sys/rpc/clnt_vc.c

   1 /*      $NetBSD: clnt_vc.c,v 1.4 2000/07/14 08:40:42 fvdl Exp $ */
   2
   3 /*-
   4  * Copyright (c) 2009, Sun Microsystems, Inc.
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions are met:
   9  * - Redistributions of source code must retain the above copyright notice,
  10  *   this list of conditions and the following disclaimer.
  11  * - Redistributions in binary form must reproduce the above copyright notice,
  12  *   this list of conditions and the following disclaimer in the documentation
  13  *   and/or other materials provided with the distribution.
  14  * - Neither the name of Sun Microsystems, Inc. nor the names of its
  15  *   contributors may be used to endorse or promote products derived
  16  *   from this software without specific prior written permission.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  19  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  21  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
  22  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  23  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  24  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  25  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  26  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  27  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  28  * POSSIBILITY OF SUCH DAMAGE.
  29  */
  30
  31 #if defined(LIBC_SCCS) && !defined(lint)
  32 static char *sccsid2 = "@(#)clnt_tcp.c 1.37 87/10/05 Copyr 1984 Sun Micro";
  33 static char *sccsid = "@(#)clnt_tcp.c   2.2 88/08/01 4.0 RPCSRC";
  34 static char sccsid3[] = "@(#)clnt_vc.c 1.19 89/03/16 Copyr 1988 Sun Micro";
  35 #endif
  36 #include <sys/cdefs.h>
  37 __FBSDID("$FreeBSD$");
  38
  39 /*
  40  * clnt_tcp.c, Implements a TCP/IP based, client side RPC.
  41  *
  42  * Copyright (C) 1984, Sun Microsystems, Inc.
  43  *
  44  * TCP based RPC supports 'batched calls'.
  45  * A sequence of calls may be batched-up in a send buffer.  The rpc call
  46  * return immediately to the client even though the call was not necessarily
  47  * sent.  The batching occurs if the results' xdr routine is NULL (0) AND
  48  * the rpc timeout value is zero (see clnt.h, rpc).
  49  *
  50  * Clients should NOT casually batch calls that in fact return results; that is,
  51  * the server side should be aware that a call is batched and not produce any
  52  * return message.  Batched calls that produce many result messages can
  53  * deadlock (netlock) the client and the server....
  54  *
  55  * Now go hang yourself.
  56  */
  57
  58 #include <sys/param.h>
  59 #include <sys/systm.h>
  60 #include <sys/lock.h>
  61 #include <sys/malloc.h>
  62 #include <sys/mbuf.h>
  63 #include <sys/mutex.h>
  64 #include <sys/pcpu.h>
  65 #include <sys/proc.h>
  66 #include <sys/protosw.h>
  67 #include <sys/socket.h>
  68 #include <sys/socketvar.h>
  69 #include <sys/syslog.h>
  70 #include <sys/time.h>
  71 #include <sys/uio.h>
  72
  73 #include <net/vnet.h>
  74
  75 #include <netinet/tcp.h>
  76
  77 #include <rpc/rpc.h>
  78 #include <rpc/rpc_com.h>
  79
  80 #define MCALL_MSG_SIZE 24
  81
  82 struct cmessage {
  83         struct cmsghdr cmsg;
  84         struct cmsgcred cmcred;
  85 };
  86
  87 static enum clnt_stat clnt_vc_call(CLIENT *, struct rpc_callextra *,
  88     rpcproc_t, struct mbuf *, struct mbuf **, struct timeval);
  89 static void clnt_vc_geterr(CLIENT *, struct rpc_err *);
  90 static bool_t clnt_vc_freeres(CLIENT *, xdrproc_t, void *);
  91 static void clnt_vc_abort(CLIENT *);
  92 static bool_t clnt_vc_control(CLIENT *, u_int, void *);
  93 static void clnt_vc_close(CLIENT *);
  94 static void clnt_vc_destroy(CLIENT *);
  95 static bool_t time_not_ok(struct timeval *);
  96 static int clnt_vc_soupcall(struct socket *so, void *arg, int waitflag);
  97
  98 static struct clnt_ops clnt_vc_ops = {
  99         .cl_call =      clnt_vc_call,
 100         .cl_abort =     clnt_vc_abort,
 101         .cl_geterr =    clnt_vc_geterr,
 102         .cl_freeres =   clnt_vc_freeres,
 103         .cl_close =     clnt_vc_close,
 104         .cl_destroy =   clnt_vc_destroy,
 105         .cl_control =   clnt_vc_control
 106 };
 107
 108 /*
 109  * A pending RPC request which awaits a reply. Requests which have
 110  * received their reply will have cr_xid set to zero and cr_mrep to
 111  * the mbuf chain of the reply.
 112  */
 113 struct ct_request {
 114         TAILQ_ENTRY(ct_request) cr_link;
 115         uint32_t                cr_xid;         /* XID of request */
 116         struct mbuf             *cr_mrep;       /* reply received by upcall */
 117         int                     cr_error;       /* any error from upcall */
 118         char                    cr_verf[MAX_AUTH_BYTES]; /* reply verf */
 119 };
 120
 121 TAILQ_HEAD(ct_request_list, ct_request);
 122
 123 struct ct_data {
 124         struct mtx      ct_lock;
 125         int             ct_threads;     /* number of threads in clnt_vc_call */
 126         bool_t          ct_closing;     /* TRUE if we are closing */
 127         bool_t          ct_closed;      /* TRUE if we are closed */
 128         struct socket   *ct_socket;     /* connection socket */
 129         bool_t          ct_closeit;     /* close it on destroy */
 130         struct timeval  ct_wait;        /* wait interval in milliseconds */
 131         struct sockaddr_storage ct_addr; /* remote addr */
 132         struct rpc_err  ct_error;
 133         uint32_t        ct_xid;
 134         char            ct_mcallc[MCALL_MSG_SIZE]; /* marshalled callmsg */
 135         size_t          ct_mpos;        /* pos after marshal */
 136         const char      *ct_waitchan;
 137         int             ct_waitflag;
 138         struct mbuf     *ct_record;     /* current reply record */
 139         size_t          ct_record_resid; /* how much left of reply to read */
 140         bool_t          ct_record_eor;   /* true if reading last fragment */
 141         struct ct_request_list ct_pending;
 142         int             ct_upcallrefs;  /* Ref cnt of upcalls in prog. */
 143 };
 144
 145 static void clnt_vc_upcallsdone(struct ct_data *);
 146
 147 /*
 148  * Create a client handle for a connection.
 149  * Default options are set, which the user can change using clnt_control()'s.
 150  * The rpc/vc package does buffering similar to stdio, so the client
 151  * must pick send and receive buffer sizes, 0 => use the default.
 152  * NB: fd is copied into a private area.
 153  * NB: The rpch->cl_auth is set null authentication. Caller may wish to
 154  * set this something more useful.
 155  *
 156  * fd should be an open socket
 157  */
 158 CLIENT *
 159 clnt_vc_create(
 160         struct socket *so,              /* open file descriptor */
 161         struct sockaddr *raddr,         /* servers address */
 162         const rpcprog_t prog,           /* program number */
 163         const rpcvers_t vers,           /* version number */
 164         size_t sendsz,                  /* buffer recv size */
 165         size_t recvsz,                  /* buffer send size */
 166         int intrflag)                   /* interruptible */
 167 {
 168         CLIENT *cl;                     /* client handle */
 169         struct ct_data *ct = NULL;      /* client handle */
 170         struct timeval now;
 171         struct rpc_msg call_msg;
 172         static uint32_t disrupt;
 173         struct __rpc_sockinfo si;
 174         XDR xdrs;
 175         int error, interrupted, one = 1, sleep_flag;
 176         struct sockopt sopt;
 177
 178         if (disrupt == 0)
 179                 disrupt = (uint32_t)(long)raddr;
 180
 181         cl = (CLIENT *)mem_alloc(sizeof (*cl));
 182         ct = (struct ct_data *)mem_alloc(sizeof (*ct));
 183
 184         mtx_init(&ct->ct_lock, "ct->ct_lock", NULL, MTX_DEF);
 185         ct->ct_threads = 0;
 186         ct->ct_closing = FALSE;
 187         ct->ct_closed = FALSE;
 188         ct->ct_upcallrefs = 0;
 189
 190         if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
 191                 error = soconnect(so, raddr, curthread);
 192                 SOCK_LOCK(so);
 193                 interrupted = 0;
 194                 sleep_flag = PSOCK;
 195                 if (intrflag != 0)
 196                         sleep_flag |= (PCATCH | PBDRY);
 197                 while ((so->so_state & SS_ISCONNECTING)
 198                     && so->so_error == 0) {
 199                         error = msleep(&so->so_timeo, SOCK_MTX(so),
 200                             sleep_flag, "connec", 0);
 201                         if (error) {
 202                                 if (error == EINTR || error == ERESTART)
 203                                         interrupted = 1;
 204                                 break;
 205                         }
 206                 }
 207                 if (error == 0) {
 208                         error = so->so_error;
 209                         so->so_error = 0;
 210                 }
 211                 SOCK_UNLOCK(so);
 212                 if (error) {
 213                         if (!interrupted)
 214                                 so->so_state &= ~SS_ISCONNECTING;
 215                         rpc_createerr.cf_stat = RPC_SYSTEMERROR;
 216                         rpc_createerr.cf_error.re_errno = error;
 217                         goto err;
 218                 }
 219         }
 220
 221         if (!__rpc_socket2sockinfo(so, &si)) {
 222                 goto err;
 223         }
 224
 225         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
 226                 bzero(&sopt, sizeof(sopt));
 227                 sopt.sopt_dir = SOPT_SET;
 228                 sopt.sopt_level = SOL_SOCKET;
 229                 sopt.sopt_name = SO_KEEPALIVE;
 230                 sopt.sopt_val = &one;
 231                 sopt.sopt_valsize = sizeof(one);
 232                 sosetopt(so, &sopt);
 233         }
 234
 235         if (so->so_proto->pr_protocol == IPPROTO_TCP) {
 236                 bzero(&sopt, sizeof(sopt));
 237                 sopt.sopt_dir = SOPT_SET;
 238                 sopt.sopt_level = IPPROTO_TCP;
 239                 sopt.sopt_name = TCP_NODELAY;
 240                 sopt.sopt_val = &one;
 241                 sopt.sopt_valsize = sizeof(one);
 242                 sosetopt(so, &sopt);
 243         }
 244
 245         ct->ct_closeit = FALSE;
 246
 247         /*
 248          * Set up private data struct
 249          */
 250         ct->ct_socket = so;
 251         ct->ct_wait.tv_sec = -1;
 252         ct->ct_wait.tv_usec = -1;
 253         memcpy(&ct->ct_addr, raddr, raddr->sa_len);
 254
 255         /*
 256          * Initialize call message
 257          */
 258         getmicrotime(&now);
 259         ct->ct_xid = ((uint32_t)++disrupt) ^ __RPC_GETXID(&now);
 260         call_msg.rm_xid = ct->ct_xid;
 261         call_msg.rm_direction = CALL;
 262         call_msg.rm_call.cb_rpcvers = RPC_MSG_VERSION;
 263         call_msg.rm_call.cb_prog = (uint32_t)prog;
 264         call_msg.rm_call.cb_vers = (uint32_t)vers;
 265
 266         /*
 267          * pre-serialize the static part of the call msg and stash it away
 268          */
 269         xdrmem_create(&xdrs, ct->ct_mcallc, MCALL_MSG_SIZE,
 270             XDR_ENCODE);
 271         if (! xdr_callhdr(&xdrs, &call_msg)) {
 272                 if (ct->ct_closeit) {
 273                         soclose(ct->ct_socket);
 274                 }
 275                 goto err;
 276         }
 277         ct->ct_mpos = XDR_GETPOS(&xdrs);
 278         XDR_DESTROY(&xdrs);
 279         ct->ct_waitchan = "rpcrecv";
 280         ct->ct_waitflag = 0;
 281
 282         /*
 283          * Create a client handle which uses xdrrec for serialization
 284          * and authnone for authentication.
 285          */
 286         sendsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)sendsz);
 287         recvsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)recvsz);
 288         error = soreserve(ct->ct_socket, sendsz, recvsz);
 289         if (error != 0) {
 290                 if (ct->ct_closeit) {
 291                         soclose(ct->ct_socket);
 292                 }
 293                 goto err;
 294         }
 295         cl->cl_refs = 1;
 296         cl->cl_ops = &clnt_vc_ops;
 297         cl->cl_private = ct;
 298         cl->cl_auth = authnone_create();
 299
 300         SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
 301         soupcall_set(ct->ct_socket, SO_RCV, clnt_vc_soupcall, ct);
 302         SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
 303
 304         ct->ct_record = NULL;
 305         ct->ct_record_resid = 0;
 306         TAILQ_INIT(&ct->ct_pending);
 307         return (cl);
 308
 309 err:
 310         if (cl) {
 311                 if (ct) {
 312                         mtx_destroy(&ct->ct_lock);
 313                         mem_free(ct, sizeof (struct ct_data));
 314                 }
 315                 if (cl)
 316                         mem_free(cl, sizeof (CLIENT));
 317         }
 318         return ((CLIENT *)NULL);
 319 }
 320
 321 static enum clnt_stat
 322 clnt_vc_call(
 323         CLIENT          *cl,            /* client handle */
 324         struct rpc_callextra *ext,      /* call metadata */
 325         rpcproc_t       proc,           /* procedure number */
 326         struct mbuf     *args,          /* pointer to args */
 327         struct mbuf     **resultsp,     /* pointer to results */
 328         struct timeval  utimeout)
 329 {
 330         struct ct_data *ct = (struct ct_data *) cl->cl_private;
 331         AUTH *auth;
 332         struct rpc_err *errp;
 333         enum clnt_stat stat;
 334         XDR xdrs;
 335         struct rpc_msg reply_msg;
 336         bool_t ok;
 337         int nrefreshes = 2;             /* number of times to refresh cred */
 338         struct timeval timeout;
 339         uint32_t xid;
 340         struct mbuf *mreq = NULL, *results;
 341         struct ct_request *cr;
 342         int error;
 343
 344         cr = malloc(sizeof(struct ct_request), M_RPC, M_WAITOK);
 345
 346         mtx_lock(&ct->ct_lock);
 347
 348         if (ct->ct_closing || ct->ct_closed) {
 349                 mtx_unlock(&ct->ct_lock);
 350                 free(cr, M_RPC);
 351                 return (RPC_CANTSEND);
 352         }
 353         ct->ct_threads++;
 354
 355         if (ext) {
 356                 auth = ext->rc_auth;
 357                 errp = &ext->rc_err;
 358         } else {
 359                 auth = cl->cl_auth;
 360                 errp = &ct->ct_error;
 361         }
 362
 363         cr->cr_mrep = NULL;
 364         cr->cr_error = 0;
 365
 366         if (ct->ct_wait.tv_usec == -1) {
 367                 timeout = utimeout;     /* use supplied timeout */
 368         } else {
 369                 timeout = ct->ct_wait;  /* use default timeout */
 370         }
 371
 372 call_again:
 373         mtx_assert(&ct->ct_lock, MA_OWNED);
 374
 375         ct->ct_xid++;
 376         xid = ct->ct_xid;
 377
 378         mtx_unlock(&ct->ct_lock);
 379
 380         /*
 381          * Leave space to pre-pend the record mark.
 382          */
 383         MGETHDR(mreq, M_WAIT, MT_DATA);
 384         mreq->m_data += sizeof(uint32_t);
 385         KASSERT(ct->ct_mpos + sizeof(uint32_t) <= MHLEN,
 386             ("RPC header too big"));
 387         bcopy(ct->ct_mcallc, mreq->m_data, ct->ct_mpos);
 388         mreq->m_len = ct->ct_mpos;
 389
 390         /*
 391          * The XID is the first thing in the request.
 392          */
 393         *mtod(mreq, uint32_t *) = htonl(xid);
 394
 395         xdrmbuf_create(&xdrs, mreq, XDR_ENCODE);
 396
 397         errp->re_status = stat = RPC_SUCCESS;
 398
 399         if ((! XDR_PUTINT32(&xdrs, &proc)) ||
 400             (! AUTH_MARSHALL(auth, xid, &xdrs,
 401                 m_copym(args, 0, M_COPYALL, M_WAITOK)))) {
 402                 errp->re_status = stat = RPC_CANTENCODEARGS;
 403                 mtx_lock(&ct->ct_lock);
 404                 goto out;
 405         }
 406         mreq->m_pkthdr.len = m_length(mreq, NULL);
 407
 408         /*
 409          * Prepend a record marker containing the packet length.
 410          */
 411         M_PREPEND(mreq, sizeof(uint32_t), M_WAIT);
 412         *mtod(mreq, uint32_t *) =
 413                 htonl(0x80000000 | (mreq->m_pkthdr.len - sizeof(uint32_t)));
 414
 415         cr->cr_xid = xid;
 416         mtx_lock(&ct->ct_lock);
 417         /*
 418          * Check to see if the other end has already started to close down
 419          * the connection. The upcall will have set ct_error.re_status
 420          * to RPC_CANTRECV if this is the case.
 421          * If the other end starts to close down the connection after this
 422          * point, it will be detected later when cr_error is checked,
 423          * since the request is in the ct_pending queue.
 424          */
 425         if (ct->ct_error.re_status == RPC_CANTRECV) {
 426                 if (errp != &ct->ct_error) {
 427                         errp->re_errno = ct->ct_error.re_errno;
 428                         errp->re_status = RPC_CANTRECV;
 429                 }
 430                 stat = RPC_CANTRECV;
 431                 goto out;
 432         }
 433         TAILQ_INSERT_TAIL(&ct->ct_pending, cr, cr_link);
 434         mtx_unlock(&ct->ct_lock);
 435
 436         /*
 437          * sosend consumes mreq.
 438          */
 439         error = sosend(ct->ct_socket, NULL, NULL, mreq, NULL, 0, curthread);
 440         mreq = NULL;
 441         if (error == EMSGSIZE) {
 442                 SOCKBUF_LOCK(&ct->ct_socket->so_snd);
 443                 sbwait(&ct->ct_socket->so_snd);
 444                 SOCKBUF_UNLOCK(&ct->ct_socket->so_snd);
 445                 AUTH_VALIDATE(auth, xid, NULL, NULL);
 446                 mtx_lock(&ct->ct_lock);
 447                 TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
 448                 goto call_again;
 449         }
 450
 451         reply_msg.acpted_rply.ar_verf.oa_flavor = AUTH_NULL;
 452         reply_msg.acpted_rply.ar_verf.oa_base = cr->cr_verf;
 453         reply_msg.acpted_rply.ar_verf.oa_length = 0;
 454         reply_msg.acpted_rply.ar_results.where = NULL;
 455         reply_msg.acpted_rply.ar_results.proc = (xdrproc_t)xdr_void;
 456
 457         mtx_lock(&ct->ct_lock);
 458         if (error) {
 459                 TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
 460                 errp->re_errno = error;
 461                 errp->re_status = stat = RPC_CANTSEND;
 462                 goto out;
 463         }
 464
 465         /*
 466          * Check to see if we got an upcall while waiting for the
 467          * lock. In both these cases, the request has been removed
 468          * from ct->ct_pending.
 469          */
 470         if (cr->cr_error) {
 471                 TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
 472                 errp->re_errno = cr->cr_error;
 473                 errp->re_status = stat = RPC_CANTRECV;
 474                 goto out;
 475         }
 476         if (cr->cr_mrep) {
 477                 TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
 478                 goto got_reply;
 479         }
 480
 481         /*
 482          * Hack to provide rpc-based message passing
 483          */
 484         if (timeout.tv_sec == 0 && timeout.tv_usec == 0) {
 485                 TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
 486                 errp->re_status = stat = RPC_TIMEDOUT;
 487                 goto out;
 488         }
 489
 490         error = msleep(cr, &ct->ct_lock, ct->ct_waitflag, ct->ct_waitchan,
 491             tvtohz(&timeout));
 492
 493         TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
 494
 495         if (error) {
 496                 /*
 497                  * The sleep returned an error so our request is still
 498                  * on the list. Turn the error code into an
 499                  * appropriate client status.
 500                  */
 501                 errp->re_errno = error;
 502                 switch (error) {
 503                 case EINTR:
 504                 case ERESTART:
 505                         stat = RPC_INTR;
 506                         break;
 507                 case EWOULDBLOCK:
 508                         stat = RPC_TIMEDOUT;
 509                         break;
 510                 default:
 511                         stat = RPC_CANTRECV;
 512                 }
 513                 errp->re_status = stat;
 514                 goto out;
 515         } else {
 516                 /*
 517                  * We were woken up by the upcall.  If the
 518                  * upcall had a receive error, report that,
 519                  * otherwise we have a reply.
 520                  */
 521                 if (cr->cr_error) {
 522                         errp->re_errno = cr->cr_error;
 523                         errp->re_status = stat = RPC_CANTRECV;
 524                         goto out;
 525                 }
 526         }
 527
 528 got_reply:
 529         /*
 530          * Now decode and validate the response. We need to drop the
 531          * lock since xdr_replymsg may end up sleeping in malloc.
 532          */
 533         mtx_unlock(&ct->ct_lock);
 534
 535         if (ext && ext->rc_feedback)
 536                 ext->rc_feedback(FEEDBACK_OK, proc, ext->rc_feedback_arg);
 537
 538         xdrmbuf_create(&xdrs, cr->cr_mrep, XDR_DECODE);
 539         ok = xdr_replymsg(&xdrs, &reply_msg);
 540         cr->cr_mrep = NULL;
 541
 542         if (ok) {
 543                 if ((reply_msg.rm_reply.rp_stat == MSG_ACCEPTED) &&
 544                     (reply_msg.acpted_rply.ar_stat == SUCCESS))
 545                         errp->re_status = stat = RPC_SUCCESS;
 546                 else
 547                         stat = _seterr_reply(&reply_msg, errp);
 548
 549                 if (stat == RPC_SUCCESS) {
 550                         results = xdrmbuf_getall(&xdrs);
 551                         if (!AUTH_VALIDATE(auth, xid,
 552                                 &reply_msg.acpted_rply.ar_verf,
 553                                 &results)) {
 554                                 errp->re_status = stat = RPC_AUTHERROR;
 555                                 errp->re_why = AUTH_INVALIDRESP;
 556                         } else {
 557                                 KASSERT(results,
 558                                     ("auth validated but no result"));
 559                                 *resultsp = results;
 560                         }
 561                 }               /* end successful completion */
 562                 /*
 563                  * If unsuccesful AND error is an authentication error
 564                  * then refresh credentials and try again, else break
 565                  */
 566                 else if (stat == RPC_AUTHERROR)
 567                         /* maybe our credentials need to be refreshed ... */
 568                         if (nrefreshes > 0 &&
 569                             AUTH_REFRESH(auth, &reply_msg)) {
 570                                 nrefreshes--;
 571                                 XDR_DESTROY(&xdrs);
 572                                 mtx_lock(&ct->ct_lock);
 573                                 goto call_again;
 574                         }
 575                 /* end of unsuccessful completion */
 576         }       /* end of valid reply message */
 577         else {
 578                 errp->re_status = stat = RPC_CANTDECODERES;
 579         }
 580         XDR_DESTROY(&xdrs);
 581         mtx_lock(&ct->ct_lock);
 582 out:
 583         mtx_assert(&ct->ct_lock, MA_OWNED);
 584
 585         KASSERT(stat != RPC_SUCCESS || *resultsp,
 586             ("RPC_SUCCESS without reply"));
 587
 588         if (mreq)
 589                 m_freem(mreq);
 590         if (cr->cr_mrep)
 591                 m_freem(cr->cr_mrep);
 592
 593         ct->ct_threads--;
 594         if (ct->ct_closing)
 595                 wakeup(ct);
 596
 597         mtx_unlock(&ct->ct_lock);
 598
 599         if (auth && stat != RPC_SUCCESS)
 600                 AUTH_VALIDATE(auth, xid, NULL, NULL);
 601
 602         free(cr, M_RPC);
 603
 604         return (stat);
 605 }
 606
 607 static void
 608 clnt_vc_geterr(CLIENT *cl, struct rpc_err *errp)
 609 {
 610         struct ct_data *ct = (struct ct_data *) cl->cl_private;
 611
 612         *errp = ct->ct_error;
 613 }
 614
 615 static bool_t
 616 clnt_vc_freeres(CLIENT *cl, xdrproc_t xdr_res, void *res_ptr)
 617 {
 618         XDR xdrs;
 619         bool_t dummy;
 620
 621         xdrs.x_op = XDR_FREE;
 622         dummy = (*xdr_res)(&xdrs, res_ptr);
 623
 624         return (dummy);
 625 }
 626
 627 /*ARGSUSED*/
 628 static void
 629 clnt_vc_abort(CLIENT *cl)
 630 {
 631 }
 632
 633 static bool_t
 634 clnt_vc_control(CLIENT *cl, u_int request, void *info)
 635 {
 636         struct ct_data *ct = (struct ct_data *)cl->cl_private;
 637         void *infop = info;
 638
 639         mtx_lock(&ct->ct_lock);
 640
 641         switch (request) {
 642         case CLSET_FD_CLOSE:
 643                 ct->ct_closeit = TRUE;
 644                 mtx_unlock(&ct->ct_lock);
 645                 return (TRUE);
 646         case CLSET_FD_NCLOSE:
 647                 ct->ct_closeit = FALSE;
 648                 mtx_unlock(&ct->ct_lock);
 649                 return (TRUE);
 650         default:
 651                 break;
 652         }
 653
 654         /* for other requests which use info */
 655         if (info == NULL) {
 656                 mtx_unlock(&ct->ct_lock);
 657                 return (FALSE);
 658         }
 659         switch (request) {
 660         case CLSET_TIMEOUT:
 661                 if (time_not_ok((struct timeval *)info)) {
 662                         mtx_unlock(&ct->ct_lock);
 663                         return (FALSE);
 664                 }
 665                 ct->ct_wait = *(struct timeval *)infop;
 666                 break;
 667         case CLGET_TIMEOUT:
 668                 *(struct timeval *)infop = ct->ct_wait;
 669                 break;
 670         case CLGET_SERVER_ADDR:
 671                 (void) memcpy(info, &ct->ct_addr, (size_t)ct->ct_addr.ss_len);
 672                 break;
 673         case CLGET_SVC_ADDR:
 674                 /*
 675                  * Slightly different semantics to userland - we use
 676                  * sockaddr instead of netbuf.
 677                  */
 678                 memcpy(info, &ct->ct_addr, ct->ct_addr.ss_len);
 679                 break;
 680         case CLSET_SVC_ADDR:            /* set to new address */
 681                 mtx_unlock(&ct->ct_lock);
 682                 return (FALSE);
 683         case CLGET_XID:
 684                 *(uint32_t *)info = ct->ct_xid;
 685                 break;
 686         case CLSET_XID:
 687                 /* This will set the xid of the NEXT call */
 688                 /* decrement by 1 as clnt_vc_call() increments once */
 689                 ct->ct_xid = *(uint32_t *)info - 1;
 690                 break;
 691         case CLGET_VERS:
 692                 /*
 693                  * This RELIES on the information that, in the call body,
 694                  * the version number field is the fifth field from the
 695                  * begining of the RPC header. MUST be changed if the
 696                  * call_struct is changed
 697                  */
 698                 *(uint32_t *)info =
 699                     ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
 700                     4 * BYTES_PER_XDR_UNIT));
 701                 break;
 702
 703         case CLSET_VERS:
 704                 *(uint32_t *)(void *)(ct->ct_mcallc +
 705                     4 * BYTES_PER_XDR_UNIT) =
 706                     htonl(*(uint32_t *)info);
 707                 break;
 708
 709         case CLGET_PROG:
 710                 /*
 711                  * This RELIES on the information that, in the call body,
 712                  * the program number field is the fourth field from the
 713                  * begining of the RPC header. MUST be changed if the
 714                  * call_struct is changed
 715                  */
 716                 *(uint32_t *)info =
 717                     ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
 718                     3 * BYTES_PER_XDR_UNIT));
 719                 break;
 720
 721         case CLSET_PROG:
 722                 *(uint32_t *)(void *)(ct->ct_mcallc +
 723                     3 * BYTES_PER_XDR_UNIT) =
 724                     htonl(*(uint32_t *)info);
 725                 break;
 726
 727         case CLSET_WAITCHAN:
 728                 ct->ct_waitchan = (const char *)info;
 729                 break;
 730
 731         case CLGET_WAITCHAN:
 732                 *(const char **) info = ct->ct_waitchan;
 733                 break;
 734
 735         case CLSET_INTERRUPTIBLE:
 736                 if (*(int *) info)
 737                         ct->ct_waitflag = PCATCH | PBDRY;
 738                 else
 739                         ct->ct_waitflag = 0;
 740                 break;
 741
 742         case CLGET_INTERRUPTIBLE:
 743                 if (ct->ct_waitflag)
 744                         *(int *) info = TRUE;
 745                 else
 746                         *(int *) info = FALSE;
 747                 break;
 748
 749         default:
 750                 mtx_unlock(&ct->ct_lock);
 751                 return (FALSE);
 752         }
 753
 754         mtx_unlock(&ct->ct_lock);
 755         return (TRUE);
 756 }
 757
 758 static void
 759 clnt_vc_close(CLIENT *cl)
 760 {
 761         struct ct_data *ct = (struct ct_data *) cl->cl_private;
 762         struct ct_request *cr;
 763
 764         mtx_lock(&ct->ct_lock);
 765
 766         if (ct->ct_closed) {
 767                 mtx_unlock(&ct->ct_lock);
 768                 return;
 769         }
 770
 771         if (ct->ct_closing) {
 772                 while (ct->ct_closing)
 773                         msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
 774                 KASSERT(ct->ct_closed, ("client should be closed"));
 775                 mtx_unlock(&ct->ct_lock);
 776                 return;
 777         }
 778
 779         if (ct->ct_socket) {
 780                 ct->ct_closing = TRUE;
 781                 mtx_unlock(&ct->ct_lock);
 782
 783                 SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
 784                 soupcall_clear(ct->ct_socket, SO_RCV);
 785                 clnt_vc_upcallsdone(ct);
 786                 SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
 787
 788                 /*
 789                  * Abort any pending requests and wait until everyone
 790                  * has finished with clnt_vc_call.
 791                  */
 792                 mtx_lock(&ct->ct_lock);
 793                 TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
 794                         cr->cr_xid = 0;
 795                         cr->cr_error = ESHUTDOWN;
 796                         wakeup(cr);
 797                 }
 798
 799                 while (ct->ct_threads)
 800                         msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
 801         }
 802
 803         ct->ct_closing = FALSE;
 804         ct->ct_closed = TRUE;
 805         mtx_unlock(&ct->ct_lock);
 806         wakeup(ct);
 807 }
 808
 809 static void
 810 clnt_vc_destroy(CLIENT *cl)
 811 {
 812         struct ct_data *ct = (struct ct_data *) cl->cl_private;
 813         struct socket *so = NULL;
 814
 815         clnt_vc_close(cl);
 816
 817         mtx_lock(&ct->ct_lock);
 818
 819         if (ct->ct_socket) {
 820                 if (ct->ct_closeit) {
 821                         so = ct->ct_socket;
 822                 }
 823         }
 824
 825         mtx_unlock(&ct->ct_lock);
 826
 827         mtx_destroy(&ct->ct_lock);
 828         if (so) {
 829                 soshutdown(so, SHUT_WR);
 830                 soclose(so);
 831         }
 832         mem_free(ct, sizeof(struct ct_data));
 833         if (cl->cl_netid && cl->cl_netid[0])
 834                 mem_free(cl->cl_netid, strlen(cl->cl_netid) +1);
 835         if (cl->cl_tp && cl->cl_tp[0])
 836                 mem_free(cl->cl_tp, strlen(cl->cl_tp) +1);
 837         mem_free(cl, sizeof(CLIENT));
 838 }
 839
 840 /*
 841  * Make sure that the time is not garbage.   -1 value is disallowed.
 842  * Note this is different from time_not_ok in clnt_dg.c
 843  */
 844 static bool_t
 845 time_not_ok(struct timeval *t)
 846 {
 847         return (t->tv_sec <= -1 || t->tv_sec > 100000000 ||
 848                 t->tv_usec <= -1 || t->tv_usec > 1000000);
 849 }
 850
 851 int
 852 clnt_vc_soupcall(struct socket *so, void *arg, int waitflag)
 853 {
 854         struct ct_data *ct = (struct ct_data *) arg;
 855         struct uio uio;
 856         struct mbuf *m;
 857         struct ct_request *cr;
 858         int error, rcvflag, foundreq;
 859         uint32_t xid, header;
 860         bool_t do_read;
 861
 862         ct->ct_upcallrefs++;
 863         uio.uio_td = curthread;
 864         do {
 865                 /*
 866                  * If ct_record_resid is zero, we are waiting for a
 867                  * record mark.
 868                  */
 869                 if (ct->ct_record_resid == 0) {
 870
 871                         /*
 872                          * Make sure there is either a whole record
 873                          * mark in the buffer or there is some other
 874                          * error condition
 875                          */
 876                         do_read = FALSE;
 877                         if (so->so_rcv.sb_cc >= sizeof(uint32_t)
 878                             || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
 879                             || so->so_error)
 880                                 do_read = TRUE;
 881
 882                         if (!do_read)
 883                                 break;
 884
 885                         SOCKBUF_UNLOCK(&so->so_rcv);
 886                         uio.uio_resid = sizeof(uint32_t);
 887                         m = NULL;
 888                         rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
 889                         error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
 890                         SOCKBUF_LOCK(&so->so_rcv);
 891
 892                         if (error == EWOULDBLOCK)
 893                                 break;
 894
 895                         /*
 896                          * If there was an error, wake up all pending
 897                          * requests.
 898                          */
 899                         if (error || uio.uio_resid > 0) {
 900                         wakeup_all:
 901                                 mtx_lock(&ct->ct_lock);
 902                                 if (!error) {
 903                                         /*
 904                                          * We must have got EOF trying
 905                                          * to read from the stream.
 906                                          */
 907                                         error = ECONNRESET;
 908                                 }
 909                                 ct->ct_error.re_status = RPC_CANTRECV;
 910                                 ct->ct_error.re_errno = error;
 911                                 TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
 912                                         cr->cr_error = error;
 913                                         wakeup(cr);
 914                                 }
 915                                 mtx_unlock(&ct->ct_lock);
 916                                 break;
 917                         }
 918                         m_copydata(m, 0, sizeof(uint32_t), (char *)&header);
 919                         header = ntohl(header);
 920                         ct->ct_record = NULL;
 921                         ct->ct_record_resid = header & 0x7fffffff;
 922                         ct->ct_record_eor = ((header & 0x80000000) != 0);
 923                         m_freem(m);
 924                 } else {
 925                         /*
 926                          * Wait until the socket has the whole record
 927                          * buffered.
 928                          */
 929                         do_read = FALSE;
 930                         if (so->so_rcv.sb_cc >= ct->ct_record_resid
 931                             || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
 932                             || so->so_error)
 933                                 do_read = TRUE;
 934
 935                         if (!do_read)
 936                                 break;
 937
 938                         /*
 939                          * We have the record mark. Read as much as
 940                          * the socket has buffered up to the end of
 941                          * this record.
 942                          */
 943                         SOCKBUF_UNLOCK(&so->so_rcv);
 944                         uio.uio_resid = ct->ct_record_resid;
 945                         m = NULL;
 946                         rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
 947                         error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
 948                         SOCKBUF_LOCK(&so->so_rcv);
 949
 950                         if (error == EWOULDBLOCK)
 951                                 break;
 952
 953                         if (error || uio.uio_resid == ct->ct_record_resid)
 954                                 goto wakeup_all;
 955
 956                         /*
 957                          * If we have part of the record already,
 958                          * chain this bit onto the end.
 959                          */
 960                         if (ct->ct_record)
 961                                 m_last(ct->ct_record)->m_next = m;
 962                         else
 963                                 ct->ct_record = m;
 964
 965                         ct->ct_record_resid = uio.uio_resid;
 966
 967                         /*
 968                          * If we have the entire record, see if we can
 969                          * match it to a request.
 970                          */
 971                         if (ct->ct_record_resid == 0
 972                             && ct->ct_record_eor) {
 973                                 /*
 974                                  * The XID is in the first uint32_t of
 975                                  * the reply.
 976                                  */
 977                                 if (ct->ct_record->m_len < sizeof(xid) &&
 978                                     m_length(ct->ct_record, NULL) <
 979                                     sizeof(xid)) {
 980                                         m_freem(ct->ct_record);
 981                                         break;
 982                                 }
 983                                 m_copydata(ct->ct_record, 0, sizeof(xid),
 984                                     (char *)&xid);
 985                                 xid = ntohl(xid);
 986
 987                                 mtx_lock(&ct->ct_lock);
 988                                 foundreq = 0;
 989                                 TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
 990                                         if (cr->cr_xid == xid) {
 991                                                 /*
 992                                                  * This one
 993                                                  * matches. We leave
 994                                                  * the reply mbuf in
 995                                                  * cr->cr_mrep. Set
 996                                                  * the XID to zero so
 997                                                  * that we will ignore
 998                                                  * any duplicaed
 999                                                  * replies.
1000                                                  */
1001                                                 cr->cr_xid = 0;
1002                                                 cr->cr_mrep = ct->ct_record;
1003                                                 cr->cr_error = 0;
1004                                                 foundreq = 1;
1005                                                 wakeup(cr);
1006                                                 break;
1007                                         }
1008                                 }
1009                                 mtx_unlock(&ct->ct_lock);
1010
1011                                 if (!foundreq)
1012                                         m_freem(ct->ct_record);
1013                                 ct->ct_record = NULL;
1014                         }
1015                 }
1016         } while (m);
1017         ct->ct_upcallrefs--;
1018         if (ct->ct_upcallrefs < 0)
1019                 panic("rpcvc upcall refcnt");
1020         if (ct->ct_upcallrefs == 0)
1021                 wakeup(&ct->ct_upcallrefs);
1022         return (SU_OK);
1023 }
1024
1025 /*
1026  * Wait for all upcalls in progress to complete.
1027  */
1028 static void
1029 clnt_vc_upcallsdone(struct ct_data *ct)
1030 {
1031
1032         SOCKBUF_LOCK_ASSERT(&ct->ct_socket->so_rcv);
1033
1034         while (ct->ct_upcallrefs > 0)
1035                 (void) msleep(&ct->ct_upcallrefs,
1036                     SOCKBUF_MTX(&ct->ct_socket->so_rcv), 0, "rpcvcup", 0);
1037 }