1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25
26 /*
27 * Copyright 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T.
28 * All rights reserved.
29 */
30
31 /*
32 * Copyright 2018 Nexenta Systems, Inc.
33 */
34
35 #include <sys/types.h>
36 #include <sys/param.h>
37 #include <sys/time.h>
38 #include <sys/vfs.h>
39 #include <sys/vnode.h>
40 #include <sys/socket.h>
41 #include <sys/errno.h>
42 #include <sys/uio.h>
43 #include <sys/proc.h>
44 #include <sys/user.h>
45 #include <sys/file.h>
46 #include <sys/tiuser.h>
47 #include <sys/kmem.h>
48 #include <sys/pathname.h>
49 #include <sys/debug.h>
50 #include <sys/vtrace.h>
51 #include <sys/cmn_err.h>
52 #include <sys/acl.h>
53 #include <sys/utsname.h>
54 #include <sys/sdt.h>
55 #include <netinet/in.h>
56 #include <sys/avl.h>
57
58 #include <rpc/types.h>
59 #include <rpc/auth.h>
60 #include <rpc/svc.h>
61
62 #include <nfs/nfs.h>
63 #include <nfs/export.h>
64 #include <nfs/nfssys.h>
65 #include <nfs/nfs_clnt.h>
66 #include <nfs/nfs_acl.h>
67 #include <nfs/nfs_log.h>
68 #include <nfs/lm.h>
69 #include <sys/sunddi.h>
70
71 static zone_key_t nfs_export_key;
72
73 /*
74 * exi_id support
75 *
76 * exi_id_next The next exi_id available.
77 * exi_id_overflow The exi_id_next already overflowed, so we should
78 * thoroughly check for duplicates.
79 * exi_id_tree AVL tree indexed by exi_id.
80 * nfs_exi_id_lock Lock to protect the export ID list
81 *
82 * All exi_id_next, exi_id_overflow, and exi_id_tree are protected by
83 * nfs_exi_id_lock.
84 */
85 static int exi_id_next;
86 static bool_t exi_id_overflow;
87 avl_tree_t exi_id_tree;
88 kmutex_t nfs_exi_id_lock;
89
90 static int unexport(nfs_export_t *, exportinfo_t *);
91 static void exportfree(exportinfo_t *);
92 static int loadindex(exportdata_t *);
93
94 extern void nfsauth_cache_free(exportinfo_t *);
95 extern int sec_svc_loadrootnames(int, int, caddr_t **, model_t);
96 extern void sec_svc_freerootnames(int, int, caddr_t *);
97
98 static int build_seclist_nodups(exportdata_t *, secinfo_t *, int);
99 static void srv_secinfo_add(secinfo_t **, int *, secinfo_t *, int, int);
100 static void srv_secinfo_remove(secinfo_t **, int *, secinfo_t *, int);
101 static void srv_secinfo_treeclimb(nfs_export_t *, exportinfo_t *,
102 secinfo_t *, int, bool_t);
103
104 #ifdef VOLATILE_FH_TEST
105 static struct ex_vol_rename *find_volrnm_fh(exportinfo_t *, nfs_fh4 *);
106 static uint32_t find_volrnm_fh_id(exportinfo_t *, nfs_fh4 *);
107 static void free_volrnm_list(exportinfo_t *);
108 #endif /* VOLATILE_FH_TEST */
109
110 fhandle_t nullfh2; /* for comparing V2 filehandles */
111
112 /*
113 * macro for static dtrace probes to trace server namespace ref count mods.
114 */
115 #define SECREF_TRACE(seclist, tag, flav, aftcnt) \
116 DTRACE_PROBE4(nfss__i__nmspc__secref, struct secinfo *, (seclist), \
117 char *, (tag), int, (int)(flav), int, (int)(aftcnt))
118
119
120 #define exptablehash(fsid, fid) (nfs_fhhash((fsid), (fid)) & (EXPTABLESIZE - 1))
121
122 extern nfs_export_t *
123 nfs_get_export(void)
124 {
125 return (zone_getspecific(nfs_export_key, curzone));
126 }
127
128 static uint8_t
129 xor_hash(uint8_t *data, int len)
130 {
131 uint8_t h = 0;
132
133 while (len--)
134 h ^= *data++;
135
136 return (h);
137 }
138
139 /*
140 * File handle hash function, XOR over all bytes in fsid and fid.
141 */
142 static unsigned
143 nfs_fhhash(fsid_t *fsid, fid_t *fid)
144 {
145 int len;
146 uint8_t h;
147
148 h = xor_hash((uint8_t *)fsid, sizeof (fsid_t));
149
150 /*
151 * Sanity check the length before using it
152 * blindly in case the client trashed it.
153 */
154 len = fid->fid_len > NFS_FH4MAXDATA ? 0 : fid->fid_len;
155 h ^= xor_hash((uint8_t *)fid->fid_data, len);
156
157 return ((unsigned)h);
158 }
159
160 /*
161 * Free the memory allocated within a secinfo entry.
162 */
163 void
164 srv_secinfo_entry_free(struct secinfo *secp)
165 {
166 if (secp->s_rootcnt > 0 && secp->s_rootnames != NULL) {
167 sec_svc_freerootnames(secp->s_secinfo.sc_rpcnum,
168 secp->s_rootcnt, secp->s_rootnames);
169 secp->s_rootcnt = 0;
170 }
171
172 if ((secp->s_secinfo.sc_rpcnum == RPCSEC_GSS) &&
173 (secp->s_secinfo.sc_gss_mech_type)) {
174 kmem_free(secp->s_secinfo.sc_gss_mech_type->elements,
175 secp->s_secinfo.sc_gss_mech_type->length);
176 kmem_free(secp->s_secinfo.sc_gss_mech_type,
177 sizeof (rpc_gss_OID_desc));
178 secp->s_secinfo.sc_gss_mech_type = NULL;
179 }
180 }
181
182 /*
183 * Free a list of secinfo allocated in the exportdata structure.
184 */
185 void
186 srv_secinfo_list_free(struct secinfo *secinfo, int cnt)
187 {
188 int i;
189
190 if (cnt == 0)
191 return;
192
193 for (i = 0; i < cnt; i++)
194 srv_secinfo_entry_free(&secinfo[i]);
195
196 kmem_free(secinfo, cnt * sizeof (struct secinfo));
197 }
198
199 /*
200 * Allocate and copy a secinfo data from "from" to "to".
201 *
202 * This routine is used by srv_secinfo_add() to add a new flavor to an
203 * ancestor's export node. The rootnames are not copied because the
204 * allowable rootname access only applies to the explicit exported node,
205 * not its ancestor's.
206 *
207 * "to" should have already been allocated and zeroed before calling
208 * this routine.
209 *
210 * This routine is used under the protection of exported_lock (RW_WRITER).
211 */
212 void
213 srv_secinfo_copy(struct secinfo *from, struct secinfo *to)
214 {
215 to->s_secinfo.sc_nfsnum = from->s_secinfo.sc_nfsnum;
216 to->s_secinfo.sc_rpcnum = from->s_secinfo.sc_rpcnum;
217
218 if (from->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
219 to->s_secinfo.sc_service = from->s_secinfo.sc_service;
220 bcopy(from->s_secinfo.sc_name, to->s_secinfo.sc_name,
221 strlen(from->s_secinfo.sc_name));
222 bcopy(from->s_secinfo.sc_gss_mech, to->s_secinfo.sc_gss_mech,
223 strlen(from->s_secinfo.sc_gss_mech));
224
225 /* copy mechanism oid */
226 to->s_secinfo.sc_gss_mech_type =
227 kmem_alloc(sizeof (rpc_gss_OID_desc), KM_SLEEP);
228 to->s_secinfo.sc_gss_mech_type->length =
229 from->s_secinfo.sc_gss_mech_type->length;
230 to->s_secinfo.sc_gss_mech_type->elements =
231 kmem_alloc(from->s_secinfo.sc_gss_mech_type->length,
232 KM_SLEEP);
233 bcopy(from->s_secinfo.sc_gss_mech_type->elements,
234 to->s_secinfo.sc_gss_mech_type->elements,
235 from->s_secinfo.sc_gss_mech_type->length);
236 }
237
238 to->s_refcnt = from->s_refcnt;
239 to->s_window = from->s_window;
240 /* no need to copy the mode bits - s_flags */
241 }
242
243 /*
244 * Create a secinfo array without duplicates. The condensed
245 * flavor list is used to propagate flavor ref counts to an
246 * export's ancestor pseudonodes.
247 */
248 static int
249 build_seclist_nodups(exportdata_t *exd, secinfo_t *nodups, int exponly)
250 {
251 int ccnt, c;
252 int ncnt, n;
253 struct secinfo *cursec;
254
255 ncnt = 0;
256 ccnt = exd->ex_seccnt;
257 cursec = exd->ex_secinfo;
258
259 for (c = 0; c < ccnt; c++) {
260
261 if (exponly && ! SEC_REF_EXPORTED(&cursec[c]))
262 continue;
263
264 for (n = 0; n < ncnt; n++) {
265 if (nodups[n].s_secinfo.sc_nfsnum ==
266 cursec[c].s_secinfo.sc_nfsnum)
267 break;
268 }
269
270 /*
271 * The structure copy below also copys ptrs embedded
272 * within struct secinfo. The ptrs are copied but
273 * they are never freed from the nodups array. If
274 * an ancestor's secinfo array doesn't contain one
275 * of the nodups flavors, then the entry is properly
276 * copied into the ancestor's secinfo array.
277 * (see srv_secinfo_copy)
278 */
279 if (n == ncnt) {
280 nodups[n] = cursec[c];
281 ncnt++;
282 }
283 }
284 return (ncnt);
285 }
286
287 /*
288 * Add the new security flavors from newdata to the current list, pcursec.
289 * Upon return, *pcursec has the newly merged secinfo list.
290 *
291 * There should be at least 1 secinfo entry in newsec.
292 *
293 * This routine is used under the protection of exported_lock (RW_WRITER).
294 */
295 static void
296 srv_secinfo_add(secinfo_t **pcursec, int *pcurcnt, secinfo_t *newsec,
297 int newcnt, int is_pseudo)
298 {
299 int ccnt, c; /* sec count in current data - curdata */
300 int n; /* index for newsec - newsecinfo */
301 int tcnt; /* total sec count after merge */
302 int mcnt; /* total sec count after merge */
303 struct secinfo *msec; /* merged secinfo list */
304 struct secinfo *cursec;
305
306 cursec = *pcursec;
307 ccnt = *pcurcnt;
308
309 ASSERT(newcnt > 0);
310 tcnt = ccnt + newcnt;
311
312 for (n = 0; n < newcnt; n++) {
313 for (c = 0; c < ccnt; c++) {
314 if (newsec[n].s_secinfo.sc_nfsnum ==
315 cursec[c].s_secinfo.sc_nfsnum) {
316 cursec[c].s_refcnt += newsec[n].s_refcnt;
317 SECREF_TRACE(cursec, "add_ref",
318 cursec[c].s_secinfo.sc_nfsnum,
319 cursec[c].s_refcnt);
320 tcnt--;
321 break;
322 }
323 }
324 }
325
326 if (tcnt == ccnt)
327 return; /* no change; no new flavors */
328
329 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
330
331 /* move current secinfo list data to the new list */
332 for (c = 0; c < ccnt; c++)
333 msec[c] = cursec[c];
334
335 /* Add the flavor that's not in the current data */
336 mcnt = ccnt;
337 for (n = 0; n < newcnt; n++) {
338 for (c = 0; c < ccnt; c++) {
339 if (newsec[n].s_secinfo.sc_nfsnum ==
340 cursec[c].s_secinfo.sc_nfsnum)
341 break;
342 }
343
344 /* This is the one. Add it. */
345 if (c == ccnt) {
346 srv_secinfo_copy(&newsec[n], &msec[mcnt]);
347
348 if (is_pseudo)
349 msec[mcnt].s_flags = M_RO;
350
351 SECREF_TRACE(msec, "new_ref",
352 msec[mcnt].s_secinfo.sc_nfsnum,
353 msec[mcnt].s_refcnt);
354 mcnt++;
355 }
356 }
357
358 ASSERT(mcnt == tcnt);
359
360 /*
361 * Done. Update curdata. Free the old secinfo list in
362 * curdata and return the new sec array info
363 */
364 if (ccnt > 0)
365 kmem_free(cursec, ccnt * sizeof (struct secinfo));
366 *pcurcnt = tcnt;
367 *pcursec = msec;
368 }
369
370 /*
371 * For NFS V4.
372 * Remove the security data of the unexported node from its ancestors.
373 * Assume there is at least one flavor entry in the current sec list
374 * (pcursec).
375 *
376 * This routine is used under the protection of exported_lock (RW_WRITER).
377 *
378 * Every element of remsec is an explicitly exported flavor. If
379 * srv_secinfo_remove() is called fom an exportfs error path, then
380 * the flavor list was derived from the user's share cmdline,
381 * and all flavors are explicit. If it was called from the unshare path,
382 * build_seclist_nodups() was called with the exponly flag.
383 */
384 static void
385 srv_secinfo_remove(secinfo_t **pcursec, int *pcurcnt, secinfo_t *remsec,
386 int remcnt)
387 {
388 int ccnt, c; /* sec count in current data - cursec */
389 int r; /* sec count in removal data - remsec */
390 int tcnt, mcnt; /* total sec count after removing */
391 struct secinfo *msec; /* final secinfo list after removing */
392 struct secinfo *cursec;
393
394 cursec = *pcursec;
395 ccnt = *pcurcnt;
396 tcnt = ccnt;
397
398 for (r = 0; r < remcnt; r++) {
399 /*
400 * At unshare/reshare time, only explicitly shared flavor ref
401 * counts are decremented and propagated to ancestors.
402 * Implicit flavor refs came from shared descendants, and
403 * they must be kept.
404 */
405 if (! SEC_REF_EXPORTED(&remsec[r]))
406 continue;
407
408 for (c = 0; c < ccnt; c++) {
409 if (remsec[r].s_secinfo.sc_nfsnum ==
410 cursec[c].s_secinfo.sc_nfsnum) {
411
412 /*
413 * Decrement secinfo reference count by 1.
414 * If this entry is invalid after decrementing
415 * the count (i.e. count < 1), this entry will
416 * be removed.
417 */
418 cursec[c].s_refcnt--;
419
420 SECREF_TRACE(cursec, "del_ref",
421 cursec[c].s_secinfo.sc_nfsnum,
422 cursec[c].s_refcnt);
423
424 ASSERT(cursec[c].s_refcnt >= 0);
425
426 if (SEC_REF_INVALID(&cursec[c]))
427 tcnt--;
428 break;
429 }
430 }
431 }
432
433 ASSERT(tcnt >= 0);
434 if (tcnt == ccnt)
435 return; /* no change; no flavors to remove */
436
437 if (tcnt == 0) {
438 srv_secinfo_list_free(cursec, ccnt);
439 *pcurcnt = 0;
440 *pcursec = NULL;
441 return;
442 }
443
444 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
445
446 /* walk thru the given secinfo list to remove the flavors */
447 mcnt = 0;
448 for (c = 0; c < ccnt; c++) {
449 if (SEC_REF_INVALID(&cursec[c])) {
450 srv_secinfo_entry_free(&cursec[c]);
451 } else {
452 msec[mcnt] = cursec[c];
453 mcnt++;
454 }
455 }
456
457 ASSERT(mcnt == tcnt);
458 /*
459 * Done. Update curdata.
460 * Free the existing secinfo list in curdata. All pointers
461 * within the list have either been moved to msec or freed
462 * if it's invalid.
463 */
464 kmem_free(*pcursec, ccnt * sizeof (struct secinfo));
465 *pcursec = msec;
466 *pcurcnt = tcnt;
467 }
468
469
470 /*
471 * For the reshare case, sec flavor accounting happens in 3 steps:
472 * 1) propagate addition of new flavor refs up the ancestor tree
473 * 2) transfer flavor refs of descendants to new/reshared exportdata
474 * 3) propagate removal of old flavor refs up the ancestor tree
475 *
476 * srv_secinfo_exp2exp() implements step 2 of a reshare. At this point,
477 * the new flavor list has already been propagated up through the
478 * ancestor tree via srv_secinfo_treeclimb().
479 *
480 * If there is more than 1 export reference to an old flavor (i.e. some
481 * of its children shared with this flavor), this flavor information
482 * needs to be transferred to the new exportdata struct. A flavor in
483 * the old exportdata has descendant refs when its s_refcnt > 1 or it
484 * is implicitly shared (M_SEC4_EXPORTED not set in s_flags).
485 *
486 * SEC_REF_EXPORTED() is only true when M_SEC4_EXPORTED is set
487 * SEC_REF_SELF() is only true when both M_SEC4_EXPORTED is set and s_refcnt==1
488 *
489 * Transferring descendant flavor refcnts happens in 2 passes:
490 * a) flavors used before (oldsecinfo) and after (curdata->ex_secinfo) reshare
491 * b) flavors used before but not after reshare
492 *
493 * This routine is used under the protection of exported_lock (RW_WRITER).
494 */
495 void
496 srv_secinfo_exp2exp(exportdata_t *curdata, secinfo_t *oldsecinfo, int ocnt)
497 {
498 int ccnt, c; /* sec count in current data - curdata */
499 int o; /* sec count in old data - oldsecinfo */
500 int tcnt, mcnt; /* total sec count after the transfer */
501 struct secinfo *msec; /* merged secinfo list */
502
503 ccnt = curdata->ex_seccnt;
504
505 ASSERT(ocnt > 0);
506 ASSERT(!(curdata->ex_flags & EX_PSEUDO));
507
508 /*
509 * If the oldsecinfo has flavors with more than 1 reference count
510 * and the flavor is specified in the reshare, transfer the flavor
511 * refs to the new seclist (curdata.ex_secinfo).
512 */
513 tcnt = ccnt + ocnt;
514
515 for (o = 0; o < ocnt; o++) {
516
517 if (SEC_REF_SELF(&oldsecinfo[o])) {
518 tcnt--;
519 continue;
520 }
521
522 for (c = 0; c < ccnt; c++) {
523 if (oldsecinfo[o].s_secinfo.sc_nfsnum ==
524 curdata->ex_secinfo[c].s_secinfo.sc_nfsnum) {
525
526 /*
527 * add old reference to the current
528 * secinfo count
529 */
530 curdata->ex_secinfo[c].s_refcnt +=
531 oldsecinfo[o].s_refcnt;
532
533 /*
534 * Delete the old export flavor
535 * reference. The initial reference
536 * was created during srv_secinfo_add,
537 * and the count is decremented below
538 * to account for the initial reference.
539 */
540 if (SEC_REF_EXPORTED(&oldsecinfo[o]))
541 curdata->ex_secinfo[c].s_refcnt--;
542
543 SECREF_TRACE(curdata->ex_path,
544 "reshare_xfer_common_child_refs",
545 curdata->ex_secinfo[c].s_secinfo.sc_nfsnum,
546 curdata->ex_secinfo[c].s_refcnt);
547
548 ASSERT(curdata->ex_secinfo[c].s_refcnt >= 0);
549
550 tcnt--;
551 break;
552 }
553 }
554 }
555
556 if (tcnt == ccnt)
557 return; /* no more transfer to do */
558
559 /*
560 * oldsecinfo has flavors referenced by its children that are not
561 * in the current (new) export flavor list. Add these flavors.
562 */
563 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
564
565 /* move current secinfo list data to the new list */
566 for (c = 0; c < ccnt; c++)
567 msec[c] = curdata->ex_secinfo[c];
568
569 /*
570 * Add the flavor that's not in the new export, but still
571 * referenced by its children.
572 */
573 mcnt = ccnt;
574 for (o = 0; o < ocnt; o++) {
575 if (! SEC_REF_SELF(&oldsecinfo[o])) {
576 for (c = 0; c < ccnt; c++) {
577 if (oldsecinfo[o].s_secinfo.sc_nfsnum ==
578 curdata->ex_secinfo[c].s_secinfo.sc_nfsnum)
579 break;
580 }
581
582 /*
583 * This is the one. Add it. Decrement the ref count
584 * by 1 if the flavor is an explicitly shared flavor
585 * for the oldsecinfo export node.
586 */
587 if (c == ccnt) {
588 srv_secinfo_copy(&oldsecinfo[o], &msec[mcnt]);
589 if (SEC_REF_EXPORTED(&oldsecinfo[o]))
590 msec[mcnt].s_refcnt--;
591
592 SECREF_TRACE(curdata,
593 "reshare_xfer_implicit_child_refs",
594 msec[mcnt].s_secinfo.sc_nfsnum,
595 msec[mcnt].s_refcnt);
596
597 ASSERT(msec[mcnt].s_refcnt >= 0);
598 mcnt++;
599 }
600 }
601 }
602
603 ASSERT(mcnt == tcnt);
604 /*
605 * Done. Update curdata, free the existing secinfo list in
606 * curdata and set the new value.
607 */
608 if (ccnt > 0)
609 kmem_free(curdata->ex_secinfo, ccnt * sizeof (struct secinfo));
610 curdata->ex_seccnt = tcnt;
611 curdata->ex_secinfo = msec;
612 }
613
614 /*
615 * When unsharing an old export node and the old node becomes a pseudo node,
616 * if there is more than 1 export reference to an old flavor (i.e. some of
617 * its children shared with this flavor), this flavor information needs to
618 * be transferred to the new shared node.
619 *
620 * This routine is used under the protection of exported_lock (RW_WRITER).
621 */
622 void
623 srv_secinfo_exp2pseu(exportdata_t *curdata, exportdata_t *olddata)
624 {
625 int ocnt, o; /* sec count in transfer data - trandata */
626 int tcnt, mcnt; /* total sec count after transfer */
627 struct secinfo *msec; /* merged secinfo list */
628
629 ASSERT(curdata->ex_flags & EX_PSEUDO);
630 ASSERT(curdata->ex_seccnt == 0);
631
632 ocnt = olddata->ex_seccnt;
633
634 /*
635 * If the olddata has flavors with more than 1 reference count,
636 * transfer the information to the curdata.
637 */
638 tcnt = ocnt;
639
640 for (o = 0; o < ocnt; o++) {
641 if (SEC_REF_SELF(&olddata->ex_secinfo[o]))
642 tcnt--;
643 }
644
645 if (tcnt == 0)
646 return; /* no transfer to do */
647
648 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
649
650 mcnt = 0;
651 for (o = 0; o < ocnt; o++) {
652 if (! SEC_REF_SELF(&olddata->ex_secinfo[o])) {
653
654 /*
655 * Decrement the reference count by 1 if the flavor is
656 * an explicitly shared flavor for the olddata export
657 * node.
658 */
659 srv_secinfo_copy(&olddata->ex_secinfo[o], &msec[mcnt]);
660 msec[mcnt].s_flags = M_RO;
661 if (SEC_REF_EXPORTED(&olddata->ex_secinfo[o]))
662 msec[mcnt].s_refcnt--;
663
664 SECREF_TRACE(curdata, "unshare_morph_pseudo",
665 msec[mcnt].s_secinfo.sc_nfsnum,
666 msec[mcnt].s_refcnt);
667
668 ASSERT(msec[mcnt].s_refcnt >= 0);
669 mcnt++;
670 }
671 }
672
673 ASSERT(mcnt == tcnt);
674 /*
675 * Done. Update curdata.
676 * Free up the existing secinfo list in curdata and
677 * set the new value.
678 */
679 curdata->ex_seccnt = tcnt;
680 curdata->ex_secinfo = msec;
681 }
682
683 /*
684 * Find for given treenode the exportinfo which has its
685 * exp_visible linked on its exi_visible list.
686 *
687 * Note: We could add new pointer either to treenode or
688 * to exp_visible, which will point there directly.
689 * This would buy some speed for some memory.
690 */
691 exportinfo_t *
692 vis2exi(treenode_t *tnode)
693 {
694 exportinfo_t *exi_ret = NULL;
695 #ifdef DEBUG
696 zone_t *zone = NULL;
697 #endif
698
699 for (;;) {
700 tnode = tnode->tree_parent;
701 #ifdef DEBUG
702 if (zone == NULL && tnode->tree_exi != NULL)
703 zone = tnode->tree_exi->exi_zone;
704 #endif
705 if (TREE_ROOT(tnode)) {
706 ASSERT3P(zone, ==, tnode->tree_exi->exi_zone);
707 exi_ret = tnode->tree_exi;
708 break;
709 }
710 }
711
712 ASSERT(exi_ret); /* Every visible should have its home exportinfo */
713 return (exi_ret);
714 }
715
716 /*
717 * For NFS V4.
718 * Add or remove the newly exported or unexported security flavors of the
719 * given exportinfo from its ancestors upto the system root.
720 */
721 void
722 srv_secinfo_treeclimb(nfs_export_t *ne, exportinfo_t *exip, secinfo_t *sec,
723 int seccnt, bool_t isadd)
724 {
725 treenode_t *tnode = exip->exi_tree;
726
727 ASSERT(RW_WRITE_HELD(&ne->exported_lock));
728 ASSERT(tnode != NULL);
729
730 if (seccnt == 0)
731 return;
732
733 /*
734 * If flavors are being added and the new export root isn't
735 * also VROOT, its implicitly allowed flavors are inherited from
736 * its pseudonode.
737 * Note - for VROOT exports the implicitly allowed flavors were
738 * transferred from the PSEUDO export in exportfs()
739 */
740 if (isadd && !(exip->exi_vp->v_flag & VROOT) &&
741 !VN_IS_CURZONEROOT(exip->exi_vp) &&
742 tnode->tree_vis->vis_seccnt > 0) {
743 srv_secinfo_add(&exip->exi_export.ex_secinfo,
744 &exip->exi_export.ex_seccnt, tnode->tree_vis->vis_secinfo,
745 tnode->tree_vis->vis_seccnt, FALSE);
746 }
747
748 /*
749 * Move to parent node and propagate sec flavor
750 * to exportinfo and to visible structures.
751 */
752 tnode = tnode->tree_parent;
753
754 while (tnode != NULL) {
755
756 /* If there is exportinfo, update it */
757 if (tnode->tree_exi != NULL) {
758 secinfo_t **pxsec =
759 &tnode->tree_exi->exi_export.ex_secinfo;
760 int *pxcnt = &tnode->tree_exi->exi_export.ex_seccnt;
761 int is_pseudo = PSEUDO(tnode->tree_exi);
762 if (isadd)
763 srv_secinfo_add(pxsec, pxcnt, sec, seccnt,
764 is_pseudo);
765 else
766 srv_secinfo_remove(pxsec, pxcnt, sec, seccnt);
767 }
768
769 /* Update every visible - only root node has no visible */
770 if (tnode->tree_vis != NULL) {
771 secinfo_t **pxsec = &tnode->tree_vis->vis_secinfo;
772 int *pxcnt = &tnode->tree_vis->vis_seccnt;
773 if (isadd)
774 srv_secinfo_add(pxsec, pxcnt, sec, seccnt,
775 FALSE);
776 else
777 srv_secinfo_remove(pxsec, pxcnt, sec, seccnt);
778 }
779 tnode = tnode->tree_parent;
780 }
781 }
782
783 /* hash_name is a text substitution for either fid_hash or path_hash */
784 #define exp_hash_unlink(exi, hash_name) \
785 if (*(exi)->hash_name.bckt == (exi)) \
786 *(exi)->hash_name.bckt = (exi)->hash_name.next; \
787 if ((exi)->hash_name.prev) \
788 (exi)->hash_name.prev->hash_name.next = (exi)->hash_name.next; \
789 if ((exi)->hash_name.next) \
790 (exi)->hash_name.next->hash_name.prev = (exi)->hash_name.prev; \
791 (exi)->hash_name.bckt = NULL;
792
793 #define exp_hash_link(exi, hash_name, bucket) \
794 (exi)->hash_name.bckt = (bucket); \
795 (exi)->hash_name.prev = NULL; \
796 (exi)->hash_name.next = *(bucket); \
797 if ((exi)->hash_name.next) \
798 (exi)->hash_name.next->hash_name.prev = (exi); \
799 *(bucket) = (exi);
800
801 void
802 export_link(nfs_export_t *ne, exportinfo_t *exi)
803 {
804 exportinfo_t **bckt;
805
806 ASSERT(RW_WRITE_HELD(&ne->exported_lock));
807
808 bckt = &ne->exptable[exptablehash(&exi->exi_fsid, &exi->exi_fid)];
809 exp_hash_link(exi, fid_hash, bckt);
810
811 bckt = &ne->exptable_path_hash[pkp_tab_hash(exi->exi_export.ex_path,
812 strlen(exi->exi_export.ex_path))];
813 exp_hash_link(exi, path_hash, bckt);
814 }
815
816 /*
817 * Helper functions for exi_id handling
818 */
819 static int
820 exi_id_compar(const void *v1, const void *v2)
821 {
822 const struct exportinfo *e1 = v1;
823 const struct exportinfo *e2 = v2;
824
825 if (e1->exi_id < e2->exi_id)
826 return (-1);
827 if (e1->exi_id > e2->exi_id)
828 return (1);
829
830 return (0);
831 }
832
833 int
834 exi_id_get_next()
835 {
836 struct exportinfo e;
837 int ret = exi_id_next;
838
839 ASSERT(MUTEX_HELD(&nfs_exi_id_lock));
840
841 do {
842 exi_id_next++;
843 if (exi_id_next == 0)
844 exi_id_overflow = TRUE;
845
846 if (!exi_id_overflow)
847 break;
848
849 if (exi_id_next == ret)
850 cmn_err(CE_PANIC, "exi_id exhausted");
851
852 e.exi_id = exi_id_next;
853 } while (avl_find(&exi_id_tree, &e, NULL) != NULL);
854
855 return (ret);
856 }
857
858 /*ARGSUSED*/
859 static void *
860 nfs_export_zone_init(zoneid_t zoneid)
861 {
862 int i;
863 nfs_export_t *ne;
864
865 ne = kmem_zalloc(sizeof (*ne), KM_SLEEP);
866
867 rw_init(&ne->exported_lock, NULL, RW_DEFAULT, NULL);
868
869 /*
870 * Allocate the place holder for the public file handle, which
871 * is all zeroes. It is initially set to the root filesystem.
872 */
873 ne->exi_root = kmem_zalloc(sizeof (*ne->exi_root), KM_SLEEP);
874 ne->exi_public = ne->exi_root;
875
876 ne->exi_root->exi_export.ex_flags = EX_PUBLIC;
877 ne->exi_root->exi_export.ex_pathlen = 1; /* length of "/" */
878 ne->exi_root->exi_export.ex_path =
879 kmem_alloc(ne->exi_root->exi_export.ex_pathlen + 1, KM_SLEEP);
880 ne->exi_root->exi_export.ex_path[0] = '/';
881 ne->exi_root->exi_export.ex_path[1] = '\0';
882
883 ne->exi_root->exi_count = 1;
884 mutex_init(&ne->exi_root->exi_lock, NULL, MUTEX_DEFAULT, NULL);
885
886 ne->exi_root->exi_zone = zone_find_by_id_nolock(zoneid);
887 ne->exi_root->exi_vp = ne->exi_root->exi_zone->zone_rootvp;
888 ne->exi_rootfid.fid_len = MAXFIDSZ;
889 if (vop_fid_pseudo(ne->exi_root->exi_vp, &ne->exi_rootfid) != 0) {
890 mutex_destroy(&ne->exi_root->exi_lock);
891 kmem_free(ne->exi_root->exi_export.ex_path,
892 ne->exi_root->exi_export.ex_pathlen + 1);
893 kmem_free(ne->exi_root, sizeof (*ne->exi_root));
894 return (NULL);
895 }
896
897 /* Initialize auth cache and auth cache lock */
898 for (i = 0; i < AUTH_TABLESIZE; i++) {
899 ne->exi_root->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t),
900 KM_SLEEP);
901 avl_create(ne->exi_root->exi_cache[i],
902 nfsauth_cache_clnt_compar, sizeof (struct auth_cache_clnt),
903 offsetof(struct auth_cache_clnt, authc_link));
904 }
905 rw_init(&ne->exi_root->exi_cache_lock, NULL, RW_DEFAULT, NULL);
906
907 /* Setup the fhandle template */
908 ne->exi_root->exi_fh.fh_fsid = rootdir->v_vfsp->vfs_fsid;
909 ne->exi_root->exi_fh.fh_xlen = ne->exi_rootfid.fid_len;
910 bcopy(ne->exi_rootfid.fid_data, ne->exi_root->exi_fh.fh_xdata,
911 ne->exi_rootfid.fid_len);
912 ne->exi_root->exi_fh.fh_len = sizeof (ne->exi_root->exi_fh.fh_data);
913
914 rw_enter(&ne->exported_lock, RW_WRITER);
915
916 /* Publish the exportinfo in the hash table */
917 export_link(ne, ne->exi_root);
918
919 /* Initialize exi_id and exi_kstats */
920 mutex_enter(&nfs_exi_id_lock);
921 ne->exi_root->exi_id = exi_id_get_next();
922 avl_add(&exi_id_tree, ne->exi_root);
923 mutex_exit(&nfs_exi_id_lock);
924
925 rw_exit(&ne->exported_lock);
926 ne->ns_root = NULL;
927
928 return (ne);
929 }
930
931 /*ARGSUSED*/
932 static void
933 nfs_export_zone_fini(zoneid_t zoneid, void *data)
934 {
935 int i;
936 nfs_export_t *ne = data;
937 struct exportinfo *exi;
938
939 rw_enter(&ne->exported_lock, RW_WRITER);
940 mutex_enter(&nfs_exi_id_lock);
941
942 avl_remove(&exi_id_tree, ne->exi_root);
943 export_unlink(ne, ne->exi_root);
944
945 mutex_exit(&nfs_exi_id_lock);
946 rw_exit(&ne->exported_lock);
947
948 /* Deallocate the place holder for the public file handle */
949 srv_secinfo_list_free(ne->exi_root->exi_export.ex_secinfo,
950 ne->exi_root->exi_export.ex_seccnt);
951 mutex_destroy(&ne->exi_root->exi_lock);
952
953 rw_destroy(&ne->exi_root->exi_cache_lock);
954 for (i = 0; i < AUTH_TABLESIZE; i++) {
955 avl_destroy(ne->exi_root->exi_cache[i]);
956 kmem_free(ne->exi_root->exi_cache[i], sizeof (avl_tree_t));
957 }
958
959 kmem_free(ne->exi_root->exi_export.ex_path,
960 ne->exi_root->exi_export.ex_pathlen + 1);
961 kmem_free(ne->exi_root, sizeof (*ne->exi_root));
962
963 exi = avl_first(&exi_id_tree);
964 while (exi != NULL) {
965 struct exportinfo *nexi = AVL_NEXT(&exi_id_tree, exi);
966 if (zoneid == exi->exi_zoneid)
967 (void) unexport(ne, exi);
968 exi = nexi;
969 }
970
971 rw_destroy(&ne->exported_lock);
972 kmem_free(ne, sizeof (*ne));
973 }
974
975 /*
976 * Initialization routine for export routines.
977 * Should only be called once.
978 */
979 void
980 nfs_exportinit(void)
981 {
982 mutex_init(&nfs_exi_id_lock, NULL, MUTEX_DEFAULT, NULL);
983
984 /* exi_id handling initialization */
985 exi_id_next = 0;
986 exi_id_overflow = FALSE;
987 avl_create(&exi_id_tree, exi_id_compar, sizeof (struct exportinfo),
988 offsetof(struct exportinfo, exi_id_link));
989
990 zone_key_create(&nfs_export_key, nfs_export_zone_init,
991 NULL, nfs_export_zone_fini);
992
993 nfslog_init();
994 }
995
996 /*
997 * Finalization routine for export routines.
998 */
999 void
1000 nfs_exportfini(void)
1001 {
1002 (void) zone_key_delete(nfs_export_key);
1003 avl_destroy(&exi_id_tree);
1004 mutex_destroy(&nfs_exi_id_lock);
1005 }
1006
1007 /*
1008 * Check if 2 gss mechanism identifiers are the same.
1009 *
1010 * return FALSE if not the same.
1011 * return TRUE if the same.
1012 */
1013 static bool_t
1014 nfs_mech_equal(rpc_gss_OID mech1, rpc_gss_OID mech2)
1015 {
1016 if ((mech1->length == 0) && (mech2->length == 0))
1017 return (TRUE);
1018
1019 if (mech1->length != mech2->length)
1020 return (FALSE);
1021
1022 return (bcmp(mech1->elements, mech2->elements, mech1->length) == 0);
1023 }
1024
1025 /*
1026 * This routine is used by rpc to map rpc security number
1027 * to nfs specific security flavor number.
1028 *
1029 * The gss callback prototype is
1030 * callback(struct svc_req *, gss_cred_id_t *, gss_ctx_id_t *,
1031 * rpc_gss_lock_t *, void **),
1032 * since nfs does not use the gss_cred_id_t/gss_ctx_id_t arguments
1033 * we cast them to void.
1034 */
1035 /*ARGSUSED*/
1036 bool_t
1037 rfs_gsscallback(struct svc_req *req, gss_cred_id_t deleg, void *gss_context,
1038 rpc_gss_lock_t *lock, void **cookie)
1039 {
1040 int i, j;
1041 rpc_gss_rawcred_t *raw_cred;
1042 struct exportinfo *exi;
1043 nfs_export_t *ne = nfs_get_export();
1044
1045 /*
1046 * We don't deal with delegated credentials.
1047 */
1048 if (deleg != GSS_C_NO_CREDENTIAL)
1049 return (FALSE);
1050
1051 raw_cred = lock->raw_cred;
1052 *cookie = NULL;
1053
1054 rw_enter(&ne->exported_lock, RW_READER);
1055
1056 for (i = 0; i < EXPTABLESIZE; i++) {
1057 exi = ne->exptable[i];
1058 while (exi) {
1059 if (exi->exi_export.ex_seccnt > 0) {
1060 struct secinfo *secp;
1061 seconfig_t *se;
1062 int seccnt;
1063
1064 secp = exi->exi_export.ex_secinfo;
1065 seccnt = exi->exi_export.ex_seccnt;
1066 for (j = 0; j < seccnt; j++) {
1067 /*
1068 * If there is a map of the triplet
1069 * (mechanism, service, qop) between
1070 * raw_cred and the exported flavor,
1071 * get the psudo flavor number.
1072 * Also qop should not be NULL, it
1073 * should be "default" or something
1074 * else.
1075 */
1076 se = &secp[j].s_secinfo;
1077 if ((se->sc_rpcnum == RPCSEC_GSS) &&
1078
1079 (nfs_mech_equal(
1080 se->sc_gss_mech_type,
1081 raw_cred->mechanism)) &&
1082
1083 (se->sc_service ==
1084 raw_cred->service) &&
1085 (raw_cred->qop == se->sc_qop)) {
1086
1087 *cookie = (void *)(uintptr_t)
1088 se->sc_nfsnum;
1089 goto done;
1090 }
1091 }
1092 }
1093 exi = exi->fid_hash.next;
1094 }
1095 }
1096 done:
1097 rw_exit(&ne->exported_lock);
1098
1099 /*
1100 * If no nfs pseudo number mapping can be found in the export
1101 * table, assign the nfsflavor to NFS_FLAVOR_NOMAP. In V4, we may
1102 * recover the flavor mismatch from NFS layer (NFS4ERR_WRONGSEC).
1103 *
1104 * For example:
1105 * server first shares with krb5i;
1106 * client mounts with krb5i;
1107 * server re-shares with krb5p;
1108 * client tries with krb5i, but no mapping can be found;
1109 * rpcsec_gss module calls this routine to do the mapping,
1110 * if this routine fails, request is rejected from
1111 * the rpc layer.
1112 * What we need is to let the nfs layer rejects the request.
1113 * For V4, we can reject with NFS4ERR_WRONGSEC and the client
1114 * may recover from it by getting the new flavor via SECINFO.
1115 *
1116 * nfs pseudo number for RPCSEC_GSS mapping (see nfssec.conf)
1117 * is owned by IANA (see RFC 2623).
1118 *
1119 * XXX NFS_FLAVOR_NOMAP is defined in Solaris to work around
1120 * the implementation issue. This number should not overlap with
1121 * any new IANA defined pseudo flavor numbers.
1122 */
1123 if (*cookie == NULL)
1124 *cookie = (void *)NFS_FLAVOR_NOMAP;
1125
1126 lock->locked = TRUE;
1127
1128 return (TRUE);
1129 }
1130
1131
1132 /*
1133 * Exportfs system call; credentials should be checked before
1134 * calling this function.
1135 */
1136 int
1137 exportfs(struct exportfs_args *args, model_t model, cred_t *cr)
1138 {
1139 vnode_t *vp;
1140 vnode_t *dvp;
1141 struct exportdata *kex;
1142 struct exportinfo *exi = NULL;
1143 struct exportinfo *ex, *ex1, *ex2;
1144 fid_t fid;
1145 fsid_t fsid;
1146 int error;
1147 size_t allocsize;
1148 struct secinfo *sp;
1149 struct secinfo *exs;
1150 rpc_gss_callback_t cb;
1151 char *pathbuf;
1152 char *log_buffer;
1153 char *tagbuf;
1154 int callback;
1155 int allocd_seccnt;
1156 STRUCT_HANDLE(exportfs_args, uap);
1157 STRUCT_DECL(exportdata, uexi);
1158 struct secinfo newsec[MAX_FLAVORS];
1159 int newcnt;
1160 struct secinfo oldsec[MAX_FLAVORS];
1161 int oldcnt;
1162 int i;
1163 struct pathname lookpn;
1164 nfs_export_t *ne = nfs_get_export();
1165
1166 STRUCT_SET_HANDLE(uap, model, args);
1167
1168 /* Read in pathname from userspace */
1169 if (error = pn_get(STRUCT_FGETP(uap, dname), UIO_USERSPACE, &lookpn))
1170 return (error);
1171
1172 /* Walk the export list looking for that pathname */
1173 rw_enter(&ne->exported_lock, RW_READER);
1174 DTRACE_PROBE(nfss__i__exported_lock1_start);
1175 for (ex1 = ne->exptable_path_hash[pkp_tab_hash(lookpn.pn_path,
1176 strlen(lookpn.pn_path))]; ex1; ex1 = ex1->path_hash.next) {
1177 if (ex1 != ne->exi_root && 0 ==
1178 strcmp(ex1->exi_export.ex_path, lookpn.pn_path)) {
1179 exi_hold(ex1);
1180 break;
1181 }
1182 }
1183 DTRACE_PROBE(nfss__i__exported_lock1_stop);
1184 rw_exit(&ne->exported_lock);
1185
1186 /* Is this an unshare? */
1187 if (STRUCT_FGETP(uap, uex) == NULL) {
1188 pn_free(&lookpn);
1189 if (ex1 == NULL)
1190 return (EINVAL);
1191 error = unexport(ne, ex1);
1192 exi_rele(ex1);
1193 return (error);
1194 }
1195
1196 /* It is a share or a re-share */
1197 error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1198 FOLLOW, &dvp, &vp);
1199 if (error == EINVAL) {
1200 /*
1201 * if fname resolves to / we get EINVAL error
1202 * since we wanted the parent vnode. Try again
1203 * with NULL dvp.
1204 */
1205 error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1206 FOLLOW, NULL, &vp);
1207 dvp = NULL;
1208 }
1209 if (!error && vp == NULL) {
1210 /* Last component of fname not found */
1211 if (dvp != NULL)
1212 VN_RELE(dvp);
1213 error = ENOENT;
1214 }
1215 if (error) {
1216 pn_free(&lookpn);
1217 if (ex1)
1218 exi_rele(ex1);
1219 return (error);
1220 }
1221
1222 /*
1223 * 'vp' may be an AUTOFS node, so we perform a
1224 * VOP_ACCESS() to trigger the mount of the
1225 * intended filesystem, so we can share the intended
1226 * filesystem instead of the AUTOFS filesystem.
1227 */
1228 (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
1229
1230 /*
1231 * We're interested in the top most filesystem.
1232 * This is specially important when uap->dname is a trigger
1233 * AUTOFS node, since we're really interested in sharing the
1234 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
1235 * call not the AUTOFS node itself.
1236 */
1237 if (vn_mountedvfs(vp) != NULL) {
1238 if (error = traverse(&vp)) {
1239 VN_RELE(vp);
1240 if (dvp != NULL)
1241 VN_RELE(dvp);
1242 pn_free(&lookpn);
1243 if (ex1)
1244 exi_rele(ex1);
1245 return (error);
1246 }
1247 }
1248
1249 /* Do not allow sharing another vnode for already shared path */
1250 if (ex1 && !PSEUDO(ex1) && !VN_CMP(ex1->exi_vp, vp)) {
1251 VN_RELE(vp);
1252 if (dvp != NULL)
1253 VN_RELE(dvp);
1254 pn_free(&lookpn);
1255 exi_rele(ex1);
1256 return (EEXIST);
1257 }
1258 if (ex1)
1259 exi_rele(ex1);
1260
1261 /*
1262 * Get the vfs id
1263 */
1264 bzero(&fid, sizeof (fid));
1265 fid.fid_len = MAXFIDSZ;
1266 error = VOP_FID(vp, &fid, NULL);
1267 fsid = vp->v_vfsp->vfs_fsid;
1268
1269 if (error) {
1270 VN_RELE(vp);
1271 if (dvp != NULL)
1272 VN_RELE(dvp);
1273 /*
1274 * If VOP_FID returns ENOSPC then the fid supplied
1275 * is too small. For now we simply return EREMOTE.
1276 */
1277 if (error == ENOSPC)
1278 error = EREMOTE;
1279 pn_free(&lookpn);
1280 return (error);
1281 }
1282
1283 /*
1284 * Do not allow re-sharing a shared vnode under a different path
1285 * PSEUDO export has ex_path fabricated, e.g. "/tmp (pseudo)", skip it.
1286 */
1287 rw_enter(&ne->exported_lock, RW_READER);
1288 DTRACE_PROBE(nfss__i__exported_lock2_start);
1289 for (ex2 = ne->exptable[exptablehash(&fsid, &fid)]; ex2;
1290 ex2 = ex2->fid_hash.next) {
1291 if (ex2 != ne->exi_root && !PSEUDO(ex2) &&
1292 VN_CMP(ex2->exi_vp, vp) &&
1293 strcmp(ex2->exi_export.ex_path, lookpn.pn_path) != 0) {
1294 DTRACE_PROBE(nfss__i__exported_lock2_stop);
1295 rw_exit(&ne->exported_lock);
1296 VN_RELE(vp);
1297 if (dvp != NULL)
1298 VN_RELE(dvp);
1299 pn_free(&lookpn);
1300 return (EEXIST);
1301 }
1302 }
1303 DTRACE_PROBE(nfss__i__exported_lock2_stop);
1304 rw_exit(&ne->exported_lock);
1305 pn_free(&lookpn);
1306
1307 exi = kmem_zalloc(sizeof (*exi), KM_SLEEP);
1308 exi->exi_fsid = fsid;
1309 exi->exi_fid = fid;
1310 exi->exi_vp = vp;
1311 exi->exi_count = 1;
1312 exi->exi_zone = crgetzone(cr);
1313 ASSERT(exi->exi_zone != NULL); /* XXX KEBE ASKS... */
1314 ASSERT3P(exi->exi_zone, ==, curzone); /* ... are these legit? */
1315 exi->exi_volatile_dev = (vfssw[vp->v_vfsp->vfs_fstype].vsw_flag &
1316 VSW_VOLATILEDEV) ? 1 : 0;
1317 mutex_init(&exi->exi_lock, NULL, MUTEX_DEFAULT, NULL);
1318 exi->exi_dvp = dvp;
1319
1320 /*
1321 * Initialize auth cache and auth cache lock
1322 */
1323 for (i = 0; i < AUTH_TABLESIZE; i++) {
1324 exi->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1325 avl_create(exi->exi_cache[i], nfsauth_cache_clnt_compar,
1326 sizeof (struct auth_cache_clnt),
1327 offsetof(struct auth_cache_clnt, authc_link));
1328 }
1329 rw_init(&exi->exi_cache_lock, NULL, RW_DEFAULT, NULL);
1330
1331 /*
1332 * Build up the template fhandle
1333 */
1334 exi->exi_fh.fh_fsid = fsid;
1335 if (exi->exi_fid.fid_len > sizeof (exi->exi_fh.fh_xdata)) {
1336 error = EREMOTE;
1337 goto out1;
1338 }
1339 exi->exi_fh.fh_xlen = exi->exi_fid.fid_len;
1340 bcopy(exi->exi_fid.fid_data, exi->exi_fh.fh_xdata,
1341 exi->exi_fid.fid_len);
1342
1343 exi->exi_fh.fh_len = sizeof (exi->exi_fh.fh_data);
1344
1345 kex = &exi->exi_export;
1346
1347 /*
1348 * Load in everything, and do sanity checking
1349 */
1350 STRUCT_INIT(uexi, model);
1351 if (copyin(STRUCT_FGETP(uap, uex), STRUCT_BUF(uexi),
1352 STRUCT_SIZE(uexi))) {
1353 error = EFAULT;
1354 goto out1;
1355 }
1356
1357 kex->ex_version = STRUCT_FGET(uexi, ex_version);
1358 if (kex->ex_version != EX_CURRENT_VERSION) {
1359 error = EINVAL;
1360 cmn_err(CE_WARN,
1361 "NFS: exportfs requires export struct version 2 - got %d\n",
1362 kex->ex_version);
1363 goto out1;
1364 }
1365
1366 /*
1367 * Must have at least one security entry
1368 */
1369 kex->ex_seccnt = STRUCT_FGET(uexi, ex_seccnt);
1370 if (kex->ex_seccnt < 1) {
1371 error = EINVAL;
1372 goto out1;
1373 }
1374
1375 kex->ex_path = STRUCT_FGETP(uexi, ex_path);
1376 kex->ex_pathlen = STRUCT_FGET(uexi, ex_pathlen);
1377 kex->ex_flags = STRUCT_FGET(uexi, ex_flags);
1378 kex->ex_anon = STRUCT_FGET(uexi, ex_anon);
1379 kex->ex_secinfo = STRUCT_FGETP(uexi, ex_secinfo);
1380 kex->ex_index = STRUCT_FGETP(uexi, ex_index);
1381 kex->ex_log_buffer = STRUCT_FGETP(uexi, ex_log_buffer);
1382 kex->ex_log_bufferlen = STRUCT_FGET(uexi, ex_log_bufferlen);
1383 kex->ex_tag = STRUCT_FGETP(uexi, ex_tag);
1384 kex->ex_taglen = STRUCT_FGET(uexi, ex_taglen);
1385
1386 /*
1387 * Copy the exported pathname into
1388 * an appropriately sized buffer.
1389 */
1390 pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1391 if (copyinstr(kex->ex_path, pathbuf, MAXPATHLEN, &kex->ex_pathlen)) {
1392 kmem_free(pathbuf, MAXPATHLEN);
1393 error = EFAULT;
1394 goto out1;
1395 }
1396 kex->ex_path = kmem_alloc(kex->ex_pathlen + 1, KM_SLEEP);
1397 bcopy(pathbuf, kex->ex_path, kex->ex_pathlen);
1398 kex->ex_path[kex->ex_pathlen] = '\0';
1399 kmem_free(pathbuf, MAXPATHLEN);
1400
1401 /*
1402 * Get the path to the logging buffer and the tag
1403 */
1404 if (kex->ex_flags & EX_LOG) {
1405 log_buffer = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1406 if (copyinstr(kex->ex_log_buffer, log_buffer, MAXPATHLEN,
1407 &kex->ex_log_bufferlen)) {
1408 kmem_free(log_buffer, MAXPATHLEN);
1409 error = EFAULT;
1410 goto out2;
1411 }
1412 kex->ex_log_buffer =
1413 kmem_alloc(kex->ex_log_bufferlen + 1, KM_SLEEP);
1414 bcopy(log_buffer, kex->ex_log_buffer, kex->ex_log_bufferlen);
1415 kex->ex_log_buffer[kex->ex_log_bufferlen] = '\0';
1416 kmem_free(log_buffer, MAXPATHLEN);
1417
1418 tagbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1419 if (copyinstr(kex->ex_tag, tagbuf, MAXPATHLEN,
1420 &kex->ex_taglen)) {
1421 kmem_free(tagbuf, MAXPATHLEN);
1422 error = EFAULT;
1423 goto out3;
1424 }
1425 kex->ex_tag = kmem_alloc(kex->ex_taglen + 1, KM_SLEEP);
1426 bcopy(tagbuf, kex->ex_tag, kex->ex_taglen);
1427 kex->ex_tag[kex->ex_taglen] = '\0';
1428 kmem_free(tagbuf, MAXPATHLEN);
1429 }
1430
1431 /*
1432 * Load the security information for each flavor
1433 */
1434 allocsize = kex->ex_seccnt * SIZEOF_STRUCT(secinfo, model);
1435 sp = kmem_zalloc(allocsize, KM_SLEEP);
1436 if (copyin(kex->ex_secinfo, sp, allocsize)) {
1437 kmem_free(sp, allocsize);
1438 error = EFAULT;
1439 goto out4;
1440 }
1441
1442 /*
1443 * All of these nested structures need to be converted to
1444 * the kernel native format.
1445 */
1446 if (model != DATAMODEL_NATIVE) {
1447 size_t allocsize2;
1448 struct secinfo *sp2;
1449
1450 allocsize2 = kex->ex_seccnt * sizeof (struct secinfo);
1451 sp2 = kmem_zalloc(allocsize2, KM_SLEEP);
1452
1453 for (i = 0; i < kex->ex_seccnt; i++) {
1454 STRUCT_HANDLE(secinfo, usi);
1455
1456 STRUCT_SET_HANDLE(usi, model,
1457 (struct secinfo *)((caddr_t)sp +
1458 (i * SIZEOF_STRUCT(secinfo, model))));
1459 bcopy(STRUCT_FGET(usi, s_secinfo.sc_name),
1460 sp2[i].s_secinfo.sc_name, MAX_NAME_LEN);
1461 sp2[i].s_secinfo.sc_nfsnum =
1462 STRUCT_FGET(usi, s_secinfo.sc_nfsnum);
1463 sp2[i].s_secinfo.sc_rpcnum =
1464 STRUCT_FGET(usi, s_secinfo.sc_rpcnum);
1465 bcopy(STRUCT_FGET(usi, s_secinfo.sc_gss_mech),
1466 sp2[i].s_secinfo.sc_gss_mech, MAX_NAME_LEN);
1467 sp2[i].s_secinfo.sc_gss_mech_type =
1468 STRUCT_FGETP(usi, s_secinfo.sc_gss_mech_type);
1469 sp2[i].s_secinfo.sc_qop =
1470 STRUCT_FGET(usi, s_secinfo.sc_qop);
1471 sp2[i].s_secinfo.sc_service =
1472 STRUCT_FGET(usi, s_secinfo.sc_service);
1473
1474 sp2[i].s_flags = STRUCT_FGET(usi, s_flags);
1475 sp2[i].s_window = STRUCT_FGET(usi, s_window);
1476 sp2[i].s_rootid = STRUCT_FGET(usi, s_rootid);
1477 sp2[i].s_rootcnt = STRUCT_FGET(usi, s_rootcnt);
1478 sp2[i].s_rootnames = STRUCT_FGETP(usi, s_rootnames);
1479 }
1480 kmem_free(sp, allocsize);
1481 sp = sp2;
1482 allocsize = allocsize2;
1483 }
1484
1485 kex->ex_secinfo = sp;
1486
1487 /*
1488 * And now copy rootnames for each individual secinfo.
1489 */
1490 callback = 0;
1491 allocd_seccnt = 0;
1492 while (allocd_seccnt < kex->ex_seccnt) {
1493
1494 exs = &sp[allocd_seccnt];
1495 if (exs->s_rootcnt > 0) {
1496 if (!sec_svc_loadrootnames(exs->s_secinfo.sc_rpcnum,
1497 exs->s_rootcnt, &exs->s_rootnames, model)) {
1498 error = EFAULT;
1499 goto out5;
1500 }
1501 }
1502
1503 if (exs->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
1504 rpc_gss_OID mech_tmp;
1505 STRUCT_DECL(rpc_gss_OID_s, umech_tmp);
1506 caddr_t elements_tmp;
1507
1508 /* Copyin mechanism type */
1509 STRUCT_INIT(umech_tmp, model);
1510 mech_tmp = kmem_alloc(sizeof (*mech_tmp), KM_SLEEP);
1511 if (copyin(exs->s_secinfo.sc_gss_mech_type,
1512 STRUCT_BUF(umech_tmp), STRUCT_SIZE(umech_tmp))) {
1513 kmem_free(mech_tmp, sizeof (*mech_tmp));
1514 error = EFAULT;
1515 goto out5;
1516 }
1517 mech_tmp->length = STRUCT_FGET(umech_tmp, length);
1518 mech_tmp->elements = STRUCT_FGETP(umech_tmp, elements);
1519
1520 elements_tmp = kmem_alloc(mech_tmp->length, KM_SLEEP);
1521 if (copyin(mech_tmp->elements, elements_tmp,
1522 mech_tmp->length)) {
1523 kmem_free(elements_tmp, mech_tmp->length);
1524 kmem_free(mech_tmp, sizeof (*mech_tmp));
1525 error = EFAULT;
1526 goto out5;
1527 }
1528 mech_tmp->elements = elements_tmp;
1529 exs->s_secinfo.sc_gss_mech_type = mech_tmp;
1530 allocd_seccnt++;
1531
1532 callback = 1;
1533 } else
1534 allocd_seccnt++;
1535 }
1536
1537 /*
1538 * Init the secinfo reference count and mark these flavors
1539 * explicitly exported flavors.
1540 */
1541 for (i = 0; i < kex->ex_seccnt; i++) {
1542 kex->ex_secinfo[i].s_flags |= M_4SEC_EXPORTED;
1543 kex->ex_secinfo[i].s_refcnt = 1;
1544 }
1545
1546 /*
1547 * Set up rpcsec_gss callback routine entry if any.
1548 */
1549 if (callback) {
1550 cb.callback = rfs_gsscallback;
1551 cb.program = NFS_ACL_PROGRAM;
1552 for (cb.version = NFS_ACL_VERSMIN;
1553 cb.version <= NFS_ACL_VERSMAX; cb.version++) {
1554 (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1555 (void *)&cb);
1556 }
1557
1558 cb.program = NFS_PROGRAM;
1559 for (cb.version = NFS_VERSMIN;
1560 cb.version <= NFS_VERSMAX; cb.version++) {
1561 (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1562 (void *)&cb);
1563 }
1564 }
1565
1566 /*
1567 * Check the index flag. Do this here to avoid holding the
1568 * lock while dealing with the index option (as we do with
1569 * the public option).
1570 */
1571 if (kex->ex_flags & EX_INDEX) {
1572 if (!kex->ex_index) { /* sanity check */
1573 error = EINVAL;
1574 goto out5;
1575 }
1576 if (error = loadindex(kex))
1577 goto out5;
1578 }
1579
1580 if (kex->ex_flags & EX_LOG) {
1581 if (error = nfslog_setup(exi))
1582 goto out6;
1583 }
1584
1585 /*
1586 * Insert the new entry at the front of the export list
1587 */
1588 rw_enter(&ne->exported_lock, RW_WRITER);
1589 DTRACE_PROBE(nfss__i__exported_lock3_start);
1590
1591 export_link(ne, exi);
1592
1593 /*
1594 * Check the rest of the list for an old entry for the fs.
1595 * If one is found then unlink it, wait until this is the
1596 * only reference and then free it.
1597 */
1598 for (ex = exi->fid_hash.next; ex != NULL; ex = ex->fid_hash.next) {
1599 if (ex != ne->exi_root && VN_CMP(ex->exi_vp, vp)) {
1600 mutex_enter(&nfs_exi_id_lock);
1601 avl_remove(&exi_id_tree, ex);
1602 mutex_exit(&nfs_exi_id_lock);
1603 export_unlink(ne, ex);
1604 break;
1605 }
1606 }
1607
1608 /*
1609 * If the public filehandle is pointing at the
1610 * old entry, then point it back at the root.
1611 */
1612 if (ex != NULL && ex == ne->exi_public)
1613 ne->exi_public = ne->exi_root;
1614
1615 /*
1616 * If the public flag is on, make the global exi_public
1617 * point to this entry and turn off the public bit so that
1618 * we can distinguish it from the place holder export.
1619 */
1620 if (kex->ex_flags & EX_PUBLIC) {
1621 ne->exi_public = exi;
1622 kex->ex_flags &= ~EX_PUBLIC;
1623 }
1624
1625 #ifdef VOLATILE_FH_TEST
1626 /*
1627 * Set up the volatile_id value if volatile on share.
1628 * The list of volatile renamed filehandles is always destroyed,
1629 * if the fs was reshared.
1630 */
1631 if (kex->ex_flags & EX_VOLFH)
1632 exi->exi_volatile_id = gethrestime_sec();
1633
1634 mutex_init(&exi->exi_vol_rename_lock, NULL, MUTEX_DEFAULT, NULL);
1635 #endif /* VOLATILE_FH_TEST */
1636
1637 /*
1638 * If this is a new export, then climb up
1639 * the tree and check if any pseudo exports
1640 * need to be created to provide a path for
1641 * NFS v4 clients.
1642 */
1643 if (ex == NULL) {
1644 error = treeclimb_export(exi);
1645 if (error)
1646 goto out7;
1647 } else {
1648 /* If it's a re-export update namespace tree */
1649 exi->exi_tree = ex->exi_tree;
1650 exi->exi_tree->tree_exi = exi;
1651
1652 /* Update the change timestamp */
1653 tree_update_change(ne, exi->exi_tree, NULL);
1654 }
1655
1656 /*
1657 * build a unique flavor list from the flavors specified
1658 * in the share cmd. unique means that each flavor only
1659 * appears once in the secinfo list -- no duplicates allowed.
1660 */
1661 newcnt = build_seclist_nodups(&exi->exi_export, newsec, FALSE);
1662
1663 srv_secinfo_treeclimb(ne, exi, newsec, newcnt, TRUE);
1664
1665 /*
1666 * If re-sharing an old export entry, update the secinfo data
1667 * depending on if the old entry is a pseudo node or not.
1668 */
1669 if (ex != NULL) {
1670 oldcnt = build_seclist_nodups(&ex->exi_export, oldsec, FALSE);
1671 if (PSEUDO(ex)) {
1672 /*
1673 * The dir being shared is a pseudo export root (which
1674 * will be transformed into a real export root). The
1675 * flavor(s) of the new share were propagated to the
1676 * ancestors by srv_secinfo_treeclimb() above. Now
1677 * transfer the implicit flavor refs from the old
1678 * pseudo exprot root to the new (real) export root.
1679 */
1680 srv_secinfo_add(&exi->exi_export.ex_secinfo,
1681 &exi->exi_export.ex_seccnt, oldsec, oldcnt, TRUE);
1682 } else {
1683 /*
1684 * First transfer implicit flavor refs to new export.
1685 * Remove old flavor refs last.
1686 */
1687 srv_secinfo_exp2exp(&exi->exi_export, oldsec, oldcnt);
1688 srv_secinfo_treeclimb(ne, ex, oldsec, oldcnt, FALSE);
1689 }
1690 }
1691
1692 /*
1693 * If it's a re-export and the old entry has a pseudonode list,
1694 * transfer it to the new export.
1695 */
1696 if (ex != NULL && (ex->exi_visible != NULL)) {
1697 exi->exi_visible = ex->exi_visible;
1698 ex->exi_visible = NULL;
1699 }
1700
1701 /*
1702 * Initialize exi_id and exi_kstats
1703 */
1704 if (ex != NULL) {
1705 exi->exi_id = ex->exi_id;
1706 } else {
1707 mutex_enter(&nfs_exi_id_lock);
1708 exi->exi_id = exi_id_get_next();
1709 mutex_exit(&nfs_exi_id_lock);
1710 }
1711 mutex_enter(&nfs_exi_id_lock);
1712 avl_add(&exi_id_tree, exi);
1713 mutex_exit(&nfs_exi_id_lock);
1714
1715 DTRACE_PROBE(nfss__i__exported_lock3_stop);
1716 rw_exit(&ne->exported_lock);
1717
1718 if (ne->exi_public == exi || kex->ex_flags & EX_LOG) {
1719 /*
1720 * Log share operation to this buffer only.
1721 */
1722 nfslog_share_record(exi, cr);
1723 }
1724
1725 if (ex != NULL)
1726 exi_rele(ex);
1727
1728 return (0);
1729
1730 out7:
1731 /* Unlink the new export in exptable. */
1732 export_unlink(ne, exi);
1733 DTRACE_PROBE(nfss__i__exported_lock3_stop);
1734 rw_exit(&ne->exported_lock);
1735 out6:
1736 if (kex->ex_flags & EX_INDEX)
1737 kmem_free(kex->ex_index, strlen(kex->ex_index) + 1);
1738 out5:
1739 /* free partially completed allocation */
1740 while (--allocd_seccnt >= 0) {
1741 exs = &kex->ex_secinfo[allocd_seccnt];
1742 srv_secinfo_entry_free(exs);
1743 }
1744
1745 if (kex->ex_secinfo) {
1746 kmem_free(kex->ex_secinfo,
1747 kex->ex_seccnt * sizeof (struct secinfo));
1748 }
1749
1750 out4:
1751 if ((kex->ex_flags & EX_LOG) && kex->ex_tag != NULL)
1752 kmem_free(kex->ex_tag, kex->ex_taglen + 1);
1753 out3:
1754 if ((kex->ex_flags & EX_LOG) && kex->ex_log_buffer != NULL)
1755 kmem_free(kex->ex_log_buffer, kex->ex_log_bufferlen + 1);
1756 out2:
1757 kmem_free(kex->ex_path, kex->ex_pathlen + 1);
1758 out1:
1759 VN_RELE(vp);
1760 if (dvp != NULL)
1761 VN_RELE(dvp);
1762 mutex_destroy(&exi->exi_lock);
1763 rw_destroy(&exi->exi_cache_lock);
1764 for (i = 0; i < AUTH_TABLESIZE; i++) {
1765 avl_destroy(exi->exi_cache[i]);
1766 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
1767 }
1768
1769 kmem_free(exi, sizeof (*exi));
1770
1771 return (error);
1772 }
1773
1774 /*
1775 * Remove the exportinfo from the export list
1776 */
1777 void
1778 export_unlink(nfs_export_t *ne, struct exportinfo *exi)
1779 {
1780 ASSERT(RW_WRITE_HELD(&ne->exported_lock));
1781
1782 exp_hash_unlink(exi, fid_hash);
1783 exp_hash_unlink(exi, path_hash);
1784 }
1785
1786 /*
1787 * Unexport an exported filesystem
1788 */
1789 static int
1790 unexport(nfs_export_t *ne, struct exportinfo *exi)
1791 {
1792 struct secinfo cursec[MAX_FLAVORS];
1793 int curcnt;
1794
1795 rw_enter(&ne->exported_lock, RW_WRITER);
1796
1797 /* Check if exi is still linked in the export table */
1798 if (!EXP_LINKED(exi) || PSEUDO(exi)) {
1799 rw_exit(&ne->exported_lock);
1800 return (EINVAL);
1801 }
1802
1803 mutex_enter(&nfs_exi_id_lock);
1804 avl_remove(&exi_id_tree, exi);
1805 mutex_exit(&nfs_exi_id_lock);
1806 export_unlink(ne, exi);
1807
1808 /*
1809 * Remove security flavors before treeclimb_unexport() is called
1810 * because srv_secinfo_treeclimb needs the namespace tree
1811 */
1812 curcnt = build_seclist_nodups(&exi->exi_export, cursec, TRUE);
1813 srv_secinfo_treeclimb(ne, exi, cursec, curcnt, FALSE);
1814
1815 /*
1816 * If there's a visible list, then need to leave
1817 * a pseudo export here to retain the visible list
1818 * for paths to exports below.
1819 */
1820 if (exi->exi_visible != NULL) {
1821 struct exportinfo *newexi;
1822
1823 newexi = pseudo_exportfs(ne, exi->exi_vp, &exi->exi_fid,
1824 exi->exi_visible, &exi->exi_export);
1825 exi->exi_visible = NULL;
1826
1827 /* interconnect the existing treenode with the new exportinfo */
1828 newexi->exi_zone = exi->exi_zone;
1829 newexi->exi_tree = exi->exi_tree;
1830 newexi->exi_tree->tree_exi = newexi;
1831
1832 /* Update the change timestamp */
1833 tree_update_change(ne, exi->exi_tree, NULL);
1834 } else {
1835 treeclimb_unexport(ne, exi);
1836 }
1837
1838 rw_exit(&ne->exported_lock);
1839
1840 /*
1841 * Need to call into the NFSv4 server and release all data
1842 * held on this particular export. This is important since
1843 * the v4 server may be holding file locks or vnodes under
1844 * this export.
1845 */
1846 rfs4_clean_state_exi(exi);
1847
1848 /*
1849 * Notify the lock manager that the filesystem is being
1850 * unexported.
1851 */
1852 lm_unexport(exi);
1853
1854 /*
1855 * If this was a public export, restore
1856 * the public filehandle to the root.
1857 */
1858
1859 /*
1860 * XXX KEBE ASKS --> Should CRED() instead be
1861 * exi->exi_zone->zone_kcred?
1862 */
1863 if (exi == ne->exi_public) {
1864 ne->exi_public = ne->exi_root;
1865
1866 nfslog_share_record(ne->exi_public, CRED());
1867 }
1868
1869 if (exi->exi_export.ex_flags & EX_LOG)
1870 nfslog_unshare_record(exi, CRED());
1871
1872 exi_rele(exi);
1873 return (0);
1874 }
1875
1876 /*
1877 * Get file handle system call.
1878 * Takes file name and returns a file handle for it.
1879 * Credentials must be verified before calling.
1880 */
1881 int
1882 nfs_getfh(struct nfs_getfh_args *args, model_t model, cred_t *cr)
1883 {
1884 nfs_fh3 fh;
1885 char buf[NFS3_MAXFHSIZE];
1886 char *logptr, logbuf[NFS3_MAXFHSIZE];
1887 int l = NFS3_MAXFHSIZE;
1888 vnode_t *vp;
1889 vnode_t *dvp;
1890 struct exportinfo *exi;
1891 int error;
1892 int vers;
1893 STRUCT_HANDLE(nfs_getfh_args, uap);
1894
1895 #ifdef lint
1896 model = model; /* STRUCT macros don't always use it */
1897 #endif
1898
1899 STRUCT_SET_HANDLE(uap, model, args);
1900
1901 error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
1902 FOLLOW, &dvp, &vp);
1903 if (error == EINVAL) {
1904 /*
1905 * if fname resolves to / we get EINVAL error
1906 * since we wanted the parent vnode. Try again
1907 * with NULL dvp.
1908 */
1909 error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
1910 FOLLOW, NULL, &vp);
1911 dvp = NULL;
1912 }
1913 if (!error && vp == NULL) {
1914 /*
1915 * Last component of fname not found
1916 */
1917 if (dvp != NULL) {
1918 VN_RELE(dvp);
1919 }
1920 error = ENOENT;
1921 }
1922 if (error)
1923 return (error);
1924
1925 /*
1926 * 'vp' may be an AUTOFS node, so we perform a
1927 * VOP_ACCESS() to trigger the mount of the
1928 * intended filesystem, so we can share the intended
1929 * filesystem instead of the AUTOFS filesystem.
1930 */
1931 (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
1932
1933 /*
1934 * We're interested in the top most filesystem.
1935 * This is specially important when uap->dname is a trigger
1936 * AUTOFS node, since we're really interested in sharing the
1937 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
1938 * call not the AUTOFS node itself.
1939 */
1940 if (vn_mountedvfs(vp) != NULL) {
1941 if (error = traverse(&vp)) {
1942 VN_RELE(vp);
1943 if (dvp != NULL)
1944 VN_RELE(dvp);
1945 return (error);
1946 }
1947 }
1948
1949 vers = STRUCT_FGET(uap, vers);
1950 exi = nfs_vptoexi(dvp, vp, cr, NULL, &error, FALSE);
1951 if (!error) {
1952 if (vers == NFS_VERSION) {
1953 error = makefh((fhandle_t *)buf, vp, exi);
1954 l = NFS_FHSIZE;
1955 logptr = buf;
1956 } else if (vers == NFS_V3) {
1957 int i, sz, pad;
1958
1959 error = makefh3(&fh, vp, exi);
1960 l = RNDUP(fh.fh3_length);
1961 if (!error && (l > sizeof (fhandle3_t)))
1962 error = EREMOTE;
1963 logptr = logbuf;
1964 if (!error) {
1965 i = 0;
1966 sz = sizeof (fsid_t);
1967 bcopy(&fh.fh3_fsid, &buf[i], sz);
1968 i += sz;
1969
1970 /*
1971 * For backwards compatibility, the
1972 * fid length may be less than
1973 * NFS_FHMAXDATA, but it was always
1974 * encoded as NFS_FHMAXDATA bytes.
1975 */
1976
1977 sz = sizeof (ushort_t);
1978 bcopy(&fh.fh3_len, &buf[i], sz);
1979 i += sz;
1980 bcopy(fh.fh3_data, &buf[i], fh.fh3_len);
1981 i += fh.fh3_len;
1982 pad = (NFS_FHMAXDATA - fh.fh3_len);
1983 if (pad > 0) {
1984 bzero(&buf[i], pad);
1985 i += pad;
1986 l += pad;
1987 }
1988
1989 sz = sizeof (ushort_t);
1990 bcopy(&fh.fh3_xlen, &buf[i], sz);
1991 i += sz;
1992 bcopy(fh.fh3_xdata, &buf[i], fh.fh3_xlen);
1993 i += fh.fh3_xlen;
1994 pad = (NFS_FHMAXDATA - fh.fh3_xlen);
1995 if (pad > 0) {
1996 bzero(&buf[i], pad);
1997 i += pad;
1998 l += pad;
1999 }
2000 }
2001 /*
2002 * If we need to do NFS logging, the filehandle
2003 * must be downsized to 32 bytes.
2004 */
2005 if (!error && exi->exi_export.ex_flags & EX_LOG) {
2006 i = 0;
2007 sz = sizeof (fsid_t);
2008 bcopy(&fh.fh3_fsid, &logbuf[i], sz);
2009 i += sz;
2010 sz = sizeof (ushort_t);
2011 bcopy(&fh.fh3_len, &logbuf[i], sz);
2012 i += sz;
2013 sz = NFS_FHMAXDATA;
2014 bcopy(fh.fh3_data, &logbuf[i], sz);
2015 i += sz;
2016 sz = sizeof (ushort_t);
2017 bcopy(&fh.fh3_xlen, &logbuf[i], sz);
2018 i += sz;
2019 sz = NFS_FHMAXDATA;
2020 bcopy(fh.fh3_xdata, &logbuf[i], sz);
2021 i += sz;
2022 }
2023 }
2024 if (!error && exi->exi_export.ex_flags & EX_LOG) {
2025 nfslog_getfh(exi, (fhandle_t *)logptr,
2026 STRUCT_FGETP(uap, fname), UIO_USERSPACE, cr);
2027 }
2028 exi_rele(exi);
2029 if (!error) {
2030 if (copyout(&l, STRUCT_FGETP(uap, lenp), sizeof (int)))
2031 error = EFAULT;
2032 if (copyout(buf, STRUCT_FGETP(uap, fhp), l))
2033 error = EFAULT;
2034 }
2035 }
2036 VN_RELE(vp);
2037 if (dvp != NULL) {
2038 VN_RELE(dvp);
2039 }
2040 return (error);
2041 }
2042
2043 /*
2044 * Strategy: if vp is in the export list, then
2045 * return the associated file handle. Otherwise, ".."
2046 * once up the vp and try again, until the root of the
2047 * filesystem is reached.
2048 */
2049 struct exportinfo *
2050 nfs_vptoexi(vnode_t *dvp, vnode_t *vp, cred_t *cr, int *walk,
2051 int *err, bool_t v4srv)
2052 {
2053 fid_t fid;
2054 int error;
2055 struct exportinfo *exi;
2056
2057 ASSERT(vp);
2058 VN_HOLD(vp);
2059 if (dvp != NULL) {
2060 VN_HOLD(dvp);
2061 }
2062 if (walk != NULL)
2063 *walk = 0;
2064
2065 for (;;) {
2066 bzero(&fid, sizeof (fid));
2067 fid.fid_len = MAXFIDSZ;
2068 error = vop_fid_pseudo(vp, &fid);
2069 if (error) {
2070 /*
2071 * If vop_fid_pseudo returns ENOSPC then the fid
2072 * supplied is too small. For now we simply
2073 * return EREMOTE.
2074 */
2075 if (error == ENOSPC)
2076 error = EREMOTE;
2077 break;
2078 }
2079
2080 if (v4srv)
2081 exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
2082 else
2083 exi = checkexport(&vp->v_vfsp->vfs_fsid, &fid);
2084
2085 if (exi != NULL) {
2086 /*
2087 * Found the export info
2088 */
2089 break;
2090 }
2091
2092 /*
2093 * We have just failed finding a matching export.
2094 * If we're at the root of this filesystem, then
2095 * it's time to stop (with failure).
2096 */
2097 if ((vp->v_flag & VROOT) || VN_IS_CURZONEROOT(vp)) {
2098 error = EINVAL;
2099 break;
2100 }
2101
2102 if (walk != NULL)
2103 (*walk)++;
2104
2105 /*
2106 * Now, do a ".." up vp. If dvp is supplied, use it,
2107 * otherwise, look it up.
2108 */
2109 if (dvp == NULL) {
2110 error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, cr,
2111 NULL, NULL, NULL);
2112 if (error)
2113 break;
2114 }
2115 VN_RELE(vp);
2116 vp = dvp;
2117 dvp = NULL;
2118 }
2119 VN_RELE(vp);
2120 if (dvp != NULL) {
2121 VN_RELE(dvp);
2122 }
2123 if (error != 0) {
2124 if (err != NULL)
2125 *err = error;
2126 return (NULL);
2127 }
2128 return (exi);
2129 }
2130
2131 int
2132 chk_clnt_sec(exportinfo_t *exi, struct svc_req *req)
2133 {
2134 int i, nfsflavor;
2135 struct secinfo *sp;
2136
2137 /*
2138 * Get the nfs flavor number from xprt.
2139 */
2140 nfsflavor = (int)(uintptr_t)req->rq_xprt->xp_cookie;
2141
2142 sp = exi->exi_export.ex_secinfo;
2143 for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2144 if ((nfsflavor == sp[i].s_secinfo.sc_nfsnum) &&
2145 SEC_REF_EXPORTED(sp + i))
2146 return (TRUE);
2147 }
2148 return (FALSE);
2149 }
2150
2151 /*
2152 * Make an fhandle from a vnode
2153 */
2154 int
2155 makefh(fhandle_t *fh, vnode_t *vp, exportinfo_t *exi)
2156 {
2157 int error;
2158
2159 *fh = exi->exi_fh; /* struct copy */
2160
2161 error = VOP_FID(vp, (fid_t *)&fh->fh_len, NULL);
2162 if (error) {
2163 /*
2164 * Should be something other than EREMOTE
2165 */
2166 return (EREMOTE);
2167 }
2168 return (0);
2169 }
2170
2171 /*
2172 * This routine makes an overloaded V2 fhandle which contains
2173 * sec modes.
2174 *
2175 * Note that the first four octets contain the length octet,
2176 * the status octet, and two padded octets to make them XDR
2177 * four-octet aligned.
2178 *
2179 * 1 2 3 4 32
2180 * +---+---+---+---+---+---+---+---+ +---+---+---+---+ +---+
2181 * | l | s | | | sec_1 |...| sec_n |...| |
2182 * +---+---+---+---+---+---+---+---+ +---+---+---+---+ +---+
2183 *
2184 * where
2185 *
2186 * the status octet s indicates whether there are more security
2187 * flavors (1 means yes, 0 means no) that require the client to
2188 * perform another 0x81 LOOKUP to get them,
2189 *
2190 * the length octet l is the length describing the number of
2191 * valid octets that follow. (l = 4 * n, where n is the number
2192 * of security flavors sent in the current overloaded filehandle.)
2193 *
2194 * sec_index should always be in the inclusive range: [1 - ex_seccnt],
2195 * and it tells server where to start within the secinfo array.
2196 * Usually it will always be 1; however, if more flavors are used
2197 * for the public export than can be encoded in the overloaded FH
2198 * (7 for NFS2), subsequent SNEGO MCLs will have a larger index
2199 * so the server will pick up where it left off from the previous
2200 * MCL reply.
2201 *
2202 * With NFS4 support, implicitly allowed flavors are also in
2203 * the secinfo array; however, they should not be returned in
2204 * SNEGO MCL replies.
2205 */
2206 int
2207 makefh_ol(fhandle_t *fh, exportinfo_t *exi, uint_t sec_index)
2208 {
2209 secinfo_t sec[MAX_FLAVORS];
2210 int totalcnt, i, *ipt, cnt, seccnt, secidx, fh_max_cnt;
2211 char *c;
2212
2213 if (fh == NULL || exi == NULL || sec_index < 1)
2214 return (EREMOTE);
2215
2216 /*
2217 * WebNFS clients need to know the unique set of explicitly
2218 * shared flavors in used for the public export. When
2219 * "TRUE" is passed to build_seclist_nodups(), only explicitly
2220 * shared flavors are included in the list.
2221 */
2222 seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2223 if (sec_index > seccnt)
2224 return (EREMOTE);
2225
2226 fh_max_cnt = (NFS_FHSIZE / sizeof (int)) - 1;
2227 totalcnt = seccnt - sec_index + 1;
2228 cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2229
2230 c = (char *)fh;
2231 /*
2232 * Encode the length octet representing the number of
2233 * security flavors (in bytes) in this overloaded fh.
2234 */
2235 *c = cnt * sizeof (int);
2236
2237 /*
2238 * Encode the status octet that indicates whether there
2239 * are more security flavors the client needs to get.
2240 */
2241 *(c + 1) = totalcnt > fh_max_cnt;
2242
2243 /*
2244 * put security flavors in the overloaded fh
2245 */
2246 ipt = (int *)(c + sizeof (int32_t));
2247 secidx = sec_index - 1;
2248 for (i = 0; i < cnt; i++) {
2249 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2250 }
2251 return (0);
2252 }
2253
2254 /*
2255 * Make an nfs_fh3 from a vnode
2256 */
2257 int
2258 makefh3(nfs_fh3 *fh, vnode_t *vp, struct exportinfo *exi)
2259 {
2260 int error;
2261 fid_t fid;
2262
2263 bzero(&fid, sizeof (fid));
2264 fid.fid_len = sizeof (fh->fh3_data);
2265 error = VOP_FID(vp, &fid, NULL);
2266 if (error)
2267 return (EREMOTE);
2268
2269 bzero(fh, sizeof (nfs_fh3));
2270 fh->fh3_fsid = exi->exi_fsid;
2271 fh->fh3_len = fid.fid_len;
2272 bcopy(fid.fid_data, fh->fh3_data, fh->fh3_len);
2273
2274 fh->fh3_xlen = exi->exi_fid.fid_len;
2275 ASSERT(fh->fh3_xlen <= sizeof (fh->fh3_xdata));
2276 bcopy(exi->exi_fid.fid_data, fh->fh3_xdata, fh->fh3_xlen);
2277
2278 fh->fh3_length = sizeof (fh->fh3_fsid)
2279 + sizeof (fh->fh3_len) + fh->fh3_len
2280 + sizeof (fh->fh3_xlen) + fh->fh3_xlen;
2281 fh->fh3_flags = 0;
2282
2283 return (0);
2284 }
2285
2286 /*
2287 * This routine makes an overloaded V3 fhandle which contains
2288 * sec modes.
2289 *
2290 * 1 4
2291 * +--+--+--+--+
2292 * | len |
2293 * +--+--+--+--+
2294 * up to 64
2295 * +--+--+--+--+--+--+--+--+--+--+--+--+ +--+--+--+--+
2296 * |s | | | | sec_1 | sec_2 | ... | sec_n |
2297 * +--+--+--+--+--+--+--+--+--+--+--+--+ +--+--+--+--+
2298 *
2299 * len = 4 * (n+1), where n is the number of security flavors
2300 * sent in the current overloaded filehandle.
2301 *
2302 * the status octet s indicates whether there are more security
2303 * mechanisms (1 means yes, 0 means no) that require the client
2304 * to perform another 0x81 LOOKUP to get them.
2305 *
2306 * Three octets are padded after the status octet.
2307 */
2308 int
2309 makefh3_ol(nfs_fh3 *fh, struct exportinfo *exi, uint_t sec_index)
2310 {
2311 secinfo_t sec[MAX_FLAVORS];
2312 int totalcnt, cnt, *ipt, i, seccnt, fh_max_cnt, secidx;
2313 char *c;
2314
2315 if (fh == NULL || exi == NULL || sec_index < 1)
2316 return (EREMOTE);
2317
2318 /*
2319 * WebNFS clients need to know the unique set of explicitly
2320 * shared flavors in used for the public export. When
2321 * "TRUE" is passed to build_seclist_nodups(), only explicitly
2322 * shared flavors are included in the list.
2323 */
2324 seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2325
2326 if (sec_index > seccnt)
2327 return (EREMOTE);
2328
2329 fh_max_cnt = (NFS3_FHSIZE / sizeof (int)) - 1;
2330 totalcnt = seccnt - sec_index + 1;
2331 cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2332
2333 /*
2334 * Place the length in fh3_length representing the number
2335 * of security flavors (in bytes) in this overloaded fh.
2336 */
2337 fh->fh3_flags = FH_WEBNFS;
2338 fh->fh3_length = (cnt+1) * sizeof (int32_t);
2339
2340 c = (char *)&fh->fh3_u.nfs_fh3_i.fh3_i;
2341 /*
2342 * Encode the status octet that indicates whether there
2343 * are more security flavors the client needs to get.
2344 */
2345 *c = totalcnt > fh_max_cnt;
2346
2347 /*
2348 * put security flavors in the overloaded fh
2349 */
2350 secidx = sec_index - 1;
2351 ipt = (int *)(c + sizeof (int32_t));
2352 for (i = 0; i < cnt; i++) {
2353 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2354 }
2355 return (0);
2356 }
2357
2358 /*
2359 * Make an nfs_fh4 from a vnode
2360 */
2361 int
2362 makefh4(nfs_fh4 *fh, vnode_t *vp, struct exportinfo *exi)
2363 {
2364 int error;
2365 nfs_fh4_fmt_t *fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2366 fid_t fid;
2367
2368 bzero(&fid, sizeof (fid));
2369 fid.fid_len = MAXFIDSZ;
2370 /*
2371 * vop_fid_pseudo() is used to set up NFSv4 namespace, so
2372 * use vop_fid_pseudo() here to get the fid instead of VOP_FID.
2373 */
2374 error = vop_fid_pseudo(vp, &fid);
2375 if (error)
2376 return (error);
2377
2378 fh->nfs_fh4_len = NFS_FH4_LEN;
2379
2380 fh_fmtp->fh4_i.fhx_fsid = exi->exi_fh.fh_fsid;
2381 fh_fmtp->fh4_i.fhx_xlen = exi->exi_fh.fh_xlen;
2382
2383 bzero(fh_fmtp->fh4_i.fhx_data, sizeof (fh_fmtp->fh4_i.fhx_data));
2384 bzero(fh_fmtp->fh4_i.fhx_xdata, sizeof (fh_fmtp->fh4_i.fhx_xdata));
2385 ASSERT(exi->exi_fh.fh_xlen <= sizeof (fh_fmtp->fh4_i.fhx_xdata));
2386 bcopy(exi->exi_fh.fh_xdata, fh_fmtp->fh4_i.fhx_xdata,
2387 exi->exi_fh.fh_xlen);
2388
2389 fh_fmtp->fh4_len = fid.fid_len;
2390 ASSERT(fid.fid_len <= sizeof (fh_fmtp->fh4_data));
2391 bcopy(fid.fid_data, fh_fmtp->fh4_data, fid.fid_len);
2392 fh_fmtp->fh4_flag = 0;
2393
2394 #ifdef VOLATILE_FH_TEST
2395 /*
2396 * XXX (temporary?)
2397 * Use the rnode volatile_id value to add volatility to the fh.
2398 *
2399 * For testing purposes there are currently two scenarios, based
2400 * on whether the filesystem was shared with "volatile_fh"
2401 * or "expire_on_rename". In the first case, use the value of
2402 * export struct share_time as the volatile_id. In the second
2403 * case use the vnode volatile_id value (which is set to the
2404 * time in which the file was renamed).
2405 *
2406 * Note that the above are temporary constructs for testing only
2407 * XXX
2408 */
2409 if (exi->exi_export.ex_flags & EX_VOLRNM) {
2410 fh_fmtp->fh4_volatile_id = find_volrnm_fh_id(exi, fh);
2411 } else if (exi->exi_export.ex_flags & EX_VOLFH) {
2412 fh_fmtp->fh4_volatile_id = exi->exi_volatile_id;
2413 } else {
2414 fh_fmtp->fh4_volatile_id = 0;
2415 }
2416 #endif /* VOLATILE_FH_TEST */
2417
2418 return (0);
2419 }
2420
2421 /*
2422 * Convert an fhandle into a vnode.
2423 * Uses the file id (fh_len + fh_data) in the fhandle to get the vnode.
2424 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2425 * are done with it.
2426 */
2427 vnode_t *
2428 nfs_fhtovp(fhandle_t *fh, struct exportinfo *exi)
2429 {
2430 vfs_t *vfsp;
2431 vnode_t *vp;
2432 int error;
2433 fid_t *fidp;
2434
2435 TRACE_0(TR_FAC_NFS, TR_FHTOVP_START,
2436 "fhtovp_start");
2437
2438 if (exi == NULL) {
2439 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2440 "fhtovp_end:(%S)", "exi NULL");
2441 return (NULL); /* not exported */
2442 }
2443
2444 ASSERT(exi->exi_vp != NULL);
2445
2446 if (PUBLIC_FH2(fh)) {
2447 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2448 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2449 "fhtovp_end:(%S)", "root not exported");
2450 return (NULL);
2451 }
2452 vp = exi->exi_vp;
2453 VN_HOLD(vp);
2454 return (vp);
2455 }
2456
2457 vfsp = exi->exi_vp->v_vfsp;
2458 ASSERT(vfsp != NULL);
2459 fidp = (fid_t *)&fh->fh_len;
2460
2461 error = VFS_VGET(vfsp, &vp, fidp);
2462 if (error || vp == NULL) {
2463 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2464 "fhtovp_end:(%S)", "VFS_GET failed or vp NULL");
2465 return (NULL);
2466 }
2467 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2468 "fhtovp_end:(%S)", "end");
2469 return (vp);
2470 }
2471
2472 /*
2473 * Convert an nfs_fh3 into a vnode.
2474 * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2475 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2476 * are done with it.
2477 */
2478 vnode_t *
2479 nfs3_fhtovp(nfs_fh3 *fh, struct exportinfo *exi)
2480 {
2481 vfs_t *vfsp;
2482 vnode_t *vp;
2483 int error;
2484 fid_t *fidp;
2485
2486 if (exi == NULL)
2487 return (NULL); /* not exported */
2488
2489 ASSERT(exi->exi_vp != NULL);
2490
2491 if (PUBLIC_FH3(fh)) {
2492 if (exi->exi_export.ex_flags & EX_PUBLIC)
2493 return (NULL);
2494 vp = exi->exi_vp;
2495 VN_HOLD(vp);
2496 return (vp);
2497 }
2498
2499 if (fh->fh3_length < NFS3_OLDFHSIZE ||
2500 fh->fh3_length > NFS3_MAXFHSIZE)
2501 return (NULL);
2502
2503 vfsp = exi->exi_vp->v_vfsp;
2504 ASSERT(vfsp != NULL);
2505 fidp = FH3TOFIDP(fh);
2506
2507 error = VFS_VGET(vfsp, &vp, fidp);
2508 if (error || vp == NULL)
2509 return (NULL);
2510
2511 return (vp);
2512 }
2513
2514 /*
2515 * Convert an nfs_fh4 into a vnode.
2516 * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2517 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2518 * are done with it.
2519 */
2520 vnode_t *
2521 nfs4_fhtovp(nfs_fh4 *fh, struct exportinfo *exi, nfsstat4 *statp)
2522 {
2523 vfs_t *vfsp;
2524 vnode_t *vp = NULL;
2525 int error;
2526 fid_t *fidp;
2527 nfs_fh4_fmt_t *fh_fmtp;
2528 #ifdef VOLATILE_FH_TEST
2529 uint32_t volatile_id = 0;
2530 #endif /* VOLATILE_FH_TEST */
2531
2532 if (exi == NULL) {
2533 *statp = NFS4ERR_STALE;
2534 return (NULL); /* not exported */
2535 }
2536 ASSERT(exi->exi_vp != NULL);
2537
2538 /* caller should have checked this */
2539 ASSERT(fh->nfs_fh4_len >= NFS_FH4_LEN);
2540
2541 fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2542 vfsp = exi->exi_vp->v_vfsp;
2543 ASSERT(vfsp != NULL);
2544 fidp = (fid_t *)&fh_fmtp->fh4_len;
2545
2546 #ifdef VOLATILE_FH_TEST
2547 /* XXX check if volatile - should be changed later */
2548 if (exi->exi_export.ex_flags & (EX_VOLRNM | EX_VOLFH)) {
2549 /*
2550 * Filesystem is shared with volatile filehandles
2551 */
2552 if (exi->exi_export.ex_flags & EX_VOLRNM)
2553 volatile_id = find_volrnm_fh_id(exi, fh);
2554 else
2555 volatile_id = exi->exi_volatile_id;
2556
2557 if (fh_fmtp->fh4_volatile_id != volatile_id) {
2558 *statp = NFS4ERR_FHEXPIRED;
2559 return (NULL);
2560 }
2561 }
2562 /*
2563 * XXX even if test_volatile_fh false, the fh may contain a
2564 * volatile id if obtained when the test was set.
2565 */
2566 fh_fmtp->fh4_volatile_id = (uchar_t)0;
2567 #endif /* VOLATILE_FH_TEST */
2568
2569 error = VFS_VGET(vfsp, &vp, fidp);
2570 /*
2571 * If we can not get vp from VFS_VGET, perhaps this is
2572 * an nfs v2/v3/v4 node in an nfsv4 pseudo filesystem.
2573 * Check it out.
2574 */
2575 if (error && PSEUDO(exi))
2576 error = nfs4_vget_pseudo(exi, &vp, fidp);
2577
2578 if (error || vp == NULL) {
2579 *statp = NFS4ERR_STALE;
2580 return (NULL);
2581 }
2582 /* XXX - disgusting hack */
2583 if (vp->v_type == VNON && vp->v_flag & V_XATTRDIR)
2584 vp->v_type = VDIR;
2585 *statp = NFS4_OK;
2586 return (vp);
2587 }
2588
2589 /*
2590 * Find the export structure associated with the given filesystem.
2591 * If found, then increment the ref count (exi_count).
2592 */
2593 struct exportinfo *
2594 checkexport(fsid_t *fsid, fid_t *fid)
2595 {
2596 struct exportinfo *exi;
2597 nfs_export_t *ne = nfs_get_export();
2598
2599 rw_enter(&ne->exported_lock, RW_READER);
2600 for (exi = ne->exptable[exptablehash(fsid, fid)];
2601 exi != NULL;
2602 exi = exi->fid_hash.next) {
2603 if (exportmatch(exi, fsid, fid)) {
2604 /*
2605 * If this is the place holder for the
2606 * public file handle, then return the
2607 * real export entry for the public file
2608 * handle.
2609 */
2610 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2611 exi = ne->exi_public;
2612 }
2613
2614 exi_hold(exi);
2615 rw_exit(&ne->exported_lock);
2616 return (exi);
2617 }
2618 }
2619 rw_exit(&ne->exported_lock);
2620 return (NULL);
2621 }
2622
2623
2624 /*
2625 * "old school" version of checkexport() for NFS4. NFS4
2626 * rfs4_compound holds exported_lock for duration of compound
2627 * processing. This version doesn't manipulate exi_count
2628 * since NFS4 breaks fundamental assumptions in the exi_count
2629 * design.
2630 */
2631 struct exportinfo *
2632 checkexport4(fsid_t *fsid, fid_t *fid, vnode_t *vp)
2633 {
2634 struct exportinfo *exi;
2635 nfs_export_t *ne = nfs_get_export();
2636
2637 ASSERT(RW_LOCK_HELD(&ne->exported_lock));
2638
2639 for (exi = ne->exptable[exptablehash(fsid, fid)];
2640 exi != NULL;
2641 exi = exi->fid_hash.next) {
2642 if (exportmatch(exi, fsid, fid)) {
2643 /*
2644 * If this is the place holder for the
2645 * public file handle, then return the
2646 * real export entry for the public file
2647 * handle.
2648 */
2649 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2650 exi = ne->exi_public;
2651 }
2652
2653 /*
2654 * If vp is given, check if vp is the
2655 * same vnode as the exported node.
2656 *
2657 * Since VOP_FID of a lofs node returns the
2658 * fid of its real node (ufs), the exported
2659 * node for lofs and (pseudo) ufs may have
2660 * the same fsid and fid.
2661 */
2662 if (vp == NULL || vp == exi->exi_vp)
2663 return (exi);
2664 }
2665 }
2666
2667 return (NULL);
2668 }
2669
2670 /*
2671 * Free an entire export list node
2672 */
2673 void
2674 exportfree(struct exportinfo *exi)
2675 {
2676 struct exportdata *ex;
2677 struct charset_cache *cache;
2678 int i;
2679
2680 ex = &exi->exi_export;
2681
2682 ASSERT(exi->exi_vp != NULL && !(exi->exi_export.ex_flags & EX_PUBLIC));
2683 VN_RELE(exi->exi_vp);
2684 if (exi->exi_dvp != NULL)
2685 VN_RELE(exi->exi_dvp);
2686
2687 if (ex->ex_flags & EX_INDEX)
2688 kmem_free(ex->ex_index, strlen(ex->ex_index) + 1);
2689
2690 kmem_free(ex->ex_path, ex->ex_pathlen + 1);
2691 nfsauth_cache_free(exi);
2692
2693 /*
2694 * if there is a character set mapping cached, clean it up.
2695 */
2696 for (cache = exi->exi_charset; cache != NULL;
2697 cache = exi->exi_charset) {
2698 if (cache->inbound != (kiconv_t)-1)
2699 (void) kiconv_close(cache->inbound);
2700 if (cache->outbound != (kiconv_t)-1)
2701 (void) kiconv_close(cache->outbound);
2702 exi->exi_charset = cache->next;
2703 kmem_free(cache, sizeof (struct charset_cache));
2704 }
2705
2706 if (exi->exi_logbuffer != NULL)
2707 nfslog_disable(exi);
2708
2709 if (ex->ex_flags & EX_LOG) {
2710 kmem_free(ex->ex_log_buffer, ex->ex_log_bufferlen + 1);
2711 kmem_free(ex->ex_tag, ex->ex_taglen + 1);
2712 }
2713
2714 if (exi->exi_visible)
2715 free_visible(exi->exi_visible);
2716
2717 srv_secinfo_list_free(ex->ex_secinfo, ex->ex_seccnt);
2718
2719 #ifdef VOLATILE_FH_TEST
2720 free_volrnm_list(exi);
2721 mutex_destroy(&exi->exi_vol_rename_lock);
2722 #endif /* VOLATILE_FH_TEST */
2723
2724 mutex_destroy(&exi->exi_lock);
2725 rw_destroy(&exi->exi_cache_lock);
2726 /*
2727 * All nodes in the exi_cache AVL trees were removed and freed in the
2728 * nfsauth_cache_free() call above. We will just destroy and free the
2729 * empty AVL trees here.
2730 */
2731 for (i = 0; i < AUTH_TABLESIZE; i++) {
2732 avl_destroy(exi->exi_cache[i]);
2733 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
2734 }
2735
2736 kmem_free(exi, sizeof (*exi));
2737 }
2738
2739 /*
2740 * load the index file from user space into kernel space.
2741 */
2742 static int
2743 loadindex(struct exportdata *kex)
2744 {
2745 int error;
2746 char index[MAXNAMELEN+1];
2747 size_t len;
2748
2749 /*
2750 * copyinstr copies the complete string including the NULL and
2751 * returns the len with the NULL byte included in the calculation
2752 * as long as the max length is not exceeded.
2753 */
2754 if (error = copyinstr(kex->ex_index, index, sizeof (index), &len))
2755 return (error);
2756
2757 kex->ex_index = kmem_alloc(len, KM_SLEEP);
2758 bcopy(index, kex->ex_index, len);
2759
2760 return (0);
2761 }
2762
2763 void
2764 exi_hold(struct exportinfo *exi)
2765 {
2766 mutex_enter(&exi->exi_lock);
2767 exi->exi_count++;
2768 mutex_exit(&exi->exi_lock);
2769 }
2770
2771 /*
2772 * When a thread completes using exi, it should call exi_rele().
2773 * exi_rele() decrements exi_count. It releases exi if exi_count == 0, i.e.
2774 * if this is the last user of exi and exi is not on exportinfo list anymore
2775 */
2776 void
2777 exi_rele(struct exportinfo *exi)
2778 {
2779 mutex_enter(&exi->exi_lock);
2780 exi->exi_count--;
2781 if (exi->exi_count == 0) {
2782 mutex_exit(&exi->exi_lock);
2783 exportfree(exi);
2784 } else
2785 mutex_exit(&exi->exi_lock);
2786 }
2787
2788 #ifdef VOLATILE_FH_TEST
2789 /*
2790 * Test for volatile fh's - add file handle to list and set its volatile id
2791 * to time it was renamed. If EX_VOLFH is also on and the fs is reshared,
2792 * the vol_rename queue is purged.
2793 *
2794 * XXX This code is for unit testing purposes only... To correctly use it, it
2795 * needs to tie a rename list to the export struct and (more
2796 * important), protect access to the exi rename list using a write lock.
2797 */
2798
2799 /*
2800 * get the fh vol record if it's in the volatile on rename list. Don't check
2801 * volatile_id in the file handle - compare only the file handles.
2802 */
2803 static struct ex_vol_rename *
2804 find_volrnm_fh(struct exportinfo *exi, nfs_fh4 *fh4p)
2805 {
2806 struct ex_vol_rename *p = NULL;
2807 fhandle4_t *fhp;
2808
2809 /* XXX shouldn't we assert &exported_lock held? */
2810 ASSERT(MUTEX_HELD(&exi->exi_vol_rename_lock));
2811
2812 if (fh4p->nfs_fh4_len != NFS_FH4_LEN) {
2813 return (NULL);
2814 }
2815 fhp = &((nfs_fh4_fmt_t *)fh4p->nfs_fh4_val)->fh4_i;
2816 for (p = exi->exi_vol_rename; p != NULL; p = p->vrn_next) {
2817 if (bcmp(fhp, &p->vrn_fh_fmt.fh4_i,
2818 sizeof (fhandle4_t)) == 0)
2819 break;
2820 }
2821 return (p);
2822 }
2823
2824 /*
2825 * get the volatile id for the fh (if there is - else return 0). Ignore the
2826 * volatile_id in the file handle - compare only the file handles.
2827 */
2828 static uint32_t
2829 find_volrnm_fh_id(struct exportinfo *exi, nfs_fh4 *fh4p)
2830 {
2831 struct ex_vol_rename *p;
2832 uint32_t volatile_id;
2833
2834 mutex_enter(&exi->exi_vol_rename_lock);
2835 p = find_volrnm_fh(exi, fh4p);
2836 volatile_id = (p ? p->vrn_fh_fmt.fh4_volatile_id :
2837 exi->exi_volatile_id);
2838 mutex_exit(&exi->exi_vol_rename_lock);
2839 return (volatile_id);
2840 }
2841
2842 /*
2843 * Free the volatile on rename list - will be called if a filesystem is
2844 * unshared or reshared without EX_VOLRNM
2845 */
2846 static void
2847 free_volrnm_list(struct exportinfo *exi)
2848 {
2849 struct ex_vol_rename *p, *pnext;
2850
2851 /* no need to hold mutex lock - this one is called from exportfree */
2852 for (p = exi->exi_vol_rename; p != NULL; p = pnext) {
2853 pnext = p->vrn_next;
2854 kmem_free(p, sizeof (*p));
2855 }
2856 exi->exi_vol_rename = NULL;
2857 }
2858
2859 /*
2860 * Add a file handle to the volatile on rename list.
2861 */
2862 void
2863 add_volrnm_fh(struct exportinfo *exi, vnode_t *vp)
2864 {
2865 struct ex_vol_rename *p;
2866 char fhbuf[NFS4_FHSIZE];
2867 nfs_fh4 fh4;
2868 int error;
2869
2870 fh4.nfs_fh4_val = fhbuf;
2871 error = makefh4(&fh4, vp, exi);
2872 if ((error) || (fh4.nfs_fh4_len != sizeof (p->vrn_fh_fmt))) {
2873 return;
2874 }
2875
2876 mutex_enter(&exi->exi_vol_rename_lock);
2877
2878 p = find_volrnm_fh(exi, &fh4);
2879
2880 if (p == NULL) {
2881 p = kmem_alloc(sizeof (*p), KM_SLEEP);
2882 bcopy(fh4.nfs_fh4_val, &p->vrn_fh_fmt, sizeof (p->vrn_fh_fmt));
2883 p->vrn_next = exi->exi_vol_rename;
2884 exi->exi_vol_rename = p;
2885 }
2886
2887 p->vrn_fh_fmt.fh4_volatile_id = gethrestime_sec();
2888 mutex_exit(&exi->exi_vol_rename_lock);
2889 }
2890
2891 #endif /* VOLATILE_FH_TEST */