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