sys/rpc/clnt_dg.c

   1 /*      $NetBSD: clnt_dg.c,v 1.4 2000/07/14 08:40:41 fvdl Exp $ */
   2
   3 /*-
   4  * SPDX-License-Identifier: BSD-3-Clause
   5  *
   6  * Copyright (c) 2009, Sun Microsystems, Inc.
   7  * All rights reserved.
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions are met:
  11  * - Redistributions of source code must retain the above copyright notice,
  12  *   this list of conditions and the following disclaimer.
  13  * - Redistributions in binary form must reproduce the above copyright notice,
  14  *   this list of conditions and the following disclaimer in the documentation
  15  *   and/or other materials provided with the distribution.
  16  * - Neither the name of Sun Microsystems, Inc. nor the names of its
  17  *   contributors may be used to endorse or promote products derived
  18  *   from this software without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  21  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  23  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
  24  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  30  * POSSIBILITY OF SUCH DAMAGE.
  31  */
  32 /*
  33  * Copyright (c) 1986-1991 by Sun Microsystems Inc.
  34  */
  35
  36 #if defined(LIBC_SCCS) && !defined(lint)
  37 #ident  "@(#)clnt_dg.c  1.23    94/04/22 SMI"
  38 static char sccsid[] = "@(#)clnt_dg.c 1.19 89/03/16 Copyr 1988 Sun Micro";
  39 #endif
  40 #include <sys/cdefs.h>
  41 /*
  42  * Implements a connectionless client side RPC.
  43  */
  44
  45 #include <sys/param.h>
  46 #include <sys/systm.h>
  47 #include <sys/kernel.h>
  48 #include <sys/lock.h>
  49 #include <sys/malloc.h>
  50 #include <sys/mbuf.h>
  51 #include <sys/mutex.h>
  52 #include <sys/pcpu.h>
  53 #include <sys/proc.h>
  54 #include <sys/socket.h>
  55 #include <sys/socketvar.h>
  56 #include <sys/time.h>
  57 #include <sys/uio.h>
  58
  59 #include <net/vnet.h>
  60
  61 #include <rpc/rpc.h>
  62 #include <rpc/rpc_com.h>
  63
  64
  65 #ifdef _FREEFALL_CONFIG
  66 /*
  67  * Disable RPC exponential back-off for FreeBSD.org systems.
  68  */
  69 #define RPC_MAX_BACKOFF         1 /* second */
  70 #else
  71 #define RPC_MAX_BACKOFF         30 /* seconds */
  72 #endif
  73
  74 static bool_t time_not_ok(struct timeval *);
  75 static enum clnt_stat clnt_dg_call(CLIENT *, struct rpc_callextra *,
  76     rpcproc_t, struct mbuf *, struct mbuf **, struct timeval);
  77 static void clnt_dg_geterr(CLIENT *, struct rpc_err *);
  78 static bool_t clnt_dg_freeres(CLIENT *, xdrproc_t, void *);
  79 static void clnt_dg_abort(CLIENT *);
  80 static bool_t clnt_dg_control(CLIENT *, u_int, void *);
  81 static void clnt_dg_close(CLIENT *);
  82 static void clnt_dg_destroy(CLIENT *);
  83 static int clnt_dg_soupcall(struct socket *so, void *arg, int waitflag);
  84
  85 static const struct clnt_ops clnt_dg_ops = {
  86         .cl_call =      clnt_dg_call,
  87         .cl_abort =     clnt_dg_abort,
  88         .cl_geterr =    clnt_dg_geterr,
  89         .cl_freeres =   clnt_dg_freeres,
  90         .cl_close =     clnt_dg_close,
  91         .cl_destroy =   clnt_dg_destroy,
  92         .cl_control =   clnt_dg_control
  93 };
  94
  95 static volatile uint32_t rpc_xid = 0;
  96
  97 /*
  98  * A pending RPC request which awaits a reply. Requests which have
  99  * received their reply will have cr_xid set to zero and cr_mrep to
 100  * the mbuf chain of the reply.
 101  */
 102 struct cu_request {
 103         TAILQ_ENTRY(cu_request) cr_link;
 104         CLIENT                  *cr_client;     /* owner */
 105         uint32_t                cr_xid;         /* XID of request */
 106         struct mbuf             *cr_mrep;       /* reply received by upcall */
 107         int                     cr_error;       /* any error from upcall */
 108         char                    cr_verf[MAX_AUTH_BYTES]; /* reply verf */
 109 };
 110
 111 TAILQ_HEAD(cu_request_list, cu_request);
 112
 113 #define MCALL_MSG_SIZE 24
 114
 115 /*
 116  * This structure is pointed to by the socket buffer's sb_upcallarg
 117  * member. It is separate from the client private data to facilitate
 118  * multiple clients sharing the same socket. The cs_lock mutex is used
 119  * to protect all fields of this structure, the socket's receive
 120  * buffer SOCKBUF_LOCK is used to ensure that exactly one of these
 121  * structures is installed on the socket.
 122  */
 123 struct cu_socket {
 124         struct mtx              cs_lock;
 125         int                     cs_refs;        /* Count of clients */
 126         struct cu_request_list  cs_pending;     /* Requests awaiting replies */
 127         int                     cs_upcallrefs;  /* Refcnt of upcalls in prog.*/
 128 };
 129
 130 static void clnt_dg_upcallsdone(struct socket *, struct cu_socket *);
 131
 132 /*
 133  * Private data kept per client handle
 134  */
 135 struct cu_data {
 136         int                     cu_threads;     /* # threads in clnt_vc_call */
 137         bool_t                  cu_closing;     /* TRUE if we are closing */
 138         bool_t                  cu_closed;      /* TRUE if we are closed */
 139         struct socket           *cu_socket;     /* connection socket */
 140         bool_t                  cu_closeit;     /* opened by library */
 141         struct sockaddr_storage cu_raddr;       /* remote address */
 142         int                     cu_rlen;
 143         struct timeval          cu_wait;        /* retransmit interval */
 144         struct timeval          cu_total;       /* total time for the call */
 145         struct rpc_err          cu_error;
 146         uint32_t                cu_xid;
 147         char                    cu_mcallc[MCALL_MSG_SIZE]; /* marshalled callmsg */
 148         size_t                  cu_mcalllen;
 149         size_t                  cu_sendsz;      /* send size */
 150         size_t                  cu_recvsz;      /* recv size */
 151         int                     cu_async;
 152         int                     cu_connect;     /* Use connect(). */
 153         int                     cu_connected;   /* Have done connect(). */
 154         const char              *cu_waitchan;
 155         int                     cu_waitflag;
 156         int                     cu_cwnd;        /* congestion window */
 157         int                     cu_sent;        /* number of in-flight RPCs */
 158         bool_t                  cu_cwnd_wait;
 159 };
 160
 161 #define CWNDSCALE       256
 162 #define MAXCWND         (32 * CWNDSCALE)
 163
 164 /*
 165  * Connection less client creation returns with client handle parameters.
 166  * Default options are set, which the user can change using clnt_control().
 167  * fd should be open and bound.
 168  * NB: The rpch->cl_auth is initialized to null authentication.
 169  *      Caller may wish to set this something more useful.
 170  *
 171  * sendsz and recvsz are the maximum allowable packet sizes that can be
 172  * sent and received. Normally they are the same, but they can be
 173  * changed to improve the program efficiency and buffer allocation.
 174  * If they are 0, use the transport default.
 175  *
 176  * If svcaddr is NULL, returns NULL.
 177  */
 178 CLIENT *
 179 clnt_dg_create(
 180         struct socket *so,
 181         struct sockaddr *svcaddr,       /* servers address */
 182         rpcprog_t program,              /* program number */
 183         rpcvers_t version,              /* version number */
 184         size_t sendsz,                  /* buffer recv size */
 185         size_t recvsz)                  /* buffer send size */
 186 {
 187         CLIENT *cl = NULL;              /* client handle */
 188         struct cu_data *cu = NULL;      /* private data */
 189         struct cu_socket *cs = NULL;
 190         struct sockbuf *sb;
 191         struct timeval now;
 192         struct rpc_msg call_msg;
 193         struct __rpc_sockinfo si;
 194         XDR xdrs;
 195         int error;
 196         uint32_t newxid;
 197
 198         if (svcaddr == NULL) {
 199                 rpc_createerr.cf_stat = RPC_UNKNOWNADDR;
 200                 return (NULL);
 201         }
 202
 203         if (!__rpc_socket2sockinfo(so, &si)) {
 204                 rpc_createerr.cf_stat = RPC_TLIERROR;
 205                 rpc_createerr.cf_error.re_errno = 0;
 206                 return (NULL);
 207         }
 208
 209         /*
 210          * Find the receive and the send size
 211          */
 212         sendsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)sendsz);
 213         recvsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)recvsz);
 214         if ((sendsz == 0) || (recvsz == 0)) {
 215                 rpc_createerr.cf_stat = RPC_TLIERROR; /* XXX */
 216                 rpc_createerr.cf_error.re_errno = 0;
 217                 return (NULL);
 218         }
 219
 220         cl = mem_alloc(sizeof (CLIENT));
 221
 222         /*
 223          * Should be multiple of 4 for XDR.
 224          */
 225         sendsz = rounddown(sendsz + 3, 4);
 226         recvsz = rounddown(recvsz + 3, 4);
 227         cu = mem_alloc(sizeof (*cu));
 228         cu->cu_threads = 0;
 229         cu->cu_closing = FALSE;
 230         cu->cu_closed = FALSE;
 231         (void) memcpy(&cu->cu_raddr, svcaddr, (size_t)svcaddr->sa_len);
 232         cu->cu_rlen = svcaddr->sa_len;
 233         /* Other values can also be set through clnt_control() */
 234         cu->cu_wait.tv_sec = 3; /* heuristically chosen */
 235         cu->cu_wait.tv_usec = 0;
 236         cu->cu_total.tv_sec = -1;
 237         cu->cu_total.tv_usec = -1;
 238         cu->cu_sendsz = sendsz;
 239         cu->cu_recvsz = recvsz;
 240         cu->cu_async = FALSE;
 241         cu->cu_connect = FALSE;
 242         cu->cu_connected = FALSE;
 243         cu->cu_waitchan = "rpcrecv";
 244         cu->cu_waitflag = 0;
 245         cu->cu_cwnd = MAXCWND / 2;
 246         cu->cu_sent = 0;
 247         cu->cu_cwnd_wait = FALSE;
 248         (void) getmicrotime(&now);
 249         /* Clip at 28bits so that it will not wrap around. */
 250         newxid = __RPC_GETXID(&now) & 0xfffffff;
 251         atomic_cmpset_32(&rpc_xid, 0, newxid);
 252         call_msg.rm_xid = atomic_fetchadd_32(&rpc_xid, 1);
 253         call_msg.rm_call.cb_prog = program;
 254         call_msg.rm_call.cb_vers = version;
 255         xdrmem_create(&xdrs, cu->cu_mcallc, MCALL_MSG_SIZE, XDR_ENCODE);
 256         if (! xdr_callhdr(&xdrs, &call_msg)) {
 257                 rpc_createerr.cf_stat = RPC_CANTENCODEARGS;  /* XXX */
 258                 rpc_createerr.cf_error.re_errno = 0;
 259                 goto err2;
 260         }
 261         cu->cu_mcalllen = XDR_GETPOS(&xdrs);
 262
 263         /*
 264          * By default, closeit is always FALSE. It is users responsibility
 265          * to do a close on it, else the user may use clnt_control
 266          * to let clnt_destroy do it for him/her.
 267          */
 268         cu->cu_closeit = FALSE;
 269         cu->cu_socket = so;
 270         error = soreserve(so, (u_long)sendsz, (u_long)recvsz);
 271         if (error != 0) {
 272                 rpc_createerr.cf_stat = RPC_FAILED;
 273                 rpc_createerr.cf_error.re_errno = error;
 274                 goto err2;
 275         }
 276
 277         sb = &so->so_rcv;
 278         SOCKBUF_LOCK(&so->so_rcv);
 279 recheck_socket:
 280         if (sb->sb_upcall) {
 281                 if (sb->sb_upcall != clnt_dg_soupcall) {
 282                         SOCKBUF_UNLOCK(&so->so_rcv);
 283                         printf("clnt_dg_create(): socket already has an incompatible upcall\n");
 284                         goto err2;
 285                 }
 286                 cs = (struct cu_socket *) sb->sb_upcallarg;
 287                 mtx_lock(&cs->cs_lock);
 288                 cs->cs_refs++;
 289                 mtx_unlock(&cs->cs_lock);
 290         } else {
 291                 /*
 292                  * We are the first on this socket - allocate the
 293                  * structure and install it in the socket.
 294                  */
 295                 SOCKBUF_UNLOCK(&so->so_rcv);
 296                 cs = mem_alloc(sizeof(*cs));
 297                 SOCKBUF_LOCK(&so->so_rcv);
 298                 if (sb->sb_upcall) {
 299                         /*
 300                          * We have lost a race with some other client.
 301                          */
 302                         mem_free(cs, sizeof(*cs));
 303                         goto recheck_socket;
 304                 }
 305                 mtx_init(&cs->cs_lock, "cs->cs_lock", NULL, MTX_DEF);
 306                 cs->cs_refs = 1;
 307                 cs->cs_upcallrefs = 0;
 308                 TAILQ_INIT(&cs->cs_pending);
 309                 soupcall_set(so, SO_RCV, clnt_dg_soupcall, cs);
 310         }
 311         SOCKBUF_UNLOCK(&so->so_rcv);
 312
 313         cl->cl_refs = 1;
 314         cl->cl_ops = &clnt_dg_ops;
 315         cl->cl_private = (caddr_t)(void *)cu;
 316         cl->cl_auth = authnone_create();
 317         cl->cl_tp = NULL;
 318         cl->cl_netid = NULL;
 319         return (cl);
 320 err2:
 321         mem_free(cl, sizeof (CLIENT));
 322         mem_free(cu, sizeof (*cu));
 323
 324         return (NULL);
 325 }
 326
 327 static enum clnt_stat
 328 clnt_dg_call(
 329         CLIENT          *cl,            /* client handle */
 330         struct rpc_callextra *ext,      /* call metadata */
 331         rpcproc_t       proc,           /* procedure number */
 332         struct mbuf     *args,          /* pointer to args */
 333         struct mbuf     **resultsp,     /* pointer to results */
 334         struct timeval  utimeout)       /* seconds to wait before giving up */
 335 {
 336         struct cu_data *cu = (struct cu_data *)cl->cl_private;
 337         struct cu_socket *cs;
 338         struct rpc_timers *rt;
 339         AUTH *auth;
 340         struct rpc_err *errp;
 341         enum clnt_stat stat;
 342         XDR xdrs;
 343         struct rpc_msg reply_msg;
 344         bool_t ok;
 345         int retrans;                    /* number of re-transmits so far */
 346         int nrefreshes = 2;             /* number of times to refresh cred */
 347         struct timeval *tvp;
 348         int timeout;
 349         int retransmit_time;
 350         int next_sendtime, starttime, rtt, time_waited, tv = 0;
 351         struct sockaddr *sa;
 352         uint32_t xid = 0;
 353         struct mbuf *mreq = NULL, *results;
 354         struct cu_request *cr;
 355         int error;
 356
 357         cs = cu->cu_socket->so_rcv.sb_upcallarg;
 358         cr = malloc(sizeof(struct cu_request), M_RPC, M_WAITOK);
 359
 360         mtx_lock(&cs->cs_lock);
 361
 362         if (cu->cu_closing || cu->cu_closed) {
 363                 mtx_unlock(&cs->cs_lock);
 364                 free(cr, M_RPC);
 365                 return (RPC_CANTSEND);
 366         }
 367         cu->cu_threads++;
 368
 369         if (ext) {
 370                 auth = ext->rc_auth;
 371                 errp = &ext->rc_err;
 372         } else {
 373                 auth = cl->cl_auth;
 374                 errp = &cu->cu_error;
 375         }
 376
 377         cr->cr_client = cl;
 378         cr->cr_mrep = NULL;
 379         cr->cr_error = 0;
 380
 381         if (cu->cu_total.tv_usec == -1) {
 382                 tvp = &utimeout; /* use supplied timeout */
 383         } else {
 384                 tvp = &cu->cu_total; /* use default timeout */
 385         }
 386         if (tvp->tv_sec || tvp->tv_usec)
 387                 timeout = tvtohz(tvp);
 388         else
 389                 timeout = 0;
 390
 391         if (cu->cu_connect && !cu->cu_connected) {
 392                 mtx_unlock(&cs->cs_lock);
 393                 error = soconnect(cu->cu_socket,
 394                     (struct sockaddr *)&cu->cu_raddr, curthread);
 395                 mtx_lock(&cs->cs_lock);
 396                 if (error) {
 397                         errp->re_errno = error;
 398                         errp->re_status = stat = RPC_CANTSEND;
 399                         goto out;
 400                 }
 401                 cu->cu_connected = 1;
 402         }
 403         if (cu->cu_connected)
 404                 sa = NULL;
 405         else
 406                 sa = (struct sockaddr *)&cu->cu_raddr;
 407         time_waited = 0;
 408         retrans = 0;
 409         if (ext && ext->rc_timers) {
 410                 rt = ext->rc_timers;
 411                 if (!rt->rt_rtxcur)
 412                         rt->rt_rtxcur = tvtohz(&cu->cu_wait);
 413                 retransmit_time = next_sendtime = rt->rt_rtxcur;
 414         } else {
 415                 rt = NULL;
 416                 retransmit_time = next_sendtime = tvtohz(&cu->cu_wait);
 417         }
 418
 419         starttime = ticks;
 420
 421 call_again:
 422         mtx_assert(&cs->cs_lock, MA_OWNED);
 423
 424         xid = atomic_fetchadd_32(&rpc_xid, 1);
 425
 426 send_again:
 427         mtx_unlock(&cs->cs_lock);
 428
 429         mreq = m_gethdr(M_WAITOK, MT_DATA);
 430         KASSERT(cu->cu_mcalllen <= MHLEN, ("RPC header too big"));
 431         bcopy(cu->cu_mcallc, mreq->m_data, cu->cu_mcalllen);
 432         mreq->m_len = cu->cu_mcalllen;
 433
 434         /*
 435          * The XID is the first thing in the request.
 436          */
 437         *mtod(mreq, uint32_t *) = htonl(xid);
 438
 439         xdrmbuf_create(&xdrs, mreq, XDR_ENCODE);
 440
 441         if (cu->cu_async == TRUE && args == NULL)
 442                 goto get_reply;
 443
 444         if ((! XDR_PUTINT32(&xdrs, &proc)) ||
 445             (! AUTH_MARSHALL(auth, xid, &xdrs,
 446                 m_copym(args, 0, M_COPYALL, M_WAITOK)))) {
 447                 errp->re_status = stat = RPC_CANTENCODEARGS;
 448                 mtx_lock(&cs->cs_lock);
 449                 goto out;
 450         }
 451         mreq->m_pkthdr.len = m_length(mreq, NULL);
 452
 453         cr->cr_xid = xid;
 454         mtx_lock(&cs->cs_lock);
 455
 456         /*
 457          * Try to get a place in the congestion window.
 458          */
 459         while (cu->cu_sent >= cu->cu_cwnd) {
 460                 cu->cu_cwnd_wait = TRUE;
 461                 error = msleep(&cu->cu_cwnd_wait, &cs->cs_lock,
 462                     cu->cu_waitflag, "rpccwnd", 0);
 463                 if (error) {
 464                         errp->re_errno = error;
 465                         if (error == EINTR || error == ERESTART)
 466                                 errp->re_status = stat = RPC_INTR;
 467                         else
 468                                 errp->re_status = stat = RPC_CANTSEND;
 469                         goto out;
 470                 }
 471         }
 472         cu->cu_sent += CWNDSCALE;
 473
 474         TAILQ_INSERT_TAIL(&cs->cs_pending, cr, cr_link);
 475         mtx_unlock(&cs->cs_lock);
 476
 477         /*
 478          * sosend consumes mreq.
 479          */
 480         error = sosend(cu->cu_socket, sa, NULL, mreq, NULL, 0, curthread);
 481         mreq = NULL;
 482
 483         /*
 484          * sub-optimal code appears here because we have
 485          * some clock time to spare while the packets are in flight.
 486          * (We assume that this is actually only executed once.)
 487          */
 488         reply_msg.acpted_rply.ar_verf.oa_flavor = AUTH_NULL;
 489         reply_msg.acpted_rply.ar_verf.oa_base = cr->cr_verf;
 490         reply_msg.acpted_rply.ar_verf.oa_length = 0;
 491         reply_msg.acpted_rply.ar_results.where = NULL;
 492         reply_msg.acpted_rply.ar_results.proc = (xdrproc_t)xdr_void;
 493
 494         mtx_lock(&cs->cs_lock);
 495         if (error) {
 496                 TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
 497                 errp->re_errno = error;
 498                 errp->re_status = stat = RPC_CANTSEND;
 499                 cu->cu_sent -= CWNDSCALE;
 500                 if (cu->cu_cwnd_wait) {
 501                         cu->cu_cwnd_wait = FALSE;
 502                         wakeup(&cu->cu_cwnd_wait);
 503                 }
 504                 goto out;
 505         }
 506
 507         /*
 508          * Check to see if we got an upcall while waiting for the
 509          * lock.
 510          */
 511         if (cr->cr_error) {
 512                 TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
 513                 errp->re_errno = cr->cr_error;
 514                 errp->re_status = stat = RPC_CANTRECV;
 515                 cu->cu_sent -= CWNDSCALE;
 516                 if (cu->cu_cwnd_wait) {
 517                         cu->cu_cwnd_wait = FALSE;
 518                         wakeup(&cu->cu_cwnd_wait);
 519                 }
 520                 goto out;
 521         }
 522         if (cr->cr_mrep) {
 523                 TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
 524                 cu->cu_sent -= CWNDSCALE;
 525                 if (cu->cu_cwnd_wait) {
 526                         cu->cu_cwnd_wait = FALSE;
 527                         wakeup(&cu->cu_cwnd_wait);
 528                 }
 529                 goto got_reply;
 530         }
 531
 532         /*
 533          * Hack to provide rpc-based message passing
 534          */
 535         if (timeout == 0) {
 536                 TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
 537                 errp->re_status = stat = RPC_TIMEDOUT;
 538                 cu->cu_sent -= CWNDSCALE;
 539                 if (cu->cu_cwnd_wait) {
 540                         cu->cu_cwnd_wait = FALSE;
 541                         wakeup(&cu->cu_cwnd_wait);
 542                 }
 543                 goto out;
 544         }
 545
 546 get_reply:
 547         for (;;) {
 548                 /* Decide how long to wait. */
 549                 if (next_sendtime < timeout)
 550                         tv = next_sendtime;
 551                 else
 552                         tv = timeout;
 553                 tv -= time_waited;
 554
 555                 if (tv > 0) {
 556                         if (cu->cu_closing || cu->cu_closed) {
 557                                 error = 0;
 558                                 cr->cr_error = ESHUTDOWN;
 559                         } else {
 560                                 error = msleep(cr, &cs->cs_lock,
 561                                     cu->cu_waitflag, cu->cu_waitchan, tv);
 562                         }
 563                 } else {
 564                         error = EWOULDBLOCK;
 565                 }
 566
 567                 TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
 568                 cu->cu_sent -= CWNDSCALE;
 569                 if (cu->cu_cwnd_wait) {
 570                         cu->cu_cwnd_wait = FALSE;
 571                         wakeup(&cu->cu_cwnd_wait);
 572                 }
 573
 574                 if (!error) {
 575                         /*
 576                          * We were woken up by the upcall.  If the
 577                          * upcall had a receive error, report that,
 578                          * otherwise we have a reply.
 579                          */
 580                         if (cr->cr_error) {
 581                                 errp->re_errno = cr->cr_error;
 582                                 errp->re_status = stat = RPC_CANTRECV;
 583                                 goto out;
 584                         }
 585
 586                         cu->cu_cwnd += (CWNDSCALE * CWNDSCALE
 587                             + cu->cu_cwnd / 2) / cu->cu_cwnd;
 588                         if (cu->cu_cwnd > MAXCWND)
 589                                 cu->cu_cwnd = MAXCWND;
 590
 591                         if (rt) {
 592                                 /*
 593                                  * Add one to the time since a tick
 594                                  * count of N means that the actual
 595                                  * time taken was somewhere between N
 596                                  * and N+1.
 597                                  */
 598                                 rtt = ticks - starttime + 1;
 599
 600                                 /*
 601                                  * Update our estimate of the round
 602                                  * trip time using roughly the
 603                                  * algorithm described in RFC
 604                                  * 2988. Given an RTT sample R:
 605                                  *
 606                                  * RTTVAR = (1-beta) * RTTVAR + beta * |SRTT-R|
 607                                  * SRTT = (1-alpha) * SRTT + alpha * R
 608                                  *
 609                                  * where alpha = 0.125 and beta = 0.25.
 610                                  *
 611                                  * The initial retransmit timeout is
 612                                  * SRTT + 4*RTTVAR and doubles on each
 613                                  * retransmision.
 614                                  */
 615                                 if (rt->rt_srtt == 0) {
 616                                         rt->rt_srtt = rtt;
 617                                         rt->rt_deviate = rtt / 2;
 618                                 } else {
 619                                         int32_t error = rtt - rt->rt_srtt;
 620                                         rt->rt_srtt += error / 8;
 621                                         error = abs(error) - rt->rt_deviate;
 622                                         rt->rt_deviate += error / 4;
 623                                 }
 624                                 rt->rt_rtxcur = rt->rt_srtt + 4*rt->rt_deviate;
 625                         }
 626
 627                         break;
 628                 }
 629
 630                 /*
 631                  * The sleep returned an error so our request is still
 632                  * on the list. If we got EWOULDBLOCK, we may want to
 633                  * re-send the request.
 634                  */
 635                 if (error != EWOULDBLOCK) {
 636                         errp->re_errno = error;
 637                         if (error == EINTR || error == ERESTART)
 638                                 errp->re_status = stat = RPC_INTR;
 639                         else
 640                                 errp->re_status = stat = RPC_CANTRECV;
 641                         goto out;
 642                 }
 643
 644                 time_waited = ticks - starttime;
 645
 646                 /* Check for timeout. */
 647                 if (time_waited > timeout) {
 648                         errp->re_errno = EWOULDBLOCK;
 649                         errp->re_status = stat = RPC_TIMEDOUT;
 650                         goto out;
 651                 }
 652
 653                 /* Retransmit if necessary. */
 654                 if (time_waited >= next_sendtime) {
 655                         cu->cu_cwnd /= 2;
 656                         if (cu->cu_cwnd < CWNDSCALE)
 657                                 cu->cu_cwnd = CWNDSCALE;
 658                         if (ext && ext->rc_feedback) {
 659                                 mtx_unlock(&cs->cs_lock);
 660                                 if (retrans == 0)
 661                                         ext->rc_feedback(FEEDBACK_REXMIT1,
 662                                             proc, ext->rc_feedback_arg);
 663                                 else
 664                                         ext->rc_feedback(FEEDBACK_REXMIT2,
 665                                             proc, ext->rc_feedback_arg);
 666                                 mtx_lock(&cs->cs_lock);
 667                         }
 668                         if (cu->cu_closing || cu->cu_closed) {
 669                                 errp->re_errno = ESHUTDOWN;
 670                                 errp->re_status = stat = RPC_CANTRECV;
 671                                 goto out;
 672                         }
 673                         retrans++;
 674                         /* update retransmit_time */
 675                         if (retransmit_time < RPC_MAX_BACKOFF * hz)
 676                                 retransmit_time = 2 * retransmit_time;
 677                         next_sendtime += retransmit_time;
 678                         goto send_again;
 679                 }
 680                 cu->cu_sent += CWNDSCALE;
 681                 TAILQ_INSERT_TAIL(&cs->cs_pending, cr, cr_link);
 682         }
 683
 684 got_reply:
 685         /*
 686          * Now decode and validate the response. We need to drop the
 687          * lock since xdr_replymsg may end up sleeping in malloc.
 688          */
 689         mtx_unlock(&cs->cs_lock);
 690
 691         if (ext && ext->rc_feedback)
 692                 ext->rc_feedback(FEEDBACK_OK, proc, ext->rc_feedback_arg);
 693
 694         xdrmbuf_create(&xdrs, cr->cr_mrep, XDR_DECODE);
 695         ok = xdr_replymsg(&xdrs, &reply_msg);
 696         cr->cr_mrep = NULL;
 697
 698         if (ok) {
 699                 if ((reply_msg.rm_reply.rp_stat == MSG_ACCEPTED) &&
 700                     (reply_msg.acpted_rply.ar_stat == SUCCESS))
 701                         errp->re_status = stat = RPC_SUCCESS;
 702                 else
 703                         stat = _seterr_reply(&reply_msg, &(cu->cu_error));
 704
 705                 if (errp->re_status == RPC_SUCCESS) {
 706                         results = xdrmbuf_getall(&xdrs);
 707                         if (! AUTH_VALIDATE(auth, xid,
 708                                 &reply_msg.acpted_rply.ar_verf,
 709                                 &results)) {
 710                                 errp->re_status = stat = RPC_AUTHERROR;
 711                                 errp->re_why = AUTH_INVALIDRESP;
 712                                 if (retrans &&
 713                                     auth->ah_cred.oa_flavor == RPCSEC_GSS) {
 714                                         /*
 715                                          * If we retransmitted, its
 716                                          * possible that we will
 717                                          * receive a reply for one of
 718                                          * the earlier transmissions
 719                                          * (which will use an older
 720                                          * RPCSEC_GSS sequence
 721                                          * number). In this case, just
 722                                          * go back and listen for a
 723                                          * new reply. We could keep a
 724                                          * record of all the seq
 725                                          * numbers we have transmitted
 726                                          * so far so that we could
 727                                          * accept a reply for any of
 728                                          * them here.
 729                                          */
 730                                         XDR_DESTROY(&xdrs);
 731                                         mtx_lock(&cs->cs_lock);
 732                                         cu->cu_sent += CWNDSCALE;
 733                                         TAILQ_INSERT_TAIL(&cs->cs_pending,
 734                                             cr, cr_link);
 735                                         cr->cr_mrep = NULL;
 736                                         goto get_reply;
 737                                 }
 738                         } else {
 739                                 *resultsp = results;
 740                         }
 741                 }               /* end successful completion */
 742                 /*
 743                  * If unsuccessful AND error is an authentication error
 744                  * then refresh credentials and try again, else break
 745                  */
 746                 else if (stat == RPC_AUTHERROR)
 747                         /* maybe our credentials need to be refreshed ... */
 748                         if (nrefreshes > 0 &&
 749                             AUTH_REFRESH(auth, &reply_msg)) {
 750                                 nrefreshes--;
 751                                 XDR_DESTROY(&xdrs);
 752                                 mtx_lock(&cs->cs_lock);
 753                                 goto call_again;
 754                         }
 755                 /* end of unsuccessful completion */
 756         }       /* end of valid reply message */
 757         else {
 758                 errp->re_status = stat = RPC_CANTDECODERES;
 759
 760         }
 761         XDR_DESTROY(&xdrs);
 762         mtx_lock(&cs->cs_lock);
 763 out:
 764         mtx_assert(&cs->cs_lock, MA_OWNED);
 765
 766         if (mreq)
 767                 m_freem(mreq);
 768         if (cr->cr_mrep)
 769                 m_freem(cr->cr_mrep);
 770
 771         cu->cu_threads--;
 772         if (cu->cu_closing)
 773                 wakeup(cu);
 774
 775         mtx_unlock(&cs->cs_lock);
 776
 777         if (auth && stat != RPC_SUCCESS)
 778                 AUTH_VALIDATE(auth, xid, NULL, NULL);
 779
 780         free(cr, M_RPC);
 781
 782         return (stat);
 783 }
 784
 785 static void
 786 clnt_dg_geterr(CLIENT *cl, struct rpc_err *errp)
 787 {
 788         struct cu_data *cu = (struct cu_data *)cl->cl_private;
 789
 790         *errp = cu->cu_error;
 791 }
 792
 793 static bool_t
 794 clnt_dg_freeres(CLIENT *cl, xdrproc_t xdr_res, void *res_ptr)
 795 {
 796         XDR xdrs;
 797         bool_t dummy;
 798
 799         xdrs.x_op = XDR_FREE;
 800         dummy = (*xdr_res)(&xdrs, res_ptr);
 801
 802         return (dummy);
 803 }
 804
 805 /*ARGSUSED*/
 806 static void
 807 clnt_dg_abort(CLIENT *h)
 808 {
 809 }
 810
 811 static bool_t
 812 clnt_dg_control(CLIENT *cl, u_int request, void *info)
 813 {
 814         struct cu_data *cu = (struct cu_data *)cl->cl_private;
 815         struct cu_socket *cs;
 816         struct sockaddr *addr;
 817
 818         cs = cu->cu_socket->so_rcv.sb_upcallarg;
 819         mtx_lock(&cs->cs_lock);
 820
 821         switch (request) {
 822         case CLSET_FD_CLOSE:
 823                 cu->cu_closeit = TRUE;
 824                 mtx_unlock(&cs->cs_lock);
 825                 return (TRUE);
 826         case CLSET_FD_NCLOSE:
 827                 cu->cu_closeit = FALSE;
 828                 mtx_unlock(&cs->cs_lock);
 829                 return (TRUE);
 830         }
 831
 832         /* for other requests which use info */
 833         if (info == NULL) {
 834                 mtx_unlock(&cs->cs_lock);
 835                 return (FALSE);
 836         }
 837         switch (request) {
 838         case CLSET_TIMEOUT:
 839                 if (time_not_ok((struct timeval *)info)) {
 840                         mtx_unlock(&cs->cs_lock);
 841                         return (FALSE);
 842                 }
 843                 cu->cu_total = *(struct timeval *)info;
 844                 break;
 845         case CLGET_TIMEOUT:
 846                 *(struct timeval *)info = cu->cu_total;
 847                 break;
 848         case CLSET_RETRY_TIMEOUT:
 849                 if (time_not_ok((struct timeval *)info)) {
 850                         mtx_unlock(&cs->cs_lock);
 851                         return (FALSE);
 852                 }
 853                 cu->cu_wait = *(struct timeval *)info;
 854                 break;
 855         case CLGET_RETRY_TIMEOUT:
 856                 *(struct timeval *)info = cu->cu_wait;
 857                 break;
 858         case CLGET_SVC_ADDR:
 859                 /*
 860                  * Slightly different semantics to userland - we use
 861                  * sockaddr instead of netbuf.
 862                  */
 863                 memcpy(info, &cu->cu_raddr, cu->cu_raddr.ss_len);
 864                 break;
 865         case CLSET_SVC_ADDR:            /* set to new address */
 866                 addr = (struct sockaddr *)info;
 867                 (void) memcpy(&cu->cu_raddr, addr, addr->sa_len);
 868                 break;
 869         case CLGET_XID:
 870                 *(uint32_t *)info = atomic_load_32(&rpc_xid);
 871                 break;
 872
 873         case CLSET_XID:
 874                 /* This will set the xid of the NEXT call */
 875                 /* decrement by 1 as clnt_dg_call() increments once */
 876                 atomic_store_32(&rpc_xid, *(uint32_t *)info - 1);
 877                 break;
 878
 879         case CLGET_VERS:
 880                 /*
 881                  * This RELIES on the information that, in the call body,
 882                  * the version number field is the fifth field from the
 883                  * beginning of the RPC header. MUST be changed if the
 884                  * call_struct is changed
 885                  */
 886                 *(uint32_t *)info =
 887                     ntohl(*(uint32_t *)(void *)(cu->cu_mcallc +
 888                     4 * BYTES_PER_XDR_UNIT));
 889                 break;
 890
 891         case CLSET_VERS:
 892                 *(uint32_t *)(void *)(cu->cu_mcallc + 4 * BYTES_PER_XDR_UNIT)
 893                         = htonl(*(uint32_t *)info);
 894                 break;
 895
 896         case CLGET_PROG:
 897                 /*
 898                  * This RELIES on the information that, in the call body,
 899                  * the program number field is the fourth field from the
 900                  * beginning of the RPC header. MUST be changed if the
 901                  * call_struct is changed
 902                  */
 903                 *(uint32_t *)info =
 904                     ntohl(*(uint32_t *)(void *)(cu->cu_mcallc +
 905                     3 * BYTES_PER_XDR_UNIT));
 906                 break;
 907
 908         case CLSET_PROG:
 909                 *(uint32_t *)(void *)(cu->cu_mcallc + 3 * BYTES_PER_XDR_UNIT)
 910                         = htonl(*(uint32_t *)info);
 911                 break;
 912         case CLSET_ASYNC:
 913                 cu->cu_async = *(int *)info;
 914                 break;
 915         case CLSET_CONNECT:
 916                 cu->cu_connect = *(int *)info;
 917                 break;
 918         case CLSET_WAITCHAN:
 919                 cu->cu_waitchan = (const char *)info;
 920                 break;
 921         case CLGET_WAITCHAN:
 922                 *(const char **) info = cu->cu_waitchan;
 923                 break;
 924         case CLSET_INTERRUPTIBLE:
 925                 if (*(int *) info)
 926                         cu->cu_waitflag = PCATCH;
 927                 else
 928                         cu->cu_waitflag = 0;
 929                 break;
 930         case CLGET_INTERRUPTIBLE:
 931                 if (cu->cu_waitflag)
 932                         *(int *) info = TRUE;
 933                 else
 934                         *(int *) info = FALSE;
 935                 break;
 936         default:
 937                 mtx_unlock(&cs->cs_lock);
 938                 return (FALSE);
 939         }
 940         mtx_unlock(&cs->cs_lock);
 941         return (TRUE);
 942 }
 943
 944 static void
 945 clnt_dg_close(CLIENT *cl)
 946 {
 947         struct cu_data *cu = (struct cu_data *)cl->cl_private;
 948         struct cu_socket *cs;
 949         struct cu_request *cr;
 950
 951         cs = cu->cu_socket->so_rcv.sb_upcallarg;
 952         mtx_lock(&cs->cs_lock);
 953
 954         if (cu->cu_closed) {
 955                 mtx_unlock(&cs->cs_lock);
 956                 return;
 957         }
 958
 959         if (cu->cu_closing) {
 960                 while (cu->cu_closing)
 961                         msleep(cu, &cs->cs_lock, 0, "rpcclose", 0);
 962                 KASSERT(cu->cu_closed, ("client should be closed"));
 963                 mtx_unlock(&cs->cs_lock);
 964                 return;
 965         }
 966
 967         /*
 968          * Abort any pending requests and wait until everyone
 969          * has finished with clnt_vc_call.
 970          */
 971         cu->cu_closing = TRUE;
 972         TAILQ_FOREACH(cr, &cs->cs_pending, cr_link) {
 973                 if (cr->cr_client == cl) {
 974                         cr->cr_xid = 0;
 975                         cr->cr_error = ESHUTDOWN;
 976                         wakeup(cr);
 977                 }
 978         }
 979
 980         while (cu->cu_threads)
 981                 msleep(cu, &cs->cs_lock, 0, "rpcclose", 0);
 982
 983         cu->cu_closing = FALSE;
 984         cu->cu_closed = TRUE;
 985
 986         mtx_unlock(&cs->cs_lock);
 987         wakeup(cu);
 988 }
 989
 990 static void
 991 clnt_dg_destroy(CLIENT *cl)
 992 {
 993         struct cu_data *cu = (struct cu_data *)cl->cl_private;
 994         struct cu_socket *cs;
 995         struct socket *so = NULL;
 996         bool_t lastsocketref;
 997
 998         cs = cu->cu_socket->so_rcv.sb_upcallarg;
 999         clnt_dg_close(cl);
1000
1001         SOCKBUF_LOCK(&cu->cu_socket->so_rcv);
1002         mtx_lock(&cs->cs_lock);
1003
1004         cs->cs_refs--;
1005         if (cs->cs_refs == 0) {
1006                 mtx_unlock(&cs->cs_lock);
1007                 soupcall_clear(cu->cu_socket, SO_RCV);
1008                 clnt_dg_upcallsdone(cu->cu_socket, cs);
1009                 SOCKBUF_UNLOCK(&cu->cu_socket->so_rcv);
1010                 mtx_destroy(&cs->cs_lock);
1011                 mem_free(cs, sizeof(*cs));
1012                 lastsocketref = TRUE;
1013         } else {
1014                 mtx_unlock(&cs->cs_lock);
1015                 SOCKBUF_UNLOCK(&cu->cu_socket->so_rcv);
1016                 lastsocketref = FALSE;
1017         }
1018
1019         if (cu->cu_closeit && lastsocketref) {
1020                 so = cu->cu_socket;
1021                 cu->cu_socket = NULL;
1022         }
1023
1024         if (so)
1025                 soclose(so);
1026
1027         if (cl->cl_netid && cl->cl_netid[0])
1028                 mem_free(cl->cl_netid, strlen(cl->cl_netid) +1);
1029         if (cl->cl_tp && cl->cl_tp[0])
1030                 mem_free(cl->cl_tp, strlen(cl->cl_tp) +1);
1031         mem_free(cu, sizeof (*cu));
1032         mem_free(cl, sizeof (CLIENT));
1033 }
1034
1035 /*
1036  * Make sure that the time is not garbage.  -1 value is allowed.
1037  */
1038 static bool_t
1039 time_not_ok(struct timeval *t)
1040 {
1041         return (t->tv_sec < -1 || t->tv_sec > 100000000 ||
1042                 t->tv_usec < -1 || t->tv_usec > 1000000);
1043 }
1044
1045 int
1046 clnt_dg_soupcall(struct socket *so, void *arg, int waitflag)
1047 {
1048         struct cu_socket *cs = (struct cu_socket *) arg;
1049         struct uio uio;
1050         struct mbuf *m;
1051         struct mbuf *control;
1052         struct cu_request *cr;
1053         int error, rcvflag, foundreq;
1054         uint32_t xid;
1055
1056         cs->cs_upcallrefs++;
1057         uio.uio_resid = 1000000000;
1058         uio.uio_td = curthread;
1059         do {
1060                 SOCKBUF_UNLOCK(&so->so_rcv);
1061                 m = NULL;
1062                 control = NULL;
1063                 rcvflag = MSG_DONTWAIT;
1064                 error = soreceive(so, NULL, &uio, &m, &control, &rcvflag);
1065                 if (control)
1066                         m_freem(control);
1067                 SOCKBUF_LOCK(&so->so_rcv);
1068
1069                 if (error == EWOULDBLOCK)
1070                         break;
1071
1072                 /*
1073                  * If there was an error, wake up all pending
1074                  * requests.
1075                  */
1076                 if (error) {
1077                         mtx_lock(&cs->cs_lock);
1078                         TAILQ_FOREACH(cr, &cs->cs_pending, cr_link) {
1079                                 cr->cr_xid = 0;
1080                                 cr->cr_error = error;
1081                                 wakeup(cr);
1082                         }
1083                         mtx_unlock(&cs->cs_lock);
1084                         break;
1085                 }
1086
1087                 /*
1088                  * The XID is in the first uint32_t of the reply.
1089                  */
1090                 if (m->m_len < sizeof(xid) && m_length(m, NULL) < sizeof(xid)) {
1091                         /*
1092                          * Should never happen.
1093                          */
1094                         m_freem(m);
1095                         continue;
1096                 }
1097
1098                 m_copydata(m, 0, sizeof(xid), (char *)&xid);
1099                 xid = ntohl(xid);
1100
1101                 /*
1102                  * Attempt to match this reply with a pending request.
1103                  */
1104                 mtx_lock(&cs->cs_lock);
1105                 foundreq = 0;
1106                 TAILQ_FOREACH(cr, &cs->cs_pending, cr_link) {
1107                         if (cr->cr_xid == xid) {
1108                                 /*
1109                                  * This one matches. We leave the
1110                                  * reply mbuf in cr->cr_mrep. Set the
1111                                  * XID to zero so that we will ignore
1112                                  * any duplicated replies that arrive
1113                                  * before clnt_dg_call removes it from
1114                                  * the queue.
1115                                  */
1116                                 cr->cr_xid = 0;
1117                                 cr->cr_mrep = m;
1118                                 cr->cr_error = 0;
1119                                 foundreq = 1;
1120                                 wakeup(cr);
1121                                 break;
1122                         }
1123                 }
1124                 mtx_unlock(&cs->cs_lock);
1125
1126                 /*
1127                  * If we didn't find the matching request, just drop
1128                  * it - its probably a repeated reply.
1129                  */
1130                 if (!foundreq)
1131                         m_freem(m);
1132         } while (m);
1133         cs->cs_upcallrefs--;
1134         if (cs->cs_upcallrefs < 0)
1135                 panic("rpcdg upcall refcnt");
1136         if (cs->cs_upcallrefs == 0)
1137                 wakeup(&cs->cs_upcallrefs);
1138         return (SU_OK);
1139 }
1140
1141 /*
1142  * Wait for all upcalls in progress to complete.
1143  */
1144 static void
1145 clnt_dg_upcallsdone(struct socket *so, struct cu_socket *cs)
1146 {
1147
1148         SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1149
1150         while (cs->cs_upcallrefs > 0)
1151                 (void) msleep(&cs->cs_upcallrefs, SOCKBUF_MTX(&so->so_rcv), 0,
1152                     "rpcdgup", 0);
1153 }