1 /*
   2  * Copyright (c) 2008 Isilon Inc http://www.isilon.com/
   3  * Authors: Doug Rabson <dfr@rabson.org>
   4  * Developed with Red Inc: Alfred Perlstein <alfred@freebsd.org>
   5  *
   6  * Redistribution and use in source and binary forms, with or without
   7  * modification, are permitted provided that the following conditions
   8  * are met:
   9  * 1. Redistributions of source code must retain the above copyright
  10  *    notice, this list of conditions and the following disclaimer.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  *
  15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  25  * SUCH DAMAGE.
  26  */
  27 
  28 /*
  29  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  30  * Copyright (c) 2012 by Delphix. All rights reserved.
  31  */
  32 
  33 /*
  34  * NFS LockManager, start/stop, support functions, etc.
  35  * Most of the interesting code is here.
  36  *
  37  * Source code derived from FreeBSD nlm_prot_impl.c
  38  */
  39 
  40 #include <sys/param.h>
  41 #include <sys/systm.h>
  42 #include <sys/thread.h>
  43 #include <sys/fcntl.h>
  44 #include <sys/flock.h>
  45 #include <sys/mount.h>
  46 #include <sys/priv.h>
  47 #include <sys/proc.h>
  48 #include <sys/share.h>
  49 #include <sys/socket.h>
  50 #include <sys/syscall.h>
  51 #include <sys/syslog.h>
  52 #include <sys/systm.h>
  53 #include <sys/class.h>
  54 #include <sys/unistd.h>
  55 #include <sys/vnode.h>
  56 #include <sys/vfs.h>
  57 #include <sys/queue.h>
  58 #include <sys/bitmap.h>
  59 #include <sys/sdt.h>
  60 #include <netinet/in.h>
  61 
  62 #include <rpc/rpc.h>
  63 #include <rpc/xdr.h>
  64 #include <rpc/pmap_prot.h>
  65 #include <rpc/pmap_clnt.h>
  66 #include <rpc/rpcb_prot.h>
  67 
  68 #include <rpcsvc/nlm_prot.h>
  69 #include <rpcsvc/sm_inter.h>
  70 #include <rpcsvc/nsm_addr.h>
  71 
  72 #include <nfs/nfs.h>
  73 #include <nfs/nfs_clnt.h>
  74 #include <nfs/export.h>
  75 #include <nfs/rnode.h>
  76 #include <nfs/lm.h>
  77 
  78 #include "nlm_impl.h"
  79 
  80 struct nlm_knc {
  81         struct knetconfig       n_knc;
  82         const char              *n_netid;
  83 };
  84 
  85 /*
  86  * Number of attempts NLM tries to obtain RPC binding
  87  * of local statd.
  88  */
  89 #define NLM_NSM_RPCBIND_RETRIES 10
  90 
  91 /*
  92  * Timeout (in seconds) NLM waits before making another
  93  * attempt to obtain RPC binding of local statd.
  94  */
  95 #define NLM_NSM_RPCBIND_TIMEOUT 5
  96 
  97 /*
  98  * Total number of sysids in NLM sysid bitmap
  99  */
 100 #define NLM_BMAP_NITEMS (LM_SYSID_MAX + 1)
 101 
 102 /*
 103  * Number of ulong_t words in bitmap that is used
 104  * for allocation of sysid numbers.
 105  */
 106 #define NLM_BMAP_WORDS  (NLM_BMAP_NITEMS / BT_NBIPUL)
 107 
 108 /*
 109  * Given an integer x, the macro returns
 110  * -1 if x is negative,
 111  *  0 if x is zero
 112  *  1 if x is positive
 113  */
 114 #define SIGN(x) (((x) > 0) - ((x) < 0))
 115 
 116 #define ARRSIZE(arr)    (sizeof (arr) / sizeof ((arr)[0]))
 117 #define NLM_KNCS        ARRSIZE(nlm_netconfigs)
 118 
 119 krwlock_t lm_lck;
 120 
 121 /*
 122  * Zero timeout for asynchronous NLM RPC operations
 123  */
 124 static const struct timeval nlm_rpctv_zero = { 0,  0 };
 125 
 126 /*
 127  * List of all Zone globals nlm_globals instences
 128  * linked together.
 129  */
 130 static struct nlm_globals_list nlm_zones_list; /* (g) */
 131 
 132 /*
 133  * NLM kmem caches
 134  */
 135 static struct kmem_cache *nlm_hosts_cache = NULL;
 136 static struct kmem_cache *nlm_vhold_cache = NULL;
 137 
 138 /*
 139  * A bitmap for allocation of new sysids.
 140  * Sysid is a unique number between LM_SYSID
 141  * and LM_SYSID_MAX. Sysid represents unique remote
 142  * host that does file locks on the given host.
 143  */
 144 static ulong_t  nlm_sysid_bmap[NLM_BMAP_WORDS]; /* (g) */
 145 static int      nlm_sysid_nidx;                 /* (g) */
 146 
 147 /*
 148  * RPC service registration for all transports
 149  */
 150 static SVC_CALLOUT nlm_svcs[] = {
 151         { NLM_PROG, 4, 4, nlm_prog_4 }, /* NLM4_VERS */
 152         { NLM_PROG, 1, 3, nlm_prog_3 }  /* NLM_VERS - NLM_VERSX */
 153 };
 154 
 155 static SVC_CALLOUT_TABLE nlm_sct = {
 156         ARRSIZE(nlm_svcs),
 157         FALSE,
 158         nlm_svcs
 159 };
 160 
 161 /*
 162  * Static table of all netid/knetconfig network
 163  * lock manager can work with. nlm_netconfigs table
 164  * is used when we need to get valid knetconfig by
 165  * netid and vice versa.
 166  *
 167  * Knetconfigs are activated either by the call from
 168  * user-space lockd daemon (server side) or by taking
 169  * knetconfig from NFS mountinfo (client side)
 170  */
 171 static struct nlm_knc nlm_netconfigs[] = { /* (g) */
 172         /* UDP */
 173         {
 174                 { NC_TPI_CLTS, NC_INET, NC_UDP, NODEV },
 175                 "udp",
 176         },
 177         /* TCP */
 178         {
 179                 { NC_TPI_COTS_ORD, NC_INET, NC_TCP, NODEV },
 180                 "tcp",
 181         },
 182         /* UDP over IPv6 */
 183         {
 184                 { NC_TPI_CLTS, NC_INET6, NC_UDP, NODEV },
 185                 "udp6",
 186         },
 187         /* TCP over IPv6 */
 188         {
 189                 { NC_TPI_COTS_ORD, NC_INET6, NC_TCP, NODEV },
 190                 "tcp6",
 191         },
 192         /* ticlts (loopback over UDP) */
 193         {
 194                 { NC_TPI_CLTS, NC_LOOPBACK, NC_NOPROTO, NODEV },
 195                 "ticlts",
 196         },
 197         /* ticotsord (loopback over TCP) */
 198         {
 199                 { NC_TPI_COTS_ORD, NC_LOOPBACK, NC_NOPROTO, NODEV },
 200                 "ticotsord",
 201         },
 202 };
 203 
 204 /*
 205  * NLM misc. function
 206  */
 207 static void nlm_copy_netbuf(struct netbuf *, struct netbuf *);
 208 static int nlm_netbuf_addrs_cmp(struct netbuf *, struct netbuf *);
 209 static void nlm_kmem_reclaim(void *);
 210 static void nlm_pool_shutdown(void);
 211 static void nlm_suspend_zone(struct nlm_globals *);
 212 static void nlm_resume_zone(struct nlm_globals *);
 213 static void nlm_nsm_clnt_init(CLIENT *, struct nlm_nsm *);
 214 static void nlm_netbuf_to_netobj(struct netbuf *, int *, netobj *);
 215 
 216 /*
 217  * NLM thread functions
 218  */
 219 static void nlm_gc(struct nlm_globals *);
 220 static void nlm_reclaimer(struct nlm_host *);
 221 
 222 /*
 223  * NLM NSM functions
 224  */
 225 static int nlm_init_local_knc(struct knetconfig *);
 226 static int nlm_nsm_init_local(struct nlm_nsm *);
 227 static int nlm_nsm_init(struct nlm_nsm *, struct knetconfig *, struct netbuf *);
 228 static void nlm_nsm_fini(struct nlm_nsm *);
 229 static enum clnt_stat nlm_nsm_simu_crash(struct nlm_nsm *);
 230 static enum clnt_stat nlm_nsm_stat(struct nlm_nsm *, int32_t *);
 231 static enum clnt_stat nlm_nsm_mon(struct nlm_nsm *, char *, uint16_t);
 232 static enum clnt_stat nlm_nsm_unmon(struct nlm_nsm *, char *);
 233 
 234 /*
 235  * NLM host functions
 236  */
 237 static int nlm_host_ctor(void *, void *, int);
 238 static void nlm_host_dtor(void *, void *);
 239 static void nlm_host_destroy(struct nlm_host *);
 240 static struct nlm_host *nlm_host_create(char *, const char *,
 241     struct knetconfig *, struct netbuf *);
 242 static struct nlm_host *nlm_host_find_locked(struct nlm_globals *,
 243     const char *, struct netbuf *, avl_index_t *);
 244 static void nlm_host_unregister(struct nlm_globals *, struct nlm_host *);
 245 static void nlm_host_gc_vholds(struct nlm_host *);
 246 static bool_t nlm_host_has_srv_locks(struct nlm_host *);
 247 static bool_t nlm_host_has_cli_locks(struct nlm_host *);
 248 static bool_t nlm_host_has_locks(struct nlm_host *);
 249 
 250 /*
 251  * NLM vhold functions
 252  */
 253 static int nlm_vhold_ctor(void *, void *, int);
 254 static void nlm_vhold_dtor(void *, void *);
 255 static void nlm_vhold_destroy(struct nlm_host *,
 256     struct nlm_vhold *);
 257 static bool_t nlm_vhold_busy(struct nlm_host *, struct nlm_vhold *);
 258 static void nlm_vhold_clean(struct nlm_vhold *, int);
 259 
 260 /*
 261  * NLM client/server sleeping locks/share reservation functions
 262  */
 263 struct nlm_slreq *nlm_slreq_find_locked(struct nlm_host *,
 264     struct nlm_vhold *, struct flock64 *);
 265 static struct nlm_shres *nlm_shres_create_item(struct shrlock *, vnode_t *);
 266 static void nlm_shres_destroy_item(struct nlm_shres *);
 267 static bool_t nlm_shres_equal(struct shrlock *, struct shrlock *);
 268 
 269 /*
 270  * NLM initialization functions.
 271  */
 272 void
 273 nlm_init(void)
 274 {
 275         nlm_hosts_cache = kmem_cache_create("nlm_host_cache",
 276             sizeof (struct nlm_host), 0, nlm_host_ctor, nlm_host_dtor,
 277             nlm_kmem_reclaim, NULL, NULL, 0);
 278 
 279         nlm_vhold_cache = kmem_cache_create("nlm_vhold_cache",
 280             sizeof (struct nlm_vhold), 0, nlm_vhold_ctor, nlm_vhold_dtor,
 281             NULL, NULL, NULL, 0);
 282 
 283         nlm_rpc_init();
 284         TAILQ_INIT(&nlm_zones_list);
 285 
 286         /* initialize sysids bitmap */
 287         bzero(nlm_sysid_bmap, sizeof (nlm_sysid_bmap));
 288         nlm_sysid_nidx = 1;
 289 
 290         /*
 291          * Reserv the sysid #0, because it's associated
 292          * with local locks only. Don't let to allocate
 293          * it for remote locks.
 294          */
 295         BT_SET(nlm_sysid_bmap, 0);
 296 }
 297 
 298 void
 299 nlm_globals_register(struct nlm_globals *g)
 300 {
 301         rw_enter(&lm_lck, RW_WRITER);
 302         TAILQ_INSERT_TAIL(&nlm_zones_list, g, nlm_link);
 303         rw_exit(&lm_lck);
 304 }
 305 
 306 void
 307 nlm_globals_unregister(struct nlm_globals *g)
 308 {
 309         rw_enter(&lm_lck, RW_WRITER);
 310         TAILQ_REMOVE(&nlm_zones_list, g, nlm_link);
 311         rw_exit(&lm_lck);
 312 }
 313 
 314 /* ARGSUSED */
 315 static void
 316 nlm_kmem_reclaim(void *cdrarg)
 317 {
 318         struct nlm_globals *g;
 319 
 320         rw_enter(&lm_lck, RW_READER);
 321         TAILQ_FOREACH(g, &nlm_zones_list, nlm_link)
 322                 cv_broadcast(&g->nlm_gc_sched_cv);
 323 
 324         rw_exit(&lm_lck);
 325 }
 326 
 327 /*
 328  * NLM garbage collector thread (GC).
 329  *
 330  * NLM GC periodically checks whether there're any host objects
 331  * that can be cleaned up. It also releases stale vnodes that
 332  * live on the server side (under protection of vhold objects).
 333  *
 334  * NLM host objects are cleaned up from GC thread because
 335  * operations helping us to determine whether given host has
 336  * any locks can be quite expensive and it's not good to call
 337  * them every time the very last reference to the host is dropped.
 338  * Thus we use "lazy" approach for hosts cleanup.
 339  *
 340  * The work of GC is to release stale vnodes on the server side
 341  * and destroy hosts that haven't any locks and any activity for
 342  * some time (i.e. idle hosts).
 343  */
 344 static void
 345 nlm_gc(struct nlm_globals *g)
 346 {
 347         struct nlm_host *hostp;
 348         clock_t now, idle_period;
 349 
 350         idle_period = SEC_TO_TICK(g->cn_idle_tmo);
 351         mutex_enter(&g->lock);
 352         for (;;) {
 353                 /*
 354                  * GC thread can be explicitly scheduled from
 355                  * memory reclamation function.
 356                  */
 357                 (void) cv_timedwait(&g->nlm_gc_sched_cv, &g->lock,
 358                     ddi_get_lbolt() + idle_period);
 359 
 360                 /*
 361                  * NLM is shutting down, time to die.
 362                  */
 363                 if (g->run_status == NLM_ST_STOPPING)
 364                         break;
 365 
 366                 now = ddi_get_lbolt();
 367                 DTRACE_PROBE2(gc__start, struct nlm_globals *, g,
 368                     clock_t, now);
 369 
 370                 /*
 371                  * Find all obviously unused vholds and destroy them.
 372                  */
 373                 for (hostp = avl_first(&g->nlm_hosts_tree); hostp != NULL;
 374                     hostp = AVL_NEXT(&g->nlm_hosts_tree, hostp)) {
 375                         struct nlm_vhold *nvp;
 376 
 377                         mutex_enter(&hostp->nh_lock);
 378 
 379                         nvp = TAILQ_FIRST(&hostp->nh_vholds_list);
 380                         while (nvp != NULL) {
 381                                 struct nlm_vhold *new_nvp;
 382 
 383                                 new_nvp = TAILQ_NEXT(nvp, nv_link);
 384 
 385                                 /*
 386                                  * If these conditions are met, the vhold is
 387                                  * obviously unused and we will destroy it.  In
 388                                  * a case either v_filocks and/or v_shrlocks is
 389                                  * non-NULL the vhold might still be unused by
 390                                  * the host, but it is expensive to check that.
 391                                  * We defer such check until the host is idle.
 392                                  * The expensive check is done below without
 393                                  * the global lock held.
 394                                  */
 395                                 if (nvp->nv_refcnt == 0 &&
 396                                     nvp->nv_vp->v_filocks == NULL &&
 397                                     nvp->nv_vp->v_shrlocks == NULL) {
 398                                         nlm_vhold_destroy(hostp, nvp);
 399                                 }
 400 
 401                                 nvp = new_nvp;
 402                         }
 403 
 404                         mutex_exit(&hostp->nh_lock);
 405                 }
 406 
 407                 /*
 408                  * Handle all hosts that are unused at the moment
 409                  * until we meet one with idle timeout in future.
 410                  */
 411                 while ((hostp = TAILQ_FIRST(&g->nlm_idle_hosts)) != NULL) {
 412                         bool_t has_locks;
 413 
 414                         if (hostp->nh_idle_timeout > now)
 415                                 break;
 416 
 417                         /*
 418                          * Drop global lock while doing expensive work
 419                          * on this host. We'll re-check any conditions
 420                          * that might change after retaking the global
 421                          * lock.
 422                          */
 423                         mutex_exit(&g->lock);
 424                         mutex_enter(&hostp->nh_lock);
 425 
 426                         /*
 427                          * nlm_globals lock was dropped earlier because
 428                          * garbage collecting of vholds and checking whether
 429                          * host has any locks/shares are expensive operations.
 430                          */
 431                         nlm_host_gc_vholds(hostp);
 432                         has_locks = nlm_host_has_locks(hostp);
 433 
 434                         mutex_exit(&hostp->nh_lock);
 435                         mutex_enter(&g->lock);
 436 
 437                         /*
 438                          * While we were doing expensive operations
 439                          * outside of nlm_globals critical section,
 440                          * somebody could take the host and remove it
 441                          * from the idle list.  Whether its been
 442                          * reinserted or not, our information about
 443                          * the host is outdated, and we should take no
 444                          * further action.
 445                          */
 446                         if ((hostp->nh_flags & NLM_NH_INIDLE) == 0 ||
 447                             hostp->nh_idle_timeout > now)
 448                                 continue;
 449 
 450                         /*
 451                          * If the host has locks we have to renew the
 452                          * host's timeout and put it at the end of LRU
 453                          * list.
 454                          */
 455                         if (has_locks) {
 456                                 TAILQ_REMOVE(&g->nlm_idle_hosts,
 457                                     hostp, nh_link);
 458                                 hostp->nh_idle_timeout = now + idle_period;
 459                                 TAILQ_INSERT_TAIL(&g->nlm_idle_hosts,
 460                                     hostp, nh_link);
 461                                 continue;
 462                         }
 463 
 464                         /*
 465                          * We're here if all the following conditions hold:
 466                          * 1) Host hasn't any locks or share reservations
 467                          * 2) Host is unused
 468                          * 3) Host wasn't touched by anyone at least for
 469                          *    g->cn_idle_tmo seconds.
 470                          *
 471                          * So, now we can destroy it.
 472                          */
 473                         nlm_host_unregister(g, hostp);
 474                         mutex_exit(&g->lock);
 475 
 476                         nlm_host_unmonitor(g, hostp);
 477                         nlm_host_destroy(hostp);
 478                         mutex_enter(&g->lock);
 479                         if (g->run_status == NLM_ST_STOPPING)
 480                                 break;
 481 
 482                 }
 483 
 484                 DTRACE_PROBE(gc__end);
 485         }
 486 
 487         DTRACE_PROBE1(gc__exit, struct nlm_globals *, g);
 488 
 489         /* Let others know that GC has died */
 490         g->nlm_gc_thread = NULL;
 491         mutex_exit(&g->lock);
 492 
 493         cv_broadcast(&g->nlm_gc_finish_cv);
 494         zthread_exit();
 495 }
 496 
 497 /*
 498  * Thread reclaim locks/shares acquired by the client side
 499  * on the given server represented by hostp.
 500  */
 501 static void
 502 nlm_reclaimer(struct nlm_host *hostp)
 503 {
 504         struct nlm_globals *g;
 505 
 506         mutex_enter(&hostp->nh_lock);
 507         hostp->nh_reclaimer = curthread;
 508         mutex_exit(&hostp->nh_lock);
 509 
 510         g = zone_getspecific(nlm_zone_key, curzone);
 511         nlm_reclaim_client(g, hostp);
 512 
 513         mutex_enter(&hostp->nh_lock);
 514         hostp->nh_flags &= ~NLM_NH_RECLAIM;
 515         hostp->nh_reclaimer = NULL;
 516         cv_broadcast(&hostp->nh_recl_cv);
 517         mutex_exit(&hostp->nh_lock);
 518 
 519         /*
 520          * Host was explicitly referenced before
 521          * nlm_reclaim() was called, release it
 522          * here.
 523          */
 524         nlm_host_release(g, hostp);
 525         zthread_exit();
 526 }
 527 
 528 /*
 529  * Copy a struct netobj.  (see xdr.h)
 530  */
 531 void
 532 nlm_copy_netobj(struct netobj *dst, struct netobj *src)
 533 {
 534         dst->n_len = src->n_len;
 535         dst->n_bytes = kmem_alloc(src->n_len, KM_SLEEP);
 536         bcopy(src->n_bytes, dst->n_bytes, src->n_len);
 537 }
 538 
 539 /*
 540  * An NLM specificw replacement for clnt_call().
 541  * nlm_clnt_call() is used by all RPC functions generated
 542  * from nlm_prot.x specification. The function is aware
 543  * about some pitfalls of NLM RPC procedures and has a logic
 544  * that handles them properly.
 545  */
 546 enum clnt_stat
 547 nlm_clnt_call(CLIENT *clnt, rpcproc_t procnum, xdrproc_t xdr_args,
 548     caddr_t argsp, xdrproc_t xdr_result, caddr_t resultp, struct timeval wait)
 549 {
 550         k_sigset_t oldmask;
 551         enum clnt_stat stat;
 552         bool_t sig_blocked = FALSE;
 553 
 554         /*
 555          * If NLM RPC procnum is one of the NLM _RES procedures
 556          * that are used to reply to asynchronous NLM RPC
 557          * (MSG calls), explicitly set RPC timeout to zero.
 558          * Client doesn't send a reply to RES procedures, so
 559          * we don't need to wait anything.
 560          *
 561          * NOTE: we ignore NLM4_*_RES procnums because they are
 562          * equal to NLM_*_RES numbers.
 563          */
 564         if (procnum >= NLM_TEST_RES && procnum <= NLM_GRANTED_RES)
 565                 wait = nlm_rpctv_zero;
 566 
 567         /*
 568          * We need to block signals in case of NLM_CANCEL RPC
 569          * in order to prevent interruption of network RPC
 570          * calls.
 571          */
 572         if (procnum == NLM_CANCEL) {
 573                 k_sigset_t newmask;
 574 
 575                 sigfillset(&newmask);
 576                 sigreplace(&newmask, &oldmask);
 577                 sig_blocked = TRUE;
 578         }
 579 
 580         stat = clnt_call(clnt, procnum, xdr_args,
 581             argsp, xdr_result, resultp, wait);
 582 
 583         /*
 584          * Restore signal mask back if signals were blocked
 585          */
 586         if (sig_blocked)
 587                 sigreplace(&oldmask, (k_sigset_t *)NULL);
 588 
 589         return (stat);
 590 }
 591 
 592 /*
 593  * Suspend NLM client/server in the given zone.
 594  *
 595  * During suspend operation we mark those hosts
 596  * that have any locks with NLM_NH_SUSPEND flags,
 597  * so that they can be checked later, when resume
 598  * operation occurs.
 599  */
 600 static void
 601 nlm_suspend_zone(struct nlm_globals *g)
 602 {
 603         struct nlm_host *hostp;
 604         struct nlm_host_list all_hosts;
 605 
 606         /*
 607          * Note that while we're doing suspend, GC thread is active
 608          * and it can destroy some hosts while we're walking through
 609          * the hosts tree. To prevent that and make suspend logic
 610          * a bit more simple we put all hosts to local "all_hosts"
 611          * list and increment reference counter of each host.
 612          * This guaranties that no hosts will be released while
 613          * we're doing suspend.
 614          * NOTE: reference of each host must be dropped during
 615          * resume operation.
 616          */
 617         TAILQ_INIT(&all_hosts);
 618         mutex_enter(&g->lock);
 619         for (hostp = avl_first(&g->nlm_hosts_tree); hostp != NULL;
 620             hostp = AVL_NEXT(&g->nlm_hosts_tree, hostp)) {
 621                 /*
 622                  * If host is idle, remove it from idle list and
 623                  * clear idle flag. That is done to prevent GC
 624                  * from touching this host.
 625                  */
 626                 if (hostp->nh_flags & NLM_NH_INIDLE) {
 627                         TAILQ_REMOVE(&g->nlm_idle_hosts, hostp, nh_link);
 628                         hostp->nh_flags &= ~NLM_NH_INIDLE;
 629                 }
 630 
 631                 hostp->nh_refs++;
 632                 TAILQ_INSERT_TAIL(&all_hosts, hostp, nh_link);
 633         }
 634 
 635         /*
 636          * Now we can walk through all hosts on the system
 637          * with zone globals lock released. The fact the
 638          * we have taken a reference to each host guaranties
 639          * that no hosts can be destroyed during that process.
 640          */
 641         mutex_exit(&g->lock);
 642         while ((hostp = TAILQ_FIRST(&all_hosts)) != NULL) {
 643                 mutex_enter(&hostp->nh_lock);
 644                 if (nlm_host_has_locks(hostp))
 645                         hostp->nh_flags |= NLM_NH_SUSPEND;
 646 
 647                 mutex_exit(&hostp->nh_lock);
 648                 TAILQ_REMOVE(&all_hosts, hostp, nh_link);
 649         }
 650 }
 651 
 652 /*
 653  * Resume NLM hosts for the given zone.
 654  *
 655  * nlm_resume_zone() is called after hosts were suspended
 656  * (see nlm_suspend_zone) and its main purpose to check
 657  * whether remote locks owned by hosts are still in consistent
 658  * state. If they aren't, resume function tries to reclaim
 659  * locks (for client side hosts) and clean locks (for
 660  * server side hosts).
 661  */
 662 static void
 663 nlm_resume_zone(struct nlm_globals *g)
 664 {
 665         struct nlm_host *hostp, *h_next;
 666 
 667         mutex_enter(&g->lock);
 668         hostp = avl_first(&g->nlm_hosts_tree);
 669 
 670         /*
 671          * In nlm_suspend_zone() the reference counter of each
 672          * host was incremented, so we can safely iterate through
 673          * all hosts without worrying that any host we touch will
 674          * be removed at the moment.
 675          */
 676         while (hostp != NULL) {
 677                 struct nlm_nsm nsm;
 678                 enum clnt_stat stat;
 679                 int32_t sm_state;
 680                 int error;
 681                 bool_t resume_failed = FALSE;
 682 
 683                 h_next = AVL_NEXT(&g->nlm_hosts_tree, hostp);
 684                 mutex_exit(&g->lock);
 685 
 686                 DTRACE_PROBE1(resume__host, struct nlm_host *, hostp);
 687 
 688                 /*
 689                  * Suspend operation marked that the host doesn't
 690                  * have any locks. Skip it.
 691                  */
 692                 if (!(hostp->nh_flags & NLM_NH_SUSPEND))
 693                         goto cycle_end;
 694 
 695                 error = nlm_nsm_init(&nsm, &hostp->nh_knc, &hostp->nh_addr);
 696                 if (error != 0) {
 697                         NLM_ERR("Resume: Failed to contact to NSM of host %s "
 698                             "[error=%d]\n", hostp->nh_name, error);
 699                         resume_failed = TRUE;
 700                         goto cycle_end;
 701                 }
 702 
 703                 stat = nlm_nsm_stat(&nsm, &sm_state);
 704                 if (stat != RPC_SUCCESS) {
 705                         NLM_ERR("Resume: Failed to call SM_STAT operation for "
 706                             "host %s [stat=%d]\n", hostp->nh_name, stat);
 707                         resume_failed = TRUE;
 708                         nlm_nsm_fini(&nsm);
 709                         goto cycle_end;
 710                 }
 711 
 712                 if (sm_state != hostp->nh_state) {
 713                         /*
 714                          * Current SM state of the host isn't equal
 715                          * to the one host had when it was suspended.
 716                          * Probably it was rebooted. Try to reclaim
 717                          * locks if the host has any on its client side.
 718                          * Also try to clean up its server side locks
 719                          * (if the host has any).
 720                          */
 721                         nlm_host_notify_client(hostp, sm_state);
 722                         nlm_host_notify_server(hostp, sm_state);
 723                 }
 724 
 725                 nlm_nsm_fini(&nsm);
 726 
 727 cycle_end:
 728                 if (resume_failed) {
 729                         /*
 730                          * Resume failed for the given host.
 731                          * Just clean up all resources it owns.
 732                          */
 733                         nlm_host_notify_server(hostp, 0);
 734                         nlm_client_cancel_all(g, hostp);
 735                 }
 736 
 737                 hostp->nh_flags &= ~NLM_NH_SUSPEND;
 738                 nlm_host_release(g, hostp);
 739                 hostp = h_next;
 740                 mutex_enter(&g->lock);
 741         }
 742 
 743         mutex_exit(&g->lock);
 744 }
 745 
 746 /*
 747  * NLM functions responsible for operations on NSM handle.
 748  */
 749 
 750 /*
 751  * Initialize knetconfig that is used for communication
 752  * with local statd via loopback interface.
 753  */
 754 static int
 755 nlm_init_local_knc(struct knetconfig *knc)
 756 {
 757         int error;
 758         vnode_t *vp;
 759 
 760         bzero(knc, sizeof (*knc));
 761         error = lookupname("/dev/tcp", UIO_SYSSPACE,
 762             FOLLOW, NULLVPP, &vp);
 763         if (error != 0)
 764                 return (error);
 765 
 766         knc->knc_semantics = NC_TPI_COTS;
 767         knc->knc_protofmly = NC_INET;
 768         knc->knc_proto = NC_TCP;
 769         knc->knc_rdev = vp->v_rdev;
 770         VN_RELE(vp);
 771 
 772 
 773         return (0);
 774 }
 775 
 776 /*
 777  * Initialize NSM handle that will be used to talk
 778  * to local statd via loopback interface.
 779  */
 780 static int
 781 nlm_nsm_init_local(struct nlm_nsm *nsm)
 782 {
 783         int error;
 784         struct knetconfig knc;
 785         struct sockaddr_in sin;
 786         struct netbuf nb;
 787 
 788         error = nlm_init_local_knc(&knc);
 789         if (error != 0)
 790                 return (error);
 791 
 792         bzero(&sin, sizeof (sin));
 793         sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
 794         sin.sin_family = AF_INET;
 795 
 796         nb.buf = (char *)&sin;
 797         nb.len = nb.maxlen = sizeof (sin);
 798 
 799         return (nlm_nsm_init(nsm, &knc, &nb));
 800 }
 801 
 802 /*
 803  * Initialize NSM handle used for talking to statd
 804  */
 805 static int
 806 nlm_nsm_init(struct nlm_nsm *nsm, struct knetconfig *knc, struct netbuf *nb)
 807 {
 808         enum clnt_stat stat;
 809         int error, retries;
 810 
 811         bzero(nsm, sizeof (*nsm));
 812         nsm->ns_knc = *knc;
 813         nlm_copy_netbuf(&nsm->ns_addr, nb);
 814 
 815         /*
 816          * Try several times to get the port of statd service,
 817          * If rpcbind_getaddr returns  RPC_PROGNOTREGISTERED,
 818          * retry an attempt, but wait for NLM_NSM_RPCBIND_TIMEOUT
 819          * seconds berofore.
 820          */
 821         for (retries = 0; retries < NLM_NSM_RPCBIND_RETRIES; retries++) {
 822                 stat = rpcbind_getaddr(&nsm->ns_knc, SM_PROG,
 823                     SM_VERS, &nsm->ns_addr);
 824                 if (stat != RPC_SUCCESS) {
 825                         if (stat == RPC_PROGNOTREGISTERED) {
 826                                 delay(SEC_TO_TICK(NLM_NSM_RPCBIND_TIMEOUT));
 827                                 continue;
 828                         }
 829                 }
 830 
 831                 break;
 832         }
 833 
 834         if (stat != RPC_SUCCESS) {
 835                 DTRACE_PROBE2(rpcbind__error, enum clnt_stat, stat,
 836                     int, retries);
 837                 error = ENOENT;
 838                 goto error;
 839         }
 840 
 841         /*
 842          * Create an RPC handle that'll be used for communication with local
 843          * statd using the status monitor protocol.
 844          */
 845         error = clnt_tli_kcreate(&nsm->ns_knc, &nsm->ns_addr, SM_PROG, SM_VERS,
 846             0, NLM_RPC_RETRIES, kcred, &nsm->ns_handle);
 847         if (error != 0)
 848                 goto error;
 849 
 850         /*
 851          * Create an RPC handle that'll be used for communication with the
 852          * local statd using the address registration protocol.
 853          */
 854         error = clnt_tli_kcreate(&nsm->ns_knc, &nsm->ns_addr, NSM_ADDR_PROGRAM,
 855             NSM_ADDR_V1, 0, NLM_RPC_RETRIES, kcred, &nsm->ns_addr_handle);
 856         if (error != 0)
 857                 goto error;
 858 
 859         sema_init(&nsm->ns_sem, 1, NULL, SEMA_DEFAULT, NULL);
 860         return (0);
 861 
 862 error:
 863         kmem_free(nsm->ns_addr.buf, nsm->ns_addr.maxlen);
 864         if (nsm->ns_handle)
 865                 CLNT_DESTROY(nsm->ns_handle);
 866 
 867         return (error);
 868 }
 869 
 870 static void
 871 nlm_nsm_fini(struct nlm_nsm *nsm)
 872 {
 873         kmem_free(nsm->ns_addr.buf, nsm->ns_addr.maxlen);
 874         CLNT_DESTROY(nsm->ns_addr_handle);
 875         nsm->ns_addr_handle = NULL;
 876         CLNT_DESTROY(nsm->ns_handle);
 877         nsm->ns_handle = NULL;
 878         sema_destroy(&nsm->ns_sem);
 879 }
 880 
 881 static enum clnt_stat
 882 nlm_nsm_simu_crash(struct nlm_nsm *nsm)
 883 {
 884         enum clnt_stat stat;
 885 
 886         sema_p(&nsm->ns_sem);
 887         nlm_nsm_clnt_init(nsm->ns_handle, nsm);
 888         stat = sm_simu_crash_1(NULL, NULL, nsm->ns_handle);
 889         sema_v(&nsm->ns_sem);
 890 
 891         return (stat);
 892 }
 893 
 894 static enum clnt_stat
 895 nlm_nsm_stat(struct nlm_nsm *nsm, int32_t *out_stat)
 896 {
 897         struct sm_name args;
 898         struct sm_stat_res res;
 899         enum clnt_stat stat;
 900 
 901         args.mon_name = uts_nodename();
 902         bzero(&res, sizeof (res));
 903 
 904         sema_p(&nsm->ns_sem);
 905         nlm_nsm_clnt_init(nsm->ns_handle, nsm);
 906         stat = sm_stat_1(&args, &res, nsm->ns_handle);
 907         sema_v(&nsm->ns_sem);
 908 
 909         if (stat == RPC_SUCCESS)
 910                 *out_stat = res.state;
 911 
 912         return (stat);
 913 }
 914 
 915 static enum clnt_stat
 916 nlm_nsm_mon(struct nlm_nsm *nsm, char *hostname, uint16_t priv)
 917 {
 918         struct mon args;
 919         struct sm_stat_res res;
 920         enum clnt_stat stat;
 921 
 922         bzero(&args, sizeof (args));
 923         bzero(&res, sizeof (res));
 924 
 925         args.mon_id.mon_name = hostname;
 926         args.mon_id.my_id.my_name = uts_nodename();
 927         args.mon_id.my_id.my_prog = NLM_PROG;
 928         args.mon_id.my_id.my_vers = NLM_SM;
 929         args.mon_id.my_id.my_proc = NLM_SM_NOTIFY1;
 930         bcopy(&priv, args.priv, sizeof (priv));
 931 
 932         sema_p(&nsm->ns_sem);
 933         nlm_nsm_clnt_init(nsm->ns_handle, nsm);
 934         stat = sm_mon_1(&args, &res, nsm->ns_handle);
 935         sema_v(&nsm->ns_sem);
 936 
 937         return (stat);
 938 }
 939 
 940 static enum clnt_stat
 941 nlm_nsm_unmon(struct nlm_nsm *nsm, char *hostname)
 942 {
 943         struct mon_id args;
 944         struct sm_stat res;
 945         enum clnt_stat stat;
 946 
 947         bzero(&args, sizeof (args));
 948         bzero(&res, sizeof (res));
 949 
 950         args.mon_name = hostname;
 951         args.my_id.my_name = uts_nodename();
 952         args.my_id.my_prog = NLM_PROG;
 953         args.my_id.my_vers = NLM_SM;
 954         args.my_id.my_proc = NLM_SM_NOTIFY1;
 955 
 956         sema_p(&nsm->ns_sem);
 957         nlm_nsm_clnt_init(nsm->ns_handle, nsm);
 958         stat = sm_unmon_1(&args, &res, nsm->ns_handle);
 959         sema_v(&nsm->ns_sem);
 960 
 961         return (stat);
 962 }
 963 
 964 static enum clnt_stat
 965 nlm_nsmaddr_reg(struct nlm_nsm *nsm, char *name, int family, netobj *address)
 966 {
 967         struct reg1args args = { 0 };
 968         struct reg1res res = { 0 };
 969         enum clnt_stat stat;
 970 
 971         args.family = family;
 972         args.name = name;
 973         args.address = *address;
 974 
 975         sema_p(&nsm->ns_sem);
 976         nlm_nsm_clnt_init(nsm->ns_addr_handle, nsm);
 977         stat = nsmaddrproc1_reg_1(&args, &res, nsm->ns_addr_handle);
 978         sema_v(&nsm->ns_sem);
 979 
 980         return (stat);
 981 }
 982 
 983 /*
 984  * Get NLM vhold object corresponding to vnode "vp".
 985  * If no such object was found, create a new one.
 986  *
 987  * The purpose of this function is to associate vhold
 988  * object with given vnode, so that:
 989  * 1) vnode is hold (VN_HOLD) while vhold object is alive.
 990  * 2) host has a track of all vnodes it touched by lock
 991  *    or share operations. These vnodes are accessible
 992  *    via collection of vhold objects.
 993  */
 994 struct nlm_vhold *
 995 nlm_vhold_get(struct nlm_host *hostp, vnode_t *vp)
 996 {
 997         struct nlm_vhold *nvp, *new_nvp = NULL;
 998 
 999         mutex_enter(&hostp->nh_lock);
1000         nvp = nlm_vhold_find_locked(hostp, vp);
1001         if (nvp != NULL)
1002                 goto out;
1003 
1004         /* nlm_vhold wasn't found, then create a new one */
1005         mutex_exit(&hostp->nh_lock);
1006         new_nvp = kmem_cache_alloc(nlm_vhold_cache, KM_SLEEP);
1007 
1008         /*
1009          * Check if another thread has already
1010          * created the same nlm_vhold.
1011          */
1012         mutex_enter(&hostp->nh_lock);
1013         nvp = nlm_vhold_find_locked(hostp, vp);
1014         if (nvp == NULL) {
1015                 nvp = new_nvp;
1016                 new_nvp = NULL;
1017 
1018                 TAILQ_INIT(&nvp->nv_slreqs);
1019                 nvp->nv_vp = vp;
1020                 nvp->nv_refcnt = 1;
1021                 VN_HOLD(nvp->nv_vp);
1022 
1023                 VERIFY(mod_hash_insert(hostp->nh_vholds_by_vp,
1024                     (mod_hash_key_t)vp, (mod_hash_val_t)nvp) == 0);
1025                 TAILQ_INSERT_TAIL(&hostp->nh_vholds_list, nvp, nv_link);
1026         }
1027 
1028 out:
1029         mutex_exit(&hostp->nh_lock);
1030         if (new_nvp != NULL)
1031                 kmem_cache_free(nlm_vhold_cache, new_nvp);
1032 
1033         return (nvp);
1034 }
1035 
1036 /*
1037  * Drop a reference to vhold object nvp.
1038  */
1039 void
1040 nlm_vhold_release(struct nlm_host *hostp, struct nlm_vhold *nvp)
1041 {
1042         if (nvp == NULL)
1043                 return;
1044 
1045         mutex_enter(&hostp->nh_lock);
1046         ASSERT(nvp->nv_refcnt > 0);
1047         nvp->nv_refcnt--;
1048 
1049         /*
1050          * If these conditions are met, the vhold is obviously unused and we
1051          * will destroy it.  In a case either v_filocks and/or v_shrlocks is
1052          * non-NULL the vhold might still be unused by the host, but it is
1053          * expensive to check that.  We defer such check until the host is
1054          * idle.  The expensive check is done in the NLM garbage collector.
1055          */
1056         if (nvp->nv_refcnt == 0 &&
1057             nvp->nv_vp->v_filocks == NULL &&
1058             nvp->nv_vp->v_shrlocks == NULL) {
1059                 nlm_vhold_destroy(hostp, nvp);
1060         }
1061 
1062         mutex_exit(&hostp->nh_lock);
1063 }
1064 
1065 /*
1066  * Clean all locks and share reservations on the
1067  * given vhold object that were acquired by the
1068  * given sysid
1069  */
1070 static void
1071 nlm_vhold_clean(struct nlm_vhold *nvp, int sysid)
1072 {
1073         cleanlocks(nvp->nv_vp, IGN_PID, sysid);
1074         cleanshares_by_sysid(nvp->nv_vp, sysid);
1075 }
1076 
1077 static void
1078 nlm_vhold_destroy(struct nlm_host *hostp, struct nlm_vhold *nvp)
1079 {
1080         ASSERT(MUTEX_HELD(&hostp->nh_lock));
1081 
1082         ASSERT(nvp->nv_refcnt == 0);
1083         ASSERT(TAILQ_EMPTY(&nvp->nv_slreqs));
1084 
1085         VERIFY(mod_hash_remove(hostp->nh_vholds_by_vp,
1086             (mod_hash_key_t)nvp->nv_vp,
1087             (mod_hash_val_t)&nvp) == 0);
1088 
1089         TAILQ_REMOVE(&hostp->nh_vholds_list, nvp, nv_link);
1090         VN_RELE(nvp->nv_vp);
1091         nvp->nv_vp = NULL;
1092 
1093         kmem_cache_free(nlm_vhold_cache, nvp);
1094 }
1095 
1096 /*
1097  * Return TRUE if the given vhold is busy.
1098  * Vhold object is considered to be "busy" when
1099  * all the following conditions hold:
1100  * 1) No one uses it at the moment;
1101  * 2) It hasn't any locks;
1102  * 3) It hasn't any share reservations;
1103  */
1104 static bool_t
1105 nlm_vhold_busy(struct nlm_host *hostp, struct nlm_vhold *nvp)
1106 {
1107         vnode_t *vp;
1108         int sysid;
1109 
1110         ASSERT(MUTEX_HELD(&hostp->nh_lock));
1111 
1112         if (nvp->nv_refcnt > 0)
1113                 return (TRUE);
1114 
1115         vp = nvp->nv_vp;
1116         sysid = hostp->nh_sysid;
1117         if (flk_has_remote_locks_for_sysid(vp, sysid) ||
1118             shr_has_remote_shares(vp, sysid))
1119                 return (TRUE);
1120 
1121         return (FALSE);
1122 }
1123 
1124 /* ARGSUSED */
1125 static int
1126 nlm_vhold_ctor(void *datap, void *cdrarg, int kmflags)
1127 {
1128         struct nlm_vhold *nvp = (struct nlm_vhold *)datap;
1129 
1130         bzero(nvp, sizeof (*nvp));
1131         return (0);
1132 }
1133 
1134 /* ARGSUSED */
1135 static void
1136 nlm_vhold_dtor(void *datap, void *cdrarg)
1137 {
1138         struct nlm_vhold *nvp = (struct nlm_vhold *)datap;
1139 
1140         ASSERT(nvp->nv_refcnt == 0);
1141         ASSERT(TAILQ_EMPTY(&nvp->nv_slreqs));
1142         ASSERT(nvp->nv_vp == NULL);
1143 }
1144 
1145 struct nlm_vhold *
1146 nlm_vhold_find_locked(struct nlm_host *hostp, const vnode_t *vp)
1147 {
1148         struct nlm_vhold *nvp = NULL;
1149 
1150         ASSERT(MUTEX_HELD(&hostp->nh_lock));
1151         (void) mod_hash_find(hostp->nh_vholds_by_vp,
1152             (mod_hash_key_t)vp,
1153             (mod_hash_val_t)&nvp);
1154 
1155         if (nvp != NULL)
1156                 nvp->nv_refcnt++;
1157 
1158         return (nvp);
1159 }
1160 
1161 /*
1162  * NLM host functions
1163  */
1164 static void
1165 nlm_copy_netbuf(struct netbuf *dst, struct netbuf *src)
1166 {
1167         ASSERT(src->len <= src->maxlen);
1168 
1169         dst->maxlen = src->maxlen;
1170         dst->len = src->len;
1171         dst->buf = kmem_zalloc(src->maxlen, KM_SLEEP);
1172         bcopy(src->buf, dst->buf, src->len);
1173 }
1174 
1175 /* ARGSUSED */
1176 static int
1177 nlm_host_ctor(void *datap, void *cdrarg, int kmflags)
1178 {
1179         struct nlm_host *hostp = (struct nlm_host *)datap;
1180 
1181         bzero(hostp, sizeof (*hostp));
1182         return (0);
1183 }
1184 
1185 /* ARGSUSED */
1186 static void
1187 nlm_host_dtor(void *datap, void *cdrarg)
1188 {
1189         struct nlm_host *hostp = (struct nlm_host *)datap;
1190         ASSERT(hostp->nh_refs == 0);
1191 }
1192 
1193 static void
1194 nlm_host_unregister(struct nlm_globals *g, struct nlm_host *hostp)
1195 {
1196         ASSERT(hostp->nh_refs == 0);
1197         ASSERT(hostp->nh_flags & NLM_NH_INIDLE);
1198 
1199         avl_remove(&g->nlm_hosts_tree, hostp);
1200         VERIFY(mod_hash_remove(g->nlm_hosts_hash,
1201             (mod_hash_key_t)(uintptr_t)hostp->nh_sysid,
1202             (mod_hash_val_t)&hostp) == 0);
1203         TAILQ_REMOVE(&g->nlm_idle_hosts, hostp, nh_link);
1204         hostp->nh_flags &= ~NLM_NH_INIDLE;
1205 }
1206 
1207 /*
1208  * Free resources used by a host. This is called after the reference
1209  * count has reached zero so it doesn't need to worry about locks.
1210  */
1211 static void
1212 nlm_host_destroy(struct nlm_host *hostp)
1213 {
1214         ASSERT(hostp->nh_name != NULL);
1215         ASSERT(hostp->nh_netid != NULL);
1216         ASSERT(TAILQ_EMPTY(&hostp->nh_vholds_list));
1217 
1218         strfree(hostp->nh_name);
1219         strfree(hostp->nh_netid);
1220         kmem_free(hostp->nh_addr.buf, hostp->nh_addr.maxlen);
1221 
1222         if (hostp->nh_sysid != LM_NOSYSID)
1223                 nlm_sysid_free(hostp->nh_sysid);
1224 
1225         nlm_rpc_cache_destroy(hostp);
1226 
1227         ASSERT(TAILQ_EMPTY(&hostp->nh_vholds_list));
1228         mod_hash_destroy_ptrhash(hostp->nh_vholds_by_vp);
1229 
1230         mutex_destroy(&hostp->nh_lock);
1231         cv_destroy(&hostp->nh_rpcb_cv);
1232         cv_destroy(&hostp->nh_recl_cv);
1233 
1234         kmem_cache_free(nlm_hosts_cache, hostp);
1235 }
1236 
1237 /*
1238  * Cleanup SERVER-side state after a client restarts,
1239  * or becomes unresponsive, or whatever.
1240  *
1241  * We unlock any active locks owned by the host.
1242  * When rpc.lockd is shutting down,
1243  * this function is called with newstate set to zero
1244  * which allows us to cancel any pending async locks
1245  * and clear the locking state.
1246  *
1247  * When "state" is 0, we don't update host's state,
1248  * but cleanup all remote locks on the host.
1249  * It's useful to call this function for resources
1250  * cleanup.
1251  */
1252 void
1253 nlm_host_notify_server(struct nlm_host *hostp, int32_t state)
1254 {
1255         struct nlm_vhold *nvp;
1256         struct nlm_slreq *slr;
1257         struct nlm_slreq_list slreqs2free;
1258 
1259         TAILQ_INIT(&slreqs2free);
1260         mutex_enter(&hostp->nh_lock);
1261         if (state != 0)
1262                 hostp->nh_state = state;
1263 
1264         TAILQ_FOREACH(nvp, &hostp->nh_vholds_list, nv_link) {
1265 
1266                 /* cleanup sleeping requests at first */
1267                 while ((slr = TAILQ_FIRST(&nvp->nv_slreqs)) != NULL) {
1268                         TAILQ_REMOVE(&nvp->nv_slreqs, slr, nsr_link);
1269 
1270                         /*
1271                          * Instead of freeing cancelled sleeping request
1272                          * here, we add it to the linked list created
1273                          * on the stack in order to do all frees outside
1274                          * the critical section.
1275                          */
1276                         TAILQ_INSERT_TAIL(&slreqs2free, slr, nsr_link);
1277                 }
1278 
1279                 nvp->nv_refcnt++;
1280                 mutex_exit(&hostp->nh_lock);
1281 
1282                 nlm_vhold_clean(nvp, hostp->nh_sysid);
1283 
1284                 mutex_enter(&hostp->nh_lock);
1285                 nvp->nv_refcnt--;
1286         }
1287 
1288         mutex_exit(&hostp->nh_lock);
1289         while ((slr = TAILQ_FIRST(&slreqs2free)) != NULL) {
1290                 TAILQ_REMOVE(&slreqs2free, slr, nsr_link);
1291                 kmem_free(slr, sizeof (*slr));
1292         }
1293 }
1294 
1295 /*
1296  * Cleanup CLIENT-side state after a server restarts,
1297  * or becomes unresponsive, or whatever.
1298  *
1299  * This is called by the local NFS statd when we receive a
1300  * host state change notification.  (also nlm_svc_stopping)
1301  *
1302  * Deal with a server restart.  If we are stopping the
1303  * NLM service, we'll have newstate == 0, and will just
1304  * cancel all our client-side lock requests.  Otherwise,
1305  * start the "recovery" process to reclaim any locks
1306  * we hold on this server.
1307  */
1308 void
1309 nlm_host_notify_client(struct nlm_host *hostp, int32_t state)
1310 {
1311         mutex_enter(&hostp->nh_lock);
1312         hostp->nh_state = state;
1313         if (hostp->nh_flags & NLM_NH_RECLAIM) {
1314                 /*
1315                  * Either host's state is up to date or
1316                  * host is already in recovery.
1317                  */
1318                 mutex_exit(&hostp->nh_lock);
1319                 return;
1320         }
1321 
1322         hostp->nh_flags |= NLM_NH_RECLAIM;
1323 
1324         /*
1325          * Host will be released by the recovery thread,
1326          * thus we need to increment refcount.
1327          */
1328         hostp->nh_refs++;
1329         mutex_exit(&hostp->nh_lock);
1330 
1331         (void) zthread_create(NULL, 0, nlm_reclaimer,
1332             hostp, 0, minclsyspri);
1333 }
1334 
1335 /*
1336  * The function is called when NLM client detects that
1337  * server has entered in grace period and client needs
1338  * to wait until reclamation process (if any) does
1339  * its job.
1340  */
1341 int
1342 nlm_host_wait_grace(struct nlm_host *hostp)
1343 {
1344         struct nlm_globals *g;
1345         int error = 0;
1346 
1347         g = zone_getspecific(nlm_zone_key, curzone);
1348         mutex_enter(&hostp->nh_lock);
1349 
1350         do {
1351                 int rc;
1352 
1353                 rc = cv_timedwait_sig(&hostp->nh_recl_cv,
1354                     &hostp->nh_lock, ddi_get_lbolt() +
1355                     SEC_TO_TICK(g->retrans_tmo));
1356 
1357                 if (rc == 0) {
1358                         error = EINTR;
1359                         break;
1360                 }
1361         } while (hostp->nh_flags & NLM_NH_RECLAIM);
1362 
1363         mutex_exit(&hostp->nh_lock);
1364         return (error);
1365 }
1366 
1367 /*
1368  * Create a new NLM host.
1369  *
1370  * NOTE: The in-kernel RPC (kRPC) subsystem uses TLI/XTI,
1371  * which needs both a knetconfig and an address when creating
1372  * endpoints. Thus host object stores both knetconfig and
1373  * netid.
1374  */
1375 static struct nlm_host *
1376 nlm_host_create(char *name, const char *netid,
1377     struct knetconfig *knc, struct netbuf *naddr)
1378 {
1379         struct nlm_host *host;
1380 
1381         host = kmem_cache_alloc(nlm_hosts_cache, KM_SLEEP);
1382 
1383         mutex_init(&host->nh_lock, NULL, MUTEX_DEFAULT, NULL);
1384         cv_init(&host->nh_rpcb_cv, NULL, CV_DEFAULT, NULL);
1385         cv_init(&host->nh_recl_cv, NULL, CV_DEFAULT, NULL);
1386 
1387         host->nh_sysid = LM_NOSYSID;
1388         host->nh_refs = 1;
1389         host->nh_name = strdup(name);
1390         host->nh_netid = strdup(netid);
1391         host->nh_knc = *knc;
1392         nlm_copy_netbuf(&host->nh_addr, naddr);
1393 
1394         host->nh_state = 0;
1395         host->nh_rpcb_state = NRPCB_NEED_UPDATE;
1396         host->nh_flags = 0;
1397 
1398         host->nh_vholds_by_vp = mod_hash_create_ptrhash("nlm vholds hash",
1399             32, mod_hash_null_valdtor, sizeof (vnode_t));
1400 
1401         TAILQ_INIT(&host->nh_vholds_list);
1402         TAILQ_INIT(&host->nh_rpchc);
1403 
1404         return (host);
1405 }
1406 
1407 /*
1408  * Cancel all client side sleeping locks owned by given host.
1409  */
1410 void
1411 nlm_host_cancel_slocks(struct nlm_globals *g, struct nlm_host *hostp)
1412 {
1413         struct nlm_slock *nslp;
1414 
1415         mutex_enter(&g->lock);
1416         TAILQ_FOREACH(nslp, &g->nlm_slocks, nsl_link) {
1417                 if (nslp->nsl_host == hostp) {
1418                         nslp->nsl_state = NLM_SL_CANCELLED;
1419                         cv_broadcast(&nslp->nsl_cond);
1420                 }
1421         }
1422 
1423         mutex_exit(&g->lock);
1424 }
1425 
1426 /*
1427  * Garbage collect stale vhold objects.
1428  *
1429  * In other words check whether vnodes that are
1430  * held by vhold objects still have any locks
1431  * or shares or still in use. If they aren't,
1432  * just destroy them.
1433  */
1434 static void
1435 nlm_host_gc_vholds(struct nlm_host *hostp)
1436 {
1437         struct nlm_vhold *nvp;
1438 
1439         ASSERT(MUTEX_HELD(&hostp->nh_lock));
1440 
1441         nvp = TAILQ_FIRST(&hostp->nh_vholds_list);
1442         while (nvp != NULL) {
1443                 struct nlm_vhold *nvp_tmp;
1444 
1445                 if (nlm_vhold_busy(hostp, nvp)) {
1446                         nvp = TAILQ_NEXT(nvp, nv_link);
1447                         continue;
1448                 }
1449 
1450                 nvp_tmp = TAILQ_NEXT(nvp, nv_link);
1451                 nlm_vhold_destroy(hostp, nvp);
1452                 nvp = nvp_tmp;
1453         }
1454 }
1455 
1456 /*
1457  * Check whether the given host has any
1458  * server side locks or share reservations.
1459  */
1460 static bool_t
1461 nlm_host_has_srv_locks(struct nlm_host *hostp)
1462 {
1463         /*
1464          * It's cheap and simple: if server has
1465          * any locks/shares there must be vhold
1466          * object storing the affected vnode.
1467          *
1468          * NOTE: We don't need to check sleeping
1469          * locks on the server side, because if
1470          * server side sleeping lock is alive,
1471          * there must be a vhold object corresponding
1472          * to target vnode.
1473          */
1474         ASSERT(MUTEX_HELD(&hostp->nh_lock));
1475         if (!TAILQ_EMPTY(&hostp->nh_vholds_list))
1476                 return (TRUE);
1477 
1478         return (FALSE);
1479 }
1480 
1481 /*
1482  * Check whether the given host has any client side
1483  * locks or share reservations.
1484  */
1485 static bool_t
1486 nlm_host_has_cli_locks(struct nlm_host *hostp)
1487 {
1488         ASSERT(MUTEX_HELD(&hostp->nh_lock));
1489 
1490         /*
1491          * XXX: It's not the way I'd like to do the check,
1492          * because flk_sysid_has_locks() can be very
1493          * expensive by design. Unfortunatelly it iterates
1494          * through all locks on the system, doesn't matter
1495          * were they made on remote system via NLM or
1496          * on local system via reclock. To understand the
1497          * problem, consider that there're dozens of thousands
1498          * of locks that are made on some ZFS dataset. And there's
1499          * another dataset shared by NFS where NLM client had locks
1500          * some time ago, but doesn't have them now.
1501          * In this case flk_sysid_has_locks() will iterate
1502          * thrught dozens of thousands locks until it returns us
1503          * FALSE.
1504          * Oh, I hope that in shiny future somebody will make
1505          * local lock manager (os/flock.c) better, so that
1506          * it'd be more friedly to remote locks and
1507          * flk_sysid_has_locks() wouldn't be so expensive.
1508          */
1509         if (flk_sysid_has_locks(hostp->nh_sysid |
1510             LM_SYSID_CLIENT, FLK_QUERY_ACTIVE))
1511                 return (TRUE);
1512 
1513         /*
1514          * Check whether host has any share reservations
1515          * registered on the client side.
1516          */
1517         if (hostp->nh_shrlist != NULL)
1518                 return (TRUE);
1519 
1520         return (FALSE);
1521 }
1522 
1523 /*
1524  * Determine whether the given host owns any
1525  * locks or share reservations.
1526  */
1527 static bool_t
1528 nlm_host_has_locks(struct nlm_host *hostp)
1529 {
1530         if (nlm_host_has_srv_locks(hostp))
1531                 return (TRUE);
1532 
1533         return (nlm_host_has_cli_locks(hostp));
1534 }
1535 
1536 /*
1537  * This function compares only addresses of two netbufs
1538  * that belong to NC_TCP[6] or NC_UDP[6] protofamily.
1539  * Port part of netbuf is ignored.
1540  *
1541  * Return values:
1542  *  -1: nb1's address is "smaller" than nb2's
1543  *   0: addresses are equal
1544  *   1: nb1's address is "greater" than nb2's
1545  */
1546 static int
1547 nlm_netbuf_addrs_cmp(struct netbuf *nb1, struct netbuf *nb2)
1548 {
1549         union nlm_addr {
1550                 struct sockaddr sa;
1551                 struct sockaddr_in sin;
1552                 struct sockaddr_in6 sin6;
1553         } *na1, *na2;
1554         int res;
1555 
1556         /* LINTED E_BAD_PTR_CAST_ALIGN */
1557         na1 = (union nlm_addr *)nb1->buf;
1558         /* LINTED E_BAD_PTR_CAST_ALIGN */
1559         na2 = (union nlm_addr *)nb2->buf;
1560 
1561         if (na1->sa.sa_family < na2->sa.sa_family)
1562                 return (-1);
1563         if (na1->sa.sa_family > na2->sa.sa_family)
1564                 return (1);
1565 
1566         switch (na1->sa.sa_family) {
1567         case AF_INET:
1568                 res = memcmp(&na1->sin.sin_addr, &na2->sin.sin_addr,
1569                     sizeof (na1->sin.sin_addr));
1570                 break;
1571         case AF_INET6:
1572                 res = memcmp(&na1->sin6.sin6_addr, &na2->sin6.sin6_addr,
1573                     sizeof (na1->sin6.sin6_addr));
1574                 break;
1575         default:
1576                 VERIFY(0);
1577                 return (0);
1578         }
1579 
1580         return (SIGN(res));
1581 }
1582 
1583 /*
1584  * Compare two nlm hosts.
1585  * Return values:
1586  * -1: host1 is "smaller" than host2
1587  *  0: host1 is equal to host2
1588  *  1: host1 is "greater" than host2
1589  */
1590 int
1591 nlm_host_cmp(const void *p1, const void *p2)
1592 {
1593         struct nlm_host *h1 = (struct nlm_host *)p1;
1594         struct nlm_host *h2 = (struct nlm_host *)p2;
1595         int res;
1596 
1597         res = strcmp(h1->nh_netid, h2->nh_netid);
1598         if (res != 0)
1599                 return (SIGN(res));
1600 
1601         res = nlm_netbuf_addrs_cmp(&h1->nh_addr, &h2->nh_addr);
1602         return (res);
1603 }
1604 
1605 /*
1606  * Find the host specified by...  (see below)
1607  * If found, increment the ref count.
1608  */
1609 static struct nlm_host *
1610 nlm_host_find_locked(struct nlm_globals *g, const char *netid,
1611     struct netbuf *naddr, avl_index_t *wherep)
1612 {
1613         struct nlm_host *hostp, key;
1614         avl_index_t pos;
1615 
1616         ASSERT(MUTEX_HELD(&g->lock));
1617 
1618         key.nh_netid = (char *)netid;
1619         key.nh_addr.buf = naddr->buf;
1620         key.nh_addr.len = naddr->len;
1621         key.nh_addr.maxlen = naddr->maxlen;
1622 
1623         hostp = avl_find(&g->nlm_hosts_tree, &key, &pos);
1624 
1625         if (hostp != NULL) {
1626                 /*
1627                  * Host is inuse now. Remove it from idle
1628                  * hosts list if needed.
1629                  */
1630                 if (hostp->nh_flags & NLM_NH_INIDLE) {
1631                         TAILQ_REMOVE(&g->nlm_idle_hosts, hostp, nh_link);
1632                         hostp->nh_flags &= ~NLM_NH_INIDLE;
1633                 }
1634 
1635                 hostp->nh_refs++;
1636         }
1637         if (wherep != NULL)
1638                 *wherep = pos;
1639 
1640         return (hostp);
1641 }
1642 
1643 /*
1644  * Find NLM host for the given name and address.
1645  */
1646 struct nlm_host *
1647 nlm_host_find(struct nlm_globals *g, const char *netid,
1648     struct netbuf *addr)
1649 {
1650         struct nlm_host *hostp = NULL;
1651 
1652         mutex_enter(&g->lock);
1653         if (g->run_status != NLM_ST_UP)
1654                 goto out;
1655 
1656         hostp = nlm_host_find_locked(g, netid, addr, NULL);
1657 
1658 out:
1659         mutex_exit(&g->lock);
1660         return (hostp);
1661 }
1662 
1663 
1664 /*
1665  * Find or create an NLM host for the given name and address.
1666  *
1667  * The remote host is determined by all of: name, netid, address.
1668  * Note that the netid is whatever nlm_svc_add_ep() gave to
1669  * svc_tli_kcreate() for the service binding.  If any of these
1670  * are different, allocate a new host (new sysid).
1671  */
1672 struct nlm_host *
1673 nlm_host_findcreate(struct nlm_globals *g, char *name,
1674     const char *netid, struct netbuf *addr)
1675 {
1676         int err;
1677         struct nlm_host *host, *newhost = NULL;
1678         struct knetconfig knc;
1679         avl_index_t where;
1680 
1681         mutex_enter(&g->lock);
1682         if (g->run_status != NLM_ST_UP) {
1683                 mutex_exit(&g->lock);
1684                 return (NULL);
1685         }
1686 
1687         host = nlm_host_find_locked(g, netid, addr, NULL);
1688         mutex_exit(&g->lock);
1689         if (host != NULL)
1690                 return (host);
1691 
1692         err = nlm_knc_from_netid(netid, &knc);
1693         if (err != 0)
1694                 return (NULL);
1695         /*
1696          * Do allocations (etc.) outside of mutex,
1697          * and then check again before inserting.
1698          */
1699         newhost = nlm_host_create(name, netid, &knc, addr);
1700         newhost->nh_sysid = nlm_sysid_alloc();
1701         if (newhost->nh_sysid == LM_NOSYSID)
1702                 goto out;
1703 
1704         mutex_enter(&g->lock);
1705         host = nlm_host_find_locked(g, netid, addr, &where);
1706         if (host == NULL) {
1707                 host = newhost;
1708                 newhost = NULL;
1709 
1710                 /*
1711                  * Insert host to the hosts AVL tree that is
1712                  * used to lookup by <netid, address> pair.
1713                  */
1714                 avl_insert(&g->nlm_hosts_tree, host, where);
1715 
1716                 /*
1717                  * Insert host to the hosts hash table that is
1718                  * used to lookup host by sysid.
1719                  */
1720                 VERIFY(mod_hash_insert(g->nlm_hosts_hash,
1721                     (mod_hash_key_t)(uintptr_t)host->nh_sysid,
1722                     (mod_hash_val_t)host) == 0);
1723         }
1724 
1725         mutex_exit(&g->lock);
1726 
1727 out:
1728         if (newhost != NULL) {
1729                 /*
1730                  * We do not need the preallocated nlm_host
1731                  * so decrement the reference counter
1732                  * and destroy it.
1733                  */
1734                 newhost->nh_refs--;
1735                 nlm_host_destroy(newhost);
1736         }
1737 
1738         return (host);
1739 }
1740 
1741 /*
1742  * Find the NLM host that matches the value of 'sysid'.
1743  * If found, return it with a new ref,
1744  * else return NULL.
1745  */
1746 struct nlm_host *
1747 nlm_host_find_by_sysid(struct nlm_globals *g, sysid_t sysid)
1748 {
1749         struct nlm_host *hostp = NULL;
1750 
1751         mutex_enter(&g->lock);
1752         if (g->run_status != NLM_ST_UP)
1753                 goto out;
1754 
1755         (void) mod_hash_find(g->nlm_hosts_hash,
1756             (mod_hash_key_t)(uintptr_t)sysid,
1757             (mod_hash_val_t)&hostp);
1758 
1759         if (hostp == NULL)
1760                 goto out;
1761 
1762         /*
1763          * Host is inuse now. Remove it
1764          * from idle hosts list if needed.
1765          */
1766         if (hostp->nh_flags & NLM_NH_INIDLE) {
1767                 TAILQ_REMOVE(&g->nlm_idle_hosts, hostp, nh_link);
1768                 hostp->nh_flags &= ~NLM_NH_INIDLE;
1769         }
1770 
1771         hostp->nh_refs++;
1772 
1773 out:
1774         mutex_exit(&g->lock);
1775         return (hostp);
1776 }
1777 
1778 /*
1779  * Release the given host.
1780  * I.e. drop a reference that was taken earlier by one of
1781  * the following functions: nlm_host_findcreate(), nlm_host_find(),
1782  * nlm_host_find_by_sysid().
1783  *
1784  * When the very last reference is dropped, host is moved to
1785  * so-called "idle state". All hosts that are in idle state
1786  * have an idle timeout. If timeout is expired, GC thread
1787  * checks whether hosts have any locks and if they heven't
1788  * any, it removes them.
1789  * NOTE: only unused hosts can be in idle state.
1790  */
1791 static void
1792 nlm_host_release_locked(struct nlm_globals *g, struct nlm_host *hostp)
1793 {
1794         if (hostp == NULL)
1795                 return;
1796 
1797         ASSERT(MUTEX_HELD(&g->lock));
1798         ASSERT(hostp->nh_refs > 0);
1799 
1800         hostp->nh_refs--;
1801         if (hostp->nh_refs != 0)
1802                 return;
1803 
1804         /*
1805          * The very last reference to the host was dropped,
1806          * thus host is unused now. Set its idle timeout
1807          * and move it to the idle hosts LRU list.
1808          */
1809         hostp->nh_idle_timeout = ddi_get_lbolt() +
1810             SEC_TO_TICK(g->cn_idle_tmo);
1811 
1812         ASSERT((hostp->nh_flags & NLM_NH_INIDLE) == 0);
1813         TAILQ_INSERT_TAIL(&g->nlm_idle_hosts, hostp, nh_link);
1814         hostp->nh_flags |= NLM_NH_INIDLE;
1815 }
1816 
1817 void
1818 nlm_host_release(struct nlm_globals *g, struct nlm_host *hostp)
1819 {
1820         if (hostp == NULL)
1821                 return;
1822 
1823         mutex_enter(&g->lock);
1824         nlm_host_release_locked(g, hostp);
1825         mutex_exit(&g->lock);
1826 }
1827 
1828 /*
1829  * Unregister this NLM host (NFS client) with the local statd
1830  * due to idleness (no locks held for a while).
1831  */
1832 void
1833 nlm_host_unmonitor(struct nlm_globals *g, struct nlm_host *host)
1834 {
1835         enum clnt_stat stat;
1836 
1837         VERIFY(host->nh_refs == 0);
1838         if (!(host->nh_flags & NLM_NH_MONITORED))
1839                 return;
1840 
1841         host->nh_flags &= ~NLM_NH_MONITORED;
1842         stat = nlm_nsm_unmon(&g->nlm_nsm, host->nh_name);
1843         if (stat != RPC_SUCCESS) {
1844                 NLM_WARN("NLM: Failed to contact statd, stat=%d\n", stat);
1845                 return;
1846         }
1847 }
1848 
1849 /*
1850  * Ask the local NFS statd to begin monitoring this host.
1851  * It will call us back when that host restarts, using the
1852  * prog,vers,proc specified below, i.e. NLM_SM_NOTIFY1,
1853  * which is handled in nlm_do_notify1().
1854  */
1855 void
1856 nlm_host_monitor(struct nlm_globals *g, struct nlm_host *host, int state)
1857 {
1858         int family;
1859         netobj obj;
1860         enum clnt_stat stat;
1861 
1862         if (state != 0 && host->nh_state == 0) {
1863                 /*
1864                  * This is the first time we have seen an NSM state
1865                  * Value for this host. We record it here to help
1866                  * detect host reboots.
1867                  */
1868                 host->nh_state = state;
1869         }
1870 
1871         mutex_enter(&host->nh_lock);
1872         if (host->nh_flags & NLM_NH_MONITORED) {
1873                 mutex_exit(&host->nh_lock);
1874                 return;
1875         }
1876 
1877         host->nh_flags |= NLM_NH_MONITORED;
1878         mutex_exit(&host->nh_lock);
1879 
1880         /*
1881          * Before we begin monitoring the host register the network address
1882          * associated with this hostname.
1883          */
1884         nlm_netbuf_to_netobj(&host->nh_addr, &family, &obj);
1885         stat = nlm_nsmaddr_reg(&g->nlm_nsm, host->nh_name, family, &obj);
1886         if (stat != RPC_SUCCESS) {
1887                 NLM_WARN("Failed to register address, stat=%d\n", stat);
1888                 mutex_enter(&g->lock);
1889                 host->nh_flags &= ~NLM_NH_MONITORED;
1890                 mutex_exit(&g->lock);
1891 
1892                 return;
1893         }
1894 
1895         /*
1896          * Tell statd how to call us with status updates for
1897          * this host. Updates arrive via nlm_do_notify1().
1898          *
1899          * We put our assigned system ID value in the priv field to
1900          * make it simpler to find the host if we are notified of a
1901          * host restart.
1902          */
1903         stat = nlm_nsm_mon(&g->nlm_nsm, host->nh_name, host->nh_sysid);
1904         if (stat != RPC_SUCCESS) {
1905                 NLM_WARN("Failed to contact local NSM, stat=%d\n", stat);
1906                 mutex_enter(&g->lock);
1907                 host->nh_flags &= ~NLM_NH_MONITORED;
1908                 mutex_exit(&g->lock);
1909 
1910                 return;
1911         }
1912 }
1913 
1914 int
1915 nlm_host_get_state(struct nlm_host *hostp)
1916 {
1917 
1918         return (hostp->nh_state);
1919 }
1920 
1921 /*
1922  * NLM client/server sleeping locks
1923  */
1924 
1925 /*
1926  * Register client side sleeping lock.
1927  *
1928  * Our client code calls this to keep information
1929  * about sleeping lock somewhere. When it receives
1930  * grant callback from server or when it just
1931  * needs to remove all sleeping locks from vnode,
1932  * it uses this information for remove/apply lock
1933  * properly.
1934  */
1935 struct nlm_slock *
1936 nlm_slock_register(
1937         struct nlm_globals *g,
1938         struct nlm_host *host,
1939         struct nlm4_lock *lock,
1940         struct vnode *vp)
1941 {
1942         struct nlm_slock *nslp;
1943 
1944         nslp = kmem_zalloc(sizeof (*nslp), KM_SLEEP);
1945         cv_init(&nslp->nsl_cond, NULL, CV_DEFAULT, NULL);
1946         nslp->nsl_lock = *lock;
1947         nlm_copy_netobj(&nslp->nsl_fh, &nslp->nsl_lock.fh);
1948         nslp->nsl_state = NLM_SL_BLOCKED;
1949         nslp->nsl_host = host;
1950         nslp->nsl_vp = vp;
1951 
1952         mutex_enter(&g->lock);
1953         TAILQ_INSERT_TAIL(&g->nlm_slocks, nslp, nsl_link);
1954         mutex_exit(&g->lock);
1955 
1956         return (nslp);
1957 }
1958 
1959 /*
1960  * Remove this lock from the wait list and destroy it.
1961  */
1962 void
1963 nlm_slock_unregister(struct nlm_globals *g, struct nlm_slock *nslp)
1964 {
1965         mutex_enter(&g->lock);
1966         TAILQ_REMOVE(&g->nlm_slocks, nslp, nsl_link);
1967         mutex_exit(&g->lock);
1968 
1969         kmem_free(nslp->nsl_fh.n_bytes, nslp->nsl_fh.n_len);
1970         cv_destroy(&nslp->nsl_cond);
1971         kmem_free(nslp, sizeof (*nslp));
1972 }
1973 
1974 /*
1975  * Wait for a granted callback or cancellation event
1976  * for a sleeping lock.
1977  *
1978  * If a signal interrupted the wait or if the lock
1979  * was cancelled, return EINTR - the caller must arrange to send
1980  * a cancellation to the server.
1981  *
1982  * If timeout occurred, return ETIMEDOUT - the caller must
1983  * resend the lock request to the server.
1984  *
1985  * On success return 0.
1986  */
1987 int
1988 nlm_slock_wait(struct nlm_globals *g,
1989     struct nlm_slock *nslp, uint_t timeo_secs)
1990 {
1991         clock_t timeo_ticks;
1992         int cv_res, error;
1993 
1994         /*
1995          * If the granted message arrived before we got here,
1996          * nslp->nsl_state will be NLM_SL_GRANTED - in that case don't sleep.
1997          */
1998         cv_res = 1;
1999         timeo_ticks = ddi_get_lbolt() + SEC_TO_TICK(timeo_secs);
2000 
2001         mutex_enter(&g->lock);
2002         while (nslp->nsl_state == NLM_SL_BLOCKED && cv_res > 0) {
2003                 cv_res = cv_timedwait_sig(&nslp->nsl_cond,
2004                     &g->lock, timeo_ticks);
2005         }
2006 
2007         /*
2008          * No matter why we wake up, if the lock was
2009          * cancelled, let the function caller to know
2010          * about it by returning EINTR.
2011          */
2012         if (nslp->nsl_state == NLM_SL_CANCELLED) {
2013                 error = EINTR;
2014                 goto out;
2015         }
2016 
2017         if (cv_res <= 0) {
2018                 /* We were woken up either by timeout or by interrupt */
2019                 error = (cv_res < 0) ? ETIMEDOUT : EINTR;
2020 
2021                 /*
2022                  * The granted message may arrive after the
2023                  * interrupt/timeout but before we manage to lock the
2024                  * mutex. Detect this by examining nslp.
2025                  */
2026                 if (nslp->nsl_state == NLM_SL_GRANTED)
2027                         error = 0;
2028         } else { /* Awaken via cv_signal()/cv_broadcast() or didn't block */
2029                 error = 0;
2030                 VERIFY(nslp->nsl_state == NLM_SL_GRANTED);
2031         }
2032 
2033 out:
2034         mutex_exit(&g->lock);
2035         return (error);
2036 }
2037 
2038 /*
2039  * Mark client side sleeping lock as granted
2040  * and wake up a process blocked on the lock.
2041  * Called from server side NLM_GRANT handler.
2042  *
2043  * If sleeping lock is found return 0, otherwise
2044  * return ENOENT.
2045  */
2046 int
2047 nlm_slock_grant(struct nlm_globals *g,
2048     struct nlm_host *hostp, struct nlm4_lock *alock)
2049 {
2050         struct nlm_slock *nslp;
2051         int error = ENOENT;
2052 
2053         mutex_enter(&g->lock);
2054         TAILQ_FOREACH(nslp, &g->nlm_slocks, nsl_link) {
2055                 if ((nslp->nsl_state != NLM_SL_BLOCKED) ||
2056                     (nslp->nsl_host != hostp))
2057                         continue;
2058 
2059                 if (alock->svid              == nslp->nsl_lock.svid &&
2060                     alock->l_offset  == nslp->nsl_lock.l_offset &&
2061                     alock->l_len     == nslp->nsl_lock.l_len &&
2062                     alock->fh.n_len  == nslp->nsl_lock.fh.n_len &&
2063                     bcmp(alock->fh.n_bytes, nslp->nsl_lock.fh.n_bytes,
2064                     nslp->nsl_lock.fh.n_len) == 0) {
2065                         nslp->nsl_state = NLM_SL_GRANTED;
2066                         cv_broadcast(&nslp->nsl_cond);
2067                         error = 0;
2068                         break;
2069                 }
2070         }
2071 
2072         mutex_exit(&g->lock);
2073         return (error);
2074 }
2075 
2076 /*
2077  * Register sleeping lock request corresponding to
2078  * flp on the given vhold object.
2079  * On success function returns 0, otherwise (if
2080  * lock request with the same flp is already
2081  * registered) function returns EEXIST.
2082  */
2083 int
2084 nlm_slreq_register(struct nlm_host *hostp, struct nlm_vhold *nvp,
2085     struct flock64 *flp)
2086 {
2087         struct nlm_slreq *slr, *new_slr = NULL;
2088         int ret = EEXIST;
2089 
2090         mutex_enter(&hostp->nh_lock);
2091         slr = nlm_slreq_find_locked(hostp, nvp, flp);
2092         if (slr != NULL)
2093                 goto out;
2094 
2095         mutex_exit(&hostp->nh_lock);
2096         new_slr = kmem_zalloc(sizeof (*slr), KM_SLEEP);
2097         bcopy(flp, &new_slr->nsr_fl, sizeof (*flp));
2098 
2099         mutex_enter(&hostp->nh_lock);
2100         slr = nlm_slreq_find_locked(hostp, nvp, flp);
2101         if (slr == NULL) {
2102                 slr = new_slr;
2103                 new_slr = NULL;
2104                 ret = 0;
2105 
2106                 TAILQ_INSERT_TAIL(&nvp->nv_slreqs, slr, nsr_link);
2107         }
2108 
2109 out:
2110         mutex_exit(&hostp->nh_lock);
2111         if (new_slr != NULL)
2112                 kmem_free(new_slr, sizeof (*new_slr));
2113 
2114         return (ret);
2115 }
2116 
2117 /*
2118  * Unregister sleeping lock request corresponding
2119  * to flp from the given vhold object.
2120  * On success function returns 0, otherwise (if
2121  * lock request corresponding to flp isn't found
2122  * on the given vhold) function returns ENOENT.
2123  */
2124 int
2125 nlm_slreq_unregister(struct nlm_host *hostp, struct nlm_vhold *nvp,
2126     struct flock64 *flp)
2127 {
2128         struct nlm_slreq *slr;
2129 
2130         mutex_enter(&hostp->nh_lock);
2131         slr = nlm_slreq_find_locked(hostp, nvp, flp);
2132         if (slr == NULL) {
2133                 mutex_exit(&hostp->nh_lock);
2134                 return (ENOENT);
2135         }
2136 
2137         TAILQ_REMOVE(&nvp->nv_slreqs, slr, nsr_link);
2138         mutex_exit(&hostp->nh_lock);
2139 
2140         kmem_free(slr, sizeof (*slr));
2141         return (0);
2142 }
2143 
2144 /*
2145  * Find sleeping lock request on the given vhold object by flp.
2146  */
2147 struct nlm_slreq *
2148 nlm_slreq_find_locked(struct nlm_host *hostp, struct nlm_vhold *nvp,
2149     struct flock64 *flp)
2150 {
2151         struct nlm_slreq *slr = NULL;
2152 
2153         ASSERT(MUTEX_HELD(&hostp->nh_lock));
2154         TAILQ_FOREACH(slr, &nvp->nv_slreqs, nsr_link) {
2155                 if (slr->nsr_fl.l_start              == flp->l_start      &&
2156                     slr->nsr_fl.l_len                == flp->l_len        &&
2157                     slr->nsr_fl.l_pid                == flp->l_pid        &&
2158                     slr->nsr_fl.l_type               == flp->l_type)
2159                         break;
2160         }
2161 
2162         return (slr);
2163 }
2164 
2165 /*
2166  * NLM tracks active share reservations made on the client side.
2167  * It needs to have a track of share reservations for two purposes
2168  * 1) to determine if nlm_host is busy (if it has active locks and/or
2169  *    share reservations, it is)
2170  * 2) to recover active share reservations when NLM server reports
2171  *    that it has rebooted.
2172  *
2173  * Unfortunately Illumos local share reservations manager (see os/share.c)
2174  * doesn't have an ability to lookup all reservations on the system
2175  * by sysid (like local lock manager) or get all reservations by sysid.
2176  * It tracks reservations per vnode and is able to get/looup them
2177  * on particular vnode. It's not what NLM needs. Thus it has that ugly
2178  * share reservations tracking scheme.
2179  */
2180 
2181 void
2182 nlm_shres_track(struct nlm_host *hostp, vnode_t *vp, struct shrlock *shrp)
2183 {
2184         struct nlm_shres *nsp, *nsp_new;
2185 
2186         /*
2187          * NFS code must fill the s_owner, so that
2188          * s_own_len is never 0.
2189          */
2190         ASSERT(shrp->s_own_len > 0);
2191         nsp_new = nlm_shres_create_item(shrp, vp);
2192 
2193         mutex_enter(&hostp->nh_lock);
2194         for (nsp = hostp->nh_shrlist; nsp != NULL; nsp = nsp->ns_next)
2195                 if (nsp->ns_vp == vp && nlm_shres_equal(shrp, nsp->ns_shr))
2196                         break;
2197 
2198         if (nsp != NULL) {
2199                 /*
2200                  * Found a duplicate. Do nothing.
2201                  */
2202 
2203                 goto out;
2204         }
2205 
2206         nsp = nsp_new;
2207         nsp_new = NULL;
2208         nsp->ns_next = hostp->nh_shrlist;
2209         hostp->nh_shrlist = nsp;
2210 
2211 out:
2212         mutex_exit(&hostp->nh_lock);
2213         if (nsp_new != NULL)
2214                 nlm_shres_destroy_item(nsp_new);
2215 }
2216 
2217 void
2218 nlm_shres_untrack(struct nlm_host *hostp, vnode_t *vp, struct shrlock *shrp)
2219 {
2220         struct nlm_shres *nsp, *nsp_prev = NULL;
2221 
2222         mutex_enter(&hostp->nh_lock);
2223         nsp = hostp->nh_shrlist;
2224         while (nsp != NULL) {
2225                 if (nsp->ns_vp == vp && nlm_shres_equal(shrp, nsp->ns_shr)) {
2226                         struct nlm_shres *nsp_del;
2227 
2228                         nsp_del = nsp;
2229                         nsp = nsp->ns_next;
2230                         if (nsp_prev != NULL)
2231                                 nsp_prev->ns_next = nsp;
2232                         else
2233                                 hostp->nh_shrlist = nsp;
2234 
2235                         nlm_shres_destroy_item(nsp_del);
2236                         continue;
2237                 }
2238 
2239                 nsp_prev = nsp;
2240                 nsp = nsp->ns_next;
2241         }
2242 
2243         mutex_exit(&hostp->nh_lock);
2244 }
2245 
2246 /*
2247  * Get a _copy_ of the list of all active share reservations
2248  * made by the given host.
2249  * NOTE: the list function returns _must_ be released using
2250  *       nlm_free_shrlist().
2251  */
2252 struct nlm_shres *
2253 nlm_get_active_shres(struct nlm_host *hostp)
2254 {
2255         struct nlm_shres *nsp, *nslist = NULL;
2256 
2257         mutex_enter(&hostp->nh_lock);
2258         for (nsp = hostp->nh_shrlist; nsp != NULL; nsp = nsp->ns_next) {
2259                 struct nlm_shres *nsp_new;
2260 
2261                 nsp_new = nlm_shres_create_item(nsp->ns_shr, nsp->ns_vp);
2262                 nsp_new->ns_next = nslist;
2263                 nslist = nsp_new;
2264         }
2265 
2266         mutex_exit(&hostp->nh_lock);
2267         return (nslist);
2268 }
2269 
2270 /*
2271  * Free memory allocated for the active share reservations
2272  * list created by nlm_get_active_shres() function.
2273  */
2274 void
2275 nlm_free_shrlist(struct nlm_shres *nslist)
2276 {
2277         struct nlm_shres *nsp;
2278 
2279         while (nslist != NULL) {
2280                 nsp =  nslist;
2281                 nslist = nslist->ns_next;
2282 
2283                 nlm_shres_destroy_item(nsp);
2284         }
2285 }
2286 
2287 static bool_t
2288 nlm_shres_equal(struct shrlock *shrp1, struct shrlock *shrp2)
2289 {
2290         if (shrp1->s_sysid   == shrp2->s_sysid    &&
2291             shrp1->s_pid     == shrp2->s_pid              &&
2292             shrp1->s_own_len == shrp2->s_own_len  &&
2293             bcmp(shrp1->s_owner, shrp2->s_owner,
2294             shrp1->s_own_len) == 0)
2295                 return (TRUE);
2296 
2297         return (FALSE);
2298 }
2299 
2300 static struct nlm_shres *
2301 nlm_shres_create_item(struct shrlock *shrp, vnode_t *vp)
2302 {
2303         struct nlm_shres *nsp;
2304 
2305         nsp = kmem_alloc(sizeof (*nsp), KM_SLEEP);
2306         nsp->ns_shr = kmem_alloc(sizeof (*shrp), KM_SLEEP);
2307         bcopy(shrp, nsp->ns_shr, sizeof (*shrp));
2308         nsp->ns_shr->s_owner = kmem_alloc(shrp->s_own_len, KM_SLEEP);
2309         bcopy(shrp->s_owner, nsp->ns_shr->s_owner, shrp->s_own_len);
2310         nsp->ns_vp = vp;
2311 
2312         return (nsp);
2313 }
2314 
2315 static void
2316 nlm_shres_destroy_item(struct nlm_shres *nsp)
2317 {
2318         kmem_free(nsp->ns_shr->s_owner,
2319             nsp->ns_shr->s_own_len);
2320         kmem_free(nsp->ns_shr, sizeof (struct shrlock));
2321         kmem_free(nsp, sizeof (*nsp));
2322 }
2323 
2324 /*
2325  * Called by klmmod.c when lockd adds a network endpoint
2326  * on which we should begin RPC services.
2327  */
2328 int
2329 nlm_svc_add_ep(struct file *fp, const char *netid, struct knetconfig *knc)
2330 {
2331         SVCMASTERXPRT *xprt = NULL;
2332         int error;
2333 
2334         error = svc_tli_kcreate(fp, 0, (char *)netid, NULL, &xprt,
2335             &nlm_sct, NULL, NLM_SVCPOOL_ID, FALSE);
2336         if (error != 0)
2337                 return (error);
2338 
2339         (void) nlm_knc_to_netid(knc);
2340         return (0);
2341 }
2342 
2343 /*
2344  * Start NLM service.
2345  */
2346 int
2347 nlm_svc_starting(struct nlm_globals *g, struct file *fp,
2348     const char *netid, struct knetconfig *knc)
2349 {
2350         int error;
2351         enum clnt_stat stat;
2352 
2353         VERIFY(g->run_status == NLM_ST_STARTING);
2354         VERIFY(g->nlm_gc_thread == NULL);
2355 
2356         error = nlm_nsm_init_local(&g->nlm_nsm);
2357         if (error != 0) {
2358                 NLM_ERR("Failed to initialize NSM handler "
2359                     "(error=%d)\n", error);
2360                 g->run_status = NLM_ST_DOWN;
2361                 return (error);
2362         }
2363 
2364         error = EIO;
2365 
2366         /*
2367          * Create an NLM garbage collector thread that will
2368          * clean up stale vholds and hosts objects.
2369          */
2370         g->nlm_gc_thread = zthread_create(NULL, 0, nlm_gc,
2371             g, 0, minclsyspri);
2372 
2373         /*
2374          * Send SIMU_CRASH to local statd to report that
2375          * NLM started, so that statd can report other hosts
2376          * about NLM state change.
2377          */
2378 
2379         stat = nlm_nsm_simu_crash(&g->nlm_nsm);
2380         if (stat != RPC_SUCCESS) {
2381                 NLM_ERR("Failed to connect to local statd "
2382                     "(rpcerr=%d)\n", stat);
2383                 goto shutdown_lm;
2384         }
2385 
2386         stat = nlm_nsm_stat(&g->nlm_nsm, &g->nsm_state);
2387         if (stat != RPC_SUCCESS) {
2388                 NLM_ERR("Failed to get the status of local statd "
2389                     "(rpcerr=%d)\n", stat);
2390                 goto shutdown_lm;
2391         }
2392 
2393         g->grace_threshold = ddi_get_lbolt() +
2394             SEC_TO_TICK(g->grace_period);
2395 
2396         /* Register endpoint used for communications with local NLM */
2397         error = nlm_svc_add_ep(fp, netid, knc);
2398         if (error != 0)
2399                 goto shutdown_lm;
2400 
2401         (void) svc_pool_control(NLM_SVCPOOL_ID,
2402             SVCPSET_SHUTDOWN_PROC, (void *)nlm_pool_shutdown);
2403         g->run_status = NLM_ST_UP;
2404         return (0);
2405 
2406 shutdown_lm:
2407         mutex_enter(&g->lock);
2408         g->run_status = NLM_ST_STOPPING;
2409         mutex_exit(&g->lock);
2410 
2411         nlm_svc_stopping(g);
2412         return (error);
2413 }
2414 
2415 /*
2416  * Called when the server pool is destroyed, so that
2417  * all transports are closed and no any server threads
2418  * exist.
2419  *
2420  * Just call lm_shutdown() to shut NLM down properly.
2421  */
2422 static void
2423 nlm_pool_shutdown(void)
2424 {
2425         (void) lm_shutdown();
2426 }
2427 
2428 /*
2429  * Stop NLM service, cleanup all resources
2430  * NLM owns at the moment.
2431  *
2432  * NOTE: NFS code can call NLM while it's
2433  * stopping or even if it's shut down. Any attempt
2434  * to lock file either on client or on the server
2435  * will fail if NLM isn't in NLM_ST_UP state.
2436  */
2437 void
2438 nlm_svc_stopping(struct nlm_globals *g)
2439 {
2440         mutex_enter(&g->lock);
2441         ASSERT(g->run_status == NLM_ST_STOPPING);
2442 
2443         /*
2444          * Ask NLM GC thread to exit and wait until it dies.
2445          */
2446         cv_signal(&g->nlm_gc_sched_cv);
2447         while (g->nlm_gc_thread != NULL)
2448                 cv_wait(&g->nlm_gc_finish_cv, &g->lock);
2449 
2450         mutex_exit(&g->lock);
2451 
2452         /*
2453          * Cleanup locks owned by NLM hosts.
2454          * NOTE: New hosts won't be created while
2455          * NLM is stopping.
2456          */
2457         while (!avl_is_empty(&g->nlm_hosts_tree)) {
2458                 struct nlm_host *hostp;
2459                 int busy_hosts = 0;
2460 
2461                 /*
2462                  * Iterate through all NLM hosts in the system
2463                  * and drop the locks they own by force.
2464                  */
2465                 hostp = avl_first(&g->nlm_hosts_tree);
2466                 while (hostp != NULL) {
2467                         /* Cleanup all client and server side locks */
2468                         nlm_client_cancel_all(g, hostp);
2469                         nlm_host_notify_server(hostp, 0);
2470 
2471                         mutex_enter(&hostp->nh_lock);
2472                         nlm_host_gc_vholds(hostp);
2473                         if (hostp->nh_refs > 0 || nlm_host_has_locks(hostp)) {
2474                                 /*
2475                                  * Oh, it seems the host is still busy, let
2476                                  * it some time to release and go to the
2477                                  * next one.
2478                                  */
2479 
2480                                 mutex_exit(&hostp->nh_lock);
2481                                 hostp = AVL_NEXT(&g->nlm_hosts_tree, hostp);
2482                                 busy_hosts++;
2483                                 continue;
2484                         }
2485 
2486                         mutex_exit(&hostp->nh_lock);
2487                         hostp = AVL_NEXT(&g->nlm_hosts_tree, hostp);
2488                 }
2489 
2490                 /*
2491                  * All hosts go to nlm_idle_hosts list after
2492                  * all locks they own are cleaned up and last refereces
2493                  * were dropped. Just destroy all hosts in nlm_idle_hosts
2494                  * list, they can not be removed from there while we're
2495                  * in stopping state.
2496                  */
2497                 while ((hostp = TAILQ_FIRST(&g->nlm_idle_hosts)) != NULL) {
2498                         nlm_host_unregister(g, hostp);
2499                         nlm_host_destroy(hostp);
2500                 }
2501 
2502                 if (busy_hosts > 0) {
2503                         /*
2504                          * There're some hosts that weren't cleaned
2505                          * up. Probably they're in resource cleanup
2506                          * process. Give them some time to do drop
2507                          * references.
2508                          */
2509                         delay(MSEC_TO_TICK(500));
2510                 }
2511         }
2512 
2513         ASSERT(TAILQ_EMPTY(&g->nlm_slocks));
2514 
2515         nlm_nsm_fini(&g->nlm_nsm);
2516         g->lockd_pid = 0;
2517         g->run_status = NLM_ST_DOWN;
2518 }
2519 
2520 /*
2521  * Returns TRUE if the given vnode has
2522  * any active or sleeping locks.
2523  */
2524 int
2525 nlm_vp_active(const vnode_t *vp)
2526 {
2527         struct nlm_globals *g;
2528         struct nlm_host *hostp;
2529         struct nlm_vhold *nvp;
2530         int active = 0;
2531 
2532         g = zone_getspecific(nlm_zone_key, curzone);
2533 
2534         /*
2535          * Server side NLM has locks on the given vnode
2536          * if there exist a vhold object that holds
2537          * the given vnode "vp" in one of NLM hosts.
2538          */
2539         mutex_enter(&g->lock);
2540         hostp = avl_first(&g->nlm_hosts_tree);
2541         while (hostp != NULL) {
2542                 mutex_enter(&hostp->nh_lock);
2543                 nvp = nlm_vhold_find_locked(hostp, vp);
2544                 mutex_exit(&hostp->nh_lock);
2545                 if (nvp != NULL) {
2546                         active = 1;
2547                         break;
2548                 }
2549 
2550                 hostp = AVL_NEXT(&g->nlm_hosts_tree, hostp);
2551         }
2552 
2553         mutex_exit(&g->lock);
2554         return (active);
2555 }
2556 
2557 /*
2558  * Called right before NFS export is going to
2559  * dissapear. The function finds all vnodes
2560  * belonging to the given export and cleans
2561  * all remote locks and share reservations
2562  * on them.
2563  */
2564 void
2565 nlm_unexport(struct exportinfo *exi)
2566 {
2567         struct nlm_globals *g;
2568         struct nlm_host *hostp;
2569 
2570         g = zone_getspecific(nlm_zone_key, curzone);
2571 
2572         mutex_enter(&g->lock);
2573         hostp = avl_first(&g->nlm_hosts_tree);
2574         while (hostp != NULL) {
2575                 struct nlm_vhold *nvp;
2576 
2577                 if (hostp->nh_flags & NLM_NH_INIDLE) {
2578                         TAILQ_REMOVE(&g->nlm_idle_hosts, hostp, nh_link);
2579                         hostp->nh_flags &= ~NLM_NH_INIDLE;
2580                 }
2581                 hostp->nh_refs++;
2582 
2583                 mutex_exit(&g->lock);
2584 
2585                 mutex_enter(&hostp->nh_lock);
2586                 TAILQ_FOREACH(nvp, &hostp->nh_vholds_list, nv_link) {
2587                         vnode_t *vp;
2588 
2589                         nvp->nv_refcnt++;
2590                         mutex_exit(&hostp->nh_lock);
2591 
2592                         vp = nvp->nv_vp;
2593 
2594                         if (!EQFSID(&exi->exi_fsid, &vp->v_vfsp->vfs_fsid))
2595                                 goto next_iter;
2596 
2597                         /*
2598                          * Ok, it we found out that vnode vp is under
2599                          * control by the exportinfo exi, now we need
2600                          * to drop all locks from this vnode, let's
2601                          * do it.
2602                          */
2603                         nlm_vhold_clean(nvp, hostp->nh_sysid);
2604 
2605                 next_iter:
2606                         mutex_enter(&hostp->nh_lock);
2607                         nvp->nv_refcnt--;
2608                 }
2609                 mutex_exit(&hostp->nh_lock);
2610 
2611                 mutex_enter(&g->lock);
2612                 nlm_host_release_locked(g, hostp);
2613 
2614                 hostp = AVL_NEXT(&g->nlm_hosts_tree, hostp);
2615         }
2616 
2617         mutex_exit(&g->lock);
2618 }
2619 
2620 /*
2621  * Allocate new unique sysid.
2622  * In case of failure (no available sysids)
2623  * return LM_NOSYSID.
2624  */
2625 sysid_t
2626 nlm_sysid_alloc(void)
2627 {
2628         sysid_t ret_sysid = LM_NOSYSID;
2629 
2630         rw_enter(&lm_lck, RW_WRITER);
2631         if (nlm_sysid_nidx > LM_SYSID_MAX)
2632                 nlm_sysid_nidx = LM_SYSID;
2633 
2634         if (!BT_TEST(nlm_sysid_bmap, nlm_sysid_nidx)) {
2635                 BT_SET(nlm_sysid_bmap, nlm_sysid_nidx);
2636                 ret_sysid = nlm_sysid_nidx++;
2637         } else {
2638                 index_t id;
2639 
2640                 id = bt_availbit(nlm_sysid_bmap, NLM_BMAP_NITEMS);
2641                 if (id > 0) {
2642                         nlm_sysid_nidx = id + 1;
2643                         ret_sysid = id;
2644                         BT_SET(nlm_sysid_bmap, id);
2645                 }
2646         }
2647 
2648         rw_exit(&lm_lck);
2649         return (ret_sysid);
2650 }
2651 
2652 void
2653 nlm_sysid_free(sysid_t sysid)
2654 {
2655         ASSERT(sysid >= LM_SYSID && sysid <= LM_SYSID_MAX);
2656 
2657         rw_enter(&lm_lck, RW_WRITER);
2658         ASSERT(BT_TEST(nlm_sysid_bmap, sysid));
2659         BT_CLEAR(nlm_sysid_bmap, sysid);
2660         rw_exit(&lm_lck);
2661 }
2662 
2663 /*
2664  * Return true if the request came from a local caller.
2665  * By necessity, this "knows" the netid names invented
2666  * in lm_svc() and nlm_netid_from_knetconfig().
2667  */
2668 bool_t
2669 nlm_caller_is_local(SVCXPRT *transp)
2670 {
2671         char *netid;
2672         struct netbuf *rtaddr;
2673 
2674         netid = svc_getnetid(transp);
2675         rtaddr = svc_getrpccaller(transp);
2676 
2677         if (netid == NULL)
2678                 return (FALSE);
2679 
2680         if (strcmp(netid, "ticlts") == 0 ||
2681             strcmp(netid, "ticotsord") == 0)
2682                 return (TRUE);
2683 
2684         if (strcmp(netid, "tcp") == 0 || strcmp(netid, "udp") == 0) {
2685                 struct sockaddr_in *sin = (void *)rtaddr->buf;
2686                 if (sin->sin_addr.s_addr == htonl(INADDR_LOOPBACK))
2687                         return (TRUE);
2688         }
2689         if (strcmp(netid, "tcp6") == 0 || strcmp(netid, "udp6") == 0) {
2690                 struct sockaddr_in6 *sin6 = (void *)rtaddr->buf;
2691                 if (IN6_IS_ADDR_LOOPBACK(&sin6->sin6_addr))
2692                         return (TRUE);
2693         }
2694 
2695         return (FALSE); /* unknown transport */
2696 }
2697 
2698 /*
2699  * Get netid string correspondig to the given knetconfig.
2700  * If not done already, save knc->knc_rdev in our table.
2701  */
2702 const char *
2703 nlm_knc_to_netid(struct knetconfig *knc)
2704 {
2705         int i;
2706         dev_t rdev;
2707         struct nlm_knc *nc;
2708         const char *netid = NULL;
2709 
2710         rw_enter(&lm_lck, RW_READER);
2711         for (i = 0; i < NLM_KNCS; i++) {
2712                 nc = &nlm_netconfigs[i];
2713 
2714                 if (nc->n_knc.knc_semantics == knc->knc_semantics &&
2715                     strcmp(nc->n_knc.knc_protofmly,
2716                     knc->knc_protofmly) == 0) {
2717                         netid = nc->n_netid;
2718                         rdev = nc->n_knc.knc_rdev;
2719                         break;
2720                 }
2721         }
2722         rw_exit(&lm_lck);
2723 
2724         if (netid != NULL && rdev == NODEV) {
2725                 rw_enter(&lm_lck, RW_WRITER);
2726                 if (nc->n_knc.knc_rdev == NODEV)
2727                         nc->n_knc.knc_rdev = knc->knc_rdev;
2728                 rw_exit(&lm_lck);
2729         }
2730 
2731         return (netid);
2732 }
2733 
2734 /*
2735  * Get a knetconfig corresponding to the given netid.
2736  * If there's no knetconfig for this netid, ENOENT
2737  * is returned.
2738  */
2739 int
2740 nlm_knc_from_netid(const char *netid, struct knetconfig *knc)
2741 {
2742         int i, ret;
2743 
2744         ret = ENOENT;
2745         for (i = 0; i < NLM_KNCS; i++) {
2746                 struct nlm_knc *nknc;
2747 
2748                 nknc = &nlm_netconfigs[i];
2749                 if (strcmp(netid, nknc->n_netid) == 0 &&
2750                     nknc->n_knc.knc_rdev != NODEV) {
2751                         *knc = nknc->n_knc;
2752                         ret = 0;
2753                         break;
2754                 }
2755         }
2756 
2757         return (ret);
2758 }
2759 
2760 void
2761 nlm_cprsuspend(void)
2762 {
2763         struct nlm_globals *g;
2764 
2765         rw_enter(&lm_lck, RW_READER);
2766         TAILQ_FOREACH(g, &nlm_zones_list, nlm_link)
2767                 nlm_suspend_zone(g);
2768 
2769         rw_exit(&lm_lck);
2770 }
2771 
2772 void
2773 nlm_cprresume(void)
2774 {
2775         struct nlm_globals *g;
2776 
2777         rw_enter(&lm_lck, RW_READER);
2778         TAILQ_FOREACH(g, &nlm_zones_list, nlm_link)
2779                 nlm_resume_zone(g);
2780 
2781         rw_exit(&lm_lck);
2782 }
2783 
2784 static void
2785 nlm_nsm_clnt_init(CLIENT *clnt, struct nlm_nsm *nsm)
2786 {
2787         (void) clnt_tli_kinit(clnt, &nsm->ns_knc, &nsm->ns_addr, 0,
2788             NLM_RPC_RETRIES, kcred);
2789 }
2790 
2791 static void
2792 nlm_netbuf_to_netobj(struct netbuf *addr, int *family, netobj *obj)
2793 {
2794         /* LINTED pointer alignment */
2795         struct sockaddr *sa = (struct sockaddr *)addr->buf;
2796 
2797         *family = sa->sa_family;
2798 
2799         switch (sa->sa_family) {
2800         case AF_INET: {
2801                 /* LINTED pointer alignment */
2802                 struct sockaddr_in *sin = (struct sockaddr_in *)sa;
2803 
2804                 obj->n_len = sizeof (sin->sin_addr);
2805                 obj->n_bytes = (char *)&sin->sin_addr;
2806                 break;
2807         }
2808 
2809         case AF_INET6: {
2810                 /* LINTED pointer alignment */
2811                 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa;
2812 
2813                 obj->n_len = sizeof (sin6->sin6_addr);
2814                 obj->n_bytes = (char *)&sin6->sin6_addr;
2815                 break;
2816         }
2817 
2818         default:
2819                 VERIFY(0);
2820                 break;
2821         }
2822 }