xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs_export.c (revision 4de2612967d06c4fdbf524a62556a1e8118a006f)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  *  	Copyright 1983, 1984, 1985, 1986, 1987, 1988, 1989  AT&T.
29  *		All rights reserved.
30  */
31 
32 
33 #pragma ident	"%Z%%M%	%I%	%E% SMI"
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 <netinet/in.h>
55 
56 #include <rpc/types.h>
57 #include <rpc/auth.h>
58 #include <rpc/svc.h>
59 
60 #include <nfs/nfs.h>
61 #include <nfs/export.h>
62 #include <nfs/nfssys.h>
63 #include <nfs/nfs_clnt.h>
64 #include <nfs/nfs_acl.h>
65 #include <nfs/nfs_log.h>
66 #include <nfs/lm.h>
67 
68 #define	EXPTABLESIZE 16
69 
70 struct exportinfo *exptable[EXPTABLESIZE];
71 
72 static int	unexport(fsid_t *, fid_t *, vnode_t *);
73 static void	exportfree(struct exportinfo *);
74 static int	loadindex(struct exportdata *);
75 
76 extern void	nfsauth_cache_free(struct exportinfo *);
77 extern int	sec_svc_loadrootnames(int, int, caddr_t **, model_t);
78 extern void	sec_svc_freerootnames(int, int, caddr_t *);
79 
80 #ifdef VOLATILE_FH_TEST
81 static struct ex_vol_rename *find_volrnm_fh(struct exportinfo *, nfs_fh4 *);
82 static uint32_t find_volrnm_fh_id(struct exportinfo *, nfs_fh4 *);
83 static void free_volrnm_list(struct exportinfo *);
84 #endif /* VOLATILE_FH_TEST */
85 
86 /*
87  * exported_lock	Read/Write lock that protects the exportinfo list.
88  *			This lock must be held when searching or modifiying
89  *			the exportinfo list.
90  */
91 krwlock_t exported_lock;
92 
93 /*
94  * "public" and default (root) location for public filehandle
95  */
96 struct exportinfo *exi_public, *exi_root;
97 
98 fid_t exi_rootfid;	/* for checking the default public file handle */
99 
100 fhandle_t nullfh2;	/* for comparing V2 filehandles */
101 
102 #define	exptablehash(fsid, fid) (nfs_fhhash((fsid), (fid)) & (EXPTABLESIZE - 1))
103 
104 /*
105  * File handle hash function, good for producing hash values 16 bits wide.
106  */
107 int
108 nfs_fhhash(fsid_t *fsid, fid_t *fid)
109 {
110 	short *data;
111 	int i, len;
112 	short h;
113 
114 	ASSERT(fid != NULL);
115 
116 	data = (short *)fid->fid_data;
117 
118 	/* fid_data must be aligned on a short */
119 	ASSERT((((uintptr_t)data) & (sizeof (short) - 1)) == 0);
120 
121 	if (fid->fid_len == 10) {
122 		/*
123 		 * probably ufs: hash on bytes 4,5 and 8,9
124 		 */
125 		return (fsid->val[0] ^ data[2] ^ data[4]);
126 	}
127 
128 	if (fid->fid_len == 6) {
129 		/*
130 		 * probably hsfs: hash on bytes 0,1 and 4,5
131 		 */
132 		return ((fsid->val[0] ^ data[0] ^ data[2]));
133 	}
134 
135 	/*
136 	 * Some other file system. Assume that every byte is
137 	 * worth hashing.
138 	 */
139 	h = (short)fsid->val[0];
140 
141 	/*
142 	 * Sanity check the length before using it
143 	 * blindly in case the client trashed it.
144 	 */
145 	if (fid->fid_len > NFS_FHMAXDATA)
146 		len = 0;
147 	else
148 		len = fid->fid_len / sizeof (short);
149 
150 	/*
151 	 * This will ignore one byte if len is not a multiple of
152 	 * of sizeof (short). No big deal since we at least get some
153 	 * variation with fsid->val[0];
154 	 */
155 	for (i = 0; i < len; i++)
156 		h ^= data[i];
157 
158 	return ((int)h);
159 }
160 
161 /*
162  * Free the memory allocated within a secinfo entry.
163  */
164 void
165 srv_secinfo_entry_free(struct secinfo *secp)
166 {
167 	if (secp->s_rootcnt > 0 && secp->s_rootnames != NULL) {
168 		sec_svc_freerootnames(secp->s_secinfo.sc_rpcnum,
169 				secp->s_rootcnt, secp->s_rootnames);
170 		secp->s_rootcnt = 0;
171 	}
172 
173 	if ((secp->s_secinfo.sc_rpcnum == RPCSEC_GSS) &&
174 	    (secp->s_secinfo.sc_gss_mech_type)) {
175 		kmem_free(secp->s_secinfo.sc_gss_mech_type->elements,
176 			secp->s_secinfo.sc_gss_mech_type->length);
177 		kmem_free(secp->s_secinfo.sc_gss_mech_type,
178 			sizeof (rpc_gss_OID_desc));
179 		secp->s_secinfo.sc_gss_mech_type = NULL;
180 	}
181 
182 }
183 
184 /*
185  * Free a list of secinfo allocated in the exportdata structure.
186  */
187 void
188 srv_secinfo_list_free(struct secinfo *secinfo, int cnt)
189 {
190 	int i;
191 
192 	if (cnt == 0)
193 		return;
194 
195 	for (i = 0; i < cnt; i++)
196 		srv_secinfo_entry_free(&secinfo[i]);
197 
198 	kmem_free(secinfo, cnt * sizeof (struct secinfo));
199 }
200 
201 /*
202  * Allocate and copy a secinfo data from "from" to "to".
203  *
204  * This routine is used by srv_secinfo_add() to add a new flavor to an
205  * ancestor's export node. The rootnames are not copied because the
206  * allowable rootname access only applies to the explicit exported node,
207  * not its ancestor's.
208  *
209  * "to" should have already been allocated and zeroed before calling
210  * this routine.
211  *
212  * This routine is used under the protection of exported_lock (RW_WRITER).
213  */
214 void
215 srv_secinfo_copy(struct secinfo *from, struct secinfo *to)
216 {
217 	to->s_secinfo.sc_nfsnum = from->s_secinfo.sc_nfsnum;
218 	to->s_secinfo.sc_rpcnum = from->s_secinfo.sc_rpcnum;
219 
220 	if (from->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
221 		to->s_secinfo.sc_service = from->s_secinfo.sc_service;
222 		bcopy(from->s_secinfo.sc_name, to->s_secinfo.sc_name,
223 			strlen(from->s_secinfo.sc_name));
224 		bcopy(from->s_secinfo.sc_gss_mech, to->s_secinfo.sc_gss_mech,
225 			strlen(from->s_secinfo.sc_gss_mech));
226 
227 		/* copy mechanism oid */
228 		to->s_secinfo.sc_gss_mech_type =
229 			kmem_alloc(sizeof (rpc_gss_OID_desc), KM_SLEEP);
230 		to->s_secinfo.sc_gss_mech_type->length =
231 			from->s_secinfo.sc_gss_mech_type->length;
232 		to->s_secinfo.sc_gss_mech_type->elements =
233 			kmem_alloc(from->s_secinfo.sc_gss_mech_type->length,
234 					KM_SLEEP);
235 		bcopy(from->s_secinfo.sc_gss_mech_type->elements,
236 			to->s_secinfo.sc_gss_mech_type->elements,
237 			from->s_secinfo.sc_gss_mech_type->length);
238 	}
239 
240 	to->s_refcnt = from->s_refcnt;
241 	to->s_window = from->s_window;
242 	/* no need to copy the mode bits - s_flags */
243 }
244 
245 /*
246  * Add the new security flavors from newdata to the current list, curdata.
247  * Upon return, curdata has the newly merged secinfo list.
248  *
249  * There should be at least 1 secinfo entry in newdata.
250  *
251  * This routine is used under the protection of exported_lock (RW_WRITER).
252  */
253 void
254 srv_secinfo_add(struct exportdata *curdata, struct exportdata *newdata)
255 {
256 	int ccnt, c;		/* sec count in current data - curdata */
257 	int ncnt, n;		/* sec count in new data - newdata */
258 	int tcnt, mcnt;		/* total sec count after merge */
259 	struct secinfo *msec;	/* merged secinfo list */
260 
261 	ccnt = curdata->ex_seccnt;
262 	ncnt = newdata->ex_seccnt;
263 
264 	ASSERT(ncnt > 0);
265 	tcnt = ccnt + ncnt;
266 
267 	for (n = 0; n < ncnt; n++) {
268 		for (c = 0; c < ccnt; c++) {
269 		    if (newdata->ex_secinfo[n].s_secinfo.sc_nfsnum ==
270 			curdata->ex_secinfo[c].s_secinfo.sc_nfsnum) {
271 
272 			/*
273 			 * add the reference count of the newdata
274 			 * to the curdata for this nfs flavor.
275 			 */
276 			curdata->ex_secinfo[c].s_refcnt +=
277 				newdata->ex_secinfo[n].s_refcnt;
278 
279 			tcnt--;
280 			break;
281 		    }
282 		}
283 	}
284 
285 	if (tcnt == ccnt)
286 		return; /* no change; no new flavors */
287 
288 	msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
289 
290 	/* move current secinfo list data to the new list */
291 	for (c = 0; c < ccnt; c++) {
292 
293 		bcopy(&curdata->ex_secinfo[c], &msec[c],
294 			sizeof (struct secinfo));
295 	}
296 
297 	/* Add the flavor that's not in the current data */
298 	mcnt = ccnt;
299 	for (n = 0; n < ncnt; n++) {
300 		for (c = 0; c < ccnt; c++) {
301 		    if (newdata->ex_secinfo[n].s_secinfo.sc_nfsnum ==
302 			curdata->ex_secinfo[c].s_secinfo.sc_nfsnum)
303 				break;
304 		}
305 
306 		/* This is the one. Add it. */
307 		if (c == ccnt) {
308 		    srv_secinfo_copy(&newdata->ex_secinfo[n], &msec[mcnt]);
309 		    if (curdata->ex_flags & EX_PSEUDO)
310 			msec[mcnt].s_flags = M_RO;
311 		    mcnt++;
312 		}
313 	}
314 
315 	ASSERT(mcnt == tcnt);
316 	/*
317 	 * Done. Update curdata.
318 	 * Free up the existing secinfo list in curdata and
319 	 * set the new value.
320 	 */
321 	if (ccnt > 0)
322 		kmem_free(curdata->ex_secinfo, ccnt * sizeof (struct secinfo));
323 	curdata->ex_seccnt = tcnt;
324 	curdata->ex_secinfo = msec;
325 }
326 
327 /*
328  * For NFS V4.
329  * Remove the security data of the unexported node from its ancestors.
330  * Assume there is at least one flavor entry in the current data, curdata.
331  *
332  * This routine is used under the protection of exported_lock (RW_WRITER).
333  */
334 void
335 srv_secinfo_remove(struct exportdata *curdata, struct exportdata *remdata)
336 {
337 	int ccnt, c;		/* sec count in current data - curdata */
338 	int rcnt, r;		/* sec count in removal data - remdata */
339 	int tcnt, mcnt;		/* total sec count after removing */
340 	struct secinfo *msec;	/* final secinfo list after removing */
341 
342 	ASSERT(curdata->ex_seccnt > 0);
343 	ccnt = curdata->ex_seccnt;
344 	rcnt = remdata->ex_seccnt;
345 	tcnt = ccnt;
346 
347 	for (r = 0; r < rcnt; r++) {
348 
349 	    if (SEC_REF_EXPORTED(&remdata->ex_secinfo[r])) {
350 		/*
351 		 * Remove a flavor only if the flavor was a shared flavor for
352 		 * the remdata exported node that's being unshared. Otherwise,
353 		 * this flavor is for the children of remdata, need to keep it.
354 		 */
355 		for (c = 0; c < ccnt; c++) {
356 		    if (remdata->ex_secinfo[r].s_secinfo.sc_nfsnum ==
357 			curdata->ex_secinfo[c].s_secinfo.sc_nfsnum) {
358 
359 			/*
360 			 * Decrement secinfo reference count by 1.
361 			 * If this entry is invalid after decrementing
362 			 * the count (i.e. count < 1), this entry will
363 			 * be removed.
364 			 */
365 			curdata->ex_secinfo[c].s_refcnt--;
366 			if (SEC_REF_INVALID(&curdata->ex_secinfo[c]))
367 				tcnt--;
368 
369 			break;
370 		    }
371 		}
372 	    }
373 	}
374 
375 	ASSERT(tcnt >= 0);
376 	if (tcnt == ccnt)
377 		return; /* no change; no flavors to remove */
378 
379 	if (tcnt == 0) {
380 		srv_secinfo_list_free(curdata->ex_secinfo, ccnt);
381 		curdata->ex_seccnt = 0;
382 		curdata->ex_secinfo = NULL;
383 		return;
384 	}
385 
386 	msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
387 
388 	/* walk thru the given secinfo list to remove the flavors */
389 	mcnt = 0;
390 	for (c = 0; c < ccnt; c++) {
391 
392 		if (SEC_REF_INVALID(&curdata->ex_secinfo[c])) {
393 			srv_secinfo_entry_free(&curdata->ex_secinfo[c]);
394 		} else {
395 			bcopy(&curdata->ex_secinfo[c], &msec[mcnt],
396 				sizeof (struct secinfo));
397 			mcnt++;
398 		}
399 	}
400 
401 	ASSERT(mcnt == tcnt);
402 	/*
403 	 * Done. Update curdata.
404 	 * Free the existing secinfo list in curdata. All pointers
405 	 * within the list have either been moved to msec or freed
406 	 * if it's invalid.
407 	 */
408 	kmem_free(curdata->ex_secinfo, ccnt * sizeof (struct secinfo));
409 	curdata->ex_seccnt = tcnt;
410 	curdata->ex_secinfo = msec;
411 }
412 
413 /*
414  * Upon re-sharing an export node, if there is more than 1 export reference
415  * to an old flavor (i.e. some of its children shared with this flavor), this
416  * flavor information needs to be transfered to the new shared node.
417  *
418  * Expect at least 1 secinfo entry in the old shared node - olddata.
419  * Expect both curdata and olddata are not pseudo nodes.
420  *
421  * This routine is used under the protection of exported_lock (RW_WRITER).
422  */
423 void
424 srv_secinfo_exp2exp(struct exportdata *curdata, struct exportdata *olddata)
425 {
426 	int ccnt, c;		/* sec count in current data - curdata */
427 	int ocnt, o;		/* sec count in old data - olddata */
428 	int tcnt, mcnt;		/* total sec count after the transfer */
429 	struct secinfo *msec;	/* merged secinfo list */
430 
431 	ccnt = curdata->ex_seccnt;
432 	ocnt = olddata->ex_seccnt;
433 
434 	ASSERT(ocnt > 0);
435 	ASSERT(!(olddata->ex_flags & EX_PSEUDO));
436 	ASSERT(!(curdata->ex_flags & EX_PSEUDO));
437 
438 	/*
439 	 * If the olddata has flavors with more than 1 reference count,
440 	 * transfer the information to the curdata.
441 	 */
442 	tcnt = ccnt + ocnt;
443 
444 	for (o = 0; o < ocnt; o++) {
445 
446 	    if (SEC_REF_SELF(&olddata->ex_secinfo[o])) {
447 		tcnt--;
448 	    } else {
449 		for (c = 0; c < ccnt; c++) {
450 		    if (olddata->ex_secinfo[o].s_secinfo.sc_nfsnum ==
451 			curdata->ex_secinfo[c].s_secinfo.sc_nfsnum) {
452 
453 			/* add old reference to the current secinfo count */
454 			curdata->ex_secinfo[c].s_refcnt +=
455 				olddata->ex_secinfo[o].s_refcnt;
456 
457 			/* delete the old export flavor reference */
458 			if (SEC_REF_EXPORTED(&olddata->ex_secinfo[o]))
459 				curdata->ex_secinfo[c].s_refcnt--;
460 			tcnt--;
461 			break;
462 		    }
463 		}
464 	    }
465 	}
466 
467 	if (tcnt == ccnt)
468 		return; /* no more transfer to do */
469 
470 	/*
471 	 * olddata has flavors refered by its children that are not
472 	 * in the current (new) export flavor list. Add these flavors.
473 	 */
474 	msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
475 
476 	/* move current secinfo list data to the new list */
477 	for (c = 0; c < ccnt; c++) {
478 		bcopy(&curdata->ex_secinfo[c], &msec[c],
479 			sizeof (struct secinfo));
480 	}
481 
482 	/*
483 	 * Add the flavor that's not in the new export, but still
484 	 * referred by its children.
485 	 */
486 	mcnt = ccnt;
487 	for (o = 0; o < ocnt; o++) {
488 	    if (! SEC_REF_SELF(&olddata->ex_secinfo[o])) {
489 		for (c = 0; c < ccnt; c++) {
490 		    if (olddata->ex_secinfo[o].s_secinfo.sc_nfsnum ==
491 			curdata->ex_secinfo[c].s_secinfo.sc_nfsnum)
492 				break;
493 		}
494 
495 		/*
496 		 * This is the one. Add it. Decrement the reference count
497 		 * by 1 if the flavor is an explicitly shared flavor for
498 		 * the olddata export node.
499 		 */
500 		if (c == ccnt) {
501 		    srv_secinfo_copy(&olddata->ex_secinfo[o], &msec[mcnt]);
502 		    if (SEC_REF_EXPORTED(&olddata->ex_secinfo[o]))
503 			msec[mcnt].s_refcnt--;
504 		    mcnt++;
505 		}
506 	    }
507 	}
508 
509 	ASSERT(mcnt == tcnt);
510 	/*
511 	 * Done. Update curdata.
512 	 * Free up the existing secinfo list in curdata and
513 	 * set the new value.
514 	 */
515 	if (ccnt > 0)
516 		kmem_free(curdata->ex_secinfo, ccnt * sizeof (struct secinfo));
517 	curdata->ex_seccnt = tcnt;
518 	curdata->ex_secinfo = msec;
519 }
520 
521 /*
522  * When unsharing an old export node and the old node becomes a pseudo node,
523  * if there is more than 1 export reference to an old flavor (i.e. some of
524  * its children shared with this flavor), this flavor information needs to
525  * be transfered to the new shared node.
526  *
527  * This routine is used under the protection of exported_lock (RW_WRITER).
528  */
529 void
530 srv_secinfo_exp2pseu(struct exportdata *curdata, struct exportdata *olddata)
531 {
532 	int ocnt, o;		/* sec count in transfer data - trandata */
533 	int tcnt, mcnt;		/* total sec count after transfer */
534 	struct secinfo *msec;	/* merged secinfo list */
535 
536 	ASSERT(curdata->ex_flags & EX_PSEUDO);
537 	ASSERT(curdata->ex_seccnt == 0);
538 
539 	ocnt = olddata->ex_seccnt;
540 
541 	/*
542 	 * If the olddata has flavors with more than 1 reference count,
543 	 * transfer the information to the curdata.
544 	 */
545 	tcnt = ocnt;
546 
547 	for (o = 0; o < ocnt; o++) {
548 	    if (SEC_REF_SELF(&olddata->ex_secinfo[o]))
549 		tcnt--;
550 	}
551 
552 	if (tcnt == 0)
553 		return; /* no transfer to do */
554 
555 	msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
556 
557 	mcnt = 0;
558 	for (o = 0; o < ocnt; o++) {
559 	    if (! SEC_REF_SELF(&olddata->ex_secinfo[o])) {
560 
561 		/*
562 		 * Decrement the reference count by 1 if the flavor is
563 		 * an explicitly shared flavor for the olddata export node.
564 		 */
565 		srv_secinfo_copy(&olddata->ex_secinfo[o], &msec[mcnt]);
566 		msec[mcnt].s_flags = M_RO; /* for a pseudo node */
567 		if (SEC_REF_EXPORTED(&olddata->ex_secinfo[o]))
568 			msec[mcnt].s_refcnt--;
569 		mcnt++;
570 	    }
571 	}
572 
573 	ASSERT(mcnt == tcnt);
574 	/*
575 	 * Done. Update curdata.
576 	 * Free up the existing secinfo list in curdata and
577 	 * set the new value.
578 	 */
579 	curdata->ex_seccnt = tcnt;
580 	curdata->ex_secinfo = msec;
581 }
582 
583 /*
584  * For NFS V4.
585  * Add or remove the newly exported or unexported security flavors of the
586  * given exportinfo from its ancestors upto the system root.
587  */
588 int
589 srv_secinfo_treeclimb(struct exportinfo *exip, bool_t isadd)
590 {
591 	vnode_t *dvp, *vp;
592 	fid_t fid;
593 	int error = 0;
594 	int exportdir;
595 	struct exportinfo *exi;
596 	struct exportdata *exdata;
597 
598 	ASSERT(RW_WRITE_HELD(&exported_lock));
599 
600 	exdata = &exip->exi_export;
601 	if (exdata->ex_seccnt == 0)
602 		return (0);
603 
604 	vp = exip->exi_vp;
605 	VN_HOLD(vp);
606 	exportdir = 1;
607 
608 	for (;;) {
609 
610 		bzero(&fid, sizeof (fid));
611 		fid.fid_len = MAXFIDSZ;
612 		error = vop_fid_pseudo(vp, &fid);
613 		if (error)
614 			break;
615 
616 		if (! exportdir) {
617 
618 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
619 
620 			if (exi != NULL) {
621 
622 			    if (isadd) {
623 				/*
624 				 * Add the new security flavors to the
625 				 * export entry of the current directory.
626 				 */
627 				srv_secinfo_add(&exi->exi_export, exdata);
628 			    } else {
629 				/*
630 				 * Remove the unexported secinfo entries.
631 				 */
632 				srv_secinfo_remove(&exi->exi_export, exdata);
633 			    }
634 			}
635 		}
636 
637 		/*
638 		 * If at the root of the filesystem, need
639 		 * to traverse across the mountpoint
640 		 * and continue the climb on the mounted-on
641 		 * filesystem.
642 		 */
643 		if (vp->v_flag & VROOT) {
644 
645 			if (VN_CMP(vp, rootdir)) {
646 				/* at system root */
647 				break;
648 			}
649 
650 			vp = untraverse(vp);
651 			exportdir = 0;
652 			continue;
653 		}
654 
655 		/*
656 		 * Now, do a ".." to find parent dir of vp.
657 		 */
658 		error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, CRED());
659 
660 		if (error == ENOTDIR && exportdir) {
661 			dvp = exip->exi_dvp;
662 			ASSERT(dvp != NULL);
663 			VN_HOLD(dvp);
664 			error = 0;
665 		}
666 
667 		if (error)
668 			break;
669 
670 		exportdir = 0;
671 		VN_RELE(vp);
672 		vp = dvp;
673 	}
674 
675 	VN_RELE(vp);
676 	return (error);
677 }
678 
679 void
680 export_link(struct exportinfo *exi) {
681 	int exporthash;
682 
683 	exporthash = exptablehash(&exi->exi_fsid, &exi->exi_fid);
684 	exi->exi_hash = exptable[exporthash];
685 	exptable[exporthash] = exi;
686 }
687 
688 /*
689  * Initialization routine for export routines. Should only be called once.
690  */
691 int
692 nfs_exportinit(void)
693 {
694 	int error;
695 
696 	rw_init(&exported_lock, NULL, RW_DEFAULT, NULL);
697 
698 	/*
699 	 * Allocate the place holder for the public file handle, which
700 	 * is all zeroes. It is initially set to the root filesystem.
701 	 */
702 	exi_root = kmem_zalloc(sizeof (*exi_root), KM_SLEEP);
703 	exi_public = exi_root;
704 
705 	exi_root->exi_export.ex_flags = EX_PUBLIC;
706 	exi_root->exi_export.ex_pathlen = 2;	/* length of "/" */
707 	exi_root->exi_export.ex_path =
708 		kmem_alloc(exi_root->exi_export.ex_pathlen, KM_SLEEP);
709 	exi_root->exi_export.ex_path[0] = '/';
710 	exi_root->exi_export.ex_path[1] = '\0';
711 
712 	exi_root->exi_count = 1;
713 	mutex_init(&exi_root->exi_lock, NULL, MUTEX_DEFAULT, NULL);
714 
715 	exi_root->exi_vp = rootdir;
716 	exi_rootfid.fid_len = MAXFIDSZ;
717 	error = vop_fid_pseudo(exi_root->exi_vp, &exi_rootfid);
718 	if (error) {
719 		mutex_destroy(&exi_root->exi_lock);
720 		kmem_free(exi_root, sizeof (*exi_root));
721 		return (error);
722 	}
723 
724 	/* setup the fhandle template */
725 	exi_root->exi_fh.fh_fsid = rootdir->v_vfsp->vfs_fsid;
726 	exi_root->exi_fh.fh_xlen = exi_rootfid.fid_len;
727 	bcopy(exi_rootfid.fid_data, exi_root->exi_fh.fh_xdata,
728 			exi_rootfid.fid_len);
729 	exi_root->exi_fh.fh_len = sizeof (exi_root->exi_fh.fh_data);
730 
731 	/*
732 	 * Publish the exportinfo in the hash table
733 	 */
734 	export_link(exi_root);
735 
736 	nfslog_init();
737 
738 	return (0);
739 }
740 
741 /*
742  * Finalization routine for export routines. Called to cleanup previoulsy
743  * initializtion work when the NFS server module could not be loaded correctly.
744  */
745 void
746 nfs_exportfini(void)
747 {
748 	/*
749 	 * Deallocate the place holder for the public file handle.
750 	 */
751 	srv_secinfo_list_free(exi_root->exi_export.ex_secinfo,
752 				exi_root->exi_export.ex_seccnt);
753 	mutex_destroy(&exi_root->exi_lock);
754 	kmem_free(exi_root, sizeof (*exi_root));
755 
756 	rw_destroy(&exported_lock);
757 }
758 
759 /*
760  *  Check if 2 gss mechanism identifiers are the same.
761  *
762  *  return FALSE if not the same.
763  *  return TRUE if the same.
764  */
765 static bool_t
766 nfs_mech_equal(rpc_gss_OID mech1, rpc_gss_OID mech2)
767 {
768 	if ((mech1->length == 0) && (mech2->length == 0))
769 		return (TRUE);
770 
771 	if (mech1->length != mech2->length)
772 		return (FALSE);
773 
774 	return (bcmp(mech1->elements, mech2->elements, mech1->length) == 0);
775 }
776 
777 /*
778  *  This routine is used by rpc to map rpc security number
779  *  to nfs specific security flavor number.
780  *
781  *  The gss callback prototype is
782  *  callback(struct svc_req *, gss_cred_id_t *, gss_ctx_id_t *,
783  *				rpc_gss_lock_t *, void **),
784  *  since nfs does not use the gss_cred_id_t/gss_ctx_id_t arguments
785  *  we cast them to void.
786  */
787 /*ARGSUSED*/
788 bool_t
789 rfs_gsscallback(struct svc_req *req, gss_cred_id_t deleg, void *gss_context,
790     rpc_gss_lock_t *lock, void **cookie)
791 {
792 	int i, j;
793 	rpc_gss_rawcred_t *raw_cred;
794 	struct exportinfo *exi;
795 
796 	/*
797 	 * We don't deal with delegated credentials.
798 	 */
799 	if (deleg != GSS_C_NO_CREDENTIAL)
800 		return (FALSE);
801 
802 	raw_cred = lock->raw_cred;
803 	*cookie = NULL;
804 
805 	rw_enter(&exported_lock, RW_READER);
806 	for (i = 0; i < EXPTABLESIZE; i++) {
807 	    exi = exptable[i];
808 	    while (exi) {
809 		if (exi->exi_export.ex_seccnt > 0) {
810 		    struct secinfo *secp;
811 
812 		    secp = exi->exi_export.ex_secinfo;
813 		    for (j = 0; j < exi->exi_export.ex_seccnt; j++) {
814 			/*
815 			 *  If there is a map of the triplet
816 			 *  (mechanism, service, qop) between raw_cred and
817 			 *  the exported flavor, get the psudo flavor number.
818 			 *  Also qop should not be NULL, it should be "default"
819 			 *  or something else.
820 			 */
821 			if ((secp[j].s_secinfo.sc_rpcnum == RPCSEC_GSS) &&
822 			(nfs_mech_equal(secp[j].s_secinfo.sc_gss_mech_type,
823 			raw_cred->mechanism)) &&
824 			(secp[j].s_secinfo.sc_service == raw_cred->service) &&
825 			(raw_cred->qop == secp[j].s_secinfo.sc_qop)) {
826 				*cookie = (void *)(uintptr_t)
827 				    secp[j].s_secinfo.sc_nfsnum;
828 				goto done;
829 			}
830 		    }
831 		}
832 		exi = exi->exi_hash;
833 	    }
834 	}
835 done:
836 	rw_exit(&exported_lock);
837 
838 	/*
839 	 * If no nfs pseudo number mapping can be found in the export
840 	 * table, assign the nfsflavor to NFS_FLAVOR_NOMAP. In V4, we may
841 	 * recover the flavor mismatch from NFS layer (NFS4ERR_WRONGSEC).
842 	 *
843 	 * For example:
844 	 *	server first shares with krb5i;
845 	 *	client mounts with krb5i;
846 	 *	server re-shares with krb5p;
847 	 *	client tries with krb5i, but no mapping can be found;
848 	 *	rpcsec_gss module calls this routine to do the mapping,
849 	 *		if this routine fails, request is rejected from
850 	 *		the rpc layer.
851 	 *	What we need is to let the nfs layer rejects the request.
852 	 *	For V4, we can reject with NFS4ERR_WRONGSEC and the client
853 	 *	may recover from it by getting the new flavor via SECINFO.
854 	 *
855 	 * nfs pseudo number for RPCSEC_GSS mapping (see nfssec.conf)
856 	 * is owned by IANA (see RFC 2623).
857 	 *
858 	 * XXX NFS_FLAVOR_NOMAP is defined in Solaris to work around
859 	 * the implementation issue. This number should not overlap with
860 	 * any new IANA defined pseudo flavor numbers.
861 	 */
862 	if (*cookie == NULL)
863 		*cookie = (void *)NFS_FLAVOR_NOMAP;
864 
865 	lock->locked = TRUE;
866 
867 	return (TRUE);
868 }
869 
870 
871 /*
872  * Exportfs system call; credentials should be checked before
873  * calling this function.
874  */
875 int
876 exportfs(struct exportfs_args *args, model_t model, cred_t *cr)
877 {
878 	vnode_t *vp;
879 	vnode_t *dvp;
880 	struct exportdata *kex;
881 	struct exportinfo *exi;
882 	struct exportinfo *ex, *prev;
883 	fid_t fid;
884 	fsid_t fsid;
885 	int error;
886 	size_t allocsize;
887 	struct secinfo *sp;
888 	struct secinfo *exs;
889 	rpc_gss_callback_t cb;
890 	char *pathbuf;
891 	char *log_buffer;
892 	char *tagbuf;
893 	int callback;
894 	int allocd_seccnt;
895 	STRUCT_HANDLE(exportfs_args, uap);
896 	STRUCT_DECL(exportdata, uexi);
897 	int i;
898 
899 	STRUCT_SET_HANDLE(uap, model, args);
900 
901 	error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
902 	    FOLLOW, &dvp, &vp);
903 	if (error == EINVAL) {
904 		/*
905 		 * if fname resolves to / we get EINVAL error
906 		 * since we wanted the parent vnode. Try again
907 		 * with NULL dvp.
908 		 */
909 		error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
910 		    FOLLOW, NULL, &vp);
911 		dvp = NULL;
912 	}
913 	if (!error && vp == NULL) {
914 		/*
915 		 * Last component of fname not found
916 		 */
917 		if (dvp != NULL) {
918 			VN_RELE(dvp);
919 		}
920 		error = ENOENT;
921 	}
922 	if (error)
923 		return (error);
924 
925 	/*
926 	 * 'vp' may be an AUTOFS node, so we perform a
927 	 * VOP_ACCESS() to trigger the mount of the
928 	 * intended filesystem, so we can share the intended
929 	 * filesystem instead of the AUTOFS filesystem.
930 	 */
931 	(void) VOP_ACCESS(vp, 0, 0, cr);
932 
933 	/*
934 	 * We're interested in the top most filesystem.
935 	 * This is specially important when uap->dname is a trigger
936 	 * AUTOFS node, since we're really interested in sharing the
937 	 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
938 	 * call not the AUTOFS node itself.
939 	 */
940 	if (vn_mountedvfs(vp) != NULL) {
941 		if (error = traverse(&vp)) {
942 			VN_RELE(vp);
943 			if (dvp != NULL)
944 				VN_RELE(dvp);
945 			return (error);
946 		}
947 	}
948 
949 	/*
950 	 * Get the vfs id
951 	 */
952 	bzero(&fid, sizeof (fid));
953 	fid.fid_len = MAXFIDSZ;
954 	error = VOP_FID(vp, &fid);
955 	fsid = vp->v_vfsp->vfs_fsid;
956 	if (error) {
957 		VN_RELE(vp);
958 		if (dvp != NULL)
959 			VN_RELE(dvp);
960 		/*
961 		 * If VOP_FID returns ENOSPC then the fid supplied
962 		 * is too small.  For now we simply return EREMOTE.
963 		 */
964 		if (error == ENOSPC)
965 			error = EREMOTE;
966 		return (error);
967 	}
968 
969 	if (STRUCT_FGETP(uap, uex) == NULL) {
970 		error = unexport(&fsid, &fid, vp);
971 		VN_RELE(vp);
972 		if (dvp != NULL)
973 			VN_RELE(dvp);
974 		return (error);
975 	}
976 	exi = kmem_zalloc(sizeof (*exi), KM_SLEEP);
977 	exi->exi_fsid = fsid;
978 	exi->exi_fid = fid;
979 	exi->exi_vp = vp;
980 	exi->exi_count = 1;
981 	exi->exi_volatile_dev = (vfssw[vp->v_vfsp->vfs_fstype].vsw_flag &
982 				VSW_VOLATILEDEV) ? 1 : 0;
983 	mutex_init(&exi->exi_lock, NULL, MUTEX_DEFAULT, NULL);
984 	exi->exi_dvp = dvp;
985 
986 	/*
987 	 * Initialize auth cache lock
988 	 */
989 	rw_init(&exi->exi_cache_lock, NULL, RW_DEFAULT, NULL);
990 
991 	/*
992 	 * Build up the template fhandle
993 	 */
994 	exi->exi_fh.fh_fsid = fsid;
995 	if (exi->exi_fid.fid_len > sizeof (exi->exi_fh.fh_xdata)) {
996 		error = EREMOTE;
997 		goto out1;
998 	}
999 	exi->exi_fh.fh_xlen = exi->exi_fid.fid_len;
1000 	bcopy(exi->exi_fid.fid_data, exi->exi_fh.fh_xdata,
1001 	    exi->exi_fid.fid_len);
1002 
1003 	exi->exi_fh.fh_len = sizeof (exi->exi_fh.fh_data);
1004 
1005 	kex = &exi->exi_export;
1006 
1007 	/*
1008 	 * Load in everything, and do sanity checking
1009 	 */
1010 	STRUCT_INIT(uexi, model);
1011 	if (copyin(STRUCT_FGETP(uap, uex), STRUCT_BUF(uexi),
1012 	    STRUCT_SIZE(uexi))) {
1013 		error = EFAULT;
1014 		goto out1;
1015 	}
1016 
1017 	kex->ex_version = STRUCT_FGET(uexi, ex_version);
1018 	if (kex->ex_version != EX_CURRENT_VERSION) {
1019 		error = EINVAL;
1020 		cmn_err(CE_WARN,
1021 		"NFS: exportfs requires export struct version 2 - got %d\n",
1022 		kex->ex_version);
1023 		goto out1;
1024 	}
1025 
1026 	/*
1027 	 * Must have at least one security entry
1028 	 */
1029 	kex->ex_seccnt = STRUCT_FGET(uexi, ex_seccnt);
1030 	if (kex->ex_seccnt < 1) {
1031 		error = EINVAL;
1032 		goto out1;
1033 	}
1034 
1035 	kex->ex_path = STRUCT_FGETP(uexi, ex_path);
1036 	kex->ex_pathlen = STRUCT_FGET(uexi, ex_pathlen);
1037 	kex->ex_flags = STRUCT_FGET(uexi, ex_flags);
1038 	kex->ex_anon = STRUCT_FGET(uexi, ex_anon);
1039 	kex->ex_secinfo = STRUCT_FGETP(uexi, ex_secinfo);
1040 	kex->ex_index = STRUCT_FGETP(uexi, ex_index);
1041 	kex->ex_log_buffer = STRUCT_FGETP(uexi, ex_log_buffer);
1042 	kex->ex_log_bufferlen = STRUCT_FGET(uexi, ex_log_bufferlen);
1043 	kex->ex_tag = STRUCT_FGETP(uexi, ex_tag);
1044 	kex->ex_taglen = STRUCT_FGET(uexi, ex_taglen);
1045 
1046 	/*
1047 	 * Copy the exported pathname into
1048 	 * an appropriately sized buffer.
1049 	 */
1050 	pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1051 	if (copyinstr(kex->ex_path, pathbuf, MAXPATHLEN, &kex->ex_pathlen)) {
1052 		kmem_free(pathbuf, MAXPATHLEN);
1053 		error = EFAULT;
1054 		goto out1;
1055 	}
1056 	kex->ex_path = kmem_alloc(kex->ex_pathlen + 1, KM_SLEEP);
1057 	bcopy(pathbuf, kex->ex_path, kex->ex_pathlen);
1058 	kex->ex_path[kex->ex_pathlen] = '\0';
1059 	kmem_free(pathbuf, MAXPATHLEN);
1060 
1061 	/*
1062 	 * Get the path to the logging buffer and the tag
1063 	 */
1064 	if (kex->ex_flags & EX_LOG) {
1065 		log_buffer = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1066 		if (copyinstr(kex->ex_log_buffer, log_buffer, MAXPATHLEN,
1067 		    &kex->ex_log_bufferlen)) {
1068 			kmem_free(log_buffer, MAXPATHLEN);
1069 			error = EFAULT;
1070 			goto out2;
1071 		}
1072 		kex->ex_log_buffer =
1073 			kmem_alloc(kex->ex_log_bufferlen + 1, KM_SLEEP);
1074 		bcopy(log_buffer, kex->ex_log_buffer, kex->ex_log_bufferlen);
1075 		kex->ex_log_buffer[kex->ex_log_bufferlen] = '\0';
1076 		kmem_free(log_buffer, MAXPATHLEN);
1077 
1078 		tagbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1079 		if (copyinstr(kex->ex_tag, tagbuf, MAXPATHLEN,
1080 		    &kex->ex_taglen)) {
1081 			kmem_free(tagbuf, MAXPATHLEN);
1082 			error = EFAULT;
1083 			goto out3;
1084 		}
1085 		kex->ex_tag = kmem_alloc(kex->ex_taglen + 1, KM_SLEEP);
1086 		bcopy(tagbuf, kex->ex_tag, kex->ex_taglen);
1087 		kex->ex_tag[kex->ex_taglen] = '\0';
1088 		kmem_free(tagbuf, MAXPATHLEN);
1089 	}
1090 
1091 	/*
1092 	 * Load the security information for each flavor
1093 	 */
1094 	allocsize = kex->ex_seccnt * SIZEOF_STRUCT(secinfo, model);
1095 	sp = kmem_zalloc(allocsize, KM_SLEEP);
1096 	if (copyin(kex->ex_secinfo, sp, allocsize)) {
1097 		kmem_free(sp, allocsize);
1098 		error = EFAULT;
1099 		goto out4;
1100 	}
1101 
1102 	/*
1103 	 * All of these nested structures need to be converted to
1104 	 * the kernel native format.
1105 	 */
1106 	if (model != DATAMODEL_NATIVE) {
1107 		size_t allocsize2;
1108 		struct secinfo *sp2;
1109 
1110 		allocsize2 = kex->ex_seccnt * sizeof (struct secinfo);
1111 		sp2 = kmem_zalloc(allocsize2, KM_SLEEP);
1112 
1113 		for (i = 0; i < kex->ex_seccnt; i++) {
1114 			STRUCT_HANDLE(secinfo, usi);
1115 
1116 			STRUCT_SET_HANDLE(usi, model,
1117 			    (struct secinfo *)((caddr_t)sp +
1118 			    (i * SIZEOF_STRUCT(secinfo, model))));
1119 			bcopy(STRUCT_FGET(usi, s_secinfo.sc_name),
1120 			    sp2[i].s_secinfo.sc_name, MAX_NAME_LEN);
1121 			sp2[i].s_secinfo.sc_nfsnum =
1122 			    STRUCT_FGET(usi, s_secinfo.sc_nfsnum);
1123 			sp2[i].s_secinfo.sc_rpcnum =
1124 			    STRUCT_FGET(usi, s_secinfo.sc_rpcnum);
1125 			bcopy(STRUCT_FGET(usi, s_secinfo.sc_gss_mech),
1126 			    sp2[i].s_secinfo.sc_gss_mech, MAX_NAME_LEN);
1127 			sp2[i].s_secinfo.sc_gss_mech_type =
1128 			    STRUCT_FGETP(usi, s_secinfo.sc_gss_mech_type);
1129 			sp2[i].s_secinfo.sc_qop =
1130 			    STRUCT_FGET(usi, s_secinfo.sc_qop);
1131 			sp2[i].s_secinfo.sc_service =
1132 			    STRUCT_FGET(usi, s_secinfo.sc_service);
1133 
1134 			sp2[i].s_flags = STRUCT_FGET(usi, s_flags);
1135 			sp2[i].s_window = STRUCT_FGET(usi, s_window);
1136 			sp2[i].s_rootcnt = STRUCT_FGET(usi, s_rootcnt);
1137 			sp2[i].s_rootnames = STRUCT_FGETP(usi, s_rootnames);
1138 		}
1139 		kmem_free(sp, allocsize);
1140 		sp = sp2;
1141 		allocsize = allocsize2;
1142 	}
1143 
1144 	kex->ex_secinfo = sp;
1145 
1146 	/*
1147 	 * And now copy rootnames for each individual secinfo.
1148 	 */
1149 	callback = 0;
1150 	allocd_seccnt = 0;
1151 	while (allocd_seccnt < kex->ex_seccnt) {
1152 
1153 		exs = &sp[allocd_seccnt];
1154 		if (exs->s_rootcnt > 0) {
1155 			if (!sec_svc_loadrootnames(exs->s_secinfo.sc_rpcnum,
1156 			    exs->s_rootcnt, &exs->s_rootnames, model)) {
1157 				error = EFAULT;
1158 				goto out5;
1159 			}
1160 		}
1161 
1162 		if (exs->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
1163 			rpc_gss_OID mech_tmp;
1164 			STRUCT_DECL(rpc_gss_OID_s, umech_tmp);
1165 			caddr_t elements_tmp;
1166 
1167 			/* Copyin mechanism type */
1168 			STRUCT_INIT(umech_tmp, model);
1169 			mech_tmp = kmem_alloc(sizeof (*mech_tmp), KM_SLEEP);
1170 			if (copyin(exs->s_secinfo.sc_gss_mech_type,
1171 			    STRUCT_BUF(umech_tmp), STRUCT_SIZE(umech_tmp))) {
1172 				kmem_free(mech_tmp, sizeof (*mech_tmp));
1173 				error = EFAULT;
1174 				goto out5;
1175 			}
1176 			mech_tmp->length = STRUCT_FGET(umech_tmp, length);
1177 			mech_tmp->elements = STRUCT_FGETP(umech_tmp, elements);
1178 
1179 			elements_tmp = kmem_alloc(mech_tmp->length, KM_SLEEP);
1180 			if (copyin(mech_tmp->elements, elements_tmp,
1181 			    mech_tmp->length)) {
1182 				kmem_free(elements_tmp, mech_tmp->length);
1183 				kmem_free(mech_tmp, sizeof (*mech_tmp));
1184 				error = EFAULT;
1185 				goto out5;
1186 			}
1187 			mech_tmp->elements = elements_tmp;
1188 			exs->s_secinfo.sc_gss_mech_type = mech_tmp;
1189 			allocd_seccnt++;
1190 
1191 			callback = 1;
1192 		} else
1193 			allocd_seccnt++;
1194 	}
1195 
1196 	/*
1197 	 * Init the secinfo reference count and mark these flavors
1198 	 * explicitly exported flavors.
1199 	 */
1200 	for (i = 0; i < kex->ex_seccnt; i++) {
1201 		kex->ex_secinfo[i].s_flags |= M_4SEC_EXPORTED;
1202 		kex->ex_secinfo[i].s_refcnt++;  /* 1 reference count */
1203 	}
1204 
1205 	/*
1206 	 *  Set up rpcsec_gss callback routine entry if any.
1207 	 */
1208 	if (callback) {
1209 		cb.callback = rfs_gsscallback;
1210 		cb.program = NFS_ACL_PROGRAM;
1211 		for (cb.version = NFS_ACL_VERSMIN;
1212 		    cb.version <= NFS_ACL_VERSMAX; cb.version++) {
1213 			(void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1214 			    (void *)&cb);
1215 		}
1216 
1217 		cb.program = NFS_PROGRAM;
1218 		for (cb.version = NFS_VERSMIN;
1219 		    cb.version <= NFS_VERSMAX; cb.version++) {
1220 			(void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1221 			    (void *)&cb);
1222 		}
1223 	}
1224 
1225 	/*
1226 	 * Check the index flag. Do this here to avoid holding the
1227 	 * lock while dealing with the index option (as we do with
1228 	 * the public option).
1229 	 */
1230 	if (kex->ex_flags & EX_INDEX) {
1231 		if (!kex->ex_index) {	/* sanity check */
1232 			error = EINVAL;
1233 			goto out5;
1234 		}
1235 		if (error = loadindex(kex))
1236 			goto out5;
1237 	}
1238 
1239 	if (kex->ex_flags & EX_LOG) {
1240 		if (error = nfslog_setup(exi))
1241 			goto out6;
1242 	}
1243 
1244 	/*
1245 	 * Insert the new entry at the front of the export list
1246 	 */
1247 	rw_enter(&exported_lock, RW_WRITER);
1248 
1249 	export_link(exi);
1250 
1251 	/*
1252 	 * Check the rest of the list for an old entry for the fs.
1253 	 * If one is found then unlink it, wait until this is the
1254 	 * only reference and then free it.
1255 	 */
1256 	prev = exi;
1257 	for (ex = prev->exi_hash; ex != NULL; prev = ex, ex = ex->exi_hash) {
1258 		if (ex != exi_root && VN_CMP(ex->exi_vp, vp)) {
1259 			prev->exi_hash = ex->exi_hash;
1260 			break;
1261 		}
1262 	}
1263 
1264 	/*
1265 	 * If the public filehandle is pointing at the
1266 	 * old entry, then point it back at the root.
1267 	 */
1268 	if (ex != NULL && ex == exi_public)
1269 		exi_public = exi_root;
1270 
1271 	/*
1272 	 * If the public flag is on, make the global exi_public
1273 	 * point to this entry and turn off the public bit so that
1274 	 * we can distinguish it from the place holder export.
1275 	 */
1276 	if (kex->ex_flags & EX_PUBLIC) {
1277 		exi_public = exi;
1278 		kex->ex_flags &= ~EX_PUBLIC;
1279 	}
1280 
1281 #ifdef VOLATILE_FH_TEST
1282 	/*
1283 	 * Set up the volatile_id value if volatile on share.
1284 	 * The list of volatile renamed filehandles is always destroyed,
1285 	 * if the fs was reshared.
1286 	 */
1287 	if (kex->ex_flags & EX_VOLFH)
1288 		exi->exi_volatile_id = gethrestime_sec();
1289 
1290 	mutex_init(&exi->exi_vol_rename_lock, NULL, MUTEX_DEFAULT, NULL);
1291 #endif /* VOLATILE_FH_TEST */
1292 
1293 	/*
1294 	 * If this is a new export, then climb up
1295 	 * the tree and check if any pseudo exports
1296 	 * need to be created to provide a path for
1297 	 * NFS v4 clients.
1298 	 */
1299 	if (ex == NULL)
1300 		error = treeclimb_export(exi);
1301 
1302 	if (!error)
1303 		error = srv_secinfo_treeclimb(exi, TRUE);
1304 
1305 	/*
1306 	 * If re-sharing an old export entry, update the secinfo data
1307 	 * depending on if the old entry is a pseudo node or not.
1308 	 */
1309 	if (!error && ex != NULL) {
1310 		if (PSEUDO(ex)) {
1311 		    srv_secinfo_add(&exi->exi_export, &ex->exi_export);
1312 		} else {
1313 		    srv_secinfo_exp2exp(&exi->exi_export, &ex->exi_export);
1314 		    error = srv_secinfo_treeclimb(ex, FALSE);
1315 		}
1316 	}
1317 
1318 	if (error)
1319 		goto out7;
1320 
1321 	/*
1322 	 * If it's a re-export and the old entry has a visible list,
1323 	 * then transfer its visible list to the new export.
1324 	 * Note: only VROOT node may have a visible list either
1325 	 * it is a PSEUDO node or a real export node.
1326 	 */
1327 	if (ex != NULL && (ex->exi_visible != NULL)) {
1328 		exi->exi_visible = ex->exi_visible;
1329 		ex->exi_visible = NULL;
1330 	}
1331 
1332 	rw_exit(&exported_lock);
1333 
1334 	if (exi_public == exi || kex->ex_flags & EX_LOG) {
1335 		/*
1336 		 * Log share operation to this buffer only.
1337 		 */
1338 		nfslog_share_record(exi, cr);
1339 	}
1340 
1341 	if (ex != NULL)
1342 		exi_rele(ex);
1343 
1344 	return (0);
1345 
1346 out7:
1347 	/*
1348 	 * Cleaning up the tree. Assuming *treeclimb* routines
1349 	 * will fail at the same place in the tree.
1350 	 */
1351 	(void) treeclimb_unexport(exi);
1352 	(void) srv_secinfo_treeclimb(exi, FALSE);
1353 
1354 	/*
1355 	 * Unlink and re-link the new and old export in exptable.
1356 	 */
1357 	(void) export_unlink(&exi->exi_fsid, &exi->exi_fid, exi->exi_vp, NULL);
1358 	if (ex != NULL)
1359 		export_link(ex);
1360 
1361 	rw_exit(&exported_lock);
1362 out6:
1363 	if (kex->ex_flags & EX_INDEX)
1364 		kmem_free(kex->ex_index, strlen(kex->ex_index) + 1);
1365 out5:
1366 	/* free partially completed allocation */
1367 	while (--allocd_seccnt >= 0) {
1368 		exs = &kex->ex_secinfo[allocd_seccnt];
1369 		srv_secinfo_entry_free(exs);
1370 	}
1371 
1372 	if (kex->ex_secinfo) {
1373 		kmem_free(kex->ex_secinfo,
1374 			kex->ex_seccnt * sizeof (struct secinfo));
1375 	}
1376 
1377 out4:
1378 	if ((kex->ex_flags & EX_LOG) && kex->ex_tag != NULL)
1379 		kmem_free(kex->ex_tag, kex->ex_taglen + 1);
1380 out3:
1381 	if ((kex->ex_flags & EX_LOG) && kex->ex_log_buffer != NULL)
1382 		kmem_free(kex->ex_log_buffer, kex->ex_log_bufferlen + 1);
1383 out2:
1384 	kmem_free(kex->ex_path, kex->ex_pathlen + 1);
1385 out1:
1386 	VN_RELE(vp);
1387 	if (dvp != NULL)
1388 		VN_RELE(dvp);
1389 	mutex_destroy(&exi->exi_lock);
1390 	rw_destroy(&exi->exi_cache_lock);
1391 	kmem_free(exi, sizeof (*exi));
1392 	return (error);
1393 }
1394 
1395 /*
1396  * Remove the exportinfo from the export list
1397  */
1398 int
1399 export_unlink(fsid_t *fsid, fid_t *fid, vnode_t *vp, struct exportinfo **exip)
1400 {
1401 	struct exportinfo **tail;
1402 
1403 	ASSERT(RW_WRITE_HELD(&exported_lock));
1404 
1405 	tail = &exptable[exptablehash(fsid, fid)];
1406 	while (*tail != NULL) {
1407 		if (exportmatch(*tail, fsid, fid)) {
1408 			/*
1409 			 * If vp is given, check if vp is the
1410 			 * same vnode as the exported node.
1411 			 *
1412 			 * Since VOP_FID of a lofs node returns the
1413 			 * fid of its real node (ufs), the exported
1414 			 * node for lofs and (pseudo) ufs may have
1415 			 * the same fsid and fid.
1416 			 */
1417 			if (vp == NULL || vp == (*tail)->exi_vp) {
1418 
1419 				if (exip != NULL)
1420 					*exip = *tail;
1421 				*tail = (*tail)->exi_hash;
1422 
1423 				return (0);
1424 			}
1425 		}
1426 		tail = &(*tail)->exi_hash;
1427 	}
1428 
1429 	return (EINVAL);
1430 }
1431 
1432 /*
1433  * Unexport an exported filesystem
1434  */
1435 int
1436 unexport(fsid_t *fsid, fid_t *fid, vnode_t *vp)
1437 {
1438 	struct exportinfo *exi = NULL;
1439 	int error;
1440 
1441 	rw_enter(&exported_lock, RW_WRITER);
1442 
1443 	error = export_unlink(fsid, fid, vp, &exi);
1444 
1445 	if (error) {
1446 		rw_exit(&exported_lock);
1447 		return (error);
1448 	}
1449 
1450 	/* pseudo node is not a real exported filesystem */
1451 	if (PSEUDO(exi)) {
1452 		/*
1453 		 * Put the pseudo node back into the export table
1454 		 * before erroring out.
1455 		 */
1456 		export_link(exi);
1457 		rw_exit(&exported_lock);
1458 		return (EINVAL);
1459 	}
1460 
1461 	/*
1462 	 * If there's a visible list, then need to leave
1463 	 * a pseudo export here to retain the visible list
1464 	 * for paths to exports below.
1465 	 */
1466 	if (exi->exi_visible) {
1467 		error = pseudo_exportfs(exi->exi_vp, exi->exi_visible,
1468 						&exi->exi_export);
1469 		if (error)
1470 			goto done;
1471 
1472 		exi->exi_visible = NULL;
1473 	} else {
1474 		error = treeclimb_unexport(exi);
1475 		if (error)
1476 			goto done;
1477 	}
1478 
1479 	error = srv_secinfo_treeclimb(exi, FALSE);
1480 	if (error)
1481 		goto done;
1482 
1483 	rw_exit(&exported_lock);
1484 
1485 	/*
1486 	 * Need to call into the NFSv4 server and release all data
1487 	 * held on this particular export.  This is important since
1488 	 * the v4 server may be holding file locks or vnodes under
1489 	 * this export.
1490 	 */
1491 	rfs4_clean_state_exi(exi);
1492 
1493 	/*
1494 	 * Notify the lock manager that the filesystem is being
1495 	 * unexported.
1496 	 */
1497 	lm_unexport(exi);
1498 
1499 	/*
1500 	 * If this was a public export, restore
1501 	 * the public filehandle to the root.
1502 	 */
1503 	if (exi == exi_public) {
1504 		exi_public = exi_root;
1505 
1506 		nfslog_share_record(exi_public, CRED());
1507 	}
1508 
1509 	if (exi->exi_export.ex_flags & EX_LOG) {
1510 		nfslog_unshare_record(exi, CRED());
1511 	}
1512 
1513 	exi_rele(exi);
1514 	return (error);
1515 
1516 done:
1517 	rw_exit(&exported_lock);
1518 	exi_rele(exi);
1519 	return (error);
1520 }
1521 
1522 /*
1523  * Get file handle system call.
1524  * Takes file name and returns a file handle for it.
1525  * Credentials must be verified before calling.
1526  */
1527 int
1528 nfs_getfh(struct nfs_getfh_args *args, model_t model, cred_t *cr)
1529 {
1530 	fhandle_t fh;
1531 	vnode_t *vp;
1532 	vnode_t *dvp;
1533 	struct exportinfo *exi;
1534 	int error;
1535 	STRUCT_HANDLE(nfs_getfh_args, uap);
1536 
1537 #ifdef lint
1538 	model = model;		/* STRUCT macros don't always use it */
1539 #endif
1540 
1541 	STRUCT_SET_HANDLE(uap, model, args);
1542 
1543 	error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
1544 	    FOLLOW, &dvp, &vp);
1545 	if (error == EINVAL) {
1546 		/*
1547 		 * if fname resolves to / we get EINVAL error
1548 		 * since we wanted the parent vnode. Try again
1549 		 * with NULL dvp.
1550 		 */
1551 		error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
1552 		    FOLLOW, NULL, &vp);
1553 		dvp = NULL;
1554 	}
1555 	if (!error && vp == NULL) {
1556 		/*
1557 		 * Last component of fname not found
1558 		 */
1559 		if (dvp != NULL) {
1560 			VN_RELE(dvp);
1561 		}
1562 		error = ENOENT;
1563 	}
1564 	if (error)
1565 		return (error);
1566 
1567 	/*
1568 	 * 'vp' may be an AUTOFS node, so we perform a
1569 	 * VOP_ACCESS() to trigger the mount of the
1570 	 * intended filesystem, so we can share the intended
1571 	 * filesystem instead of the AUTOFS filesystem.
1572 	 */
1573 	(void) VOP_ACCESS(vp, 0, 0, cr);
1574 
1575 	/*
1576 	 * We're interested in the top most filesystem.
1577 	 * This is specially important when uap->dname is a trigger
1578 	 * AUTOFS node, since we're really interested in sharing the
1579 	 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
1580 	 * call not the AUTOFS node itself.
1581 	 */
1582 	if (vn_mountedvfs(vp) != NULL) {
1583 		if (error = traverse(&vp)) {
1584 			VN_RELE(vp);
1585 			if (dvp != NULL)
1586 				VN_RELE(dvp);
1587 			return (error);
1588 		}
1589 	}
1590 
1591 	exi = nfs_vptoexi(dvp, vp, cr, NULL, &error, FALSE);
1592 	if (!error) {
1593 		error = makefh(&fh, vp, exi);
1594 		if (!error && exi->exi_export.ex_flags & EX_LOG) {
1595 			nfslog_getfh(exi, &fh, STRUCT_FGETP(uap, fname),
1596 				UIO_USERSPACE, cr);
1597 		}
1598 		exi_rele(exi);
1599 		if (!error) {
1600 			if (copyout(&fh, STRUCT_FGETP(uap, fhp), sizeof (fh)))
1601 				error = EFAULT;
1602 		}
1603 	}
1604 	VN_RELE(vp);
1605 	if (dvp != NULL) {
1606 		VN_RELE(dvp);
1607 	}
1608 	return (error);
1609 }
1610 
1611 /*
1612  * Strategy: if vp is in the export list, then
1613  * return the associated file handle. Otherwise, ".."
1614  * once up the vp and try again, until the root of the
1615  * filesystem is reached.
1616  */
1617 struct   exportinfo *
1618 nfs_vptoexi(vnode_t *dvp, vnode_t *vp, cred_t *cr, int *walk,
1619 	int *err,  bool_t v4srv)
1620 {
1621 	fid_t fid;
1622 	int error;
1623 	struct exportinfo *exi;
1624 
1625 	ASSERT(vp);
1626 	VN_HOLD(vp);
1627 	if (dvp != NULL) {
1628 		VN_HOLD(dvp);
1629 	}
1630 	if (walk != NULL)
1631 		*walk = 0;
1632 
1633 	for (;;) {
1634 		bzero(&fid, sizeof (fid));
1635 		fid.fid_len = MAXFIDSZ;
1636 		error = vop_fid_pseudo(vp, &fid);
1637 		if (error) {
1638 			/*
1639 			 * If vop_fid_pseudo returns ENOSPC then the fid
1640 			 * supplied is too small. For now we simply
1641 			 * return EREMOTE.
1642 			 */
1643 			if (error == ENOSPC)
1644 				error = EREMOTE;
1645 			break;
1646 		}
1647 
1648 		if (v4srv)
1649 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
1650 		else
1651 			exi = checkexport(&vp->v_vfsp->vfs_fsid, &fid);
1652 
1653 		if (exi != NULL) {
1654 			/*
1655 			 * Found the export info
1656 			 */
1657 			break;
1658 		}
1659 
1660 		/*
1661 		 * We have just failed finding a matching export.
1662 		 * If we're at the root of this filesystem, then
1663 		 * it's time to stop (with failure).
1664 		 */
1665 		if (vp->v_flag & VROOT) {
1666 			error = EINVAL;
1667 			break;
1668 		}
1669 
1670 		if (walk != NULL)
1671 			(*walk)++;
1672 
1673 		/*
1674 		 * Now, do a ".." up vp. If dvp is supplied, use it,
1675 		 * otherwise, look it up.
1676 		 */
1677 		if (dvp == NULL) {
1678 			error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, cr);
1679 			if (error)
1680 				break;
1681 		}
1682 		VN_RELE(vp);
1683 		vp = dvp;
1684 		dvp = NULL;
1685 	}
1686 	VN_RELE(vp);
1687 	if (dvp != NULL) {
1688 		VN_RELE(dvp);
1689 	}
1690 	if (error != 0) {
1691 		if (err != NULL)
1692 			*err = error;
1693 		return (NULL);
1694 	}
1695 	return (exi);
1696 }
1697 
1698 bool_t
1699 chk_clnt_sec(struct exportinfo *exi, struct svc_req *req)
1700 {
1701 	int i, nfsflavor;
1702 	struct secinfo *sp;
1703 	bool_t sec_found = FALSE;
1704 
1705 	/*
1706 	 *  Get the nfs flavor number from xprt.
1707 	 */
1708 	nfsflavor = (int)(uintptr_t)req->rq_xprt->xp_cookie;
1709 
1710 	sp = exi->exi_export.ex_secinfo;
1711 	for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
1712 		if (nfsflavor == sp[i].s_secinfo.sc_nfsnum) {
1713 			sec_found = TRUE;
1714 			break;
1715 		}
1716 	}
1717 	return (sec_found);
1718 }
1719 
1720 /*
1721  * Make an fhandle from a vnode
1722  */
1723 int
1724 makefh(fhandle_t *fh, vnode_t *vp, struct exportinfo *exi)
1725 {
1726 	int error;
1727 
1728 	*fh = exi->exi_fh;	/* struct copy */
1729 
1730 	error = VOP_FID(vp, (fid_t *)&fh->fh_len);
1731 	if (error) {
1732 		/*
1733 		 * Should be something other than EREMOTE
1734 		 */
1735 		return (EREMOTE);
1736 	}
1737 	return (0);
1738 }
1739 
1740 /*
1741  * This routine makes an overloaded V2 fhandle which contains
1742  * sec modes.
1743  *
1744  * Note that the first four octets contain the length octet,
1745  * the status octet, and two padded octets to make them XDR
1746  * four-octet aligned.
1747  *
1748  *   1   2   3   4                                          32
1749  * +---+---+---+---+---+---+---+---+   +---+---+---+---+   +---+
1750  * | l | s |   |   |     sec_1     |...|     sec_n     |...|   |
1751  * +---+---+---+---+---+---+---+---+   +---+---+---+---+   +---+
1752  *
1753  * where
1754  *
1755  *   the status octet s indicates whether there are more security
1756  *   flavors (1 means yes, 0 means no) that require the client to
1757  *   perform another 0x81 LOOKUP to get them,
1758  *
1759  *   the length octet l is the length describing the number of
1760  *   valid octets that follow.  (l = 4 * n, where n is the number
1761  *   of security flavors sent in the current overloaded filehandle.)
1762  */
1763 int
1764 makefh_ol(fhandle_t *fh, struct exportinfo *exi, uint_t sec_index)
1765 {
1766 	static int max_cnt = (NFS_FHSIZE/sizeof (int)) - 1;
1767 	int totalcnt, i, *ipt, cnt;
1768 	char *c;
1769 
1770 	if (fh == (fhandle_t *)NULL ||
1771 		exi == (struct exportinfo *)NULL ||
1772 		sec_index > exi->exi_export.ex_seccnt ||
1773 		sec_index < 1)
1774 		return (EREMOTE);
1775 
1776 	totalcnt = exi->exi_export.ex_seccnt-sec_index+1;
1777 	cnt = totalcnt > max_cnt? max_cnt : totalcnt;
1778 
1779 	c = (char *)fh;
1780 	/*
1781 	 * Encode the length octet representing the number of
1782 	 * security flavors (in bytes) in this overloaded fh.
1783 	 */
1784 	*c = cnt * sizeof (int);
1785 
1786 	/*
1787 	 * Encode the status octet that indicates whether there
1788 	 * are more security flavors the client needs to get.
1789 	 */
1790 	*(c+1) = totalcnt > max_cnt;
1791 
1792 	/*
1793 	 * put security flavors in the overloaded fh
1794 	 */
1795 	ipt = (int *)(c + sizeof (int32_t));
1796 	for (i = 0; i < cnt; i++) {
1797 		*ipt++ = htonl(exi->exi_export.ex_secinfo[i+sec_index-1].
1798 				s_secinfo.sc_nfsnum);
1799 	}
1800 	return (0);
1801 }
1802 
1803 /*
1804  * Make an nfs_fh3 from a vnode
1805  */
1806 int
1807 makefh3(nfs_fh3 *fh, vnode_t *vp, struct exportinfo *exi)
1808 {
1809 	int error;
1810 
1811 	fh->fh3_length = sizeof (fh->fh3_u.nfs_fh3_i);
1812 	fh->fh3_u.nfs_fh3_i.fh3_i = exi->exi_fh;	/* struct copy */
1813 
1814 	error = VOP_FID(vp, (fid_t *)&fh->fh3_len);
1815 
1816 	if (error) {
1817 		/*
1818 		 * Should be something other than EREMOTE
1819 		 */
1820 		return (EREMOTE);
1821 	}
1822 	return (0);
1823 }
1824 
1825 /*
1826  * This routine makes an overloaded V3 fhandle which contains
1827  * sec modes.
1828  *
1829  *  1        4
1830  * +--+--+--+--+
1831  * |    len    |
1832  * +--+--+--+--+
1833  *                                               up to 64
1834  * +--+--+--+--+--+--+--+--+--+--+--+--+     +--+--+--+--+
1835  * |s |  |  |  |   sec_1   |   sec_2   | ... |   sec_n   |
1836  * +--+--+--+--+--+--+--+--+--+--+--+--+     +--+--+--+--+
1837  *
1838  * len = 4 * (n+1), where n is the number of security flavors
1839  * sent in the current overloaded filehandle.
1840  *
1841  * the status octet s indicates whether there are more security
1842  * mechanisms (1 means yes, 0 means no) that require the client
1843  * to perform another 0x81 LOOKUP to get them.
1844  *
1845  * Three octets are padded after the status octet.
1846  */
1847 int
1848 makefh3_ol(nfs_fh3 *fh, struct exportinfo *exi, uint_t sec_index)
1849 {
1850 	static int max_cnt = NFS3_FHSIZE/sizeof (int) - 1;
1851 	int totalcnt, cnt, *ipt, i;
1852 	char *c;
1853 
1854 	if (fh == (nfs_fh3 *)NULL ||
1855 		exi == (struct exportinfo *)NULL ||
1856 		sec_index > exi->exi_export.ex_seccnt ||
1857 		sec_index < 1) {
1858 		return (EREMOTE);
1859 	}
1860 
1861 	totalcnt = exi->exi_export.ex_seccnt-sec_index+1;
1862 	cnt = totalcnt > max_cnt? max_cnt : totalcnt;
1863 
1864 	/*
1865 	 * Place the length in fh3_length representing the number
1866 	 * of security flavors (in bytes) in this overloaded fh.
1867 	 */
1868 	fh->fh3_length = (cnt+1) * sizeof (int32_t);
1869 
1870 	c = (char *)&fh->fh3_u.nfs_fh3_i.fh3_i;
1871 	/*
1872 	 * Encode the status octet that indicates whether there
1873 	 * are more security flavors the client needs to get.
1874 	 */
1875 	*c = totalcnt > max_cnt;
1876 
1877 	/*
1878 	 * put security flavors in the overloaded fh
1879 	 */
1880 	ipt = (int *)(c + sizeof (int32_t));
1881 	for (i = 0; i < cnt; i++) {
1882 		*(ipt+i) = htonl(
1883 		exi->exi_export.ex_secinfo[i+sec_index-1].s_secinfo.sc_nfsnum);
1884 	}
1885 	return (0);
1886 }
1887 
1888 /*
1889  * Make an nfs_fh4 from a vnode
1890  */
1891 int
1892 makefh4(nfs_fh4 *fh, vnode_t *vp, struct exportinfo *exi)
1893 {
1894 	int error;
1895 	nfs_fh4_fmt_t *fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
1896 	fid_t fid;
1897 
1898 	bzero(&fid, sizeof (fid));
1899 	fid.fid_len = MAXFIDSZ;
1900 	/*
1901 	 * vop_fid_pseudo() is used to set up NFSv4 namespace, so
1902 	 * use vop_fid_pseudo() here to get the fid instead of VOP_FID.
1903 	 */
1904 	error = vop_fid_pseudo(vp, &fid);
1905 	if (error)
1906 		return (error);
1907 
1908 	fh->nfs_fh4_len = NFS_FH4_LEN;
1909 
1910 	fh_fmtp->fh4_i.fhx_fsid = exi->exi_fh.fh_fsid;
1911 	fh_fmtp->fh4_i.fhx_xlen = exi->exi_fh.fh_xlen;
1912 
1913 	bzero(fh_fmtp->fh4_i.fhx_data, sizeof (fh_fmtp->fh4_i.fhx_data));
1914 	bzero(fh_fmtp->fh4_i.fhx_xdata, sizeof (fh_fmtp->fh4_i.fhx_xdata));
1915 	bcopy(exi->exi_fh.fh_xdata, fh_fmtp->fh4_i.fhx_xdata,
1916 		exi->exi_fh.fh_xlen);
1917 
1918 	fh_fmtp->fh4_len = fid.fid_len;
1919 	ASSERT(fid.fid_len <= sizeof (fh_fmtp->fh4_data));
1920 	bcopy(fid.fid_data, fh_fmtp->fh4_data, fid.fid_len);
1921 	fh_fmtp->fh4_flag = 0;
1922 
1923 #ifdef VOLATILE_FH_TEST
1924 	/*
1925 	 * XXX (temporary?)
1926 	 * Use the rnode volatile_id value to add volatility to the fh.
1927 	 *
1928 	 * For testing purposes there are currently two scenarios, based
1929 	 * on whether the filesystem was shared with "volatile_fh"
1930 	 * or "expire_on_rename". In the first case, use the value of
1931 	 * export struct share_time as the volatile_id. In the second
1932 	 * case use the vnode volatile_id value (which is set to the
1933 	 * time in which the file was renamed).
1934 	 *
1935 	 * Note that the above are temporary constructs for testing only
1936 	 * XXX
1937 	 */
1938 	if (exi->exi_export.ex_flags & EX_VOLRNM) {
1939 		fh_fmtp->fh4_volatile_id = find_volrnm_fh_id(exi, fh);
1940 	} else if (exi->exi_export.ex_flags & EX_VOLFH) {
1941 		fh_fmtp->fh4_volatile_id = exi->exi_volatile_id;
1942 	} else {
1943 		fh_fmtp->fh4_volatile_id = 0;
1944 	}
1945 #endif /* VOLATILE_FH_TEST */
1946 
1947 	return (0);
1948 }
1949 
1950 /*
1951  * Convert an fhandle into a vnode.
1952  * Uses the file id (fh_len + fh_data) in the fhandle to get the vnode.
1953  * WARNING: users of this routine must do a VN_RELE on the vnode when they
1954  * are done with it.
1955  */
1956 vnode_t *
1957 nfs_fhtovp(fhandle_t *fh, struct exportinfo *exi)
1958 {
1959 	vfs_t *vfsp;
1960 	vnode_t *vp;
1961 	int error;
1962 	fid_t *fidp;
1963 
1964 	TRACE_0(TR_FAC_NFS, TR_FHTOVP_START,
1965 		"fhtovp_start");
1966 
1967 	if (exi == NULL) {
1968 		TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
1969 			"fhtovp_end:(%S)", "exi NULL");
1970 		return (NULL);	/* not exported */
1971 	}
1972 
1973 	ASSERT(exi->exi_vp != NULL);
1974 
1975 	if (PUBLIC_FH2(fh)) {
1976 		if (exi->exi_export.ex_flags & EX_PUBLIC) {
1977 			TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
1978 				"fhtovp_end:(%S)", "root not exported");
1979 			return (NULL);
1980 		}
1981 		vp = exi->exi_vp;
1982 		VN_HOLD(vp);
1983 		return (vp);
1984 	}
1985 
1986 	vfsp = exi->exi_vp->v_vfsp;
1987 	ASSERT(vfsp != NULL);
1988 	fidp = (fid_t *)&fh->fh_len;
1989 
1990 	error = VFS_VGET(vfsp, &vp, fidp);
1991 	if (error || vp == NULL) {
1992 		TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
1993 			"fhtovp_end:(%S)", "VFS_GET failed or vp NULL");
1994 		return (NULL);
1995 	}
1996 	TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
1997 		"fhtovp_end:(%S)", "end");
1998 	return (vp);
1999 }
2000 
2001 /*
2002  * Convert an fhandle into a vnode.
2003  * Uses the file id (fh_len + fh_data) in the fhandle to get the vnode.
2004  * WARNING: users of this routine must do a VN_RELE on the vnode when they
2005  * are done with it.
2006  * This is just like nfs_fhtovp() but without the exportinfo argument.
2007  */
2008 
2009 vnode_t *
2010 lm_fhtovp(fhandle_t *fh)
2011 {
2012 	register vfs_t *vfsp;
2013 	vnode_t *vp;
2014 	int error;
2015 
2016 	vfsp = getvfs(&fh->fh_fsid);
2017 	if (vfsp == NULL)
2018 		return (NULL);
2019 
2020 	error = VFS_VGET(vfsp, &vp, (fid_t *)&(fh->fh_len));
2021 	VFS_RELE(vfsp);
2022 	if (error || vp == NULL)
2023 		return (NULL);
2024 
2025 	return (vp);
2026 }
2027 
2028 /*
2029  * Convert an nfs_fh3 into a vnode.
2030  * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2031  * WARNING: users of this routine must do a VN_RELE on the vnode when they
2032  * are done with it.
2033  */
2034 vnode_t *
2035 nfs3_fhtovp(nfs_fh3 *fh, struct exportinfo *exi)
2036 {
2037 	vfs_t *vfsp;
2038 	vnode_t *vp;
2039 	int error;
2040 	fid_t *fidp;
2041 
2042 	if (exi == NULL)
2043 		return (NULL);	/* not exported */
2044 
2045 	ASSERT(exi->exi_vp != NULL);
2046 
2047 	if (PUBLIC_FH3(fh)) {
2048 		if (exi->exi_export.ex_flags & EX_PUBLIC)
2049 			return (NULL);
2050 		vp = exi->exi_vp;
2051 		VN_HOLD(vp);
2052 		return (vp);
2053 	}
2054 
2055 	if (fh->fh3_length != NFS3_CURFHSIZE)
2056 		return (NULL);
2057 
2058 	vfsp = exi->exi_vp->v_vfsp;
2059 	ASSERT(vfsp != NULL);
2060 	fidp = (fid_t *)&fh->fh3_len;
2061 
2062 	error = VFS_VGET(vfsp, &vp, fidp);
2063 	if (error || vp == NULL)
2064 		return (NULL);
2065 
2066 	return (vp);
2067 }
2068 
2069 /*
2070  * Convert an nfs_fh3 into a vnode.
2071  * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2072  * WARNING: users of this routine must do a VN_RELE on the vnode when they
2073  * are done with it.
2074  * BTW: This is just like nfs3_fhtovp() but without the exportinfo arg.
2075  * Also, vfsp is accessed through getvfs() rather using exportinfo !!
2076  */
2077 
2078 vnode_t *
2079 lm_nfs3_fhtovp(nfs_fh3 *fh)
2080 {
2081 	vfs_t *vfsp;
2082 	vnode_t *vp;
2083 	int error;
2084 
2085 	if (fh->fh3_length != NFS3_CURFHSIZE)
2086 		return (NULL);
2087 
2088 	vfsp = getvfs(&fh->fh3_fsid);
2089 	if (vfsp == NULL)
2090 		return (NULL);
2091 
2092 	error = VFS_VGET(vfsp, &vp, (fid_t *)&(fh->fh3_len));
2093 	VFS_RELE(vfsp);
2094 	if (error || vp == NULL)
2095 		return (NULL);
2096 
2097 	return (vp);
2098 }
2099 
2100 /*
2101  * Convert an nfs_fh4 into a vnode.
2102  * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2103  * WARNING: users of this routine must do a VN_RELE on the vnode when they
2104  * are done with it.
2105  */
2106 vnode_t *
2107 nfs4_fhtovp(nfs_fh4 *fh, struct exportinfo *exi, nfsstat4 *statp)
2108 {
2109 	vfs_t *vfsp;
2110 	vnode_t *vp = NULL;
2111 	int error;
2112 	fid_t *fidp;
2113 	nfs_fh4_fmt_t *fh_fmtp;
2114 #ifdef VOLATILE_FH_TEST
2115 	uint32_t volatile_id = 0;
2116 #endif /* VOLATILE_FH_TEST */
2117 
2118 	if (exi == NULL) {
2119 		*statp = NFS4ERR_STALE;
2120 		return (NULL);	/* not exported */
2121 	}
2122 	ASSERT(exi->exi_vp != NULL);
2123 
2124 	/* caller should have checked this */
2125 	ASSERT(fh->nfs_fh4_len >= NFS_FH4_LEN);
2126 
2127 	fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2128 	vfsp = exi->exi_vp->v_vfsp;
2129 	ASSERT(vfsp != NULL);
2130 	fidp = (fid_t *)&fh_fmtp->fh4_len;
2131 
2132 #ifdef VOLATILE_FH_TEST
2133 	/* XXX check if volatile - should be changed later */
2134 	if (exi->exi_export.ex_flags & (EX_VOLRNM | EX_VOLFH)) {
2135 		/*
2136 		 * Filesystem is shared with volatile filehandles
2137 		 */
2138 		if (exi->exi_export.ex_flags & EX_VOLRNM)
2139 			volatile_id = find_volrnm_fh_id(exi, fh);
2140 		else
2141 			volatile_id = exi->exi_volatile_id;
2142 
2143 		if (fh_fmtp->fh4_volatile_id != volatile_id) {
2144 			*statp = NFS4ERR_FHEXPIRED;
2145 			return (NULL);
2146 		}
2147 	}
2148 	/*
2149 	 * XXX even if test_volatile_fh false, the fh may contain a
2150 	 * volatile id if obtained when the test was set.
2151 	 */
2152 	fh_fmtp->fh4_volatile_id = (uchar_t)0;
2153 #endif /* VOLATILE_FH_TEST */
2154 
2155 	error = VFS_VGET(vfsp, &vp, fidp);
2156 	/*
2157 	 * If we can not get vp from VFS_VGET, perhaps this is
2158 	 * an nfs v2/v3/v4 node in an nfsv4 pseudo filesystem.
2159 	 * Check it out.
2160 	 */
2161 	if (error && PSEUDO(exi))
2162 		error = nfs4_vget_pseudo(exi, &vp, fidp);
2163 
2164 	if (error || vp == NULL) {
2165 		*statp = NFS4ERR_STALE;
2166 		return (NULL);
2167 	}
2168 	/* XXX - disgusting hack */
2169 	if (vp->v_type == VNON && vp->v_flag & V_XATTRDIR)
2170 		vp->v_type = VDIR;
2171 	*statp = NFS4_OK;
2172 	return (vp);
2173 }
2174 
2175 /*
2176  * Find the export structure associated with the given filesystem.
2177  * If found, then increment the ref count (exi_count).
2178  */
2179 struct exportinfo *
2180 checkexport(fsid_t *fsid, fid_t *fid)
2181 {
2182 	struct exportinfo *exi;
2183 
2184 	rw_enter(&exported_lock, RW_READER);
2185 	for (exi = exptable[exptablehash(fsid, fid)];
2186 	    exi != NULL;
2187 	    exi = exi->exi_hash) {
2188 		if (exportmatch(exi, fsid, fid)) {
2189 			/*
2190 			 * If this is the place holder for the
2191 			 * public file handle, then return the
2192 			 * real export entry for the public file
2193 			 * handle.
2194 			 */
2195 			if (exi->exi_export.ex_flags & EX_PUBLIC) {
2196 				exi = exi_public;
2197 			}
2198 			mutex_enter(&exi->exi_lock);
2199 			exi->exi_count++;
2200 			mutex_exit(&exi->exi_lock);
2201 			rw_exit(&exported_lock);
2202 			return (exi);
2203 		}
2204 	}
2205 	rw_exit(&exported_lock);
2206 	return (NULL);
2207 }
2208 
2209 
2210 /*
2211  * "old school" version of checkexport() for NFS4.  NFS4
2212  * rfs4_compound holds exported_lock for duration of compound
2213  * processing.  This version doesn't manipulate exi_count
2214  * since NFS4 breaks fundamental assumptions in the exi_count
2215  * design.
2216  */
2217 struct exportinfo *
2218 checkexport4(fsid_t *fsid, fid_t *fid, vnode_t *vp)
2219 {
2220 	struct exportinfo *exi;
2221 
2222 	ASSERT(RW_LOCK_HELD(&exported_lock));
2223 
2224 	for (exi = exptable[exptablehash(fsid, fid)];
2225 	    exi != NULL;
2226 	    exi = exi->exi_hash) {
2227 		if (exportmatch(exi, fsid, fid)) {
2228 			/*
2229 			 * If this is the place holder for the
2230 			 * public file handle, then return the
2231 			 * real export entry for the public file
2232 			 * handle.
2233 			 */
2234 			if (exi->exi_export.ex_flags & EX_PUBLIC) {
2235 				exi = exi_public;
2236 			}
2237 
2238 			/*
2239 			 * If vp is given, check if vp is the
2240 			 * same vnode as the exported node.
2241 			 *
2242 			 * Since VOP_FID of a lofs node returns the
2243 			 * fid of its real node (ufs), the exported
2244 			 * node for lofs and (pseudo) ufs may have
2245 			 * the same fsid and fid.
2246 			 */
2247 			if (vp == NULL || vp == exi->exi_vp)
2248 				return (exi);
2249 		}
2250 	}
2251 
2252 	return (NULL);
2253 }
2254 
2255 /*
2256  * Free an entire export list node
2257  */
2258 void
2259 exportfree(struct exportinfo *exi)
2260 {
2261 	struct exportdata *ex;
2262 
2263 	ex = &exi->exi_export;
2264 
2265 	ASSERT(exi->exi_vp != NULL && !(exi->exi_export.ex_flags & EX_PUBLIC));
2266 	VN_RELE(exi->exi_vp);
2267 	if (exi->exi_dvp != NULL)
2268 		VN_RELE(exi->exi_dvp);
2269 
2270 	if (ex->ex_flags & EX_INDEX)
2271 		kmem_free(ex->ex_index, strlen(ex->ex_index) + 1);
2272 
2273 	kmem_free(ex->ex_path, ex->ex_pathlen + 1);
2274 	nfsauth_cache_free(exi);
2275 
2276 	if (exi->exi_logbuffer != NULL)
2277 		nfslog_disable(exi);
2278 
2279 	if (ex->ex_flags & EX_LOG) {
2280 		kmem_free(ex->ex_log_buffer, ex->ex_log_bufferlen + 1);
2281 		kmem_free(ex->ex_tag, ex->ex_taglen + 1);
2282 	}
2283 
2284 	if (exi->exi_visible)
2285 		free_visible(exi->exi_visible);
2286 
2287 	srv_secinfo_list_free(ex->ex_secinfo, ex->ex_seccnt);
2288 
2289 #ifdef VOLATILE_FH_TEST
2290 	free_volrnm_list(exi);
2291 	mutex_destroy(&exi->exi_vol_rename_lock);
2292 #endif /* VOLATILE_FH_TEST */
2293 
2294 	mutex_destroy(&exi->exi_lock);
2295 	rw_destroy(&exi->exi_cache_lock);
2296 
2297 	kmem_free(exi, sizeof (*exi));
2298 }
2299 
2300 /*
2301  * load the index file from user space into kernel space.
2302  */
2303 static int
2304 loadindex(struct exportdata *kex)
2305 {
2306 	int error;
2307 	char index[MAXNAMELEN+1];
2308 	size_t len;
2309 
2310 	/*
2311 	 * copyinstr copies the complete string including the NULL and
2312 	 * returns the len with the NULL byte included in the calculation
2313 	 * as long as the max length is not exceeded.
2314 	 */
2315 	if (error = copyinstr(kex->ex_index, index, sizeof (index), &len))
2316 		return (error);
2317 
2318 	kex->ex_index = kmem_alloc(len, KM_SLEEP);
2319 	bcopy(index, kex->ex_index, len);
2320 
2321 	return (0);
2322 }
2323 
2324 /*
2325  * When a thread completes using exi, it should call exi_rele().
2326  * exi_rele() decrements exi_count. It releases exi if exi_count == 0, i.e.
2327  * if this is the last user of exi and exi is not on exportinfo list anymore
2328  */
2329 void
2330 exi_rele(struct exportinfo *exi)
2331 {
2332 	mutex_enter(&exi->exi_lock);
2333 	exi->exi_count--;
2334 	if (exi->exi_count == 0) {
2335 		mutex_exit(&exi->exi_lock);
2336 		exportfree(exi);
2337 	} else
2338 		mutex_exit(&exi->exi_lock);
2339 }
2340 
2341 #ifdef VOLATILE_FH_TEST
2342 /*
2343  * Test for volatile fh's - add file handle to list and set its volatile id
2344  * to time it was renamed. If EX_VOLFH is also on and the fs is reshared,
2345  * the vol_rename queue is purged.
2346  *
2347  * XXX This code is for unit testing purposes only... To correctly use it, it
2348  * needs to tie a rename list to the export struct and (more
2349  * important), protect access to the exi rename list using a write lock.
2350  */
2351 
2352 /*
2353  * get the fh vol record if it's in the volatile on rename list. Don't check
2354  * volatile_id in the file handle - compare only the file handles.
2355  */
2356 static struct ex_vol_rename *
2357 find_volrnm_fh(struct exportinfo *exi, nfs_fh4 *fh4p)
2358 {
2359 	struct ex_vol_rename *p = NULL;
2360 	fhandle_ext_t *fhp;
2361 
2362 	/* XXX shouldn't we assert &exported_lock held? */
2363 	ASSERT(MUTEX_HELD(&exi->exi_vol_rename_lock));
2364 
2365 	if (fh4p->nfs_fh4_len != NFS_FH4_LEN) {
2366 		return (NULL);
2367 	}
2368 	fhp = &((nfs_fh4_fmt_t *)fh4p->nfs_fh4_val)->fh4_i;
2369 	for (p = exi->exi_vol_rename; p != NULL; p = p->vrn_next) {
2370 		if (bcmp(fhp, &p->vrn_fh_fmt.fh4_i,
2371 		    sizeof (fhandle_ext_t)) == 0)
2372 			break;
2373 	}
2374 	return (p);
2375 }
2376 
2377 /*
2378  * get the volatile id for the fh (if there is - else return 0). Ignore the
2379  * volatile_id in the file handle - compare only the file handles.
2380  */
2381 static uint32_t
2382 find_volrnm_fh_id(struct exportinfo *exi, nfs_fh4 *fh4p)
2383 {
2384 	struct ex_vol_rename *p;
2385 	uint32_t volatile_id;
2386 
2387 	mutex_enter(&exi->exi_vol_rename_lock);
2388 	p = find_volrnm_fh(exi, fh4p);
2389 	volatile_id = (p ? p->vrn_fh_fmt.fh4_volatile_id :
2390 				exi->exi_volatile_id);
2391 	mutex_exit(&exi->exi_vol_rename_lock);
2392 	return (volatile_id);
2393 }
2394 
2395 /*
2396  * Free the volatile on rename list - will be called if a filesystem is
2397  * unshared or reshared without EX_VOLRNM
2398  */
2399 static void
2400 free_volrnm_list(struct exportinfo *exi)
2401 {
2402 	struct ex_vol_rename *p, *pnext;
2403 
2404 	/* no need to hold mutex lock - this one is called from exportfree */
2405 	for (p = exi->exi_vol_rename; p != NULL; p = pnext) {
2406 		pnext = p->vrn_next;
2407 		kmem_free(p, sizeof (*p));
2408 	}
2409 	exi->exi_vol_rename = NULL;
2410 }
2411 
2412 /*
2413  * Add a file handle to the volatile on rename list.
2414  */
2415 void
2416 add_volrnm_fh(struct exportinfo *exi, vnode_t *vp)
2417 {
2418 	struct ex_vol_rename *p;
2419 	char fhbuf[NFS4_FHSIZE];
2420 	nfs_fh4 fh4;
2421 	int error;
2422 
2423 	fh4.nfs_fh4_val = fhbuf;
2424 	error = makefh4(&fh4, vp, exi);
2425 	if ((error) || (fh4.nfs_fh4_len != sizeof (p->vrn_fh_fmt))) {
2426 		return;
2427 	}
2428 
2429 	mutex_enter(&exi->exi_vol_rename_lock);
2430 
2431 	p = find_volrnm_fh(exi, &fh4);
2432 
2433 	if (p == NULL) {
2434 		p = kmem_alloc(sizeof (*p), KM_SLEEP);
2435 		bcopy(fh4.nfs_fh4_val, &p->vrn_fh_fmt, sizeof (p->vrn_fh_fmt));
2436 		p->vrn_next = exi->exi_vol_rename;
2437 		exi->exi_vol_rename = p;
2438 	}
2439 
2440 	p->vrn_fh_fmt.fh4_volatile_id = gethrestime_sec();
2441 	mutex_exit(&exi->exi_vol_rename_lock);
2442 }
2443 
2444 #endif /* VOLATILE_FH_TEST */
2445