sys/rpc/clnt_dg.c

   1 /*      $NetBSD: clnt_dg.c,v 1.4 2000/07/14 08:40:41 fvdl Exp $ */
   2
   3 /*
   4  * Sun RPC is a product of Sun Microsystems, Inc. and is provided for
   5  * unrestricted use provided that this legend is included on all tape
   6  * media and as a part of the software program in whole or part.  Users
   7  * may copy or modify Sun RPC without charge, but are not authorized
   8  * to license or distribute it to anyone else except as part of a product or
   9  * program developed by the user.
  10  *
  11  * SUN RPC IS PROVIDED AS IS WITH NO WARRANTIES OF ANY KIND INCLUDING THE
  12  * WARRANTIES OF DESIGN, MERCHANTIBILITY AND FITNESS FOR A PARTICULAR
  13  * PURPOSE, OR ARISING FROM A COURSE OF DEALING, USAGE OR TRADE PRACTICE.
  14  *
  15  * Sun RPC is provided with no support and without any obligation on the
  16  * part of Sun Microsystems, Inc. to assist in its use, correction,
  17  * modification or enhancement.
  18  *
  19  * SUN MICROSYSTEMS, INC. SHALL HAVE NO LIABILITY WITH RESPECT TO THE
  20  * INFRINGEMENT OF COPYRIGHTS, TRADE SECRETS OR ANY PATENTS BY SUN RPC
  21  * OR ANY PART THEREOF.
  22  *
  23  * In no event will Sun Microsystems, Inc. be liable for any lost revenue
  24  * or profits or other special, indirect and consequential damages, even if
  25  * Sun has been advised of the possibility of such damages.
  26  *
  27  * Sun Microsystems, Inc.
  28  * 2550 Garcia Avenue
  29  * Mountain View, California  94043
  30  */
  31 /*
  32  * Copyright (c) 1986-1991 by Sun Microsystems Inc.
  33  */
  34
  35 #if defined(LIBC_SCCS) && !defined(lint)
  36 #ident  "@(#)clnt_dg.c  1.23    94/04/22 SMI"
  37 static char sccsid[] = "@(#)clnt_dg.c 1.19 89/03/16 Copyr 1988 Sun Micro";
  38 #endif
  39 #include <sys/cdefs.h>
  40 __FBSDID("$FreeBSD$");
  41
  42 /*
  43  * Implements a connectionless client side RPC.
  44  */
  45
  46 #include <sys/param.h>
  47 #include <sys/systm.h>
  48 #include <sys/kernel.h>
  49 #include <sys/lock.h>
  50 #include <sys/malloc.h>
  51 #include <sys/mbuf.h>
  52 #include <sys/mutex.h>
  53 #include <sys/pcpu.h>
  54 #include <sys/proc.h>
  55 #include <sys/socket.h>
  56 #include <sys/socketvar.h>
  57 #include <sys/time.h>
  58 #include <sys/uio.h>
  59
  60 #include <rpc/rpc.h>
  61 #include <rpc/rpc_com.h>
  62
  63
  64 #ifdef _FREEFALL_CONFIG
  65 /*
  66  * Disable RPC exponential back-off for FreeBSD.org systems.
  67  */
  68 #define RPC_MAX_BACKOFF         1 /* second */
  69 #else
  70 #define RPC_MAX_BACKOFF         30 /* seconds */
  71 #endif
  72
  73 static bool_t time_not_ok(struct timeval *);
  74 static enum clnt_stat clnt_dg_call(CLIENT *, struct rpc_callextra *,
  75     rpcproc_t, struct mbuf *, struct mbuf **, struct timeval);
  76 static void clnt_dg_geterr(CLIENT *, struct rpc_err *);
  77 static bool_t clnt_dg_freeres(CLIENT *, xdrproc_t, void *);
  78 static void clnt_dg_abort(CLIENT *);
  79 static bool_t clnt_dg_control(CLIENT *, u_int, void *);
  80 static void clnt_dg_close(CLIENT *);
  81 static void clnt_dg_destroy(CLIENT *);
  82 static int clnt_dg_soupcall(struct socket *so, void *arg, int waitflag);
  83
  84 static struct clnt_ops clnt_dg_ops = {
  85         .cl_call =      clnt_dg_call,
  86         .cl_abort =     clnt_dg_abort,
  87         .cl_geterr =    clnt_dg_geterr,
  88         .cl_freeres =   clnt_dg_freeres,
  89         .cl_close =     clnt_dg_close,
  90         .cl_destroy =   clnt_dg_destroy,
  91         .cl_control =   clnt_dg_control
  92 };
  93
  94 static const char mem_err_clnt_dg[] = "clnt_dg_create: out of memory";
  95
  96 /*
  97  * A pending RPC request which awaits a reply. Requests which have
  98  * received their reply will have cr_xid set to zero and cr_mrep to
  99  * the mbuf chain of the reply.
 100  */
 101 struct cu_request {
 102         TAILQ_ENTRY(cu_request) cr_link;
 103         CLIENT                  *cr_client;     /* owner */
 104         uint32_t                cr_xid;         /* XID of request */
 105         struct mbuf             *cr_mrep;       /* reply received by upcall */
 106         int                     cr_error;       /* any error from upcall */
 107         char                    cr_verf[MAX_AUTH_BYTES]; /* reply verf */
 108 };
 109
 110 TAILQ_HEAD(cu_request_list, cu_request);
 111
 112 #define MCALL_MSG_SIZE 24
 113
 114 /*
 115  * This structure is pointed to by the socket buffer's sb_upcallarg
 116  * member. It is separate from the client private data to facilitate
 117  * multiple clients sharing the same socket. The cs_lock mutex is used
 118  * to protect all fields of this structure, the socket's receive
 119  * buffer SOCKBUF_LOCK is used to ensure that exactly one of these
 120  * structures is installed on the socket.
 121  */
 122 struct cu_socket {
 123         struct mtx              cs_lock;
 124         int                     cs_refs;        /* Count of clients */
 125         struct cu_request_list  cs_pending;     /* Requests awaiting replies */
 126         int                     cs_upcallrefs;  /* Refcnt of upcalls in prog.*/
 127 };
 128
 129 static void clnt_dg_upcallsdone(struct socket *, struct cu_socket *);
 130
 131 /*
 132  * Private data kept per client handle
 133  */
 134 struct cu_data {
 135         int                     cu_threads;     /* # threads in clnt_vc_call */
 136         bool_t                  cu_closing;     /* TRUE if we are closing */
 137         bool_t                  cu_closed;      /* TRUE if we are closed */
 138         struct socket           *cu_socket;     /* connection socket */
 139         bool_t                  cu_closeit;     /* opened by library */
 140         struct sockaddr_storage cu_raddr;       /* remote address */
 141         int                     cu_rlen;
 142         struct timeval          cu_wait;        /* retransmit interval */
 143         struct timeval          cu_total;       /* total time for the call */
 144         struct rpc_err          cu_error;
 145         uint32_t                cu_xid;
 146         char                    cu_mcallc[MCALL_MSG_SIZE]; /* marshalled callmsg */
 147         size_t                  cu_mcalllen;
 148         size_t                  cu_sendsz;      /* send size */
 149         size_t                  cu_recvsz;      /* recv size */
 150         int                     cu_async;
 151         int                     cu_connect;     /* Use connect(). */
 152         int                     cu_connected;   /* Have done connect(). */
 153         const char              *cu_waitchan;
 154         int                     cu_waitflag;
 155         int                     cu_cwnd;        /* congestion window */
 156         int                     cu_sent;        /* number of in-flight RPCs */
 157         bool_t                  cu_cwnd_wait;
 158 };
 159
 160 #define CWNDSCALE       256
 161 #define MAXCWND         (32 * CWNDSCALE)
 162
 163 /*
 164  * Connection less client creation returns with client handle parameters.
 165  * Default options are set, which the user can change using clnt_control().
 166  * fd should be open and bound.
 167  * NB: The rpch->cl_auth is initialized to null authentication.
 168  *      Caller may wish to set this something more useful.
 169  *
 170  * sendsz and recvsz are the maximum allowable packet sizes that can be
 171  * sent and received. Normally they are the same, but they can be
 172  * changed to improve the program efficiency and buffer allocation.
 173  * If they are 0, use the transport default.
 174  *
 175  * If svcaddr is NULL, returns NULL.
 176  */
 177 CLIENT *
 178 clnt_dg_create(
 179         struct socket *so,
 180         struct sockaddr *svcaddr,       /* servers address */
 181         rpcprog_t program,              /* program number */
 182         rpcvers_t version,              /* version number */
 183         size_t sendsz,                  /* buffer recv size */
 184         size_t recvsz)                  /* buffer send size */
 185 {
 186         CLIENT *cl = NULL;              /* client handle */
 187         struct cu_data *cu = NULL;      /* private data */
 188         struct cu_socket *cs = NULL;
 189         struct sockbuf *sb;
 190         struct timeval now;
 191         struct rpc_msg call_msg;
 192         struct __rpc_sockinfo si;
 193         XDR xdrs;
 194
 195         if (svcaddr == NULL) {
 196                 rpc_createerr.cf_stat = RPC_UNKNOWNADDR;
 197                 return (NULL);
 198         }
 199
 200         if (!__rpc_socket2sockinfo(so, &si)) {
 201                 rpc_createerr.cf_stat = RPC_TLIERROR;
 202                 rpc_createerr.cf_error.re_errno = 0;
 203                 return (NULL);
 204         }
 205
 206         /*
 207          * Find the receive and the send size
 208          */
 209         sendsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)sendsz);
 210         recvsz = __rpc_get_t_size(si.si_af, si.si_proto, (int)recvsz);
 211         if ((sendsz == 0) || (recvsz == 0)) {
 212                 rpc_createerr.cf_stat = RPC_TLIERROR; /* XXX */
 213                 rpc_createerr.cf_error.re_errno = 0;
 214                 return (NULL);
 215         }
 216
 217         cl = mem_alloc(sizeof (CLIENT));
 218
 219         /*
 220          * Should be multiple of 4 for XDR.
 221          */
 222         sendsz = ((sendsz + 3) / 4) * 4;
 223         recvsz = ((recvsz + 3) / 4) * 4;
 224         cu = mem_alloc(sizeof (*cu));
 225         cu->cu_threads = 0;
 226         cu->cu_closing = FALSE;
 227         cu->cu_closed = FALSE;
 228         (void) memcpy(&cu->cu_raddr, svcaddr, (size_t)svcaddr->sa_len);
 229         cu->cu_rlen = svcaddr->sa_len;
 230         /* Other values can also be set through clnt_control() */
 231         cu->cu_wait.tv_sec = 3; /* heuristically chosen */
 232         cu->cu_wait.tv_usec = 0;
 233         cu->cu_total.tv_sec = -1;
 234         cu->cu_total.tv_usec = -1;
 235         cu->cu_sendsz = sendsz;
 236         cu->cu_recvsz = recvsz;
 237         cu->cu_async = FALSE;
 238         cu->cu_connect = FALSE;
 239         cu->cu_connected = FALSE;
 240         cu->cu_waitchan = "rpcrecv";
 241         cu->cu_waitflag = 0;
 242         cu->cu_cwnd = MAXCWND / 2;
 243         cu->cu_sent = 0;
 244         cu->cu_cwnd_wait = FALSE;
 245         (void) getmicrotime(&now);
 246         cu->cu_xid = __RPC_GETXID(&now);
 247         call_msg.rm_xid = cu->cu_xid;
 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         soreserve(so, 256*1024, 256*1024);
 266
 267         sb = &so->so_rcv;
 268         SOCKBUF_LOCK(&so->so_rcv);
 269 recheck_socket:
 270         if (sb->sb_upcall) {
 271                 if (sb->sb_upcall != clnt_dg_soupcall) {
 272                         SOCKBUF_UNLOCK(&so->so_rcv);
 273                         printf("clnt_dg_create(): socket already has an incompatible upcall\n");
 274                         goto err2;
 275                 }
 276                 cs = (struct cu_socket *) sb->sb_upcallarg;
 277                 mtx_lock(&cs->cs_lock);
 278                 cs->cs_refs++;
 279                 mtx_unlock(&cs->cs_lock);
 280         } else {
 281                 /*
 282                  * We are the first on this socket - allocate the
 283                  * structure and install it in the socket.
 284                  */
 285                 SOCKBUF_UNLOCK(&so->so_rcv);
 286                 cs = mem_alloc(sizeof(*cs));
 287                 SOCKBUF_LOCK(&so->so_rcv);
 288                 if (sb->sb_upcall) {
 289                         /*
 290                          * We have lost a race with some other client.
 291                          */
 292                         mem_free(cs, sizeof(*cs));
 293                         goto recheck_socket;
 294                 }
 295                 mtx_init(&cs->cs_lock, "cs->cs_lock", NULL, MTX_DEF);
 296                 cs->cs_refs = 1;
 297                 cs->cs_upcallrefs = 0;
 298                 TAILQ_INIT(&cs->cs_pending);
 299                 soupcall_set(so, SO_RCV, clnt_dg_soupcall, cs);
 300         }
 301         SOCKBUF_UNLOCK(&so->so_rcv);
 302
 303         cl->cl_refs = 1;
 304         cl->cl_ops = &clnt_dg_ops;
 305         cl->cl_private = (caddr_t)(void *)cu;
 306         cl->cl_auth = authnone_create();
 307         cl->cl_tp = NULL;
 308         cl->cl_netid = NULL;
 309         return (cl);
 310 err2:
 311         if (cl) {
 312                 mem_free(cl, sizeof (CLIENT));
 313                 if (cu)
 314                         mem_free(cu, sizeof (*cu));
 315         }
 316         return (NULL);
 317 }
 318
 319 static enum clnt_stat
 320 clnt_dg_call(
 321         CLIENT          *cl,            /* client handle */
 322         struct rpc_callextra *ext,      /* call metadata */
 323         rpcproc_t       proc,           /* procedure number */
 324         struct mbuf     *args,          /* pointer to args */
 325         struct mbuf     **resultsp,     /* pointer to results */
 326         struct timeval  utimeout)       /* seconds to wait before giving up */
 327 {
 328         struct cu_data *cu = (struct cu_data *)cl->cl_private;
 329         struct cu_socket *cs;
 330         struct rpc_timers *rt;
 331         AUTH *auth;
 332         struct rpc_err *errp;
 333         enum clnt_stat stat;
 334         XDR xdrs;
 335         struct rpc_msg reply_msg;
 336         bool_t ok;
 337         int retrans;                    /* number of re-transmits so far */
 338         int nrefreshes = 2;             /* number of times to refresh cred */
 339         struct timeval *tvp;
 340         int timeout;
 341         int retransmit_time;
 342         int next_sendtime, starttime, rtt, time_waited, tv = 0;
 343         struct sockaddr *sa;
 344         socklen_t salen;
 345         uint32_t xid = 0;
 346         struct mbuf *mreq = NULL, *results;
 347         struct cu_request *cr;
 348         int error;
 349
 350         cs = cu->cu_socket->so_rcv.sb_upcallarg;
 351         cr = malloc(sizeof(struct cu_request), M_RPC, M_WAITOK);
 352
 353         mtx_lock(&cs->cs_lock);
 354
 355         if (cu->cu_closing || cu->cu_closed) {
 356                 mtx_unlock(&cs->cs_lock);
 357                 free(cr, M_RPC);
 358                 return (RPC_CANTSEND);
 359         }
 360         cu->cu_threads++;
 361
 362         if (ext) {
 363                 auth = ext->rc_auth;
 364                 errp = &ext->rc_err;
 365         } else {
 366                 auth = cl->cl_auth;
 367                 errp = &cu->cu_error;
 368         }
 369
 370         cr->cr_client = cl;
 371         cr->cr_mrep = NULL;
 372         cr->cr_error = 0;
 373
 374         if (cu->cu_total.tv_usec == -1) {
 375                 tvp = &utimeout; /* use supplied timeout */
 376         } else {
 377                 tvp = &cu->cu_total; /* use default timeout */
 378         }
 379         if (tvp->tv_sec || tvp->tv_usec)
 380                 timeout = tvtohz(tvp);
 381         else
 382                 timeout = 0;
 383
 384         if (cu->cu_connect && !cu->cu_connected) {
 385                 mtx_unlock(&cs->cs_lock);
 386                 error = soconnect(cu->cu_socket,
 387                     (struct sockaddr *)&cu->cu_raddr, curthread);
 388                 mtx_lock(&cs->cs_lock);
 389                 if (error) {
 390                         errp->re_errno = error;
 391                         errp->re_status = stat = RPC_CANTSEND;
 392                         goto out;
 393                 }
 394                 cu->cu_connected = 1;
 395         }
 396         if (cu->cu_connected) {
 397                 sa = NULL;
 398                 salen = 0;
 399         } else {
 400                 sa = (struct sockaddr *)&cu->cu_raddr;
 401                 salen = cu->cu_rlen;
 402         }
 403         time_waited = 0;
 404         retrans = 0;
 405         if (ext && ext->rc_timers) {
 406                 rt = ext->rc_timers;
 407                 if (!rt->rt_rtxcur)
 408                         rt->rt_rtxcur = tvtohz(&cu->cu_wait);
 409                 retransmit_time = next_sendtime = rt->rt_rtxcur;
 410         } else {
 411                 rt = NULL;
 412                 retransmit_time = next_sendtime = tvtohz(&cu->cu_wait);
 413         }
 414
 415         starttime = ticks;
 416
 417 call_again:
 418         mtx_assert(&cs->cs_lock, MA_OWNED);
 419
 420         cu->cu_xid++;
 421         xid = cu->cu_xid;
 422
 423 send_again:
 424         mtx_unlock(&cs->cs_lock);
 425
 426         MGETHDR(mreq, M_WAIT, MT_DATA);
 427         KASSERT(cu->cu_mcalllen <= MHLEN, ("RPC header too big"));
 428         bcopy(cu->cu_mcallc, mreq->m_data, cu->cu_mcalllen);
 429         mreq->m_len = cu->cu_mcalllen;
 430
 431         /*
 432          * The XID is the first thing in the request.
 433          */
 434         *mtod(mreq, uint32_t *) = htonl(xid);
 435
 436         xdrmbuf_create(&xdrs, mreq, XDR_ENCODE);
 437
 438         if (cu->cu_async == TRUE && args == NULL)
 439                 goto get_reply;
 440
 441         if ((! XDR_PUTINT32(&xdrs, &proc)) ||
 442             (! AUTH_MARSHALL(auth, xid, &xdrs,
 443                 m_copym(args, 0, M_COPYALL, M_WAITOK)))) {
 444                 errp->re_status = stat = RPC_CANTENCODEARGS;
 445                 mtx_lock(&cs->cs_lock);
 446                 goto out;
 447         }
 448         mreq->m_pkthdr.len = m_length(mreq, NULL);
 449
 450         cr->cr_xid = xid;
 451         mtx_lock(&cs->cs_lock);
 452
 453         /*
 454          * Try to get a place in the congestion window.
 455          */
 456         while (cu->cu_sent >= cu->cu_cwnd) {
 457                 cu->cu_cwnd_wait = TRUE;
 458                 error = msleep(&cu->cu_cwnd_wait, &cs->cs_lock,
 459                     cu->cu_waitflag, "rpccwnd", 0);
 460                 if (error) {
 461                         errp->re_errno = error;
 462                         errp->re_status = stat = RPC_CANTSEND;
 463                         goto out;
 464                 }
 465         }
 466         cu->cu_sent += CWNDSCALE;
 467
 468         TAILQ_INSERT_TAIL(&cs->cs_pending, cr, cr_link);
 469         mtx_unlock(&cs->cs_lock);
 470
 471         /*
 472          * sosend consumes mreq.
 473          */
 474         error = sosend(cu->cu_socket, sa, NULL, mreq, NULL, 0, curthread);
 475         mreq = NULL;
 476
 477         /*
 478          * sub-optimal code appears here because we have
 479          * some clock time to spare while the packets are in flight.
 480          * (We assume that this is actually only executed once.)
 481          */
 482         reply_msg.acpted_rply.ar_verf.oa_flavor = AUTH_NULL;
 483         reply_msg.acpted_rply.ar_verf.oa_base = cr->cr_verf;
 484         reply_msg.acpted_rply.ar_verf.oa_length = 0;
 485         reply_msg.acpted_rply.ar_results.where = NULL;
 486         reply_msg.acpted_rply.ar_results.proc = (xdrproc_t)xdr_void;
 487
 488         mtx_lock(&cs->cs_lock);
 489         if (error) {
 490                 TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
 491                 errp->re_errno = error;
 492                 errp->re_status = stat = RPC_CANTSEND;
 493                 cu->cu_sent -= CWNDSCALE;
 494                 if (cu->cu_cwnd_wait) {
 495                         cu->cu_cwnd_wait = FALSE;
 496                         wakeup(&cu->cu_cwnd_wait);
 497                 }
 498                 goto out;
 499         }
 500
 501         /*
 502          * Check to see if we got an upcall while waiting for the
 503          * lock.
 504          */
 505         if (cr->cr_error) {
 506                 TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
 507                 errp->re_errno = cr->cr_error;
 508                 errp->re_status = stat = RPC_CANTRECV;
 509                 cu->cu_sent -= CWNDSCALE;
 510                 if (cu->cu_cwnd_wait) {
 511                         cu->cu_cwnd_wait = FALSE;
 512                         wakeup(&cu->cu_cwnd_wait);
 513                 }
 514                 goto out;
 515         }
 516         if (cr->cr_mrep) {
 517                 TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
 518                 cu->cu_sent -= CWNDSCALE;
 519                 if (cu->cu_cwnd_wait) {
 520                         cu->cu_cwnd_wait = FALSE;
 521                         wakeup(&cu->cu_cwnd_wait);
 522                 }
 523                 goto got_reply;
 524         }
 525
 526         /*
 527          * Hack to provide rpc-based message passing
 528          */
 529         if (timeout == 0) {
 530                 TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
 531                 errp->re_status = stat = RPC_TIMEDOUT;
 532                 cu->cu_sent -= CWNDSCALE;
 533                 if (cu->cu_cwnd_wait) {
 534                         cu->cu_cwnd_wait = FALSE;
 535                         wakeup(&cu->cu_cwnd_wait);
 536                 }
 537                 goto out;
 538         }
 539
 540 get_reply:
 541         for (;;) {
 542                 /* Decide how long to wait. */
 543                 if (next_sendtime < timeout)
 544                         tv = next_sendtime;
 545                 else
 546                         tv = timeout;
 547                 tv -= time_waited;
 548
 549                 if (tv > 0) {
 550                         if (cu->cu_closing || cu->cu_closed) {
 551                                 error = 0;
 552                                 cr->cr_error = ESHUTDOWN;
 553                         } else {
 554                                 error = msleep(cr, &cs->cs_lock,
 555                                     cu->cu_waitflag, cu->cu_waitchan, tv);
 556                         }
 557                 } else {
 558                         error = EWOULDBLOCK;
 559                 }
 560
 561                 TAILQ_REMOVE(&cs->cs_pending, cr, cr_link);
 562                 cu->cu_sent -= CWNDSCALE;
 563                 if (cu->cu_cwnd_wait) {
 564                         cu->cu_cwnd_wait = FALSE;
 565                         wakeup(&cu->cu_cwnd_wait);
 566                 }
 567
 568                 if (!error) {
 569                         /*
 570                          * We were woken up by the upcall.  If the
 571                          * upcall had a receive error, report that,
 572                          * otherwise we have a reply.
 573                          */
 574                         if (cr->cr_error) {
 575                                 errp->re_errno = cr->cr_error;
 576                                 errp->re_status = stat = RPC_CANTRECV;
 577                                 goto out;
 578                         }
 579
 580                         cu->cu_cwnd += (CWNDSCALE * CWNDSCALE
 581                             + cu->cu_cwnd / 2) / cu->cu_cwnd;
 582                         if (cu->cu_cwnd > MAXCWND)
 583                                 cu->cu_cwnd = MAXCWND;
 584
 585                         if (rt) {
 586                                 /*
 587                                  * Add one to the time since a tick
 588                                  * count of N means that the actual
 589                                  * time taken was somewhere between N
 590                                  * and N+1.
 591                                  */
 592                                 rtt = ticks - starttime + 1;
 593
 594                                 /*
 595                                  * Update our estimate of the round
 596                                  * trip time using roughly the
 597                                  * algorithm described in RFC
 598                                  * 2988. Given an RTT sample R:
 599                                  *
 600                                  * RTTVAR = (1-beta) * RTTVAR + beta * |SRTT-R|
 601                                  * SRTT = (1-alpha) * SRTT + alpha * R
 602                                  *
 603                                  * where alpha = 0.125 and beta = 0.25.
 604                                  *
 605                                  * The initial retransmit timeout is
 606                                  * SRTT + 4*RTTVAR and doubles on each
 607                                  * retransmision.
 608                                  */
 609                                 if (rt->rt_srtt == 0) {
 610                                         rt->rt_srtt = rtt;
 611                                         rt->rt_deviate = rtt / 2;
 612                                 } else {
 613                                         int32_t error = rtt - rt->rt_srtt;
 614                                         rt->rt_srtt += error / 8;
 615                                         error = abs(error) - rt->rt_deviate;
 616                                         rt->rt_deviate += error / 4;
 617                                 }
 618                                 rt->rt_rtxcur = rt->rt_srtt + 4*rt->rt_deviate;
 619                         }
 620
 621                         break;
 622                 }
 623
 624                 /*
 625                  * The sleep returned an error so our request is still
 626                  * on the list. If we got EWOULDBLOCK, we may want to
 627                  * re-send the request.
 628                  */
 629                 if (error != EWOULDBLOCK) {
 630                         errp->re_errno = error;
 631                         if (error == EINTR)
 632                                 errp->re_status = stat = RPC_INTR;
 633                         else
 634                                 errp->re_status = stat = RPC_CANTRECV;
 635                         goto out;
 636                 }
 637
 638                 time_waited = ticks - starttime;
 639
 640                 /* Check for timeout. */
 641                 if (time_waited > timeout) {
 642                         errp->re_errno = EWOULDBLOCK;
 643                         errp->re_status = stat = RPC_TIMEDOUT;
 644                         goto out;
 645                 }
 646
 647                 /* Retransmit if necessary. */
 648                 if (time_waited >= next_sendtime) {
 649                         cu->cu_cwnd /= 2;
 650                         if (cu->cu_cwnd < CWNDSCALE)
 651                                 cu->cu_cwnd = CWNDSCALE;
 652                         if (ext && ext->rc_feedback) {
 653                                 mtx_unlock(&cs->cs_lock);
 654                                 if (retrans == 0)
 655                                         ext->rc_feedback(FEEDBACK_REXMIT1,
 656                                             proc, ext->rc_feedback_arg);
 657                                 else
 658                                         ext->rc_feedback(FEEDBACK_REXMIT2,
 659                                             proc, ext->rc_feedback_arg);
 660                                 mtx_lock(&cs->cs_lock);
 661                         }
 662                         if (cu->cu_closing || cu->cu_closed) {
 663                                 errp->re_errno = ESHUTDOWN;
 664                                 errp->re_status = stat = RPC_CANTRECV;
 665                                 goto out;
 666                         }
 667                         retrans++;
 668                         /* update retransmit_time */
 669                         if (retransmit_time < RPC_MAX_BACKOFF * hz)
 670                                 retransmit_time = 2 * retransmit_time;
 671                         next_sendtime += retransmit_time;
 672                         goto send_again;
 673                 }
 674                 TAILQ_INSERT_TAIL(&cs->cs_pending, cr, cr_link);
 675         }
 676
 677 got_reply:
 678         /*
 679          * Now decode and validate the response. We need to drop the
 680          * lock since xdr_replymsg may end up sleeping in malloc.
 681          */
 682         mtx_unlock(&cs->cs_lock);
 683
 684         if (ext && ext->rc_feedback)
 685                 ext->rc_feedback(FEEDBACK_OK, proc, ext->rc_feedback_arg);
 686
 687         xdrmbuf_create(&xdrs, cr->cr_mrep, XDR_DECODE);
 688         ok = xdr_replymsg(&xdrs, &reply_msg);
 689         cr->cr_mrep = NULL;
 690
 691         if (ok) {
 692                 if ((reply_msg.rm_reply.rp_stat == MSG_ACCEPTED) &&
 693                     (reply_msg.acpted_rply.ar_stat == SUCCESS))
 694                         errp->re_status = stat = RPC_SUCCESS;
 695                 else
 696                         stat = _seterr_reply(&reply_msg, &(cu->cu_error));
 697
 698                 if (errp->re_status == RPC_SUCCESS) {
 699                         results = xdrmbuf_getall(&xdrs);
 700                         if (! AUTH_VALIDATE(auth, xid,
 701                                 &reply_msg.acpted_rply.ar_verf,
 702                                 &results)) {
 703                                 errp->re_status = stat = RPC_AUTHERROR;
 704                                 errp->re_why = AUTH_INVALIDRESP;
 705                                 if (retrans &&
 706                                     auth->ah_cred.oa_flavor == RPCSEC_GSS) {
 707                                         /*
 708                                          * If we retransmitted, its
 709                                          * possible that we will
 710                                          * receive a reply for one of
 711                                          * the earlier transmissions
 712                                          * (which will use an older
 713                                          * RPCSEC_GSS sequence
 714                                          * number). In this case, just
 715                                          * go back and listen for a
 716                                          * new reply. We could keep a
 717                                          * record of all the seq
 718                                          * numbers we have transmitted
 719                                          * so far so that we could
 720                                          * accept a reply for any of
 721                                          * them here.
 722                                          */
 723                                         XDR_DESTROY(&xdrs);
 724                                         mtx_lock(&cs->cs_lock);
 725                                         TAILQ_INSERT_TAIL(&cs->cs_pending,
 726                                             cr, cr_link);
 727                                         cr->cr_mrep = NULL;
 728                                         goto get_reply;
 729                                 }
 730                         } else {
 731                                 *resultsp = results;
 732                         }
 733                 }               /* end successful completion */
 734                 /*
 735                  * If unsuccesful AND error is an authentication error
 736                  * then refresh credentials and try again, else break
 737                  */
 738                 else if (stat == RPC_AUTHERROR)
 739                         /* maybe our credentials need to be refreshed ... */
 740                         if (nrefreshes > 0 &&
 741                             AUTH_REFRESH(auth, &reply_msg)) {
 742                                 nrefreshes--;
 743                                 XDR_DESTROY(&xdrs);
 744                                 mtx_lock(&cs->cs_lock);
 745                                 goto call_again;
 746                         }
 747                 /* end of unsuccessful completion */
 748         }       /* end of valid reply message */
 749         else {
 750                 errp->re_status = stat = RPC_CANTDECODERES;
 751
 752         }
 753         XDR_DESTROY(&xdrs);
 754         mtx_lock(&cs->cs_lock);
 755 out:
 756         mtx_assert(&cs->cs_lock, MA_OWNED);
 757
 758         if (mreq)
 759                 m_freem(mreq);
 760         if (cr->cr_mrep)
 761                 m_freem(cr->cr_mrep);
 762
 763         cu->cu_threads--;
 764         if (cu->cu_closing)
 765                 wakeup(cu);
 766
 767         mtx_unlock(&cs->cs_lock);
 768
 769         if (auth && stat != RPC_SUCCESS)
 770                 AUTH_VALIDATE(auth, xid, NULL, NULL);
 771
 772         free(cr, M_RPC);
 773
 774         return (stat);
 775 }
 776
 777 static void
 778 clnt_dg_geterr(CLIENT *cl, struct rpc_err *errp)
 779 {
 780         struct cu_data *cu = (struct cu_data *)cl->cl_private;
 781
 782         *errp = cu->cu_error;
 783 }
 784
 785 static bool_t
 786 clnt_dg_freeres(CLIENT *cl, xdrproc_t xdr_res, void *res_ptr)
 787 {
 788         XDR xdrs;
 789         bool_t dummy;
 790
 791         xdrs.x_op = XDR_FREE;
 792         dummy = (*xdr_res)(&xdrs, res_ptr);
 793
 794         return (dummy);
 795 }
 796
 797 /*ARGSUSED*/
 798 static void
 799 clnt_dg_abort(CLIENT *h)
 800 {
 801 }
 802
 803 static bool_t
 804 clnt_dg_control(CLIENT *cl, u_int request, void *info)
 805 {
 806         struct cu_data *cu = (struct cu_data *)cl->cl_private;
 807         struct cu_socket *cs;
 808         struct sockaddr *addr;
 809
 810         cs = cu->cu_socket->so_rcv.sb_upcallarg;
 811         mtx_lock(&cs->cs_lock);
 812
 813         switch (request) {
 814         case CLSET_FD_CLOSE:
 815                 cu->cu_closeit = TRUE;
 816                 mtx_unlock(&cs->cs_lock);
 817                 return (TRUE);
 818         case CLSET_FD_NCLOSE:
 819                 cu->cu_closeit = FALSE;
 820                 mtx_unlock(&cs->cs_lock);
 821                 return (TRUE);
 822         }
 823
 824         /* for other requests which use info */
 825         if (info == NULL) {
 826                 mtx_unlock(&cs->cs_lock);
 827                 return (FALSE);
 828         }
 829         switch (request) {
 830         case CLSET_TIMEOUT:
 831                 if (time_not_ok((struct timeval *)info)) {
 832                         mtx_unlock(&cs->cs_lock);
 833                         return (FALSE);
 834                 }
 835                 cu->cu_total = *(struct timeval *)info;
 836                 break;
 837         case CLGET_TIMEOUT:
 838                 *(struct timeval *)info = cu->cu_total;
 839                 break;
 840         case CLSET_RETRY_TIMEOUT:
 841                 if (time_not_ok((struct timeval *)info)) {
 842                         mtx_unlock(&cs->cs_lock);
 843                         return (FALSE);
 844                 }
 845                 cu->cu_wait = *(struct timeval *)info;
 846                 break;
 847         case CLGET_RETRY_TIMEOUT:
 848                 *(struct timeval *)info = cu->cu_wait;
 849                 break;
 850         case CLGET_SVC_ADDR:
 851                 /*
 852                  * Slightly different semantics to userland - we use
 853                  * sockaddr instead of netbuf.
 854                  */
 855                 memcpy(info, &cu->cu_raddr, cu->cu_raddr.ss_len);
 856                 break;
 857         case CLSET_SVC_ADDR:            /* set to new address */
 858                 addr = (struct sockaddr *)info;
 859                 (void) memcpy(&cu->cu_raddr, addr, addr->sa_len);
 860                 break;
 861         case CLGET_XID:
 862                 *(uint32_t *)info = cu->cu_xid;
 863                 break;
 864
 865         case CLSET_XID:
 866                 /* This will set the xid of the NEXT call */
 867                 /* decrement by 1 as clnt_dg_call() increments once */
 868                 cu->cu_xid = *(uint32_t *)info - 1;
 869                 break;
 870
 871         case CLGET_VERS:
 872                 /*
 873                  * This RELIES on the information that, in the call body,
 874                  * the version number field is the fifth field from the
 875                  * begining of the RPC header. MUST be changed if the
 876                  * call_struct is changed
 877                  */
 878                 *(uint32_t *)info =
 879                     ntohl(*(uint32_t *)(void *)(cu->cu_mcallc +
 880                     4 * BYTES_PER_XDR_UNIT));
 881                 break;
 882
 883         case CLSET_VERS:
 884                 *(uint32_t *)(void *)(cu->cu_mcallc + 4 * BYTES_PER_XDR_UNIT)
 885                         = htonl(*(uint32_t *)info);
 886                 break;
 887
 888         case CLGET_PROG:
 889                 /*
 890                  * This RELIES on the information that, in the call body,
 891                  * the program number field is the fourth field from the
 892                  * begining of the RPC header. MUST be changed if the
 893                  * call_struct is changed
 894                  */
 895                 *(uint32_t *)info =
 896                     ntohl(*(uint32_t *)(void *)(cu->cu_mcallc +
 897                     3 * BYTES_PER_XDR_UNIT));
 898                 break;
 899
 900         case CLSET_PROG:
 901                 *(uint32_t *)(void *)(cu->cu_mcallc + 3 * BYTES_PER_XDR_UNIT)
 902                         = htonl(*(uint32_t *)info);
 903                 break;
 904         case CLSET_ASYNC:
 905                 cu->cu_async = *(int *)info;
 906                 break;
 907         case CLSET_CONNECT:
 908                 cu->cu_connect = *(int *)info;
 909                 break;
 910         case CLSET_WAITCHAN:
 911                 cu->cu_waitchan = (const char *)info;
 912                 break;
 913         case CLGET_WAITCHAN:
 914                 *(const char **) info = cu->cu_waitchan;
 915                 break;
 916         case CLSET_INTERRUPTIBLE:
 917                 if (*(int *) info)
 918                         cu->cu_waitflag = PCATCH;
 919                 else
 920                         cu->cu_waitflag = 0;
 921                 break;
 922         case CLGET_INTERRUPTIBLE:
 923                 if (cu->cu_waitflag)
 924                         *(int *) info = TRUE;
 925                 else
 926                         *(int *) info = FALSE;
 927                 break;
 928         default:
 929                 mtx_unlock(&cs->cs_lock);
 930                 return (FALSE);
 931         }
 932         mtx_unlock(&cs->cs_lock);
 933         return (TRUE);
 934 }
 935
 936 static void
 937 clnt_dg_close(CLIENT *cl)
 938 {
 939         struct cu_data *cu = (struct cu_data *)cl->cl_private;
 940         struct cu_socket *cs;
 941         struct cu_request *cr;
 942
 943         cs = cu->cu_socket->so_rcv.sb_upcallarg;
 944         mtx_lock(&cs->cs_lock);
 945
 946         if (cu->cu_closed) {
 947                 mtx_unlock(&cs->cs_lock);
 948                 return;
 949         }
 950
 951         if (cu->cu_closing) {
 952                 while (cu->cu_closing)
 953                         msleep(cu, &cs->cs_lock, 0, "rpcclose", 0);
 954                 KASSERT(cu->cu_closed, ("client should be closed"));
 955                 mtx_unlock(&cs->cs_lock);
 956                 return;
 957         }
 958
 959         /*
 960          * Abort any pending requests and wait until everyone
 961          * has finished with clnt_vc_call.
 962          */
 963         cu->cu_closing = TRUE;
 964         TAILQ_FOREACH(cr, &cs->cs_pending, cr_link) {
 965                 if (cr->cr_client == cl) {
 966                         cr->cr_xid = 0;
 967                         cr->cr_error = ESHUTDOWN;
 968                         wakeup(cr);
 969                 }
 970         }
 971
 972         while (cu->cu_threads)
 973                 msleep(cu, &cs->cs_lock, 0, "rpcclose", 0);
 974
 975         cu->cu_closing = FALSE;
 976         cu->cu_closed = TRUE;
 977
 978         mtx_unlock(&cs->cs_lock);
 979         wakeup(cu);
 980 }
 981
 982 static void
 983 clnt_dg_destroy(CLIENT *cl)
 984 {
 985         struct cu_data *cu = (struct cu_data *)cl->cl_private;
 986         struct cu_socket *cs;
 987         struct socket *so = NULL;
 988         bool_t lastsocketref;
 989
 990         cs = cu->cu_socket->so_rcv.sb_upcallarg;
 991         clnt_dg_close(cl);
 992
 993         mtx_lock(&cs->cs_lock);
 994
 995         cs->cs_refs--;
 996         if (cs->cs_refs == 0) {
 997                 mtx_unlock(&cs->cs_lock);
 998                 SOCKBUF_LOCK(&cu->cu_socket->so_rcv);
 999                 soupcall_clear(cu->cu_socket, SO_RCV);
1000                 clnt_dg_upcallsdone(cu->cu_socket, cs);
1001                 SOCKBUF_UNLOCK(&cu->cu_socket->so_rcv);
1002                 mtx_destroy(&cs->cs_lock);
1003                 mem_free(cs, sizeof(*cs));
1004                 lastsocketref = TRUE;
1005         } else {
1006                 mtx_unlock(&cs->cs_lock);
1007                 lastsocketref = FALSE;
1008         }
1009
1010         if (cu->cu_closeit && lastsocketref) {
1011                 so = cu->cu_socket;
1012                 cu->cu_socket = NULL;
1013         }
1014
1015         if (so)
1016                 soclose(so);
1017
1018         if (cl->cl_netid && cl->cl_netid[0])
1019                 mem_free(cl->cl_netid, strlen(cl->cl_netid) +1);
1020         if (cl->cl_tp && cl->cl_tp[0])
1021                 mem_free(cl->cl_tp, strlen(cl->cl_tp) +1);
1022         mem_free(cu, sizeof (*cu));
1023         mem_free(cl, sizeof (CLIENT));
1024 }
1025
1026 /*
1027  * Make sure that the time is not garbage.  -1 value is allowed.
1028  */
1029 static bool_t
1030 time_not_ok(struct timeval *t)
1031 {
1032         return (t->tv_sec < -1 || t->tv_sec > 100000000 ||
1033                 t->tv_usec < -1 || t->tv_usec > 1000000);
1034 }
1035
1036 int
1037 clnt_dg_soupcall(struct socket *so, void *arg, int waitflag)
1038 {
1039         struct cu_socket *cs = (struct cu_socket *) arg;
1040         struct uio uio;
1041         struct mbuf *m;
1042         struct mbuf *control;
1043         struct cu_request *cr;
1044         int error, rcvflag, foundreq;
1045         uint32_t xid;
1046
1047         cs->cs_upcallrefs++;
1048         uio.uio_resid = 1000000000;
1049         uio.uio_td = curthread;
1050         do {
1051                 SOCKBUF_UNLOCK(&so->so_rcv);
1052                 m = NULL;
1053                 control = NULL;
1054                 rcvflag = MSG_DONTWAIT;
1055                 error = soreceive(so, NULL, &uio, &m, &control, &rcvflag);
1056                 if (control)
1057                         m_freem(control);
1058                 SOCKBUF_LOCK(&so->so_rcv);
1059
1060                 if (error == EWOULDBLOCK)
1061                         break;
1062
1063                 /*
1064                  * If there was an error, wake up all pending
1065                  * requests.
1066                  */
1067                 if (error) {
1068                         mtx_lock(&cs->cs_lock);
1069                         TAILQ_FOREACH(cr, &cs->cs_pending, cr_link) {
1070                                 cr->cr_xid = 0;
1071                                 cr->cr_error = error;
1072                                 wakeup(cr);
1073                         }
1074                         mtx_unlock(&cs->cs_lock);
1075                         break;
1076                 }
1077
1078                 /*
1079                  * The XID is in the first uint32_t of the reply.
1080                  */
1081                 if (m->m_len < sizeof(xid))
1082                         m = m_pullup(m, sizeof(xid));
1083                 if (!m)
1084                         /*
1085                          * Should never happen.
1086                          */
1087                         continue;
1088
1089                 xid = ntohl(*mtod(m, uint32_t *));
1090
1091                 /*
1092                  * Attempt to match this reply with a pending request.
1093                  */
1094                 mtx_lock(&cs->cs_lock);
1095                 foundreq = 0;
1096                 TAILQ_FOREACH(cr, &cs->cs_pending, cr_link) {
1097                         if (cr->cr_xid == xid) {
1098                                 /*
1099                                  * This one matches. We leave the
1100                                  * reply mbuf in cr->cr_mrep. Set the
1101                                  * XID to zero so that we will ignore
1102                                  * any duplicated replies that arrive
1103                                  * before clnt_dg_call removes it from
1104                                  * the queue.
1105                                  */
1106                                 cr->cr_xid = 0;
1107                                 cr->cr_mrep = m;
1108                                 cr->cr_error = 0;
1109                                 foundreq = 1;
1110                                 wakeup(cr);
1111                                 break;
1112                         }
1113                 }
1114                 mtx_unlock(&cs->cs_lock);
1115
1116                 /*
1117                  * If we didn't find the matching request, just drop
1118                  * it - its probably a repeated reply.
1119                  */
1120                 if (!foundreq)
1121                         m_freem(m);
1122         } while (m);
1123         cs->cs_upcallrefs--;
1124         if (cs->cs_upcallrefs < 0)
1125                 panic("rpcdg upcall refcnt");
1126         if (cs->cs_upcallrefs == 0)
1127                 wakeup(&cs->cs_upcallrefs);
1128         return (SU_OK);
1129 }
1130
1131 /*
1132  * Wait for all upcalls in progress to complete.
1133  */
1134 static void
1135 clnt_dg_upcallsdone(struct socket *so, struct cu_socket *cs)
1136 {
1137
1138         SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1139
1140         while (cs->cs_upcallrefs > 0)
1141                 (void) msleep(&cs->cs_upcallrefs, SOCKBUF_MTX(&so->so_rcv), 0,
1142                     "rpcdgup", 0);
1143 }