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             tnode->tree_vis->vis_seccnt > 0) {
 734                 srv_secinfo_add(&exip->exi_export.ex_secinfo,
 735                     &exip->exi_export.ex_seccnt, tnode->tree_vis->vis_secinfo,
 736                     tnode->tree_vis->vis_seccnt, FALSE);
 737         }
 738 
 739         /*
 740          * Move to parent node and propagate sec flavor
 741          * to exportinfo and to visible structures.
 742          */
 743         tnode = tnode->tree_parent;
 744 
 745         while (tnode != NULL) {
 746 
 747                 /* If there is exportinfo, update it */
 748                 if (tnode->tree_exi != NULL) {
 749                         secinfo_t **pxsec =
 750                             &tnode->tree_exi->exi_export.ex_secinfo;
 751                         int *pxcnt = &tnode->tree_exi->exi_export.ex_seccnt;
 752                         int is_pseudo = PSEUDO(tnode->tree_exi);
 753                         if (isadd)
 754                                 srv_secinfo_add(pxsec, pxcnt, sec, seccnt,
 755                                     is_pseudo);
 756                         else
 757                                 srv_secinfo_remove(pxsec, pxcnt, sec, seccnt);
 758                 }
 759 
 760                 /* Update every visible - only root node has no visible */
 761                 if (tnode->tree_vis != NULL) {
 762                         secinfo_t **pxsec = &tnode->tree_vis->vis_secinfo;
 763                         int *pxcnt = &tnode->tree_vis->vis_seccnt;
 764                         if (isadd)
 765                                 srv_secinfo_add(pxsec, pxcnt, sec, seccnt,
 766                                     FALSE);
 767                         else
 768                                 srv_secinfo_remove(pxsec, pxcnt, sec, seccnt);
 769                 }
 770                 tnode = tnode->tree_parent;
 771         }
 772 }
 773 
 774 /* hash_name is a text substitution for either fid_hash or path_hash */
 775 #define exp_hash_unlink(exi, hash_name) \
 776         if (*(exi)->hash_name.bckt == (exi)) \
 777                 *(exi)->hash_name.bckt = (exi)->hash_name.next; \
 778         if ((exi)->hash_name.prev) \
 779                 (exi)->hash_name.prev->hash_name.next = (exi)->hash_name.next; \
 780         if ((exi)->hash_name.next) \
 781                 (exi)->hash_name.next->hash_name.prev = (exi)->hash_name.prev; \
 782         (exi)->hash_name.bckt = NULL;
 783 
 784 #define exp_hash_link(exi, hash_name, bucket) \
 785         (exi)->hash_name.bckt = (bucket); \
 786         (exi)->hash_name.prev = NULL; \
 787         (exi)->hash_name.next = *(bucket); \
 788         if ((exi)->hash_name.next) \
 789                 (exi)->hash_name.next->hash_name.prev = (exi); \
 790         *(bucket) = (exi);
 791 
 792 void
 793 export_link(nfs_export_t *ne, exportinfo_t *exi)
 794 {
 795         exportinfo_t **bckt;
 796 
 797         ASSERT(RW_WRITE_HELD(&ne->exported_lock));
 798 
 799         bckt = &ne->exptable[exptablehash(&exi->exi_fsid, &exi->exi_fid)];
 800         exp_hash_link(exi, fid_hash, bckt);
 801 
 802         bckt = &ne->exptable_path_hash[pkp_tab_hash(exi->exi_export.ex_path,
 803             strlen(exi->exi_export.ex_path))];
 804         exp_hash_link(exi, path_hash, bckt);
 805 }
 806 
 807 /*
 808  * Helper functions for exi_id handling
 809  */
 810 static int
 811 exi_id_compar(const void *v1, const void *v2)
 812 {
 813         const struct exportinfo *e1 = v1;
 814         const struct exportinfo *e2 = v2;
 815 
 816         if (e1->exi_id < e2->exi_id)
 817                 return (-1);
 818         if (e1->exi_id > e2->exi_id)
 819                 return (1);
 820 
 821         return (0);
 822 }
 823 
 824 int
 825 exi_id_get_next()
 826 {
 827         struct exportinfo e;
 828         int ret = exi_id_next;
 829 
 830         ASSERT(MUTEX_HELD(&nfs_exi_id_lock));
 831 
 832         do {
 833                 exi_id_next++;
 834                 if (exi_id_next == 0)
 835                         exi_id_overflow = TRUE;
 836 
 837                 if (!exi_id_overflow)
 838                         break;
 839 
 840                 if (exi_id_next == ret)
 841                         cmn_err(CE_PANIC, "exi_id exhausted");
 842 
 843                 e.exi_id = exi_id_next;
 844         } while (avl_find(&exi_id_tree, &e, NULL) != NULL);
 845 
 846         return (ret);
 847 }
 848 
 849 /*ARGSUSED*/
 850 static void *
 851 nfs_export_zone_init(zoneid_t zoneid)
 852 {
 853         int i;
 854         nfs_export_t *ne;
 855 
 856         ne = kmem_zalloc(sizeof (*ne), KM_SLEEP);
 857 
 858         rw_init(&ne->exported_lock, NULL, RW_DEFAULT, NULL);
 859 
 860         /*
 861          * Allocate the place holder for the public file handle, which
 862          * is all zeroes. It is initially set to the root filesystem.
 863          */
 864         ne->exi_root = kmem_zalloc(sizeof (*ne->exi_root), KM_SLEEP);
 865         ne->exi_public = ne->exi_root;
 866 
 867         ne->exi_root->exi_export.ex_flags = EX_PUBLIC;
 868         ne->exi_root->exi_export.ex_pathlen = 1;  /* length of "/" */
 869         ne->exi_root->exi_export.ex_path =
 870             kmem_alloc(ne->exi_root->exi_export.ex_pathlen + 1, KM_SLEEP);
 871         ne->exi_root->exi_export.ex_path[0] = '/';
 872         ne->exi_root->exi_export.ex_path[1] = '\0';
 873 
 874         ne->exi_root->exi_count = 1;
 875         mutex_init(&ne->exi_root->exi_lock, NULL, MUTEX_DEFAULT, NULL);
 876 
 877         ne->exi_root->exi_vp = ZONE_ROOTVP();
 878         ne->exi_rootfid.fid_len = MAXFIDSZ;
 879         if (vop_fid_pseudo(ne->exi_root->exi_vp, &ne->exi_rootfid) != 0) {
 880                 mutex_destroy(&ne->exi_root->exi_lock);
 881                 kmem_free(ne->exi_root->exi_export.ex_path,
 882                     ne->exi_root->exi_export.ex_pathlen + 1);
 883                 kmem_free(ne->exi_root, sizeof (*ne->exi_root));
 884                 return (NULL);
 885         }
 886 
 887         /* Initialize auth cache and auth cache lock */
 888         for (i = 0; i < AUTH_TABLESIZE; i++) {
 889                 ne->exi_root->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t),
 890                     KM_SLEEP);
 891                 avl_create(ne->exi_root->exi_cache[i],
 892                     nfsauth_cache_clnt_compar, sizeof (struct auth_cache_clnt),
 893                     offsetof(struct auth_cache_clnt, authc_link));
 894         }
 895         rw_init(&ne->exi_root->exi_cache_lock, NULL, RW_DEFAULT, NULL);
 896 
 897         /* Setup the fhandle template */
 898         ne->exi_root->exi_fh.fh_fsid = rootdir->v_vfsp->vfs_fsid;
 899         ne->exi_root->exi_fh.fh_xlen = ne->exi_rootfid.fid_len;
 900         bcopy(ne->exi_rootfid.fid_data, ne->exi_root->exi_fh.fh_xdata,
 901             ne->exi_rootfid.fid_len);
 902         ne->exi_root->exi_fh.fh_len = sizeof (ne->exi_root->exi_fh.fh_data);
 903 
 904         rw_enter(&ne->exported_lock, RW_WRITER);
 905 
 906         /* Publish the exportinfo in the hash table */
 907         export_link(ne, ne->exi_root);
 908 
 909         /* Initialize exi_id and exi_kstats */
 910         mutex_enter(&nfs_exi_id_lock);
 911         ne->exi_root->exi_id = exi_id_get_next();
 912         avl_add(&exi_id_tree, ne->exi_root);
 913         mutex_exit(&nfs_exi_id_lock);
 914         ne->exi_root->exi_kstats = exp_kstats_init(zoneid,
 915             ne->exi_root->exi_id, ne->exi_root->exi_export.ex_path,
 916             ne->exi_root->exi_export.ex_pathlen, FALSE);
 917 
 918         rw_exit(&ne->exported_lock);
 919         ne->ns_root = NULL;
 920 
 921         return (ne);
 922 }
 923 
 924 /*ARGSUSED*/
 925 static void
 926 nfs_export_zone_fini(zoneid_t zoneid, void *data)
 927 {
 928         int i;
 929         nfs_export_t *ne = data;
 930         struct exportinfo *exi;
 931 
 932         rw_enter(&ne->exported_lock, RW_WRITER);
 933         mutex_enter(&nfs_exi_id_lock);
 934 
 935         exp_kstats_delete(ne->exi_root->exi_kstats);
 936         avl_remove(&exi_id_tree, ne->exi_root);
 937         export_unlink(ne, ne->exi_root);
 938 
 939         mutex_exit(&nfs_exi_id_lock);
 940         rw_exit(&ne->exported_lock);
 941 
 942         /* Deallocate the place holder for the public file handle */
 943         srv_secinfo_list_free(ne->exi_root->exi_export.ex_secinfo,
 944             ne->exi_root->exi_export.ex_seccnt);
 945         mutex_destroy(&ne->exi_root->exi_lock);
 946 
 947         rw_destroy(&ne->exi_root->exi_cache_lock);
 948         for (i = 0; i < AUTH_TABLESIZE; i++) {
 949                 avl_destroy(ne->exi_root->exi_cache[i]);
 950                 kmem_free(ne->exi_root->exi_cache[i], sizeof (avl_tree_t));
 951         }
 952 
 953         exp_kstats_fini(ne->exi_root->exi_kstats);
 954         kmem_free(ne->exi_root->exi_export.ex_path,
 955             ne->exi_root->exi_export.ex_pathlen + 1);
 956         kmem_free(ne->exi_root, sizeof (*ne->exi_root));
 957 
 958         exi = avl_first(&exi_id_tree);
 959         while (exi != NULL) {
 960                 struct exportinfo *nexi = AVL_NEXT(&exi_id_tree, exi);
 961                 if (zoneid == exi->exi_zoneid)
 962                         (void) unexport(ne, exi);
 963                 exi = nexi;
 964         }
 965 
 966         rw_destroy(&ne->exported_lock);
 967         kmem_free(ne, sizeof (*ne));
 968 }
 969 
 970 /*
 971  * Initialization routine for export routines.
 972  * Should only be called once.
 973  */
 974 void
 975 nfs_exportinit(void)
 976 {
 977         mutex_init(&nfs_exi_id_lock, NULL, MUTEX_DEFAULT, NULL);
 978 
 979         /* exi_id handling initialization */
 980         exi_id_next = 0;
 981         exi_id_overflow = FALSE;
 982         avl_create(&exi_id_tree, exi_id_compar, sizeof (struct exportinfo),
 983             offsetof(struct exportinfo, exi_id_link));
 984 
 985         zone_key_create(&nfs_export_key, nfs_export_zone_init,
 986             NULL, nfs_export_zone_fini);
 987 
 988         nfslog_init();
 989 }
 990 
 991 /*
 992  * Finalization routine for export routines.
 993  */
 994 void
 995 nfs_exportfini(void)
 996 {
 997         (void) zone_key_delete(nfs_export_key);
 998         avl_destroy(&exi_id_tree);
 999         mutex_destroy(&nfs_exi_id_lock);
1000 }
1001 
1002 /*
1003  *  Check if 2 gss mechanism identifiers are the same.
1004  *
1005  *  return FALSE if not the same.
1006  *  return TRUE if the same.
1007  */
1008 static bool_t
1009 nfs_mech_equal(rpc_gss_OID mech1, rpc_gss_OID mech2)
1010 {
1011         if ((mech1->length == 0) && (mech2->length == 0))
1012                 return (TRUE);
1013 
1014         if (mech1->length != mech2->length)
1015                 return (FALSE);
1016 
1017         return (bcmp(mech1->elements, mech2->elements, mech1->length) == 0);
1018 }
1019 
1020 /*
1021  *  This routine is used by rpc to map rpc security number
1022  *  to nfs specific security flavor number.
1023  *
1024  *  The gss callback prototype is
1025  *  callback(struct svc_req *, gss_cred_id_t *, gss_ctx_id_t *,
1026  *                              rpc_gss_lock_t *, void **),
1027  *  since nfs does not use the gss_cred_id_t/gss_ctx_id_t arguments
1028  *  we cast them to void.
1029  */
1030 /*ARGSUSED*/
1031 bool_t
1032 rfs_gsscallback(struct svc_req *req, gss_cred_id_t deleg, void *gss_context,
1033     rpc_gss_lock_t *lock, void **cookie)
1034 {
1035         int i, j;
1036         rpc_gss_rawcred_t *raw_cred;
1037         struct exportinfo *exi;
1038         nfs_export_t *ne = nfs_get_export();
1039 
1040         /*
1041          * We don't deal with delegated credentials.
1042          */
1043         if (deleg != GSS_C_NO_CREDENTIAL)
1044                 return (FALSE);
1045 
1046         raw_cred = lock->raw_cred;
1047         *cookie = NULL;
1048 
1049         rw_enter(&ne->exported_lock, RW_READER);
1050 
1051         for (i = 0; i < EXPTABLESIZE; i++) {
1052                 exi = ne->exptable[i];
1053                 while (exi) {
1054                         if (exi->exi_export.ex_seccnt > 0) {
1055                                 struct secinfo *secp;
1056                                 seconfig_t *se;
1057                                 int seccnt;
1058 
1059                                 secp = exi->exi_export.ex_secinfo;
1060                                 seccnt = exi->exi_export.ex_seccnt;
1061                                 for (j = 0; j < seccnt; j++) {
1062                                         /*
1063                                          *  If there is a map of the triplet
1064                                          *  (mechanism, service, qop) between
1065                                          *  raw_cred and the exported flavor,
1066                                          *  get the psudo flavor number.
1067                                          *  Also qop should not be NULL, it
1068                                          *  should be "default" or something
1069                                          *  else.
1070                                          */
1071                                         se = &secp[j].s_secinfo;
1072                                         if ((se->sc_rpcnum == RPCSEC_GSS) &&
1073 
1074                                             (nfs_mech_equal(
1075                                             se->sc_gss_mech_type,
1076                                             raw_cred->mechanism)) &&
1077 
1078                                             (se->sc_service ==
1079                                             raw_cred->service) &&
1080                                             (raw_cred->qop == se->sc_qop)) {
1081 
1082                                                 *cookie = (void *)(uintptr_t)
1083                                                     se->sc_nfsnum;
1084                                                 goto done;
1085                                         }
1086                                 }
1087                         }
1088                         exi = exi->fid_hash.next;
1089                 }
1090         }
1091 done:
1092         rw_exit(&ne->exported_lock);
1093 
1094         /*
1095          * If no nfs pseudo number mapping can be found in the export
1096          * table, assign the nfsflavor to NFS_FLAVOR_NOMAP. In V4, we may
1097          * recover the flavor mismatch from NFS layer (NFS4ERR_WRONGSEC).
1098          *
1099          * For example:
1100          *      server first shares with krb5i;
1101          *      client mounts with krb5i;
1102          *      server re-shares with krb5p;
1103          *      client tries with krb5i, but no mapping can be found;
1104          *      rpcsec_gss module calls this routine to do the mapping,
1105          *              if this routine fails, request is rejected from
1106          *              the rpc layer.
1107          *      What we need is to let the nfs layer rejects the request.
1108          *      For V4, we can reject with NFS4ERR_WRONGSEC and the client
1109          *      may recover from it by getting the new flavor via SECINFO.
1110          *
1111          * nfs pseudo number for RPCSEC_GSS mapping (see nfssec.conf)
1112          * is owned by IANA (see RFC 2623).
1113          *
1114          * XXX NFS_FLAVOR_NOMAP is defined in Solaris to work around
1115          * the implementation issue. This number should not overlap with
1116          * any new IANA defined pseudo flavor numbers.
1117          */
1118         if (*cookie == NULL)
1119                 *cookie = (void *)NFS_FLAVOR_NOMAP;
1120 
1121         lock->locked = TRUE;
1122 
1123         return (TRUE);
1124 }
1125 
1126 
1127 /*
1128  * Exportfs system call; credentials should be checked before
1129  * calling this function.
1130  */
1131 int
1132 exportfs(struct exportfs_args *args, model_t model, cred_t *cr)
1133 {
1134         vnode_t *vp;
1135         vnode_t *dvp;
1136         struct exportdata *kex;
1137         struct exportinfo *exi = NULL;
1138         struct exportinfo *ex, *ex1, *ex2;
1139         fid_t fid;
1140         fsid_t fsid;
1141         int error;
1142         size_t allocsize;
1143         struct secinfo *sp;
1144         struct secinfo *exs;
1145         rpc_gss_callback_t cb;
1146         char *pathbuf;
1147         char *log_buffer;
1148         char *tagbuf;
1149         int callback;
1150         int allocd_seccnt;
1151         STRUCT_HANDLE(exportfs_args, uap);
1152         STRUCT_DECL(exportdata, uexi);
1153         struct secinfo newsec[MAX_FLAVORS];
1154         int newcnt;
1155         struct secinfo oldsec[MAX_FLAVORS];
1156         int oldcnt;
1157         int i;
1158         struct pathname lookpn;
1159         nfs_export_t *ne = nfs_get_export();
1160 
1161         STRUCT_SET_HANDLE(uap, model, args);
1162 
1163         /* Read in pathname from userspace */
1164         if (error = pn_get(STRUCT_FGETP(uap, dname), UIO_USERSPACE, &lookpn))
1165                 return (error);
1166 
1167         /* Walk the export list looking for that pathname */
1168         rw_enter(&ne->exported_lock, RW_READER);
1169         DTRACE_PROBE(nfss__i__exported_lock1_start);
1170         for (ex1 = ne->exptable_path_hash[pkp_tab_hash(lookpn.pn_path,
1171             strlen(lookpn.pn_path))]; ex1; ex1 = ex1->path_hash.next) {
1172                 if (ex1 != ne->exi_root && 0 ==
1173                     strcmp(ex1->exi_export.ex_path, lookpn.pn_path)) {
1174                         exi_hold(ex1);
1175                         break;
1176                 }
1177         }
1178         DTRACE_PROBE(nfss__i__exported_lock1_stop);
1179         rw_exit(&ne->exported_lock);
1180 
1181         /* Is this an unshare? */
1182         if (STRUCT_FGETP(uap, uex) == NULL) {
1183                 pn_free(&lookpn);
1184                 if (ex1 == NULL)
1185                         return (EINVAL);
1186                 error = unexport(ne, ex1);
1187                 exi_rele(&ex1);
1188                 return (error);
1189         }
1190 
1191         /* It is a share or a re-share */
1192         error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1193             FOLLOW, &dvp, &vp);
1194         if (error == EINVAL) {
1195                 /*
1196                  * if fname resolves to / we get EINVAL error
1197                  * since we wanted the parent vnode. Try again
1198                  * with NULL dvp.
1199                  */
1200                 error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1201                     FOLLOW, NULL, &vp);
1202                 dvp = NULL;
1203         }
1204         if (!error && vp == NULL) {
1205                 /* Last component of fname not found */
1206                 if (dvp != NULL)
1207                         VN_RELE(dvp);
1208                 error = ENOENT;
1209         }
1210         if (error) {
1211                 pn_free(&lookpn);
1212                 if (ex1)
1213                         exi_rele(&ex1);
1214                 return (error);
1215         }
1216 
1217         /*
1218          * 'vp' may be an AUTOFS node, so we perform a
1219          * VOP_ACCESS() to trigger the mount of the
1220          * intended filesystem, so we can share the intended
1221          * filesystem instead of the AUTOFS filesystem.
1222          */
1223         (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
1224 
1225         /*
1226          * We're interested in the top most filesystem.
1227          * This is specially important when uap->dname is a trigger
1228          * AUTOFS node, since we're really interested in sharing the
1229          * filesystem AUTOFS mounted as result of the VOP_ACCESS()
1230          * call not the AUTOFS node itself.
1231          */
1232         if (vn_mountedvfs(vp) != NULL) {
1233                 if (error = traverse(&vp)) {
1234                         VN_RELE(vp);
1235                         if (dvp != NULL)
1236                                 VN_RELE(dvp);
1237                         pn_free(&lookpn);
1238                         if (ex1)
1239                                 exi_rele(&ex1);
1240                         return (error);
1241                 }
1242         }
1243 
1244         /* Do not allow sharing another vnode for already shared path */
1245         if (ex1 && !PSEUDO(ex1) && !VN_CMP(ex1->exi_vp, vp)) {
1246                 VN_RELE(vp);
1247                 if (dvp != NULL)
1248                         VN_RELE(dvp);
1249                 pn_free(&lookpn);
1250                 exi_rele(&ex1);
1251                 return (EEXIST);
1252         }
1253         if (ex1)
1254                 exi_rele(&ex1);
1255 
1256         /*
1257          * Get the vfs id
1258          */
1259         bzero(&fid, sizeof (fid));
1260         fid.fid_len = MAXFIDSZ;
1261         error = VOP_FID(vp, &fid, NULL);
1262         fsid = vp->v_vfsp->vfs_fsid;
1263 
1264         if (error) {
1265                 VN_RELE(vp);
1266                 if (dvp != NULL)
1267                         VN_RELE(dvp);
1268                 /*
1269                  * If VOP_FID returns ENOSPC then the fid supplied
1270                  * is too small.  For now we simply return EREMOTE.
1271                  */
1272                 if (error == ENOSPC)
1273                         error = EREMOTE;
1274                 pn_free(&lookpn);
1275                 return (error);
1276         }
1277 
1278         /*
1279          * Do not allow re-sharing a shared vnode under a different path
1280          * PSEUDO export has ex_path fabricated, e.g. "/tmp (pseudo)", skip it.
1281          */
1282         rw_enter(&ne->exported_lock, RW_READER);
1283         DTRACE_PROBE(nfss__i__exported_lock2_start);
1284         for (ex2 = ne->exptable[exptablehash(&fsid, &fid)]; ex2;
1285             ex2 = ex2->fid_hash.next) {
1286                 if (ex2 != ne->exi_root && !PSEUDO(ex2) &&
1287                     VN_CMP(ex2->exi_vp, vp) &&
1288                     strcmp(ex2->exi_export.ex_path, lookpn.pn_path) != 0) {
1289                         DTRACE_PROBE(nfss__i__exported_lock2_stop);
1290                         rw_exit(&ne->exported_lock);
1291                         VN_RELE(vp);
1292                         if (dvp != NULL)
1293                                 VN_RELE(dvp);
1294                         pn_free(&lookpn);
1295                         return (EEXIST);
1296                 }
1297         }
1298         DTRACE_PROBE(nfss__i__exported_lock2_stop);
1299         rw_exit(&ne->exported_lock);
1300         pn_free(&lookpn);
1301 
1302         exi = kmem_zalloc(sizeof (*exi), KM_SLEEP);
1303         exi->exi_fsid = fsid;
1304         exi->exi_fid = fid;
1305         exi->exi_vp = vp;
1306         exi->exi_count = 1;
1307         exi->exi_zoneid = crgetzoneid(cr);
1308         exi->exi_volatile_dev = (vfssw[vp->v_vfsp->vfs_fstype].vsw_flag &
1309             VSW_VOLATILEDEV) ? 1 : 0;
1310         mutex_init(&exi->exi_lock, NULL, MUTEX_DEFAULT, NULL);
1311         exi->exi_dvp = dvp;
1312 
1313         /*
1314          * Initialize auth cache and auth cache lock
1315          */
1316         for (i = 0; i < AUTH_TABLESIZE; i++) {
1317                 exi->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1318                 avl_create(exi->exi_cache[i], nfsauth_cache_clnt_compar,
1319                     sizeof (struct auth_cache_clnt),
1320                     offsetof(struct auth_cache_clnt, authc_link));
1321         }
1322         rw_init(&exi->exi_cache_lock, NULL, RW_DEFAULT, NULL);
1323 
1324         /*
1325          * Build up the template fhandle
1326          */
1327         exi->exi_fh.fh_fsid = fsid;
1328         if (exi->exi_fid.fid_len > sizeof (exi->exi_fh.fh_xdata)) {
1329                 error = EREMOTE;
1330                 goto out1;
1331         }
1332         exi->exi_fh.fh_xlen = exi->exi_fid.fid_len;
1333         bcopy(exi->exi_fid.fid_data, exi->exi_fh.fh_xdata,
1334             exi->exi_fid.fid_len);
1335 
1336         exi->exi_fh.fh_len = sizeof (exi->exi_fh.fh_data);
1337 
1338         kex = &exi->exi_export;
1339 
1340         /*
1341          * Load in everything, and do sanity checking
1342          */
1343         STRUCT_INIT(uexi, model);
1344         if (copyin(STRUCT_FGETP(uap, uex), STRUCT_BUF(uexi),
1345             STRUCT_SIZE(uexi))) {
1346                 error = EFAULT;
1347                 goto out1;
1348         }
1349 
1350         kex->ex_version = STRUCT_FGET(uexi, ex_version);
1351         if (kex->ex_version != EX_CURRENT_VERSION) {
1352                 error = EINVAL;
1353                 cmn_err(CE_WARN,
1354                     "NFS: exportfs requires export struct version 2 - got %d\n",
1355                     kex->ex_version);
1356                 goto out1;
1357         }
1358 
1359         /*
1360          * Must have at least one security entry
1361          */
1362         kex->ex_seccnt = STRUCT_FGET(uexi, ex_seccnt);
1363         if (kex->ex_seccnt < 1) {
1364                 error = EINVAL;
1365                 goto out1;
1366         }
1367 
1368         kex->ex_path = STRUCT_FGETP(uexi, ex_path);
1369         kex->ex_pathlen = STRUCT_FGET(uexi, ex_pathlen);
1370         kex->ex_flags = STRUCT_FGET(uexi, ex_flags);
1371         kex->ex_anon = STRUCT_FGET(uexi, ex_anon);
1372         kex->ex_secinfo = STRUCT_FGETP(uexi, ex_secinfo);
1373         kex->ex_index = STRUCT_FGETP(uexi, ex_index);
1374         kex->ex_log_buffer = STRUCT_FGETP(uexi, ex_log_buffer);
1375         kex->ex_log_bufferlen = STRUCT_FGET(uexi, ex_log_bufferlen);
1376         kex->ex_tag = STRUCT_FGETP(uexi, ex_tag);
1377         kex->ex_taglen = STRUCT_FGET(uexi, ex_taglen);
1378 
1379         /*
1380          * Copy the exported pathname into
1381          * an appropriately sized buffer.
1382          */
1383         pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1384         if (copyinstr(kex->ex_path, pathbuf, MAXPATHLEN, &kex->ex_pathlen)) {
1385                 kmem_free(pathbuf, MAXPATHLEN);
1386                 error = EFAULT;
1387                 goto out1;
1388         }
1389         kex->ex_path = kmem_alloc(kex->ex_pathlen + 1, KM_SLEEP);
1390         bcopy(pathbuf, kex->ex_path, kex->ex_pathlen);
1391         kex->ex_path[kex->ex_pathlen] = '\0';
1392         kmem_free(pathbuf, MAXPATHLEN);
1393 
1394         /*
1395          * Get the path to the logging buffer and the tag
1396          */
1397         if (kex->ex_flags & EX_LOG) {
1398                 log_buffer = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1399                 if (copyinstr(kex->ex_log_buffer, log_buffer, MAXPATHLEN,
1400                     &kex->ex_log_bufferlen)) {
1401                         kmem_free(log_buffer, MAXPATHLEN);
1402                         error = EFAULT;
1403                         goto out2;
1404                 }
1405                 kex->ex_log_buffer =
1406                     kmem_alloc(kex->ex_log_bufferlen + 1, KM_SLEEP);
1407                 bcopy(log_buffer, kex->ex_log_buffer, kex->ex_log_bufferlen);
1408                 kex->ex_log_buffer[kex->ex_log_bufferlen] = '\0';
1409                 kmem_free(log_buffer, MAXPATHLEN);
1410 
1411                 tagbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1412                 if (copyinstr(kex->ex_tag, tagbuf, MAXPATHLEN,
1413                     &kex->ex_taglen)) {
1414                         kmem_free(tagbuf, MAXPATHLEN);
1415                         error = EFAULT;
1416                         goto out3;
1417                 }
1418                 kex->ex_tag = kmem_alloc(kex->ex_taglen + 1, KM_SLEEP);
1419                 bcopy(tagbuf, kex->ex_tag, kex->ex_taglen);
1420                 kex->ex_tag[kex->ex_taglen] = '\0';
1421                 kmem_free(tagbuf, MAXPATHLEN);
1422         }
1423 
1424         /*
1425          * Load the security information for each flavor
1426          */
1427         allocsize = kex->ex_seccnt * SIZEOF_STRUCT(secinfo, model);
1428         sp = kmem_zalloc(allocsize, KM_SLEEP);
1429         if (copyin(kex->ex_secinfo, sp, allocsize)) {
1430                 kmem_free(sp, allocsize);
1431                 error = EFAULT;
1432                 goto out4;
1433         }
1434 
1435         /*
1436          * All of these nested structures need to be converted to
1437          * the kernel native format.
1438          */
1439         if (model != DATAMODEL_NATIVE) {
1440                 size_t allocsize2;
1441                 struct secinfo *sp2;
1442 
1443                 allocsize2 = kex->ex_seccnt * sizeof (struct secinfo);
1444                 sp2 = kmem_zalloc(allocsize2, KM_SLEEP);
1445 
1446                 for (i = 0; i < kex->ex_seccnt; i++) {
1447                         STRUCT_HANDLE(secinfo, usi);
1448 
1449                         STRUCT_SET_HANDLE(usi, model,
1450                             (struct secinfo *)((caddr_t)sp +
1451                             (i * SIZEOF_STRUCT(secinfo, model))));
1452                         bcopy(STRUCT_FGET(usi, s_secinfo.sc_name),
1453                             sp2[i].s_secinfo.sc_name, MAX_NAME_LEN);
1454                         sp2[i].s_secinfo.sc_nfsnum =
1455                             STRUCT_FGET(usi, s_secinfo.sc_nfsnum);
1456                         sp2[i].s_secinfo.sc_rpcnum =
1457                             STRUCT_FGET(usi, s_secinfo.sc_rpcnum);
1458                         bcopy(STRUCT_FGET(usi, s_secinfo.sc_gss_mech),
1459                             sp2[i].s_secinfo.sc_gss_mech, MAX_NAME_LEN);
1460                         sp2[i].s_secinfo.sc_gss_mech_type =
1461                             STRUCT_FGETP(usi, s_secinfo.sc_gss_mech_type);
1462                         sp2[i].s_secinfo.sc_qop =
1463                             STRUCT_FGET(usi, s_secinfo.sc_qop);
1464                         sp2[i].s_secinfo.sc_service =
1465                             STRUCT_FGET(usi, s_secinfo.sc_service);
1466 
1467                         sp2[i].s_flags = STRUCT_FGET(usi, s_flags);
1468                         sp2[i].s_window = STRUCT_FGET(usi, s_window);
1469                         sp2[i].s_rootid = STRUCT_FGET(usi, s_rootid);
1470                         sp2[i].s_rootcnt = STRUCT_FGET(usi, s_rootcnt);
1471                         sp2[i].s_rootnames = STRUCT_FGETP(usi, s_rootnames);
1472                 }
1473                 kmem_free(sp, allocsize);
1474                 sp = sp2;
1475                 allocsize = allocsize2;
1476         }
1477 
1478         kex->ex_secinfo = sp;
1479 
1480         /*
1481          * And now copy rootnames for each individual secinfo.
1482          */
1483         callback = 0;
1484         allocd_seccnt = 0;
1485         while (allocd_seccnt < kex->ex_seccnt) {
1486 
1487                 exs = &sp[allocd_seccnt];
1488                 if (exs->s_rootcnt > 0) {
1489                         if (!sec_svc_loadrootnames(exs->s_secinfo.sc_rpcnum,
1490                             exs->s_rootcnt, &exs->s_rootnames, model)) {
1491                                 error = EFAULT;
1492                                 goto out5;
1493                         }
1494                 }
1495 
1496                 if (exs->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
1497                         rpc_gss_OID mech_tmp;
1498                         STRUCT_DECL(rpc_gss_OID_s, umech_tmp);
1499                         caddr_t elements_tmp;
1500 
1501                         /* Copyin mechanism type */
1502                         STRUCT_INIT(umech_tmp, model);
1503                         mech_tmp = kmem_alloc(sizeof (*mech_tmp), KM_SLEEP);
1504                         if (copyin(exs->s_secinfo.sc_gss_mech_type,
1505                             STRUCT_BUF(umech_tmp), STRUCT_SIZE(umech_tmp))) {
1506                                 kmem_free(mech_tmp, sizeof (*mech_tmp));
1507                                 error = EFAULT;
1508                                 goto out5;
1509                         }
1510                         mech_tmp->length = STRUCT_FGET(umech_tmp, length);
1511                         mech_tmp->elements = STRUCT_FGETP(umech_tmp, elements);
1512 
1513                         elements_tmp = kmem_alloc(mech_tmp->length, KM_SLEEP);
1514                         if (copyin(mech_tmp->elements, elements_tmp,
1515                             mech_tmp->length)) {
1516                                 kmem_free(elements_tmp, mech_tmp->length);
1517                                 kmem_free(mech_tmp, sizeof (*mech_tmp));
1518                                 error = EFAULT;
1519                                 goto out5;
1520                         }
1521                         mech_tmp->elements = elements_tmp;
1522                         exs->s_secinfo.sc_gss_mech_type = mech_tmp;
1523                         allocd_seccnt++;
1524 
1525                         callback = 1;
1526                 } else
1527                         allocd_seccnt++;
1528         }
1529 
1530         /*
1531          * Init the secinfo reference count and mark these flavors
1532          * explicitly exported flavors.
1533          */
1534         for (i = 0; i < kex->ex_seccnt; i++) {
1535                 kex->ex_secinfo[i].s_flags |= M_4SEC_EXPORTED;
1536                 kex->ex_secinfo[i].s_refcnt = 1;
1537         }
1538 
1539         /*
1540          *  Set up rpcsec_gss callback routine entry if any.
1541          */
1542         if (callback) {
1543                 cb.callback = rfs_gsscallback;
1544                 cb.program = NFS_ACL_PROGRAM;
1545                 for (cb.version = NFS_ACL_VERSMIN;
1546                     cb.version <= NFS_ACL_VERSMAX; cb.version++) {
1547                         (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1548                             (void *)&cb);
1549                 }
1550 
1551                 cb.program = NFS_PROGRAM;
1552                 for (cb.version = NFS_VERSMIN;
1553                     cb.version <= NFS_VERSMAX; cb.version++) {
1554                         (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1555                             (void *)&cb);
1556                 }
1557         }
1558 
1559         /*
1560          * Check the index flag. Do this here to avoid holding the
1561          * lock while dealing with the index option (as we do with
1562          * the public option).
1563          */
1564         if (kex->ex_flags & EX_INDEX) {
1565                 if (!kex->ex_index) {        /* sanity check */
1566                         error = EINVAL;
1567                         goto out5;
1568                 }
1569                 if (error = loadindex(kex))
1570                         goto out5;
1571         }
1572 
1573         if (kex->ex_flags & EX_LOG) {
1574                 if (error = nfslog_setup(exi))
1575                         goto out6;
1576         }
1577 
1578         /*
1579          * Insert the new entry at the front of the export list
1580          */
1581         rw_enter(&ne->exported_lock, RW_WRITER);
1582         DTRACE_PROBE(nfss__i__exported_lock3_start);
1583 
1584         export_link(ne, exi);
1585 
1586         /*
1587          * Check the rest of the list for an old entry for the fs.
1588          * If one is found then unlink it, wait until this is the
1589          * only reference and then free it.
1590          */
1591         for (ex = exi->fid_hash.next; ex != NULL; ex = ex->fid_hash.next) {
1592                 if (ex != ne->exi_root && VN_CMP(ex->exi_vp, vp)) {
1593                         mutex_enter(&nfs_exi_id_lock);
1594                         avl_remove(&exi_id_tree, ex);
1595                         mutex_exit(&nfs_exi_id_lock);
1596                         export_unlink(ne, ex);
1597                         break;
1598                 }
1599         }
1600 
1601         /*
1602          * If the public filehandle is pointing at the
1603          * old entry, then point it back at the root.
1604          */
1605         if (ex != NULL && ex == ne->exi_public)
1606                 ne->exi_public = ne->exi_root;
1607 
1608         /*
1609          * If the public flag is on, make the global exi_public
1610          * point to this entry and turn off the public bit so that
1611          * we can distinguish it from the place holder export.
1612          */
1613         if (kex->ex_flags & EX_PUBLIC) {
1614                 ne->exi_public = exi;
1615                 kex->ex_flags &= ~EX_PUBLIC;
1616         }
1617 
1618 #ifdef VOLATILE_FH_TEST
1619         /*
1620          * Set up the volatile_id value if volatile on share.
1621          * The list of volatile renamed filehandles is always destroyed,
1622          * if the fs was reshared.
1623          */
1624         if (kex->ex_flags & EX_VOLFH)
1625                 exi->exi_volatile_id = gethrestime_sec();
1626 
1627         mutex_init(&exi->exi_vol_rename_lock, NULL, MUTEX_DEFAULT, NULL);
1628 #endif /* VOLATILE_FH_TEST */
1629 
1630         /*
1631          * If this is a new export, then climb up
1632          * the tree and check if any pseudo exports
1633          * need to be created to provide a path for
1634          * NFS v4 clients.
1635          */
1636         if (ex == NULL) {
1637                 error = treeclimb_export(exi);
1638                 if (error)
1639                         goto out7;
1640         } else {
1641                 /* If it's a re-export update namespace tree */
1642                 exi->exi_tree = ex->exi_tree;
1643                 exi->exi_tree->tree_exi = exi;
1644 
1645                 /* Update the change timestamp */
1646                 tree_update_change(ne, exi->exi_tree, NULL);
1647         }
1648 
1649         /*
1650          * build a unique flavor list from the flavors specified
1651          * in the share cmd.  unique means that each flavor only
1652          * appears once in the secinfo list -- no duplicates allowed.
1653          */
1654         newcnt = build_seclist_nodups(&exi->exi_export, newsec, FALSE);
1655 
1656         srv_secinfo_treeclimb(ne, exi, newsec, newcnt, TRUE);
1657 
1658         /*
1659          * If re-sharing an old export entry, update the secinfo data
1660          * depending on if the old entry is a pseudo node or not.
1661          */
1662         if (ex != NULL) {
1663                 oldcnt = build_seclist_nodups(&ex->exi_export, oldsec, FALSE);
1664                 if (PSEUDO(ex)) {
1665                         /*
1666                          * The dir being shared is a pseudo export root (which
1667                          * will be transformed into a real export root).  The
1668                          * flavor(s) of the new share were propagated to the
1669                          * ancestors by srv_secinfo_treeclimb() above.  Now
1670                          * transfer the implicit flavor refs from the old
1671                          * pseudo exprot root to the new (real) export root.
1672                          */
1673                         srv_secinfo_add(&exi->exi_export.ex_secinfo,
1674                             &exi->exi_export.ex_seccnt, oldsec, oldcnt, TRUE);
1675                 } else {
1676                         /*
1677                          * First transfer implicit flavor refs to new export.
1678                          * Remove old flavor refs last.
1679                          */
1680                         srv_secinfo_exp2exp(&exi->exi_export, oldsec, oldcnt);
1681                         srv_secinfo_treeclimb(ne, ex, oldsec, oldcnt, FALSE);
1682                 }
1683         }
1684 
1685         /*
1686          * If it's a re-export and the old entry has a pseudonode list,
1687          * transfer it to the new export.
1688          */
1689         if (ex != NULL && (ex->exi_visible != NULL)) {
1690                 exi->exi_visible = ex->exi_visible;
1691                 ex->exi_visible = NULL;
1692         }
1693 
1694         /*
1695          * Initialize exi_id and exi_kstats
1696          */
1697         if (ex != NULL) {
1698                 exi->exi_id = ex->exi_id;
1699                 exi->exi_kstats = ex->exi_kstats;
1700                 ex->exi_kstats = NULL;
1701                 exp_kstats_reset(exi->exi_kstats, kex->ex_path,
1702                     kex->ex_pathlen, FALSE);
1703         } else {
1704                 mutex_enter(&nfs_exi_id_lock);
1705                 exi->exi_id = exi_id_get_next();
1706                 mutex_exit(&nfs_exi_id_lock);
1707                 exi->exi_kstats = exp_kstats_init(crgetzoneid(cr), exi->exi_id,
1708                     kex->ex_path, kex->ex_pathlen, FALSE);
1709         }
1710         mutex_enter(&nfs_exi_id_lock);
1711         avl_add(&exi_id_tree, exi);
1712         mutex_exit(&nfs_exi_id_lock);
1713 
1714         DTRACE_PROBE(nfss__i__exported_lock3_stop);
1715         rw_exit(&ne->exported_lock);
1716 
1717         if (ne->exi_public == exi || kex->ex_flags & EX_LOG) {
1718                 /*
1719                  * Log share operation to this buffer only.
1720                  */
1721                 nfslog_share_record(exi, cr);
1722         }
1723 
1724         if (ex != NULL)
1725                 exi_rele(&ex);
1726 
1727         return (0);
1728 
1729 out7:
1730         /* Unlink the new export in exptable. */
1731         export_unlink(ne, exi);
1732         DTRACE_PROBE(nfss__i__exported_lock3_stop);
1733         rw_exit(&ne->exported_lock);
1734 out6:
1735         if (kex->ex_flags & EX_INDEX)
1736                 kmem_free(kex->ex_index, strlen(kex->ex_index) + 1);
1737 out5:
1738         /* free partially completed allocation */
1739         while (--allocd_seccnt >= 0) {
1740                 exs = &kex->ex_secinfo[allocd_seccnt];
1741                 srv_secinfo_entry_free(exs);
1742         }
1743 
1744         if (kex->ex_secinfo) {
1745                 kmem_free(kex->ex_secinfo,
1746                     kex->ex_seccnt * sizeof (struct secinfo));
1747         }
1748 
1749 out4:
1750         if ((kex->ex_flags & EX_LOG) && kex->ex_tag != NULL)
1751                 kmem_free(kex->ex_tag, kex->ex_taglen + 1);
1752 out3:
1753         if ((kex->ex_flags & EX_LOG) && kex->ex_log_buffer != NULL)
1754                 kmem_free(kex->ex_log_buffer, kex->ex_log_bufferlen + 1);
1755 out2:
1756         kmem_free(kex->ex_path, kex->ex_pathlen + 1);
1757 out1:
1758         VN_RELE(vp);
1759         if (dvp != NULL)
1760                 VN_RELE(dvp);
1761         mutex_destroy(&exi->exi_lock);
1762         rw_destroy(&exi->exi_cache_lock);
1763         for (i = 0; i < AUTH_TABLESIZE; i++) {
1764                 avl_destroy(exi->exi_cache[i]);
1765                 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
1766         }
1767 
1768         kmem_free(exi, sizeof (*exi));
1769 
1770         return (error);
1771 }
1772 
1773 /*
1774  * Remove the exportinfo from the export list
1775  */
1776 void
1777 export_unlink(nfs_export_t *ne, struct exportinfo *exi)
1778 {
1779         ASSERT(RW_WRITE_HELD(&ne->exported_lock));
1780 
1781         exp_hash_unlink(exi, fid_hash);
1782         exp_hash_unlink(exi, path_hash);
1783 }
1784 
1785 /*
1786  * Unexport an exported filesystem
1787  */
1788 static int
1789 unexport(nfs_export_t *ne, struct exportinfo *exi)
1790 {
1791         struct secinfo cursec[MAX_FLAVORS];
1792         int curcnt;
1793 
1794         rw_enter(&ne->exported_lock, RW_WRITER);
1795 
1796         /* Check if exi is still linked in the export table */
1797         if (!EXP_LINKED(exi) || PSEUDO(exi)) {
1798                 rw_exit(&ne->exported_lock);
1799                 return (EINVAL);
1800         }
1801 
1802         exp_kstats_delete(exi->exi_kstats);
1803         mutex_enter(&nfs_exi_id_lock);
1804         avl_remove(&exi_id_tree, exi);
1805         mutex_exit(&nfs_exi_id_lock);
1806         export_unlink(ne, exi);
1807 
1808         /*
1809          * Remove security flavors before treeclimb_unexport() is called
1810          * because srv_secinfo_treeclimb needs the namespace tree
1811          */
1812         curcnt = build_seclist_nodups(&exi->exi_export, cursec, TRUE);
1813         srv_secinfo_treeclimb(ne, exi, cursec, curcnt, FALSE);
1814 
1815         /*
1816          * If there's a visible list, then need to leave
1817          * a pseudo export here to retain the visible list
1818          * for paths to exports below.
1819          */
1820         if (exi->exi_visible != NULL) {
1821                 struct exportinfo *newexi;
1822 
1823                 newexi = pseudo_exportfs(ne, exi->exi_vp, &exi->exi_fid,
1824                     exi->exi_visible, &exi->exi_export);
1825                 exi->exi_visible = NULL;
1826 
1827                 /* interconnect the existing treenode with the new exportinfo */
1828                 newexi->exi_tree = exi->exi_tree;
1829                 newexi->exi_tree->tree_exi = newexi;
1830 
1831                 /* Update the change timestamp */
1832                 tree_update_change(ne, exi->exi_tree, NULL);
1833         } else {
1834                 treeclimb_unexport(ne, exi);
1835         }
1836 
1837         rw_exit(&ne->exported_lock);
1838 
1839         /*
1840          * Need to call into the NFSv4 server and release all data
1841          * held on this particular export.  This is important since
1842          * the v4 server may be holding file locks or vnodes under
1843          * this export.
1844          */
1845         rfs4_clean_state_exi(exi);
1846 
1847         /*
1848          * Notify the lock manager that the filesystem is being
1849          * unexported.
1850          */
1851         lm_unexport(exi);
1852 
1853         /*
1854          * If this was a public export, restore
1855          * the public filehandle to the root.
1856          */
1857         if (exi == ne->exi_public) {
1858                 ne->exi_public = ne->exi_root;
1859 
1860                 nfslog_share_record(ne->exi_public, CRED());
1861         }
1862 
1863         if (exi->exi_export.ex_flags & EX_LOG)
1864                 nfslog_unshare_record(exi, CRED());
1865 
1866         exi_rele(&exi);
1867         return (0);
1868 }
1869 
1870 /*
1871  * Get file handle system call.
1872  * Takes file name and returns a file handle for it.
1873  * Credentials must be verified before calling.
1874  */
1875 int
1876 nfs_getfh(struct nfs_getfh_args *args, model_t model, cred_t *cr)
1877 {
1878         nfs_fh3 fh;
1879         char buf[NFS3_MAXFHSIZE];
1880         char *logptr, logbuf[NFS3_MAXFHSIZE];
1881         int l = NFS3_MAXFHSIZE;
1882         vnode_t *vp;
1883         vnode_t *dvp;
1884         struct exportinfo *exi;
1885         int error;
1886         int vers;
1887         STRUCT_HANDLE(nfs_getfh_args, uap);
1888 
1889 #ifdef lint
1890         model = model;          /* STRUCT macros don't always use it */
1891 #endif
1892 
1893         STRUCT_SET_HANDLE(uap, model, args);
1894 
1895         error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
1896             FOLLOW, &dvp, &vp);
1897         if (error == EINVAL) {
1898                 /*
1899                  * if fname resolves to / we get EINVAL error
1900                  * since we wanted the parent vnode. Try again
1901                  * with NULL dvp.
1902                  */
1903                 error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
1904                     FOLLOW, NULL, &vp);
1905                 dvp = NULL;
1906         }
1907         if (!error && vp == NULL) {
1908                 /*
1909                  * Last component of fname not found
1910                  */
1911                 if (dvp != NULL) {
1912                         VN_RELE(dvp);
1913                 }
1914                 error = ENOENT;
1915         }
1916         if (error)
1917                 return (error);
1918 
1919         /*
1920          * 'vp' may be an AUTOFS node, so we perform a
1921          * VOP_ACCESS() to trigger the mount of the
1922          * intended filesystem, so we can share the intended
1923          * filesystem instead of the AUTOFS filesystem.
1924          */
1925         (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
1926 
1927         /*
1928          * We're interested in the top most filesystem.
1929          * This is specially important when uap->dname is a trigger
1930          * AUTOFS node, since we're really interested in sharing the
1931          * filesystem AUTOFS mounted as result of the VOP_ACCESS()
1932          * call not the AUTOFS node itself.
1933          */
1934         if (vn_mountedvfs(vp) != NULL) {
1935                 if (error = traverse(&vp)) {
1936                         VN_RELE(vp);
1937                         if (dvp != NULL)
1938                                 VN_RELE(dvp);
1939                         return (error);
1940                 }
1941         }
1942 
1943         vers = STRUCT_FGET(uap, vers);
1944         exi = nfs_vptoexi(dvp, vp, cr, NULL, &error, FALSE);
1945         if (!error) {
1946                 if (vers == NFS_VERSION) {
1947                         error = makefh((fhandle_t *)buf, vp, exi);
1948                         l = NFS_FHSIZE;
1949                         logptr = buf;
1950                 } else if (vers == NFS_V3) {
1951                         int i, sz, pad;
1952 
1953                         error = makefh3(&fh, vp, exi);
1954                         l = RNDUP(fh.fh3_length);
1955                         if (!error && (l > sizeof (fhandle3_t)))
1956                                 error = EREMOTE;
1957                         logptr = logbuf;
1958                         if (!error) {
1959                                 i = 0;
1960                                 sz = sizeof (fsid_t);
1961                                 bcopy(&fh.fh3_fsid, &buf[i], sz);
1962                                 i += sz;
1963 
1964                                 /*
1965                                  * For backwards compatibility, the
1966                                  * fid length may be less than
1967                                  * NFS_FHMAXDATA, but it was always
1968                                  * encoded as NFS_FHMAXDATA bytes.
1969                                  */
1970 
1971                                 sz = sizeof (ushort_t);
1972                                 bcopy(&fh.fh3_len, &buf[i], sz);
1973                                 i += sz;
1974                                 bcopy(fh.fh3_data, &buf[i], fh.fh3_len);
1975                                 i += fh.fh3_len;
1976                                 pad = (NFS_FHMAXDATA - fh.fh3_len);
1977                                 if (pad > 0) {
1978                                         bzero(&buf[i], pad);
1979                                         i += pad;
1980                                         l += pad;
1981                                 }
1982 
1983                                 sz = sizeof (ushort_t);
1984                                 bcopy(&fh.fh3_xlen, &buf[i], sz);
1985                                 i += sz;
1986                                 bcopy(fh.fh3_xdata, &buf[i], fh.fh3_xlen);
1987                                 i += fh.fh3_xlen;
1988                                 pad = (NFS_FHMAXDATA - fh.fh3_xlen);
1989                                 if (pad > 0) {
1990                                         bzero(&buf[i], pad);
1991                                         i += pad;
1992                                         l += pad;
1993                                 }
1994                         }
1995                         /*
1996                          * If we need to do NFS logging, the filehandle
1997                          * must be downsized to 32 bytes.
1998                          */
1999                         if (!error && exi->exi_export.ex_flags & EX_LOG) {
2000                                 i = 0;
2001                                 sz = sizeof (fsid_t);
2002                                 bcopy(&fh.fh3_fsid, &logbuf[i], sz);
2003                                 i += sz;
2004                                 sz = sizeof (ushort_t);
2005                                 bcopy(&fh.fh3_len, &logbuf[i], sz);
2006                                 i += sz;
2007                                 sz = NFS_FHMAXDATA;
2008                                 bcopy(fh.fh3_data, &logbuf[i], sz);
2009                                 i += sz;
2010                                 sz = sizeof (ushort_t);
2011                                 bcopy(&fh.fh3_xlen, &logbuf[i], sz);
2012                                 i += sz;
2013                                 sz = NFS_FHMAXDATA;
2014                                 bcopy(fh.fh3_xdata, &logbuf[i], sz);
2015                                 i += sz;
2016                         }
2017                 }
2018                 if (!error && exi->exi_export.ex_flags & EX_LOG) {
2019                         nfslog_getfh(exi, (fhandle_t *)logptr,
2020                             STRUCT_FGETP(uap, fname), UIO_USERSPACE, cr);
2021                 }
2022                 exi_rele(&exi);
2023                 if (!error) {
2024                         if (copyout(&l, STRUCT_FGETP(uap, lenp), sizeof (int)))
2025                                 error = EFAULT;
2026                         if (copyout(buf, STRUCT_FGETP(uap, fhp), l))
2027                                 error = EFAULT;
2028                 }
2029         }
2030         VN_RELE(vp);
2031         if (dvp != NULL) {
2032                 VN_RELE(dvp);
2033         }
2034         return (error);
2035 }
2036 
2037 /*
2038  * Strategy: if vp is in the export list, then
2039  * return the associated file handle. Otherwise, ".."
2040  * once up the vp and try again, until the root of the
2041  * filesystem is reached.
2042  */
2043 struct   exportinfo *
2044 nfs_vptoexi(vnode_t *dvp, vnode_t *vp, cred_t *cr, int *walk,
2045     int *err, bool_t v4srv)
2046 {
2047         fid_t fid;
2048         int error;
2049         struct exportinfo *exi;
2050 
2051         ASSERT(vp);
2052         VN_HOLD(vp);
2053         if (dvp != NULL) {
2054                 VN_HOLD(dvp);
2055         }
2056         if (walk != NULL)
2057                 *walk = 0;
2058 
2059         for (;;) {
2060                 bzero(&fid, sizeof (fid));
2061                 fid.fid_len = MAXFIDSZ;
2062                 error = vop_fid_pseudo(vp, &fid);
2063                 if (error) {
2064                         /*
2065                          * If vop_fid_pseudo returns ENOSPC then the fid
2066                          * supplied is too small. For now we simply
2067                          * return EREMOTE.
2068                          */
2069                         if (error == ENOSPC)
2070                                 error = EREMOTE;
2071                         break;
2072                 }
2073 
2074                 if (v4srv)
2075                         exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
2076                 else
2077                         exi = checkexport(&vp->v_vfsp->vfs_fsid, &fid);
2078 
2079                 if (exi != NULL) {
2080                         /*
2081                          * Found the export info
2082                          */
2083                         break;
2084                 }
2085 
2086                 /*
2087                  * We have just failed finding a matching export.
2088                  * If we're at the root of this filesystem, then
2089                  * it's time to stop (with failure).
2090                  */
2091                 if (vp->v_flag & VROOT) {
2092                         error = EINVAL;
2093                         break;
2094                 }
2095 
2096                 if (walk != NULL)
2097                         (*walk)++;
2098 
2099                 /*
2100                  * Now, do a ".." up vp. If dvp is supplied, use it,
2101                  * otherwise, look it up.
2102                  */
2103                 if (dvp == NULL) {
2104                         error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, cr,
2105                             NULL, NULL, NULL);
2106                         if (error)
2107                                 break;
2108                 }
2109                 VN_RELE(vp);
2110                 vp = dvp;
2111                 dvp = NULL;
2112         }
2113         VN_RELE(vp);
2114         if (dvp != NULL) {
2115                 VN_RELE(dvp);
2116         }
2117         if (error != 0) {
2118                 if (err != NULL)
2119                         *err = error;
2120                 return (NULL);
2121         }
2122         return (exi);
2123 }
2124 
2125 int
2126 chk_clnt_sec(exportinfo_t *exi, struct svc_req *req)
2127 {
2128         int i, nfsflavor;
2129         struct secinfo *sp;
2130 
2131         /*
2132          *  Get the nfs flavor number from xprt.
2133          */
2134         nfsflavor = (int)(uintptr_t)req->rq_xprt->xp_cookie;
2135 
2136         sp = exi->exi_export.ex_secinfo;
2137         for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2138                 if ((nfsflavor == sp[i].s_secinfo.sc_nfsnum) &&
2139                     SEC_REF_EXPORTED(sp + i))
2140                         return (TRUE);
2141         }
2142         return (FALSE);
2143 }
2144 
2145 /*
2146  * Make an fhandle from a vnode
2147  */
2148 int
2149 makefh(fhandle_t *fh, vnode_t *vp, exportinfo_t *exi)
2150 {
2151         int error;
2152 
2153         *fh = exi->exi_fh;   /* struct copy */
2154 
2155         error = VOP_FID(vp, (fid_t *)&fh->fh_len, NULL);
2156         if (error) {
2157                 /*
2158                  * Should be something other than EREMOTE
2159                  */
2160                 return (EREMOTE);
2161         }
2162         return (0);
2163 }
2164 
2165 /*
2166  * This routine makes an overloaded V2 fhandle which contains
2167  * sec modes.
2168  *
2169  * Note that the first four octets contain the length octet,
2170  * the status octet, and two padded octets to make them XDR
2171  * four-octet aligned.
2172  *
2173  *   1   2   3   4                                          32
2174  * +---+---+---+---+---+---+---+---+   +---+---+---+---+   +---+
2175  * | l | s |   |   |     sec_1     |...|     sec_n     |...|   |
2176  * +---+---+---+---+---+---+---+---+   +---+---+---+---+   +---+
2177  *
2178  * where
2179  *
2180  *   the status octet s indicates whether there are more security
2181  *   flavors (1 means yes, 0 means no) that require the client to
2182  *   perform another 0x81 LOOKUP to get them,
2183  *
2184  *   the length octet l is the length describing the number of
2185  *   valid octets that follow.  (l = 4 * n, where n is the number
2186  *   of security flavors sent in the current overloaded filehandle.)
2187  *
2188  *   sec_index should always be in the inclusive range: [1 - ex_seccnt],
2189  *   and it tells server where to start within the secinfo array.
2190  *   Usually it will always be 1; however, if more flavors are used
2191  *   for the public export than can be encoded in the overloaded FH
2192  *   (7 for NFS2), subsequent SNEGO MCLs will have a larger index
2193  *   so the server will pick up where it left off from the previous
2194  *   MCL reply.
2195  *
2196  *   With NFS4 support, implicitly allowed flavors are also in
2197  *   the secinfo array; however, they should not be returned in
2198  *   SNEGO MCL replies.
2199  */
2200 int
2201 makefh_ol(fhandle_t *fh, exportinfo_t *exi, uint_t sec_index)
2202 {
2203         secinfo_t sec[MAX_FLAVORS];
2204         int totalcnt, i, *ipt, cnt, seccnt, secidx, fh_max_cnt;
2205         char *c;
2206 
2207         if (fh == NULL || exi == NULL || sec_index < 1)
2208                 return (EREMOTE);
2209 
2210         /*
2211          * WebNFS clients need to know the unique set of explicitly
2212          * shared flavors in used for the public export. When
2213          * "TRUE" is passed to build_seclist_nodups(), only explicitly
2214          * shared flavors are included in the list.
2215          */
2216         seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2217         if (sec_index > seccnt)
2218                 return (EREMOTE);
2219 
2220         fh_max_cnt = (NFS_FHSIZE / sizeof (int)) - 1;
2221         totalcnt = seccnt - sec_index + 1;
2222         cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2223 
2224         c = (char *)fh;
2225         /*
2226          * Encode the length octet representing the number of
2227          * security flavors (in bytes) in this overloaded fh.
2228          */
2229         *c = cnt * sizeof (int);
2230 
2231         /*
2232          * Encode the status octet that indicates whether there
2233          * are more security flavors the client needs to get.
2234          */
2235         *(c + 1) = totalcnt > fh_max_cnt;
2236 
2237         /*
2238          * put security flavors in the overloaded fh
2239          */
2240         ipt = (int *)(c + sizeof (int32_t));
2241         secidx = sec_index - 1;
2242         for (i = 0; i < cnt; i++) {
2243                 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2244         }
2245         return (0);
2246 }
2247 
2248 /*
2249  * Make an nfs_fh3 from a vnode
2250  */
2251 int
2252 makefh3(nfs_fh3 *fh, vnode_t *vp, struct exportinfo *exi)
2253 {
2254         int error;
2255         fid_t fid;
2256 
2257         bzero(&fid, sizeof (fid));
2258         fid.fid_len = sizeof (fh->fh3_data);
2259         error = VOP_FID(vp, &fid, NULL);
2260         if (error)
2261                 return (EREMOTE);
2262 
2263         bzero(fh, sizeof (nfs_fh3));
2264         fh->fh3_fsid = exi->exi_fsid;
2265         fh->fh3_len = fid.fid_len;
2266         bcopy(fid.fid_data, fh->fh3_data, fh->fh3_len);
2267 
2268         fh->fh3_xlen = exi->exi_fid.fid_len;
2269         ASSERT(fh->fh3_xlen <= sizeof (fh->fh3_xdata));
2270         bcopy(exi->exi_fid.fid_data, fh->fh3_xdata, fh->fh3_xlen);
2271 
2272         fh->fh3_length = sizeof (fh->fh3_fsid)
2273             + sizeof (fh->fh3_len) + fh->fh3_len
2274             + sizeof (fh->fh3_xlen) + fh->fh3_xlen;
2275         fh->fh3_flags = 0;
2276 
2277         return (0);
2278 }
2279 
2280 /*
2281  * This routine makes an overloaded V3 fhandle which contains
2282  * sec modes.
2283  *
2284  *  1        4
2285  * +--+--+--+--+
2286  * |    len    |
2287  * +--+--+--+--+
2288  *                                               up to 64
2289  * +--+--+--+--+--+--+--+--+--+--+--+--+     +--+--+--+--+
2290  * |s |  |  |  |   sec_1   |   sec_2   | ... |   sec_n   |
2291  * +--+--+--+--+--+--+--+--+--+--+--+--+     +--+--+--+--+
2292  *
2293  * len = 4 * (n+1), where n is the number of security flavors
2294  * sent in the current overloaded filehandle.
2295  *
2296  * the status octet s indicates whether there are more security
2297  * mechanisms (1 means yes, 0 means no) that require the client
2298  * to perform another 0x81 LOOKUP to get them.
2299  *
2300  * Three octets are padded after the status octet.
2301  */
2302 int
2303 makefh3_ol(nfs_fh3 *fh, struct exportinfo *exi, uint_t sec_index)
2304 {
2305         secinfo_t sec[MAX_FLAVORS];
2306         int totalcnt, cnt, *ipt, i, seccnt, fh_max_cnt, secidx;
2307         char *c;
2308 
2309         if (fh == NULL || exi == NULL || sec_index < 1)
2310                 return (EREMOTE);
2311 
2312         /*
2313          * WebNFS clients need to know the unique set of explicitly
2314          * shared flavors in used for the public export. When
2315          * "TRUE" is passed to build_seclist_nodups(), only explicitly
2316          * shared flavors are included in the list.
2317          */
2318         seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2319 
2320         if (sec_index > seccnt)
2321                 return (EREMOTE);
2322 
2323         fh_max_cnt = (NFS3_FHSIZE / sizeof (int)) - 1;
2324         totalcnt = seccnt - sec_index + 1;
2325         cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2326 
2327         /*
2328          * Place the length in fh3_length representing the number
2329          * of security flavors (in bytes) in this overloaded fh.
2330          */
2331         fh->fh3_flags = FH_WEBNFS;
2332         fh->fh3_length = (cnt+1) * sizeof (int32_t);
2333 
2334         c = (char *)&fh->fh3_u.nfs_fh3_i.fh3_i;
2335         /*
2336          * Encode the status octet that indicates whether there
2337          * are more security flavors the client needs to get.
2338          */
2339         *c = totalcnt > fh_max_cnt;
2340 
2341         /*
2342          * put security flavors in the overloaded fh
2343          */
2344         secidx = sec_index - 1;
2345         ipt = (int *)(c + sizeof (int32_t));
2346         for (i = 0; i < cnt; i++) {
2347                 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2348         }
2349         return (0);
2350 }
2351 
2352 /*
2353  * Make an nfs_fh4 from a vnode
2354  */
2355 int
2356 makefh4(nfs_fh4 *fh, vnode_t *vp, struct exportinfo *exi)
2357 {
2358         int error;
2359         nfs_fh4_fmt_t *fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2360         fid_t fid;
2361 
2362         bzero(&fid, sizeof (fid));
2363         fid.fid_len = MAXFIDSZ;
2364         /*
2365          * vop_fid_pseudo() is used to set up NFSv4 namespace, so
2366          * use vop_fid_pseudo() here to get the fid instead of VOP_FID.
2367          */
2368         error = vop_fid_pseudo(vp, &fid);
2369         if (error)
2370                 return (error);
2371 
2372         fh->nfs_fh4_len = NFS_FH4_LEN;
2373 
2374         fh_fmtp->fh4_i.fhx_fsid = exi->exi_fh.fh_fsid;
2375         fh_fmtp->fh4_i.fhx_xlen = exi->exi_fh.fh_xlen;
2376 
2377         bzero(fh_fmtp->fh4_i.fhx_data, sizeof (fh_fmtp->fh4_i.fhx_data));
2378         bzero(fh_fmtp->fh4_i.fhx_xdata, sizeof (fh_fmtp->fh4_i.fhx_xdata));
2379         ASSERT(exi->exi_fh.fh_xlen <= sizeof (fh_fmtp->fh4_i.fhx_xdata));
2380         bcopy(exi->exi_fh.fh_xdata, fh_fmtp->fh4_i.fhx_xdata,
2381             exi->exi_fh.fh_xlen);
2382 
2383         fh_fmtp->fh4_len = fid.fid_len;
2384         ASSERT(fid.fid_len <= sizeof (fh_fmtp->fh4_data));
2385         bcopy(fid.fid_data, fh_fmtp->fh4_data, fid.fid_len);
2386         fh_fmtp->fh4_flag = 0;
2387 
2388 #ifdef VOLATILE_FH_TEST
2389         /*
2390          * XXX (temporary?)
2391          * Use the rnode volatile_id value to add volatility to the fh.
2392          *
2393          * For testing purposes there are currently two scenarios, based
2394          * on whether the filesystem was shared with "volatile_fh"
2395          * or "expire_on_rename". In the first case, use the value of
2396          * export struct share_time as the volatile_id. In the second
2397          * case use the vnode volatile_id value (which is set to the
2398          * time in which the file was renamed).
2399          *
2400          * Note that the above are temporary constructs for testing only
2401          * XXX
2402          */
2403         if (exi->exi_export.ex_flags & EX_VOLRNM) {
2404                 fh_fmtp->fh4_volatile_id = find_volrnm_fh_id(exi, fh);
2405         } else if (exi->exi_export.ex_flags & EX_VOLFH) {
2406                 fh_fmtp->fh4_volatile_id = exi->exi_volatile_id;
2407         } else {
2408                 fh_fmtp->fh4_volatile_id = 0;
2409         }
2410 #endif /* VOLATILE_FH_TEST */
2411 
2412         return (0);
2413 }
2414 
2415 /*
2416  * Convert an fhandle into a vnode.
2417  * Uses the file id (fh_len + fh_data) in the fhandle to get the vnode.
2418  * WARNING: users of this routine must do a VN_RELE on the vnode when they
2419  * are done with it.
2420  */
2421 vnode_t *
2422 nfs_fhtovp(fhandle_t *fh, struct exportinfo *exi)
2423 {
2424         vfs_t *vfsp;
2425         vnode_t *vp;
2426         int error;
2427         fid_t *fidp;
2428 
2429         TRACE_0(TR_FAC_NFS, TR_FHTOVP_START,
2430             "fhtovp_start");
2431 
2432         if (exi == NULL) {
2433                 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2434                     "fhtovp_end:(%S)", "exi NULL");
2435                 return (NULL);  /* not exported */
2436         }
2437 
2438         ASSERT(exi->exi_vp != NULL);
2439 
2440         if (PUBLIC_FH2(fh)) {
2441                 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2442                         TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2443                             "fhtovp_end:(%S)", "root not exported");
2444                         return (NULL);
2445                 }
2446                 vp = exi->exi_vp;
2447                 VN_HOLD(vp);
2448                 return (vp);
2449         }
2450 
2451         vfsp = exi->exi_vp->v_vfsp;
2452         ASSERT(vfsp != NULL);
2453         fidp = (fid_t *)&fh->fh_len;
2454 
2455         error = VFS_VGET(vfsp, &vp, fidp);
2456         if (error || vp == NULL) {
2457                 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2458                     "fhtovp_end:(%S)", "VFS_GET failed or vp NULL");
2459                 return (NULL);
2460         }
2461         TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2462             "fhtovp_end:(%S)", "end");
2463         return (vp);
2464 }
2465 
2466 /*
2467  * Convert an nfs_fh3 into a vnode.
2468  * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2469  * WARNING: users of this routine must do a VN_RELE on the vnode when they
2470  * are done with it.
2471  */
2472 vnode_t *
2473 nfs3_fhtovp(nfs_fh3 *fh, struct exportinfo *exi)
2474 {
2475         vfs_t *vfsp;
2476         vnode_t *vp;
2477         int error;
2478         fid_t *fidp;
2479 
2480         if (exi == NULL)
2481                 return (NULL);  /* not exported */
2482 
2483         ASSERT(exi->exi_vp != NULL);
2484 
2485         if (PUBLIC_FH3(fh)) {
2486                 if (exi->exi_export.ex_flags & EX_PUBLIC)
2487                         return (NULL);
2488                 vp = exi->exi_vp;
2489                 VN_HOLD(vp);
2490                 return (vp);
2491         }
2492 
2493         if (fh->fh3_length < NFS3_OLDFHSIZE ||
2494             fh->fh3_length > NFS3_MAXFHSIZE)
2495                 return (NULL);
2496 
2497         vfsp = exi->exi_vp->v_vfsp;
2498         ASSERT(vfsp != NULL);
2499         fidp = FH3TOFIDP(fh);
2500 
2501         error = VFS_VGET(vfsp, &vp, fidp);
2502         if (error || vp == NULL)
2503                 return (NULL);
2504 
2505         return (vp);
2506 }
2507 
2508 /*
2509  * Convert an nfs_fh4 into a vnode.
2510  * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2511  * WARNING: users of this routine must do a VN_RELE on the vnode when they
2512  * are done with it.
2513  */
2514 vnode_t *
2515 nfs4_fhtovp(nfs_fh4 *fh, struct exportinfo *exi, nfsstat4 *statp)
2516 {
2517         vfs_t *vfsp;
2518         vnode_t *vp = NULL;
2519         int error;
2520         fid_t *fidp;
2521         nfs_fh4_fmt_t *fh_fmtp;
2522 #ifdef VOLATILE_FH_TEST
2523         uint32_t volatile_id = 0;
2524 #endif /* VOLATILE_FH_TEST */
2525 
2526         if (exi == NULL) {
2527                 *statp = NFS4ERR_STALE;
2528                 return (NULL);  /* not exported */
2529         }
2530         ASSERT(exi->exi_vp != NULL);
2531 
2532         /* caller should have checked this */
2533         ASSERT(fh->nfs_fh4_len >= NFS_FH4_LEN);
2534 
2535         fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2536         vfsp = exi->exi_vp->v_vfsp;
2537         ASSERT(vfsp != NULL);
2538         fidp = (fid_t *)&fh_fmtp->fh4_len;
2539 
2540 #ifdef VOLATILE_FH_TEST
2541         /* XXX check if volatile - should be changed later */
2542         if (exi->exi_export.ex_flags & (EX_VOLRNM | EX_VOLFH)) {
2543                 /*
2544                  * Filesystem is shared with volatile filehandles
2545                  */
2546                 if (exi->exi_export.ex_flags & EX_VOLRNM)
2547                         volatile_id = find_volrnm_fh_id(exi, fh);
2548                 else
2549                         volatile_id = exi->exi_volatile_id;
2550 
2551                 if (fh_fmtp->fh4_volatile_id != volatile_id) {
2552                         *statp = NFS4ERR_FHEXPIRED;
2553                         return (NULL);
2554                 }
2555         }
2556         /*
2557          * XXX even if test_volatile_fh false, the fh may contain a
2558          * volatile id if obtained when the test was set.
2559          */
2560         fh_fmtp->fh4_volatile_id = (uchar_t)0;
2561 #endif /* VOLATILE_FH_TEST */
2562 
2563         error = VFS_VGET(vfsp, &vp, fidp);
2564         /*
2565          * If we can not get vp from VFS_VGET, perhaps this is
2566          * an nfs v2/v3/v4 node in an nfsv4 pseudo filesystem.
2567          * Check it out.
2568          */
2569         if (error && PSEUDO(exi))
2570                 error = nfs4_vget_pseudo(exi, &vp, fidp);
2571 
2572         if (error || vp == NULL) {
2573                 *statp = NFS4ERR_STALE;
2574                 return (NULL);
2575         }
2576         /* XXX - disgusting hack */
2577         if (vp->v_type == VNON && vp->v_flag & V_XATTRDIR)
2578                 vp->v_type = VDIR;
2579         *statp = NFS4_OK;
2580         return (vp);
2581 }
2582 
2583 /*
2584  * Find the export structure associated with the given filesystem.
2585  * If found, then increment the ref count (exi_count).
2586  */
2587 struct exportinfo *
2588 checkexport(fsid_t *fsid, fid_t *fid)
2589 {
2590         struct exportinfo *exi;
2591         nfs_export_t *ne = nfs_get_export();
2592 
2593         rw_enter(&ne->exported_lock, RW_READER);
2594         for (exi = ne->exptable[exptablehash(fsid, fid)];
2595             exi != NULL;
2596             exi = exi->fid_hash.next) {
2597                 if (exportmatch(exi, fsid, fid)) {
2598                         /*
2599                          * If this is the place holder for the
2600                          * public file handle, then return the
2601                          * real export entry for the public file
2602                          * handle.
2603                          */
2604                         if (exi->exi_export.ex_flags & EX_PUBLIC) {
2605                                 exi = ne->exi_public;
2606                         }
2607 
2608                         exi_hold(exi);
2609                         rw_exit(&ne->exported_lock);
2610                         return (exi);
2611                 }
2612         }
2613         rw_exit(&ne->exported_lock);
2614         return (NULL);
2615 }
2616 
2617 
2618 /*
2619  * "old school" version of checkexport() for NFS4.  NFS4
2620  * rfs4_compound holds exported_lock for duration of compound
2621  * processing.  This version doesn't manipulate exi_count
2622  * since NFS4 breaks fundamental assumptions in the exi_count
2623  * design.
2624  */
2625 struct exportinfo *
2626 checkexport4(fsid_t *fsid, fid_t *fid, vnode_t *vp)
2627 {
2628         struct exportinfo *exi;
2629         nfs_export_t *ne = nfs_get_export();
2630 
2631         ASSERT(RW_LOCK_HELD(&ne->exported_lock));
2632 
2633         for (exi = ne->exptable[exptablehash(fsid, fid)];
2634             exi != NULL;
2635             exi = exi->fid_hash.next) {
2636                 if (exportmatch(exi, fsid, fid)) {
2637                         /*
2638                          * If this is the place holder for the
2639                          * public file handle, then return the
2640                          * real export entry for the public file
2641                          * handle.
2642                          */
2643                         if (exi->exi_export.ex_flags & EX_PUBLIC) {
2644                                 exi = ne->exi_public;
2645                         }
2646 
2647                         /*
2648                          * If vp is given, check if vp is the
2649                          * same vnode as the exported node.
2650                          *
2651                          * Since VOP_FID of a lofs node returns the
2652                          * fid of its real node (ufs), the exported
2653                          * node for lofs and (pseudo) ufs may have
2654                          * the same fsid and fid.
2655                          */
2656                         if (vp == NULL || vp == exi->exi_vp)
2657                                 return (exi);
2658                 }
2659         }
2660 
2661         return (NULL);
2662 }
2663 
2664 /*
2665  * Free an entire export list node
2666  */
2667 static void
2668 exportfree(struct exportinfo *exi)
2669 {
2670         struct exportdata *ex;
2671         struct charset_cache *cache;
2672         int i;
2673 
2674         ex = &exi->exi_export;
2675 
2676         ASSERT(exi->exi_vp != NULL && !(exi->exi_export.ex_flags & EX_PUBLIC));
2677         VN_RELE(exi->exi_vp);
2678         if (exi->exi_dvp != NULL)
2679                 VN_RELE(exi->exi_dvp);
2680 
2681         if (ex->ex_flags & EX_INDEX)
2682                 kmem_free(ex->ex_index, strlen(ex->ex_index) + 1);
2683 
2684         kmem_free(ex->ex_path, ex->ex_pathlen + 1);
2685         nfsauth_cache_free(exi);
2686 
2687         /*
2688          * if there is a character set mapping cached, clean it up.
2689          */
2690         for (cache = exi->exi_charset; cache != NULL;
2691             cache = exi->exi_charset) {
2692                 if (cache->inbound != (kiconv_t)-1)
2693                         (void) kiconv_close(cache->inbound);
2694                 if (cache->outbound != (kiconv_t)-1)
2695                         (void) kiconv_close(cache->outbound);
2696                 exi->exi_charset = cache->next;
2697                 kmem_free(cache, sizeof (struct charset_cache));
2698         }
2699 
2700         if (exi->exi_logbuffer != NULL)
2701                 nfslog_disable(exi);
2702 
2703         if (ex->ex_flags & EX_LOG) {
2704                 kmem_free(ex->ex_log_buffer, ex->ex_log_bufferlen + 1);
2705                 kmem_free(ex->ex_tag, ex->ex_taglen + 1);
2706         }
2707 
2708         if (exi->exi_visible)
2709                 free_visible(exi->exi_visible);
2710 
2711         srv_secinfo_list_free(ex->ex_secinfo, ex->ex_seccnt);
2712 
2713 #ifdef VOLATILE_FH_TEST
2714         free_volrnm_list(exi);
2715         mutex_destroy(&exi->exi_vol_rename_lock);
2716 #endif /* VOLATILE_FH_TEST */
2717 
2718         mutex_destroy(&exi->exi_lock);
2719         rw_destroy(&exi->exi_cache_lock);
2720         /*
2721          * All nodes in the exi_cache AVL trees were removed and freed in the
2722          * nfsauth_cache_free() call above.  We will just destroy and free the
2723          * empty AVL trees here.
2724          */
2725         for (i = 0; i < AUTH_TABLESIZE; i++) {
2726                 avl_destroy(exi->exi_cache[i]);
2727                 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
2728         }
2729 
2730         exp_kstats_fini(exi->exi_kstats);
2731 
2732         kmem_free(exi, sizeof (*exi));
2733 }
2734 
2735 /*
2736  * load the index file from user space into kernel space.
2737  */
2738 static int
2739 loadindex(struct exportdata *kex)
2740 {
2741         int error;
2742         char index[MAXNAMELEN+1];
2743         size_t len;
2744 
2745         /*
2746          * copyinstr copies the complete string including the NULL and
2747          * returns the len with the NULL byte included in the calculation
2748          * as long as the max length is not exceeded.
2749          */
2750         if (error = copyinstr(kex->ex_index, index, sizeof (index), &len))
2751                 return (error);
2752 
2753         kex->ex_index = kmem_alloc(len, KM_SLEEP);
2754         bcopy(index, kex->ex_index, len);
2755 
2756         return (0);
2757 }
2758 
2759 void
2760 exi_hold(struct exportinfo *exi)
2761 {
2762         mutex_enter(&exi->exi_lock);
2763         exi->exi_count++;
2764         mutex_exit(&exi->exi_lock);
2765 }
2766 
2767 /*
2768  * When a thread completes using exi, it should call exi_rele().
2769  * exi_rele() decrements exi_count. It releases exi if exi_count == 0, i.e.
2770  * if this is the last user of exi and exi is not on exportinfo list anymore
2771  */
2772 void
2773 exi_rele(struct exportinfo **exi)
2774 {
2775         struct exportinfo *exip = *exi;
2776         mutex_enter(&exip->exi_lock);
2777         exip->exi_count--;
2778         if (exip->exi_count == 0) {
2779                 mutex_exit(&exip->exi_lock);
2780                 /*
2781                  * The exportinfo structure needs to be cleared here
2782                  * since the control point, for when we free the structure,
2783                  * is in this function and is triggered by the reference
2784                  * count. The caller does not necessarily know when that
2785                  * will be the case.
2786                  */
2787                 *exi = NULL;
2788                 exportfree(exip);
2789         } else
2790                 mutex_exit(&exip->exi_lock);
2791 }
2792 
2793 #ifdef VOLATILE_FH_TEST
2794 /*
2795  * Test for volatile fh's - add file handle to list and set its volatile id
2796  * to time it was renamed. If EX_VOLFH is also on and the fs is reshared,
2797  * the vol_rename queue is purged.
2798  *
2799  * XXX This code is for unit testing purposes only... To correctly use it, it
2800  * needs to tie a rename list to the export struct and (more
2801  * important), protect access to the exi rename list using a write lock.
2802  */
2803 
2804 /*
2805  * get the fh vol record if it's in the volatile on rename list. Don't check
2806  * volatile_id in the file handle - compare only the file handles.
2807  */
2808 static struct ex_vol_rename *
2809 find_volrnm_fh(struct exportinfo *exi, nfs_fh4 *fh4p)
2810 {
2811         struct ex_vol_rename *p = NULL;
2812         fhandle4_t *fhp;
2813 
2814         /* XXX shouldn't we assert &exported_lock held? */
2815         ASSERT(MUTEX_HELD(&exi->exi_vol_rename_lock));
2816 
2817         if (fh4p->nfs_fh4_len != NFS_FH4_LEN) {
2818                 return (NULL);
2819         }
2820         fhp = &((nfs_fh4_fmt_t *)fh4p->nfs_fh4_val)->fh4_i;
2821         for (p = exi->exi_vol_rename; p != NULL; p = p->vrn_next) {
2822                 if (bcmp(fhp, &p->vrn_fh_fmt.fh4_i,
2823                     sizeof (fhandle4_t)) == 0)
2824                         break;
2825         }
2826         return (p);
2827 }
2828 
2829 /*
2830  * get the volatile id for the fh (if there is - else return 0). Ignore the
2831  * volatile_id in the file handle - compare only the file handles.
2832  */
2833 static uint32_t
2834 find_volrnm_fh_id(struct exportinfo *exi, nfs_fh4 *fh4p)
2835 {
2836         struct ex_vol_rename *p;
2837         uint32_t volatile_id;
2838 
2839         mutex_enter(&exi->exi_vol_rename_lock);
2840         p = find_volrnm_fh(exi, fh4p);
2841         volatile_id = (p ? p->vrn_fh_fmt.fh4_volatile_id :
2842             exi->exi_volatile_id);
2843         mutex_exit(&exi->exi_vol_rename_lock);
2844         return (volatile_id);
2845 }
2846 
2847 /*
2848  * Free the volatile on rename list - will be called if a filesystem is
2849  * unshared or reshared without EX_VOLRNM
2850  */
2851 static void
2852 free_volrnm_list(struct exportinfo *exi)
2853 {
2854         struct ex_vol_rename *p, *pnext;
2855 
2856         /* no need to hold mutex lock - this one is called from exportfree */
2857         for (p = exi->exi_vol_rename; p != NULL; p = pnext) {
2858                 pnext = p->vrn_next;
2859                 kmem_free(p, sizeof (*p));
2860         }
2861         exi->exi_vol_rename = NULL;
2862 }
2863 
2864 /*
2865  * Add a file handle to the volatile on rename list.
2866  */
2867 void
2868 add_volrnm_fh(struct exportinfo *exi, vnode_t *vp)
2869 {
2870         struct ex_vol_rename *p;
2871         char fhbuf[NFS4_FHSIZE];
2872         nfs_fh4 fh4;
2873         int error;
2874 
2875         fh4.nfs_fh4_val = fhbuf;
2876         error = makefh4(&fh4, vp, exi);
2877         if ((error) || (fh4.nfs_fh4_len != sizeof (p->vrn_fh_fmt))) {
2878                 return;
2879         }
2880 
2881         mutex_enter(&exi->exi_vol_rename_lock);
2882 
2883         p = find_volrnm_fh(exi, &fh4);
2884 
2885         if (p == NULL) {
2886                 p = kmem_alloc(sizeof (*p), KM_SLEEP);
2887                 bcopy(fh4.nfs_fh4_val, &p->vrn_fh_fmt, sizeof (p->vrn_fh_fmt));
2888                 p->vrn_next = exi->exi_vol_rename;
2889                 exi->exi_vol_rename = p;
2890         }
2891 
2892         p->vrn_fh_fmt.fh4_volatile_id = gethrestime_sec();
2893         mutex_exit(&exi->exi_vol_rename_lock);
2894 }
2895 
2896 #endif /* VOLATILE_FH_TEST */