1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
  24  */
  25 
  26 /*
  27  *      Copyright 1983, 1984, 1985, 1986, 1987, 1988, 1989  AT&T.
  28  *              All rights reserved.
  29  */
  30 
  31 /*
  32  * Copyright 2018 Nexenta Systems, Inc.
  33  */
  34 
  35 #include <sys/types.h>
  36 #include <sys/param.h>
  37 #include <sys/time.h>
  38 #include <sys/vfs.h>
  39 #include <sys/vnode.h>
  40 #include <sys/socket.h>
  41 #include <sys/errno.h>
  42 #include <sys/uio.h>
  43 #include <sys/proc.h>
  44 #include <sys/user.h>
  45 #include <sys/file.h>
  46 #include <sys/tiuser.h>
  47 #include <sys/kmem.h>
  48 #include <sys/pathname.h>
  49 #include <sys/debug.h>
  50 #include <sys/vtrace.h>
  51 #include <sys/cmn_err.h>
  52 #include <sys/acl.h>
  53 #include <sys/utsname.h>
  54 #include <sys/sdt.h>
  55 #include <netinet/in.h>
  56 #include <sys/avl.h>
  57 
  58 #include <rpc/types.h>
  59 #include <rpc/auth.h>
  60 #include <rpc/svc.h>
  61 
  62 #include <nfs/nfs.h>
  63 #include <nfs/export.h>
  64 #include <nfs/nfssys.h>
  65 #include <nfs/nfs_clnt.h>
  66 #include <nfs/nfs_acl.h>
  67 #include <nfs/nfs_log.h>
  68 #include <nfs/lm.h>
  69 #include <sys/sunddi.h>
  70 
  71 static zone_key_t nfs_export_key;
  72 
  73 /*
  74  * exi_id support
  75  *
  76  * exi_id_next          The next exi_id available.
  77  * exi_id_overflow      The exi_id_next already overflowed, so we should
  78  *                      thoroughly check for duplicates.
  79  * exi_id_tree          AVL tree indexed by exi_id.
  80  * nfs_exi_id_lock      Lock to protect the export ID list
  81  *
  82  * All exi_id_next, exi_id_overflow, and exi_id_tree are protected by
  83  * nfs_exi_id_lock.
  84  */
  85 static int exi_id_next;
  86 static bool_t exi_id_overflow;
  87 avl_tree_t exi_id_tree;
  88 kmutex_t nfs_exi_id_lock;
  89 
  90 static int      unexport(nfs_export_t *, exportinfo_t *);
  91 static void     exportfree(exportinfo_t *);
  92 static int      loadindex(exportdata_t *);
  93 
  94 extern void     nfsauth_cache_free(exportinfo_t *);
  95 extern int      sec_svc_loadrootnames(int, int, caddr_t **, model_t);
  96 extern void     sec_svc_freerootnames(int, int, caddr_t *);
  97 
  98 static int      build_seclist_nodups(exportdata_t *, secinfo_t *, int);
  99 static void     srv_secinfo_add(secinfo_t **, int *, secinfo_t *, int, int);
 100 static void     srv_secinfo_remove(secinfo_t **, int *, secinfo_t *, int);
 101 static void     srv_secinfo_treeclimb(nfs_export_t *, exportinfo_t *,
 102                     secinfo_t *, int, bool_t);
 103 
 104 #ifdef VOLATILE_FH_TEST
 105 static struct ex_vol_rename *find_volrnm_fh(exportinfo_t *, nfs_fh4 *);
 106 static uint32_t find_volrnm_fh_id(exportinfo_t *, nfs_fh4 *);
 107 static void     free_volrnm_list(exportinfo_t *);
 108 #endif /* VOLATILE_FH_TEST */
 109 
 110 fhandle_t nullfh2;      /* for comparing V2 filehandles */
 111 
 112 /*
 113  * macro for static dtrace probes to trace server namespace ref count mods.
 114  */
 115 #define SECREF_TRACE(seclist, tag, flav, aftcnt) \
 116         DTRACE_PROBE4(nfss__i__nmspc__secref, struct secinfo *, (seclist), \
 117                 char *, (tag), int, (int)(flav), int, (int)(aftcnt))
 118 
 119 
 120 #define exptablehash(fsid, fid) (nfs_fhhash((fsid), (fid)) & (EXPTABLESIZE - 1))
 121 
 122 extern nfs_export_t *
 123 nfs_get_export(void)
 124 {
 125         return (zone_getspecific(nfs_export_key, curzone));
 126 }
 127 
 128 static uint8_t
 129 xor_hash(uint8_t *data, int len)
 130 {
 131         uint8_t h = 0;
 132 
 133         while (len--)
 134                 h ^= *data++;
 135 
 136         return (h);
 137 }
 138 
 139 /*
 140  * File handle hash function, XOR over all bytes in fsid and fid.
 141  */
 142 static unsigned
 143 nfs_fhhash(fsid_t *fsid, fid_t *fid)
 144 {
 145         int len;
 146         uint8_t h;
 147 
 148         h = xor_hash((uint8_t *)fsid, sizeof (fsid_t));
 149 
 150         /*
 151          * Sanity check the length before using it
 152          * blindly in case the client trashed it.
 153          */
 154         len = fid->fid_len > NFS_FH4MAXDATA ? 0 : fid->fid_len;
 155         h ^= xor_hash((uint8_t *)fid->fid_data, len);
 156 
 157         return ((unsigned)h);
 158 }
 159 
 160 /*
 161  * Free the memory allocated within a secinfo entry.
 162  */
 163 void
 164 srv_secinfo_entry_free(struct secinfo *secp)
 165 {
 166         if (secp->s_rootcnt > 0 && secp->s_rootnames != NULL) {
 167                 sec_svc_freerootnames(secp->s_secinfo.sc_rpcnum,
 168                     secp->s_rootcnt, secp->s_rootnames);
 169                 secp->s_rootcnt = 0;
 170         }
 171 
 172         if ((secp->s_secinfo.sc_rpcnum == RPCSEC_GSS) &&
 173             (secp->s_secinfo.sc_gss_mech_type)) {
 174                 kmem_free(secp->s_secinfo.sc_gss_mech_type->elements,
 175                     secp->s_secinfo.sc_gss_mech_type->length);
 176                 kmem_free(secp->s_secinfo.sc_gss_mech_type,
 177                     sizeof (rpc_gss_OID_desc));
 178                 secp->s_secinfo.sc_gss_mech_type = NULL;
 179         }
 180 }
 181 
 182 /*
 183  * Free a list of secinfo allocated in the exportdata structure.
 184  */
 185 void
 186 srv_secinfo_list_free(struct secinfo *secinfo, int cnt)
 187 {
 188         int i;
 189 
 190         if (cnt == 0)
 191                 return;
 192 
 193         for (i = 0; i < cnt; i++)
 194                 srv_secinfo_entry_free(&secinfo[i]);
 195 
 196         kmem_free(secinfo, cnt * sizeof (struct secinfo));
 197 }
 198 
 199 /*
 200  * Allocate and copy a secinfo data from "from" to "to".
 201  *
 202  * This routine is used by srv_secinfo_add() to add a new flavor to an
 203  * ancestor's export node. The rootnames are not copied because the
 204  * allowable rootname access only applies to the explicit exported node,
 205  * not its ancestor's.
 206  *
 207  * "to" should have already been allocated and zeroed before calling
 208  * this routine.
 209  *
 210  * This routine is used under the protection of exported_lock (RW_WRITER).
 211  */
 212 void
 213 srv_secinfo_copy(struct secinfo *from, struct secinfo *to)
 214 {
 215         to->s_secinfo.sc_nfsnum = from->s_secinfo.sc_nfsnum;
 216         to->s_secinfo.sc_rpcnum = from->s_secinfo.sc_rpcnum;
 217 
 218         if (from->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
 219                 to->s_secinfo.sc_service = from->s_secinfo.sc_service;
 220                 bcopy(from->s_secinfo.sc_name, to->s_secinfo.sc_name,
 221                     strlen(from->s_secinfo.sc_name));
 222                 bcopy(from->s_secinfo.sc_gss_mech, to->s_secinfo.sc_gss_mech,
 223                     strlen(from->s_secinfo.sc_gss_mech));
 224 
 225                 /* copy mechanism oid */
 226                 to->s_secinfo.sc_gss_mech_type =
 227                     kmem_alloc(sizeof (rpc_gss_OID_desc), KM_SLEEP);
 228                 to->s_secinfo.sc_gss_mech_type->length =
 229                     from->s_secinfo.sc_gss_mech_type->length;
 230                 to->s_secinfo.sc_gss_mech_type->elements =
 231                     kmem_alloc(from->s_secinfo.sc_gss_mech_type->length,
 232                     KM_SLEEP);
 233                 bcopy(from->s_secinfo.sc_gss_mech_type->elements,
 234                     to->s_secinfo.sc_gss_mech_type->elements,
 235                     from->s_secinfo.sc_gss_mech_type->length);
 236         }
 237 
 238         to->s_refcnt = from->s_refcnt;
 239         to->s_window = from->s_window;
 240         /* no need to copy the mode bits - s_flags */
 241 }
 242 
 243 /*
 244  * Create a secinfo array without duplicates.  The condensed
 245  * flavor list is used to propagate flavor ref counts  to an
 246  * export's ancestor pseudonodes.
 247  */
 248 static int
 249 build_seclist_nodups(exportdata_t *exd, secinfo_t *nodups, int exponly)
 250 {
 251         int ccnt, c;
 252         int ncnt, n;
 253         struct secinfo *cursec;
 254 
 255         ncnt = 0;
 256         ccnt = exd->ex_seccnt;
 257         cursec = exd->ex_secinfo;
 258 
 259         for (c = 0; c < ccnt; c++) {
 260 
 261                 if (exponly && ! SEC_REF_EXPORTED(&cursec[c]))
 262                         continue;
 263 
 264                 for (n = 0; n < ncnt; n++) {
 265                         if (nodups[n].s_secinfo.sc_nfsnum ==
 266                             cursec[c].s_secinfo.sc_nfsnum)
 267                                 break;
 268                 }
 269 
 270                 /*
 271                  * The structure copy below also copys ptrs embedded
 272                  * within struct secinfo.  The ptrs are copied but
 273                  * they are never freed from the nodups array.  If
 274                  * an ancestor's secinfo array doesn't contain one
 275                  * of the nodups flavors, then the entry is properly
 276                  * copied into the ancestor's secinfo array.
 277                  * (see srv_secinfo_copy)
 278                  */
 279                 if (n == ncnt) {
 280                         nodups[n] = cursec[c];
 281                         ncnt++;
 282                 }
 283         }
 284         return (ncnt);
 285 }
 286 
 287 /*
 288  * Add the new security flavors from newdata to the current list, pcursec.
 289  * Upon return, *pcursec has the newly merged secinfo list.
 290  *
 291  * There should be at least 1 secinfo entry in newsec.
 292  *
 293  * This routine is used under the protection of exported_lock (RW_WRITER).
 294  */
 295 static void
 296 srv_secinfo_add(secinfo_t **pcursec, int *pcurcnt, secinfo_t *newsec,
 297     int newcnt, int is_pseudo)
 298 {
 299         int ccnt, c;            /* sec count in current data - curdata */
 300         int n;                  /* index for newsec  - newsecinfo */
 301         int tcnt;               /* total sec count after merge */
 302         int mcnt;               /* total sec count after merge */
 303         struct secinfo *msec;   /* merged secinfo list */
 304         struct secinfo *cursec;
 305 
 306         cursec = *pcursec;
 307         ccnt = *pcurcnt;
 308 
 309         ASSERT(newcnt > 0);
 310         tcnt = ccnt + newcnt;
 311 
 312         for (n = 0; n < newcnt; n++) {
 313                 for (c = 0; c < ccnt; c++) {
 314                         if (newsec[n].s_secinfo.sc_nfsnum ==
 315                             cursec[c].s_secinfo.sc_nfsnum) {
 316                                 cursec[c].s_refcnt += newsec[n].s_refcnt;
 317                                 SECREF_TRACE(cursec, "add_ref",
 318                                     cursec[c].s_secinfo.sc_nfsnum,
 319                                     cursec[c].s_refcnt);
 320                                 tcnt--;
 321                                 break;
 322                         }
 323                 }
 324         }
 325 
 326         if (tcnt == ccnt)
 327                 return; /* no change; no new flavors */
 328 
 329         msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
 330 
 331         /* move current secinfo list data to the new list */
 332         for (c = 0; c < ccnt; c++)
 333                 msec[c] = cursec[c];
 334 
 335         /* Add the flavor that's not in the current data */
 336         mcnt = ccnt;
 337         for (n = 0; n < newcnt; n++) {
 338                 for (c = 0; c < ccnt; c++) {
 339                         if (newsec[n].s_secinfo.sc_nfsnum ==
 340                             cursec[c].s_secinfo.sc_nfsnum)
 341                                 break;
 342                 }
 343 
 344                 /* This is the one. Add it. */
 345                 if (c == ccnt) {
 346                         srv_secinfo_copy(&newsec[n], &msec[mcnt]);
 347 
 348                         if (is_pseudo)
 349                                 msec[mcnt].s_flags = M_RO;
 350 
 351                         SECREF_TRACE(msec, "new_ref",
 352                             msec[mcnt].s_secinfo.sc_nfsnum,
 353                             msec[mcnt].s_refcnt);
 354                         mcnt++;
 355                 }
 356         }
 357 
 358         ASSERT(mcnt == tcnt);
 359 
 360         /*
 361          * Done. Update curdata. Free the old secinfo list in
 362          * curdata and return the new sec array info
 363          */
 364         if (ccnt > 0)
 365                 kmem_free(cursec, ccnt * sizeof (struct secinfo));
 366         *pcurcnt = tcnt;
 367         *pcursec = msec;
 368 }
 369 
 370 /*
 371  * For NFS V4.
 372  * Remove the security data of the unexported node from its ancestors.
 373  * Assume there is at least one flavor entry in the current sec list
 374  * (pcursec).
 375  *
 376  * This routine is used under the protection of exported_lock (RW_WRITER).
 377  *
 378  * Every element of remsec is an explicitly exported flavor.  If
 379  * srv_secinfo_remove() is called fom an exportfs error path, then
 380  * the flavor list was derived from the user's share cmdline,
 381  * and all flavors are explicit.  If it was called from the unshare path,
 382  * build_seclist_nodups() was called with the exponly flag.
 383  */
 384 static void
 385 srv_secinfo_remove(secinfo_t **pcursec, int *pcurcnt, secinfo_t *remsec,
 386     int remcnt)
 387 {
 388         int ccnt, c;            /* sec count in current data - cursec */
 389         int r;                  /* sec count in removal data - remsec */
 390         int tcnt, mcnt;         /* total sec count after removing */
 391         struct secinfo *msec;   /* final secinfo list after removing */
 392         struct secinfo *cursec;
 393 
 394         cursec = *pcursec;
 395         ccnt = *pcurcnt;
 396         tcnt = ccnt;
 397 
 398         for (r = 0; r < remcnt; r++) {
 399                 /*
 400                  * At unshare/reshare time, only explicitly shared flavor ref
 401                  * counts are decremented and propagated to ancestors.
 402                  * Implicit flavor refs came from shared descendants, and
 403                  * they must be kept.
 404                  */
 405                 if (! SEC_REF_EXPORTED(&remsec[r]))
 406                         continue;
 407 
 408                 for (c = 0; c < ccnt; c++) {
 409                         if (remsec[r].s_secinfo.sc_nfsnum ==
 410                             cursec[c].s_secinfo.sc_nfsnum) {
 411 
 412                                 /*
 413                                  * Decrement secinfo reference count by 1.
 414                                  * If this entry is invalid after decrementing
 415                                  * the count (i.e. count < 1), this entry will
 416                                  * be removed.
 417                                  */
 418                                 cursec[c].s_refcnt--;
 419 
 420                                 SECREF_TRACE(cursec, "del_ref",
 421                                     cursec[c].s_secinfo.sc_nfsnum,
 422                                     cursec[c].s_refcnt);
 423 
 424                                 ASSERT(cursec[c].s_refcnt >= 0);
 425 
 426                                 if (SEC_REF_INVALID(&cursec[c]))
 427                                         tcnt--;
 428                                 break;
 429                         }
 430                 }
 431         }
 432 
 433         ASSERT(tcnt >= 0);
 434         if (tcnt == ccnt)
 435                 return; /* no change; no flavors to remove */
 436 
 437         if (tcnt == 0) {
 438                 srv_secinfo_list_free(cursec, ccnt);
 439                 *pcurcnt = 0;
 440                 *pcursec = NULL;
 441                 return;
 442         }
 443 
 444         msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
 445 
 446         /* walk thru the given secinfo list to remove the flavors */
 447         mcnt = 0;
 448         for (c = 0; c < ccnt; c++) {
 449                 if (SEC_REF_INVALID(&cursec[c])) {
 450                         srv_secinfo_entry_free(&cursec[c]);
 451                 } else {
 452                         msec[mcnt] = cursec[c];
 453                         mcnt++;
 454                 }
 455         }
 456 
 457         ASSERT(mcnt == tcnt);
 458         /*
 459          * Done. Update curdata.
 460          * Free the existing secinfo list in curdata. All pointers
 461          * within the list have either been moved to msec or freed
 462          * if it's invalid.
 463          */
 464         kmem_free(*pcursec, ccnt * sizeof (struct secinfo));
 465         *pcursec = msec;
 466         *pcurcnt = tcnt;
 467 }
 468 
 469 
 470 /*
 471  * For the reshare case, sec flavor accounting happens in 3 steps:
 472  * 1) propagate addition of new flavor refs up the ancestor tree
 473  * 2) transfer flavor refs of descendants to new/reshared exportdata
 474  * 3) propagate removal of old flavor refs up the ancestor tree
 475  *
 476  * srv_secinfo_exp2exp() implements step 2 of a reshare.  At this point,
 477  * the new flavor list has already been propagated up through the
 478  * ancestor tree via srv_secinfo_treeclimb().
 479  *
 480  * If there is more than 1 export reference to an old flavor (i.e. some
 481  * of its children shared with this flavor), this flavor information
 482  * needs to be transferred to the new exportdata struct.  A flavor in
 483  * the old exportdata has descendant refs when its s_refcnt > 1 or it
 484  * is implicitly shared (M_SEC4_EXPORTED not set in s_flags).
 485  *
 486  * SEC_REF_EXPORTED() is only true when  M_SEC4_EXPORTED is set
 487  * SEC_REF_SELF() is only true when both M_SEC4_EXPORTED is set and s_refcnt==1
 488  *
 489  * Transferring descendant flavor refcnts happens in 2 passes:
 490  * a) flavors used before (oldsecinfo) and after (curdata->ex_secinfo) reshare
 491  * b) flavors used before but not after reshare
 492  *
 493  * This routine is used under the protection of exported_lock (RW_WRITER).
 494  */
 495 void
 496 srv_secinfo_exp2exp(exportdata_t *curdata, secinfo_t *oldsecinfo, int ocnt)
 497 {
 498         int ccnt, c;            /* sec count in current data - curdata */
 499         int o;                  /* sec count in old data - oldsecinfo */
 500         int tcnt, mcnt;         /* total sec count after the transfer */
 501         struct secinfo *msec;   /* merged secinfo list */
 502 
 503         ccnt = curdata->ex_seccnt;
 504 
 505         ASSERT(ocnt > 0);
 506         ASSERT(!(curdata->ex_flags & EX_PSEUDO));
 507 
 508         /*
 509          * If the oldsecinfo has flavors with more than 1 reference count
 510          * and the flavor is specified in the reshare, transfer the flavor
 511          * refs to the new seclist (curdata.ex_secinfo).
 512          */
 513         tcnt = ccnt + ocnt;
 514 
 515         for (o = 0; o < ocnt; o++) {
 516 
 517                 if (SEC_REF_SELF(&oldsecinfo[o])) {
 518                         tcnt--;
 519                         continue;
 520                 }
 521 
 522                 for (c = 0; c < ccnt; c++) {
 523                         if (oldsecinfo[o].s_secinfo.sc_nfsnum ==
 524                             curdata->ex_secinfo[c].s_secinfo.sc_nfsnum) {
 525 
 526                                 /*
 527                                  * add old reference to the current
 528                                  * secinfo count
 529                                  */
 530                                 curdata->ex_secinfo[c].s_refcnt +=
 531                                     oldsecinfo[o].s_refcnt;
 532 
 533                                 /*
 534                                  * Delete the old export flavor
 535                                  * reference.  The initial reference
 536                                  * was created during srv_secinfo_add,
 537                                  * and the count is decremented below
 538                                  * to account for the initial reference.
 539                                  */
 540                                 if (SEC_REF_EXPORTED(&oldsecinfo[o]))
 541                                         curdata->ex_secinfo[c].s_refcnt--;
 542 
 543                                 SECREF_TRACE(curdata->ex_path,
 544                                     "reshare_xfer_common_child_refs",
 545                                     curdata->ex_secinfo[c].s_secinfo.sc_nfsnum,
 546                                     curdata->ex_secinfo[c].s_refcnt);
 547 
 548                                 ASSERT(curdata->ex_secinfo[c].s_refcnt >= 0);
 549 
 550                                 tcnt--;
 551                                 break;
 552                         }
 553                 }
 554         }
 555 
 556         if (tcnt == ccnt)
 557                 return; /* no more transfer to do */
 558 
 559         /*
 560          * oldsecinfo has flavors referenced by its children that are not
 561          * in the current (new) export flavor list.  Add these flavors.
 562          */
 563         msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
 564 
 565         /* move current secinfo list data to the new list */
 566         for (c = 0; c < ccnt; c++)
 567                 msec[c] = curdata->ex_secinfo[c];
 568 
 569         /*
 570          * Add the flavor that's not in the new export, but still
 571          * referenced by its children.
 572          */
 573         mcnt = ccnt;
 574         for (o = 0; o < ocnt; o++) {
 575                 if (! SEC_REF_SELF(&oldsecinfo[o])) {
 576                         for (c = 0; c < ccnt; c++) {
 577                                 if (oldsecinfo[o].s_secinfo.sc_nfsnum ==
 578                                     curdata->ex_secinfo[c].s_secinfo.sc_nfsnum)
 579                                         break;
 580                         }
 581 
 582                         /*
 583                          * This is the one. Add it. Decrement the ref count
 584                          * by 1 if the flavor is an explicitly shared flavor
 585                          * for the oldsecinfo export node.
 586                          */
 587                         if (c == ccnt) {
 588                                 srv_secinfo_copy(&oldsecinfo[o], &msec[mcnt]);
 589                                 if (SEC_REF_EXPORTED(&oldsecinfo[o]))
 590                                         msec[mcnt].s_refcnt--;
 591 
 592                                 SECREF_TRACE(curdata,
 593                                     "reshare_xfer_implicit_child_refs",
 594                                     msec[mcnt].s_secinfo.sc_nfsnum,
 595                                     msec[mcnt].s_refcnt);
 596 
 597                                 ASSERT(msec[mcnt].s_refcnt >= 0);
 598                                 mcnt++;
 599                         }
 600                 }
 601         }
 602 
 603         ASSERT(mcnt == tcnt);
 604         /*
 605          * Done. Update curdata, free the existing secinfo list in
 606          * curdata and set the new value.
 607          */
 608         if (ccnt > 0)
 609                 kmem_free(curdata->ex_secinfo, ccnt * sizeof (struct secinfo));
 610         curdata->ex_seccnt = tcnt;
 611         curdata->ex_secinfo = msec;
 612 }
 613 
 614 /*
 615  * When unsharing an old export node and the old node becomes a pseudo node,
 616  * if there is more than 1 export reference to an old flavor (i.e. some of
 617  * its children shared with this flavor), this flavor information needs to
 618  * be transferred to the new shared node.
 619  *
 620  * This routine is used under the protection of exported_lock (RW_WRITER).
 621  */
 622 void
 623 srv_secinfo_exp2pseu(exportdata_t *curdata, exportdata_t *olddata)
 624 {
 625         int ocnt, o;            /* sec count in transfer data - trandata */
 626         int tcnt, mcnt;         /* total sec count after transfer */
 627         struct secinfo *msec;   /* merged secinfo list */
 628 
 629         ASSERT(curdata->ex_flags & EX_PSEUDO);
 630         ASSERT(curdata->ex_seccnt == 0);
 631 
 632         ocnt = olddata->ex_seccnt;
 633 
 634         /*
 635          * If the olddata has flavors with more than 1 reference count,
 636          * transfer the information to the curdata.
 637          */
 638         tcnt = ocnt;
 639 
 640         for (o = 0; o < ocnt; o++) {
 641                 if (SEC_REF_SELF(&olddata->ex_secinfo[o]))
 642                         tcnt--;
 643         }
 644 
 645         if (tcnt == 0)
 646                 return; /* no transfer to do */
 647 
 648         msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
 649 
 650         mcnt = 0;
 651         for (o = 0; o < ocnt; o++) {
 652                 if (! SEC_REF_SELF(&olddata->ex_secinfo[o])) {
 653 
 654                         /*
 655                          * Decrement the reference count by 1 if the flavor is
 656                          * an explicitly shared flavor for the olddata export
 657                          * node.
 658                          */
 659                         srv_secinfo_copy(&olddata->ex_secinfo[o], &msec[mcnt]);
 660                         msec[mcnt].s_flags = M_RO;
 661                         if (SEC_REF_EXPORTED(&olddata->ex_secinfo[o]))
 662                                 msec[mcnt].s_refcnt--;
 663 
 664                         SECREF_TRACE(curdata, "unshare_morph_pseudo",
 665                             msec[mcnt].s_secinfo.sc_nfsnum,
 666                             msec[mcnt].s_refcnt);
 667 
 668                         ASSERT(msec[mcnt].s_refcnt >= 0);
 669                         mcnt++;
 670                 }
 671         }
 672 
 673         ASSERT(mcnt == tcnt);
 674         /*
 675          * Done. Update curdata.
 676          * Free up the existing secinfo list in curdata and
 677          * set the new value.
 678          */
 679         curdata->ex_seccnt = tcnt;
 680         curdata->ex_secinfo = msec;
 681 }
 682 
 683 /*
 684  * Find for given treenode the exportinfo which has its
 685  * exp_visible linked on its exi_visible list.
 686  *
 687  * Note: We could add new pointer either to treenode or
 688  * to exp_visible, which will point there directly.
 689  * This would buy some speed for some memory.
 690  */
 691 exportinfo_t *
 692 vis2exi(treenode_t *tnode)
 693 {
 694         exportinfo_t *exi_ret = NULL;
 695 
 696         for (;;) {
 697                 tnode = tnode->tree_parent;
 698                 if (TREE_ROOT(tnode)) {
 699                         exi_ret = tnode->tree_exi;
 700                         break;
 701                 }
 702         }
 703 
 704         ASSERT(exi_ret); /* Every visible should have its home exportinfo */
 705         return (exi_ret);
 706 }
 707 
 708 /*
 709  * For NFS V4.
 710  * Add or remove the newly exported or unexported security flavors of the
 711  * given exportinfo from its ancestors upto the system root.
 712  */
 713 void
 714 srv_secinfo_treeclimb(nfs_export_t *ne, exportinfo_t *exip, secinfo_t *sec,
 715     int seccnt, bool_t isadd)
 716 {
 717         treenode_t *tnode = exip->exi_tree;
 718 
 719         ASSERT(RW_WRITE_HELD(&ne->exported_lock));
 720         ASSERT(tnode != NULL);
 721 
 722         if (seccnt == 0)
 723                 return;
 724 
 725         /*
 726          * If flavors are being added and the new export root isn't
 727          * also VROOT, its implicitly allowed flavors are inherited from
 728          * its pseudonode.
 729          * Note - for VROOT exports the implicitly allowed flavors were
 730          * transferred from the PSEUDO export in exportfs()
 731          */
 732         if (isadd && !(exip->exi_vp->v_flag & VROOT) &&
 733             !VN_IS_CURZONEROOT(exip->exi_vp) &&
 734             tnode->tree_vis->vis_seccnt > 0) {
 735                 srv_secinfo_add(&exip->exi_export.ex_secinfo,
 736                     &exip->exi_export.ex_seccnt, tnode->tree_vis->vis_secinfo,
 737                     tnode->tree_vis->vis_seccnt, FALSE);
 738         }
 739 
 740         /*
 741          * Move to parent node and propagate sec flavor
 742          * to exportinfo and to visible structures.
 743          */
 744         tnode = tnode->tree_parent;
 745 
 746         while (tnode != NULL) {
 747 
 748                 /* If there is exportinfo, update it */
 749                 if (tnode->tree_exi != NULL) {
 750                         secinfo_t **pxsec =
 751                             &tnode->tree_exi->exi_export.ex_secinfo;
 752                         int *pxcnt = &tnode->tree_exi->exi_export.ex_seccnt;
 753                         int is_pseudo = PSEUDO(tnode->tree_exi);
 754                         if (isadd)
 755                                 srv_secinfo_add(pxsec, pxcnt, sec, seccnt,
 756                                     is_pseudo);
 757                         else
 758                                 srv_secinfo_remove(pxsec, pxcnt, sec, seccnt);
 759                 }
 760 
 761                 /* Update every visible - only root node has no visible */
 762                 if (tnode->tree_vis != NULL) {
 763                         secinfo_t **pxsec = &tnode->tree_vis->vis_secinfo;
 764                         int *pxcnt = &tnode->tree_vis->vis_seccnt;
 765                         if (isadd)
 766                                 srv_secinfo_add(pxsec, pxcnt, sec, seccnt,
 767                                     FALSE);
 768                         else
 769                                 srv_secinfo_remove(pxsec, pxcnt, sec, seccnt);
 770                 }
 771                 tnode = tnode->tree_parent;
 772         }
 773 }
 774 
 775 /* hash_name is a text substitution for either fid_hash or path_hash */
 776 #define exp_hash_unlink(exi, hash_name) \
 777         if (*(exi)->hash_name.bckt == (exi)) \
 778                 *(exi)->hash_name.bckt = (exi)->hash_name.next; \
 779         if ((exi)->hash_name.prev) \
 780                 (exi)->hash_name.prev->hash_name.next = (exi)->hash_name.next; \
 781         if ((exi)->hash_name.next) \
 782                 (exi)->hash_name.next->hash_name.prev = (exi)->hash_name.prev; \
 783         (exi)->hash_name.bckt = NULL;
 784 
 785 #define exp_hash_link(exi, hash_name, bucket) \
 786         (exi)->hash_name.bckt = (bucket); \
 787         (exi)->hash_name.prev = NULL; \
 788         (exi)->hash_name.next = *(bucket); \
 789         if ((exi)->hash_name.next) \
 790                 (exi)->hash_name.next->hash_name.prev = (exi); \
 791         *(bucket) = (exi);
 792 
 793 void
 794 export_link(nfs_export_t *ne, exportinfo_t *exi)
 795 {
 796         exportinfo_t **bckt;
 797 
 798         ASSERT(RW_WRITE_HELD(&ne->exported_lock));
 799 
 800         bckt = &ne->exptable[exptablehash(&exi->exi_fsid, &exi->exi_fid)];
 801         exp_hash_link(exi, fid_hash, bckt);
 802 
 803         bckt = &ne->exptable_path_hash[pkp_tab_hash(exi->exi_export.ex_path,
 804             strlen(exi->exi_export.ex_path))];
 805         exp_hash_link(exi, path_hash, bckt);
 806 }
 807 
 808 /*
 809  * Helper functions for exi_id handling
 810  */
 811 static int
 812 exi_id_compar(const void *v1, const void *v2)
 813 {
 814         const struct exportinfo *e1 = v1;
 815         const struct exportinfo *e2 = v2;
 816 
 817         if (e1->exi_id < e2->exi_id)
 818                 return (-1);
 819         if (e1->exi_id > e2->exi_id)
 820                 return (1);
 821 
 822         return (0);
 823 }
 824 
 825 int
 826 exi_id_get_next()
 827 {
 828         struct exportinfo e;
 829         int ret = exi_id_next;
 830 
 831         ASSERT(MUTEX_HELD(&nfs_exi_id_lock));
 832 
 833         do {
 834                 exi_id_next++;
 835                 if (exi_id_next == 0)
 836                         exi_id_overflow = TRUE;
 837 
 838                 if (!exi_id_overflow)
 839                         break;
 840 
 841                 if (exi_id_next == ret)
 842                         cmn_err(CE_PANIC, "exi_id exhausted");
 843 
 844                 e.exi_id = exi_id_next;
 845         } while (avl_find(&exi_id_tree, &e, NULL) != NULL);
 846 
 847         return (ret);
 848 }
 849 
 850 /*ARGSUSED*/
 851 static void *
 852 nfs_export_zone_init(zoneid_t zoneid)
 853 {
 854         int i;
 855         nfs_export_t *ne;
 856 
 857         ne = kmem_zalloc(sizeof (*ne), KM_SLEEP);
 858 
 859         rw_init(&ne->exported_lock, NULL, RW_DEFAULT, NULL);
 860 
 861         /*
 862          * Allocate the place holder for the public file handle, which
 863          * is all zeroes. It is initially set to the root filesystem.
 864          */
 865         ne->exi_root = kmem_zalloc(sizeof (*ne->exi_root), KM_SLEEP);
 866         ne->exi_public = ne->exi_root;
 867 
 868         ne->exi_root->exi_export.ex_flags = EX_PUBLIC;
 869         ne->exi_root->exi_export.ex_pathlen = 1;  /* length of "/" */
 870         ne->exi_root->exi_export.ex_path =
 871             kmem_alloc(ne->exi_root->exi_export.ex_pathlen + 1, KM_SLEEP);
 872         ne->exi_root->exi_export.ex_path[0] = '/';
 873         ne->exi_root->exi_export.ex_path[1] = '\0';
 874 
 875         ne->exi_root->exi_count = 1;
 876         mutex_init(&ne->exi_root->exi_lock, NULL, MUTEX_DEFAULT, NULL);
 877 
 878         ne->exi_root->exi_vp = ZONE_ROOTVP();
 879         ne->exi_rootfid.fid_len = MAXFIDSZ;
 880         if (vop_fid_pseudo(ne->exi_root->exi_vp, &ne->exi_rootfid) != 0) {
 881                 mutex_destroy(&ne->exi_root->exi_lock);
 882                 kmem_free(ne->exi_root->exi_export.ex_path,
 883                     ne->exi_root->exi_export.ex_pathlen + 1);
 884                 kmem_free(ne->exi_root, sizeof (*ne->exi_root));
 885                 return (NULL);
 886         }
 887 
 888         /* Initialize auth cache and auth cache lock */
 889         for (i = 0; i < AUTH_TABLESIZE; i++) {
 890                 ne->exi_root->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t),
 891                     KM_SLEEP);
 892                 avl_create(ne->exi_root->exi_cache[i],
 893                     nfsauth_cache_clnt_compar, sizeof (struct auth_cache_clnt),
 894                     offsetof(struct auth_cache_clnt, authc_link));
 895         }
 896         rw_init(&ne->exi_root->exi_cache_lock, NULL, RW_DEFAULT, NULL);
 897 
 898         /* Setup the fhandle template */
 899         ne->exi_root->exi_fh.fh_fsid = rootdir->v_vfsp->vfs_fsid;
 900         ne->exi_root->exi_fh.fh_xlen = ne->exi_rootfid.fid_len;
 901         bcopy(ne->exi_rootfid.fid_data, ne->exi_root->exi_fh.fh_xdata,
 902             ne->exi_rootfid.fid_len);
 903         ne->exi_root->exi_fh.fh_len = sizeof (ne->exi_root->exi_fh.fh_data);
 904 
 905         rw_enter(&ne->exported_lock, RW_WRITER);
 906 
 907         /* Publish the exportinfo in the hash table */
 908         export_link(ne, ne->exi_root);
 909 
 910         /* Initialize exi_id and exi_kstats */
 911         mutex_enter(&nfs_exi_id_lock);
 912         ne->exi_root->exi_id = exi_id_get_next();
 913         avl_add(&exi_id_tree, ne->exi_root);
 914         mutex_exit(&nfs_exi_id_lock);
 915 
 916         rw_exit(&ne->exported_lock);
 917         ne->ns_root = NULL;
 918 
 919         return (ne);
 920 }
 921 
 922 /*ARGSUSED*/
 923 static void
 924 nfs_export_zone_fini(zoneid_t zoneid, void *data)
 925 {
 926         int i;
 927         nfs_export_t *ne = data;
 928         struct exportinfo *exi;
 929 
 930         rw_enter(&ne->exported_lock, RW_WRITER);
 931         mutex_enter(&nfs_exi_id_lock);
 932 
 933         avl_remove(&exi_id_tree, ne->exi_root);
 934         export_unlink(ne, ne->exi_root);
 935 
 936         mutex_exit(&nfs_exi_id_lock);
 937         rw_exit(&ne->exported_lock);
 938 
 939         /* Deallocate the place holder for the public file handle */
 940         srv_secinfo_list_free(ne->exi_root->exi_export.ex_secinfo,
 941             ne->exi_root->exi_export.ex_seccnt);
 942         mutex_destroy(&ne->exi_root->exi_lock);
 943 
 944         rw_destroy(&ne->exi_root->exi_cache_lock);
 945         for (i = 0; i < AUTH_TABLESIZE; i++) {
 946                 avl_destroy(ne->exi_root->exi_cache[i]);
 947                 kmem_free(ne->exi_root->exi_cache[i], sizeof (avl_tree_t));
 948         }
 949 
 950         kmem_free(ne->exi_root->exi_export.ex_path,
 951             ne->exi_root->exi_export.ex_pathlen + 1);
 952         kmem_free(ne->exi_root, sizeof (*ne->exi_root));
 953 
 954         exi = avl_first(&exi_id_tree);
 955         while (exi != NULL) {
 956                 struct exportinfo *nexi = AVL_NEXT(&exi_id_tree, exi);
 957                 if (zoneid == exi->exi_zoneid)
 958                         (void) unexport(ne, exi);
 959                 exi = nexi;
 960         }
 961 
 962         rw_destroy(&ne->exported_lock);
 963         kmem_free(ne, sizeof (*ne));
 964 }
 965 
 966 /*
 967  * Initialization routine for export routines.
 968  * Should only be called once.
 969  */
 970 void
 971 nfs_exportinit(void)
 972 {
 973         mutex_init(&nfs_exi_id_lock, NULL, MUTEX_DEFAULT, NULL);
 974 
 975         /* exi_id handling initialization */
 976         exi_id_next = 0;
 977         exi_id_overflow = FALSE;
 978         avl_create(&exi_id_tree, exi_id_compar, sizeof (struct exportinfo),
 979             offsetof(struct exportinfo, exi_id_link));
 980 
 981         zone_key_create(&nfs_export_key, nfs_export_zone_init,
 982             NULL, nfs_export_zone_fini);
 983 
 984         nfslog_init();
 985 }
 986 
 987 /*
 988  * Finalization routine for export routines.
 989  */
 990 void
 991 nfs_exportfini(void)
 992 {
 993         (void) zone_key_delete(nfs_export_key);
 994         avl_destroy(&exi_id_tree);
 995         mutex_destroy(&nfs_exi_id_lock);
 996 }
 997 
 998 /*
 999  *  Check if 2 gss mechanism identifiers are the same.
1000  *
1001  *  return FALSE if not the same.
1002  *  return TRUE if the same.
1003  */
1004 static bool_t
1005 nfs_mech_equal(rpc_gss_OID mech1, rpc_gss_OID mech2)
1006 {
1007         if ((mech1->length == 0) && (mech2->length == 0))
1008                 return (TRUE);
1009 
1010         if (mech1->length != mech2->length)
1011                 return (FALSE);
1012 
1013         return (bcmp(mech1->elements, mech2->elements, mech1->length) == 0);
1014 }
1015 
1016 /*
1017  *  This routine is used by rpc to map rpc security number
1018  *  to nfs specific security flavor number.
1019  *
1020  *  The gss callback prototype is
1021  *  callback(struct svc_req *, gss_cred_id_t *, gss_ctx_id_t *,
1022  *                              rpc_gss_lock_t *, void **),
1023  *  since nfs does not use the gss_cred_id_t/gss_ctx_id_t arguments
1024  *  we cast them to void.
1025  */
1026 /*ARGSUSED*/
1027 bool_t
1028 rfs_gsscallback(struct svc_req *req, gss_cred_id_t deleg, void *gss_context,
1029     rpc_gss_lock_t *lock, void **cookie)
1030 {
1031         int i, j;
1032         rpc_gss_rawcred_t *raw_cred;
1033         struct exportinfo *exi;
1034         nfs_export_t *ne = nfs_get_export();
1035 
1036         /*
1037          * We don't deal with delegated credentials.
1038          */
1039         if (deleg != GSS_C_NO_CREDENTIAL)
1040                 return (FALSE);
1041 
1042         raw_cred = lock->raw_cred;
1043         *cookie = NULL;
1044 
1045         rw_enter(&ne->exported_lock, RW_READER);
1046 
1047         for (i = 0; i < EXPTABLESIZE; i++) {
1048                 exi = ne->exptable[i];
1049                 while (exi) {
1050                         if (exi->exi_export.ex_seccnt > 0) {
1051                                 struct secinfo *secp;
1052                                 seconfig_t *se;
1053                                 int seccnt;
1054 
1055                                 secp = exi->exi_export.ex_secinfo;
1056                                 seccnt = exi->exi_export.ex_seccnt;
1057                                 for (j = 0; j < seccnt; j++) {
1058                                         /*
1059                                          *  If there is a map of the triplet
1060                                          *  (mechanism, service, qop) between
1061                                          *  raw_cred and the exported flavor,
1062                                          *  get the psudo flavor number.
1063                                          *  Also qop should not be NULL, it
1064                                          *  should be "default" or something
1065                                          *  else.
1066                                          */
1067                                         se = &secp[j].s_secinfo;
1068                                         if ((se->sc_rpcnum == RPCSEC_GSS) &&
1069 
1070                                             (nfs_mech_equal(
1071                                             se->sc_gss_mech_type,
1072                                             raw_cred->mechanism)) &&
1073 
1074                                             (se->sc_service ==
1075                                             raw_cred->service) &&
1076                                             (raw_cred->qop == se->sc_qop)) {
1077 
1078                                                 *cookie = (void *)(uintptr_t)
1079                                                     se->sc_nfsnum;
1080                                                 goto done;
1081                                         }
1082                                 }
1083                         }
1084                         exi = exi->fid_hash.next;
1085                 }
1086         }
1087 done:
1088         rw_exit(&ne->exported_lock);
1089 
1090         /*
1091          * If no nfs pseudo number mapping can be found in the export
1092          * table, assign the nfsflavor to NFS_FLAVOR_NOMAP. In V4, we may
1093          * recover the flavor mismatch from NFS layer (NFS4ERR_WRONGSEC).
1094          *
1095          * For example:
1096          *      server first shares with krb5i;
1097          *      client mounts with krb5i;
1098          *      server re-shares with krb5p;
1099          *      client tries with krb5i, but no mapping can be found;
1100          *      rpcsec_gss module calls this routine to do the mapping,
1101          *              if this routine fails, request is rejected from
1102          *              the rpc layer.
1103          *      What we need is to let the nfs layer rejects the request.
1104          *      For V4, we can reject with NFS4ERR_WRONGSEC and the client
1105          *      may recover from it by getting the new flavor via SECINFO.
1106          *
1107          * nfs pseudo number for RPCSEC_GSS mapping (see nfssec.conf)
1108          * is owned by IANA (see RFC 2623).
1109          *
1110          * XXX NFS_FLAVOR_NOMAP is defined in Solaris to work around
1111          * the implementation issue. This number should not overlap with
1112          * any new IANA defined pseudo flavor numbers.
1113          */
1114         if (*cookie == NULL)
1115                 *cookie = (void *)NFS_FLAVOR_NOMAP;
1116 
1117         lock->locked = TRUE;
1118 
1119         return (TRUE);
1120 }
1121 
1122 
1123 /*
1124  * Exportfs system call; credentials should be checked before
1125  * calling this function.
1126  */
1127 int
1128 exportfs(struct exportfs_args *args, model_t model, cred_t *cr)
1129 {
1130         vnode_t *vp;
1131         vnode_t *dvp;
1132         struct exportdata *kex;
1133         struct exportinfo *exi = NULL;
1134         struct exportinfo *ex, *ex1, *ex2;
1135         fid_t fid;
1136         fsid_t fsid;
1137         int error;
1138         size_t allocsize;
1139         struct secinfo *sp;
1140         struct secinfo *exs;
1141         rpc_gss_callback_t cb;
1142         char *pathbuf;
1143         char *log_buffer;
1144         char *tagbuf;
1145         int callback;
1146         int allocd_seccnt;
1147         STRUCT_HANDLE(exportfs_args, uap);
1148         STRUCT_DECL(exportdata, uexi);
1149         struct secinfo newsec[MAX_FLAVORS];
1150         int newcnt;
1151         struct secinfo oldsec[MAX_FLAVORS];
1152         int oldcnt;
1153         int i;
1154         struct pathname lookpn;
1155         nfs_export_t *ne = nfs_get_export();
1156 
1157         STRUCT_SET_HANDLE(uap, model, args);
1158 
1159         /* Read in pathname from userspace */
1160         if (error = pn_get(STRUCT_FGETP(uap, dname), UIO_USERSPACE, &lookpn))
1161                 return (error);
1162 
1163         /* Walk the export list looking for that pathname */
1164         rw_enter(&ne->exported_lock, RW_READER);
1165         DTRACE_PROBE(nfss__i__exported_lock1_start);
1166         for (ex1 = ne->exptable_path_hash[pkp_tab_hash(lookpn.pn_path,
1167             strlen(lookpn.pn_path))]; ex1; ex1 = ex1->path_hash.next) {
1168                 if (ex1 != ne->exi_root && 0 ==
1169                     strcmp(ex1->exi_export.ex_path, lookpn.pn_path)) {
1170                         exi_hold(ex1);
1171                         break;
1172                 }
1173         }
1174         DTRACE_PROBE(nfss__i__exported_lock1_stop);
1175         rw_exit(&ne->exported_lock);
1176 
1177         /* Is this an unshare? */
1178         if (STRUCT_FGETP(uap, uex) == NULL) {
1179                 pn_free(&lookpn);
1180                 if (ex1 == NULL)
1181                         return (EINVAL);
1182                 error = unexport(ne, ex1);
1183                 exi_rele(ex1);
1184                 return (error);
1185         }
1186 
1187         /* It is a share or a re-share */
1188         error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1189             FOLLOW, &dvp, &vp);
1190         if (error == EINVAL) {
1191                 /*
1192                  * if fname resolves to / we get EINVAL error
1193                  * since we wanted the parent vnode. Try again
1194                  * with NULL dvp.
1195                  */
1196                 error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1197                     FOLLOW, NULL, &vp);
1198                 dvp = NULL;
1199         }
1200         if (!error && vp == NULL) {
1201                 /* Last component of fname not found */
1202                 if (dvp != NULL)
1203                         VN_RELE(dvp);
1204                 error = ENOENT;
1205         }
1206         if (error) {
1207                 pn_free(&lookpn);
1208                 if (ex1)
1209                         exi_rele(ex1);
1210                 return (error);
1211         }
1212 
1213         /*
1214          * 'vp' may be an AUTOFS node, so we perform a
1215          * VOP_ACCESS() to trigger the mount of the
1216          * intended filesystem, so we can share the intended
1217          * filesystem instead of the AUTOFS filesystem.
1218          */
1219         (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
1220 
1221         /*
1222          * We're interested in the top most filesystem.
1223          * This is specially important when uap->dname is a trigger
1224          * AUTOFS node, since we're really interested in sharing the
1225          * filesystem AUTOFS mounted as result of the VOP_ACCESS()
1226          * call not the AUTOFS node itself.
1227          */
1228         if (vn_mountedvfs(vp) != NULL) {
1229                 if (error = traverse(&vp)) {
1230                         VN_RELE(vp);
1231                         if (dvp != NULL)
1232                                 VN_RELE(dvp);
1233                         pn_free(&lookpn);
1234                         if (ex1)
1235                                 exi_rele(ex1);
1236                         return (error);
1237                 }
1238         }
1239 
1240         /* Do not allow sharing another vnode for already shared path */
1241         if (ex1 && !PSEUDO(ex1) && !VN_CMP(ex1->exi_vp, vp)) {
1242                 VN_RELE(vp);
1243                 if (dvp != NULL)
1244                         VN_RELE(dvp);
1245                 pn_free(&lookpn);
1246                 exi_rele(ex1);
1247                 return (EEXIST);
1248         }
1249         if (ex1)
1250                 exi_rele(ex1);
1251 
1252         /*
1253          * Get the vfs id
1254          */
1255         bzero(&fid, sizeof (fid));
1256         fid.fid_len = MAXFIDSZ;
1257         error = VOP_FID(vp, &fid, NULL);
1258         fsid = vp->v_vfsp->vfs_fsid;
1259 
1260         if (error) {
1261                 VN_RELE(vp);
1262                 if (dvp != NULL)
1263                         VN_RELE(dvp);
1264                 /*
1265                  * If VOP_FID returns ENOSPC then the fid supplied
1266                  * is too small.  For now we simply return EREMOTE.
1267                  */
1268                 if (error == ENOSPC)
1269                         error = EREMOTE;
1270                 pn_free(&lookpn);
1271                 return (error);
1272         }
1273 
1274         /*
1275          * Do not allow re-sharing a shared vnode under a different path
1276          * PSEUDO export has ex_path fabricated, e.g. "/tmp (pseudo)", skip it.
1277          */
1278         rw_enter(&ne->exported_lock, RW_READER);
1279         DTRACE_PROBE(nfss__i__exported_lock2_start);
1280         for (ex2 = ne->exptable[exptablehash(&fsid, &fid)]; ex2;
1281             ex2 = ex2->fid_hash.next) {
1282                 if (ex2 != ne->exi_root && !PSEUDO(ex2) &&
1283                     VN_CMP(ex2->exi_vp, vp) &&
1284                     strcmp(ex2->exi_export.ex_path, lookpn.pn_path) != 0) {
1285                         DTRACE_PROBE(nfss__i__exported_lock2_stop);
1286                         rw_exit(&ne->exported_lock);
1287                         VN_RELE(vp);
1288                         if (dvp != NULL)
1289                                 VN_RELE(dvp);
1290                         pn_free(&lookpn);
1291                         return (EEXIST);
1292                 }
1293         }
1294         DTRACE_PROBE(nfss__i__exported_lock2_stop);
1295         rw_exit(&ne->exported_lock);
1296         pn_free(&lookpn);
1297 
1298         exi = kmem_zalloc(sizeof (*exi), KM_SLEEP);
1299         exi->exi_fsid = fsid;
1300         exi->exi_fid = fid;
1301         exi->exi_vp = vp;
1302         exi->exi_count = 1;
1303         exi->exi_zoneid = crgetzoneid(cr);
1304         exi->exi_volatile_dev = (vfssw[vp->v_vfsp->vfs_fstype].vsw_flag &
1305             VSW_VOLATILEDEV) ? 1 : 0;
1306         mutex_init(&exi->exi_lock, NULL, MUTEX_DEFAULT, NULL);
1307         exi->exi_dvp = dvp;
1308 
1309         /*
1310          * Initialize auth cache and auth cache lock
1311          */
1312         for (i = 0; i < AUTH_TABLESIZE; i++) {
1313                 exi->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1314                 avl_create(exi->exi_cache[i], nfsauth_cache_clnt_compar,
1315                     sizeof (struct auth_cache_clnt),
1316                     offsetof(struct auth_cache_clnt, authc_link));
1317         }
1318         rw_init(&exi->exi_cache_lock, NULL, RW_DEFAULT, NULL);
1319 
1320         /*
1321          * Build up the template fhandle
1322          */
1323         exi->exi_fh.fh_fsid = fsid;
1324         if (exi->exi_fid.fid_len > sizeof (exi->exi_fh.fh_xdata)) {
1325                 error = EREMOTE;
1326                 goto out1;
1327         }
1328         exi->exi_fh.fh_xlen = exi->exi_fid.fid_len;
1329         bcopy(exi->exi_fid.fid_data, exi->exi_fh.fh_xdata,
1330             exi->exi_fid.fid_len);
1331 
1332         exi->exi_fh.fh_len = sizeof (exi->exi_fh.fh_data);
1333 
1334         kex = &exi->exi_export;
1335 
1336         /*
1337          * Load in everything, and do sanity checking
1338          */
1339         STRUCT_INIT(uexi, model);
1340         if (copyin(STRUCT_FGETP(uap, uex), STRUCT_BUF(uexi),
1341             STRUCT_SIZE(uexi))) {
1342                 error = EFAULT;
1343                 goto out1;
1344         }
1345 
1346         kex->ex_version = STRUCT_FGET(uexi, ex_version);
1347         if (kex->ex_version != EX_CURRENT_VERSION) {
1348                 error = EINVAL;
1349                 cmn_err(CE_WARN,
1350                     "NFS: exportfs requires export struct version 2 - got %d\n",
1351                     kex->ex_version);
1352                 goto out1;
1353         }
1354 
1355         /*
1356          * Must have at least one security entry
1357          */
1358         kex->ex_seccnt = STRUCT_FGET(uexi, ex_seccnt);
1359         if (kex->ex_seccnt < 1) {
1360                 error = EINVAL;
1361                 goto out1;
1362         }
1363 
1364         kex->ex_path = STRUCT_FGETP(uexi, ex_path);
1365         kex->ex_pathlen = STRUCT_FGET(uexi, ex_pathlen);
1366         kex->ex_flags = STRUCT_FGET(uexi, ex_flags);
1367         kex->ex_anon = STRUCT_FGET(uexi, ex_anon);
1368         kex->ex_secinfo = STRUCT_FGETP(uexi, ex_secinfo);
1369         kex->ex_index = STRUCT_FGETP(uexi, ex_index);
1370         kex->ex_log_buffer = STRUCT_FGETP(uexi, ex_log_buffer);
1371         kex->ex_log_bufferlen = STRUCT_FGET(uexi, ex_log_bufferlen);
1372         kex->ex_tag = STRUCT_FGETP(uexi, ex_tag);
1373         kex->ex_taglen = STRUCT_FGET(uexi, ex_taglen);
1374 
1375         /*
1376          * Copy the exported pathname into
1377          * an appropriately sized buffer.
1378          */
1379         pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1380         if (copyinstr(kex->ex_path, pathbuf, MAXPATHLEN, &kex->ex_pathlen)) {
1381                 kmem_free(pathbuf, MAXPATHLEN);
1382                 error = EFAULT;
1383                 goto out1;
1384         }
1385         kex->ex_path = kmem_alloc(kex->ex_pathlen + 1, KM_SLEEP);
1386         bcopy(pathbuf, kex->ex_path, kex->ex_pathlen);
1387         kex->ex_path[kex->ex_pathlen] = '\0';
1388         kmem_free(pathbuf, MAXPATHLEN);
1389 
1390         /*
1391          * Get the path to the logging buffer and the tag
1392          */
1393         if (kex->ex_flags & EX_LOG) {
1394                 log_buffer = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1395                 if (copyinstr(kex->ex_log_buffer, log_buffer, MAXPATHLEN,
1396                     &kex->ex_log_bufferlen)) {
1397                         kmem_free(log_buffer, MAXPATHLEN);
1398                         error = EFAULT;
1399                         goto out2;
1400                 }
1401                 kex->ex_log_buffer =
1402                     kmem_alloc(kex->ex_log_bufferlen + 1, KM_SLEEP);
1403                 bcopy(log_buffer, kex->ex_log_buffer, kex->ex_log_bufferlen);
1404                 kex->ex_log_buffer[kex->ex_log_bufferlen] = '\0';
1405                 kmem_free(log_buffer, MAXPATHLEN);
1406 
1407                 tagbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1408                 if (copyinstr(kex->ex_tag, tagbuf, MAXPATHLEN,
1409                     &kex->ex_taglen)) {
1410                         kmem_free(tagbuf, MAXPATHLEN);
1411                         error = EFAULT;
1412                         goto out3;
1413                 }
1414                 kex->ex_tag = kmem_alloc(kex->ex_taglen + 1, KM_SLEEP);
1415                 bcopy(tagbuf, kex->ex_tag, kex->ex_taglen);
1416                 kex->ex_tag[kex->ex_taglen] = '\0';
1417                 kmem_free(tagbuf, MAXPATHLEN);
1418         }
1419 
1420         /*
1421          * Load the security information for each flavor
1422          */
1423         allocsize = kex->ex_seccnt * SIZEOF_STRUCT(secinfo, model);
1424         sp = kmem_zalloc(allocsize, KM_SLEEP);
1425         if (copyin(kex->ex_secinfo, sp, allocsize)) {
1426                 kmem_free(sp, allocsize);
1427                 error = EFAULT;
1428                 goto out4;
1429         }
1430 
1431         /*
1432          * All of these nested structures need to be converted to
1433          * the kernel native format.
1434          */
1435         if (model != DATAMODEL_NATIVE) {
1436                 size_t allocsize2;
1437                 struct secinfo *sp2;
1438 
1439                 allocsize2 = kex->ex_seccnt * sizeof (struct secinfo);
1440                 sp2 = kmem_zalloc(allocsize2, KM_SLEEP);
1441 
1442                 for (i = 0; i < kex->ex_seccnt; i++) {
1443                         STRUCT_HANDLE(secinfo, usi);
1444 
1445                         STRUCT_SET_HANDLE(usi, model,
1446                             (struct secinfo *)((caddr_t)sp +
1447                             (i * SIZEOF_STRUCT(secinfo, model))));
1448                         bcopy(STRUCT_FGET(usi, s_secinfo.sc_name),
1449                             sp2[i].s_secinfo.sc_name, MAX_NAME_LEN);
1450                         sp2[i].s_secinfo.sc_nfsnum =
1451                             STRUCT_FGET(usi, s_secinfo.sc_nfsnum);
1452                         sp2[i].s_secinfo.sc_rpcnum =
1453                             STRUCT_FGET(usi, s_secinfo.sc_rpcnum);
1454                         bcopy(STRUCT_FGET(usi, s_secinfo.sc_gss_mech),
1455                             sp2[i].s_secinfo.sc_gss_mech, MAX_NAME_LEN);
1456                         sp2[i].s_secinfo.sc_gss_mech_type =
1457                             STRUCT_FGETP(usi, s_secinfo.sc_gss_mech_type);
1458                         sp2[i].s_secinfo.sc_qop =
1459                             STRUCT_FGET(usi, s_secinfo.sc_qop);
1460                         sp2[i].s_secinfo.sc_service =
1461                             STRUCT_FGET(usi, s_secinfo.sc_service);
1462 
1463                         sp2[i].s_flags = STRUCT_FGET(usi, s_flags);
1464                         sp2[i].s_window = STRUCT_FGET(usi, s_window);
1465                         sp2[i].s_rootid = STRUCT_FGET(usi, s_rootid);
1466                         sp2[i].s_rootcnt = STRUCT_FGET(usi, s_rootcnt);
1467                         sp2[i].s_rootnames = STRUCT_FGETP(usi, s_rootnames);
1468                 }
1469                 kmem_free(sp, allocsize);
1470                 sp = sp2;
1471                 allocsize = allocsize2;
1472         }
1473 
1474         kex->ex_secinfo = sp;
1475 
1476         /*
1477          * And now copy rootnames for each individual secinfo.
1478          */
1479         callback = 0;
1480         allocd_seccnt = 0;
1481         while (allocd_seccnt < kex->ex_seccnt) {
1482 
1483                 exs = &sp[allocd_seccnt];
1484                 if (exs->s_rootcnt > 0) {
1485                         if (!sec_svc_loadrootnames(exs->s_secinfo.sc_rpcnum,
1486                             exs->s_rootcnt, &exs->s_rootnames, model)) {
1487                                 error = EFAULT;
1488                                 goto out5;
1489                         }
1490                 }
1491 
1492                 if (exs->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
1493                         rpc_gss_OID mech_tmp;
1494                         STRUCT_DECL(rpc_gss_OID_s, umech_tmp);
1495                         caddr_t elements_tmp;
1496 
1497                         /* Copyin mechanism type */
1498                         STRUCT_INIT(umech_tmp, model);
1499                         mech_tmp = kmem_alloc(sizeof (*mech_tmp), KM_SLEEP);
1500                         if (copyin(exs->s_secinfo.sc_gss_mech_type,
1501                             STRUCT_BUF(umech_tmp), STRUCT_SIZE(umech_tmp))) {
1502                                 kmem_free(mech_tmp, sizeof (*mech_tmp));
1503                                 error = EFAULT;
1504                                 goto out5;
1505                         }
1506                         mech_tmp->length = STRUCT_FGET(umech_tmp, length);
1507                         mech_tmp->elements = STRUCT_FGETP(umech_tmp, elements);
1508 
1509                         elements_tmp = kmem_alloc(mech_tmp->length, KM_SLEEP);
1510                         if (copyin(mech_tmp->elements, elements_tmp,
1511                             mech_tmp->length)) {
1512                                 kmem_free(elements_tmp, mech_tmp->length);
1513                                 kmem_free(mech_tmp, sizeof (*mech_tmp));
1514                                 error = EFAULT;
1515                                 goto out5;
1516                         }
1517                         mech_tmp->elements = elements_tmp;
1518                         exs->s_secinfo.sc_gss_mech_type = mech_tmp;
1519                         allocd_seccnt++;
1520 
1521                         callback = 1;
1522                 } else
1523                         allocd_seccnt++;
1524         }
1525 
1526         /*
1527          * Init the secinfo reference count and mark these flavors
1528          * explicitly exported flavors.
1529          */
1530         for (i = 0; i < kex->ex_seccnt; i++) {
1531                 kex->ex_secinfo[i].s_flags |= M_4SEC_EXPORTED;
1532                 kex->ex_secinfo[i].s_refcnt = 1;
1533         }
1534 
1535         /*
1536          *  Set up rpcsec_gss callback routine entry if any.
1537          */
1538         if (callback) {
1539                 cb.callback = rfs_gsscallback;
1540                 cb.program = NFS_ACL_PROGRAM;
1541                 for (cb.version = NFS_ACL_VERSMIN;
1542                     cb.version <= NFS_ACL_VERSMAX; cb.version++) {
1543                         (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1544                             (void *)&cb);
1545                 }
1546 
1547                 cb.program = NFS_PROGRAM;
1548                 for (cb.version = NFS_VERSMIN;
1549                     cb.version <= NFS_VERSMAX; cb.version++) {
1550                         (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1551                             (void *)&cb);
1552                 }
1553         }
1554 
1555         /*
1556          * Check the index flag. Do this here to avoid holding the
1557          * lock while dealing with the index option (as we do with
1558          * the public option).
1559          */
1560         if (kex->ex_flags & EX_INDEX) {
1561                 if (!kex->ex_index) {        /* sanity check */
1562                         error = EINVAL;
1563                         goto out5;
1564                 }
1565                 if (error = loadindex(kex))
1566                         goto out5;
1567         }
1568 
1569         if (kex->ex_flags & EX_LOG) {
1570                 if (error = nfslog_setup(exi))
1571                         goto out6;
1572         }
1573 
1574         /*
1575          * Insert the new entry at the front of the export list
1576          */
1577         rw_enter(&ne->exported_lock, RW_WRITER);
1578         DTRACE_PROBE(nfss__i__exported_lock3_start);
1579 
1580         export_link(ne, exi);
1581 
1582         /*
1583          * Check the rest of the list for an old entry for the fs.
1584          * If one is found then unlink it, wait until this is the
1585          * only reference and then free it.
1586          */
1587         for (ex = exi->fid_hash.next; ex != NULL; ex = ex->fid_hash.next) {
1588                 if (ex != ne->exi_root && VN_CMP(ex->exi_vp, vp)) {
1589                         mutex_enter(&nfs_exi_id_lock);
1590                         avl_remove(&exi_id_tree, ex);
1591                         mutex_exit(&nfs_exi_id_lock);
1592                         export_unlink(ne, ex);
1593                         break;
1594                 }
1595         }
1596 
1597         /*
1598          * If the public filehandle is pointing at the
1599          * old entry, then point it back at the root.
1600          */
1601         if (ex != NULL && ex == ne->exi_public)
1602                 ne->exi_public = ne->exi_root;
1603 
1604         /*
1605          * If the public flag is on, make the global exi_public
1606          * point to this entry and turn off the public bit so that
1607          * we can distinguish it from the place holder export.
1608          */
1609         if (kex->ex_flags & EX_PUBLIC) {
1610                 ne->exi_public = exi;
1611                 kex->ex_flags &= ~EX_PUBLIC;
1612         }
1613 
1614 #ifdef VOLATILE_FH_TEST
1615         /*
1616          * Set up the volatile_id value if volatile on share.
1617          * The list of volatile renamed filehandles is always destroyed,
1618          * if the fs was reshared.
1619          */
1620         if (kex->ex_flags & EX_VOLFH)
1621                 exi->exi_volatile_id = gethrestime_sec();
1622 
1623         mutex_init(&exi->exi_vol_rename_lock, NULL, MUTEX_DEFAULT, NULL);
1624 #endif /* VOLATILE_FH_TEST */
1625 
1626         /*
1627          * If this is a new export, then climb up
1628          * the tree and check if any pseudo exports
1629          * need to be created to provide a path for
1630          * NFS v4 clients.
1631          */
1632         if (ex == NULL) {
1633                 error = treeclimb_export(exi);
1634                 if (error)
1635                         goto out7;
1636         } else {
1637                 /* If it's a re-export update namespace tree */
1638                 exi->exi_tree = ex->exi_tree;
1639                 exi->exi_tree->tree_exi = exi;
1640 
1641                 /* Update the change timestamp */
1642                 tree_update_change(ne, exi->exi_tree, NULL);
1643         }
1644 
1645         /*
1646          * build a unique flavor list from the flavors specified
1647          * in the share cmd.  unique means that each flavor only
1648          * appears once in the secinfo list -- no duplicates allowed.
1649          */
1650         newcnt = build_seclist_nodups(&exi->exi_export, newsec, FALSE);
1651 
1652         srv_secinfo_treeclimb(ne, exi, newsec, newcnt, TRUE);
1653 
1654         /*
1655          * If re-sharing an old export entry, update the secinfo data
1656          * depending on if the old entry is a pseudo node or not.
1657          */
1658         if (ex != NULL) {
1659                 oldcnt = build_seclist_nodups(&ex->exi_export, oldsec, FALSE);
1660                 if (PSEUDO(ex)) {
1661                         /*
1662                          * The dir being shared is a pseudo export root (which
1663                          * will be transformed into a real export root).  The
1664                          * flavor(s) of the new share were propagated to the
1665                          * ancestors by srv_secinfo_treeclimb() above.  Now
1666                          * transfer the implicit flavor refs from the old
1667                          * pseudo exprot root to the new (real) export root.
1668                          */
1669                         srv_secinfo_add(&exi->exi_export.ex_secinfo,
1670                             &exi->exi_export.ex_seccnt, oldsec, oldcnt, TRUE);
1671                 } else {
1672                         /*
1673                          * First transfer implicit flavor refs to new export.
1674                          * Remove old flavor refs last.
1675                          */
1676                         srv_secinfo_exp2exp(&exi->exi_export, oldsec, oldcnt);
1677                         srv_secinfo_treeclimb(ne, ex, oldsec, oldcnt, FALSE);
1678                 }
1679         }
1680 
1681         /*
1682          * If it's a re-export and the old entry has a pseudonode list,
1683          * transfer it to the new export.
1684          */
1685         if (ex != NULL && (ex->exi_visible != NULL)) {
1686                 exi->exi_visible = ex->exi_visible;
1687                 ex->exi_visible = NULL;
1688         }
1689 
1690         /*
1691          * Initialize exi_id and exi_kstats
1692          */
1693         if (ex != NULL) {
1694                 exi->exi_id = ex->exi_id;
1695         } else {
1696                 mutex_enter(&nfs_exi_id_lock);
1697                 exi->exi_id = exi_id_get_next();
1698                 mutex_exit(&nfs_exi_id_lock);
1699         }
1700         mutex_enter(&nfs_exi_id_lock);
1701         avl_add(&exi_id_tree, exi);
1702         mutex_exit(&nfs_exi_id_lock);
1703 
1704         DTRACE_PROBE(nfss__i__exported_lock3_stop);
1705         rw_exit(&ne->exported_lock);
1706 
1707         if (ne->exi_public == exi || kex->ex_flags & EX_LOG) {
1708                 /*
1709                  * Log share operation to this buffer only.
1710                  */
1711                 nfslog_share_record(exi, cr);
1712         }
1713 
1714         if (ex != NULL)
1715                 exi_rele(ex);
1716 
1717         return (0);
1718 
1719 out7:
1720         /* Unlink the new export in exptable. */
1721         export_unlink(ne, exi);
1722         DTRACE_PROBE(nfss__i__exported_lock3_stop);
1723         rw_exit(&ne->exported_lock);
1724 out6:
1725         if (kex->ex_flags & EX_INDEX)
1726                 kmem_free(kex->ex_index, strlen(kex->ex_index) + 1);
1727 out5:
1728         /* free partially completed allocation */
1729         while (--allocd_seccnt >= 0) {
1730                 exs = &kex->ex_secinfo[allocd_seccnt];
1731                 srv_secinfo_entry_free(exs);
1732         }
1733 
1734         if (kex->ex_secinfo) {
1735                 kmem_free(kex->ex_secinfo,
1736                     kex->ex_seccnt * sizeof (struct secinfo));
1737         }
1738 
1739 out4:
1740         if ((kex->ex_flags & EX_LOG) && kex->ex_tag != NULL)
1741                 kmem_free(kex->ex_tag, kex->ex_taglen + 1);
1742 out3:
1743         if ((kex->ex_flags & EX_LOG) && kex->ex_log_buffer != NULL)
1744                 kmem_free(kex->ex_log_buffer, kex->ex_log_bufferlen + 1);
1745 out2:
1746         kmem_free(kex->ex_path, kex->ex_pathlen + 1);
1747 out1:
1748         VN_RELE(vp);
1749         if (dvp != NULL)
1750                 VN_RELE(dvp);
1751         mutex_destroy(&exi->exi_lock);
1752         rw_destroy(&exi->exi_cache_lock);
1753         for (i = 0; i < AUTH_TABLESIZE; i++) {
1754                 avl_destroy(exi->exi_cache[i]);
1755                 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
1756         }
1757 
1758         kmem_free(exi, sizeof (*exi));
1759 
1760         return (error);
1761 }
1762 
1763 /*
1764  * Remove the exportinfo from the export list
1765  */
1766 void
1767 export_unlink(nfs_export_t *ne, struct exportinfo *exi)
1768 {
1769         ASSERT(RW_WRITE_HELD(&ne->exported_lock));
1770 
1771         exp_hash_unlink(exi, fid_hash);
1772         exp_hash_unlink(exi, path_hash);
1773 }
1774 
1775 /*
1776  * Unexport an exported filesystem
1777  */
1778 static int
1779 unexport(nfs_export_t *ne, struct exportinfo *exi)
1780 {
1781         struct secinfo cursec[MAX_FLAVORS];
1782         int curcnt;
1783 
1784         rw_enter(&ne->exported_lock, RW_WRITER);
1785 
1786         /* Check if exi is still linked in the export table */
1787         if (!EXP_LINKED(exi) || PSEUDO(exi)) {
1788                 rw_exit(&ne->exported_lock);
1789                 return (EINVAL);
1790         }
1791 
1792         mutex_enter(&nfs_exi_id_lock);
1793         avl_remove(&exi_id_tree, exi);
1794         mutex_exit(&nfs_exi_id_lock);
1795         export_unlink(ne, exi);
1796 
1797         /*
1798          * Remove security flavors before treeclimb_unexport() is called
1799          * because srv_secinfo_treeclimb needs the namespace tree
1800          */
1801         curcnt = build_seclist_nodups(&exi->exi_export, cursec, TRUE);
1802         srv_secinfo_treeclimb(ne, exi, cursec, curcnt, FALSE);
1803 
1804         /*
1805          * If there's a visible list, then need to leave
1806          * a pseudo export here to retain the visible list
1807          * for paths to exports below.
1808          */
1809         if (exi->exi_visible != NULL) {
1810                 struct exportinfo *newexi;
1811 
1812                 newexi = pseudo_exportfs(ne, exi->exi_vp, &exi->exi_fid,
1813                     exi->exi_visible, &exi->exi_export);
1814                 exi->exi_visible = NULL;
1815 
1816                 /* interconnect the existing treenode with the new exportinfo */
1817                 newexi->exi_tree = exi->exi_tree;
1818                 newexi->exi_tree->tree_exi = newexi;
1819 
1820                 /* Update the change timestamp */
1821                 tree_update_change(ne, exi->exi_tree, NULL);
1822         } else {
1823                 treeclimb_unexport(ne, exi);
1824         }
1825 
1826         rw_exit(&ne->exported_lock);
1827 
1828         /*
1829          * Need to call into the NFSv4 server and release all data
1830          * held on this particular export.  This is important since
1831          * the v4 server may be holding file locks or vnodes under
1832          * this export.
1833          */
1834         rfs4_clean_state_exi(exi);
1835 
1836         /*
1837          * Notify the lock manager that the filesystem is being
1838          * unexported.
1839          */
1840         lm_unexport(exi);
1841 
1842         /*
1843          * If this was a public export, restore
1844          * the public filehandle to the root.
1845          */
1846         if (exi == ne->exi_public) {
1847                 ne->exi_public = ne->exi_root;
1848 
1849                 nfslog_share_record(ne->exi_public, CRED());
1850         }
1851 
1852         if (exi->exi_export.ex_flags & EX_LOG)
1853                 nfslog_unshare_record(exi, CRED());
1854 
1855         exi_rele(exi);
1856         return (0);
1857 }
1858 
1859 /*
1860  * Get file handle system call.
1861  * Takes file name and returns a file handle for it.
1862  * Credentials must be verified before calling.
1863  */
1864 int
1865 nfs_getfh(struct nfs_getfh_args *args, model_t model, cred_t *cr)
1866 {
1867         nfs_fh3 fh;
1868         char buf[NFS3_MAXFHSIZE];
1869         char *logptr, logbuf[NFS3_MAXFHSIZE];
1870         int l = NFS3_MAXFHSIZE;
1871         vnode_t *vp;
1872         vnode_t *dvp;
1873         struct exportinfo *exi;
1874         int error;
1875         int vers;
1876         STRUCT_HANDLE(nfs_getfh_args, uap);
1877 
1878 #ifdef lint
1879         model = model;          /* STRUCT macros don't always use it */
1880 #endif
1881 
1882         STRUCT_SET_HANDLE(uap, model, args);
1883 
1884         error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
1885             FOLLOW, &dvp, &vp);
1886         if (error == EINVAL) {
1887                 /*
1888                  * if fname resolves to / we get EINVAL error
1889                  * since we wanted the parent vnode. Try again
1890                  * with NULL dvp.
1891                  */
1892                 error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
1893                     FOLLOW, NULL, &vp);
1894                 dvp = NULL;
1895         }
1896         if (!error && vp == NULL) {
1897                 /*
1898                  * Last component of fname not found
1899                  */
1900                 if (dvp != NULL) {
1901                         VN_RELE(dvp);
1902                 }
1903                 error = ENOENT;
1904         }
1905         if (error)
1906                 return (error);
1907 
1908         /*
1909          * 'vp' may be an AUTOFS node, so we perform a
1910          * VOP_ACCESS() to trigger the mount of the
1911          * intended filesystem, so we can share the intended
1912          * filesystem instead of the AUTOFS filesystem.
1913          */
1914         (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
1915 
1916         /*
1917          * We're interested in the top most filesystem.
1918          * This is specially important when uap->dname is a trigger
1919          * AUTOFS node, since we're really interested in sharing the
1920          * filesystem AUTOFS mounted as result of the VOP_ACCESS()
1921          * call not the AUTOFS node itself.
1922          */
1923         if (vn_mountedvfs(vp) != NULL) {
1924                 if (error = traverse(&vp)) {
1925                         VN_RELE(vp);
1926                         if (dvp != NULL)
1927                                 VN_RELE(dvp);
1928                         return (error);
1929                 }
1930         }
1931 
1932         vers = STRUCT_FGET(uap, vers);
1933         exi = nfs_vptoexi(dvp, vp, cr, NULL, &error, FALSE);
1934         if (!error) {
1935                 if (vers == NFS_VERSION) {
1936                         error = makefh((fhandle_t *)buf, vp, exi);
1937                         l = NFS_FHSIZE;
1938                         logptr = buf;
1939                 } else if (vers == NFS_V3) {
1940                         int i, sz, pad;
1941 
1942                         error = makefh3(&fh, vp, exi);
1943                         l = RNDUP(fh.fh3_length);
1944                         if (!error && (l > sizeof (fhandle3_t)))
1945                                 error = EREMOTE;
1946                         logptr = logbuf;
1947                         if (!error) {
1948                                 i = 0;
1949                                 sz = sizeof (fsid_t);
1950                                 bcopy(&fh.fh3_fsid, &buf[i], sz);
1951                                 i += sz;
1952 
1953                                 /*
1954                                  * For backwards compatibility, the
1955                                  * fid length may be less than
1956                                  * NFS_FHMAXDATA, but it was always
1957                                  * encoded as NFS_FHMAXDATA bytes.
1958                                  */
1959 
1960                                 sz = sizeof (ushort_t);
1961                                 bcopy(&fh.fh3_len, &buf[i], sz);
1962                                 i += sz;
1963                                 bcopy(fh.fh3_data, &buf[i], fh.fh3_len);
1964                                 i += fh.fh3_len;
1965                                 pad = (NFS_FHMAXDATA - fh.fh3_len);
1966                                 if (pad > 0) {
1967                                         bzero(&buf[i], pad);
1968                                         i += pad;
1969                                         l += pad;
1970                                 }
1971 
1972                                 sz = sizeof (ushort_t);
1973                                 bcopy(&fh.fh3_xlen, &buf[i], sz);
1974                                 i += sz;
1975                                 bcopy(fh.fh3_xdata, &buf[i], fh.fh3_xlen);
1976                                 i += fh.fh3_xlen;
1977                                 pad = (NFS_FHMAXDATA - fh.fh3_xlen);
1978                                 if (pad > 0) {
1979                                         bzero(&buf[i], pad);
1980                                         i += pad;
1981                                         l += pad;
1982                                 }
1983                         }
1984                         /*
1985                          * If we need to do NFS logging, the filehandle
1986                          * must be downsized to 32 bytes.
1987                          */
1988                         if (!error && exi->exi_export.ex_flags & EX_LOG) {
1989                                 i = 0;
1990                                 sz = sizeof (fsid_t);
1991                                 bcopy(&fh.fh3_fsid, &logbuf[i], sz);
1992                                 i += sz;
1993                                 sz = sizeof (ushort_t);
1994                                 bcopy(&fh.fh3_len, &logbuf[i], sz);
1995                                 i += sz;
1996                                 sz = NFS_FHMAXDATA;
1997                                 bcopy(fh.fh3_data, &logbuf[i], sz);
1998                                 i += sz;
1999                                 sz = sizeof (ushort_t);
2000                                 bcopy(&fh.fh3_xlen, &logbuf[i], sz);
2001                                 i += sz;
2002                                 sz = NFS_FHMAXDATA;
2003                                 bcopy(fh.fh3_xdata, &logbuf[i], sz);
2004                                 i += sz;
2005                         }
2006                 }
2007                 if (!error && exi->exi_export.ex_flags & EX_LOG) {
2008                         nfslog_getfh(exi, (fhandle_t *)logptr,
2009                             STRUCT_FGETP(uap, fname), UIO_USERSPACE, cr);
2010                 }
2011                 exi_rele(exi);
2012                 if (!error) {
2013                         if (copyout(&l, STRUCT_FGETP(uap, lenp), sizeof (int)))
2014                                 error = EFAULT;
2015                         if (copyout(buf, STRUCT_FGETP(uap, fhp), l))
2016                                 error = EFAULT;
2017                 }
2018         }
2019         VN_RELE(vp);
2020         if (dvp != NULL) {
2021                 VN_RELE(dvp);
2022         }
2023         return (error);
2024 }
2025 
2026 /*
2027  * Strategy: if vp is in the export list, then
2028  * return the associated file handle. Otherwise, ".."
2029  * once up the vp and try again, until the root of the
2030  * filesystem is reached.
2031  */
2032 struct   exportinfo *
2033 nfs_vptoexi(vnode_t *dvp, vnode_t *vp, cred_t *cr, int *walk,
2034     int *err, bool_t v4srv)
2035 {
2036         fid_t fid;
2037         int error;
2038         struct exportinfo *exi;
2039 
2040         ASSERT(vp);
2041         VN_HOLD(vp);
2042         if (dvp != NULL) {
2043                 VN_HOLD(dvp);
2044         }
2045         if (walk != NULL)
2046                 *walk = 0;
2047 
2048         for (;;) {
2049                 bzero(&fid, sizeof (fid));
2050                 fid.fid_len = MAXFIDSZ;
2051                 error = vop_fid_pseudo(vp, &fid);
2052                 if (error) {
2053                         /*
2054                          * If vop_fid_pseudo returns ENOSPC then the fid
2055                          * supplied is too small. For now we simply
2056                          * return EREMOTE.
2057                          */
2058                         if (error == ENOSPC)
2059                                 error = EREMOTE;
2060                         break;
2061                 }
2062 
2063                 if (v4srv)
2064                         exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
2065                 else
2066                         exi = checkexport(&vp->v_vfsp->vfs_fsid, &fid);
2067 
2068                 if (exi != NULL) {
2069                         /*
2070                          * Found the export info
2071                          */
2072                         break;
2073                 }
2074 
2075                 /*
2076                  * We have just failed finding a matching export.
2077                  * If we're at the root of this filesystem, then
2078                  * it's time to stop (with failure).
2079                  */
2080                 if ((vp->v_flag & VROOT) || VN_IS_CURZONEROOT(vp)) {
2081                         error = EINVAL;
2082                         break;
2083                 }
2084 
2085                 if (walk != NULL)
2086                         (*walk)++;
2087 
2088                 /*
2089                  * Now, do a ".." up vp. If dvp is supplied, use it,
2090                  * otherwise, look it up.
2091                  */
2092                 if (dvp == NULL) {
2093                         error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, cr,
2094                             NULL, NULL, NULL);
2095                         if (error)
2096                                 break;
2097                 }
2098                 VN_RELE(vp);
2099                 vp = dvp;
2100                 dvp = NULL;
2101         }
2102         VN_RELE(vp);
2103         if (dvp != NULL) {
2104                 VN_RELE(dvp);
2105         }
2106         if (error != 0) {
2107                 if (err != NULL)
2108                         *err = error;
2109                 return (NULL);
2110         }
2111         return (exi);
2112 }
2113 
2114 int
2115 chk_clnt_sec(exportinfo_t *exi, struct svc_req *req)
2116 {
2117         int i, nfsflavor;
2118         struct secinfo *sp;
2119 
2120         /*
2121          *  Get the nfs flavor number from xprt.
2122          */
2123         nfsflavor = (int)(uintptr_t)req->rq_xprt->xp_cookie;
2124 
2125         sp = exi->exi_export.ex_secinfo;
2126         for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2127                 if ((nfsflavor == sp[i].s_secinfo.sc_nfsnum) &&
2128                     SEC_REF_EXPORTED(sp + i))
2129                         return (TRUE);
2130         }
2131         return (FALSE);
2132 }
2133 
2134 /*
2135  * Make an fhandle from a vnode
2136  */
2137 int
2138 makefh(fhandle_t *fh, vnode_t *vp, exportinfo_t *exi)
2139 {
2140         int error;
2141 
2142         *fh = exi->exi_fh;   /* struct copy */
2143 
2144         error = VOP_FID(vp, (fid_t *)&fh->fh_len, NULL);
2145         if (error) {
2146                 /*
2147                  * Should be something other than EREMOTE
2148                  */
2149                 return (EREMOTE);
2150         }
2151         return (0);
2152 }
2153 
2154 /*
2155  * This routine makes an overloaded V2 fhandle which contains
2156  * sec modes.
2157  *
2158  * Note that the first four octets contain the length octet,
2159  * the status octet, and two padded octets to make them XDR
2160  * four-octet aligned.
2161  *
2162  *   1   2   3   4                                          32
2163  * +---+---+---+---+---+---+---+---+   +---+---+---+---+   +---+
2164  * | l | s |   |   |     sec_1     |...|     sec_n     |...|   |
2165  * +---+---+---+---+---+---+---+---+   +---+---+---+---+   +---+
2166  *
2167  * where
2168  *
2169  *   the status octet s indicates whether there are more security
2170  *   flavors (1 means yes, 0 means no) that require the client to
2171  *   perform another 0x81 LOOKUP to get them,
2172  *
2173  *   the length octet l is the length describing the number of
2174  *   valid octets that follow.  (l = 4 * n, where n is the number
2175  *   of security flavors sent in the current overloaded filehandle.)
2176  *
2177  *   sec_index should always be in the inclusive range: [1 - ex_seccnt],
2178  *   and it tells server where to start within the secinfo array.
2179  *   Usually it will always be 1; however, if more flavors are used
2180  *   for the public export than can be encoded in the overloaded FH
2181  *   (7 for NFS2), subsequent SNEGO MCLs will have a larger index
2182  *   so the server will pick up where it left off from the previous
2183  *   MCL reply.
2184  *
2185  *   With NFS4 support, implicitly allowed flavors are also in
2186  *   the secinfo array; however, they should not be returned in
2187  *   SNEGO MCL replies.
2188  */
2189 int
2190 makefh_ol(fhandle_t *fh, exportinfo_t *exi, uint_t sec_index)
2191 {
2192         secinfo_t sec[MAX_FLAVORS];
2193         int totalcnt, i, *ipt, cnt, seccnt, secidx, fh_max_cnt;
2194         char *c;
2195 
2196         if (fh == NULL || exi == NULL || sec_index < 1)
2197                 return (EREMOTE);
2198 
2199         /*
2200          * WebNFS clients need to know the unique set of explicitly
2201          * shared flavors in used for the public export. When
2202          * "TRUE" is passed to build_seclist_nodups(), only explicitly
2203          * shared flavors are included in the list.
2204          */
2205         seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2206         if (sec_index > seccnt)
2207                 return (EREMOTE);
2208 
2209         fh_max_cnt = (NFS_FHSIZE / sizeof (int)) - 1;
2210         totalcnt = seccnt - sec_index + 1;
2211         cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2212 
2213         c = (char *)fh;
2214         /*
2215          * Encode the length octet representing the number of
2216          * security flavors (in bytes) in this overloaded fh.
2217          */
2218         *c = cnt * sizeof (int);
2219 
2220         /*
2221          * Encode the status octet that indicates whether there
2222          * are more security flavors the client needs to get.
2223          */
2224         *(c + 1) = totalcnt > fh_max_cnt;
2225 
2226         /*
2227          * put security flavors in the overloaded fh
2228          */
2229         ipt = (int *)(c + sizeof (int32_t));
2230         secidx = sec_index - 1;
2231         for (i = 0; i < cnt; i++) {
2232                 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2233         }
2234         return (0);
2235 }
2236 
2237 /*
2238  * Make an nfs_fh3 from a vnode
2239  */
2240 int
2241 makefh3(nfs_fh3 *fh, vnode_t *vp, struct exportinfo *exi)
2242 {
2243         int error;
2244         fid_t fid;
2245 
2246         bzero(&fid, sizeof (fid));
2247         fid.fid_len = sizeof (fh->fh3_data);
2248         error = VOP_FID(vp, &fid, NULL);
2249         if (error)
2250                 return (EREMOTE);
2251 
2252         bzero(fh, sizeof (nfs_fh3));
2253         fh->fh3_fsid = exi->exi_fsid;
2254         fh->fh3_len = fid.fid_len;
2255         bcopy(fid.fid_data, fh->fh3_data, fh->fh3_len);
2256 
2257         fh->fh3_xlen = exi->exi_fid.fid_len;
2258         ASSERT(fh->fh3_xlen <= sizeof (fh->fh3_xdata));
2259         bcopy(exi->exi_fid.fid_data, fh->fh3_xdata, fh->fh3_xlen);
2260 
2261         fh->fh3_length = sizeof (fh->fh3_fsid)
2262             + sizeof (fh->fh3_len) + fh->fh3_len
2263             + sizeof (fh->fh3_xlen) + fh->fh3_xlen;
2264         fh->fh3_flags = 0;
2265 
2266         return (0);
2267 }
2268 
2269 /*
2270  * This routine makes an overloaded V3 fhandle which contains
2271  * sec modes.
2272  *
2273  *  1        4
2274  * +--+--+--+--+
2275  * |    len    |
2276  * +--+--+--+--+
2277  *                                               up to 64
2278  * +--+--+--+--+--+--+--+--+--+--+--+--+     +--+--+--+--+
2279  * |s |  |  |  |   sec_1   |   sec_2   | ... |   sec_n   |
2280  * +--+--+--+--+--+--+--+--+--+--+--+--+     +--+--+--+--+
2281  *
2282  * len = 4 * (n+1), where n is the number of security flavors
2283  * sent in the current overloaded filehandle.
2284  *
2285  * the status octet s indicates whether there are more security
2286  * mechanisms (1 means yes, 0 means no) that require the client
2287  * to perform another 0x81 LOOKUP to get them.
2288  *
2289  * Three octets are padded after the status octet.
2290  */
2291 int
2292 makefh3_ol(nfs_fh3 *fh, struct exportinfo *exi, uint_t sec_index)
2293 {
2294         secinfo_t sec[MAX_FLAVORS];
2295         int totalcnt, cnt, *ipt, i, seccnt, fh_max_cnt, secidx;
2296         char *c;
2297 
2298         if (fh == NULL || exi == NULL || sec_index < 1)
2299                 return (EREMOTE);
2300 
2301         /*
2302          * WebNFS clients need to know the unique set of explicitly
2303          * shared flavors in used for the public export. When
2304          * "TRUE" is passed to build_seclist_nodups(), only explicitly
2305          * shared flavors are included in the list.
2306          */
2307         seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2308 
2309         if (sec_index > seccnt)
2310                 return (EREMOTE);
2311 
2312         fh_max_cnt = (NFS3_FHSIZE / sizeof (int)) - 1;
2313         totalcnt = seccnt - sec_index + 1;
2314         cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2315 
2316         /*
2317          * Place the length in fh3_length representing the number
2318          * of security flavors (in bytes) in this overloaded fh.
2319          */
2320         fh->fh3_flags = FH_WEBNFS;
2321         fh->fh3_length = (cnt+1) * sizeof (int32_t);
2322 
2323         c = (char *)&fh->fh3_u.nfs_fh3_i.fh3_i;
2324         /*
2325          * Encode the status octet that indicates whether there
2326          * are more security flavors the client needs to get.
2327          */
2328         *c = totalcnt > fh_max_cnt;
2329 
2330         /*
2331          * put security flavors in the overloaded fh
2332          */
2333         secidx = sec_index - 1;
2334         ipt = (int *)(c + sizeof (int32_t));
2335         for (i = 0; i < cnt; i++) {
2336                 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2337         }
2338         return (0);
2339 }
2340 
2341 /*
2342  * Make an nfs_fh4 from a vnode
2343  */
2344 int
2345 makefh4(nfs_fh4 *fh, vnode_t *vp, struct exportinfo *exi)
2346 {
2347         int error;
2348         nfs_fh4_fmt_t *fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2349         fid_t fid;
2350 
2351         bzero(&fid, sizeof (fid));
2352         fid.fid_len = MAXFIDSZ;
2353         /*
2354          * vop_fid_pseudo() is used to set up NFSv4 namespace, so
2355          * use vop_fid_pseudo() here to get the fid instead of VOP_FID.
2356          */
2357         error = vop_fid_pseudo(vp, &fid);
2358         if (error)
2359                 return (error);
2360 
2361         fh->nfs_fh4_len = NFS_FH4_LEN;
2362 
2363         fh_fmtp->fh4_i.fhx_fsid = exi->exi_fh.fh_fsid;
2364         fh_fmtp->fh4_i.fhx_xlen = exi->exi_fh.fh_xlen;
2365 
2366         bzero(fh_fmtp->fh4_i.fhx_data, sizeof (fh_fmtp->fh4_i.fhx_data));
2367         bzero(fh_fmtp->fh4_i.fhx_xdata, sizeof (fh_fmtp->fh4_i.fhx_xdata));
2368         ASSERT(exi->exi_fh.fh_xlen <= sizeof (fh_fmtp->fh4_i.fhx_xdata));
2369         bcopy(exi->exi_fh.fh_xdata, fh_fmtp->fh4_i.fhx_xdata,
2370             exi->exi_fh.fh_xlen);
2371 
2372         fh_fmtp->fh4_len = fid.fid_len;
2373         ASSERT(fid.fid_len <= sizeof (fh_fmtp->fh4_data));
2374         bcopy(fid.fid_data, fh_fmtp->fh4_data, fid.fid_len);
2375         fh_fmtp->fh4_flag = 0;
2376 
2377 #ifdef VOLATILE_FH_TEST
2378         /*
2379          * XXX (temporary?)
2380          * Use the rnode volatile_id value to add volatility to the fh.
2381          *
2382          * For testing purposes there are currently two scenarios, based
2383          * on whether the filesystem was shared with "volatile_fh"
2384          * or "expire_on_rename". In the first case, use the value of
2385          * export struct share_time as the volatile_id. In the second
2386          * case use the vnode volatile_id value (which is set to the
2387          * time in which the file was renamed).
2388          *
2389          * Note that the above are temporary constructs for testing only
2390          * XXX
2391          */
2392         if (exi->exi_export.ex_flags & EX_VOLRNM) {
2393                 fh_fmtp->fh4_volatile_id = find_volrnm_fh_id(exi, fh);
2394         } else if (exi->exi_export.ex_flags & EX_VOLFH) {
2395                 fh_fmtp->fh4_volatile_id = exi->exi_volatile_id;
2396         } else {
2397                 fh_fmtp->fh4_volatile_id = 0;
2398         }
2399 #endif /* VOLATILE_FH_TEST */
2400 
2401         return (0);
2402 }
2403 
2404 /*
2405  * Convert an fhandle into a vnode.
2406  * Uses the file id (fh_len + fh_data) in the fhandle to get the vnode.
2407  * WARNING: users of this routine must do a VN_RELE on the vnode when they
2408  * are done with it.
2409  */
2410 vnode_t *
2411 nfs_fhtovp(fhandle_t *fh, struct exportinfo *exi)
2412 {
2413         vfs_t *vfsp;
2414         vnode_t *vp;
2415         int error;
2416         fid_t *fidp;
2417 
2418         TRACE_0(TR_FAC_NFS, TR_FHTOVP_START,
2419             "fhtovp_start");
2420 
2421         if (exi == NULL) {
2422                 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2423                     "fhtovp_end:(%S)", "exi NULL");
2424                 return (NULL);  /* not exported */
2425         }
2426 
2427         ASSERT(exi->exi_vp != NULL);
2428 
2429         if (PUBLIC_FH2(fh)) {
2430                 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2431                         TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2432                             "fhtovp_end:(%S)", "root not exported");
2433                         return (NULL);
2434                 }
2435                 vp = exi->exi_vp;
2436                 VN_HOLD(vp);
2437                 return (vp);
2438         }
2439 
2440         vfsp = exi->exi_vp->v_vfsp;
2441         ASSERT(vfsp != NULL);
2442         fidp = (fid_t *)&fh->fh_len;
2443 
2444         error = VFS_VGET(vfsp, &vp, fidp);
2445         if (error || vp == NULL) {
2446                 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2447                     "fhtovp_end:(%S)", "VFS_GET failed or vp NULL");
2448                 return (NULL);
2449         }
2450         TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2451             "fhtovp_end:(%S)", "end");
2452         return (vp);
2453 }
2454 
2455 /*
2456  * Convert an nfs_fh3 into a vnode.
2457  * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2458  * WARNING: users of this routine must do a VN_RELE on the vnode when they
2459  * are done with it.
2460  */
2461 vnode_t *
2462 nfs3_fhtovp(nfs_fh3 *fh, struct exportinfo *exi)
2463 {
2464         vfs_t *vfsp;
2465         vnode_t *vp;
2466         int error;
2467         fid_t *fidp;
2468 
2469         if (exi == NULL)
2470                 return (NULL);  /* not exported */
2471 
2472         ASSERT(exi->exi_vp != NULL);
2473 
2474         if (PUBLIC_FH3(fh)) {
2475                 if (exi->exi_export.ex_flags & EX_PUBLIC)
2476                         return (NULL);
2477                 vp = exi->exi_vp;
2478                 VN_HOLD(vp);
2479                 return (vp);
2480         }
2481 
2482         if (fh->fh3_length < NFS3_OLDFHSIZE ||
2483             fh->fh3_length > NFS3_MAXFHSIZE)
2484                 return (NULL);
2485 
2486         vfsp = exi->exi_vp->v_vfsp;
2487         ASSERT(vfsp != NULL);
2488         fidp = FH3TOFIDP(fh);
2489 
2490         error = VFS_VGET(vfsp, &vp, fidp);
2491         if (error || vp == NULL)
2492                 return (NULL);
2493 
2494         return (vp);
2495 }
2496 
2497 /*
2498  * Convert an nfs_fh4 into a vnode.
2499  * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2500  * WARNING: users of this routine must do a VN_RELE on the vnode when they
2501  * are done with it.
2502  */
2503 vnode_t *
2504 nfs4_fhtovp(nfs_fh4 *fh, struct exportinfo *exi, nfsstat4 *statp)
2505 {
2506         vfs_t *vfsp;
2507         vnode_t *vp = NULL;
2508         int error;
2509         fid_t *fidp;
2510         nfs_fh4_fmt_t *fh_fmtp;
2511 #ifdef VOLATILE_FH_TEST
2512         uint32_t volatile_id = 0;
2513 #endif /* VOLATILE_FH_TEST */
2514 
2515         if (exi == NULL) {
2516                 *statp = NFS4ERR_STALE;
2517                 return (NULL);  /* not exported */
2518         }
2519         ASSERT(exi->exi_vp != NULL);
2520 
2521         /* caller should have checked this */
2522         ASSERT(fh->nfs_fh4_len >= NFS_FH4_LEN);
2523 
2524         fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2525         vfsp = exi->exi_vp->v_vfsp;
2526         ASSERT(vfsp != NULL);
2527         fidp = (fid_t *)&fh_fmtp->fh4_len;
2528 
2529 #ifdef VOLATILE_FH_TEST
2530         /* XXX check if volatile - should be changed later */
2531         if (exi->exi_export.ex_flags & (EX_VOLRNM | EX_VOLFH)) {
2532                 /*
2533                  * Filesystem is shared with volatile filehandles
2534                  */
2535                 if (exi->exi_export.ex_flags & EX_VOLRNM)
2536                         volatile_id = find_volrnm_fh_id(exi, fh);
2537                 else
2538                         volatile_id = exi->exi_volatile_id;
2539 
2540                 if (fh_fmtp->fh4_volatile_id != volatile_id) {
2541                         *statp = NFS4ERR_FHEXPIRED;
2542                         return (NULL);
2543                 }
2544         }
2545         /*
2546          * XXX even if test_volatile_fh false, the fh may contain a
2547          * volatile id if obtained when the test was set.
2548          */
2549         fh_fmtp->fh4_volatile_id = (uchar_t)0;
2550 #endif /* VOLATILE_FH_TEST */
2551 
2552         error = VFS_VGET(vfsp, &vp, fidp);
2553         /*
2554          * If we can not get vp from VFS_VGET, perhaps this is
2555          * an nfs v2/v3/v4 node in an nfsv4 pseudo filesystem.
2556          * Check it out.
2557          */
2558         if (error && PSEUDO(exi))
2559                 error = nfs4_vget_pseudo(exi, &vp, fidp);
2560 
2561         if (error || vp == NULL) {
2562                 *statp = NFS4ERR_STALE;
2563                 return (NULL);
2564         }
2565         /* XXX - disgusting hack */
2566         if (vp->v_type == VNON && vp->v_flag & V_XATTRDIR)
2567                 vp->v_type = VDIR;
2568         *statp = NFS4_OK;
2569         return (vp);
2570 }
2571 
2572 /*
2573  * Find the export structure associated with the given filesystem.
2574  * If found, then increment the ref count (exi_count).
2575  */
2576 struct exportinfo *
2577 checkexport(fsid_t *fsid, fid_t *fid)
2578 {
2579         struct exportinfo *exi;
2580         nfs_export_t *ne = nfs_get_export();
2581 
2582         rw_enter(&ne->exported_lock, RW_READER);
2583         for (exi = ne->exptable[exptablehash(fsid, fid)];
2584             exi != NULL;
2585             exi = exi->fid_hash.next) {
2586                 if (exportmatch(exi, fsid, fid)) {
2587                         /*
2588                          * If this is the place holder for the
2589                          * public file handle, then return the
2590                          * real export entry for the public file
2591                          * handle.
2592                          */
2593                         if (exi->exi_export.ex_flags & EX_PUBLIC) {
2594                                 exi = ne->exi_public;
2595                         }
2596 
2597                         exi_hold(exi);
2598                         rw_exit(&ne->exported_lock);
2599                         return (exi);
2600                 }
2601         }
2602         rw_exit(&ne->exported_lock);
2603         return (NULL);
2604 }
2605 
2606 
2607 /*
2608  * "old school" version of checkexport() for NFS4.  NFS4
2609  * rfs4_compound holds exported_lock for duration of compound
2610  * processing.  This version doesn't manipulate exi_count
2611  * since NFS4 breaks fundamental assumptions in the exi_count
2612  * design.
2613  */
2614 struct exportinfo *
2615 checkexport4(fsid_t *fsid, fid_t *fid, vnode_t *vp)
2616 {
2617         struct exportinfo *exi;
2618         nfs_export_t *ne = nfs_get_export();
2619 
2620         ASSERT(RW_LOCK_HELD(&ne->exported_lock));
2621 
2622         for (exi = ne->exptable[exptablehash(fsid, fid)];
2623             exi != NULL;
2624             exi = exi->fid_hash.next) {
2625                 if (exportmatch(exi, fsid, fid)) {
2626                         /*
2627                          * If this is the place holder for the
2628                          * public file handle, then return the
2629                          * real export entry for the public file
2630                          * handle.
2631                          */
2632                         if (exi->exi_export.ex_flags & EX_PUBLIC) {
2633                                 exi = ne->exi_public;
2634                         }
2635 
2636                         /*
2637                          * If vp is given, check if vp is the
2638                          * same vnode as the exported node.
2639                          *
2640                          * Since VOP_FID of a lofs node returns the
2641                          * fid of its real node (ufs), the exported
2642                          * node for lofs and (pseudo) ufs may have
2643                          * the same fsid and fid.
2644                          */
2645                         if (vp == NULL || vp == exi->exi_vp)
2646                                 return (exi);
2647                 }
2648         }
2649 
2650         return (NULL);
2651 }
2652 
2653 /*
2654  * Free an entire export list node
2655  */
2656 void
2657 exportfree(struct exportinfo *exi)
2658 {
2659         struct exportdata *ex;
2660         struct charset_cache *cache;
2661         int i;
2662 
2663         ex = &exi->exi_export;
2664 
2665         ASSERT(exi->exi_vp != NULL && !(exi->exi_export.ex_flags & EX_PUBLIC));
2666         VN_RELE(exi->exi_vp);
2667         if (exi->exi_dvp != NULL)
2668                 VN_RELE(exi->exi_dvp);
2669 
2670         if (ex->ex_flags & EX_INDEX)
2671                 kmem_free(ex->ex_index, strlen(ex->ex_index) + 1);
2672 
2673         kmem_free(ex->ex_path, ex->ex_pathlen + 1);
2674         nfsauth_cache_free(exi);
2675 
2676         /*
2677          * if there is a character set mapping cached, clean it up.
2678          */
2679         for (cache = exi->exi_charset; cache != NULL;
2680             cache = exi->exi_charset) {
2681                 if (cache->inbound != (kiconv_t)-1)
2682                         (void) kiconv_close(cache->inbound);
2683                 if (cache->outbound != (kiconv_t)-1)
2684                         (void) kiconv_close(cache->outbound);
2685                 exi->exi_charset = cache->next;
2686                 kmem_free(cache, sizeof (struct charset_cache));
2687         }
2688 
2689         if (exi->exi_logbuffer != NULL)
2690                 nfslog_disable(exi);
2691 
2692         if (ex->ex_flags & EX_LOG) {
2693                 kmem_free(ex->ex_log_buffer, ex->ex_log_bufferlen + 1);
2694                 kmem_free(ex->ex_tag, ex->ex_taglen + 1);
2695         }
2696 
2697         if (exi->exi_visible)
2698                 free_visible(exi->exi_visible);
2699 
2700         srv_secinfo_list_free(ex->ex_secinfo, ex->ex_seccnt);
2701 
2702 #ifdef VOLATILE_FH_TEST
2703         free_volrnm_list(exi);
2704         mutex_destroy(&exi->exi_vol_rename_lock);
2705 #endif /* VOLATILE_FH_TEST */
2706 
2707         mutex_destroy(&exi->exi_lock);
2708         rw_destroy(&exi->exi_cache_lock);
2709         /*
2710          * All nodes in the exi_cache AVL trees were removed and freed in the
2711          * nfsauth_cache_free() call above.  We will just destroy and free the
2712          * empty AVL trees here.
2713          */
2714         for (i = 0; i < AUTH_TABLESIZE; i++) {
2715                 avl_destroy(exi->exi_cache[i]);
2716                 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
2717         }
2718 
2719         kmem_free(exi, sizeof (*exi));
2720 }
2721 
2722 /*
2723  * load the index file from user space into kernel space.
2724  */
2725 static int
2726 loadindex(struct exportdata *kex)
2727 {
2728         int error;
2729         char index[MAXNAMELEN+1];
2730         size_t len;
2731 
2732         /*
2733          * copyinstr copies the complete string including the NULL and
2734          * returns the len with the NULL byte included in the calculation
2735          * as long as the max length is not exceeded.
2736          */
2737         if (error = copyinstr(kex->ex_index, index, sizeof (index), &len))
2738                 return (error);
2739 
2740         kex->ex_index = kmem_alloc(len, KM_SLEEP);
2741         bcopy(index, kex->ex_index, len);
2742 
2743         return (0);
2744 }
2745 
2746 void
2747 exi_hold(struct exportinfo *exi)
2748 {
2749         mutex_enter(&exi->exi_lock);
2750         exi->exi_count++;
2751         mutex_exit(&exi->exi_lock);
2752 }
2753 
2754 /*
2755  * When a thread completes using exi, it should call exi_rele().
2756  * exi_rele() decrements exi_count. It releases exi if exi_count == 0, i.e.
2757  * if this is the last user of exi and exi is not on exportinfo list anymore
2758  */
2759 void
2760 exi_rele(struct exportinfo *exi)
2761 {
2762         mutex_enter(&exi->exi_lock);
2763         exi->exi_count--;
2764         if (exi->exi_count == 0) {
2765                 mutex_exit(&exi->exi_lock);
2766                 exportfree(exi);
2767         } else
2768                 mutex_exit(&exi->exi_lock);
2769 }
2770 
2771 #ifdef VOLATILE_FH_TEST
2772 /*
2773  * Test for volatile fh's - add file handle to list and set its volatile id
2774  * to time it was renamed. If EX_VOLFH is also on and the fs is reshared,
2775  * the vol_rename queue is purged.
2776  *
2777  * XXX This code is for unit testing purposes only... To correctly use it, it
2778  * needs to tie a rename list to the export struct and (more
2779  * important), protect access to the exi rename list using a write lock.
2780  */
2781 
2782 /*
2783  * get the fh vol record if it's in the volatile on rename list. Don't check
2784  * volatile_id in the file handle - compare only the file handles.
2785  */
2786 static struct ex_vol_rename *
2787 find_volrnm_fh(struct exportinfo *exi, nfs_fh4 *fh4p)
2788 {
2789         struct ex_vol_rename *p = NULL;
2790         fhandle4_t *fhp;
2791 
2792         /* XXX shouldn't we assert &exported_lock held? */
2793         ASSERT(MUTEX_HELD(&exi->exi_vol_rename_lock));
2794 
2795         if (fh4p->nfs_fh4_len != NFS_FH4_LEN) {
2796                 return (NULL);
2797         }
2798         fhp = &((nfs_fh4_fmt_t *)fh4p->nfs_fh4_val)->fh4_i;
2799         for (p = exi->exi_vol_rename; p != NULL; p = p->vrn_next) {
2800                 if (bcmp(fhp, &p->vrn_fh_fmt.fh4_i,
2801                     sizeof (fhandle4_t)) == 0)
2802                         break;
2803         }
2804         return (p);
2805 }
2806 
2807 /*
2808  * get the volatile id for the fh (if there is - else return 0). Ignore the
2809  * volatile_id in the file handle - compare only the file handles.
2810  */
2811 static uint32_t
2812 find_volrnm_fh_id(struct exportinfo *exi, nfs_fh4 *fh4p)
2813 {
2814         struct ex_vol_rename *p;
2815         uint32_t volatile_id;
2816 
2817         mutex_enter(&exi->exi_vol_rename_lock);
2818         p = find_volrnm_fh(exi, fh4p);
2819         volatile_id = (p ? p->vrn_fh_fmt.fh4_volatile_id :
2820             exi->exi_volatile_id);
2821         mutex_exit(&exi->exi_vol_rename_lock);
2822         return (volatile_id);
2823 }
2824 
2825 /*
2826  * Free the volatile on rename list - will be called if a filesystem is
2827  * unshared or reshared without EX_VOLRNM
2828  */
2829 static void
2830 free_volrnm_list(struct exportinfo *exi)
2831 {
2832         struct ex_vol_rename *p, *pnext;
2833 
2834         /* no need to hold mutex lock - this one is called from exportfree */
2835         for (p = exi->exi_vol_rename; p != NULL; p = pnext) {
2836                 pnext = p->vrn_next;
2837                 kmem_free(p, sizeof (*p));
2838         }
2839         exi->exi_vol_rename = NULL;
2840 }
2841 
2842 /*
2843  * Add a file handle to the volatile on rename list.
2844  */
2845 void
2846 add_volrnm_fh(struct exportinfo *exi, vnode_t *vp)
2847 {
2848         struct ex_vol_rename *p;
2849         char fhbuf[NFS4_FHSIZE];
2850         nfs_fh4 fh4;
2851         int error;
2852 
2853         fh4.nfs_fh4_val = fhbuf;
2854         error = makefh4(&fh4, vp, exi);
2855         if ((error) || (fh4.nfs_fh4_len != sizeof (p->vrn_fh_fmt))) {
2856                 return;
2857         }
2858 
2859         mutex_enter(&exi->exi_vol_rename_lock);
2860 
2861         p = find_volrnm_fh(exi, &fh4);
2862 
2863         if (p == NULL) {
2864                 p = kmem_alloc(sizeof (*p), KM_SLEEP);
2865                 bcopy(fh4.nfs_fh4_val, &p->vrn_fh_fmt, sizeof (p->vrn_fh_fmt));
2866                 p->vrn_next = exi->exi_vol_rename;
2867                 exi->exi_vol_rename = p;
2868         }
2869 
2870         p->vrn_fh_fmt.fh4_volatile_id = gethrestime_sec();
2871         mutex_exit(&exi->exi_vol_rename_lock);
2872 }
2873 
2874 #endif /* VOLATILE_FH_TEST */