp->rx_opt.user_mss);
 471
 472         return (user_mss && user_mss < mss) ? user_mss : mss;
 473 }
 474 #endif  /* _LINUX_TCP_H */