sys/rpc/svc.c

   1 /*      $NetBSD: svc.c,v 1.21 2000/07/06 03:10:35 christos Exp $        */
   2
   3 /*-
   4  * Copyright (c) 2009, Sun Microsystems, Inc.
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions are met:
   9  * - Redistributions of source code must retain the above copyright notice,
  10  *   this list of conditions and the following disclaimer.
  11  * - Redistributions in binary form must reproduce the above copyright notice,
  12  *   this list of conditions and the following disclaimer in the documentation
  13  *   and/or other materials provided with the distribution.
  14  * - Neither the name of Sun Microsystems, Inc. nor the names of its
  15  *   contributors may be used to endorse or promote products derived
  16  *   from this software without specific prior written permission.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  19  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  21  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
  22  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  23  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  24  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
  25  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
  26  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
  27  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
  28  * POSSIBILITY OF SUCH DAMAGE.
  29  */
  30
  31 #if defined(LIBC_SCCS) && !defined(lint)
  32 static char *sccsid2 = "@(#)svc.c 1.44 88/02/08 Copyr 1984 Sun Micro";
  33 static char *sccsid = "@(#)svc.c        2.4 88/08/11 4.0 RPCSRC";
  34 #endif
  35 #include <sys/cdefs.h>
  36 __FBSDID("$FreeBSD$");
  37
  38 /*
  39  * svc.c, Server-side remote procedure call interface.
  40  *
  41  * There are two sets of procedures here.  The xprt routines are
  42  * for handling transport handles.  The svc routines handle the
  43  * list of service routines.
  44  *
  45  * Copyright (C) 1984, Sun Microsystems, Inc.
  46  */
  47
  48 #include <sys/param.h>
  49 #include <sys/lock.h>
  50 #include <sys/kernel.h>
  51 #include <sys/kthread.h>
  52 #include <sys/malloc.h>
  53 #include <sys/mbuf.h>
  54 #include <sys/mutex.h>
  55 #include <sys/proc.h>
  56 #include <sys/queue.h>
  57 #include <sys/socketvar.h>
  58 #include <sys/systm.h>
  59 #include <sys/smp.h>
  60 #include <sys/sx.h>
  61 #include <sys/ucred.h>
  62
  63 #include <rpc/rpc.h>
  64 #include <rpc/rpcb_clnt.h>
  65 #include <rpc/replay.h>
  66
  67 #include <rpc/rpc_com.h>
  68
  69 #define SVC_VERSQUIET 0x0001            /* keep quiet about vers mismatch */
  70 #define version_keepquiet(xp) (SVC_EXT(xp)->xp_flags & SVC_VERSQUIET)
  71
  72 static struct svc_callout *svc_find(SVCPOOL *pool, rpcprog_t, rpcvers_t,
  73     char *);
  74 static void svc_new_thread(SVCGROUP *grp);
  75 static void xprt_unregister_locked(SVCXPRT *xprt);
  76 static void svc_change_space_used(SVCPOOL *pool, int delta);
  77 static bool_t svc_request_space_available(SVCPOOL *pool);
  78
  79 /* ***************  SVCXPRT related stuff **************** */
  80
  81 static int svcpool_minthread_sysctl(SYSCTL_HANDLER_ARGS);
  82 static int svcpool_maxthread_sysctl(SYSCTL_HANDLER_ARGS);
  83 static int svcpool_threads_sysctl(SYSCTL_HANDLER_ARGS);
  84
  85 SVCPOOL*
  86 svcpool_create(const char *name, struct sysctl_oid_list *sysctl_base)
  87 {
  88         SVCPOOL *pool;
  89         SVCGROUP *grp;
  90         int g;
  91
  92         pool = malloc(sizeof(SVCPOOL), M_RPC, M_WAITOK|M_ZERO);
  93
  94         mtx_init(&pool->sp_lock, "sp_lock", NULL, MTX_DEF);
  95         pool->sp_name = name;
  96         pool->sp_state = SVCPOOL_INIT;
  97         pool->sp_proc = NULL;
  98         TAILQ_INIT(&pool->sp_callouts);
  99         TAILQ_INIT(&pool->sp_lcallouts);
 100         pool->sp_minthreads = 1;
 101         pool->sp_maxthreads = 1;
 102         pool->sp_groupcount = 1;
 103         for (g = 0; g < SVC_MAXGROUPS; g++) {
 104                 grp = &pool->sp_groups[g];
 105                 mtx_init(&grp->sg_lock, "sg_lock", NULL, MTX_DEF);
 106                 grp->sg_pool = pool;
 107                 grp->sg_state = SVCPOOL_ACTIVE;
 108                 TAILQ_INIT(&grp->sg_xlist);
 109                 TAILQ_INIT(&grp->sg_active);
 110                 LIST_INIT(&grp->sg_idlethreads);
 111                 grp->sg_minthreads = 1;
 112                 grp->sg_maxthreads = 1;
 113         }
 114
 115         /*
 116          * Don't use more than a quarter of mbuf clusters or more than
 117          * 45Mb buffering requests.
 118          */
 119         pool->sp_space_high = nmbclusters * MCLBYTES / 4;
 120         if (pool->sp_space_high > 45 << 20)
 121                 pool->sp_space_high = 45 << 20;
 122         pool->sp_space_low = 2 * pool->sp_space_high / 3;
 123
 124         sysctl_ctx_init(&pool->sp_sysctl);
 125         if (sysctl_base) {
 126                 SYSCTL_ADD_PROC(&pool->sp_sysctl, sysctl_base, OID_AUTO,
 127                     "minthreads", CTLTYPE_INT | CTLFLAG_RW,
 128                     pool, 0, svcpool_minthread_sysctl, "I",
 129                     "Minimal number of threads");
 130                 SYSCTL_ADD_PROC(&pool->sp_sysctl, sysctl_base, OID_AUTO,
 131                     "maxthreads", CTLTYPE_INT | CTLFLAG_RW,
 132                     pool, 0, svcpool_maxthread_sysctl, "I",
 133                     "Maximal number of threads");
 134                 SYSCTL_ADD_PROC(&pool->sp_sysctl, sysctl_base, OID_AUTO,
 135                     "threads", CTLTYPE_INT | CTLFLAG_RD,
 136                     pool, 0, svcpool_threads_sysctl, "I",
 137                     "Current number of threads");
 138                 SYSCTL_ADD_INT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
 139                     "groups", CTLFLAG_RD, &pool->sp_groupcount, 0,
 140                     "Number of thread groups");
 141
 142                 SYSCTL_ADD_UINT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
 143                     "request_space_used", CTLFLAG_RD,
 144                     &pool->sp_space_used, 0,
 145                     "Space in parsed but not handled requests.");
 146
 147                 SYSCTL_ADD_UINT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
 148                     "request_space_used_highest", CTLFLAG_RD,
 149                     &pool->sp_space_used_highest, 0,
 150                     "Highest space used since reboot.");
 151
 152                 SYSCTL_ADD_UINT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
 153                     "request_space_high", CTLFLAG_RW,
 154                     &pool->sp_space_high, 0,
 155                     "Maximum space in parsed but not handled requests.");
 156
 157                 SYSCTL_ADD_UINT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
 158                     "request_space_low", CTLFLAG_RW,
 159                     &pool->sp_space_low, 0,
 160                     "Low water mark for request space.");
 161
 162                 SYSCTL_ADD_INT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
 163                     "request_space_throttled", CTLFLAG_RD,
 164                     &pool->sp_space_throttled, 0,
 165                     "Whether nfs requests are currently throttled");
 166
 167                 SYSCTL_ADD_INT(&pool->sp_sysctl, sysctl_base, OID_AUTO,
 168                     "request_space_throttle_count", CTLFLAG_RD,
 169                     &pool->sp_space_throttle_count, 0,
 170                     "Count of times throttling based on request space has occurred");
 171         }
 172
 173         return pool;
 174 }
 175
 176 void
 177 svcpool_destroy(SVCPOOL *pool)
 178 {
 179         SVCGROUP *grp;
 180         SVCXPRT *xprt, *nxprt;
 181         struct svc_callout *s;
 182         struct svc_loss_callout *sl;
 183         struct svcxprt_list cleanup;
 184         int g;
 185
 186         TAILQ_INIT(&cleanup);
 187
 188         for (g = 0; g < SVC_MAXGROUPS; g++) {
 189                 grp = &pool->sp_groups[g];
 190                 mtx_lock(&grp->sg_lock);
 191                 while ((xprt = TAILQ_FIRST(&grp->sg_xlist)) != NULL) {
 192                         xprt_unregister_locked(xprt);
 193                         TAILQ_INSERT_TAIL(&cleanup, xprt, xp_link);
 194                 }
 195                 mtx_unlock(&grp->sg_lock);
 196         }
 197         TAILQ_FOREACH_SAFE(xprt, &cleanup, xp_link, nxprt) {
 198                 SVC_RELEASE(xprt);
 199         }
 200
 201         mtx_lock(&pool->sp_lock);
 202         while ((s = TAILQ_FIRST(&pool->sp_callouts)) != NULL) {
 203                 mtx_unlock(&pool->sp_lock);
 204                 svc_unreg(pool, s->sc_prog, s->sc_vers);
 205                 mtx_lock(&pool->sp_lock);
 206         }
 207         while ((sl = TAILQ_FIRST(&pool->sp_lcallouts)) != NULL) {
 208                 mtx_unlock(&pool->sp_lock);
 209                 svc_loss_unreg(pool, sl->slc_dispatch);
 210                 mtx_lock(&pool->sp_lock);
 211         }
 212         mtx_unlock(&pool->sp_lock);
 213
 214         for (g = 0; g < SVC_MAXGROUPS; g++) {
 215                 grp = &pool->sp_groups[g];
 216                 mtx_destroy(&grp->sg_lock);
 217         }
 218         mtx_destroy(&pool->sp_lock);
 219
 220         if (pool->sp_rcache)
 221                 replay_freecache(pool->sp_rcache);
 222
 223         sysctl_ctx_free(&pool->sp_sysctl);
 224         free(pool, M_RPC);
 225 }
 226
 227 /*
 228  * Sysctl handler to get the present thread count on a pool
 229  */
 230 static int
 231 svcpool_threads_sysctl(SYSCTL_HANDLER_ARGS)
 232 {
 233         SVCPOOL *pool;
 234         int threads, error, g;
 235
 236         pool = oidp->oid_arg1;
 237         threads = 0;
 238         mtx_lock(&pool->sp_lock);
 239         for (g = 0; g < pool->sp_groupcount; g++)
 240                 threads += pool->sp_groups[g].sg_threadcount;
 241         mtx_unlock(&pool->sp_lock);
 242         error = sysctl_handle_int(oidp, &threads, 0, req);
 243         return (error);
 244 }
 245
 246 /*
 247  * Sysctl handler to set the minimum thread count on a pool
 248  */
 249 static int
 250 svcpool_minthread_sysctl(SYSCTL_HANDLER_ARGS)
 251 {
 252         SVCPOOL *pool;
 253         int newminthreads, error, g;
 254
 255         pool = oidp->oid_arg1;
 256         newminthreads = pool->sp_minthreads;
 257         error = sysctl_handle_int(oidp, &newminthreads, 0, req);
 258         if (error == 0 && newminthreads != pool->sp_minthreads) {
 259                 if (newminthreads > pool->sp_maxthreads)
 260                         return (EINVAL);
 261                 mtx_lock(&pool->sp_lock);
 262                 pool->sp_minthreads = newminthreads;
 263                 for (g = 0; g < pool->sp_groupcount; g++) {
 264                         pool->sp_groups[g].sg_minthreads = max(1,
 265                             pool->sp_minthreads / pool->sp_groupcount);
 266                 }
 267                 mtx_unlock(&pool->sp_lock);
 268         }
 269         return (error);
 270 }
 271
 272 /*
 273  * Sysctl handler to set the maximum thread count on a pool
 274  */
 275 static int
 276 svcpool_maxthread_sysctl(SYSCTL_HANDLER_ARGS)
 277 {
 278         SVCPOOL *pool;
 279         int newmaxthreads, error, g;
 280
 281         pool = oidp->oid_arg1;
 282         newmaxthreads = pool->sp_maxthreads;
 283         error = sysctl_handle_int(oidp, &newmaxthreads, 0, req);
 284         if (error == 0 && newmaxthreads != pool->sp_maxthreads) {
 285                 if (newmaxthreads < pool->sp_minthreads)
 286                         return (EINVAL);
 287                 mtx_lock(&pool->sp_lock);
 288                 pool->sp_maxthreads = newmaxthreads;
 289                 for (g = 0; g < pool->sp_groupcount; g++) {
 290                         pool->sp_groups[g].sg_maxthreads = max(1,
 291                             pool->sp_maxthreads / pool->sp_groupcount);
 292                 }
 293                 mtx_unlock(&pool->sp_lock);
 294         }
 295         return (error);
 296 }
 297
 298 /*
 299  * Activate a transport handle.
 300  */
 301 void
 302 xprt_register(SVCXPRT *xprt)
 303 {
 304         SVCPOOL *pool = xprt->xp_pool;
 305         SVCGROUP *grp;
 306         int g;
 307
 308         SVC_ACQUIRE(xprt);
 309         g = atomic_fetchadd_int(&pool->sp_nextgroup, 1) % pool->sp_groupcount;
 310         xprt->xp_group = grp = &pool->sp_groups[g];
 311         mtx_lock(&grp->sg_lock);
 312         xprt->xp_registered = TRUE;
 313         xprt->xp_active = FALSE;
 314         TAILQ_INSERT_TAIL(&grp->sg_xlist, xprt, xp_link);
 315         mtx_unlock(&grp->sg_lock);
 316 }
 317
 318 /*
 319  * De-activate a transport handle. Note: the locked version doesn't
 320  * release the transport - caller must do that after dropping the pool
 321  * lock.
 322  */
 323 static void
 324 xprt_unregister_locked(SVCXPRT *xprt)
 325 {
 326         SVCGROUP *grp = xprt->xp_group;
 327
 328         mtx_assert(&grp->sg_lock, MA_OWNED);
 329         KASSERT(xprt->xp_registered == TRUE,
 330             ("xprt_unregister_locked: not registered"));
 331         xprt_inactive_locked(xprt);
 332         TAILQ_REMOVE(&grp->sg_xlist, xprt, xp_link);
 333         xprt->xp_registered = FALSE;
 334 }
 335
 336 void
 337 xprt_unregister(SVCXPRT *xprt)
 338 {
 339         SVCGROUP *grp = xprt->xp_group;
 340
 341         mtx_lock(&grp->sg_lock);
 342         if (xprt->xp_registered == FALSE) {
 343                 /* Already unregistered by another thread */
 344                 mtx_unlock(&grp->sg_lock);
 345                 return;
 346         }
 347         xprt_unregister_locked(xprt);
 348         mtx_unlock(&grp->sg_lock);
 349
 350         SVC_RELEASE(xprt);
 351 }
 352
 353 /*
 354  * Attempt to assign a service thread to this transport.
 355  */
 356 static int
 357 xprt_assignthread(SVCXPRT *xprt)
 358 {
 359         SVCGROUP *grp = xprt->xp_group;
 360         SVCTHREAD *st;
 361
 362         mtx_assert(&grp->sg_lock, MA_OWNED);
 363         st = LIST_FIRST(&grp->sg_idlethreads);
 364         if (st) {
 365                 LIST_REMOVE(st, st_ilink);
 366                 SVC_ACQUIRE(xprt);
 367                 xprt->xp_thread = st;
 368                 st->st_xprt = xprt;
 369                 cv_signal(&st->st_cond);
 370                 return (TRUE);
 371         } else {
 372                 /*
 373                  * See if we can create a new thread. The
 374                  * actual thread creation happens in
 375                  * svc_run_internal because our locking state
 376                  * is poorly defined (we are typically called
 377                  * from a socket upcall). Don't create more
 378                  * than one thread per second.
 379                  */
 380                 if (grp->sg_state == SVCPOOL_ACTIVE
 381                     && grp->sg_lastcreatetime < time_uptime
 382                     && grp->sg_threadcount < grp->sg_maxthreads) {
 383                         grp->sg_state = SVCPOOL_THREADWANTED;
 384                 }
 385         }
 386         return (FALSE);
 387 }
 388
 389 void
 390 xprt_active(SVCXPRT *xprt)
 391 {
 392         SVCGROUP *grp = xprt->xp_group;
 393
 394         mtx_lock(&grp->sg_lock);
 395
 396         if (!xprt->xp_registered) {
 397                 /*
 398                  * Race with xprt_unregister - we lose.
 399                  */
 400                 mtx_unlock(&grp->sg_lock);
 401                 return;
 402         }
 403
 404         if (!xprt->xp_active) {
 405                 xprt->xp_active = TRUE;
 406                 if (xprt->xp_thread == NULL) {
 407                         if (!svc_request_space_available(xprt->xp_pool) ||
 408                             !xprt_assignthread(xprt))
 409                                 TAILQ_INSERT_TAIL(&grp->sg_active, xprt,
 410                                     xp_alink);
 411                 }
 412         }
 413
 414         mtx_unlock(&grp->sg_lock);
 415 }
 416
 417 void
 418 xprt_inactive_locked(SVCXPRT *xprt)
 419 {
 420         SVCGROUP *grp = xprt->xp_group;
 421
 422         mtx_assert(&grp->sg_lock, MA_OWNED);
 423         if (xprt->xp_active) {
 424                 if (xprt->xp_thread == NULL)
 425                         TAILQ_REMOVE(&grp->sg_active, xprt, xp_alink);
 426                 xprt->xp_active = FALSE;
 427         }
 428 }
 429
 430 void
 431 xprt_inactive(SVCXPRT *xprt)
 432 {
 433         SVCGROUP *grp = xprt->xp_group;
 434
 435         mtx_lock(&grp->sg_lock);
 436         xprt_inactive_locked(xprt);
 437         mtx_unlock(&grp->sg_lock);
 438 }
 439
 440 /*
 441  * Variant of xprt_inactive() for use only when sure that port is
 442  * assigned to thread. For example, withing receive handlers.
 443  */
 444 void
 445 xprt_inactive_self(SVCXPRT *xprt)
 446 {
 447
 448         KASSERT(xprt->xp_thread != NULL,
 449             ("xprt_inactive_self(%p) with NULL xp_thread", xprt));
 450         xprt->xp_active = FALSE;
 451 }
 452
 453 /*
 454  * Add a service program to the callout list.
 455  * The dispatch routine will be called when a rpc request for this
 456  * program number comes in.
 457  */
 458 bool_t
 459 svc_reg(SVCXPRT *xprt, const rpcprog_t prog, const rpcvers_t vers,
 460     void (*dispatch)(struct svc_req *, SVCXPRT *),
 461     const struct netconfig *nconf)
 462 {
 463         SVCPOOL *pool = xprt->xp_pool;
 464         struct svc_callout *s;
 465         char *netid = NULL;
 466         int flag = 0;
 467
 468 /* VARIABLES PROTECTED BY svc_lock: s, svc_head */
 469
 470         if (xprt->xp_netid) {
 471                 netid = strdup(xprt->xp_netid, M_RPC);
 472                 flag = 1;
 473         } else if (nconf && nconf->nc_netid) {
 474                 netid = strdup(nconf->nc_netid, M_RPC);
 475                 flag = 1;
 476         } /* must have been created with svc_raw_create */
 477         if ((netid == NULL) && (flag == 1)) {
 478                 return (FALSE);
 479         }
 480
 481         mtx_lock(&pool->sp_lock);
 482         if ((s = svc_find(pool, prog, vers, netid)) != NULL) {
 483                 if (netid)
 484                         free(netid, M_RPC);
 485                 if (s->sc_dispatch == dispatch)
 486                         goto rpcb_it; /* he is registering another xptr */
 487                 mtx_unlock(&pool->sp_lock);
 488                 return (FALSE);
 489         }
 490         s = malloc(sizeof (struct svc_callout), M_RPC, M_NOWAIT);
 491         if (s == NULL) {
 492                 if (netid)
 493                         free(netid, M_RPC);
 494                 mtx_unlock(&pool->sp_lock);
 495                 return (FALSE);
 496         }
 497
 498         s->sc_prog = prog;
 499         s->sc_vers = vers;
 500         s->sc_dispatch = dispatch;
 501         s->sc_netid = netid;
 502         TAILQ_INSERT_TAIL(&pool->sp_callouts, s, sc_link);
 503
 504         if ((xprt->xp_netid == NULL) && (flag == 1) && netid)
 505                 ((SVCXPRT *) xprt)->xp_netid = strdup(netid, M_RPC);
 506
 507 rpcb_it:
 508         mtx_unlock(&pool->sp_lock);
 509         /* now register the information with the local binder service */
 510         if (nconf) {
 511                 bool_t dummy;
 512                 struct netconfig tnc;
 513                 struct netbuf nb;
 514                 tnc = *nconf;
 515                 nb.buf = &xprt->xp_ltaddr;
 516                 nb.len = xprt->xp_ltaddr.ss_len;
 517                 dummy = rpcb_set(prog, vers, &tnc, &nb);
 518                 return (dummy);
 519         }
 520         return (TRUE);
 521 }
 522
 523 /*
 524  * Remove a service program from the callout list.
 525  */
 526 void
 527 svc_unreg(SVCPOOL *pool, const rpcprog_t prog, const rpcvers_t vers)
 528 {
 529         struct svc_callout *s;
 530
 531         /* unregister the information anyway */
 532         (void) rpcb_unset(prog, vers, NULL);
 533         mtx_lock(&pool->sp_lock);
 534         while ((s = svc_find(pool, prog, vers, NULL)) != NULL) {
 535                 TAILQ_REMOVE(&pool->sp_callouts, s, sc_link);
 536                 if (s->sc_netid)
 537                         mem_free(s->sc_netid, sizeof (s->sc_netid) + 1);
 538                 mem_free(s, sizeof (struct svc_callout));
 539         }
 540         mtx_unlock(&pool->sp_lock);
 541 }
 542
 543 /*
 544  * Add a service connection loss program to the callout list.
 545  * The dispatch routine will be called when some port in ths pool die.
 546  */
 547 bool_t
 548 svc_loss_reg(SVCXPRT *xprt, void (*dispatch)(SVCXPRT *))
 549 {
 550         SVCPOOL *pool = xprt->xp_pool;
 551         struct svc_loss_callout *s;
 552
 553         mtx_lock(&pool->sp_lock);
 554         TAILQ_FOREACH(s, &pool->sp_lcallouts, slc_link) {
 555                 if (s->slc_dispatch == dispatch)
 556                         break;
 557         }
 558         if (s != NULL) {
 559                 mtx_unlock(&pool->sp_lock);
 560                 return (TRUE);
 561         }
 562         s = malloc(sizeof(struct svc_loss_callout), M_RPC, M_NOWAIT);
 563         if (s == NULL) {
 564                 mtx_unlock(&pool->sp_lock);
 565                 return (FALSE);
 566         }
 567         s->slc_dispatch = dispatch;
 568         TAILQ_INSERT_TAIL(&pool->sp_lcallouts, s, slc_link);
 569         mtx_unlock(&pool->sp_lock);
 570         return (TRUE);
 571 }
 572
 573 /*
 574  * Remove a service connection loss program from the callout list.
 575  */
 576 void
 577 svc_loss_unreg(SVCPOOL *pool, void (*dispatch)(SVCXPRT *))
 578 {
 579         struct svc_loss_callout *s;
 580
 581         mtx_lock(&pool->sp_lock);
 582         TAILQ_FOREACH(s, &pool->sp_lcallouts, slc_link) {
 583                 if (s->slc_dispatch == dispatch) {
 584                         TAILQ_REMOVE(&pool->sp_lcallouts, s, slc_link);
 585                         free(s, M_RPC);
 586                         break;
 587                 }
 588         }
 589         mtx_unlock(&pool->sp_lock);
 590 }
 591
 592 /* ********************** CALLOUT list related stuff ************* */
 593
 594 /*
 595  * Search the callout list for a program number, return the callout
 596  * struct.
 597  */
 598 static struct svc_callout *
 599 svc_find(SVCPOOL *pool, rpcprog_t prog, rpcvers_t vers, char *netid)
 600 {
 601         struct svc_callout *s;
 602
 603         mtx_assert(&pool->sp_lock, MA_OWNED);
 604         TAILQ_FOREACH(s, &pool->sp_callouts, sc_link) {
 605                 if (s->sc_prog == prog && s->sc_vers == vers
 606                     && (netid == NULL || s->sc_netid == NULL ||
 607                         strcmp(netid, s->sc_netid) == 0))
 608                         break;
 609         }
 610
 611         return (s);
 612 }
 613
 614 /* ******************* REPLY GENERATION ROUTINES  ************ */
 615
 616 static bool_t
 617 svc_sendreply_common(struct svc_req *rqstp, struct rpc_msg *rply,
 618     struct mbuf *body)
 619 {
 620         SVCXPRT *xprt = rqstp->rq_xprt;
 621         bool_t ok;
 622
 623         if (rqstp->rq_args) {
 624                 m_freem(rqstp->rq_args);
 625                 rqstp->rq_args = NULL;
 626         }
 627
 628         if (xprt->xp_pool->sp_rcache)
 629                 replay_setreply(xprt->xp_pool->sp_rcache,
 630                     rply, svc_getrpccaller(rqstp), body);
 631
 632         if (!SVCAUTH_WRAP(&rqstp->rq_auth, &body))
 633                 return (FALSE);
 634
 635         ok = SVC_REPLY(xprt, rply, rqstp->rq_addr, body, &rqstp->rq_reply_seq);
 636         if (rqstp->rq_addr) {
 637                 free(rqstp->rq_addr, M_SONAME);
 638                 rqstp->rq_addr = NULL;
 639         }
 640
 641         return (ok);
 642 }
 643
 644 /*
 645  * Send a reply to an rpc request
 646  */
 647 bool_t
 648 svc_sendreply(struct svc_req *rqstp, xdrproc_t xdr_results, void * xdr_location)
 649 {
 650         struct rpc_msg rply;
 651         struct mbuf *m;
 652         XDR xdrs;
 653         bool_t ok;
 654
 655         rply.rm_xid = rqstp->rq_xid;
 656         rply.rm_direction = REPLY;
 657         rply.rm_reply.rp_stat = MSG_ACCEPTED;
 658         rply.acpted_rply.ar_verf = rqstp->rq_verf;
 659         rply.acpted_rply.ar_stat = SUCCESS;
 660         rply.acpted_rply.ar_results.where = NULL;
 661         rply.acpted_rply.ar_results.proc = (xdrproc_t) xdr_void;
 662
 663         MGET(m, M_WAIT, MT_DATA);
 664         MCLGET(m, M_WAIT);
 665         m->m_len = 0;
 666         xdrmbuf_create(&xdrs, m, XDR_ENCODE);
 667         ok = xdr_results(&xdrs, xdr_location);
 668         XDR_DESTROY(&xdrs);
 669
 670         if (ok) {
 671                 return (svc_sendreply_common(rqstp, &rply, m));
 672         } else {
 673                 m_freem(m);
 674                 return (FALSE);
 675         }
 676 }
 677
 678 bool_t
 679 svc_sendreply_mbuf(struct svc_req *rqstp, struct mbuf *m)
 680 {
 681         struct rpc_msg rply;
 682
 683         rply.rm_xid = rqstp->rq_xid;
 684         rply.rm_direction = REPLY;
 685         rply.rm_reply.rp_stat = MSG_ACCEPTED;
 686         rply.acpted_rply.ar_verf = rqstp->rq_verf;
 687         rply.acpted_rply.ar_stat = SUCCESS;
 688         rply.acpted_rply.ar_results.where = NULL;
 689         rply.acpted_rply.ar_results.proc = (xdrproc_t) xdr_void;
 690
 691         return (svc_sendreply_common(rqstp, &rply, m));
 692 }
 693
 694 /*
 695  * No procedure error reply
 696  */
 697 void
 698 svcerr_noproc(struct svc_req *rqstp)
 699 {
 700         SVCXPRT *xprt = rqstp->rq_xprt;
 701         struct rpc_msg rply;
 702
 703         rply.rm_xid = rqstp->rq_xid;
 704         rply.rm_direction = REPLY;
 705         rply.rm_reply.rp_stat = MSG_ACCEPTED;
 706         rply.acpted_rply.ar_verf = rqstp->rq_verf;
 707         rply.acpted_rply.ar_stat = PROC_UNAVAIL;
 708
 709         if (xprt->xp_pool->sp_rcache)
 710                 replay_setreply(xprt->xp_pool->sp_rcache,
 711                     &rply, svc_getrpccaller(rqstp), NULL);
 712
 713         svc_sendreply_common(rqstp, &rply, NULL);
 714 }
 715
 716 /*
 717  * Can't decode args error reply
 718  */
 719 void
 720 svcerr_decode(struct svc_req *rqstp)
 721 {
 722         SVCXPRT *xprt = rqstp->rq_xprt;
 723         struct rpc_msg rply;
 724
 725         rply.rm_xid = rqstp->rq_xid;
 726         rply.rm_direction = REPLY;
 727         rply.rm_reply.rp_stat = MSG_ACCEPTED;
 728         rply.acpted_rply.ar_verf = rqstp->rq_verf;
 729         rply.acpted_rply.ar_stat = GARBAGE_ARGS;
 730
 731         if (xprt->xp_pool->sp_rcache)
 732                 replay_setreply(xprt->xp_pool->sp_rcache,
 733                     &rply, (struct sockaddr *) &xprt->xp_rtaddr, NULL);
 734
 735         svc_sendreply_common(rqstp, &rply, NULL);
 736 }
 737
 738 /*
 739  * Some system error
 740  */
 741 void
 742 svcerr_systemerr(struct svc_req *rqstp)
 743 {
 744         SVCXPRT *xprt = rqstp->rq_xprt;
 745         struct rpc_msg rply;
 746
 747         rply.rm_xid = rqstp->rq_xid;
 748         rply.rm_direction = REPLY;
 749         rply.rm_reply.rp_stat = MSG_ACCEPTED;
 750         rply.acpted_rply.ar_verf = rqstp->rq_verf;
 751         rply.acpted_rply.ar_stat = SYSTEM_ERR;
 752
 753         if (xprt->xp_pool->sp_rcache)
 754                 replay_setreply(xprt->xp_pool->sp_rcache,
 755                     &rply, svc_getrpccaller(rqstp), NULL);
 756
 757         svc_sendreply_common(rqstp, &rply, NULL);
 758 }
 759
 760 /*
 761  * Authentication error reply
 762  */
 763 void
 764 svcerr_auth(struct svc_req *rqstp, enum auth_stat why)
 765 {
 766         SVCXPRT *xprt = rqstp->rq_xprt;
 767         struct rpc_msg rply;
 768
 769         rply.rm_xid = rqstp->rq_xid;
 770         rply.rm_direction = REPLY;
 771         rply.rm_reply.rp_stat = MSG_DENIED;
 772         rply.rjcted_rply.rj_stat = AUTH_ERROR;
 773         rply.rjcted_rply.rj_why = why;
 774
 775         if (xprt->xp_pool->sp_rcache)
 776                 replay_setreply(xprt->xp_pool->sp_rcache,
 777                     &rply, svc_getrpccaller(rqstp), NULL);
 778
 779         svc_sendreply_common(rqstp, &rply, NULL);
 780 }
 781
 782 /*
 783  * Auth too weak error reply
 784  */
 785 void
 786 svcerr_weakauth(struct svc_req *rqstp)
 787 {
 788
 789         svcerr_auth(rqstp, AUTH_TOOWEAK);
 790 }
 791
 792 /*
 793  * Program unavailable error reply
 794  */
 795 void
 796 svcerr_noprog(struct svc_req *rqstp)
 797 {
 798         SVCXPRT *xprt = rqstp->rq_xprt;
 799         struct rpc_msg rply;
 800
 801         rply.rm_xid = rqstp->rq_xid;
 802         rply.rm_direction = REPLY;
 803         rply.rm_reply.rp_stat = MSG_ACCEPTED;
 804         rply.acpted_rply.ar_verf = rqstp->rq_verf;
 805         rply.acpted_rply.ar_stat = PROG_UNAVAIL;
 806
 807         if (xprt->xp_pool->sp_rcache)
 808                 replay_setreply(xprt->xp_pool->sp_rcache,
 809                     &rply, svc_getrpccaller(rqstp), NULL);
 810
 811         svc_sendreply_common(rqstp, &rply, NULL);
 812 }
 813
 814 /*
 815  * Program version mismatch error reply
 816  */
 817 void
 818 svcerr_progvers(struct svc_req *rqstp, rpcvers_t low_vers, rpcvers_t high_vers)
 819 {
 820         SVCXPRT *xprt = rqstp->rq_xprt;
 821         struct rpc_msg rply;
 822
 823         rply.rm_xid = rqstp->rq_xid;
 824         rply.rm_direction = REPLY;
 825         rply.rm_reply.rp_stat = MSG_ACCEPTED;
 826         rply.acpted_rply.ar_verf = rqstp->rq_verf;
 827         rply.acpted_rply.ar_stat = PROG_MISMATCH;
 828         rply.acpted_rply.ar_vers.low = (uint32_t)low_vers;
 829         rply.acpted_rply.ar_vers.high = (uint32_t)high_vers;
 830
 831         if (xprt->xp_pool->sp_rcache)
 832                 replay_setreply(xprt->xp_pool->sp_rcache,
 833                     &rply, svc_getrpccaller(rqstp), NULL);
 834
 835         svc_sendreply_common(rqstp, &rply, NULL);
 836 }
 837
 838 /*
 839  * Allocate a new server transport structure. All fields are
 840  * initialized to zero and xp_p3 is initialized to point at an
 841  * extension structure to hold various flags and authentication
 842  * parameters.
 843  */
 844 SVCXPRT *
 845 svc_xprt_alloc(void)
 846 {
 847         SVCXPRT *xprt;
 848         SVCXPRT_EXT *ext;
 849
 850         xprt = mem_alloc(sizeof(SVCXPRT));
 851         ext = mem_alloc(sizeof(SVCXPRT_EXT));
 852         xprt->xp_p3 = ext;
 853         refcount_init(&xprt->xp_refs, 1);
 854
 855         return (xprt);
 856 }
 857
 858 /*
 859  * Free a server transport structure.
 860  */
 861 void
 862 svc_xprt_free(SVCXPRT *xprt)
 863 {
 864
 865         mem_free(xprt->xp_p3, sizeof(SVCXPRT_EXT));
 866         mem_free(xprt, sizeof(SVCXPRT));
 867 }
 868
 869 /* ******************* SERVER INPUT STUFF ******************* */
 870
 871 /*
 872  * Read RPC requests from a transport and queue them to be
 873  * executed. We handle authentication and replay cache replies here.
 874  * Actually dispatching the RPC is deferred till svc_executereq.
 875  */
 876 static enum xprt_stat
 877 svc_getreq(SVCXPRT *xprt, struct svc_req **rqstp_ret)
 878 {
 879         SVCPOOL *pool = xprt->xp_pool;
 880         struct svc_req *r;
 881         struct rpc_msg msg;
 882         struct mbuf *args;
 883         struct svc_loss_callout *s;
 884         enum xprt_stat stat;
 885
 886         /* now receive msgs from xprtprt (support batch calls) */
 887         r = malloc(sizeof(*r), M_RPC, M_WAITOK|M_ZERO);
 888
 889         msg.rm_call.cb_cred.oa_base = r->rq_credarea;
 890         msg.rm_call.cb_verf.oa_base = &r->rq_credarea[MAX_AUTH_BYTES];
 891         r->rq_clntcred = &r->rq_credarea[2*MAX_AUTH_BYTES];
 892         if (SVC_RECV(xprt, &msg, &r->rq_addr, &args)) {
 893                 enum auth_stat why;
 894
 895                 /*
 896                  * Handle replays and authenticate before queuing the
 897                  * request to be executed.
 898                  */
 899                 SVC_ACQUIRE(xprt);
 900                 r->rq_xprt = xprt;
 901                 if (pool->sp_rcache) {
 902                         struct rpc_msg repmsg;
 903                         struct mbuf *repbody;
 904                         enum replay_state rs;
 905                         rs = replay_find(pool->sp_rcache, &msg,
 906                             svc_getrpccaller(r), &repmsg, &repbody);
 907                         switch (rs) {
 908                         case RS_NEW:
 909                                 break;
 910                         case RS_DONE:
 911                                 SVC_REPLY(xprt, &repmsg, r->rq_addr,
 912                                     repbody, &r->rq_reply_seq);
 913                                 if (r->rq_addr) {
 914                                         free(r->rq_addr, M_SONAME);
 915                                         r->rq_addr = NULL;
 916                                 }
 917                                 m_freem(args);
 918                                 goto call_done;
 919
 920                         default:
 921                                 m_freem(args);
 922                                 goto call_done;
 923                         }
 924                 }
 925
 926                 r->rq_xid = msg.rm_xid;
 927                 r->rq_prog = msg.rm_call.cb_prog;
 928                 r->rq_vers = msg.rm_call.cb_vers;
 929                 r->rq_proc = msg.rm_call.cb_proc;
 930                 r->rq_size = sizeof(*r) + m_length(args, NULL);
 931                 r->rq_args = args;
 932                 if ((why = _authenticate(r, &msg)) != AUTH_OK) {
 933                         /*
 934                          * RPCSEC_GSS uses this return code
 935                          * for requests that form part of its
 936                          * context establishment protocol and
 937                          * should not be dispatched to the
 938                          * application.
 939                          */
 940                         if (why != RPCSEC_GSS_NODISPATCH)
 941                                 svcerr_auth(r, why);
 942                         goto call_done;
 943                 }
 944
 945                 if (!SVCAUTH_UNWRAP(&r->rq_auth, &r->rq_args)) {
 946                         svcerr_decode(r);
 947                         goto call_done;
 948                 }
 949
 950                 /*
 951                  * Everything checks out, return request to caller.
 952                  */
 953                 *rqstp_ret = r;
 954                 r = NULL;
 955         }
 956 call_done:
 957         if (r) {
 958                 svc_freereq(r);
 959                 r = NULL;
 960         }
 961         if ((stat = SVC_STAT(xprt)) == XPRT_DIED) {
 962                 TAILQ_FOREACH(s, &pool->sp_lcallouts, slc_link)
 963                         (*s->slc_dispatch)(xprt);
 964                 xprt_unregister(xprt);
 965         }
 966
 967         return (stat);
 968 }
 969
 970 static void
 971 svc_executereq(struct svc_req *rqstp)
 972 {
 973         SVCXPRT *xprt = rqstp->rq_xprt;
 974         SVCPOOL *pool = xprt->xp_pool;
 975         int prog_found;
 976         rpcvers_t low_vers;
 977         rpcvers_t high_vers;
 978         struct svc_callout *s;
 979
 980         /* now match message with a registered service*/
 981         prog_found = FALSE;
 982         low_vers = (rpcvers_t) -1L;
 983         high_vers = (rpcvers_t) 0L;
 984         TAILQ_FOREACH(s, &pool->sp_callouts, sc_link) {
 985                 if (s->sc_prog == rqstp->rq_prog) {
 986                         if (s->sc_vers == rqstp->rq_vers) {
 987                                 /*
 988                                  * We hand ownership of r to the
 989                                  * dispatch method - they must call
 990                                  * svc_freereq.
 991                                  */
 992                                 (*s->sc_dispatch)(rqstp, xprt);
 993                                 return;
 994                         }  /* found correct version */
 995                         prog_found = TRUE;
 996                         if (s->sc_vers < low_vers)
 997                                 low_vers = s->sc_vers;
 998                         if (s->sc_vers > high_vers)
 999                                 high_vers = s->sc_vers;
1000                 }   /* found correct program */
1001         }
1002
1003         /*
1004          * if we got here, the program or version
1005          * is not served ...
1006          */
1007         if (prog_found)
1008                 svcerr_progvers(rqstp, low_vers, high_vers);
1009         else
1010                 svcerr_noprog(rqstp);
1011
1012         svc_freereq(rqstp);
1013 }
1014
1015 static void
1016 svc_checkidle(SVCGROUP *grp)
1017 {
1018         SVCXPRT *xprt, *nxprt;
1019         time_t timo;
1020         struct svcxprt_list cleanup;
1021
1022         TAILQ_INIT(&cleanup);
1023         TAILQ_FOREACH_SAFE(xprt, &grp->sg_xlist, xp_link, nxprt) {
1024                 /*
1025                  * Only some transports have idle timers. Don't time
1026                  * something out which is just waking up.
1027                  */
1028                 if (!xprt->xp_idletimeout || xprt->xp_thread)
1029                         continue;
1030
1031                 timo = xprt->xp_lastactive + xprt->xp_idletimeout;
1032                 if (time_uptime > timo) {
1033                         xprt_unregister_locked(xprt);
1034                         TAILQ_INSERT_TAIL(&cleanup, xprt, xp_link);
1035                 }
1036         }
1037
1038         mtx_unlock(&grp->sg_lock);
1039         TAILQ_FOREACH_SAFE(xprt, &cleanup, xp_link, nxprt) {
1040                 SVC_RELEASE(xprt);
1041         }
1042         mtx_lock(&grp->sg_lock);
1043 }
1044
1045 static void
1046 svc_assign_waiting_sockets(SVCPOOL *pool)
1047 {
1048         SVCGROUP *grp;
1049         SVCXPRT *xprt;
1050         int g;
1051
1052         for (g = 0; g < pool->sp_groupcount; g++) {
1053                 grp = &pool->sp_groups[g];
1054                 mtx_lock(&grp->sg_lock);
1055                 while ((xprt = TAILQ_FIRST(&grp->sg_active)) != NULL) {
1056                         if (xprt_assignthread(xprt))
1057                                 TAILQ_REMOVE(&grp->sg_active, xprt, xp_alink);
1058                         else
1059                                 break;
1060                 }
1061                 mtx_unlock(&grp->sg_lock);
1062         }
1063 }
1064
1065 static void
1066 svc_change_space_used(SVCPOOL *pool, int delta)
1067 {
1068         unsigned int value;
1069
1070         value = atomic_fetchadd_int(&pool->sp_space_used, delta) + delta;
1071         if (delta > 0) {
1072                 if (value >= pool->sp_space_high && !pool->sp_space_throttled) {
1073                         pool->sp_space_throttled = TRUE;
1074                         pool->sp_space_throttle_count++;
1075                 }
1076                 if (value > pool->sp_space_used_highest)
1077                         pool->sp_space_used_highest = value;
1078         } else {
1079                 if (value < pool->sp_space_low && pool->sp_space_throttled) {
1080                         pool->sp_space_throttled = FALSE;
1081                         svc_assign_waiting_sockets(pool);
1082                 }
1083         }
1084 }
1085
1086 static bool_t
1087 svc_request_space_available(SVCPOOL *pool)
1088 {
1089
1090         if (pool->sp_space_throttled)
1091                 return (FALSE);
1092         return (TRUE);
1093 }
1094
1095 static void
1096 svc_run_internal(SVCGROUP *grp, bool_t ismaster)
1097 {
1098         SVCPOOL *pool = grp->sg_pool;
1099         SVCTHREAD *st, *stpref;
1100         SVCXPRT *xprt;
1101         enum xprt_stat stat;
1102         struct svc_req *rqstp;
1103         size_t sz;
1104         int error;
1105
1106         st = mem_alloc(sizeof(*st));
1107         mtx_init(&st->st_lock, "st_lock", NULL, MTX_DEF);
1108         st->st_pool = pool;
1109         st->st_xprt = NULL;
1110         STAILQ_INIT(&st->st_reqs);
1111         cv_init(&st->st_cond, "rpcsvc");
1112
1113         mtx_lock(&grp->sg_lock);
1114
1115         /*
1116          * If we are a new thread which was spawned to cope with
1117          * increased load, set the state back to SVCPOOL_ACTIVE.
1118          */
1119         if (grp->sg_state == SVCPOOL_THREADSTARTING)
1120                 grp->sg_state = SVCPOOL_ACTIVE;
1121
1122         while (grp->sg_state != SVCPOOL_CLOSING) {
1123                 /*
1124                  * Create new thread if requested.
1125                  */
1126                 if (grp->sg_state == SVCPOOL_THREADWANTED) {
1127                         grp->sg_state = SVCPOOL_THREADSTARTING;
1128                         grp->sg_lastcreatetime = time_uptime;
1129                         mtx_unlock(&grp->sg_lock);
1130                         svc_new_thread(grp);
1131                         mtx_lock(&grp->sg_lock);
1132                         continue;
1133                 }
1134
1135                 /*
1136                  * Check for idle transports once per second.
1137                  */
1138                 if (time_uptime > grp->sg_lastidlecheck) {
1139                         grp->sg_lastidlecheck = time_uptime;
1140                         svc_checkidle(grp);
1141                 }
1142
1143                 xprt = st->st_xprt;
1144                 if (!xprt) {
1145                         /*
1146                          * Enforce maxthreads count.
1147                          */
1148                         if (grp->sg_threadcount > grp->sg_maxthreads)
1149                                 break;
1150
1151                         /*
1152                          * Before sleeping, see if we can find an
1153                          * active transport which isn't being serviced
1154                          * by a thread.
1155                          */
1156                         if (svc_request_space_available(pool) &&
1157                             (xprt = TAILQ_FIRST(&grp->sg_active)) != NULL) {
1158                                 TAILQ_REMOVE(&grp->sg_active, xprt, xp_alink);
1159                                 SVC_ACQUIRE(xprt);
1160                                 xprt->xp_thread = st;
1161                                 st->st_xprt = xprt;
1162                                 continue;
1163                         }
1164
1165                         LIST_INSERT_HEAD(&grp->sg_idlethreads, st, st_ilink);
1166                         if (ismaster || (!ismaster &&
1167                             grp->sg_threadcount > grp->sg_minthreads))
1168                                 error = cv_timedwait_sig(&st->st_cond,
1169                                     &grp->sg_lock, 5 * hz);
1170                         else
1171                                 error = cv_wait_sig(&st->st_cond,
1172                                     &grp->sg_lock);
1173                         if (st->st_xprt == NULL)
1174                                 LIST_REMOVE(st, st_ilink);
1175
1176                         /*
1177                          * Reduce worker thread count when idle.
1178                          */
1179                         if (error == EWOULDBLOCK) {
1180                                 if (!ismaster
1181                                     && (grp->sg_threadcount
1182                                         > grp->sg_minthreads)
1183                                         && !st->st_xprt)
1184                                         break;
1185                         } else if (error) {
1186                                 mtx_unlock(&grp->sg_lock);
1187                                 svc_exit(pool);
1188                                 mtx_lock(&grp->sg_lock);
1189                                 break;
1190                         }
1191                         continue;
1192                 }
1193                 mtx_unlock(&grp->sg_lock);
1194
1195                 /*
1196                  * Drain the transport socket and queue up any RPCs.
1197                  */
1198                 xprt->xp_lastactive = time_uptime;
1199                 do {
1200                         if (!svc_request_space_available(pool))
1201                                 break;
1202                         rqstp = NULL;
1203                         stat = svc_getreq(xprt, &rqstp);
1204                         if (rqstp) {
1205                                 svc_change_space_used(pool, rqstp->rq_size);
1206                                 /*
1207                                  * See if the application has a preference
1208                                  * for some other thread.
1209                                  */
1210                                 if (pool->sp_assign) {
1211                                         stpref = pool->sp_assign(st, rqstp);
1212                                         rqstp->rq_thread = stpref;
1213                                         STAILQ_INSERT_TAIL(&stpref->st_reqs,
1214                                             rqstp, rq_link);
1215                                         mtx_unlock(&stpref->st_lock);
1216                                         if (stpref != st)
1217                                                 rqstp = NULL;
1218                                 } else {
1219                                         rqstp->rq_thread = st;
1220                                         STAILQ_INSERT_TAIL(&st->st_reqs,
1221                                             rqstp, rq_link);
1222                                 }
1223                         }
1224                 } while (rqstp == NULL && stat == XPRT_MOREREQS
1225                     && grp->sg_state != SVCPOOL_CLOSING);
1226
1227                 /*
1228                  * Move this transport to the end of the active list to
1229                  * ensure fairness when multiple transports are active.
1230                  * If this was the last queued request, svc_getreq will end
1231                  * up calling xprt_inactive to remove from the active list.
1232                  */
1233                 mtx_lock(&grp->sg_lock);
1234                 xprt->xp_thread = NULL;
1235                 st->st_xprt = NULL;
1236                 if (xprt->xp_active) {
1237                         if (!svc_request_space_available(pool) ||
1238                             !xprt_assignthread(xprt))
1239                                 TAILQ_INSERT_TAIL(&grp->sg_active,
1240                                     xprt, xp_alink);
1241                 }
1242                 mtx_unlock(&grp->sg_lock);
1243                 SVC_RELEASE(xprt);
1244
1245                 /*
1246                  * Execute what we have queued.
1247                  */
1248                 sz = 0;
1249                 mtx_lock(&st->st_lock);
1250                 while ((rqstp = STAILQ_FIRST(&st->st_reqs)) != NULL) {
1251                         STAILQ_REMOVE_HEAD(&st->st_reqs, rq_link);
1252                         mtx_unlock(&st->st_lock);
1253                         sz += rqstp->rq_size;
1254                         svc_executereq(rqstp);
1255                         mtx_lock(&st->st_lock);
1256                 }
1257                 mtx_unlock(&st->st_lock);
1258                 svc_change_space_used(pool, -sz);
1259                 mtx_lock(&grp->sg_lock);
1260         }
1261
1262         if (st->st_xprt) {
1263                 xprt = st->st_xprt;
1264                 st->st_xprt = NULL;
1265                 SVC_RELEASE(xprt);
1266         }
1267         KASSERT(STAILQ_EMPTY(&st->st_reqs), ("stray reqs on exit"));
1268         mtx_destroy(&st->st_lock);
1269         cv_destroy(&st->st_cond);
1270         mem_free(st, sizeof(*st));
1271
1272         grp->sg_threadcount--;
1273         if (!ismaster)
1274                 wakeup(grp);
1275         mtx_unlock(&grp->sg_lock);
1276 }
1277
1278 static void
1279 svc_thread_start(void *arg)
1280 {
1281
1282         svc_run_internal((SVCGROUP *) arg, FALSE);
1283         kthread_exit();
1284 }
1285
1286 static void
1287 svc_new_thread(SVCGROUP *grp)
1288 {
1289         SVCPOOL *pool = grp->sg_pool;
1290         struct thread *td;
1291
1292         mtx_lock(&grp->sg_lock);
1293         grp->sg_threadcount++;
1294         mtx_unlock(&grp->sg_lock);
1295         kthread_add(svc_thread_start, grp, pool->sp_proc, &td, 0, 0,
1296             "%s: service", pool->sp_name);
1297 }
1298
1299 void
1300 svc_run(SVCPOOL *pool)
1301 {
1302         int g, i;
1303         struct proc *p;
1304         struct thread *td;
1305         SVCGROUP *grp;
1306
1307         p = curproc;
1308         td = curthread;
1309         snprintf(td->td_name, sizeof(td->td_name),
1310             "%s: master", pool->sp_name);
1311         pool->sp_state = SVCPOOL_ACTIVE;
1312         pool->sp_proc = p;
1313
1314         /* Choose group count based on number of threads and CPUs. */
1315         pool->sp_groupcount = max(1, min(SVC_MAXGROUPS,
1316             min(pool->sp_maxthreads / 2, mp_ncpus) / 6));
1317         for (g = 0; g < pool->sp_groupcount; g++) {
1318                 grp = &pool->sp_groups[g];
1319                 grp->sg_minthreads = max(1,
1320                     pool->sp_minthreads / pool->sp_groupcount);
1321                 grp->sg_maxthreads = max(1,
1322                     pool->sp_maxthreads / pool->sp_groupcount);
1323                 grp->sg_lastcreatetime = time_uptime;
1324         }
1325
1326         /* Starting threads */
1327         pool->sp_groups[0].sg_threadcount++;
1328         for (g = 0; g < pool->sp_groupcount; g++) {
1329                 grp = &pool->sp_groups[g];
1330                 for (i = ((g == 0) ? 1 : 0); i < grp->sg_minthreads; i++)
1331                         svc_new_thread(grp);
1332         }
1333         svc_run_internal(&pool->sp_groups[0], TRUE);
1334
1335         /* Waiting for threads to stop. */
1336         for (g = 0; g < pool->sp_groupcount; g++) {
1337                 grp = &pool->sp_groups[g];
1338                 mtx_lock(&grp->sg_lock);
1339                 while (grp->sg_threadcount > 0)
1340                         msleep(grp, &grp->sg_lock, 0, "svcexit", 0);
1341                 mtx_unlock(&grp->sg_lock);
1342         }
1343 }
1344
1345 void
1346 svc_exit(SVCPOOL *pool)
1347 {
1348         SVCGROUP *grp;
1349         SVCTHREAD *st;
1350         int g;
1351
1352         pool->sp_state = SVCPOOL_CLOSING;
1353         for (g = 0; g < pool->sp_groupcount; g++) {
1354                 grp = &pool->sp_groups[g];
1355                 mtx_lock(&grp->sg_lock);
1356                 if (grp->sg_state != SVCPOOL_CLOSING) {
1357                         grp->sg_state = SVCPOOL_CLOSING;
1358                         LIST_FOREACH(st, &grp->sg_idlethreads, st_ilink)
1359                                 cv_signal(&st->st_cond);
1360                 }
1361                 mtx_unlock(&grp->sg_lock);
1362         }
1363 }
1364
1365 bool_t
1366 svc_getargs(struct svc_req *rqstp, xdrproc_t xargs, void *args)
1367 {
1368         struct mbuf *m;
1369         XDR xdrs;
1370         bool_t stat;
1371
1372         m = rqstp->rq_args;
1373         rqstp->rq_args = NULL;
1374
1375         xdrmbuf_create(&xdrs, m, XDR_DECODE);
1376         stat = xargs(&xdrs, args);
1377         XDR_DESTROY(&xdrs);
1378
1379         return (stat);
1380 }
1381
1382 bool_t
1383 svc_freeargs(struct svc_req *rqstp, xdrproc_t xargs, void *args)
1384 {
1385         XDR xdrs;
1386
1387         if (rqstp->rq_addr) {
1388                 free(rqstp->rq_addr, M_SONAME);
1389                 rqstp->rq_addr = NULL;
1390         }
1391
1392         xdrs.x_op = XDR_FREE;
1393         return (xargs(&xdrs, args));
1394 }
1395
1396 void
1397 svc_freereq(struct svc_req *rqstp)
1398 {
1399         SVCTHREAD *st;
1400         SVCPOOL *pool;
1401
1402         st = rqstp->rq_thread;
1403         if (st) {
1404                 pool = st->st_pool;
1405                 if (pool->sp_done)
1406                         pool->sp_done(st, rqstp);
1407         }
1408
1409         if (rqstp->rq_auth.svc_ah_ops)
1410                 SVCAUTH_RELEASE(&rqstp->rq_auth);
1411
1412         if (rqstp->rq_xprt) {
1413                 SVC_RELEASE(rqstp->rq_xprt);
1414         }
1415
1416         if (rqstp->rq_addr)
1417                 free(rqstp->rq_addr, M_SONAME);
1418
1419         if (rqstp->rq_args)
1420                 m_freem(rqstp->rq_args);
1421
1422         free(rqstp, M_RPC);
1423 }