1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  */
  24 
  25 #ifndef _INET_TCP_IMPL_H
  26 #define _INET_TCP_IMPL_H
  27 
  28 /*
  29  * TCP implementation private declarations.  These interfaces are
  30  * used to build the IP module and are not meant to be accessed
  31  * by any modules except IP itself.  They are undocumented and are
  32  * subject to change without notice.
  33  */
  34 
  35 #ifdef  __cplusplus
  36 extern "C" {
  37 #endif
  38 
  39 #ifdef _KERNEL
  40 
  41 #include <sys/cpuvar.h>
  42 #include <sys/clock_impl.h>       /* For LBOLT_FASTPATH{,64} */
  43 #include <inet/optcom.h>
  44 #include <inet/tcp.h>
  45 #include <inet/tunables.h>
  46 
  47 #define TCP_MOD_ID      5105
  48 
  49 extern struct qinit     tcp_sock_winit;
  50 extern struct qinit     tcp_winit;
  51 
  52 extern sock_downcalls_t sock_tcp_downcalls;
  53 
  54 /*
  55  * Bind hash list size and has function.  It has to be a power of 2 for
  56  * hashing.
  57  */
  58 #define TCP_BIND_FANOUT_SIZE    512
  59 #define TCP_BIND_HASH(lport) (ntohs(lport) & (TCP_BIND_FANOUT_SIZE - 1))
  60 
  61 /*
  62  * This implementation follows the 4.3BSD interpretation of the urgent
  63  * pointer and not RFC 1122. Switching to RFC 1122 behavior would cause
  64  * incompatible changes in protocols like telnet and rlogin.
  65  */
  66 #define TCP_OLD_URP_INTERPRETATION      1
  67 
  68 /* TCP option length */
  69 #define TCPOPT_NOP_LEN          1
  70 #define TCPOPT_MAXSEG_LEN       4
  71 #define TCPOPT_WS_LEN           3
  72 #define TCPOPT_REAL_WS_LEN      (TCPOPT_WS_LEN+1)
  73 #define TCPOPT_TSTAMP_LEN       10
  74 #define TCPOPT_REAL_TS_LEN      (TCPOPT_TSTAMP_LEN+2)
  75 #define TCPOPT_SACK_OK_LEN      2
  76 #define TCPOPT_REAL_SACK_OK_LEN (TCPOPT_SACK_OK_LEN+2)
  77 #define TCPOPT_REAL_SACK_LEN    4
  78 #define TCPOPT_MAX_SACK_LEN     36
  79 #define TCPOPT_HEADER_LEN       2
  80 
  81 /* Round up the value to the nearest mss. */
  82 #define MSS_ROUNDUP(value, mss)         ((((value) - 1) / (mss) + 1) * (mss))
  83 
  84 /*
  85  * Was this tcp created via socket() interface?
  86  */
  87 #define TCP_IS_SOCKET(tcp)      ((tcp)->tcp_issocket)
  88 
  89 /*
  90  * Is this tcp not attached to any upper client?
  91  */
  92 #define TCP_IS_DETACHED(tcp)    ((tcp)->tcp_detached)
  93 
  94 /* TCP timers related data strucutres.  Refer to tcp_timers.c. */
  95 typedef struct tcp_timer_s {
  96         conn_t  *connp;
  97         void    (*tcpt_proc)(void *);
  98         callout_id_t   tcpt_tid;
  99 } tcp_timer_t;
 100 
 101 extern kmem_cache_t *tcp_timercache;
 102 
 103 /*
 104  * Macro for starting various timers.  Retransmission timer has its own macro,
 105  * TCP_TIMER_RESTART().  tim is in millisec.
 106  */
 107 #define TCP_TIMER(tcp, f, tim)          \
 108         tcp_timeout(tcp->tcp_connp, f, tim)
 109 #define TCP_TIMER_CANCEL(tcp, id)       \
 110         tcp_timeout_cancel(tcp->tcp_connp, id)
 111 
 112 /*
 113  * To restart the TCP retransmission timer.  intvl is in millisec.
 114  */
 115 #define TCP_TIMER_RESTART(tcp, intvl) {                                 \
 116         if ((tcp)->tcp_timer_tid != 0)                                       \
 117                 (void) TCP_TIMER_CANCEL((tcp), (tcp)->tcp_timer_tid);        \
 118         (tcp)->tcp_timer_tid = TCP_TIMER((tcp), tcp_timer, (intvl)); \
 119 }
 120 
 121 /*
 122  * For scalability, we must not run a timer for every TCP connection
 123  * in TIME_WAIT state.  To see why, consider (for time wait interval of
 124  * 1 minutes):
 125  *      10,000 connections/sec * 60 seconds/time wait = 600,000 active conn's
 126  *
 127  * This list is ordered by time, so you need only delete from the head
 128  * until you get to entries which aren't old enough to delete yet.
 129  * The list consists of only the detached TIME_WAIT connections.
 130  *
 131  * When a tcp_t enters TIME_WAIT state, a timer is started (timeout is
 132  * tcps_time_wait_interval).  When the tcp_t is detached (upper layer closes
 133  * the end point), it is moved to the time wait list and another timer is
 134  * started (expiry time is set at tcp_time_wait_expire, which is
 135  * also calculated using tcps_time_wait_interval).  This means that the
 136  * TIME_WAIT state can be extended (up to doubled) if the tcp_t doesn't
 137  * become detached for a long time.
 138  *
 139  * The list manipulations (including tcp_time_wait_next/prev)
 140  * are protected by the tcp_time_wait_lock. The content of the
 141  * detached TIME_WAIT connections is protected by the normal perimeters.
 142  *
 143  * This list is per squeue and squeues are shared across the tcp_stack_t's.
 144  * Things on tcp_time_wait_head remain associated with the tcp_stack_t
 145  * and conn_netstack.
 146  * The tcp_t's that are added to tcp_free_list are disassociated and
 147  * have NULL tcp_tcps and conn_netstack pointers.
 148  */
 149 typedef struct tcp_squeue_priv_s {
 150         kmutex_t        tcp_time_wait_lock;
 151         callout_id_t    tcp_time_wait_tid;
 152         tcp_t           *tcp_time_wait_head;
 153         tcp_t           *tcp_time_wait_tail;
 154         tcp_t           *tcp_free_list;
 155         uint_t          tcp_free_list_cnt;
 156 #ifdef DEBUG
 157         /*
 158          * For debugging purpose, true when tcp_time_wait_collector() is
 159          * running.
 160          */
 161         boolean_t       tcp_time_wait_running;
 162 #endif
 163 } tcp_squeue_priv_t;
 164 
 165 /*
 166  * Parameters for TCP Initial Send Sequence number (ISS) generation.  When
 167  * tcp_strong_iss is set to 1, which is the default, the ISS is calculated
 168  * by adding three components: a time component which grows by 1 every 4096
 169  * nanoseconds (versus every 4 microseconds suggested by RFC 793, page 27);
 170  * a per-connection component which grows by 125000 for every new connection;
 171  * and an "extra" component that grows by a random amount centered
 172  * approximately on 64000.  This causes the ISS generator to cycle every
 173  * 4.89 hours if no TCP connections are made, and faster if connections are
 174  * made.
 175  *
 176  * When tcp_strong_iss is set to 0, ISS is calculated by adding two
 177  * components: a time component which grows by 250000 every second; and
 178  * a per-connection component which grows by 125000 for every new connections.
 179  *
 180  * A third method, when tcp_strong_iss is set to 2, for generating ISS is
 181  * prescribed by Steve Bellovin.  This involves adding time, the 125000 per
 182  * connection, and a one-way hash (MD5) of the connection ID <sport, dport,
 183  * src, dst>, a "truly" random (per RFC 1750) number, and a console-entered
 184  * password.
 185  */
 186 #define ISS_INCR        250000
 187 #define ISS_NSEC_SHT    12
 188 
 189 /* Macros for timestamp comparisons */
 190 #define TSTMP_GEQ(a, b) ((int32_t)((a)-(b)) >= 0)
 191 #define TSTMP_LT(a, b)  ((int32_t)((a)-(b)) < 0)
 192 
 193 /*
 194  * Initialize cwnd according to RFC 3390.  def_max_init_cwnd is
 195  * either tcp_slow_start_initial or tcp_slow_start_after idle
 196  * depending on the caller.  If the upper layer has not used the
 197  * TCP_INIT_CWND option to change the initial cwnd, tcp_init_cwnd
 198  * should be 0 and we use the formula in RFC 3390 to set tcp_cwnd.
 199  * If the upper layer has changed set the tcp_init_cwnd, just use
 200  * it to calculate the tcp_cwnd.
 201  */
 202 #define TCP_SET_INIT_CWND(tcp, mss, def_max_init_cwnd)                  \
 203 {                                                                       \
 204         if ((tcp)->tcp_init_cwnd == 0) {                             \
 205                 (tcp)->tcp_cwnd = MIN(def_max_init_cwnd * (mss),     \
 206                     MIN(4 * (mss), MAX(2 * (mss), 4380 / (mss) * (mss)))); \
 207         } else {                                                        \
 208                 (tcp)->tcp_cwnd = (tcp)->tcp_init_cwnd * (mss);           \
 209         }                                                               \
 210         tcp->tcp_cwnd_cnt = 0;                                               \
 211 }
 212 
 213 /*
 214  * Set ECN capable transport (ECT) code point in IP header.
 215  *
 216  * Note that there are 2 ECT code points '01' and '10', which are called
 217  * ECT(1) and ECT(0) respectively.  Here we follow the original ECT code
 218  * point ECT(0) for TCP as described in RFC 2481.
 219  */
 220 #define TCP_SET_ECT(tcp, iph) \
 221         if ((tcp)->tcp_connp->conn_ipversion == IPV4_VERSION) { \
 222                 /* We need to clear the code point first. */ \
 223                 ((ipha_t *)(iph))->ipha_type_of_service &= 0xFC; \
 224                 ((ipha_t *)(iph))->ipha_type_of_service |= IPH_ECN_ECT0; \
 225         } else { \
 226                 ((ip6_t *)(iph))->ip6_vcf &= htonl(0xFFCFFFFF); \
 227                 ((ip6_t *)(iph))->ip6_vcf |= htonl(IPH_ECN_ECT0 << 20); \
 228         }
 229 
 230 /*
 231  * Set tcp_rto with boundary checking.
 232  */
 233 #define TCP_SET_RTO(tcp, rto) \
 234         if ((rto) < (tcp)->tcp_rto_min)                   \
 235                 (tcp)->tcp_rto = (tcp)->tcp_rto_min;      \
 236         else if ((rto) > (tcp)->tcp_rto_max)              \
 237                 (tcp)->tcp_rto = (tcp)->tcp_rto_max;      \
 238         else                                            \
 239                 (tcp)->tcp_rto = (rto);
 240 
 241 /*
 242  * TCP options struct returned from tcp_parse_options.
 243  */
 244 typedef struct tcp_opt_s {
 245         uint32_t        tcp_opt_mss;
 246         uint32_t        tcp_opt_wscale;
 247         uint32_t        tcp_opt_ts_val;
 248         uint32_t        tcp_opt_ts_ecr;
 249         tcp_t           *tcp;
 250 } tcp_opt_t;
 251 
 252 /*
 253  * Write-side flow-control is implemented via the per instance STREAMS
 254  * write-side Q by explicitly setting QFULL to stop the flow of mblk_t(s)
 255  * and clearing QFULL and calling qbackenable() to restart the flow based
 256  * on the number of TCP unsent bytes (i.e. those not on the wire waiting
 257  * for a remote ACK).
 258  *
 259  * This is different than a standard STREAMS kmod which when using the
 260  * STREAMS Q the framework would automatictly flow-control based on the
 261  * defined hiwat/lowat values as mblk_t's are enqueued/dequeued.
 262  *
 263  * As of FireEngine TCP write-side flow-control needs to take into account
 264  * both the unsent tcp_xmit list bytes but also any squeue_t enqueued bytes
 265  * (i.e. from tcp_wput() -> tcp_output()).
 266  *
 267  * This is accomplished by adding a new tcp_t fields, tcp_squeue_bytes, to
 268  * count the number of bytes enqueued by tcp_wput() and the number of bytes
 269  * dequeued and processed by tcp_output().
 270  *
 271  * So, the total number of bytes unsent is (squeue_bytes + unsent) with all
 272  * flow-control uses of unsent replaced with the macro TCP_UNSENT_BYTES.
 273  */
 274 extern void     tcp_clrqfull(tcp_t *);
 275 extern void     tcp_setqfull(tcp_t *);
 276 
 277 #define TCP_UNSENT_BYTES(tcp) \
 278         ((tcp)->tcp_squeue_bytes + (tcp)->tcp_unsent)
 279 
 280 /*
 281  * Linked list struct to store listener connection limit configuration per
 282  * IP stack.  The list is stored at tcps_listener_conf in tcp_stack_t.
 283  *
 284  * tl_port: the listener port of this limit configuration
 285  * tl_ratio: the maximum amount of memory consumed by all concurrent TCP
 286  *           connections created by a listener does not exceed 1/tl_ratio
 287  *           of the total system memory.  Note that this is only an
 288  *           approximation.
 289  * tl_link: linked list struct
 290  */
 291 typedef struct tcp_listener_s {
 292         in_port_t       tl_port;
 293         uint32_t        tl_ratio;
 294         list_node_t     tl_link;
 295 } tcp_listener_t;
 296 
 297 /*
 298  * If there is a limit set on the number of connections allowed per each
 299  * listener, the following struct is used to store that counter.  It keeps
 300  * the number of TCP connection created by a listener.  Note that this needs
 301  * to be separated from the listener since the listener can go away before
 302  * all the connections are gone.
 303  *
 304  * When the struct is allocated, tlc_cnt is set to 1.  When a new connection
 305  * is created by the listener, tlc_cnt is incremented by 1.  When a connection
 306  * created by the listener goes away, tlc_count is decremented by 1.  When the
 307  * listener itself goes away, tlc_cnt is decremented  by one.  The last
 308  * connection (or the listener) which decrements tlc_cnt to zero frees the
 309  * struct.
 310  *
 311  * tlc_max is the maximum number of concurrent TCP connections created from a
 312  * listner.  It is calculated when the tcp_listen_cnt_t is allocated.
 313  *
 314  * tlc_report_time stores the time when cmn_err() is called to report that the
 315  * max has been exceeeded.  Report is done at most once every
 316  * TCP_TLC_REPORT_INTERVAL mins for a listener.
 317  *
 318  * tlc_drop stores the number of connection attempt dropped because the
 319  * limit has reached.
 320  */
 321 typedef struct tcp_listen_cnt_s {
 322         uint32_t        tlc_max;
 323         uint32_t        tlc_cnt;
 324         int64_t         tlc_report_time;
 325         uint32_t        tlc_drop;
 326 } tcp_listen_cnt_t;
 327 
 328 #define TCP_TLC_REPORT_INTERVAL (30 * MINUTES)
 329 
 330 #define TCP_DECR_LISTEN_CNT(tcp)                                        \
 331 {                                                                       \
 332         ASSERT((tcp)->tcp_listen_cnt->tlc_cnt > 0);                    \
 333         if (atomic_add_32_nv(&(tcp)->tcp_listen_cnt->tlc_cnt, -1) == 0) \
 334                 kmem_free((tcp)->tcp_listen_cnt, sizeof (tcp_listen_cnt_t)); \
 335         (tcp)->tcp_listen_cnt = NULL;                                        \
 336 }
 337 
 338 /* Increment and decrement the number of connections in tcp_stack_t. */
 339 #define TCPS_CONN_INC(tcps)                                             \
 340         atomic_inc_64(                                                  \
 341             (uint64_t *)&(tcps)->tcps_sc[CPU->cpu_seqid]->tcp_sc_conn_cnt)
 342 
 343 #define TCPS_CONN_DEC(tcps)                                             \
 344         atomic_dec_64(                                                  \
 345             (uint64_t *)&(tcps)->tcps_sc[CPU->cpu_seqid]->tcp_sc_conn_cnt)
 346 
 347 /*
 348  * When the system is under memory pressure, stack variable tcps_reclaim is
 349  * true, we shorten the connection timeout abort interval to tcp_early_abort
 350  * seconds.  Defined in tcp.c.
 351  */
 352 extern uint32_t tcp_early_abort;
 353 
 354 /*
 355  * To reach to an eager in Q0 which can be dropped due to an incoming
 356  * new SYN request when Q0 is full, a new doubly linked list is
 357  * introduced. This list allows to select an eager from Q0 in O(1) time.
 358  * This is needed to avoid spending too much time walking through the
 359  * long list of eagers in Q0 when tcp_drop_q0() is called. Each member of
 360  * this new list has to be a member of Q0.
 361  * This list is headed by listener's tcp_t. When the list is empty,
 362  * both the pointers - tcp_eager_next_drop_q0 and tcp_eager_prev_drop_q0,
 363  * of listener's tcp_t point to listener's tcp_t itself.
 364  *
 365  * Given an eager in Q0 and a listener, MAKE_DROPPABLE() puts the eager
 366  * in the list. MAKE_UNDROPPABLE() takes the eager out of the list.
 367  * These macros do not affect the eager's membership to Q0.
 368  */
 369 #define MAKE_DROPPABLE(listener, eager)                                 \
 370         if ((eager)->tcp_eager_next_drop_q0 == NULL) {                       \
 371                 (listener)->tcp_eager_next_drop_q0->tcp_eager_prev_drop_q0\
 372                     = (eager);                                          \
 373                 (eager)->tcp_eager_prev_drop_q0 = (listener);                \
 374                 (eager)->tcp_eager_next_drop_q0 =                    \
 375                     (listener)->tcp_eager_next_drop_q0;                      \
 376                 (listener)->tcp_eager_next_drop_q0 = (eager);                \
 377         }
 378 
 379 #define MAKE_UNDROPPABLE(eager)                                         \
 380         if ((eager)->tcp_eager_next_drop_q0 != NULL) {                       \
 381                 (eager)->tcp_eager_next_drop_q0->tcp_eager_prev_drop_q0   \
 382                     = (eager)->tcp_eager_prev_drop_q0;                       \
 383                 (eager)->tcp_eager_prev_drop_q0->tcp_eager_next_drop_q0   \
 384                     = (eager)->tcp_eager_next_drop_q0;                       \
 385                 (eager)->tcp_eager_prev_drop_q0 = NULL;                      \
 386                 (eager)->tcp_eager_next_drop_q0 = NULL;                      \
 387         }
 388 
 389 /*
 390  * The format argument to pass to tcp_display().
 391  * DISP_PORT_ONLY means that the returned string has only port info.
 392  * DISP_ADDR_AND_PORT means that the returned string also contains the
 393  * remote and local IP address.
 394  */
 395 #define DISP_PORT_ONLY          1
 396 #define DISP_ADDR_AND_PORT      2
 397 
 398 #define IP_ADDR_CACHE_SIZE      2048
 399 #define IP_ADDR_CACHE_HASH(faddr)                                       \
 400         (ntohl(faddr) & (IP_ADDR_CACHE_SIZE -1))
 401 
 402 /* TCP cwnd burst factor. */
 403 #define TCP_CWND_INFINITE       65535
 404 #define TCP_CWND_SS             3
 405 #define TCP_CWND_NORMAL         5
 406 
 407 /*
 408  * TCP reassembly macros.  We hide starting and ending sequence numbers in
 409  * b_next and b_prev of messages on the reassembly queue.  The messages are
 410  * chained using b_cont.  These macros are used in tcp_reass() so we don't
 411  * have to see the ugly casts and assignments.
 412  */
 413 #define TCP_REASS_SEQ(mp)               ((uint32_t)(uintptr_t)((mp)->b_next))
 414 #define TCP_REASS_SET_SEQ(mp, u)        ((mp)->b_next = \
 415                                         (mblk_t *)(uintptr_t)(u))
 416 #define TCP_REASS_END(mp)               ((uint32_t)(uintptr_t)((mp)->b_prev))
 417 #define TCP_REASS_SET_END(mp, u)        ((mp)->b_prev = \
 418                                         (mblk_t *)(uintptr_t)(u))
 419 
 420 #define tcps_time_wait_interval         tcps_propinfo_tbl[0].prop_cur_uval
 421 #define tcps_conn_req_max_q             tcps_propinfo_tbl[1].prop_cur_uval
 422 #define tcps_conn_req_max_q0            tcps_propinfo_tbl[2].prop_cur_uval
 423 #define tcps_conn_req_min               tcps_propinfo_tbl[3].prop_cur_uval
 424 #define tcps_conn_grace_period          tcps_propinfo_tbl[4].prop_cur_uval
 425 #define tcps_cwnd_max_                  tcps_propinfo_tbl[5].prop_cur_uval
 426 #define tcps_dbg                        tcps_propinfo_tbl[6].prop_cur_uval
 427 #define tcps_smallest_nonpriv_port      tcps_propinfo_tbl[7].prop_cur_uval
 428 #define tcps_ip_abort_cinterval         tcps_propinfo_tbl[8].prop_cur_uval
 429 #define tcps_ip_abort_linterval         tcps_propinfo_tbl[9].prop_cur_uval
 430 #define tcps_ip_abort_interval          tcps_propinfo_tbl[10].prop_cur_uval
 431 #define tcps_ip_notify_cinterval        tcps_propinfo_tbl[11].prop_cur_uval
 432 #define tcps_ip_notify_interval         tcps_propinfo_tbl[12].prop_cur_uval
 433 #define tcps_ipv4_ttl                   tcps_propinfo_tbl[13].prop_cur_uval
 434 #define tcps_keepalive_interval_high    tcps_propinfo_tbl[14].prop_max_uval
 435 #define tcps_keepalive_interval         tcps_propinfo_tbl[14].prop_cur_uval
 436 #define tcps_keepalive_interval_low     tcps_propinfo_tbl[14].prop_min_uval
 437 #define tcps_maxpsz_multiplier          tcps_propinfo_tbl[15].prop_cur_uval
 438 #define tcps_mss_def_ipv4               tcps_propinfo_tbl[16].prop_cur_uval
 439 #define tcps_mss_max_ipv4               tcps_propinfo_tbl[17].prop_cur_uval
 440 #define tcps_mss_min                    tcps_propinfo_tbl[18].prop_cur_uval
 441 #define tcps_naglim_def                 tcps_propinfo_tbl[19].prop_cur_uval
 442 #define tcps_rexmit_interval_initial_high       \
 443                                         tcps_propinfo_tbl[20].prop_max_uval
 444 #define tcps_rexmit_interval_initial    tcps_propinfo_tbl[20].prop_cur_uval
 445 #define tcps_rexmit_interval_initial_low        \
 446                                         tcps_propinfo_tbl[20].prop_min_uval
 447 #define tcps_rexmit_interval_max_high   tcps_propinfo_tbl[21].prop_max_uval
 448 #define tcps_rexmit_interval_max        tcps_propinfo_tbl[21].prop_cur_uval
 449 #define tcps_rexmit_interval_max_low    tcps_propinfo_tbl[21].prop_min_uval
 450 #define tcps_rexmit_interval_min_high   tcps_propinfo_tbl[22].prop_max_uval
 451 #define tcps_rexmit_interval_min        tcps_propinfo_tbl[22].prop_cur_uval
 452 #define tcps_rexmit_interval_min_low    tcps_propinfo_tbl[22].prop_min_uval
 453 #define tcps_deferred_ack_interval      tcps_propinfo_tbl[23].prop_cur_uval
 454 #define tcps_snd_lowat_fraction         tcps_propinfo_tbl[24].prop_cur_uval
 455 #define tcps_dupack_fast_retransmit     tcps_propinfo_tbl[25].prop_cur_uval
 456 #define tcps_ignore_path_mtu            tcps_propinfo_tbl[26].prop_cur_bval
 457 #define tcps_smallest_anon_port         tcps_propinfo_tbl[27].prop_cur_uval
 458 #define tcps_largest_anon_port          tcps_propinfo_tbl[28].prop_cur_uval
 459 #define tcps_xmit_hiwat                 tcps_propinfo_tbl[29].prop_cur_uval
 460 #define tcps_xmit_lowat                 tcps_propinfo_tbl[30].prop_cur_uval
 461 #define tcps_recv_hiwat                 tcps_propinfo_tbl[31].prop_cur_uval
 462 #define tcps_recv_hiwat_minmss          tcps_propinfo_tbl[32].prop_cur_uval
 463 #define tcps_fin_wait_2_flush_interval_high     \
 464                                         tcps_propinfo_tbl[33].prop_max_uval
 465 #define tcps_fin_wait_2_flush_interval  tcps_propinfo_tbl[33].prop_cur_uval
 466 #define tcps_fin_wait_2_flush_interval_low      \
 467                                         tcps_propinfo_tbl[33].prop_min_uval
 468 #define tcps_max_buf                    tcps_propinfo_tbl[34].prop_cur_uval
 469 #define tcps_strong_iss                 tcps_propinfo_tbl[35].prop_cur_uval
 470 #define tcps_rtt_updates                tcps_propinfo_tbl[36].prop_cur_uval
 471 #define tcps_wscale_always              tcps_propinfo_tbl[37].prop_cur_bval
 472 #define tcps_tstamp_always              tcps_propinfo_tbl[38].prop_cur_bval
 473 #define tcps_tstamp_if_wscale           tcps_propinfo_tbl[39].prop_cur_bval
 474 #define tcps_rexmit_interval_extra      tcps_propinfo_tbl[40].prop_cur_uval
 475 #define tcps_deferred_acks_max          tcps_propinfo_tbl[41].prop_cur_uval
 476 #define tcps_slow_start_after_idle      tcps_propinfo_tbl[42].prop_cur_uval
 477 #define tcps_slow_start_initial         tcps_propinfo_tbl[43].prop_cur_uval
 478 #define tcps_sack_permitted             tcps_propinfo_tbl[44].prop_cur_uval
 479 #define tcps_ipv6_hoplimit              tcps_propinfo_tbl[45].prop_cur_uval
 480 #define tcps_mss_def_ipv6               tcps_propinfo_tbl[46].prop_cur_uval
 481 #define tcps_mss_max_ipv6               tcps_propinfo_tbl[47].prop_cur_uval
 482 #define tcps_rev_src_routes             tcps_propinfo_tbl[48].prop_cur_bval
 483 #define tcps_local_dack_interval        tcps_propinfo_tbl[49].prop_cur_uval
 484 #define tcps_local_dacks_max            tcps_propinfo_tbl[50].prop_cur_uval
 485 #define tcps_ecn_permitted              tcps_propinfo_tbl[51].prop_cur_uval
 486 #define tcps_rst_sent_rate_enabled      tcps_propinfo_tbl[52].prop_cur_bval
 487 #define tcps_rst_sent_rate              tcps_propinfo_tbl[53].prop_cur_uval
 488 #define tcps_push_timer_interval        tcps_propinfo_tbl[54].prop_cur_uval
 489 #define tcps_use_smss_as_mss_opt        tcps_propinfo_tbl[55].prop_cur_bval
 490 #define tcps_keepalive_abort_interval_high \
 491                                         tcps_propinfo_tbl[56].prop_max_uval
 492 #define tcps_keepalive_abort_interval \
 493                                         tcps_propinfo_tbl[56].prop_cur_uval
 494 #define tcps_keepalive_abort_interval_low \
 495                                         tcps_propinfo_tbl[56].prop_min_uval
 496 #define tcps_wroff_xtra                 tcps_propinfo_tbl[57].prop_cur_uval
 497 #define tcps_dev_flow_ctl               tcps_propinfo_tbl[58].prop_cur_bval
 498 #define tcps_reass_timeout              tcps_propinfo_tbl[59].prop_cur_uval
 499 
 500 extern struct qinit tcp_rinitv4, tcp_rinitv6;
 501 extern boolean_t do_tcp_fusion;
 502 
 503 /*
 504  * Object to represent database of options to search passed to
 505  * {sock,tpi}optcom_req() interface routine to take care of option
 506  * management and associated methods.
 507  */
 508 extern optdb_obj_t      tcp_opt_obj;
 509 extern uint_t           tcp_max_optsize;
 510 
 511 extern int tcp_squeue_flag;
 512 
 513 extern uint_t tcp_free_list_max_cnt;
 514 
 515 /*
 516  * Functions in tcp.c.
 517  */
 518 extern void     tcp_acceptor_hash_insert(t_uscalar_t, tcp_t *);
 519 extern tcp_t    *tcp_acceptor_hash_lookup(t_uscalar_t, tcp_stack_t *);
 520 extern void     tcp_acceptor_hash_remove(tcp_t *);
 521 extern mblk_t   *tcp_ack_mp(tcp_t *);
 522 extern int      tcp_build_hdrs(tcp_t *);
 523 extern void     tcp_cleanup(tcp_t *);
 524 extern int      tcp_clean_death(tcp_t *, int);
 525 extern void     tcp_clean_death_wrapper(void *, mblk_t *, void *,
 526                     ip_recv_attr_t *);
 527 extern void     tcp_close_common(conn_t *, int);
 528 extern void     tcp_close_detached(tcp_t *);
 529 extern void     tcp_close_mpp(mblk_t **);
 530 extern void     tcp_closei_local(tcp_t *);
 531 extern sock_lower_handle_t tcp_create(int, int, int, sock_downcalls_t **,
 532                     uint_t *, int *, int, cred_t *);
 533 extern conn_t   *tcp_create_common(cred_t *, boolean_t, boolean_t, int *);
 534 extern void     tcp_disconnect(tcp_t *, mblk_t *);
 535 extern char     *tcp_display(tcp_t *, char *, char);
 536 extern int      tcp_do_bind(conn_t *, struct sockaddr *, socklen_t, cred_t *,
 537                     boolean_t);
 538 extern int      tcp_do_connect(conn_t *, const struct sockaddr *, socklen_t,
 539                     cred_t *, pid_t);
 540 extern int      tcp_do_listen(conn_t *, struct sockaddr *, socklen_t, int,
 541                     cred_t *, boolean_t);
 542 extern int      tcp_do_unbind(conn_t *);
 543 extern boolean_t        tcp_eager_blowoff(tcp_t *, t_scalar_t);
 544 extern void     tcp_eager_cleanup(tcp_t *, boolean_t);
 545 extern void     tcp_eager_kill(void *, mblk_t *, void *, ip_recv_attr_t *);
 546 extern void     tcp_eager_unlink(tcp_t *);
 547 extern int      tcp_getpeername(sock_lower_handle_t, struct sockaddr *,
 548                     socklen_t *, cred_t *);
 549 extern int      tcp_getsockname(sock_lower_handle_t, struct sockaddr *,
 550                     socklen_t *, cred_t *);
 551 extern void     tcp_init_values(tcp_t *, tcp_t *);
 552 extern void     tcp_ipsec_cleanup(tcp_t *);
 553 extern int      tcp_maxpsz_set(tcp_t *, boolean_t);
 554 extern void     tcp_mss_set(tcp_t *, uint32_t);
 555 extern void     tcp_reinput(conn_t *, mblk_t *, ip_recv_attr_t *, ip_stack_t *);
 556 extern void     tcp_rsrv(queue_t *);
 557 extern uint_t   tcp_rwnd_reopen(tcp_t *);
 558 extern int      tcp_rwnd_set(tcp_t *, uint32_t);
 559 extern int      tcp_set_destination(tcp_t *);
 560 extern void     tcp_set_ws_value(tcp_t *);
 561 extern void     tcp_stop_lingering(tcp_t *);
 562 extern void     tcp_update_pmtu(tcp_t *, boolean_t);
 563 extern mblk_t   *tcp_zcopy_backoff(tcp_t *, mblk_t *, boolean_t);
 564 extern boolean_t        tcp_zcopy_check(tcp_t *);
 565 extern void     tcp_zcopy_notify(tcp_t *);
 566 extern void     tcp_get_proto_props(tcp_t *, struct sock_proto_props *);
 567 
 568 /*
 569  * Bind related functions in tcp_bind.c
 570  */
 571 extern int      tcp_bind_check(conn_t *, struct sockaddr *, socklen_t,
 572                     cred_t *, boolean_t);
 573 extern void     tcp_bind_hash_insert(tf_t *, tcp_t *, int);
 574 extern void     tcp_bind_hash_remove(tcp_t *);
 575 extern in_port_t        tcp_bindi(tcp_t *, in_port_t, const in6_addr_t *,
 576                             int, boolean_t, boolean_t, boolean_t);
 577 extern in_port_t        tcp_update_next_port(in_port_t, const tcp_t *,
 578                             boolean_t);
 579 
 580 /*
 581  * Fusion related functions in tcp_fusion.c.
 582  */
 583 extern void     tcp_fuse(tcp_t *, uchar_t *, tcpha_t *);
 584 extern void     tcp_unfuse(tcp_t *);
 585 extern boolean_t tcp_fuse_output(tcp_t *, mblk_t *, uint32_t);
 586 extern void     tcp_fuse_output_urg(tcp_t *, mblk_t *);
 587 extern boolean_t tcp_fuse_rcv_drain(queue_t *, tcp_t *, mblk_t **);
 588 extern size_t   tcp_fuse_set_rcv_hiwat(tcp_t *, size_t);
 589 extern int      tcp_fuse_maxpsz(tcp_t *);
 590 extern void     tcp_fuse_backenable(tcp_t *);
 591 extern void     tcp_iss_key_init(uint8_t *, int, tcp_stack_t *);
 592 
 593 /*
 594  * Output related functions in tcp_output.c.
 595  */
 596 extern void     tcp_close_output(void *, mblk_t *, void *, ip_recv_attr_t *);
 597 extern void     tcp_output(void *, mblk_t *, void *, ip_recv_attr_t *);
 598 extern void     tcp_output_urgent(void *, mblk_t *, void *, ip_recv_attr_t *);
 599 extern void     tcp_rexmit_after_error(tcp_t *);
 600 extern void     tcp_sack_rexmit(tcp_t *, uint_t *);
 601 extern void     tcp_send_data(tcp_t *, mblk_t *);
 602 extern void     tcp_send_synack(void *, mblk_t *, void *, ip_recv_attr_t *);
 603 extern void     tcp_shutdown_output(void *, mblk_t *, void *, ip_recv_attr_t *);
 604 extern void     tcp_ss_rexmit(tcp_t *);
 605 extern void     tcp_update_xmit_tail(tcp_t *, uint32_t);
 606 extern void     tcp_wput(queue_t *, mblk_t *);
 607 extern void     tcp_wput_data(tcp_t *, mblk_t *, boolean_t);
 608 extern void     tcp_wput_sock(queue_t *, mblk_t *);
 609 extern void     tcp_wput_fallback(queue_t *, mblk_t *);
 610 extern void     tcp_xmit_ctl(char *, tcp_t *, uint32_t, uint32_t, int);
 611 extern void     tcp_xmit_listeners_reset(mblk_t *, ip_recv_attr_t *,
 612                     ip_stack_t *i, conn_t *);
 613 extern mblk_t   *tcp_xmit_mp(tcp_t *, mblk_t *, int32_t, int32_t *,
 614                     mblk_t **, uint32_t, boolean_t, uint32_t *, boolean_t);
 615 
 616 /*
 617  * Input related functions in tcp_input.c.
 618  */
 619 extern void     tcp_icmp_input(void *, mblk_t *, void *, ip_recv_attr_t *);
 620 extern void     tcp_input_data(void *, mblk_t *, void *, ip_recv_attr_t *);
 621 extern void     tcp_input_listener_unbound(void *, mblk_t *, void *,
 622                     ip_recv_attr_t *);
 623 extern boolean_t        tcp_paws_check(tcp_t *, tcpha_t *, tcp_opt_t *);
 624 extern uint_t   tcp_rcv_drain(tcp_t *);
 625 extern void     tcp_rcv_enqueue(tcp_t *, mblk_t *, uint_t, cred_t *);
 626 extern boolean_t        tcp_verifyicmp(conn_t *, void *, icmph_t *, icmp6_t *,
 627                             ip_recv_attr_t *);
 628 
 629 /*
 630  * Kernel socket related functions in tcp_socket.c.
 631  */
 632 extern int      tcp_fallback(sock_lower_handle_t, queue_t *, boolean_t,
 633                     so_proto_quiesced_cb_t, sock_quiesce_arg_t *);
 634 extern boolean_t tcp_newconn_notify(tcp_t *, ip_recv_attr_t *);
 635 
 636 /*
 637  * Timer related functions in tcp_timers.c.
 638  */
 639 extern void     tcp_ack_timer(void *);
 640 extern void     tcp_close_linger_timeout(void *);
 641 extern void     tcp_keepalive_timer(void *);
 642 extern void     tcp_push_timer(void *);
 643 extern void     tcp_reass_timer(void *);
 644 extern mblk_t   *tcp_timermp_alloc(int);
 645 extern void     tcp_timermp_free(tcp_t *);
 646 extern timeout_id_t tcp_timeout(conn_t *, void (*)(void *), hrtime_t);
 647 extern clock_t  tcp_timeout_cancel(conn_t *, timeout_id_t);
 648 extern void     tcp_timer(void *arg);
 649 extern void     tcp_timers_stop(tcp_t *);
 650 
 651 /*
 652  * TCP TPI related functions in tcp_tpi.c.
 653  */
 654 extern void     tcp_addr_req(tcp_t *, mblk_t *);
 655 extern void     tcp_capability_req(tcp_t *, mblk_t *);
 656 extern boolean_t        tcp_conn_con(tcp_t *, uchar_t *, mblk_t *,
 657                             mblk_t **, ip_recv_attr_t *);
 658 extern void     tcp_err_ack(tcp_t *, mblk_t *, int, int);
 659 extern void     tcp_err_ack_prim(tcp_t *, mblk_t *, int, int, int);
 660 extern void     tcp_info_req(tcp_t *, mblk_t *);
 661 extern void     tcp_send_conn_ind(void *, mblk_t *, void *);
 662 extern void     tcp_send_pending(void *, mblk_t *, void *, ip_recv_attr_t *);
 663 extern void     tcp_tpi_accept(queue_t *, mblk_t *);
 664 extern void     tcp_tpi_bind(tcp_t *, mblk_t *);
 665 extern int      tcp_tpi_close(queue_t *, int);
 666 extern int      tcp_tpi_close_accept(queue_t *);
 667 extern void     tcp_tpi_connect(tcp_t *, mblk_t *);
 668 extern int      tcp_tpi_opt_get(queue_t *, t_scalar_t, t_scalar_t, uchar_t *);
 669 extern int      tcp_tpi_opt_set(queue_t *, uint_t, int, int, uint_t, uchar_t *,
 670                     uint_t *, uchar_t *, void *, cred_t *);
 671 extern void     tcp_tpi_unbind(tcp_t *, mblk_t *);
 672 extern void     tcp_tli_accept(tcp_t *, mblk_t *);
 673 extern void     tcp_use_pure_tpi(tcp_t *);
 674 extern void     tcp_do_capability_ack(tcp_t *, struct T_capability_ack *,
 675                     t_uscalar_t);
 676 
 677 /*
 678  * TCP option processing related functions in tcp_opt_data.c
 679  */
 680 extern int      tcp_opt_default(queue_t *, t_scalar_t, t_scalar_t, uchar_t *);
 681 extern int      tcp_opt_get(conn_t *, int, int, uchar_t *);
 682 extern int      tcp_opt_set(conn_t *, uint_t, int, int, uint_t, uchar_t *,
 683                     uint_t *, uchar_t *, void *, cred_t *);
 684 
 685 /*
 686  * TCP time wait processing related functions in tcp_time_wait.c.
 687  */
 688 extern void             tcp_time_wait_append(tcp_t *);
 689 extern void             tcp_time_wait_collector(void *);
 690 extern boolean_t        tcp_time_wait_remove(tcp_t *, tcp_squeue_priv_t *);
 691 extern void             tcp_time_wait_processing(tcp_t *, mblk_t *, uint32_t,
 692                             uint32_t, int, tcpha_t *, ip_recv_attr_t *);
 693 
 694 /*
 695  * Misc functions in tcp_misc.c.
 696  */
 697 extern uint32_t tcp_find_listener_conf(tcp_stack_t *, in_port_t);
 698 extern void     tcp_ioctl_abort_conn(queue_t *, mblk_t *);
 699 extern void     tcp_listener_conf_cleanup(tcp_stack_t *);
 700 extern void     tcp_stack_cpu_add(tcp_stack_t *, processorid_t);
 701 
 702 #endif  /* _KERNEL */
 703 
 704 #ifdef  __cplusplus
 705 }
 706 #endif
 707 
 708 #endif  /* _INET_TCP_IMPL_H */