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 <net/sock.h>
  23 #include <net/inet_connection_sock.h>
  24 #include <net/inet_timewait_sock.h>
  25 #include <uapi/linux/tcp.h>
  26
  27 static inline struct tcphdr *tcp_hdr(const struct sk_buff *skb)
  28 {
  29         return (struct tcphdr *)skb_transport_header(skb);
  30 }
  31
  32 static inline unsigned int tcp_hdrlen(const struct sk_buff *skb)
  33 {
  34         return tcp_hdr(skb)->doff * 4;
  35 }
  36
  37 static inline struct tcphdr *inner_tcp_hdr(const struct sk_buff *skb)
  38 {
  39         return (struct tcphdr *)skb_inner_transport_header(skb);
  40 }
  41
  42 static inline unsigned int inner_tcp_hdrlen(const struct sk_buff *skb)
  43 {
  44         return inner_tcp_hdr(skb)->doff * 4;
  45 }
  46
  47 static inline unsigned int tcp_optlen(const struct sk_buff *skb)
  48 {
  49         return (tcp_hdr(skb)->doff - 5) * 4;
  50 }
  51
  52 /* TCP Fast Open */
  53 #define TCP_FASTOPEN_COOKIE_MIN 4       /* Min Fast Open Cookie size in bytes */
  54 #define TCP_FASTOPEN_COOKIE_MAX 16      /* Max Fast Open Cookie size in bytes */
  55 #define TCP_FASTOPEN_COOKIE_SIZE 8      /* the size employed by this impl. */
  56
  57 /* TCP Fast Open Cookie as stored in memory */
  58 struct tcp_fastopen_cookie {
  59         s8      len;
  60         u8      val[TCP_FASTOPEN_COOKIE_MAX];
  61         bool    exp;    /* In RFC6994 experimental option format */
  62 };
  63
  64 /* This defines a selective acknowledgement block. */
  65 struct tcp_sack_block_wire {
  66         __be32  start_seq;
  67         __be32  end_seq;
  68 };
  69
  70 struct tcp_sack_block {
  71         u32     start_seq;
  72         u32     end_seq;
  73 };
  74
  75 /*These are used to set the sack_ok field in struct tcp_options_received */
  76 #define TCP_SACK_SEEN     (1 << 0)   /*1 = peer is SACK capable, */
  77 #define TCP_FACK_ENABLED  (1 << 1)   /*1 = FACK is enabled locally*/
  78 #define TCP_DSACK_SEEN    (1 << 2)   /*1 = DSACK was received from peer*/
  79
  80 struct tcp_options_received {
  81 /*      PAWS/RTTM data  */
  82         long    ts_recent_stamp;/* Time we stored ts_recent (for aging) */
  83         u32     ts_recent;      /* Time stamp to echo next              */
  84         u32     rcv_tsval;      /* Time stamp value                     */
  85         u32     rcv_tsecr;      /* Time stamp echo reply                */
  86         u16     saw_tstamp : 1, /* Saw TIMESTAMP on last packet         */
  87                 tstamp_ok : 1,  /* TIMESTAMP seen on SYN packet         */
  88                 dsack : 1,      /* D-SACK is scheduled                  */
  89                 wscale_ok : 1,  /* Wscale seen on SYN packet            */
  90                 sack_ok : 4,    /* SACK seen on SYN packet              */
  91                 snd_wscale : 4, /* Window scaling received from sender  */
  92                 rcv_wscale : 4; /* Window scaling to send to receiver   */
  93         u8      num_sacks;      /* Number of SACK blocks                */
  94         u16     user_mss;       /* mss requested by user in ioctl       */
  95         u16     mss_clamp;      /* Maximal mss, negotiated at connection setup */
  96 };
  97
  98 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
  99 {
 100         rx_opt->tstamp_ok = rx_opt->sack_ok = 0;
 101         rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
 102 }
 103
 104 /* This is the max number of SACKS that we'll generate and process. It's safe
 105  * to increase this, although since:
 106  *   size = TCPOLEN_SACK_BASE_ALIGNED (4) + n * TCPOLEN_SACK_PERBLOCK (8)
 107  * only four options will fit in a standard TCP header */
 108 #define TCP_NUM_SACKS 4
 109
 110 struct tcp_request_sock_ops;
 111
 112 struct tcp_request_sock {
 113         struct inet_request_sock        req;
 114         const struct tcp_request_sock_ops *af_specific;
 115         bool                            tfo_listener;
 116         u32                             rcv_isn;
 117         u32                             snt_isn;
 118         u32                             snt_synack; /* synack sent time */
 119         u32                             last_oow_ack_time; /* last SYNACK */
 120         u32                             rcv_nxt; /* the ack # by SYNACK. For
 121                                                   * FastOpen it's the seq#
 122                                                   * after data-in-SYN.
 123                                                   */
 124 };
 125
 126 static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
 127 {
 128         return (struct tcp_request_sock *)req;
 129 }
 130
 131 struct tcp_sock {
 132         /* inet_connection_sock has to be the first member of tcp_sock */
 133         struct inet_connection_sock     inet_conn;
 134         u16     tcp_header_len; /* Bytes of tcp header to send          */
 135         u16     gso_segs;       /* Max number of segs per GSO packet    */
 136
 137 /*
 138  *      Header prediction flags
 139  *      0x5?10 << 16 + snd_wnd in net byte order
 140  */
 141         __be32  pred_flags;
 142
 143 /*
 144  *      RFC793 variables by their proper names. This means you can
 145  *      read the code and the spec side by side (and laugh ...)
 146  *      See RFC793 and RFC1122. The RFC writes these in capitals.
 147  */
 148         u64     bytes_received; /* RFC4898 tcpEStatsAppHCThruOctetsReceived
 149                                  * sum(delta(rcv_nxt)), or how many bytes
 150                                  * were acked.
 151                                  */
 152         u32     segs_in;        /* RFC4898 tcpEStatsPerfSegsIn
 153                                  * total number of segments in.
 154                                  */
 155         u32     rcv_nxt;        /* What we want to receive next         */
 156         u32     copied_seq;     /* Head of yet unread data              */
 157         u32     rcv_wup;        /* rcv_nxt on last window update sent   */
 158         u32     snd_nxt;        /* Next sequence we send                */
 159         u32     segs_out;       /* RFC4898 tcpEStatsPerfSegsOut
 160                                  * The total number of segments sent.
 161                                  */
 162         u64     bytes_acked;    /* RFC4898 tcpEStatsAppHCThruOctetsAcked
 163                                  * sum(delta(snd_una)), or how many bytes
 164                                  * were acked.
 165                                  */
 166         struct u64_stats_sync syncp; /* protects 64bit vars (cf tcp_get_info()) */
 167
 168         u32     snd_una;        /* First byte we want an ack for        */
 169         u32     snd_sml;        /* Last byte of the most recently transmitted small packet */
 170         u32     rcv_tstamp;     /* timestamp of last received ACK (for keepalives) */
 171         u32     lsndtime;       /* timestamp of last sent data packet (for restart window) */
 172         u32     last_oow_ack_time;  /* timestamp of last out-of-window ACK */
 173
 174         u32     tsoffset;       /* timestamp offset */
 175
 176         struct list_head tsq_node; /* anchor in tsq_tasklet.head list */
 177         unsigned long   tsq_flags;
 178
 179         /* Data for direct copy to user */
 180         struct {
 181                 struct sk_buff_head     prequeue;
 182                 struct task_struct      *task;
 183                 struct msghdr           *msg;
 184                 int                     memory;
 185                 int                     len;
 186         } ucopy;
 187
 188         u32     snd_wl1;        /* Sequence for window update           */
 189         u32     snd_wnd;        /* The window we expect to receive      */
 190         u32     max_window;     /* Maximal window ever seen from peer   */
 191         u32     mss_cache;      /* Cached effective mss, not including SACKS */
 192
 193         u32     window_clamp;   /* Maximal window to advertise          */
 194         u32     rcv_ssthresh;   /* Current window clamp                 */
 195
 196         u16     advmss;         /* Advertised MSS                       */
 197         u8      unused;
 198         u8      nonagle     : 4,/* Disable Nagle algorithm?             */
 199                 thin_lto    : 1,/* Use linear timeouts for thin streams */
 200                 thin_dupack : 1,/* Fast retransmit on first dupack      */
 201                 repair      : 1,
 202                 frto        : 1;/* F-RTO (RFC5682) activated in CA_Loss */
 203         u8      repair_queue;
 204         u8      do_early_retrans:1,/* Enable RFC5827 early-retransmit  */
 205                 syn_data:1,     /* SYN includes data */
 206                 syn_fastopen:1, /* SYN includes Fast Open option */
 207                 syn_fastopen_exp:1,/* SYN includes Fast Open exp. option */
 208                 syn_data_acked:1,/* data in SYN is acked by SYN-ACK */
 209                 save_syn:1,     /* Save headers of SYN packet */
 210                 is_cwnd_limited:1;/* forward progress limited by snd_cwnd? */
 211         u32     tlp_high_seq;   /* snd_nxt at the time of TLP retransmit. */
 212
 213 /* RTT measurement */
 214         u32     srtt_us;        /* smoothed round trip time << 3 in usecs */
 215         u32     mdev_us;        /* medium deviation                     */
 216         u32     mdev_max_us;    /* maximal mdev for the last rtt period */
 217         u32     rttvar_us;      /* smoothed mdev_max                    */
 218         u32     rtt_seq;        /* sequence number to update rttvar     */
 219
 220         u32     packets_out;    /* Packets which are "in flight"        */
 221         u32     retrans_out;    /* Retransmitted packets out            */
 222         u32     max_packets_out;  /* max packets_out in last window */
 223         u32     max_packets_seq;  /* right edge of max_packets_out flight */
 224
 225         u16     urg_data;       /* Saved octet of OOB data and control flags */
 226         u8      ecn_flags;      /* ECN status bits.                     */
 227         u8      keepalive_probes; /* num of allowed keep alive probes   */
 228         u32     reordering;     /* Packet reordering metric.            */
 229         u32     snd_up;         /* Urgent pointer               */
 230
 231 /*
 232  *      Options received (usually on last packet, some only on SYN packets).
 233  */
 234         struct tcp_options_received rx_opt;
 235
 236 /*
 237  *      Slow start and congestion control (see also Nagle, and Karn & Partridge)
 238  */
 239         u32     snd_ssthresh;   /* Slow start size threshold            */
 240         u32     snd_cwnd;       /* Sending congestion window            */
 241         u32     snd_cwnd_cnt;   /* Linear increase counter              */
 242         u32     snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
 243         u32     snd_cwnd_used;
 244         u32     snd_cwnd_stamp;
 245         u32     prior_cwnd;     /* Congestion window at start of Recovery. */
 246         u32     prr_delivered;  /* Number of newly delivered packets to
 247                                  * receiver in Recovery. */
 248         u32     prr_out;        /* Total number of pkts sent during Recovery. */
 249
 250         u32     rcv_wnd;        /* Current receiver window              */
 251         u32     write_seq;      /* Tail(+1) of data held in tcp send buffer */
 252         u32     notsent_lowat;  /* TCP_NOTSENT_LOWAT */
 253         u32     pushed_seq;     /* Last pushed seq, required to talk to windows */
 254         u32     lost_out;       /* Lost packets                 */
 255         u32     sacked_out;     /* SACK'd packets                       */
 256         u32     fackets_out;    /* FACK'd packets                       */
 257
 258         /* from STCP, retrans queue hinting */
 259         struct sk_buff* lost_skb_hint;
 260         struct sk_buff *retransmit_skb_hint;
 261
 262         /* OOO segments go in this list. Note that socket lock must be held,
 263          * as we do not use sk_buff_head lock.
 264          */
 265         struct sk_buff_head     out_of_order_queue;
 266
 267         /* SACKs data, these 2 need to be together (see tcp_options_write) */
 268         struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
 269         struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
 270
 271         struct tcp_sack_block recv_sack_cache[4];
 272
 273         struct sk_buff *highest_sack;   /* skb just after the highest
 274                                          * skb with SACKed bit set
 275                                          * (validity guaranteed only if
 276                                          * sacked_out > 0)
 277                                          */
 278
 279         int     lost_cnt_hint;
 280         u32     retransmit_high;        /* L-bits may be on up to this seqno */
 281
 282         u32     lost_retrans_low;       /* Sent seq after any rxmit (lowest) */
 283
 284         u32     prior_ssthresh; /* ssthresh saved at recovery start     */
 285         u32     high_seq;       /* snd_nxt at onset of congestion       */
 286
 287         u32     retrans_stamp;  /* Timestamp of the last retransmit,
 288                                  * also used in SYN-SENT to remember stamp of
 289                                  * the first SYN. */
 290         u32     undo_marker;    /* snd_una upon a new recovery episode. */
 291         int     undo_retrans;   /* number of undoable retransmissions. */
 292         u32     total_retrans;  /* Total retransmits for entire connection */
 293
 294         u32     urg_seq;        /* Seq of received urgent pointer */
 295         unsigned int            keepalive_time;   /* time before keep alive takes place */
 296         unsigned int            keepalive_intvl;  /* time interval between keep alive probes */
 297
 298         int                     linger2;
 299
 300 /* Receiver side RTT estimation */
 301         struct {
 302                 u32     rtt;
 303                 u32     seq;
 304                 u32     time;
 305         } rcv_rtt_est;
 306
 307 /* Receiver queue space */
 308         struct {
 309                 int     space;
 310                 u32     seq;
 311                 u32     time;
 312         } rcvq_space;
 313
 314 /* TCP-specific MTU probe information. */
 315         struct {
 316                 u32               probe_seq_start;
 317                 u32               probe_seq_end;
 318         } mtu_probe;
 319         u32     mtu_info; /* We received an ICMP_FRAG_NEEDED / ICMPV6_PKT_TOOBIG
 320                            * while socket was owned by user.
 321                            */
 322
 323 #ifdef CONFIG_TCP_MD5SIG
 324 /* TCP AF-Specific parts; only used by MD5 Signature support so far */
 325         const struct tcp_sock_af_ops    *af_specific;
 326
 327 /* TCP MD5 Signature Option information */
 328         struct tcp_md5sig_info  __rcu *md5sig_info;
 329 #endif
 330
 331 /* TCP fastopen related information */
 332         struct tcp_fastopen_request *fastopen_req;
 333         /* fastopen_rsk points to request_sock that resulted in this big
 334          * socket. Used to retransmit SYNACKs etc.
 335          */
 336         struct request_sock *fastopen_rsk;
 337         u32     *saved_syn;
 338 };
 339
 340 enum tsq_flags {
 341         TSQ_THROTTLED,
 342         TSQ_QUEUED,
 343         TCP_TSQ_DEFERRED,          /* tcp_tasklet_func() found socket was owned */
 344         TCP_WRITE_TIMER_DEFERRED,  /* tcp_write_timer() found socket was owned */
 345         TCP_DELACK_TIMER_DEFERRED, /* tcp_delack_timer() found socket was owned */
 346         TCP_MTU_REDUCED_DEFERRED,  /* tcp_v{4|6}_err() could not call
 347                                     * tcp_v{4|6}_mtu_reduced()
 348                                     */
 349 };
 350
 351 static inline struct tcp_sock *tcp_sk(const struct sock *sk)
 352 {
 353         return (struct tcp_sock *)sk;
 354 }
 355
 356 struct tcp_timewait_sock {
 357         struct inet_timewait_sock tw_sk;
 358         u32                       tw_rcv_nxt;
 359         u32                       tw_snd_nxt;
 360         u32                       tw_rcv_wnd;
 361         u32                       tw_ts_offset;
 362         u32                       tw_ts_recent;
 363
 364         /* The time we sent the last out-of-window ACK: */
 365         u32                       tw_last_oow_ack_time;
 366
 367         long                      tw_ts_recent_stamp;
 368 #ifdef CONFIG_TCP_MD5SIG
 369         struct tcp_md5sig_key     *tw_md5_key;
 370 #endif
 371 };
 372
 373 static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
 374 {
 375         return (struct tcp_timewait_sock *)sk;
 376 }
 377
 378 static inline bool tcp_passive_fastopen(const struct sock *sk)
 379 {
 380         return (sk->sk_state == TCP_SYN_RECV &&
 381                 tcp_sk(sk)->fastopen_rsk != NULL);
 382 }
 383
 384 extern void tcp_sock_destruct(struct sock *sk);
 385
 386 static inline int fastopen_init_queue(struct sock *sk, int backlog)
 387 {
 388         struct request_sock_queue *queue =
 389             &inet_csk(sk)->icsk_accept_queue;
 390
 391         if (queue->fastopenq == NULL) {
 392                 queue->fastopenq = kzalloc(
 393                     sizeof(struct fastopen_queue),
 394                     sk->sk_allocation);
 395                 if (queue->fastopenq == NULL)
 396                         return -ENOMEM;
 397
 398                 sk->sk_destruct = tcp_sock_destruct;
 399                 spin_lock_init(&queue->fastopenq->lock);
 400         }
 401         queue->fastopenq->max_qlen = backlog;
 402         return 0;
 403 }
 404
 405 static inline void tcp_saved_syn_free(struct tcp_sock *tp)
 406 {
 407         kfree(tp->saved_syn);
 408         tp->saved_syn = NULL;
 409 }
 410
 411 #endif  /* _LINUX_TCP_H */