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/kernel.h>
  61 #include <sys/lock.h>
  62 #include <sys/malloc.h>
  63 #include <sys/mbuf.h>
  64 #include <sys/mutex.h>
  65 #include <sys/pcpu.h>
  66 #include <sys/proc.h>
  67 #include <sys/protosw.h>
  68 #include <sys/socket.h>
  69 #include <sys/socketvar.h>
  70 #include <sys/sx.h>
  71 #include <sys/syslog.h>
  72 #include <sys/time.h>
  73 #include <sys/uio.h>
  74
  75 #include <net/vnet.h>
  76
  77 #include <netinet/tcp.h>
  78
  79 #include <rpc/rpc.h>
  80 #include <rpc/rpc_com.h>
  81 #include <rpc/krpc.h>
  82
  83 struct cmessage {
  84         struct cmsghdr cmsg;
  85         struct cmsgcred cmcred;
  86 };
  87
  88 static enum clnt_stat clnt_vc_call(CLIENT *, struct rpc_callextra *,
  89     rpcproc_t, struct mbuf *, struct mbuf **, struct timeval);
  90 static void clnt_vc_geterr(CLIENT *, struct rpc_err *);
  91 static bool_t clnt_vc_freeres(CLIENT *, xdrproc_t, void *);
  92 static void clnt_vc_abort(CLIENT *);
  93 static bool_t clnt_vc_control(CLIENT *, u_int, void *);
  94 static void clnt_vc_close(CLIENT *);
  95 static void clnt_vc_destroy(CLIENT *);
  96 static bool_t time_not_ok(struct timeval *);
  97 static int clnt_vc_soupcall(struct socket *so, void *arg, int waitflag);
  98
  99 static struct clnt_ops clnt_vc_ops = {
 100         .cl_call =      clnt_vc_call,
 101         .cl_abort =     clnt_vc_abort,
 102         .cl_geterr =    clnt_vc_geterr,
 103         .cl_freeres =   clnt_vc_freeres,
 104         .cl_close =     clnt_vc_close,
 105         .cl_destroy =   clnt_vc_destroy,
 106         .cl_control =   clnt_vc_control
 107 };
 108
 109 static void clnt_vc_upcallsdone(struct ct_data *);
 110
 111 static int      fake_wchan;
 112
 113 /*
 114  * Create a client handle for a connection.
 115  * Default options are set, which the user can change using clnt_control()'s.
 116  * The rpc/vc package does buffering similar to stdio, so the client
 117  * must pick send and receive buffer sizes, 0 => use the default.
 118  * NB: fd is copied into a private area.
 119  * NB: The rpch->cl_auth is set null authentication. Caller may wish to
 120  * set this something more useful.
 121  *
 122  * fd should be an open socket
 123  */
 124 CLIENT *
 125 clnt_vc_create(
 126         struct socket *so,              /* open file descriptor */
 127         struct sockaddr *raddr,         /* servers address */
 128         const rpcprog_t prog,           /* program number */
 129         const rpcvers_t vers,           /* version number */
 130         size_t sendsz,                  /* buffer recv size */
 131         size_t recvsz,                  /* buffer send size */
 132         int intrflag)                   /* interruptible */
 133 {
 134         CLIENT *cl;                     /* client handle */
 135         struct ct_data *ct = NULL;      /* client handle */
 136         struct timeval now;
 137         struct rpc_msg call_msg;
 138         static uint32_t disrupt;
 139         struct __rpc_sockinfo si;
 140         XDR xdrs;
 141         int error, interrupted, one = 1, sleep_flag;
 142         struct sockopt sopt;
 143
 144         if (disrupt == 0)
 145                 disrupt = (uint32_t)(long)raddr;
 146
 147         cl = (CLIENT *)mem_alloc(sizeof (*cl));
 148         ct = (struct ct_data *)mem_alloc(sizeof (*ct));
 149
 150         mtx_init(&ct->ct_lock, "ct->ct_lock", NULL, MTX_DEF);
 151         ct->ct_threads = 0;
 152         ct->ct_closing = FALSE;
 153         ct->ct_closed = FALSE;
 154         ct->ct_upcallrefs = 0;
 155
 156         if ((so->so_state & (SS_ISCONNECTED|SS_ISCONFIRMING)) == 0) {
 157                 error = soconnect(so, raddr, curthread);
 158                 SOCK_LOCK(so);
 159                 interrupted = 0;
 160                 sleep_flag = PSOCK;
 161                 if (intrflag != 0)
 162                         sleep_flag |= PCATCH;
 163                 while ((so->so_state & SS_ISCONNECTING)
 164                     && so->so_error == 0) {
 165                         error = msleep(&so->so_timeo, SOCK_MTX(so),
 166                             sleep_flag, "connec", 0);
 167                         if (error) {
 168                                 if (error == EINTR || error == ERESTART)
 169                                         interrupted = 1;
 170                                 break;
 171                         }
 172                 }
 173                 if (error == 0) {
 174                         error = so->so_error;
 175                         so->so_error = 0;
 176                 }
 177                 SOCK_UNLOCK(so);
 178                 if (error) {
 179                         if (!interrupted)
 180                                 so->so_state &= ~SS_ISCONNECTING;
 181                         rpc_createerr.cf_stat = RPC_SYSTEMERROR;
 182                         rpc_createerr.cf_error.re_errno = error;
 183                         goto err;
 184                 }
 185         }
 186
 187         if (!__rpc_socket2sockinfo(so, &si)) {
 188                 goto err;
 189         }
 190
 191         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
 192                 bzero(&sopt, sizeof(sopt));
 193                 sopt.sopt_dir = SOPT_SET;
 194                 sopt.sopt_level = SOL_SOCKET;
 195                 sopt.sopt_name = SO_KEEPALIVE;
 196                 sopt.sopt_val = &one;
 197                 sopt.sopt_valsize = sizeof(one);
 198                 sosetopt(so, &sopt);
 199         }
 200
 201         if (so->so_proto->pr_protocol == IPPROTO_TCP) {
 202                 bzero(&sopt, sizeof(sopt));
 203                 sopt.sopt_dir = SOPT_SET;
 204                 sopt.sopt_level = IPPROTO_TCP;
 205                 sopt.sopt_name = TCP_NODELAY;
 206                 sopt.sopt_val = &one;
 207                 sopt.sopt_valsize = sizeof(one);
 208                 sosetopt(so, &sopt);
 209         }
 210
 211         ct->ct_closeit = FALSE;
 212
 213         /*
 214          * Set up private data struct
 215          */
 216         ct->ct_socket = so;
 217         ct->ct_wait.tv_sec = -1;
 218         ct->ct_wait.tv_usec = -1;
 219         memcpy(&ct->ct_addr, raddr, raddr->sa_len);
 220
 221         /*
 222          * Initialize call message
 223          */
 224         getmicrotime(&now);
 225         ct->ct_xid = ((uint32_t)++disrupt) ^ __RPC_GETXID(&now);
 226         call_msg.rm_xid = ct->ct_xid;
 227         call_msg.rm_direction = CALL;
 228         call_msg.rm_call.cb_rpcvers = RPC_MSG_VERSION;
 229         call_msg.rm_call.cb_prog = (uint32_t)prog;
 230         call_msg.rm_call.cb_vers = (uint32_t)vers;
 231
 232         /*
 233          * pre-serialize the static part of the call msg and stash it away
 234          */
 235         xdrmem_create(&xdrs, ct->ct_mcallc, MCALL_MSG_SIZE,
 236             XDR_ENCODE);
 237         if (! xdr_callhdr(&xdrs, &call_msg)) {
 238                 if (ct->ct_closeit) {
 239                         soclose(ct->ct_socket);
 240                 }
 241                 goto err;
 242         }
 243         ct->ct_mpos = XDR_GETPOS(&xdrs);
 244         XDR_DESTROY(&xdrs);
 245         ct->ct_waitchan = "rpcrecv";
 246         ct->ct_waitflag = 0;
 247
 248         /*
 249          * Create a client handle which uses xdrrec for serialization
 250          * and authnone for authentication.
 251          */
 252         sendsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)sendsz);
 253         recvsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)recvsz);
 254         error = soreserve(ct->ct_socket, sendsz, recvsz);
 255         if (error != 0) {
 256                 if (ct->ct_closeit) {
 257                         soclose(ct->ct_socket);
 258                 }
 259                 goto err;
 260         }
 261         cl->cl_refs = 1;
 262         cl->cl_ops = &clnt_vc_ops;
 263         cl->cl_private = ct;
 264         cl->cl_auth = authnone_create();
 265
 266         SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
 267         soupcall_set(ct->ct_socket, SO_RCV, clnt_vc_soupcall, ct);
 268         SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
 269
 270         ct->ct_record = NULL;
 271         ct->ct_record_resid = 0;
 272         TAILQ_INIT(&ct->ct_pending);
 273         return (cl);
 274
 275 err:
 276         mtx_destroy(&ct->ct_lock);
 277         mem_free(ct, sizeof (struct ct_data));
 278         mem_free(cl, sizeof (CLIENT));
 279
 280         return ((CLIENT *)NULL);
 281 }
 282
 283 static enum clnt_stat
 284 clnt_vc_call(
 285         CLIENT          *cl,            /* client handle */
 286         struct rpc_callextra *ext,      /* call metadata */
 287         rpcproc_t       proc,           /* procedure number */
 288         struct mbuf     *args,          /* pointer to args */
 289         struct mbuf     **resultsp,     /* pointer to results */
 290         struct timeval  utimeout)
 291 {
 292         struct ct_data *ct = (struct ct_data *) cl->cl_private;
 293         AUTH *auth;
 294         struct rpc_err *errp;
 295         enum clnt_stat stat;
 296         XDR xdrs;
 297         struct rpc_msg reply_msg;
 298         bool_t ok;
 299         int nrefreshes = 2;             /* number of times to refresh cred */
 300         struct timeval timeout;
 301         uint32_t xid;
 302         struct mbuf *mreq = NULL, *results;
 303         struct ct_request *cr;
 304         int error, trycnt;
 305
 306         cr = malloc(sizeof(struct ct_request), M_RPC, M_WAITOK);
 307
 308         mtx_lock(&ct->ct_lock);
 309
 310         if (ct->ct_closing || ct->ct_closed) {
 311                 mtx_unlock(&ct->ct_lock);
 312                 free(cr, M_RPC);
 313                 return (RPC_CANTSEND);
 314         }
 315         ct->ct_threads++;
 316
 317         if (ext) {
 318                 auth = ext->rc_auth;
 319                 errp = &ext->rc_err;
 320         } else {
 321                 auth = cl->cl_auth;
 322                 errp = &ct->ct_error;
 323         }
 324
 325         cr->cr_mrep = NULL;
 326         cr->cr_error = 0;
 327
 328         if (ct->ct_wait.tv_usec == -1) {
 329                 timeout = utimeout;     /* use supplied timeout */
 330         } else {
 331                 timeout = ct->ct_wait;  /* use default timeout */
 332         }
 333
 334         /*
 335          * After 15sec of looping, allow it to return RPC_CANTSEND, which will
 336          * cause the clnt_reconnect layer to create a new TCP connection.
 337          */
 338         trycnt = 15 * hz;
 339 call_again:
 340         mtx_assert(&ct->ct_lock, MA_OWNED);
 341         if (ct->ct_closing || ct->ct_closed) {
 342                 ct->ct_threads--;
 343                 wakeup(ct);
 344                 mtx_unlock(&ct->ct_lock);
 345                 free(cr, M_RPC);
 346                 return (RPC_CANTSEND);
 347         }
 348
 349         ct->ct_xid++;
 350         xid = ct->ct_xid;
 351
 352         mtx_unlock(&ct->ct_lock);
 353
 354         /*
 355          * Leave space to pre-pend the record mark.
 356          */
 357         mreq = m_gethdr(M_WAITOK, MT_DATA);
 358         mreq->m_data += sizeof(uint32_t);
 359         KASSERT(ct->ct_mpos + sizeof(uint32_t) <= MHLEN,
 360             ("RPC header too big"));
 361         bcopy(ct->ct_mcallc, mreq->m_data, ct->ct_mpos);
 362         mreq->m_len = ct->ct_mpos;
 363
 364         /*
 365          * The XID is the first thing in the request.
 366          */
 367         *mtod(mreq, uint32_t *) = htonl(xid);
 368
 369         xdrmbuf_create(&xdrs, mreq, XDR_ENCODE);
 370
 371         errp->re_status = stat = RPC_SUCCESS;
 372
 373         if ((! XDR_PUTINT32(&xdrs, &proc)) ||
 374             (! AUTH_MARSHALL(auth, xid, &xdrs,
 375                 m_copym(args, 0, M_COPYALL, M_WAITOK)))) {
 376                 errp->re_status = stat = RPC_CANTENCODEARGS;
 377                 mtx_lock(&ct->ct_lock);
 378                 goto out;
 379         }
 380         mreq->m_pkthdr.len = m_length(mreq, NULL);
 381
 382         /*
 383          * Prepend a record marker containing the packet length.
 384          */
 385         M_PREPEND(mreq, sizeof(uint32_t), M_WAITOK);
 386         *mtod(mreq, uint32_t *) =
 387                 htonl(0x80000000 | (mreq->m_pkthdr.len - sizeof(uint32_t)));
 388
 389         cr->cr_xid = xid;
 390         mtx_lock(&ct->ct_lock);
 391         /*
 392          * Check to see if the other end has already started to close down
 393          * the connection. The upcall will have set ct_error.re_status
 394          * to RPC_CANTRECV if this is the case.
 395          * If the other end starts to close down the connection after this
 396          * point, it will be detected later when cr_error is checked,
 397          * since the request is in the ct_pending queue.
 398          */
 399         if (ct->ct_error.re_status == RPC_CANTRECV) {
 400                 if (errp != &ct->ct_error) {
 401                         errp->re_errno = ct->ct_error.re_errno;
 402                         errp->re_status = RPC_CANTRECV;
 403                 }
 404                 stat = RPC_CANTRECV;
 405                 goto out;
 406         }
 407         TAILQ_INSERT_TAIL(&ct->ct_pending, cr, cr_link);
 408         mtx_unlock(&ct->ct_lock);
 409
 410         /*
 411          * sosend consumes mreq.
 412          */
 413         error = sosend(ct->ct_socket, NULL, NULL, mreq, NULL, 0, curthread);
 414         mreq = NULL;
 415         if (error == EMSGSIZE || (error == ERESTART &&
 416             (ct->ct_waitflag & PCATCH) == 0 && trycnt-- > 0)) {
 417                 SOCKBUF_LOCK(&ct->ct_socket->so_snd);
 418                 sbwait(&ct->ct_socket->so_snd);
 419                 SOCKBUF_UNLOCK(&ct->ct_socket->so_snd);
 420                 AUTH_VALIDATE(auth, xid, NULL, NULL);
 421                 mtx_lock(&ct->ct_lock);
 422                 TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
 423                 /* Sleep for 1 clock tick before trying the sosend() again. */
 424                 msleep(&fake_wchan, &ct->ct_lock, 0, "rpclpsnd", 1);
 425                 goto call_again;
 426         }
 427
 428         reply_msg.acpted_rply.ar_verf.oa_flavor = AUTH_NULL;
 429         reply_msg.acpted_rply.ar_verf.oa_base = cr->cr_verf;
 430         reply_msg.acpted_rply.ar_verf.oa_length = 0;
 431         reply_msg.acpted_rply.ar_results.where = NULL;
 432         reply_msg.acpted_rply.ar_results.proc = (xdrproc_t)xdr_void;
 433
 434         mtx_lock(&ct->ct_lock);
 435         if (error) {
 436                 TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
 437                 errp->re_errno = error;
 438                 errp->re_status = stat = RPC_CANTSEND;
 439                 goto out;
 440         }
 441
 442         /*
 443          * Check to see if we got an upcall while waiting for the
 444          * lock. In both these cases, the request has been removed
 445          * from ct->ct_pending.
 446          */
 447         if (cr->cr_error) {
 448                 TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
 449                 errp->re_errno = cr->cr_error;
 450                 errp->re_status = stat = RPC_CANTRECV;
 451                 goto out;
 452         }
 453         if (cr->cr_mrep) {
 454                 TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
 455                 goto got_reply;
 456         }
 457
 458         /*
 459          * Hack to provide rpc-based message passing
 460          */
 461         if (timeout.tv_sec == 0 && timeout.tv_usec == 0) {
 462                 TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
 463                 errp->re_status = stat = RPC_TIMEDOUT;
 464                 goto out;
 465         }
 466
 467         error = msleep(cr, &ct->ct_lock, ct->ct_waitflag, ct->ct_waitchan,
 468             tvtohz(&timeout));
 469
 470         TAILQ_REMOVE(&ct->ct_pending, cr, cr_link);
 471
 472         if (error) {
 473                 /*
 474                  * The sleep returned an error so our request is still
 475                  * on the list. Turn the error code into an
 476                  * appropriate client status.
 477                  */
 478                 errp->re_errno = error;
 479                 switch (error) {
 480                 case EINTR:
 481                         stat = RPC_INTR;
 482                         break;
 483                 case EWOULDBLOCK:
 484                         stat = RPC_TIMEDOUT;
 485                         break;
 486                 default:
 487                         stat = RPC_CANTRECV;
 488                 }
 489                 errp->re_status = stat;
 490                 goto out;
 491         } else {
 492                 /*
 493                  * We were woken up by the upcall.  If the
 494                  * upcall had a receive error, report that,
 495                  * otherwise we have a reply.
 496                  */
 497                 if (cr->cr_error) {
 498                         errp->re_errno = cr->cr_error;
 499                         errp->re_status = stat = RPC_CANTRECV;
 500                         goto out;
 501                 }
 502         }
 503
 504 got_reply:
 505         /*
 506          * Now decode and validate the response. We need to drop the
 507          * lock since xdr_replymsg may end up sleeping in malloc.
 508          */
 509         mtx_unlock(&ct->ct_lock);
 510
 511         if (ext && ext->rc_feedback)
 512                 ext->rc_feedback(FEEDBACK_OK, proc, ext->rc_feedback_arg);
 513
 514         xdrmbuf_create(&xdrs, cr->cr_mrep, XDR_DECODE);
 515         ok = xdr_replymsg(&xdrs, &reply_msg);
 516         cr->cr_mrep = NULL;
 517
 518         if (ok) {
 519                 if ((reply_msg.rm_reply.rp_stat == MSG_ACCEPTED) &&
 520                     (reply_msg.acpted_rply.ar_stat == SUCCESS))
 521                         errp->re_status = stat = RPC_SUCCESS;
 522                 else
 523                         stat = _seterr_reply(&reply_msg, errp);
 524
 525                 if (stat == RPC_SUCCESS) {
 526                         results = xdrmbuf_getall(&xdrs);
 527                         if (!AUTH_VALIDATE(auth, xid,
 528                                 &reply_msg.acpted_rply.ar_verf,
 529                                 &results)) {
 530                                 errp->re_status = stat = RPC_AUTHERROR;
 531                                 errp->re_why = AUTH_INVALIDRESP;
 532                         } else {
 533                                 KASSERT(results,
 534                                     ("auth validated but no result"));
 535                                 *resultsp = results;
 536                         }
 537                 }               /* end successful completion */
 538                 /*
 539                  * If unsuccesful AND error is an authentication error
 540                  * then refresh credentials and try again, else break
 541                  */
 542                 else if (stat == RPC_AUTHERROR)
 543                         /* maybe our credentials need to be refreshed ... */
 544                         if (nrefreshes > 0 &&
 545                             AUTH_REFRESH(auth, &reply_msg)) {
 546                                 nrefreshes--;
 547                                 XDR_DESTROY(&xdrs);
 548                                 mtx_lock(&ct->ct_lock);
 549                                 goto call_again;
 550                         }
 551                 /* end of unsuccessful completion */
 552         }       /* end of valid reply message */
 553         else {
 554                 errp->re_status = stat = RPC_CANTDECODERES;
 555         }
 556         XDR_DESTROY(&xdrs);
 557         mtx_lock(&ct->ct_lock);
 558 out:
 559         mtx_assert(&ct->ct_lock, MA_OWNED);
 560
 561         KASSERT(stat != RPC_SUCCESS || *resultsp,
 562             ("RPC_SUCCESS without reply"));
 563
 564         if (mreq)
 565                 m_freem(mreq);
 566         if (cr->cr_mrep)
 567                 m_freem(cr->cr_mrep);
 568
 569         ct->ct_threads--;
 570         if (ct->ct_closing)
 571                 wakeup(ct);
 572
 573         mtx_unlock(&ct->ct_lock);
 574
 575         if (auth && stat != RPC_SUCCESS)
 576                 AUTH_VALIDATE(auth, xid, NULL, NULL);
 577
 578         free(cr, M_RPC);
 579
 580         return (stat);
 581 }
 582
 583 static void
 584 clnt_vc_geterr(CLIENT *cl, struct rpc_err *errp)
 585 {
 586         struct ct_data *ct = (struct ct_data *) cl->cl_private;
 587
 588         *errp = ct->ct_error;
 589 }
 590
 591 static bool_t
 592 clnt_vc_freeres(CLIENT *cl, xdrproc_t xdr_res, void *res_ptr)
 593 {
 594         XDR xdrs;
 595         bool_t dummy;
 596
 597         xdrs.x_op = XDR_FREE;
 598         dummy = (*xdr_res)(&xdrs, res_ptr);
 599
 600         return (dummy);
 601 }
 602
 603 /*ARGSUSED*/
 604 static void
 605 clnt_vc_abort(CLIENT *cl)
 606 {
 607 }
 608
 609 static bool_t
 610 clnt_vc_control(CLIENT *cl, u_int request, void *info)
 611 {
 612         struct ct_data *ct = (struct ct_data *)cl->cl_private;
 613         void *infop = info;
 614         SVCXPRT *xprt;
 615
 616         mtx_lock(&ct->ct_lock);
 617
 618         switch (request) {
 619         case CLSET_FD_CLOSE:
 620                 ct->ct_closeit = TRUE;
 621                 mtx_unlock(&ct->ct_lock);
 622                 return (TRUE);
 623         case CLSET_FD_NCLOSE:
 624                 ct->ct_closeit = FALSE;
 625                 mtx_unlock(&ct->ct_lock);
 626                 return (TRUE);
 627         default:
 628                 break;
 629         }
 630
 631         /* for other requests which use info */
 632         if (info == NULL) {
 633                 mtx_unlock(&ct->ct_lock);
 634                 return (FALSE);
 635         }
 636         switch (request) {
 637         case CLSET_TIMEOUT:
 638                 if (time_not_ok((struct timeval *)info)) {
 639                         mtx_unlock(&ct->ct_lock);
 640                         return (FALSE);
 641                 }
 642                 ct->ct_wait = *(struct timeval *)infop;
 643                 break;
 644         case CLGET_TIMEOUT:
 645                 *(struct timeval *)infop = ct->ct_wait;
 646                 break;
 647         case CLGET_SERVER_ADDR:
 648                 (void) memcpy(info, &ct->ct_addr, (size_t)ct->ct_addr.ss_len);
 649                 break;
 650         case CLGET_SVC_ADDR:
 651                 /*
 652                  * Slightly different semantics to userland - we use
 653                  * sockaddr instead of netbuf.
 654                  */
 655                 memcpy(info, &ct->ct_addr, ct->ct_addr.ss_len);
 656                 break;
 657         case CLSET_SVC_ADDR:            /* set to new address */
 658                 mtx_unlock(&ct->ct_lock);
 659                 return (FALSE);
 660         case CLGET_XID:
 661                 *(uint32_t *)info = ct->ct_xid;
 662                 break;
 663         case CLSET_XID:
 664                 /* This will set the xid of the NEXT call */
 665                 /* decrement by 1 as clnt_vc_call() increments once */
 666                 ct->ct_xid = *(uint32_t *)info - 1;
 667                 break;
 668         case CLGET_VERS:
 669                 /*
 670                  * This RELIES on the information that, in the call body,
 671                  * the version number field is the fifth field from the
 672                  * begining of the RPC header. MUST be changed if the
 673                  * call_struct is changed
 674                  */
 675                 *(uint32_t *)info =
 676                     ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
 677                     4 * BYTES_PER_XDR_UNIT));
 678                 break;
 679
 680         case CLSET_VERS:
 681                 *(uint32_t *)(void *)(ct->ct_mcallc +
 682                     4 * BYTES_PER_XDR_UNIT) =
 683                     htonl(*(uint32_t *)info);
 684                 break;
 685
 686         case CLGET_PROG:
 687                 /*
 688                  * This RELIES on the information that, in the call body,
 689                  * the program number field is the fourth field from the
 690                  * begining of the RPC header. MUST be changed if the
 691                  * call_struct is changed
 692                  */
 693                 *(uint32_t *)info =
 694                     ntohl(*(uint32_t *)(void *)(ct->ct_mcallc +
 695                     3 * BYTES_PER_XDR_UNIT));
 696                 break;
 697
 698         case CLSET_PROG:
 699                 *(uint32_t *)(void *)(ct->ct_mcallc +
 700                     3 * BYTES_PER_XDR_UNIT) =
 701                     htonl(*(uint32_t *)info);
 702                 break;
 703
 704         case CLSET_WAITCHAN:
 705                 ct->ct_waitchan = (const char *)info;
 706                 break;
 707
 708         case CLGET_WAITCHAN:
 709                 *(const char **) info = ct->ct_waitchan;
 710                 break;
 711
 712         case CLSET_INTERRUPTIBLE:
 713                 if (*(int *) info)
 714                         ct->ct_waitflag = PCATCH;
 715                 else
 716                         ct->ct_waitflag = 0;
 717                 break;
 718
 719         case CLGET_INTERRUPTIBLE:
 720                 if (ct->ct_waitflag)
 721                         *(int *) info = TRUE;
 722                 else
 723                         *(int *) info = FALSE;
 724                 break;
 725
 726         case CLSET_BACKCHANNEL:
 727                 xprt = (SVCXPRT *)info;
 728                 if (ct->ct_backchannelxprt == NULL) {
 729                         xprt->xp_p2 = ct;
 730                         ct->ct_backchannelxprt = xprt;
 731                 }
 732                 break;
 733
 734         default:
 735                 mtx_unlock(&ct->ct_lock);
 736                 return (FALSE);
 737         }
 738
 739         mtx_unlock(&ct->ct_lock);
 740         return (TRUE);
 741 }
 742
 743 static void
 744 clnt_vc_close(CLIENT *cl)
 745 {
 746         struct ct_data *ct = (struct ct_data *) cl->cl_private;
 747         struct ct_request *cr;
 748
 749         mtx_lock(&ct->ct_lock);
 750
 751         if (ct->ct_closed) {
 752                 mtx_unlock(&ct->ct_lock);
 753                 return;
 754         }
 755
 756         if (ct->ct_closing) {
 757                 while (ct->ct_closing)
 758                         msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
 759                 KASSERT(ct->ct_closed, ("client should be closed"));
 760                 mtx_unlock(&ct->ct_lock);
 761                 return;
 762         }
 763
 764         if (ct->ct_socket) {
 765                 ct->ct_closing = TRUE;
 766                 mtx_unlock(&ct->ct_lock);
 767
 768                 SOCKBUF_LOCK(&ct->ct_socket->so_rcv);
 769                 soupcall_clear(ct->ct_socket, SO_RCV);
 770                 clnt_vc_upcallsdone(ct);
 771                 SOCKBUF_UNLOCK(&ct->ct_socket->so_rcv);
 772
 773                 /*
 774                  * Abort any pending requests and wait until everyone
 775                  * has finished with clnt_vc_call.
 776                  */
 777                 mtx_lock(&ct->ct_lock);
 778                 TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
 779                         cr->cr_xid = 0;
 780                         cr->cr_error = ESHUTDOWN;
 781                         wakeup(cr);
 782                 }
 783
 784                 while (ct->ct_threads)
 785                         msleep(ct, &ct->ct_lock, 0, "rpcclose", 0);
 786         }
 787
 788         ct->ct_closing = FALSE;
 789         ct->ct_closed = TRUE;
 790         mtx_unlock(&ct->ct_lock);
 791         wakeup(ct);
 792 }
 793
 794 static void
 795 clnt_vc_destroy(CLIENT *cl)
 796 {
 797         struct ct_data *ct = (struct ct_data *) cl->cl_private;
 798         struct socket *so = NULL;
 799         SVCXPRT *xprt;
 800
 801         clnt_vc_close(cl);
 802
 803         mtx_lock(&ct->ct_lock);
 804         xprt = ct->ct_backchannelxprt;
 805         ct->ct_backchannelxprt = NULL;
 806         if (xprt != NULL) {
 807                 mtx_unlock(&ct->ct_lock);       /* To avoid a LOR. */
 808                 sx_xlock(&xprt->xp_lock);
 809                 mtx_lock(&ct->ct_lock);
 810                 xprt->xp_p2 = NULL;
 811                 sx_xunlock(&xprt->xp_lock);
 812         }
 813
 814         if (ct->ct_socket) {
 815                 if (ct->ct_closeit) {
 816                         so = ct->ct_socket;
 817                 }
 818         }
 819
 820         mtx_unlock(&ct->ct_lock);
 821
 822         mtx_destroy(&ct->ct_lock);
 823         if (so) {
 824                 soshutdown(so, SHUT_WR);
 825                 soclose(so);
 826         }
 827         mem_free(ct, sizeof(struct ct_data));
 828         if (cl->cl_netid && cl->cl_netid[0])
 829                 mem_free(cl->cl_netid, strlen(cl->cl_netid) +1);
 830         if (cl->cl_tp && cl->cl_tp[0])
 831                 mem_free(cl->cl_tp, strlen(cl->cl_tp) +1);
 832         mem_free(cl, sizeof(CLIENT));
 833 }
 834
 835 /*
 836  * Make sure that the time is not garbage.   -1 value is disallowed.
 837  * Note this is different from time_not_ok in clnt_dg.c
 838  */
 839 static bool_t
 840 time_not_ok(struct timeval *t)
 841 {
 842         return (t->tv_sec <= -1 || t->tv_sec > 100000000 ||
 843                 t->tv_usec <= -1 || t->tv_usec > 1000000);
 844 }
 845
 846 int
 847 clnt_vc_soupcall(struct socket *so, void *arg, int waitflag)
 848 {
 849         struct ct_data *ct = (struct ct_data *) arg;
 850         struct uio uio;
 851         struct mbuf *m, *m2;
 852         struct ct_request *cr;
 853         int error, rcvflag, foundreq;
 854         uint32_t xid_plus_direction[2], header;
 855         bool_t do_read;
 856         SVCXPRT *xprt;
 857         struct cf_conn *cd;
 858
 859         CTASSERT(sizeof(xid_plus_direction) == 2 * sizeof(uint32_t));
 860         ct->ct_upcallrefs++;
 861         uio.uio_td = curthread;
 862         do {
 863                 /*
 864                  * If ct_record_resid is zero, we are waiting for a
 865                  * record mark.
 866                  */
 867                 if (ct->ct_record_resid == 0) {
 868
 869                         /*
 870                          * Make sure there is either a whole record
 871                          * mark in the buffer or there is some other
 872                          * error condition
 873                          */
 874                         do_read = FALSE;
 875                         if (so->so_rcv.sb_cc >= sizeof(uint32_t)
 876                             || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
 877                             || so->so_error)
 878                                 do_read = TRUE;
 879
 880                         if (!do_read)
 881                                 break;
 882
 883                         SOCKBUF_UNLOCK(&so->so_rcv);
 884                         uio.uio_resid = sizeof(uint32_t);
 885                         m = NULL;
 886                         rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
 887                         error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
 888                         SOCKBUF_LOCK(&so->so_rcv);
 889
 890                         if (error == EWOULDBLOCK)
 891                                 break;
 892
 893                         /*
 894                          * If there was an error, wake up all pending
 895                          * requests.
 896                          */
 897                         if (error || uio.uio_resid > 0) {
 898                         wakeup_all:
 899                                 mtx_lock(&ct->ct_lock);
 900                                 if (!error) {
 901                                         /*
 902                                          * We must have got EOF trying
 903                                          * to read from the stream.
 904                                          */
 905                                         error = ECONNRESET;
 906                                 }
 907                                 ct->ct_error.re_status = RPC_CANTRECV;
 908                                 ct->ct_error.re_errno = error;
 909                                 TAILQ_FOREACH(cr, &ct->ct_pending, cr_link) {
 910                                         cr->cr_error = error;
 911                                         wakeup(cr);
 912                                 }
 913                                 mtx_unlock(&ct->ct_lock);
 914                                 break;
 915                         }
 916                         m_copydata(m, 0, sizeof(uint32_t), (char *)&header);
 917                         header = ntohl(header);
 918                         ct->ct_record = NULL;
 919                         ct->ct_record_resid = header & 0x7fffffff;
 920                         ct->ct_record_eor = ((header & 0x80000000) != 0);
 921                         m_freem(m);
 922                 } else {
 923                         /*
 924                          * Wait until the socket has the whole record
 925                          * buffered.
 926                          */
 927                         do_read = FALSE;
 928                         if (so->so_rcv.sb_cc >= ct->ct_record_resid
 929                             || (so->so_rcv.sb_state & SBS_CANTRCVMORE)
 930                             || so->so_error)
 931                                 do_read = TRUE;
 932
 933                         if (!do_read)
 934                                 break;
 935
 936                         /*
 937                          * We have the record mark. Read as much as
 938                          * the socket has buffered up to the end of
 939                          * this record.
 940                          */
 941                         SOCKBUF_UNLOCK(&so->so_rcv);
 942                         uio.uio_resid = ct->ct_record_resid;
 943                         m = NULL;
 944                         rcvflag = MSG_DONTWAIT | MSG_SOCALLBCK;
 945                         error = soreceive(so, NULL, &uio, &m, NULL, &rcvflag);
 946                         SOCKBUF_LOCK(&so->so_rcv);
 947
 948                         if (error == EWOULDBLOCK)
 949                                 break;
 950
 951                         if (error || uio.uio_resid == ct->ct_record_resid)
 952                                 goto wakeup_all;
 953
 954                         /*
 955                          * If we have part of the record already,
 956                          * chain this bit onto the end.
 957                          */
 958                         if (ct->ct_record)
 959                                 m_last(ct->ct_record)->m_next = m;
 960                         else
 961                                 ct->ct_record = m;
 962
 963                         ct->ct_record_resid = uio.uio_resid;
 964
 965                         /*
 966                          * If we have the entire record, see if we can
 967                          * match it to a request.
 968                          */
 969                         if (ct->ct_record_resid == 0
 970                             && ct->ct_record_eor) {
 971                                 /*
 972                                  * The XID is in the first uint32_t of
 973                                  * the reply and the message direction
 974                                  * is the second one.
 975                                  */
 976                                 if (ct->ct_record->m_len <
 977                                     sizeof(xid_plus_direction) &&
 978                                     m_length(ct->ct_record, NULL) <
 979                                     sizeof(xid_plus_direction)) {
 980                                         m_freem(ct->ct_record);
 981                                         break;
 982                                 }
 983                                 m_copydata(ct->ct_record, 0,
 984                                     sizeof(xid_plus_direction),
 985                                     (char *)xid_plus_direction);
 986                                 xid_plus_direction[0] =
 987                                     ntohl(xid_plus_direction[0]);
 988                                 xid_plus_direction[1] =
 989                                     ntohl(xid_plus_direction[1]);
 990                                 /* Check message direction. */
 991                                 if (xid_plus_direction[1] == CALL) {
 992                                         /* This is a backchannel request. */
 993                                         mtx_lock(&ct->ct_lock);
 994                                         xprt = ct->ct_backchannelxprt;
 995                                         if (xprt == NULL) {
 996                                                 mtx_unlock(&ct->ct_lock);
 997                                                 /* Just throw it away. */
 998                                                 m_freem(ct->ct_record);
 999                                                 ct->ct_record = NULL;
1000                                         } else {
1001                                                 cd = (struct cf_conn *)
1002                                                     xprt->xp_p1;
1003                                                 m2 = cd->mreq;
1004                                                 /*
1005                                                  * The requests are chained
1006                                                  * in the m_nextpkt list.
1007                                                  */
1008                                                 while (m2 != NULL &&
1009                                                     m2->m_nextpkt != NULL)
1010                                                         /* Find end of list. */
1011                                                         m2 = m2->m_nextpkt;
1012                                                 if (m2 != NULL)
1013                                                         m2->m_nextpkt =
1014                                                             ct->ct_record;
1015                                                 else
1016                                                         cd->mreq =
1017                                                             ct->ct_record;
1018                                                 ct->ct_record->m_nextpkt =
1019                                                     NULL;
1020                                                 ct->ct_record = NULL;
1021                                                 xprt_active(xprt);
1022                                                 mtx_unlock(&ct->ct_lock);
1023                                         }
1024                                 } else {
1025                                         mtx_lock(&ct->ct_lock);
1026                                         foundreq = 0;
1027                                         TAILQ_FOREACH(cr, &ct->ct_pending,
1028                                             cr_link) {
1029                                                 if (cr->cr_xid ==
1030                                                     xid_plus_direction[0]) {
1031                                                         /*
1032                                                          * This one
1033                                                          * matches. We leave
1034                                                          * the reply mbuf in
1035                                                          * cr->cr_mrep. Set
1036                                                          * the XID to zero so
1037                                                          * that we will ignore
1038                                                          * any duplicated
1039                                                          * replies.
1040                                                          */
1041                                                         cr->cr_xid = 0;
1042                                                         cr->cr_mrep =
1043                                                             ct->ct_record;
1044                                                         cr->cr_error = 0;
1045                                                         foundreq = 1;
1046                                                         wakeup(cr);
1047                                                         break;
1048                                                 }
1049                                         }
1050                                         mtx_unlock(&ct->ct_lock);
1051
1052                                         if (!foundreq)
1053                                                 m_freem(ct->ct_record);
1054                                         ct->ct_record = NULL;
1055                                 }
1056                         }
1057                 }
1058         } while (m);
1059         ct->ct_upcallrefs--;
1060         if (ct->ct_upcallrefs < 0)
1061                 panic("rpcvc upcall refcnt");
1062         if (ct->ct_upcallrefs == 0)
1063                 wakeup(&ct->ct_upcallrefs);
1064         return (SU_OK);
1065 }
1066
1067 /*
1068  * Wait for all upcalls in progress to complete.
1069  */
1070 static void
1071 clnt_vc_upcallsdone(struct ct_data *ct)
1072 {
1073
1074         SOCKBUF_LOCK_ASSERT(&ct->ct_socket->so_rcv);
1075
1076         while (ct->ct_upcallrefs > 0)
1077                 (void) msleep(&ct->ct_upcallrefs,
1078                     SOCKBUF_MTX(&ct->ct_socket->so_rcv), 0, "rpcvcup", 0);
1079 }