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