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