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 /*
72 * exi_id support
73 *
74 * exi_id_next The next exi_id available.
75 * exi_id_overflow The exi_id_next already overflowed, so we should
76 * thoroughly check for duplicates.
77 * exi_id_tree AVL tree indexed by exi_id.
78 * nfs_exi_id_lock Lock to protect the export ID list
79 *
80 * All exi_id_next, exi_id_overflow, and exi_id_tree are protected by
81 * nfs_exi_id_lock.
82 */
83 static int exi_id_next;
84 static bool_t exi_id_overflow;
85 avl_tree_t exi_id_tree;
86 kmutex_t nfs_exi_id_lock;
87
88 static int unexport(nfs_export_t *, exportinfo_t *);
89 static void exportfree(exportinfo_t *);
90 static int loadindex(exportdata_t *);
91
92 extern void nfsauth_cache_free(exportinfo_t *);
93 extern int sec_svc_loadrootnames(int, int, caddr_t **, model_t);
94 extern void sec_svc_freerootnames(int, int, caddr_t *);
95
96 static int build_seclist_nodups(exportdata_t *, secinfo_t *, int);
97 static void srv_secinfo_add(secinfo_t **, int *, secinfo_t *, int, int);
98 static void srv_secinfo_remove(secinfo_t **, int *, secinfo_t *, int);
99 static void srv_secinfo_treeclimb(nfs_export_t *, exportinfo_t *,
100 secinfo_t *, int, bool_t);
101
102 #ifdef VOLATILE_FH_TEST
103 static struct ex_vol_rename *find_volrnm_fh(exportinfo_t *, nfs_fh4 *);
104 static uint32_t find_volrnm_fh_id(exportinfo_t *, nfs_fh4 *);
105 static void free_volrnm_list(exportinfo_t *);
106 #endif /* VOLATILE_FH_TEST */
107
108 fhandle_t nullfh2; /* for comparing V2 filehandles */
109
110 /*
111 * macro for static dtrace probes to trace server namespace ref count mods.
112 */
113 #define SECREF_TRACE(seclist, tag, flav, aftcnt) \
114 DTRACE_PROBE4(nfss__i__nmspc__secref, struct secinfo *, (seclist), \
115 char *, (tag), int, (int)(flav), int, (int)(aftcnt))
116
117
118 #define exptablehash(fsid, fid) (nfs_fhhash((fsid), (fid)) & (EXPTABLESIZE - 1))
119
120 extern nfs_export_t *
121 nfs_get_export(void)
122 {
123 nfs_globals_t *ng = zone_getspecific(nfssrv_zone_key, curzone);
124 nfs_export_t *ne = ng->nfs_export;
125 ASSERT(ne != NULL);
126 return (ne);
127 }
128
129 static uint8_t
130 xor_hash(uint8_t *data, int len)
131 {
132 uint8_t h = 0;
133
134 while (len--)
135 h ^= *data++;
136
137 return (h);
138 }
139
140 /*
141 * File handle hash function, XOR over all bytes in fsid and fid.
142 */
143 static unsigned
144 nfs_fhhash(fsid_t *fsid, fid_t *fid)
145 {
146 int len;
147 uint8_t h;
148
149 h = xor_hash((uint8_t *)fsid, sizeof (fsid_t));
150
151 /*
152 * Sanity check the length before using it
153 * blindly in case the client trashed it.
154 */
155 len = fid->fid_len > NFS_FH4MAXDATA ? 0 : fid->fid_len;
156 h ^= xor_hash((uint8_t *)fid->fid_data, len);
157
158 return ((unsigned)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 * Free a list of secinfo allocated in the exportdata structure.
185 */
186 void
187 srv_secinfo_list_free(struct secinfo *secinfo, int cnt)
188 {
189 int i;
190
191 if (cnt == 0)
192 return;
193
194 for (i = 0; i < cnt; i++)
195 srv_secinfo_entry_free(&secinfo[i]);
196
197 kmem_free(secinfo, cnt * sizeof (struct secinfo));
198 }
199
200 /*
201 * Allocate and copy a secinfo data from "from" to "to".
202 *
203 * This routine is used by srv_secinfo_add() to add a new flavor to an
204 * ancestor's export node. The rootnames are not copied because the
205 * allowable rootname access only applies to the explicit exported node,
206 * not its ancestor's.
207 *
208 * "to" should have already been allocated and zeroed before calling
209 * this routine.
210 *
211 * This routine is used under the protection of exported_lock (RW_WRITER).
212 */
213 void
214 srv_secinfo_copy(struct secinfo *from, struct secinfo *to)
215 {
216 to->s_secinfo.sc_nfsnum = from->s_secinfo.sc_nfsnum;
217 to->s_secinfo.sc_rpcnum = from->s_secinfo.sc_rpcnum;
218
219 if (from->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
220 to->s_secinfo.sc_service = from->s_secinfo.sc_service;
221 bcopy(from->s_secinfo.sc_name, to->s_secinfo.sc_name,
222 strlen(from->s_secinfo.sc_name));
223 bcopy(from->s_secinfo.sc_gss_mech, to->s_secinfo.sc_gss_mech,
224 strlen(from->s_secinfo.sc_gss_mech));
225
226 /* copy mechanism oid */
227 to->s_secinfo.sc_gss_mech_type =
228 kmem_alloc(sizeof (rpc_gss_OID_desc), KM_SLEEP);
229 to->s_secinfo.sc_gss_mech_type->length =
230 from->s_secinfo.sc_gss_mech_type->length;
231 to->s_secinfo.sc_gss_mech_type->elements =
232 kmem_alloc(from->s_secinfo.sc_gss_mech_type->length,
233 KM_SLEEP);
234 bcopy(from->s_secinfo.sc_gss_mech_type->elements,
235 to->s_secinfo.sc_gss_mech_type->elements,
236 from->s_secinfo.sc_gss_mech_type->length);
237 }
238
239 to->s_refcnt = from->s_refcnt;
240 to->s_window = from->s_window;
241 /* no need to copy the mode bits - s_flags */
242 }
243
244 /*
245 * Create a secinfo array without duplicates. The condensed
246 * flavor list is used to propagate flavor ref counts to an
247 * export's ancestor pseudonodes.
248 */
249 static int
250 build_seclist_nodups(exportdata_t *exd, secinfo_t *nodups, int exponly)
251 {
252 int ccnt, c;
253 int ncnt, n;
254 struct secinfo *cursec;
255
256 ncnt = 0;
257 ccnt = exd->ex_seccnt;
258 cursec = exd->ex_secinfo;
259
260 for (c = 0; c < ccnt; c++) {
261
262 if (exponly && ! SEC_REF_EXPORTED(&cursec[c]))
263 continue;
264
265 for (n = 0; n < ncnt; n++) {
266 if (nodups[n].s_secinfo.sc_nfsnum ==
267 cursec[c].s_secinfo.sc_nfsnum)
268 break;
269 }
270
271 /*
272 * The structure copy below also copys ptrs embedded
273 * within struct secinfo. The ptrs are copied but
274 * they are never freed from the nodups array. If
275 * an ancestor's secinfo array doesn't contain one
276 * of the nodups flavors, then the entry is properly
277 * copied into the ancestor's secinfo array.
278 * (see srv_secinfo_copy)
279 */
280 if (n == ncnt) {
281 nodups[n] = cursec[c];
282 ncnt++;
283 }
284 }
285 return (ncnt);
286 }
287
288 /*
289 * Add the new security flavors from newdata to the current list, pcursec.
290 * Upon return, *pcursec has the newly merged secinfo list.
291 *
292 * There should be at least 1 secinfo entry in newsec.
293 *
294 * This routine is used under the protection of exported_lock (RW_WRITER).
295 */
296 static void
297 srv_secinfo_add(secinfo_t **pcursec, int *pcurcnt, secinfo_t *newsec,
298 int newcnt, int is_pseudo)
299 {
300 int ccnt, c; /* sec count in current data - curdata */
301 int n; /* index for newsec - newsecinfo */
302 int tcnt; /* total sec count after merge */
303 int mcnt; /* total sec count after merge */
304 struct secinfo *msec; /* merged secinfo list */
305 struct secinfo *cursec;
306
307 cursec = *pcursec;
308 ccnt = *pcurcnt;
309
310 ASSERT(newcnt > 0);
311 tcnt = ccnt + newcnt;
312
313 for (n = 0; n < newcnt; n++) {
314 for (c = 0; c < ccnt; c++) {
315 if (newsec[n].s_secinfo.sc_nfsnum ==
316 cursec[c].s_secinfo.sc_nfsnum) {
317 cursec[c].s_refcnt += newsec[n].s_refcnt;
318 SECREF_TRACE(cursec, "add_ref",
319 cursec[c].s_secinfo.sc_nfsnum,
320 cursec[c].s_refcnt);
321 tcnt--;
322 break;
323 }
324 }
325 }
326
327 if (tcnt == ccnt)
328 return; /* no change; no new flavors */
329
330 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
331
332 /* move current secinfo list data to the new list */
333 for (c = 0; c < ccnt; c++)
334 msec[c] = cursec[c];
335
336 /* Add the flavor that's not in the current data */
337 mcnt = ccnt;
338 for (n = 0; n < newcnt; n++) {
339 for (c = 0; c < ccnt; c++) {
340 if (newsec[n].s_secinfo.sc_nfsnum ==
341 cursec[c].s_secinfo.sc_nfsnum)
342 break;
343 }
344
345 /* This is the one. Add it. */
346 if (c == ccnt) {
347 srv_secinfo_copy(&newsec[n], &msec[mcnt]);
348
349 if (is_pseudo)
350 msec[mcnt].s_flags = M_RO;
351
352 SECREF_TRACE(msec, "new_ref",
353 msec[mcnt].s_secinfo.sc_nfsnum,
354 msec[mcnt].s_refcnt);
355 mcnt++;
356 }
357 }
358
359 ASSERT(mcnt == tcnt);
360
361 /*
362 * Done. Update curdata. Free the old secinfo list in
363 * curdata and return the new sec array info
364 */
365 if (ccnt > 0)
366 kmem_free(cursec, ccnt * sizeof (struct secinfo));
367 *pcurcnt = tcnt;
368 *pcursec = msec;
369 }
370
371 /*
372 * For NFS V4.
373 * Remove the security data of the unexported node from its ancestors.
374 * Assume there is at least one flavor entry in the current sec list
375 * (pcursec).
376 *
377 * This routine is used under the protection of exported_lock (RW_WRITER).
378 *
379 * Every element of remsec is an explicitly exported flavor. If
380 * srv_secinfo_remove() is called fom an exportfs error path, then
381 * the flavor list was derived from the user's share cmdline,
382 * and all flavors are explicit. If it was called from the unshare path,
383 * build_seclist_nodups() was called with the exponly flag.
384 */
385 static void
386 srv_secinfo_remove(secinfo_t **pcursec, int *pcurcnt, secinfo_t *remsec,
387 int remcnt)
388 {
389 int ccnt, c; /* sec count in current data - cursec */
390 int r; /* sec count in removal data - remsec */
391 int tcnt, mcnt; /* total sec count after removing */
392 struct secinfo *msec; /* final secinfo list after removing */
393 struct secinfo *cursec;
394
395 cursec = *pcursec;
396 ccnt = *pcurcnt;
397 tcnt = ccnt;
398
399 for (r = 0; r < remcnt; r++) {
400 /*
401 * At unshare/reshare time, only explicitly shared flavor ref
402 * counts are decremented and propagated to ancestors.
403 * Implicit flavor refs came from shared descendants, and
404 * they must be kept.
405 */
406 if (! SEC_REF_EXPORTED(&remsec[r]))
407 continue;
408
409 for (c = 0; c < ccnt; c++) {
410 if (remsec[r].s_secinfo.sc_nfsnum ==
411 cursec[c].s_secinfo.sc_nfsnum) {
412
413 /*
414 * Decrement secinfo reference count by 1.
415 * If this entry is invalid after decrementing
416 * the count (i.e. count < 1), this entry will
417 * be removed.
418 */
419 cursec[c].s_refcnt--;
420
421 SECREF_TRACE(cursec, "del_ref",
422 cursec[c].s_secinfo.sc_nfsnum,
423 cursec[c].s_refcnt);
424
425 ASSERT(cursec[c].s_refcnt >= 0);
426
427 if (SEC_REF_INVALID(&cursec[c]))
428 tcnt--;
429 break;
430 }
431 }
432 }
433
434 ASSERT(tcnt >= 0);
435 if (tcnt == ccnt)
436 return; /* no change; no flavors to remove */
437
438 if (tcnt == 0) {
439 srv_secinfo_list_free(cursec, ccnt);
440 *pcurcnt = 0;
441 *pcursec = NULL;
442 return;
443 }
444
445 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
446
447 /* walk thru the given secinfo list to remove the flavors */
448 mcnt = 0;
449 for (c = 0; c < ccnt; c++) {
450 if (SEC_REF_INVALID(&cursec[c])) {
451 srv_secinfo_entry_free(&cursec[c]);
452 } else {
453 msec[mcnt] = cursec[c];
454 mcnt++;
455 }
456 }
457
458 ASSERT(mcnt == tcnt);
459 /*
460 * Done. Update curdata.
461 * Free the existing secinfo list in curdata. All pointers
462 * within the list have either been moved to msec or freed
463 * if it's invalid.
464 */
465 kmem_free(*pcursec, ccnt * sizeof (struct secinfo));
466 *pcursec = msec;
467 *pcurcnt = tcnt;
468 }
469
470
471 /*
472 * For the reshare case, sec flavor accounting happens in 3 steps:
473 * 1) propagate addition of new flavor refs up the ancestor tree
474 * 2) transfer flavor refs of descendants to new/reshared exportdata
475 * 3) propagate removal of old flavor refs up the ancestor tree
476 *
477 * srv_secinfo_exp2exp() implements step 2 of a reshare. At this point,
478 * the new flavor list has already been propagated up through the
479 * ancestor tree via srv_secinfo_treeclimb().
480 *
481 * If there is more than 1 export reference to an old flavor (i.e. some
482 * of its children shared with this flavor), this flavor information
483 * needs to be transferred to the new exportdata struct. A flavor in
484 * the old exportdata has descendant refs when its s_refcnt > 1 or it
485 * is implicitly shared (M_SEC4_EXPORTED not set in s_flags).
486 *
487 * SEC_REF_EXPORTED() is only true when M_SEC4_EXPORTED is set
488 * SEC_REF_SELF() is only true when both M_SEC4_EXPORTED is set and s_refcnt==1
489 *
490 * Transferring descendant flavor refcnts happens in 2 passes:
491 * a) flavors used before (oldsecinfo) and after (curdata->ex_secinfo) reshare
492 * b) flavors used before but not after reshare
493 *
494 * This routine is used under the protection of exported_lock (RW_WRITER).
495 */
496 void
497 srv_secinfo_exp2exp(exportdata_t *curdata, secinfo_t *oldsecinfo, int ocnt)
498 {
499 int ccnt, c; /* sec count in current data - curdata */
500 int o; /* sec count in old data - oldsecinfo */
501 int tcnt, mcnt; /* total sec count after the transfer */
502 struct secinfo *msec; /* merged secinfo list */
503
504 ccnt = curdata->ex_seccnt;
505
506 ASSERT(ocnt > 0);
507 ASSERT(!(curdata->ex_flags & EX_PSEUDO));
508
509 /*
510 * If the oldsecinfo has flavors with more than 1 reference count
511 * and the flavor is specified in the reshare, transfer the flavor
512 * refs to the new seclist (curdata.ex_secinfo).
513 */
514 tcnt = ccnt + ocnt;
515
516 for (o = 0; o < ocnt; o++) {
517
518 if (SEC_REF_SELF(&oldsecinfo[o])) {
519 tcnt--;
520 continue;
521 }
522
523 for (c = 0; c < ccnt; c++) {
524 if (oldsecinfo[o].s_secinfo.sc_nfsnum ==
525 curdata->ex_secinfo[c].s_secinfo.sc_nfsnum) {
526
527 /*
528 * add old reference to the current
529 * secinfo count
530 */
531 curdata->ex_secinfo[c].s_refcnt +=
532 oldsecinfo[o].s_refcnt;
533
534 /*
535 * Delete the old export flavor
536 * reference. The initial reference
537 * was created during srv_secinfo_add,
538 * and the count is decremented below
539 * to account for the initial reference.
540 */
541 if (SEC_REF_EXPORTED(&oldsecinfo[o]))
542 curdata->ex_secinfo[c].s_refcnt--;
543
544 SECREF_TRACE(curdata->ex_path,
545 "reshare_xfer_common_child_refs",
546 curdata->ex_secinfo[c].s_secinfo.sc_nfsnum,
547 curdata->ex_secinfo[c].s_refcnt);
548
549 ASSERT(curdata->ex_secinfo[c].s_refcnt >= 0);
550
551 tcnt--;
552 break;
553 }
554 }
555 }
556
557 if (tcnt == ccnt)
558 return; /* no more transfer to do */
559
560 /*
561 * oldsecinfo has flavors referenced by its children that are not
562 * in the current (new) export flavor list. Add these flavors.
563 */
564 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
565
566 /* move current secinfo list data to the new list */
567 for (c = 0; c < ccnt; c++)
568 msec[c] = curdata->ex_secinfo[c];
569
570 /*
571 * Add the flavor that's not in the new export, but still
572 * referenced by its children.
573 */
574 mcnt = ccnt;
575 for (o = 0; o < ocnt; o++) {
576 if (! SEC_REF_SELF(&oldsecinfo[o])) {
577 for (c = 0; c < ccnt; c++) {
578 if (oldsecinfo[o].s_secinfo.sc_nfsnum ==
579 curdata->ex_secinfo[c].s_secinfo.sc_nfsnum)
580 break;
581 }
582
583 /*
584 * This is the one. Add it. Decrement the ref count
585 * by 1 if the flavor is an explicitly shared flavor
586 * for the oldsecinfo export node.
587 */
588 if (c == ccnt) {
589 srv_secinfo_copy(&oldsecinfo[o], &msec[mcnt]);
590 if (SEC_REF_EXPORTED(&oldsecinfo[o]))
591 msec[mcnt].s_refcnt--;
592
593 SECREF_TRACE(curdata,
594 "reshare_xfer_implicit_child_refs",
595 msec[mcnt].s_secinfo.sc_nfsnum,
596 msec[mcnt].s_refcnt);
597
598 ASSERT(msec[mcnt].s_refcnt >= 0);
599 mcnt++;
600 }
601 }
602 }
603
604 ASSERT(mcnt == tcnt);
605 /*
606 * Done. Update curdata, free the existing secinfo list in
607 * curdata and set the new value.
608 */
609 if (ccnt > 0)
610 kmem_free(curdata->ex_secinfo, ccnt * sizeof (struct secinfo));
611 curdata->ex_seccnt = tcnt;
612 curdata->ex_secinfo = msec;
613 }
614
615 /*
616 * When unsharing an old export node and the old node becomes a pseudo node,
617 * if there is more than 1 export reference to an old flavor (i.e. some of
618 * its children shared with this flavor), this flavor information needs to
619 * be transferred to the new shared node.
620 *
621 * This routine is used under the protection of exported_lock (RW_WRITER).
622 */
623 void
624 srv_secinfo_exp2pseu(exportdata_t *curdata, exportdata_t *olddata)
625 {
626 int ocnt, o; /* sec count in transfer data - trandata */
627 int tcnt, mcnt; /* total sec count after transfer */
628 struct secinfo *msec; /* merged secinfo list */
629
630 ASSERT(curdata->ex_flags & EX_PSEUDO);
631 ASSERT(curdata->ex_seccnt == 0);
632
633 ocnt = olddata->ex_seccnt;
634
635 /*
636 * If the olddata has flavors with more than 1 reference count,
637 * transfer the information to the curdata.
638 */
639 tcnt = ocnt;
640
641 for (o = 0; o < ocnt; o++) {
642 if (SEC_REF_SELF(&olddata->ex_secinfo[o]))
643 tcnt--;
644 }
645
646 if (tcnt == 0)
647 return; /* no transfer to do */
648
649 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
650
651 mcnt = 0;
652 for (o = 0; o < ocnt; o++) {
653 if (! SEC_REF_SELF(&olddata->ex_secinfo[o])) {
654
655 /*
656 * Decrement the reference count by 1 if the flavor is
657 * an explicitly shared flavor for the olddata export
658 * node.
659 */
660 srv_secinfo_copy(&olddata->ex_secinfo[o], &msec[mcnt]);
661 msec[mcnt].s_flags = M_RO;
662 if (SEC_REF_EXPORTED(&olddata->ex_secinfo[o]))
663 msec[mcnt].s_refcnt--;
664
665 SECREF_TRACE(curdata, "unshare_morph_pseudo",
666 msec[mcnt].s_secinfo.sc_nfsnum,
667 msec[mcnt].s_refcnt);
668
669 ASSERT(msec[mcnt].s_refcnt >= 0);
670 mcnt++;
671 }
672 }
673
674 ASSERT(mcnt == tcnt);
675 /*
676 * Done. Update curdata.
677 * Free up the existing secinfo list in curdata and
678 * set the new value.
679 */
680 curdata->ex_seccnt = tcnt;
681 curdata->ex_secinfo = msec;
682 }
683
684 /*
685 * Find for given treenode the exportinfo which has its
686 * exp_visible linked on its exi_visible list.
687 *
688 * Note: We could add new pointer either to treenode or
689 * to exp_visible, which will point there directly.
690 * This would buy some speed for some memory.
691 */
692 exportinfo_t *
693 vis2exi(treenode_t *tnode)
694 {
695 exportinfo_t *exi_ret = NULL;
696
697 for (;;) {
698 tnode = tnode->tree_parent;
699 if (TREE_ROOT(tnode)) {
700 exi_ret = tnode->tree_exi;
701 break;
702 }
703 }
704
705 ASSERT(exi_ret); /* Every visible should have its home exportinfo */
706 return (exi_ret);
707 }
708
709 /*
710 * For NFS V4.
711 * Add or remove the newly exported or unexported security flavors of the
712 * given exportinfo from its ancestors upto the system root.
713 */
714 void
715 srv_secinfo_treeclimb(nfs_export_t *ne, exportinfo_t *exip, secinfo_t *sec,
716 int seccnt, bool_t isadd)
717 {
718 treenode_t *tnode;
719
720 ASSERT(RW_WRITE_HELD(&ne->exported_lock));
721
722 /*
723 * exi_tree can be null for the zone root
724 * which means we're already at the "top"
725 * and there's nothing more to "climb".
726 */
727 tnode = exip->exi_tree;
728 if (tnode == NULL) {
729 /* Should only happen for... */
730 ASSERT(exip == ne->exi_root);
731 return;
732 }
733
734 if (seccnt == 0)
735 return;
736
737 /*
738 * If flavors are being added and the new export root isn't
739 * also VROOT, its implicitly allowed flavors are inherited from
740 * its pseudonode.
741 * Note - for VROOT exports the implicitly allowed flavors were
742 * transferred from the PSEUDO export in exportfs()
743 */
744 if (isadd && !(exip->exi_vp->v_flag & VROOT) &&
745 tnode->tree_vis->vis_seccnt > 0) {
746 srv_secinfo_add(&exip->exi_export.ex_secinfo,
747 &exip->exi_export.ex_seccnt, tnode->tree_vis->vis_secinfo,
748 tnode->tree_vis->vis_seccnt, FALSE);
749 }
750
751 /*
752 * Move to parent node and propagate sec flavor
753 * to exportinfo and to visible structures.
754 */
755 tnode = tnode->tree_parent;
756
757 while (tnode != NULL) {
758
759 /* If there is exportinfo, update it */
760 if (tnode->tree_exi != NULL) {
761 secinfo_t **pxsec =
762 &tnode->tree_exi->exi_export.ex_secinfo;
763 int *pxcnt = &tnode->tree_exi->exi_export.ex_seccnt;
764 int is_pseudo = PSEUDO(tnode->tree_exi);
765 if (isadd)
766 srv_secinfo_add(pxsec, pxcnt, sec, seccnt,
767 is_pseudo);
768 else
769 srv_secinfo_remove(pxsec, pxcnt, sec, seccnt);
770 }
771
772 /* Update every visible - only root node has no visible */
773 if (tnode->tree_vis != NULL) {
774 secinfo_t **pxsec = &tnode->tree_vis->vis_secinfo;
775 int *pxcnt = &tnode->tree_vis->vis_seccnt;
776 if (isadd)
777 srv_secinfo_add(pxsec, pxcnt, sec, seccnt,
778 FALSE);
779 else
780 srv_secinfo_remove(pxsec, pxcnt, sec, seccnt);
781 }
782 tnode = tnode->tree_parent;
783 }
784 }
785
786 /* hash_name is a text substitution for either fid_hash or path_hash */
787 #define exp_hash_unlink(exi, hash_name) \
788 if (*(exi)->hash_name.bckt == (exi)) \
789 *(exi)->hash_name.bckt = (exi)->hash_name.next; \
790 if ((exi)->hash_name.prev) \
791 (exi)->hash_name.prev->hash_name.next = (exi)->hash_name.next; \
792 if ((exi)->hash_name.next) \
793 (exi)->hash_name.next->hash_name.prev = (exi)->hash_name.prev; \
794 (exi)->hash_name.bckt = NULL;
795
796 #define exp_hash_link(exi, hash_name, bucket) \
797 (exi)->hash_name.bckt = (bucket); \
798 (exi)->hash_name.prev = NULL; \
799 (exi)->hash_name.next = *(bucket); \
800 if ((exi)->hash_name.next) \
801 (exi)->hash_name.next->hash_name.prev = (exi); \
802 *(bucket) = (exi);
803
804 void
805 export_link(nfs_export_t *ne, exportinfo_t *exi)
806 {
807 exportinfo_t **bckt;
808
809 ASSERT(RW_WRITE_HELD(&ne->exported_lock));
810 ASSERT(exi->exi_zoneid == ne->ne_globals->nfs_zoneid);
811
812 bckt = &ne->exptable[exptablehash(&exi->exi_fsid, &exi->exi_fid)];
813 exp_hash_link(exi, fid_hash, bckt);
814
815 bckt = &ne->exptable_path_hash[pkp_tab_hash(exi->exi_export.ex_path,
816 strlen(exi->exi_export.ex_path))];
817 exp_hash_link(exi, path_hash, bckt);
818 }
819
820 /*
821 * Helper functions for exi_id handling
822 */
823 static int
824 exi_id_compar(const void *v1, const void *v2)
825 {
826 const struct exportinfo *e1 = v1;
827 const struct exportinfo *e2 = v2;
828
829 if (e1->exi_id < e2->exi_id)
830 return (-1);
831 if (e1->exi_id > e2->exi_id)
832 return (1);
833
834 return (0);
835 }
836
837 int
838 exi_id_get_next()
839 {
840 struct exportinfo e;
841 int ret = exi_id_next;
842
843 ASSERT(MUTEX_HELD(&nfs_exi_id_lock));
844
845 do {
846 exi_id_next++;
847 if (exi_id_next == 0)
848 exi_id_overflow = TRUE;
849
850 if (!exi_id_overflow)
851 break;
852
853 if (exi_id_next == ret)
854 cmn_err(CE_PANIC, "exi_id exhausted");
855
856 e.exi_id = exi_id_next;
857 } while (avl_find(&exi_id_tree, &e, NULL) != NULL);
858
859 return (ret);
860 }
861
862 /*
863 * Get the root file handle for this zone.
864 * Called when nfs_svc() starts
865 */
866 int
867 nfs_export_get_rootfh(nfs_globals_t *g)
868 {
869 nfs_export_t *ne = g->nfs_export;
870 int err;
871
872 ne->exi_rootfid.fid_len = MAXFIDSZ;
873 err = vop_fid_pseudo(ne->exi_root->exi_vp, &ne->exi_rootfid);
874 if (err != 0) {
875 ne->exi_rootfid.fid_len = 0;
876 return (err);
877 }
878
879 /* Setup the fhandle template exi_fh */
880 ne->exi_root->exi_fh.fh_fsid = rootdir->v_vfsp->vfs_fsid;
881 ne->exi_root->exi_fh.fh_xlen = ne->exi_rootfid.fid_len;
882 bcopy(ne->exi_rootfid.fid_data, ne->exi_root->exi_fh.fh_xdata,
883 ne->exi_rootfid.fid_len);
884 ne->exi_root->exi_fh.fh_len = sizeof (ne->exi_root->exi_fh.fh_data);
885
886 return (0);
887 }
888
889 void
890 nfs_export_zone_init(nfs_globals_t *ng)
891 {
892 int i;
893 nfs_export_t *ne;
894
895 ne = kmem_zalloc(sizeof (*ne), KM_SLEEP);
896
897 rw_init(&ne->exported_lock, NULL, RW_DEFAULT, NULL);
898
899 ne->ne_globals = ng; /* "up" pointer */
900
901 /*
902 * Allocate the place holder for the public file handle, which
903 * is all zeroes. It is initially set to the root filesystem.
904 */
905 ne->exi_root = kmem_zalloc(sizeof (*ne->exi_root), KM_SLEEP);
906 ne->exi_public = ne->exi_root;
907
908 ne->exi_root->exi_export.ex_flags = EX_PUBLIC;
909 ne->exi_root->exi_export.ex_pathlen = 1; /* length of "/" */
910 ne->exi_root->exi_export.ex_path =
911 kmem_alloc(ne->exi_root->exi_export.ex_pathlen + 1, KM_SLEEP);
912 ne->exi_root->exi_export.ex_path[0] = '/';
913 ne->exi_root->exi_export.ex_path[1] = '\0';
914
915 ne->exi_root->exi_count = 1;
916 mutex_init(&ne->exi_root->exi_lock, NULL, MUTEX_DEFAULT, NULL);
917
918 ne->exi_root->exi_vp = ZONE_ROOTVP();
919 ne->exi_root->exi_zoneid = ng->nfs_zoneid;
920
921 /*
922 * Fill in ne->exi_rootfid later, in nfs_export_get_rootfid
923 * because we can't correctly return errors here.
924 */
925
926 /* Initialize auth cache and auth cache lock */
927 for (i = 0; i < AUTH_TABLESIZE; i++) {
928 ne->exi_root->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t),
929 KM_SLEEP);
930 avl_create(ne->exi_root->exi_cache[i],
931 nfsauth_cache_clnt_compar, sizeof (struct auth_cache_clnt),
932 offsetof(struct auth_cache_clnt, authc_link));
933 }
934 rw_init(&ne->exi_root->exi_cache_lock, NULL, RW_DEFAULT, NULL);
935
936 /* setup exi_fh later, in nfs_export_get_rootfid */
937
938 rw_enter(&ne->exported_lock, RW_WRITER);
939
940 /* Publish the exportinfo in the hash table */
941 export_link(ne, ne->exi_root);
942
943 /* Initialize exi_id and exi_kstats */
944 mutex_enter(&nfs_exi_id_lock);
945 ne->exi_root->exi_id = exi_id_get_next();
946 avl_add(&exi_id_tree, ne->exi_root);
947 mutex_exit(&nfs_exi_id_lock);
948
949 rw_exit(&ne->exported_lock);
950 ne->ns_root = NULL;
951
952 ng->nfs_export = ne;
953 }
954
955 /*
956 * During zone shutdown, remove exports
957 */
958 void
959 nfs_export_zone_shutdown(nfs_globals_t *ng)
960 {
961 nfs_export_t *ne = ng->nfs_export;
962 struct exportinfo *exi, *nexi;
963 int i, errors;
964
965 rw_enter(&ne->exported_lock, RW_READER);
966
967 errors = 0;
968 for (i = 0; i < EXPTABLESIZE; i++) {
969
970 exi = ne->exptable[i];
971 if (exi != NULL)
972 exi_hold(exi);
973
974 while (exi != NULL) {
975
976 /*
977 * Get and hold next export before
978 * dropping the rwlock and unexport
979 */
980 nexi = exi->fid_hash.next;
981 if (nexi != NULL)
982 exi_hold(nexi);
983
984 rw_exit(&ne->exported_lock);
985
986 /*
987 * Skip ne->exi_root which gets special
988 * create/destroy handling.
989 */
990 if (exi != ne->exi_root &&
991 unexport(ne, exi) != 0)
992 errors++;
993 exi_rele(exi);
994
995 rw_enter(&ne->exported_lock, RW_READER);
996 exi = nexi;
997 }
998 }
999 if (errors > 0) {
1000 cmn_err(CE_NOTE,
1001 "NFS: failed un-exports in zone %d",
1002 (int) ng->nfs_zoneid);
1003 }
1004
1005 rw_exit(&ne->exported_lock);
1006 }
1007
1008 void
1009 nfs_export_zone_fini(nfs_globals_t *ng)
1010 {
1011 int i;
1012 nfs_export_t *ne = ng->nfs_export;
1013 struct exportinfo *exi;
1014
1015 ng->nfs_export = NULL;
1016
1017 rw_enter(&ne->exported_lock, RW_WRITER);
1018
1019 mutex_enter(&nfs_exi_id_lock);
1020 avl_remove(&exi_id_tree, ne->exi_root);
1021 mutex_exit(&nfs_exi_id_lock);
1022
1023 export_unlink(ne, ne->exi_root);
1024
1025 rw_exit(&ne->exported_lock);
1026
1027 /* Deallocate the place holder for the public file handle */
1028 srv_secinfo_list_free(ne->exi_root->exi_export.ex_secinfo,
1029 ne->exi_root->exi_export.ex_seccnt);
1030 mutex_destroy(&ne->exi_root->exi_lock);
1031
1032 rw_destroy(&ne->exi_root->exi_cache_lock);
1033 for (i = 0; i < AUTH_TABLESIZE; i++) {
1034 avl_destroy(ne->exi_root->exi_cache[i]);
1035 kmem_free(ne->exi_root->exi_cache[i], sizeof (avl_tree_t));
1036 }
1037
1038 kmem_free(ne->exi_root->exi_export.ex_path,
1039 ne->exi_root->exi_export.ex_pathlen + 1);
1040 kmem_free(ne->exi_root, sizeof (*ne->exi_root));
1041
1042 /*
1043 * The shutdown hook should have left the exi_id_tree
1044 * with nothing belonging to this zone.
1045 */
1046 mutex_enter(&nfs_exi_id_lock);
1047 i = 0;
1048 exi = avl_first(&exi_id_tree);
1049 while (exi != NULL) {
1050 if (exi->exi_zoneid == ng->nfs_zoneid)
1051 i++;
1052 exi = AVL_NEXT(&exi_id_tree, exi);
1053 }
1054 mutex_exit(&nfs_exi_id_lock);
1055 if (i > 0) {
1056 cmn_err(CE_NOTE,
1057 "NFS: zone %d has %d export IDs left after shutdown",
1058 (int) ng->nfs_zoneid, i);
1059 }
1060 rw_destroy(&ne->exported_lock);
1061 kmem_free(ne, sizeof (*ne));
1062 }
1063
1064 /*
1065 * Initialization routine for export routines.
1066 * Should only be called once.
1067 */
1068 void
1069 nfs_exportinit(void)
1070 {
1071 mutex_init(&nfs_exi_id_lock, NULL, MUTEX_DEFAULT, NULL);
1072
1073 /* exi_id handling initialization */
1074 exi_id_next = 0;
1075 exi_id_overflow = FALSE;
1076 avl_create(&exi_id_tree, exi_id_compar, sizeof (struct exportinfo),
1077 offsetof(struct exportinfo, exi_id_link));
1078
1079 nfslog_init();
1080 }
1081
1082 /*
1083 * Finalization routine for export routines.
1084 */
1085 void
1086 nfs_exportfini(void)
1087 {
1088 avl_destroy(&exi_id_tree);
1089 mutex_destroy(&nfs_exi_id_lock);
1090 }
1091
1092 /*
1093 * Check if 2 gss mechanism identifiers are the same.
1094 *
1095 * return FALSE if not the same.
1096 * return TRUE if the same.
1097 */
1098 static bool_t
1099 nfs_mech_equal(rpc_gss_OID mech1, rpc_gss_OID mech2)
1100 {
1101 if ((mech1->length == 0) && (mech2->length == 0))
1102 return (TRUE);
1103
1104 if (mech1->length != mech2->length)
1105 return (FALSE);
1106
1107 return (bcmp(mech1->elements, mech2->elements, mech1->length) == 0);
1108 }
1109
1110 /*
1111 * This routine is used by rpc to map rpc security number
1112 * to nfs specific security flavor number.
1113 *
1114 * The gss callback prototype is
1115 * callback(struct svc_req *, gss_cred_id_t *, gss_ctx_id_t *,
1116 * rpc_gss_lock_t *, void **),
1117 * since nfs does not use the gss_cred_id_t/gss_ctx_id_t arguments
1118 * we cast them to void.
1119 */
1120 /*ARGSUSED*/
1121 bool_t
1122 rfs_gsscallback(struct svc_req *req, gss_cred_id_t deleg, void *gss_context,
1123 rpc_gss_lock_t *lock, void **cookie)
1124 {
1125 int i, j;
1126 rpc_gss_rawcred_t *raw_cred;
1127 struct exportinfo *exi;
1128 nfs_export_t *ne = nfs_get_export();
1129
1130 /*
1131 * We don't deal with delegated credentials.
1132 */
1133 if (deleg != GSS_C_NO_CREDENTIAL)
1134 return (FALSE);
1135
1136 raw_cred = lock->raw_cred;
1137 *cookie = NULL;
1138
1139 rw_enter(&ne->exported_lock, RW_READER);
1140
1141 for (i = 0; i < EXPTABLESIZE; i++) {
1142 exi = ne->exptable[i];
1143 while (exi) {
1144 if (exi->exi_export.ex_seccnt > 0) {
1145 struct secinfo *secp;
1146 seconfig_t *se;
1147 int seccnt;
1148
1149 secp = exi->exi_export.ex_secinfo;
1150 seccnt = exi->exi_export.ex_seccnt;
1151 for (j = 0; j < seccnt; j++) {
1152 /*
1153 * If there is a map of the triplet
1154 * (mechanism, service, qop) between
1155 * raw_cred and the exported flavor,
1156 * get the psudo flavor number.
1157 * Also qop should not be NULL, it
1158 * should be "default" or something
1159 * else.
1160 */
1161 se = &secp[j].s_secinfo;
1162 if ((se->sc_rpcnum == RPCSEC_GSS) &&
1163
1164 (nfs_mech_equal(
1165 se->sc_gss_mech_type,
1166 raw_cred->mechanism)) &&
1167
1168 (se->sc_service ==
1169 raw_cred->service) &&
1170 (raw_cred->qop == se->sc_qop)) {
1171
1172 *cookie = (void *)(uintptr_t)
1173 se->sc_nfsnum;
1174 goto done;
1175 }
1176 }
1177 }
1178 exi = exi->fid_hash.next;
1179 }
1180 }
1181 done:
1182 rw_exit(&ne->exported_lock);
1183
1184 /*
1185 * If no nfs pseudo number mapping can be found in the export
1186 * table, assign the nfsflavor to NFS_FLAVOR_NOMAP. In V4, we may
1187 * recover the flavor mismatch from NFS layer (NFS4ERR_WRONGSEC).
1188 *
1189 * For example:
1190 * server first shares with krb5i;
1191 * client mounts with krb5i;
1192 * server re-shares with krb5p;
1193 * client tries with krb5i, but no mapping can be found;
1194 * rpcsec_gss module calls this routine to do the mapping,
1195 * if this routine fails, request is rejected from
1196 * the rpc layer.
1197 * What we need is to let the nfs layer rejects the request.
1198 * For V4, we can reject with NFS4ERR_WRONGSEC and the client
1199 * may recover from it by getting the new flavor via SECINFO.
1200 *
1201 * nfs pseudo number for RPCSEC_GSS mapping (see nfssec.conf)
1202 * is owned by IANA (see RFC 2623).
1203 *
1204 * XXX NFS_FLAVOR_NOMAP is defined in Solaris to work around
1205 * the implementation issue. This number should not overlap with
1206 * any new IANA defined pseudo flavor numbers.
1207 */
1208 if (*cookie == NULL)
1209 *cookie = (void *)NFS_FLAVOR_NOMAP;
1210
1211 lock->locked = TRUE;
1212
1213 return (TRUE);
1214 }
1215
1216
1217 /*
1218 * Exportfs system call; credentials should be checked before
1219 * calling this function.
1220 */
1221 int
1222 exportfs(struct exportfs_args *args, model_t model, cred_t *cr)
1223 {
1224 vnode_t *vp;
1225 vnode_t *dvp;
1226 struct exportdata *kex;
1227 struct exportinfo *exi = NULL;
1228 struct exportinfo *ex, *ex1, *ex2;
1229 fid_t fid;
1230 fsid_t fsid;
1231 int error;
1232 size_t allocsize;
1233 struct secinfo *sp;
1234 struct secinfo *exs;
1235 rpc_gss_callback_t cb;
1236 char *pathbuf;
1237 char *log_buffer;
1238 char *tagbuf;
1239 int callback;
1240 int allocd_seccnt;
1241 STRUCT_HANDLE(exportfs_args, uap);
1242 STRUCT_DECL(exportdata, uexi);
1243 struct secinfo newsec[MAX_FLAVORS];
1244 int newcnt;
1245 struct secinfo oldsec[MAX_FLAVORS];
1246 int oldcnt;
1247 int i;
1248 struct pathname lookpn;
1249 nfs_export_t *ne = nfs_get_export();
1250
1251 STRUCT_SET_HANDLE(uap, model, args);
1252
1253 /* Read in pathname from userspace */
1254 if (error = pn_get(STRUCT_FGETP(uap, dname), UIO_USERSPACE, &lookpn))
1255 return (error);
1256
1257 /* Walk the export list looking for that pathname */
1258 rw_enter(&ne->exported_lock, RW_READER);
1259 DTRACE_PROBE(nfss__i__exported_lock1_start);
1260 for (ex1 = ne->exptable_path_hash[pkp_tab_hash(lookpn.pn_path,
1261 strlen(lookpn.pn_path))]; ex1; ex1 = ex1->path_hash.next) {
1262 if (ex1 != ne->exi_root && 0 ==
1263 strcmp(ex1->exi_export.ex_path, lookpn.pn_path)) {
1264 exi_hold(ex1);
1265 break;
1266 }
1267 }
1268 DTRACE_PROBE(nfss__i__exported_lock1_stop);
1269 rw_exit(&ne->exported_lock);
1270
1271 /* Is this an unshare? */
1272 if (STRUCT_FGETP(uap, uex) == NULL) {
1273 pn_free(&lookpn);
1274 if (ex1 == NULL)
1275 return (EINVAL);
1276 error = unexport(ne, ex1);
1277 exi_rele(ex1);
1278 return (error);
1279 }
1280
1281 /* It is a share or a re-share */
1282 error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1283 FOLLOW, &dvp, &vp);
1284 if (error == EINVAL) {
1285 /*
1286 * if fname resolves to / we get EINVAL error
1287 * since we wanted the parent vnode. Try again
1288 * with NULL dvp.
1289 */
1290 error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1291 FOLLOW, NULL, &vp);
1292 dvp = NULL;
1293 }
1294 if (!error && vp == NULL) {
1295 /* Last component of fname not found */
1296 if (dvp != NULL)
1297 VN_RELE(dvp);
1298 error = ENOENT;
1299 }
1300 if (error) {
1301 pn_free(&lookpn);
1302 if (ex1)
1303 exi_rele(ex1);
1304 return (error);
1305 }
1306
1307 /*
1308 * 'vp' may be an AUTOFS node, so we perform a
1309 * VOP_ACCESS() to trigger the mount of the
1310 * intended filesystem, so we can share the intended
1311 * filesystem instead of the AUTOFS filesystem.
1312 */
1313 (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
1314
1315 /*
1316 * We're interested in the top most filesystem.
1317 * This is specially important when uap->dname is a trigger
1318 * AUTOFS node, since we're really interested in sharing the
1319 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
1320 * call not the AUTOFS node itself.
1321 */
1322 if (vn_mountedvfs(vp) != NULL) {
1323 if (error = traverse(&vp)) {
1324 VN_RELE(vp);
1325 if (dvp != NULL)
1326 VN_RELE(dvp);
1327 pn_free(&lookpn);
1328 if (ex1)
1329 exi_rele(ex1);
1330 return (error);
1331 }
1332 }
1333
1334 /* Do not allow sharing another vnode for already shared path */
1335 if (ex1 && !PSEUDO(ex1) && !VN_CMP(ex1->exi_vp, vp)) {
1336 VN_RELE(vp);
1337 if (dvp != NULL)
1338 VN_RELE(dvp);
1339 pn_free(&lookpn);
1340 exi_rele(ex1);
1341 return (EEXIST);
1342 }
1343 if (ex1)
1344 exi_rele(ex1);
1345
1346 /*
1347 * Get the vfs id
1348 */
1349 bzero(&fid, sizeof (fid));
1350 fid.fid_len = MAXFIDSZ;
1351 error = VOP_FID(vp, &fid, NULL);
1352 fsid = vp->v_vfsp->vfs_fsid;
1353
1354 if (error) {
1355 VN_RELE(vp);
1356 if (dvp != NULL)
1357 VN_RELE(dvp);
1358 /*
1359 * If VOP_FID returns ENOSPC then the fid supplied
1360 * is too small. For now we simply return EREMOTE.
1361 */
1362 if (error == ENOSPC)
1363 error = EREMOTE;
1364 pn_free(&lookpn);
1365 return (error);
1366 }
1367
1368 /*
1369 * Do not allow re-sharing a shared vnode under a different path
1370 * PSEUDO export has ex_path fabricated, e.g. "/tmp (pseudo)", skip it.
1371 */
1372 rw_enter(&ne->exported_lock, RW_READER);
1373 DTRACE_PROBE(nfss__i__exported_lock2_start);
1374 for (ex2 = ne->exptable[exptablehash(&fsid, &fid)]; ex2;
1375 ex2 = ex2->fid_hash.next) {
1376 if (ex2 != ne->exi_root && !PSEUDO(ex2) &&
1377 VN_CMP(ex2->exi_vp, vp) &&
1378 strcmp(ex2->exi_export.ex_path, lookpn.pn_path) != 0) {
1379 DTRACE_PROBE(nfss__i__exported_lock2_stop);
1380 rw_exit(&ne->exported_lock);
1381 VN_RELE(vp);
1382 if (dvp != NULL)
1383 VN_RELE(dvp);
1384 pn_free(&lookpn);
1385 return (EEXIST);
1386 }
1387 }
1388 DTRACE_PROBE(nfss__i__exported_lock2_stop);
1389 rw_exit(&ne->exported_lock);
1390 pn_free(&lookpn);
1391
1392 exi = kmem_zalloc(sizeof (*exi), KM_SLEEP);
1393 exi->exi_fsid = fsid;
1394 exi->exi_fid = fid;
1395 exi->exi_vp = vp;
1396 exi->exi_count = 1;
1397 exi->exi_zoneid = crgetzoneid(cr);
1398 exi->exi_volatile_dev = (vfssw[vp->v_vfsp->vfs_fstype].vsw_flag &
1399 VSW_VOLATILEDEV) ? 1 : 0;
1400 mutex_init(&exi->exi_lock, NULL, MUTEX_DEFAULT, NULL);
1401 exi->exi_dvp = dvp;
1402
1403 /*
1404 * Initialize auth cache and auth cache lock
1405 */
1406 for (i = 0; i < AUTH_TABLESIZE; i++) {
1407 exi->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1408 avl_create(exi->exi_cache[i], nfsauth_cache_clnt_compar,
1409 sizeof (struct auth_cache_clnt),
1410 offsetof(struct auth_cache_clnt, authc_link));
1411 }
1412 rw_init(&exi->exi_cache_lock, NULL, RW_DEFAULT, NULL);
1413
1414 /*
1415 * Build up the template fhandle
1416 */
1417 exi->exi_fh.fh_fsid = fsid;
1418 if (exi->exi_fid.fid_len > sizeof (exi->exi_fh.fh_xdata)) {
1419 error = EREMOTE;
1420 goto out1;
1421 }
1422 exi->exi_fh.fh_xlen = exi->exi_fid.fid_len;
1423 bcopy(exi->exi_fid.fid_data, exi->exi_fh.fh_xdata,
1424 exi->exi_fid.fid_len);
1425
1426 exi->exi_fh.fh_len = sizeof (exi->exi_fh.fh_data);
1427
1428 kex = &exi->exi_export;
1429
1430 /*
1431 * Load in everything, and do sanity checking
1432 */
1433 STRUCT_INIT(uexi, model);
1434 if (copyin(STRUCT_FGETP(uap, uex), STRUCT_BUF(uexi),
1435 STRUCT_SIZE(uexi))) {
1436 error = EFAULT;
1437 goto out1;
1438 }
1439
1440 kex->ex_version = STRUCT_FGET(uexi, ex_version);
1441 if (kex->ex_version != EX_CURRENT_VERSION) {
1442 error = EINVAL;
1443 cmn_err(CE_WARN,
1444 "NFS: exportfs requires export struct version 2 - got %d\n",
1445 kex->ex_version);
1446 goto out1;
1447 }
1448
1449 /*
1450 * Must have at least one security entry
1451 */
1452 kex->ex_seccnt = STRUCT_FGET(uexi, ex_seccnt);
1453 if (kex->ex_seccnt < 1) {
1454 error = EINVAL;
1455 goto out1;
1456 }
1457
1458 kex->ex_path = STRUCT_FGETP(uexi, ex_path);
1459 kex->ex_pathlen = STRUCT_FGET(uexi, ex_pathlen);
1460 kex->ex_flags = STRUCT_FGET(uexi, ex_flags);
1461 kex->ex_anon = STRUCT_FGET(uexi, ex_anon);
1462 kex->ex_secinfo = STRUCT_FGETP(uexi, ex_secinfo);
1463 kex->ex_index = STRUCT_FGETP(uexi, ex_index);
1464 kex->ex_log_buffer = STRUCT_FGETP(uexi, ex_log_buffer);
1465 kex->ex_log_bufferlen = STRUCT_FGET(uexi, ex_log_bufferlen);
1466 kex->ex_tag = STRUCT_FGETP(uexi, ex_tag);
1467 kex->ex_taglen = STRUCT_FGET(uexi, ex_taglen);
1468
1469 /*
1470 * Copy the exported pathname into
1471 * an appropriately sized buffer.
1472 */
1473 pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1474 if (copyinstr(kex->ex_path, pathbuf, MAXPATHLEN, &kex->ex_pathlen)) {
1475 kmem_free(pathbuf, MAXPATHLEN);
1476 error = EFAULT;
1477 goto out1;
1478 }
1479 kex->ex_path = kmem_alloc(kex->ex_pathlen + 1, KM_SLEEP);
1480 bcopy(pathbuf, kex->ex_path, kex->ex_pathlen);
1481 kex->ex_path[kex->ex_pathlen] = '\0';
1482 kmem_free(pathbuf, MAXPATHLEN);
1483
1484 /*
1485 * Get the path to the logging buffer and the tag
1486 */
1487 if (kex->ex_flags & EX_LOG) {
1488 log_buffer = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1489 if (copyinstr(kex->ex_log_buffer, log_buffer, MAXPATHLEN,
1490 &kex->ex_log_bufferlen)) {
1491 kmem_free(log_buffer, MAXPATHLEN);
1492 error = EFAULT;
1493 goto out2;
1494 }
1495 kex->ex_log_buffer =
1496 kmem_alloc(kex->ex_log_bufferlen + 1, KM_SLEEP);
1497 bcopy(log_buffer, kex->ex_log_buffer, kex->ex_log_bufferlen);
1498 kex->ex_log_buffer[kex->ex_log_bufferlen] = '\0';
1499 kmem_free(log_buffer, MAXPATHLEN);
1500
1501 tagbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1502 if (copyinstr(kex->ex_tag, tagbuf, MAXPATHLEN,
1503 &kex->ex_taglen)) {
1504 kmem_free(tagbuf, MAXPATHLEN);
1505 error = EFAULT;
1506 goto out3;
1507 }
1508 kex->ex_tag = kmem_alloc(kex->ex_taglen + 1, KM_SLEEP);
1509 bcopy(tagbuf, kex->ex_tag, kex->ex_taglen);
1510 kex->ex_tag[kex->ex_taglen] = '\0';
1511 kmem_free(tagbuf, MAXPATHLEN);
1512 }
1513
1514 /*
1515 * Load the security information for each flavor
1516 */
1517 allocsize = kex->ex_seccnt * SIZEOF_STRUCT(secinfo, model);
1518 sp = kmem_zalloc(allocsize, KM_SLEEP);
1519 if (copyin(kex->ex_secinfo, sp, allocsize)) {
1520 kmem_free(sp, allocsize);
1521 error = EFAULT;
1522 goto out4;
1523 }
1524
1525 /*
1526 * All of these nested structures need to be converted to
1527 * the kernel native format.
1528 */
1529 if (model != DATAMODEL_NATIVE) {
1530 size_t allocsize2;
1531 struct secinfo *sp2;
1532
1533 allocsize2 = kex->ex_seccnt * sizeof (struct secinfo);
1534 sp2 = kmem_zalloc(allocsize2, KM_SLEEP);
1535
1536 for (i = 0; i < kex->ex_seccnt; i++) {
1537 STRUCT_HANDLE(secinfo, usi);
1538
1539 STRUCT_SET_HANDLE(usi, model,
1540 (struct secinfo *)((caddr_t)sp +
1541 (i * SIZEOF_STRUCT(secinfo, model))));
1542 bcopy(STRUCT_FGET(usi, s_secinfo.sc_name),
1543 sp2[i].s_secinfo.sc_name, MAX_NAME_LEN);
1544 sp2[i].s_secinfo.sc_nfsnum =
1545 STRUCT_FGET(usi, s_secinfo.sc_nfsnum);
1546 sp2[i].s_secinfo.sc_rpcnum =
1547 STRUCT_FGET(usi, s_secinfo.sc_rpcnum);
1548 bcopy(STRUCT_FGET(usi, s_secinfo.sc_gss_mech),
1549 sp2[i].s_secinfo.sc_gss_mech, MAX_NAME_LEN);
1550 sp2[i].s_secinfo.sc_gss_mech_type =
1551 STRUCT_FGETP(usi, s_secinfo.sc_gss_mech_type);
1552 sp2[i].s_secinfo.sc_qop =
1553 STRUCT_FGET(usi, s_secinfo.sc_qop);
1554 sp2[i].s_secinfo.sc_service =
1555 STRUCT_FGET(usi, s_secinfo.sc_service);
1556
1557 sp2[i].s_flags = STRUCT_FGET(usi, s_flags);
1558 sp2[i].s_window = STRUCT_FGET(usi, s_window);
1559 sp2[i].s_rootid = STRUCT_FGET(usi, s_rootid);
1560 sp2[i].s_rootcnt = STRUCT_FGET(usi, s_rootcnt);
1561 sp2[i].s_rootnames = STRUCT_FGETP(usi, s_rootnames);
1562 }
1563 kmem_free(sp, allocsize);
1564 sp = sp2;
1565 allocsize = allocsize2;
1566 }
1567
1568 kex->ex_secinfo = sp;
1569
1570 /*
1571 * And now copy rootnames for each individual secinfo.
1572 */
1573 callback = 0;
1574 allocd_seccnt = 0;
1575 while (allocd_seccnt < kex->ex_seccnt) {
1576
1577 exs = &sp[allocd_seccnt];
1578 if (exs->s_rootcnt > 0) {
1579 if (!sec_svc_loadrootnames(exs->s_secinfo.sc_rpcnum,
1580 exs->s_rootcnt, &exs->s_rootnames, model)) {
1581 error = EFAULT;
1582 goto out5;
1583 }
1584 }
1585
1586 if (exs->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
1587 rpc_gss_OID mech_tmp;
1588 STRUCT_DECL(rpc_gss_OID_s, umech_tmp);
1589 caddr_t elements_tmp;
1590
1591 /* Copyin mechanism type */
1592 STRUCT_INIT(umech_tmp, model);
1593 mech_tmp = kmem_alloc(sizeof (*mech_tmp), KM_SLEEP);
1594 if (copyin(exs->s_secinfo.sc_gss_mech_type,
1595 STRUCT_BUF(umech_tmp), STRUCT_SIZE(umech_tmp))) {
1596 kmem_free(mech_tmp, sizeof (*mech_tmp));
1597 error = EFAULT;
1598 goto out5;
1599 }
1600 mech_tmp->length = STRUCT_FGET(umech_tmp, length);
1601 mech_tmp->elements = STRUCT_FGETP(umech_tmp, elements);
1602
1603 elements_tmp = kmem_alloc(mech_tmp->length, KM_SLEEP);
1604 if (copyin(mech_tmp->elements, elements_tmp,
1605 mech_tmp->length)) {
1606 kmem_free(elements_tmp, mech_tmp->length);
1607 kmem_free(mech_tmp, sizeof (*mech_tmp));
1608 error = EFAULT;
1609 goto out5;
1610 }
1611 mech_tmp->elements = elements_tmp;
1612 exs->s_secinfo.sc_gss_mech_type = mech_tmp;
1613 allocd_seccnt++;
1614
1615 callback = 1;
1616 } else
1617 allocd_seccnt++;
1618 }
1619
1620 /*
1621 * Init the secinfo reference count and mark these flavors
1622 * explicitly exported flavors.
1623 */
1624 for (i = 0; i < kex->ex_seccnt; i++) {
1625 kex->ex_secinfo[i].s_flags |= M_4SEC_EXPORTED;
1626 kex->ex_secinfo[i].s_refcnt = 1;
1627 }
1628
1629 /*
1630 * Set up rpcsec_gss callback routine entry if any.
1631 */
1632 if (callback) {
1633 cb.callback = rfs_gsscallback;
1634 cb.program = NFS_ACL_PROGRAM;
1635 for (cb.version = NFS_ACL_VERSMIN;
1636 cb.version <= NFS_ACL_VERSMAX; cb.version++) {
1637 (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1638 (void *)&cb);
1639 }
1640
1641 cb.program = NFS_PROGRAM;
1642 for (cb.version = NFS_VERSMIN;
1643 cb.version <= NFS_VERSMAX; cb.version++) {
1644 (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1645 (void *)&cb);
1646 }
1647 }
1648
1649 /*
1650 * Check the index flag. Do this here to avoid holding the
1651 * lock while dealing with the index option (as we do with
1652 * the public option).
1653 */
1654 if (kex->ex_flags & EX_INDEX) {
1655 if (!kex->ex_index) { /* sanity check */
1656 error = EINVAL;
1657 goto out5;
1658 }
1659 if (error = loadindex(kex))
1660 goto out5;
1661 }
1662
1663 if (kex->ex_flags & EX_LOG) {
1664 if (error = nfslog_setup(exi))
1665 goto out6;
1666 }
1667
1668 /*
1669 * Insert the new entry at the front of the export list
1670 */
1671 rw_enter(&ne->exported_lock, RW_WRITER);
1672 DTRACE_PROBE(nfss__i__exported_lock3_start);
1673
1674 export_link(ne, exi);
1675
1676 /*
1677 * Check the rest of the list for an old entry for the fs.
1678 * If one is found then unlink it, wait until this is the
1679 * only reference and then free it.
1680 */
1681 for (ex = exi->fid_hash.next; ex != NULL; ex = ex->fid_hash.next) {
1682 if (ex != ne->exi_root && VN_CMP(ex->exi_vp, vp)) {
1683 mutex_enter(&nfs_exi_id_lock);
1684 avl_remove(&exi_id_tree, ex);
1685 mutex_exit(&nfs_exi_id_lock);
1686 export_unlink(ne, ex);
1687 break;
1688 }
1689 }
1690
1691 /*
1692 * If the public filehandle is pointing at the
1693 * old entry, then point it back at the root.
1694 */
1695 if (ex != NULL && ex == ne->exi_public)
1696 ne->exi_public = ne->exi_root;
1697
1698 /*
1699 * If the public flag is on, make the global exi_public
1700 * point to this entry and turn off the public bit so that
1701 * we can distinguish it from the place holder export.
1702 */
1703 if (kex->ex_flags & EX_PUBLIC) {
1704 ne->exi_public = exi;
1705 kex->ex_flags &= ~EX_PUBLIC;
1706 }
1707
1708 #ifdef VOLATILE_FH_TEST
1709 /*
1710 * Set up the volatile_id value if volatile on share.
1711 * The list of volatile renamed filehandles is always destroyed,
1712 * if the fs was reshared.
1713 */
1714 if (kex->ex_flags & EX_VOLFH)
1715 exi->exi_volatile_id = gethrestime_sec();
1716
1717 mutex_init(&exi->exi_vol_rename_lock, NULL, MUTEX_DEFAULT, NULL);
1718 #endif /* VOLATILE_FH_TEST */
1719
1720 /*
1721 * If this is a new export, then climb up
1722 * the tree and check if any pseudo exports
1723 * need to be created to provide a path for
1724 * NFS v4 clients.
1725 */
1726 if (ex == NULL) {
1727 error = treeclimb_export(exi);
1728 if (error)
1729 goto out7;
1730 } else {
1731 /* If it's a re-export update namespace tree */
1732 exi->exi_tree = ex->exi_tree;
1733 exi->exi_tree->tree_exi = exi;
1734
1735 /* Update the change timestamp */
1736 tree_update_change(ne, exi->exi_tree, NULL);
1737 }
1738
1739 /*
1740 * build a unique flavor list from the flavors specified
1741 * in the share cmd. unique means that each flavor only
1742 * appears once in the secinfo list -- no duplicates allowed.
1743 */
1744 newcnt = build_seclist_nodups(&exi->exi_export, newsec, FALSE);
1745
1746 srv_secinfo_treeclimb(ne, exi, newsec, newcnt, TRUE);
1747
1748 /*
1749 * If re-sharing an old export entry, update the secinfo data
1750 * depending on if the old entry is a pseudo node or not.
1751 */
1752 if (ex != NULL) {
1753 oldcnt = build_seclist_nodups(&ex->exi_export, oldsec, FALSE);
1754 if (PSEUDO(ex)) {
1755 /*
1756 * The dir being shared is a pseudo export root (which
1757 * will be transformed into a real export root). The
1758 * flavor(s) of the new share were propagated to the
1759 * ancestors by srv_secinfo_treeclimb() above. Now
1760 * transfer the implicit flavor refs from the old
1761 * pseudo exprot root to the new (real) export root.
1762 */
1763 srv_secinfo_add(&exi->exi_export.ex_secinfo,
1764 &exi->exi_export.ex_seccnt, oldsec, oldcnt, TRUE);
1765 } else {
1766 /*
1767 * First transfer implicit flavor refs to new export.
1768 * Remove old flavor refs last.
1769 */
1770 srv_secinfo_exp2exp(&exi->exi_export, oldsec, oldcnt);
1771 srv_secinfo_treeclimb(ne, ex, oldsec, oldcnt, FALSE);
1772 }
1773 }
1774
1775 /*
1776 * If it's a re-export and the old entry has a pseudonode list,
1777 * transfer it to the new export.
1778 */
1779 if (ex != NULL && (ex->exi_visible != NULL)) {
1780 exi->exi_visible = ex->exi_visible;
1781 ex->exi_visible = NULL;
1782 }
1783
1784 /*
1785 * Initialize exi_id and exi_kstats
1786 */
1787 if (ex != NULL) {
1788 exi->exi_id = ex->exi_id;
1789 } else {
1790 mutex_enter(&nfs_exi_id_lock);
1791 exi->exi_id = exi_id_get_next();
1792 mutex_exit(&nfs_exi_id_lock);
1793 }
1794 mutex_enter(&nfs_exi_id_lock);
1795 avl_add(&exi_id_tree, exi);
1796 mutex_exit(&nfs_exi_id_lock);
1797
1798 DTRACE_PROBE(nfss__i__exported_lock3_stop);
1799 rw_exit(&ne->exported_lock);
1800
1801 if (ne->exi_public == exi || kex->ex_flags & EX_LOG) {
1802 /*
1803 * Log share operation to this buffer only.
1804 */
1805 nfslog_share_record(exi, cr);
1806 }
1807
1808 if (ex != NULL)
1809 exi_rele(ex);
1810
1811 return (0);
1812
1813 out7:
1814 /* Unlink the new export in exptable. */
1815 export_unlink(ne, exi);
1816 DTRACE_PROBE(nfss__i__exported_lock3_stop);
1817 rw_exit(&ne->exported_lock);
1818 out6:
1819 if (kex->ex_flags & EX_INDEX)
1820 kmem_free(kex->ex_index, strlen(kex->ex_index) + 1);
1821 out5:
1822 /* free partially completed allocation */
1823 while (--allocd_seccnt >= 0) {
1824 exs = &kex->ex_secinfo[allocd_seccnt];
1825 srv_secinfo_entry_free(exs);
1826 }
1827
1828 if (kex->ex_secinfo) {
1829 kmem_free(kex->ex_secinfo,
1830 kex->ex_seccnt * sizeof (struct secinfo));
1831 }
1832
1833 out4:
1834 if ((kex->ex_flags & EX_LOG) && kex->ex_tag != NULL)
1835 kmem_free(kex->ex_tag, kex->ex_taglen + 1);
1836 out3:
1837 if ((kex->ex_flags & EX_LOG) && kex->ex_log_buffer != NULL)
1838 kmem_free(kex->ex_log_buffer, kex->ex_log_bufferlen + 1);
1839 out2:
1840 kmem_free(kex->ex_path, kex->ex_pathlen + 1);
1841 out1:
1842 VN_RELE(vp);
1843 if (dvp != NULL)
1844 VN_RELE(dvp);
1845 mutex_destroy(&exi->exi_lock);
1846 rw_destroy(&exi->exi_cache_lock);
1847 for (i = 0; i < AUTH_TABLESIZE; i++) {
1848 avl_destroy(exi->exi_cache[i]);
1849 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
1850 }
1851
1852 kmem_free(exi, sizeof (*exi));
1853
1854 return (error);
1855 }
1856
1857 /*
1858 * Remove the exportinfo from the export list
1859 */
1860 void
1861 export_unlink(nfs_export_t *ne, struct exportinfo *exi)
1862 {
1863 ASSERT(RW_WRITE_HELD(&ne->exported_lock));
1864
1865 exp_hash_unlink(exi, fid_hash);
1866 exp_hash_unlink(exi, path_hash);
1867 }
1868
1869 /*
1870 * Unexport an exported filesystem
1871 */
1872 static int
1873 unexport(nfs_export_t *ne, struct exportinfo *exi)
1874 {
1875 struct secinfo cursec[MAX_FLAVORS];
1876 int curcnt;
1877
1878 rw_enter(&ne->exported_lock, RW_WRITER);
1879
1880 /* Check if exi is still linked in the export table */
1881 if (!EXP_LINKED(exi) || PSEUDO(exi)) {
1882 rw_exit(&ne->exported_lock);
1883 return (EINVAL);
1884 }
1885
1886 mutex_enter(&nfs_exi_id_lock);
1887 avl_remove(&exi_id_tree, exi);
1888 mutex_exit(&nfs_exi_id_lock);
1889 export_unlink(ne, exi);
1890
1891 /*
1892 * Remove security flavors before treeclimb_unexport() is called
1893 * because srv_secinfo_treeclimb needs the namespace tree
1894 */
1895 curcnt = build_seclist_nodups(&exi->exi_export, cursec, TRUE);
1896 srv_secinfo_treeclimb(ne, exi, cursec, curcnt, FALSE);
1897
1898 /*
1899 * If there's a visible list, then need to leave
1900 * a pseudo export here to retain the visible list
1901 * for paths to exports below.
1902 */
1903 if (exi->exi_visible != NULL) {
1904 struct exportinfo *newexi;
1905
1906 newexi = pseudo_exportfs(ne, exi->exi_vp, &exi->exi_fid,
1907 exi->exi_visible, &exi->exi_export);
1908 exi->exi_visible = NULL;
1909
1910 /* interconnect the existing treenode with the new exportinfo */
1911 newexi->exi_tree = exi->exi_tree;
1912 newexi->exi_tree->tree_exi = newexi;
1913
1914 /* Update the change timestamp */
1915 tree_update_change(ne, exi->exi_tree, NULL);
1916 } else {
1917 treeclimb_unexport(ne, exi);
1918 }
1919
1920 rw_exit(&ne->exported_lock);
1921
1922 /*
1923 * Need to call into the NFSv4 server and release all data
1924 * held on this particular export. This is important since
1925 * the v4 server may be holding file locks or vnodes under
1926 * this export.
1927 */
1928 rfs4_clean_state_exi(ne, exi);
1929
1930 /*
1931 * Notify the lock manager that the filesystem is being
1932 * unexported.
1933 */
1934 lm_unexport(exi);
1935
1936 /*
1937 * If this was a public export, restore
1938 * the public filehandle to the root.
1939 */
1940 if (exi == ne->exi_public) {
1941 ne->exi_public = ne->exi_root;
1942
1943 nfslog_share_record(ne->exi_public, CRED());
1944 }
1945
1946 if (exi->exi_export.ex_flags & EX_LOG)
1947 nfslog_unshare_record(exi, CRED());
1948
1949 exi_rele(exi);
1950 return (0);
1951 }
1952
1953 /*
1954 * Get file handle system call.
1955 * Takes file name and returns a file handle for it.
1956 * Credentials must be verified before calling.
1957 */
1958 int
1959 nfs_getfh(struct nfs_getfh_args *args, model_t model, cred_t *cr)
1960 {
1961 nfs_fh3 fh;
1962 char buf[NFS3_MAXFHSIZE];
1963 char *logptr, logbuf[NFS3_MAXFHSIZE];
1964 int l = NFS3_MAXFHSIZE;
1965 vnode_t *vp;
1966 vnode_t *dvp;
1967 struct exportinfo *exi;
1968 int error;
1969 int vers;
1970 STRUCT_HANDLE(nfs_getfh_args, uap);
1971
1972 #ifdef lint
1973 model = model; /* STRUCT macros don't always use it */
1974 #endif
1975
1976 STRUCT_SET_HANDLE(uap, model, args);
1977
1978 error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
1979 FOLLOW, &dvp, &vp);
1980 if (error == EINVAL) {
1981 /*
1982 * if fname resolves to / we get EINVAL error
1983 * since we wanted the parent vnode. Try again
1984 * with NULL dvp.
1985 */
1986 error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
1987 FOLLOW, NULL, &vp);
1988 dvp = NULL;
1989 }
1990 if (!error && vp == NULL) {
1991 /*
1992 * Last component of fname not found
1993 */
1994 if (dvp != NULL) {
1995 VN_RELE(dvp);
1996 }
1997 error = ENOENT;
1998 }
1999 if (error)
2000 return (error);
2001
2002 /*
2003 * 'vp' may be an AUTOFS node, so we perform a
2004 * VOP_ACCESS() to trigger the mount of the
2005 * intended filesystem, so we can share the intended
2006 * filesystem instead of the AUTOFS filesystem.
2007 */
2008 (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
2009
2010 /*
2011 * We're interested in the top most filesystem.
2012 * This is specially important when uap->dname is a trigger
2013 * AUTOFS node, since we're really interested in sharing the
2014 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
2015 * call not the AUTOFS node itself.
2016 */
2017 if (vn_mountedvfs(vp) != NULL) {
2018 if (error = traverse(&vp)) {
2019 VN_RELE(vp);
2020 if (dvp != NULL)
2021 VN_RELE(dvp);
2022 return (error);
2023 }
2024 }
2025
2026 vers = STRUCT_FGET(uap, vers);
2027 exi = nfs_vptoexi(dvp, vp, cr, NULL, &error, FALSE);
2028 if (!error) {
2029 if (vers == NFS_VERSION) {
2030 error = makefh((fhandle_t *)buf, vp, exi);
2031 l = NFS_FHSIZE;
2032 logptr = buf;
2033 } else if (vers == NFS_V3) {
2034 int i, sz, pad;
2035
2036 error = makefh3(&fh, vp, exi);
2037 l = RNDUP(fh.fh3_length);
2038 if (!error && (l > sizeof (fhandle3_t)))
2039 error = EREMOTE;
2040 logptr = logbuf;
2041 if (!error) {
2042 i = 0;
2043 sz = sizeof (fsid_t);
2044 bcopy(&fh.fh3_fsid, &buf[i], sz);
2045 i += sz;
2046
2047 /*
2048 * For backwards compatibility, the
2049 * fid length may be less than
2050 * NFS_FHMAXDATA, but it was always
2051 * encoded as NFS_FHMAXDATA bytes.
2052 */
2053
2054 sz = sizeof (ushort_t);
2055 bcopy(&fh.fh3_len, &buf[i], sz);
2056 i += sz;
2057 bcopy(fh.fh3_data, &buf[i], fh.fh3_len);
2058 i += fh.fh3_len;
2059 pad = (NFS_FHMAXDATA - fh.fh3_len);
2060 if (pad > 0) {
2061 bzero(&buf[i], pad);
2062 i += pad;
2063 l += pad;
2064 }
2065
2066 sz = sizeof (ushort_t);
2067 bcopy(&fh.fh3_xlen, &buf[i], sz);
2068 i += sz;
2069 bcopy(fh.fh3_xdata, &buf[i], fh.fh3_xlen);
2070 i += fh.fh3_xlen;
2071 pad = (NFS_FHMAXDATA - fh.fh3_xlen);
2072 if (pad > 0) {
2073 bzero(&buf[i], pad);
2074 i += pad;
2075 l += pad;
2076 }
2077 }
2078 /*
2079 * If we need to do NFS logging, the filehandle
2080 * must be downsized to 32 bytes.
2081 */
2082 if (!error && exi->exi_export.ex_flags & EX_LOG) {
2083 i = 0;
2084 sz = sizeof (fsid_t);
2085 bcopy(&fh.fh3_fsid, &logbuf[i], sz);
2086 i += sz;
2087 sz = sizeof (ushort_t);
2088 bcopy(&fh.fh3_len, &logbuf[i], sz);
2089 i += sz;
2090 sz = NFS_FHMAXDATA;
2091 bcopy(fh.fh3_data, &logbuf[i], sz);
2092 i += sz;
2093 sz = sizeof (ushort_t);
2094 bcopy(&fh.fh3_xlen, &logbuf[i], sz);
2095 i += sz;
2096 sz = NFS_FHMAXDATA;
2097 bcopy(fh.fh3_xdata, &logbuf[i], sz);
2098 i += sz;
2099 }
2100 }
2101 if (!error && exi->exi_export.ex_flags & EX_LOG) {
2102 nfslog_getfh(exi, (fhandle_t *)logptr,
2103 STRUCT_FGETP(uap, fname), UIO_USERSPACE, cr);
2104 }
2105 exi_rele(exi);
2106 if (!error) {
2107 if (copyout(&l, STRUCT_FGETP(uap, lenp), sizeof (int)))
2108 error = EFAULT;
2109 if (copyout(buf, STRUCT_FGETP(uap, fhp), l))
2110 error = EFAULT;
2111 }
2112 }
2113 VN_RELE(vp);
2114 if (dvp != NULL) {
2115 VN_RELE(dvp);
2116 }
2117 return (error);
2118 }
2119
2120 /*
2121 * Strategy: if vp is in the export list, then
2122 * return the associated file handle. Otherwise, ".."
2123 * once up the vp and try again, until the root of the
2124 * filesystem is reached.
2125 */
2126 struct exportinfo *
2127 nfs_vptoexi(vnode_t *dvp, vnode_t *vp, cred_t *cr, int *walk,
2128 int *err, bool_t v4srv)
2129 {
2130 fid_t fid;
2131 int error;
2132 struct exportinfo *exi;
2133
2134 ASSERT(vp);
2135 VN_HOLD(vp);
2136 if (dvp != NULL) {
2137 VN_HOLD(dvp);
2138 }
2139 if (walk != NULL)
2140 *walk = 0;
2141
2142 for (;;) {
2143 bzero(&fid, sizeof (fid));
2144 fid.fid_len = MAXFIDSZ;
2145 error = vop_fid_pseudo(vp, &fid);
2146 if (error) {
2147 /*
2148 * If vop_fid_pseudo returns ENOSPC then the fid
2149 * supplied is too small. For now we simply
2150 * return EREMOTE.
2151 */
2152 if (error == ENOSPC)
2153 error = EREMOTE;
2154 break;
2155 }
2156
2157 if (v4srv)
2158 exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
2159 else
2160 exi = checkexport(&vp->v_vfsp->vfs_fsid, &fid);
2161
2162 if (exi != NULL) {
2163 /*
2164 * Found the export info
2165 */
2166 break;
2167 }
2168
2169 /*
2170 * We have just failed finding a matching export.
2171 * If we're at the root of this filesystem, then
2172 * it's time to stop (with failure).
2173 */
2174 if (vp->v_flag & VROOT) {
2175 error = EINVAL;
2176 break;
2177 }
2178
2179 if (walk != NULL)
2180 (*walk)++;
2181
2182 /*
2183 * Now, do a ".." up vp. If dvp is supplied, use it,
2184 * otherwise, look it up.
2185 */
2186 if (dvp == NULL) {
2187 error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, cr,
2188 NULL, NULL, NULL);
2189 if (error)
2190 break;
2191 }
2192 VN_RELE(vp);
2193 vp = dvp;
2194 dvp = NULL;
2195 }
2196 VN_RELE(vp);
2197 if (dvp != NULL) {
2198 VN_RELE(dvp);
2199 }
2200 if (error != 0) {
2201 if (err != NULL)
2202 *err = error;
2203 return (NULL);
2204 }
2205 return (exi);
2206 }
2207
2208 int
2209 chk_clnt_sec(exportinfo_t *exi, struct svc_req *req)
2210 {
2211 int i, nfsflavor;
2212 struct secinfo *sp;
2213
2214 /*
2215 * Get the nfs flavor number from xprt.
2216 */
2217 nfsflavor = (int)(uintptr_t)req->rq_xprt->xp_cookie;
2218
2219 sp = exi->exi_export.ex_secinfo;
2220 for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2221 if ((nfsflavor == sp[i].s_secinfo.sc_nfsnum) &&
2222 SEC_REF_EXPORTED(sp + i))
2223 return (TRUE);
2224 }
2225 return (FALSE);
2226 }
2227
2228 /*
2229 * Make an fhandle from a vnode
2230 */
2231 int
2232 makefh(fhandle_t *fh, vnode_t *vp, exportinfo_t *exi)
2233 {
2234 int error;
2235
2236 *fh = exi->exi_fh; /* struct copy */
2237
2238 error = VOP_FID(vp, (fid_t *)&fh->fh_len, NULL);
2239 if (error) {
2240 /*
2241 * Should be something other than EREMOTE
2242 */
2243 return (EREMOTE);
2244 }
2245 return (0);
2246 }
2247
2248 /*
2249 * This routine makes an overloaded V2 fhandle which contains
2250 * sec modes.
2251 *
2252 * Note that the first four octets contain the length octet,
2253 * the status octet, and two padded octets to make them XDR
2254 * four-octet aligned.
2255 *
2256 * 1 2 3 4 32
2257 * +---+---+---+---+---+---+---+---+ +---+---+---+---+ +---+
2258 * | l | s | | | sec_1 |...| sec_n |...| |
2259 * +---+---+---+---+---+---+---+---+ +---+---+---+---+ +---+
2260 *
2261 * where
2262 *
2263 * the status octet s indicates whether there are more security
2264 * flavors (1 means yes, 0 means no) that require the client to
2265 * perform another 0x81 LOOKUP to get them,
2266 *
2267 * the length octet l is the length describing the number of
2268 * valid octets that follow. (l = 4 * n, where n is the number
2269 * of security flavors sent in the current overloaded filehandle.)
2270 *
2271 * sec_index should always be in the inclusive range: [1 - ex_seccnt],
2272 * and it tells server where to start within the secinfo array.
2273 * Usually it will always be 1; however, if more flavors are used
2274 * for the public export than can be encoded in the overloaded FH
2275 * (7 for NFS2), subsequent SNEGO MCLs will have a larger index
2276 * so the server will pick up where it left off from the previous
2277 * MCL reply.
2278 *
2279 * With NFS4 support, implicitly allowed flavors are also in
2280 * the secinfo array; however, they should not be returned in
2281 * SNEGO MCL replies.
2282 */
2283 int
2284 makefh_ol(fhandle_t *fh, exportinfo_t *exi, uint_t sec_index)
2285 {
2286 secinfo_t sec[MAX_FLAVORS];
2287 int totalcnt, i, *ipt, cnt, seccnt, secidx, fh_max_cnt;
2288 char *c;
2289
2290 if (fh == NULL || exi == NULL || sec_index < 1)
2291 return (EREMOTE);
2292
2293 /*
2294 * WebNFS clients need to know the unique set of explicitly
2295 * shared flavors in used for the public export. When
2296 * "TRUE" is passed to build_seclist_nodups(), only explicitly
2297 * shared flavors are included in the list.
2298 */
2299 seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2300 if (sec_index > seccnt)
2301 return (EREMOTE);
2302
2303 fh_max_cnt = (NFS_FHSIZE / sizeof (int)) - 1;
2304 totalcnt = seccnt - sec_index + 1;
2305 cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2306
2307 c = (char *)fh;
2308 /*
2309 * Encode the length octet representing the number of
2310 * security flavors (in bytes) in this overloaded fh.
2311 */
2312 *c = cnt * sizeof (int);
2313
2314 /*
2315 * Encode the status octet that indicates whether there
2316 * are more security flavors the client needs to get.
2317 */
2318 *(c + 1) = totalcnt > fh_max_cnt;
2319
2320 /*
2321 * put security flavors in the overloaded fh
2322 */
2323 ipt = (int *)(c + sizeof (int32_t));
2324 secidx = sec_index - 1;
2325 for (i = 0; i < cnt; i++) {
2326 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2327 }
2328 return (0);
2329 }
2330
2331 /*
2332 * Make an nfs_fh3 from a vnode
2333 */
2334 int
2335 makefh3(nfs_fh3 *fh, vnode_t *vp, struct exportinfo *exi)
2336 {
2337 int error;
2338 fid_t fid;
2339
2340 bzero(&fid, sizeof (fid));
2341 fid.fid_len = sizeof (fh->fh3_data);
2342 error = VOP_FID(vp, &fid, NULL);
2343 if (error)
2344 return (EREMOTE);
2345
2346 bzero(fh, sizeof (nfs_fh3));
2347 fh->fh3_fsid = exi->exi_fsid;
2348 fh->fh3_len = fid.fid_len;
2349 bcopy(fid.fid_data, fh->fh3_data, fh->fh3_len);
2350
2351 fh->fh3_xlen = exi->exi_fid.fid_len;
2352 ASSERT(fh->fh3_xlen <= sizeof (fh->fh3_xdata));
2353 bcopy(exi->exi_fid.fid_data, fh->fh3_xdata, fh->fh3_xlen);
2354
2355 fh->fh3_length = sizeof (fh->fh3_fsid)
2356 + sizeof (fh->fh3_len) + fh->fh3_len
2357 + sizeof (fh->fh3_xlen) + fh->fh3_xlen;
2358 fh->fh3_flags = 0;
2359
2360 return (0);
2361 }
2362
2363 /*
2364 * This routine makes an overloaded V3 fhandle which contains
2365 * sec modes.
2366 *
2367 * 1 4
2368 * +--+--+--+--+
2369 * | len |
2370 * +--+--+--+--+
2371 * up to 64
2372 * +--+--+--+--+--+--+--+--+--+--+--+--+ +--+--+--+--+
2373 * |s | | | | sec_1 | sec_2 | ... | sec_n |
2374 * +--+--+--+--+--+--+--+--+--+--+--+--+ +--+--+--+--+
2375 *
2376 * len = 4 * (n+1), where n is the number of security flavors
2377 * sent in the current overloaded filehandle.
2378 *
2379 * the status octet s indicates whether there are more security
2380 * mechanisms (1 means yes, 0 means no) that require the client
2381 * to perform another 0x81 LOOKUP to get them.
2382 *
2383 * Three octets are padded after the status octet.
2384 */
2385 int
2386 makefh3_ol(nfs_fh3 *fh, struct exportinfo *exi, uint_t sec_index)
2387 {
2388 secinfo_t sec[MAX_FLAVORS];
2389 int totalcnt, cnt, *ipt, i, seccnt, fh_max_cnt, secidx;
2390 char *c;
2391
2392 if (fh == NULL || exi == NULL || sec_index < 1)
2393 return (EREMOTE);
2394
2395 /*
2396 * WebNFS clients need to know the unique set of explicitly
2397 * shared flavors in used for the public export. When
2398 * "TRUE" is passed to build_seclist_nodups(), only explicitly
2399 * shared flavors are included in the list.
2400 */
2401 seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2402
2403 if (sec_index > seccnt)
2404 return (EREMOTE);
2405
2406 fh_max_cnt = (NFS3_FHSIZE / sizeof (int)) - 1;
2407 totalcnt = seccnt - sec_index + 1;
2408 cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2409
2410 /*
2411 * Place the length in fh3_length representing the number
2412 * of security flavors (in bytes) in this overloaded fh.
2413 */
2414 fh->fh3_flags = FH_WEBNFS;
2415 fh->fh3_length = (cnt+1) * sizeof (int32_t);
2416
2417 c = (char *)&fh->fh3_u.nfs_fh3_i.fh3_i;
2418 /*
2419 * Encode the status octet that indicates whether there
2420 * are more security flavors the client needs to get.
2421 */
2422 *c = totalcnt > fh_max_cnt;
2423
2424 /*
2425 * put security flavors in the overloaded fh
2426 */
2427 secidx = sec_index - 1;
2428 ipt = (int *)(c + sizeof (int32_t));
2429 for (i = 0; i < cnt; i++) {
2430 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2431 }
2432 return (0);
2433 }
2434
2435 /*
2436 * Make an nfs_fh4 from a vnode
2437 */
2438 int
2439 makefh4(nfs_fh4 *fh, vnode_t *vp, struct exportinfo *exi)
2440 {
2441 int error;
2442 nfs_fh4_fmt_t *fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2443 fid_t fid;
2444
2445 bzero(&fid, sizeof (fid));
2446 fid.fid_len = MAXFIDSZ;
2447 /*
2448 * vop_fid_pseudo() is used to set up NFSv4 namespace, so
2449 * use vop_fid_pseudo() here to get the fid instead of VOP_FID.
2450 */
2451 error = vop_fid_pseudo(vp, &fid);
2452 if (error)
2453 return (error);
2454
2455 fh->nfs_fh4_len = NFS_FH4_LEN;
2456
2457 fh_fmtp->fh4_i.fhx_fsid = exi->exi_fh.fh_fsid;
2458 fh_fmtp->fh4_i.fhx_xlen = exi->exi_fh.fh_xlen;
2459
2460 bzero(fh_fmtp->fh4_i.fhx_data, sizeof (fh_fmtp->fh4_i.fhx_data));
2461 bzero(fh_fmtp->fh4_i.fhx_xdata, sizeof (fh_fmtp->fh4_i.fhx_xdata));
2462 ASSERT(exi->exi_fh.fh_xlen <= sizeof (fh_fmtp->fh4_i.fhx_xdata));
2463 bcopy(exi->exi_fh.fh_xdata, fh_fmtp->fh4_i.fhx_xdata,
2464 exi->exi_fh.fh_xlen);
2465
2466 fh_fmtp->fh4_len = fid.fid_len;
2467 ASSERT(fid.fid_len <= sizeof (fh_fmtp->fh4_data));
2468 bcopy(fid.fid_data, fh_fmtp->fh4_data, fid.fid_len);
2469 fh_fmtp->fh4_flag = 0;
2470
2471 #ifdef VOLATILE_FH_TEST
2472 /*
2473 * XXX (temporary?)
2474 * Use the rnode volatile_id value to add volatility to the fh.
2475 *
2476 * For testing purposes there are currently two scenarios, based
2477 * on whether the filesystem was shared with "volatile_fh"
2478 * or "expire_on_rename". In the first case, use the value of
2479 * export struct share_time as the volatile_id. In the second
2480 * case use the vnode volatile_id value (which is set to the
2481 * time in which the file was renamed).
2482 *
2483 * Note that the above are temporary constructs for testing only
2484 * XXX
2485 */
2486 if (exi->exi_export.ex_flags & EX_VOLRNM) {
2487 fh_fmtp->fh4_volatile_id = find_volrnm_fh_id(exi, fh);
2488 } else if (exi->exi_export.ex_flags & EX_VOLFH) {
2489 fh_fmtp->fh4_volatile_id = exi->exi_volatile_id;
2490 } else {
2491 fh_fmtp->fh4_volatile_id = 0;
2492 }
2493 #endif /* VOLATILE_FH_TEST */
2494
2495 return (0);
2496 }
2497
2498 /*
2499 * Convert an fhandle into a vnode.
2500 * Uses the file id (fh_len + fh_data) in the fhandle to get the vnode.
2501 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2502 * are done with it.
2503 */
2504 vnode_t *
2505 nfs_fhtovp(fhandle_t *fh, struct exportinfo *exi)
2506 {
2507 vfs_t *vfsp;
2508 vnode_t *vp;
2509 int error;
2510 fid_t *fidp;
2511
2512 TRACE_0(TR_FAC_NFS, TR_FHTOVP_START,
2513 "fhtovp_start");
2514
2515 if (exi == NULL) {
2516 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2517 "fhtovp_end:(%S)", "exi NULL");
2518 return (NULL); /* not exported */
2519 }
2520
2521 ASSERT(exi->exi_vp != NULL);
2522
2523 if (PUBLIC_FH2(fh)) {
2524 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2525 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2526 "fhtovp_end:(%S)", "root not exported");
2527 return (NULL);
2528 }
2529 vp = exi->exi_vp;
2530 VN_HOLD(vp);
2531 return (vp);
2532 }
2533
2534 vfsp = exi->exi_vp->v_vfsp;
2535 ASSERT(vfsp != NULL);
2536 fidp = (fid_t *)&fh->fh_len;
2537
2538 error = VFS_VGET(vfsp, &vp, fidp);
2539 if (error || vp == NULL) {
2540 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2541 "fhtovp_end:(%S)", "VFS_GET failed or vp NULL");
2542 return (NULL);
2543 }
2544 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2545 "fhtovp_end:(%S)", "end");
2546 return (vp);
2547 }
2548
2549 /*
2550 * Convert an nfs_fh3 into a vnode.
2551 * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2552 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2553 * are done with it.
2554 */
2555 vnode_t *
2556 nfs3_fhtovp(nfs_fh3 *fh, struct exportinfo *exi)
2557 {
2558 vfs_t *vfsp;
2559 vnode_t *vp;
2560 int error;
2561 fid_t *fidp;
2562
2563 if (exi == NULL)
2564 return (NULL); /* not exported */
2565
2566 ASSERT(exi->exi_vp != NULL);
2567
2568 if (PUBLIC_FH3(fh)) {
2569 if (exi->exi_export.ex_flags & EX_PUBLIC)
2570 return (NULL);
2571 vp = exi->exi_vp;
2572 VN_HOLD(vp);
2573 return (vp);
2574 }
2575
2576 if (fh->fh3_length < NFS3_OLDFHSIZE ||
2577 fh->fh3_length > NFS3_MAXFHSIZE)
2578 return (NULL);
2579
2580 vfsp = exi->exi_vp->v_vfsp;
2581 ASSERT(vfsp != NULL);
2582 fidp = FH3TOFIDP(fh);
2583
2584 error = VFS_VGET(vfsp, &vp, fidp);
2585 if (error || vp == NULL)
2586 return (NULL);
2587
2588 return (vp);
2589 }
2590
2591 /*
2592 * Convert an nfs_fh4 into a vnode.
2593 * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2594 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2595 * are done with it.
2596 */
2597 vnode_t *
2598 nfs4_fhtovp(nfs_fh4 *fh, struct exportinfo *exi, nfsstat4 *statp)
2599 {
2600 vfs_t *vfsp;
2601 vnode_t *vp = NULL;
2602 int error;
2603 fid_t *fidp;
2604 nfs_fh4_fmt_t *fh_fmtp;
2605 #ifdef VOLATILE_FH_TEST
2606 uint32_t volatile_id = 0;
2607 #endif /* VOLATILE_FH_TEST */
2608
2609 if (exi == NULL) {
2610 *statp = NFS4ERR_STALE;
2611 return (NULL); /* not exported */
2612 }
2613 ASSERT(exi->exi_vp != NULL);
2614
2615 /* caller should have checked this */
2616 ASSERT(fh->nfs_fh4_len >= NFS_FH4_LEN);
2617
2618 fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2619 vfsp = exi->exi_vp->v_vfsp;
2620 ASSERT(vfsp != NULL);
2621 fidp = (fid_t *)&fh_fmtp->fh4_len;
2622
2623 #ifdef VOLATILE_FH_TEST
2624 /* XXX check if volatile - should be changed later */
2625 if (exi->exi_export.ex_flags & (EX_VOLRNM | EX_VOLFH)) {
2626 /*
2627 * Filesystem is shared with volatile filehandles
2628 */
2629 if (exi->exi_export.ex_flags & EX_VOLRNM)
2630 volatile_id = find_volrnm_fh_id(exi, fh);
2631 else
2632 volatile_id = exi->exi_volatile_id;
2633
2634 if (fh_fmtp->fh4_volatile_id != volatile_id) {
2635 *statp = NFS4ERR_FHEXPIRED;
2636 return (NULL);
2637 }
2638 }
2639 /*
2640 * XXX even if test_volatile_fh false, the fh may contain a
2641 * volatile id if obtained when the test was set.
2642 */
2643 fh_fmtp->fh4_volatile_id = (uchar_t)0;
2644 #endif /* VOLATILE_FH_TEST */
2645
2646 error = VFS_VGET(vfsp, &vp, fidp);
2647 /*
2648 * If we can not get vp from VFS_VGET, perhaps this is
2649 * an nfs v2/v3/v4 node in an nfsv4 pseudo filesystem.
2650 * Check it out.
2651 */
2652 if (error && PSEUDO(exi))
2653 error = nfs4_vget_pseudo(exi, &vp, fidp);
2654
2655 if (error || vp == NULL) {
2656 *statp = NFS4ERR_STALE;
2657 return (NULL);
2658 }
2659 /* XXX - disgusting hack */
2660 if (vp->v_type == VNON && vp->v_flag & V_XATTRDIR)
2661 vp->v_type = VDIR;
2662 *statp = NFS4_OK;
2663 return (vp);
2664 }
2665
2666 /*
2667 * Find the export structure associated with the given filesystem.
2668 * If found, then increment the ref count (exi_count).
2669 */
2670 struct exportinfo *
2671 checkexport(fsid_t *fsid, fid_t *fid)
2672 {
2673 struct exportinfo *exi;
2674 nfs_export_t *ne = nfs_get_export();
2675
2676 rw_enter(&ne->exported_lock, RW_READER);
2677 for (exi = ne->exptable[exptablehash(fsid, fid)];
2678 exi != NULL;
2679 exi = exi->fid_hash.next) {
2680 if (exportmatch(exi, fsid, fid)) {
2681 /*
2682 * If this is the place holder for the
2683 * public file handle, then return the
2684 * real export entry for the public file
2685 * handle.
2686 */
2687 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2688 exi = ne->exi_public;
2689 }
2690
2691 exi_hold(exi);
2692 rw_exit(&ne->exported_lock);
2693 return (exi);
2694 }
2695 }
2696 rw_exit(&ne->exported_lock);
2697 return (NULL);
2698 }
2699
2700
2701 /*
2702 * "old school" version of checkexport() for NFS4. NFS4
2703 * rfs4_compound holds exported_lock for duration of compound
2704 * processing. This version doesn't manipulate exi_count
2705 * since NFS4 breaks fundamental assumptions in the exi_count
2706 * design.
2707 */
2708 struct exportinfo *
2709 checkexport4(fsid_t *fsid, fid_t *fid, vnode_t *vp)
2710 {
2711 struct exportinfo *exi;
2712 nfs_export_t *ne = nfs_get_export();
2713
2714 ASSERT(RW_LOCK_HELD(&ne->exported_lock));
2715
2716 for (exi = ne->exptable[exptablehash(fsid, fid)];
2717 exi != NULL;
2718 exi = exi->fid_hash.next) {
2719 if (exportmatch(exi, fsid, fid)) {
2720 /*
2721 * If this is the place holder for the
2722 * public file handle, then return the
2723 * real export entry for the public file
2724 * handle.
2725 */
2726 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2727 exi = ne->exi_public;
2728 }
2729
2730 /*
2731 * If vp is given, check if vp is the
2732 * same vnode as the exported node.
2733 *
2734 * Since VOP_FID of a lofs node returns the
2735 * fid of its real node (ufs), the exported
2736 * node for lofs and (pseudo) ufs may have
2737 * the same fsid and fid.
2738 */
2739 if (vp == NULL || vp == exi->exi_vp)
2740 return (exi);
2741 }
2742 }
2743
2744 return (NULL);
2745 }
2746
2747 /*
2748 * Free an entire export list node
2749 */
2750 void
2751 exportfree(struct exportinfo *exi)
2752 {
2753 struct exportdata *ex;
2754 struct charset_cache *cache;
2755 int i;
2756
2757 ex = &exi->exi_export;
2758
2759 ASSERT(exi->exi_vp != NULL && !(exi->exi_export.ex_flags & EX_PUBLIC));
2760 VN_RELE(exi->exi_vp);
2761 if (exi->exi_dvp != NULL)
2762 VN_RELE(exi->exi_dvp);
2763
2764 if (ex->ex_flags & EX_INDEX)
2765 kmem_free(ex->ex_index, strlen(ex->ex_index) + 1);
2766
2767 kmem_free(ex->ex_path, ex->ex_pathlen + 1);
2768 nfsauth_cache_free(exi);
2769
2770 /*
2771 * if there is a character set mapping cached, clean it up.
2772 */
2773 for (cache = exi->exi_charset; cache != NULL;
2774 cache = exi->exi_charset) {
2775 if (cache->inbound != (kiconv_t)-1)
2776 (void) kiconv_close(cache->inbound);
2777 if (cache->outbound != (kiconv_t)-1)
2778 (void) kiconv_close(cache->outbound);
2779 exi->exi_charset = cache->next;
2780 kmem_free(cache, sizeof (struct charset_cache));
2781 }
2782
2783 if (exi->exi_logbuffer != NULL)
2784 nfslog_disable(exi);
2785
2786 if (ex->ex_flags & EX_LOG) {
2787 kmem_free(ex->ex_log_buffer, ex->ex_log_bufferlen + 1);
2788 kmem_free(ex->ex_tag, ex->ex_taglen + 1);
2789 }
2790
2791 if (exi->exi_visible)
2792 free_visible(exi->exi_visible);
2793
2794 srv_secinfo_list_free(ex->ex_secinfo, ex->ex_seccnt);
2795
2796 #ifdef VOLATILE_FH_TEST
2797 free_volrnm_list(exi);
2798 mutex_destroy(&exi->exi_vol_rename_lock);
2799 #endif /* VOLATILE_FH_TEST */
2800
2801 mutex_destroy(&exi->exi_lock);
2802 rw_destroy(&exi->exi_cache_lock);
2803 /*
2804 * All nodes in the exi_cache AVL trees were removed and freed in the
2805 * nfsauth_cache_free() call above. We will just destroy and free the
2806 * empty AVL trees here.
2807 */
2808 for (i = 0; i < AUTH_TABLESIZE; i++) {
2809 avl_destroy(exi->exi_cache[i]);
2810 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
2811 }
2812
2813 kmem_free(exi, sizeof (*exi));
2814 }
2815
2816 /*
2817 * load the index file from user space into kernel space.
2818 */
2819 static int
2820 loadindex(struct exportdata *kex)
2821 {
2822 int error;
2823 char index[MAXNAMELEN+1];
2824 size_t len;
2825
2826 /*
2827 * copyinstr copies the complete string including the NULL and
2828 * returns the len with the NULL byte included in the calculation
2829 * as long as the max length is not exceeded.
2830 */
2831 if (error = copyinstr(kex->ex_index, index, sizeof (index), &len))
2832 return (error);
2833
2834 kex->ex_index = kmem_alloc(len, KM_SLEEP);
2835 bcopy(index, kex->ex_index, len);
2836
2837 return (0);
2838 }
2839
2840 void
2841 exi_hold(struct exportinfo *exi)
2842 {
2843 mutex_enter(&exi->exi_lock);
2844 exi->exi_count++;
2845 mutex_exit(&exi->exi_lock);
2846 }
2847
2848 /*
2849 * When a thread completes using exi, it should call exi_rele().
2850 * exi_rele() decrements exi_count. It releases exi if exi_count == 0, i.e.
2851 * if this is the last user of exi and exi is not on exportinfo list anymore
2852 */
2853 void
2854 exi_rele(struct exportinfo *exi)
2855 {
2856 mutex_enter(&exi->exi_lock);
2857 exi->exi_count--;
2858 if (exi->exi_count == 0) {
2859 mutex_exit(&exi->exi_lock);
2860 exportfree(exi);
2861 } else
2862 mutex_exit(&exi->exi_lock);
2863 }
2864
2865 #ifdef VOLATILE_FH_TEST
2866 /*
2867 * Test for volatile fh's - add file handle to list and set its volatile id
2868 * to time it was renamed. If EX_VOLFH is also on and the fs is reshared,
2869 * the vol_rename queue is purged.
2870 *
2871 * XXX This code is for unit testing purposes only... To correctly use it, it
2872 * needs to tie a rename list to the export struct and (more
2873 * important), protect access to the exi rename list using a write lock.
2874 */
2875
2876 /*
2877 * get the fh vol record if it's in the volatile on rename list. Don't check
2878 * volatile_id in the file handle - compare only the file handles.
2879 */
2880 static struct ex_vol_rename *
2881 find_volrnm_fh(struct exportinfo *exi, nfs_fh4 *fh4p)
2882 {
2883 struct ex_vol_rename *p = NULL;
2884 fhandle4_t *fhp;
2885
2886 /* XXX shouldn't we assert &exported_lock held? */
2887 ASSERT(MUTEX_HELD(&exi->exi_vol_rename_lock));
2888
2889 if (fh4p->nfs_fh4_len != NFS_FH4_LEN) {
2890 return (NULL);
2891 }
2892 fhp = &((nfs_fh4_fmt_t *)fh4p->nfs_fh4_val)->fh4_i;
2893 for (p = exi->exi_vol_rename; p != NULL; p = p->vrn_next) {
2894 if (bcmp(fhp, &p->vrn_fh_fmt.fh4_i,
2895 sizeof (fhandle4_t)) == 0)
2896 break;
2897 }
2898 return (p);
2899 }
2900
2901 /*
2902 * get the volatile id for the fh (if there is - else return 0). Ignore the
2903 * volatile_id in the file handle - compare only the file handles.
2904 */
2905 static uint32_t
2906 find_volrnm_fh_id(struct exportinfo *exi, nfs_fh4 *fh4p)
2907 {
2908 struct ex_vol_rename *p;
2909 uint32_t volatile_id;
2910
2911 mutex_enter(&exi->exi_vol_rename_lock);
2912 p = find_volrnm_fh(exi, fh4p);
2913 volatile_id = (p ? p->vrn_fh_fmt.fh4_volatile_id :
2914 exi->exi_volatile_id);
2915 mutex_exit(&exi->exi_vol_rename_lock);
2916 return (volatile_id);
2917 }
2918
2919 /*
2920 * Free the volatile on rename list - will be called if a filesystem is
2921 * unshared or reshared without EX_VOLRNM
2922 */
2923 static void
2924 free_volrnm_list(struct exportinfo *exi)
2925 {
2926 struct ex_vol_rename *p, *pnext;
2927
2928 /* no need to hold mutex lock - this one is called from exportfree */
2929 for (p = exi->exi_vol_rename; p != NULL; p = pnext) {
2930 pnext = p->vrn_next;
2931 kmem_free(p, sizeof (*p));
2932 }
2933 exi->exi_vol_rename = NULL;
2934 }
2935
2936 /*
2937 * Add a file handle to the volatile on rename list.
2938 */
2939 void
2940 add_volrnm_fh(struct exportinfo *exi, vnode_t *vp)
2941 {
2942 struct ex_vol_rename *p;
2943 char fhbuf[NFS4_FHSIZE];
2944 nfs_fh4 fh4;
2945 int error;
2946
2947 fh4.nfs_fh4_val = fhbuf;
2948 error = makefh4(&fh4, vp, exi);
2949 if ((error) || (fh4.nfs_fh4_len != sizeof (p->vrn_fh_fmt))) {
2950 return;
2951 }
2952
2953 mutex_enter(&exi->exi_vol_rename_lock);
2954
2955 p = find_volrnm_fh(exi, &fh4);
2956
2957 if (p == NULL) {
2958 p = kmem_alloc(sizeof (*p), KM_SLEEP);
2959 bcopy(fh4.nfs_fh4_val, &p->vrn_fh_fmt, sizeof (p->vrn_fh_fmt));
2960 p->vrn_next = exi->exi_vol_rename;
2961 exi->exi_vol_rename = p;
2962 }
2963
2964 p->vrn_fh_fmt.fh4_volatile_id = gethrestime_sec();
2965 mutex_exit(&exi->exi_vol_rename_lock);
2966 }
2967
2968 #endif /* VOLATILE_FH_TEST */