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 = nfs_srv_getzg();
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 /* Every visible should have its home exportinfo */
706 ASSERT(exi_ret != NULL);
707 return (exi_ret);
708 }
709
710 /*
711 * For NFS V4.
712 * Add or remove the newly exported or unexported security flavors of the
713 * given exportinfo from its ancestors upto the system root.
714 */
715 static void
716 srv_secinfo_treeclimb(nfs_export_t *ne, exportinfo_t *exip, secinfo_t *sec,
717 int seccnt, bool_t isadd)
718 {
719 treenode_t *tnode;
720
721 ASSERT(RW_WRITE_HELD(&ne->exported_lock));
722
723 /*
724 * exi_tree can be null for the zone root
725 * which means we're already at the "top"
726 * and there's nothing more to "climb".
727 */
728 tnode = exip->exi_tree;
729 if (tnode == NULL) {
730 /* Should only happen for... */
731 ASSERT(exip == ne->exi_root);
732 return;
733 }
734
735 if (seccnt == 0)
736 return;
737
738 /*
739 * If flavors are being added and the new export root isn't
740 * also VROOT, its implicitly allowed flavors are inherited from
741 * its pseudonode.
742 * Note - for VROOT exports the implicitly allowed flavors were
743 * transferred from the PSEUDO export in exportfs()
744 */
745 if (isadd && !(exip->exi_vp->v_flag & VROOT) &&
746 !VN_CMP(exip->exi_vp, EXI_TO_ZONEROOTVP(exip)) &&
747 tnode->tree_vis->vis_seccnt > 0) {
748 srv_secinfo_add(&exip->exi_export.ex_secinfo,
749 &exip->exi_export.ex_seccnt, tnode->tree_vis->vis_secinfo,
750 tnode->tree_vis->vis_seccnt, FALSE);
751 }
752
753 /*
754 * Move to parent node and propagate sec flavor
755 * to exportinfo and to visible structures.
756 */
757 tnode = tnode->tree_parent;
758
759 while (tnode != NULL) {
760
761 /* If there is exportinfo, update it */
762 if (tnode->tree_exi != NULL) {
763 secinfo_t **pxsec =
764 &tnode->tree_exi->exi_export.ex_secinfo;
765 int *pxcnt = &tnode->tree_exi->exi_export.ex_seccnt;
766 int is_pseudo = PSEUDO(tnode->tree_exi);
767 if (isadd)
768 srv_secinfo_add(pxsec, pxcnt, sec, seccnt,
769 is_pseudo);
770 else
771 srv_secinfo_remove(pxsec, pxcnt, sec, seccnt);
772 }
773
774 /* Update every visible - only root node has no visible */
775 if (tnode->tree_vis != NULL) {
776 secinfo_t **pxsec = &tnode->tree_vis->vis_secinfo;
777 int *pxcnt = &tnode->tree_vis->vis_seccnt;
778 if (isadd)
779 srv_secinfo_add(pxsec, pxcnt, sec, seccnt,
780 FALSE);
781 else
782 srv_secinfo_remove(pxsec, pxcnt, sec, seccnt);
783 }
784 tnode = tnode->tree_parent;
785 }
786 }
787
788 /* hash_name is a text substitution for either fid_hash or path_hash */
789 #define exp_hash_unlink(exi, hash_name) \
790 if (*(exi)->hash_name.bckt == (exi)) \
791 *(exi)->hash_name.bckt = (exi)->hash_name.next; \
792 if ((exi)->hash_name.prev) \
793 (exi)->hash_name.prev->hash_name.next = (exi)->hash_name.next; \
794 if ((exi)->hash_name.next) \
795 (exi)->hash_name.next->hash_name.prev = (exi)->hash_name.prev; \
796 (exi)->hash_name.bckt = NULL;
797
798 #define exp_hash_link(exi, hash_name, bucket) \
799 (exi)->hash_name.bckt = (bucket); \
800 (exi)->hash_name.prev = NULL; \
801 (exi)->hash_name.next = *(bucket); \
802 if ((exi)->hash_name.next) \
803 (exi)->hash_name.next->hash_name.prev = (exi); \
804 *(bucket) = (exi);
805
806 void
807 export_link(nfs_export_t *ne, exportinfo_t *exi)
808 {
809 exportinfo_t **bckt;
810
811 ASSERT(RW_WRITE_HELD(&ne->exported_lock));
812
813 bckt = &ne->exptable[exptablehash(&exi->exi_fsid, &exi->exi_fid)];
814 exp_hash_link(exi, fid_hash, bckt);
815
816 bckt = &ne->exptable_path_hash[pkp_tab_hash(exi->exi_export.ex_path,
817 strlen(exi->exi_export.ex_path))];
818 exp_hash_link(exi, path_hash, bckt);
819 exi->exi_ne = ne;
820 }
821
822 /*
823 * Helper functions for exi_id handling
824 */
825 static int
826 exi_id_compar(const void *v1, const void *v2)
827 {
828 const struct exportinfo *e1 = v1;
829 const struct exportinfo *e2 = v2;
830
831 if (e1->exi_id < e2->exi_id)
832 return (-1);
833 if (e1->exi_id > e2->exi_id)
834 return (1);
835
836 return (0);
837 }
838
839 int
840 exi_id_get_next()
841 {
842 struct exportinfo e;
843 int ret = exi_id_next;
844
845 ASSERT(MUTEX_HELD(&nfs_exi_id_lock));
846
847 do {
848 exi_id_next++;
849 if (exi_id_next == 0)
850 exi_id_overflow = TRUE;
851
852 if (!exi_id_overflow)
853 break;
854
855 if (exi_id_next == ret)
856 cmn_err(CE_PANIC, "exi_id exhausted");
857
858 e.exi_id = exi_id_next;
859 } while (avl_find(&exi_id_tree, &e, NULL) != NULL);
860
861 return (ret);
862 }
863
864 /*
865 * Get the root file handle for this zone.
866 * Called when nfs_svc() starts
867 */
868 int
869 nfs_export_get_rootfh(nfs_globals_t *g)
870 {
871 nfs_export_t *ne = g->nfs_export;
872 int err;
873
874 ne->exi_rootfid.fid_len = MAXFIDSZ;
875 err = vop_fid_pseudo(ne->exi_root->exi_vp, &ne->exi_rootfid);
876 if (err != 0) {
877 ne->exi_rootfid.fid_len = 0;
878 return (err);
879 }
880
881 /* Setup the fhandle template exi_fh */
882 ne->exi_root->exi_fh.fh_fsid = rootdir->v_vfsp->vfs_fsid;
883 ne->exi_root->exi_fh.fh_xlen = ne->exi_rootfid.fid_len;
884 bcopy(ne->exi_rootfid.fid_data, ne->exi_root->exi_fh.fh_xdata,
885 ne->exi_rootfid.fid_len);
886 ne->exi_root->exi_fh.fh_len = sizeof (ne->exi_root->exi_fh.fh_data);
887
888 return (0);
889 }
890
891 void
892 nfs_export_zone_init(nfs_globals_t *ng)
893 {
894 int i;
895 nfs_export_t *ne;
896 zone_t *zone;
897
898 ne = kmem_zalloc(sizeof (*ne), KM_SLEEP);
899
900 rw_init(&ne->exported_lock, NULL, RW_DEFAULT, NULL);
901
902 ne->ne_globals = ng; /* "up" pointer */
903
904 /*
905 * Allocate the place holder for the public file handle, which
906 * is all zeroes. It is initially set to the root filesystem.
907 */
908 ne->exi_root = kmem_zalloc(sizeof (*ne->exi_root), KM_SLEEP);
909 ne->exi_public = ne->exi_root;
910
911 ne->exi_root->exi_export.ex_flags = EX_PUBLIC;
912 ne->exi_root->exi_export.ex_pathlen = 1; /* length of "/" */
913 ne->exi_root->exi_export.ex_path =
914 kmem_alloc(ne->exi_root->exi_export.ex_pathlen + 1, KM_SLEEP);
915 ne->exi_root->exi_export.ex_path[0] = '/';
916 ne->exi_root->exi_export.ex_path[1] = '\0';
917
918 ne->exi_root->exi_count = 1;
919 mutex_init(&ne->exi_root->exi_lock, NULL, MUTEX_DEFAULT, NULL);
920
921 /*
922 * Because we cannot:
923 * ASSERT(curzone->zone_id == ng->nfs_zoneid);
924 * We grab the zone pointer explicitly (like netstacks do) and
925 * set the rootvp here.
926 *
927 * Subsequent exportinfo_t's that get export_link()ed to "ne" also
928 * will backpoint to "ne" such that exi->exi_ne->exi_root->exi_vp
929 * will get the zone's rootvp for a given exportinfo_t.
930 */
931 zone = zone_find_by_id_nolock(ng->nfs_zoneid);
932 ne->exi_root->exi_vp = zone->zone_rootvp;
933 ne->exi_root->exi_zoneid = ng->nfs_zoneid;
934
935 /*
936 * Fill in ne->exi_rootfid later, in nfs_export_get_rootfid
937 * because we can't correctly return errors here.
938 */
939
940 /* Initialize auth cache and auth cache lock */
941 for (i = 0; i < AUTH_TABLESIZE; i++) {
942 ne->exi_root->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t),
943 KM_SLEEP);
944 avl_create(ne->exi_root->exi_cache[i],
945 nfsauth_cache_clnt_compar, sizeof (struct auth_cache_clnt),
946 offsetof(struct auth_cache_clnt, authc_link));
947 }
948 rw_init(&ne->exi_root->exi_cache_lock, NULL, RW_DEFAULT, NULL);
949
950 /* setup exi_fh later, in nfs_export_get_rootfid */
951
952 rw_enter(&ne->exported_lock, RW_WRITER);
953
954 /* Publish the exportinfo in the hash table */
955 export_link(ne, ne->exi_root);
956
957 /* Initialize exi_id and exi_kstats */
958 mutex_enter(&nfs_exi_id_lock);
959 ne->exi_root->exi_id = exi_id_get_next();
960 avl_add(&exi_id_tree, ne->exi_root);
961 mutex_exit(&nfs_exi_id_lock);
962
963 rw_exit(&ne->exported_lock);
964 ne->ns_root = NULL;
965
966 ng->nfs_export = ne;
967 }
968
969 /*
970 * During zone shutdown, remove exports
971 */
972 void
973 nfs_export_zone_shutdown(nfs_globals_t *ng)
974 {
975 nfs_export_t *ne = ng->nfs_export;
976 struct exportinfo *exi, *nexi;
977 int i, errors;
978
979 rw_enter(&ne->exported_lock, RW_READER);
980
981 errors = 0;
982 for (i = 0; i < EXPTABLESIZE; i++) {
983
984 exi = ne->exptable[i];
985 if (exi != NULL)
986 exi_hold(exi);
987
988 while (exi != NULL) {
989
990 /*
991 * Get and hold next export before
992 * dropping the rwlock and unexport
993 */
994 nexi = exi->fid_hash.next;
995 if (nexi != NULL)
996 exi_hold(nexi);
997
998 rw_exit(&ne->exported_lock);
999
1000 /*
1001 * Skip ne->exi_root which gets special
1002 * create/destroy handling.
1003 */
1004 if (exi != ne->exi_root &&
1005 unexport(ne, exi) != 0)
1006 errors++;
1007 exi_rele(exi);
1008
1009 rw_enter(&ne->exported_lock, RW_READER);
1010 exi = nexi;
1011 }
1012 }
1013 if (errors > 0) {
1014 cmn_err(CE_NOTE, "NFS: failed un-exports in zone %d",
1015 (int)ng->nfs_zoneid);
1016 }
1017
1018 rw_exit(&ne->exported_lock);
1019 }
1020
1021 void
1022 nfs_export_zone_fini(nfs_globals_t *ng)
1023 {
1024 int i;
1025 nfs_export_t *ne = ng->nfs_export;
1026 struct exportinfo *exi;
1027
1028 ng->nfs_export = NULL;
1029
1030 rw_enter(&ne->exported_lock, RW_WRITER);
1031
1032 mutex_enter(&nfs_exi_id_lock);
1033 avl_remove(&exi_id_tree, ne->exi_root);
1034 mutex_exit(&nfs_exi_id_lock);
1035
1036 export_unlink(ne, ne->exi_root);
1037
1038 rw_exit(&ne->exported_lock);
1039
1040 /* Deallocate the place holder for the public file handle */
1041 srv_secinfo_list_free(ne->exi_root->exi_export.ex_secinfo,
1042 ne->exi_root->exi_export.ex_seccnt);
1043 mutex_destroy(&ne->exi_root->exi_lock);
1044
1045 rw_destroy(&ne->exi_root->exi_cache_lock);
1046 for (i = 0; i < AUTH_TABLESIZE; i++) {
1047 avl_destroy(ne->exi_root->exi_cache[i]);
1048 kmem_free(ne->exi_root->exi_cache[i], sizeof (avl_tree_t));
1049 }
1050
1051 kmem_free(ne->exi_root->exi_export.ex_path,
1052 ne->exi_root->exi_export.ex_pathlen + 1);
1053 kmem_free(ne->exi_root, sizeof (*ne->exi_root));
1054
1055 /*
1056 * The shutdown hook should have left the exi_id_tree
1057 * with nothing belonging to this zone.
1058 */
1059 mutex_enter(&nfs_exi_id_lock);
1060 i = 0;
1061 exi = avl_first(&exi_id_tree);
1062 while (exi != NULL) {
1063 if (exi->exi_zoneid == ng->nfs_zoneid)
1064 i++;
1065 exi = AVL_NEXT(&exi_id_tree, exi);
1066 }
1067 mutex_exit(&nfs_exi_id_lock);
1068 if (i > 0) {
1069 cmn_err(CE_NOTE,
1070 "NFS: zone %d has %d export IDs left after shutdown",
1071 (int)ng->nfs_zoneid, i);
1072 }
1073 rw_destroy(&ne->exported_lock);
1074 kmem_free(ne, sizeof (*ne));
1075 }
1076
1077 /*
1078 * Initialization routine for export routines.
1079 * Should only be called once.
1080 */
1081 void
1082 nfs_exportinit(void)
1083 {
1084 mutex_init(&nfs_exi_id_lock, NULL, MUTEX_DEFAULT, NULL);
1085
1086 /* exi_id handling initialization */
1087 exi_id_next = 0;
1088 exi_id_overflow = FALSE;
1089 avl_create(&exi_id_tree, exi_id_compar, sizeof (struct exportinfo),
1090 offsetof(struct exportinfo, exi_id_link));
1091
1092 nfslog_init();
1093 }
1094
1095 /*
1096 * Finalization routine for export routines.
1097 */
1098 void
1099 nfs_exportfini(void)
1100 {
1101 avl_destroy(&exi_id_tree);
1102 mutex_destroy(&nfs_exi_id_lock);
1103 }
1104
1105 /*
1106 * Check if 2 gss mechanism identifiers are the same.
1107 *
1108 * return FALSE if not the same.
1109 * return TRUE if the same.
1110 */
1111 static bool_t
1112 nfs_mech_equal(rpc_gss_OID mech1, rpc_gss_OID mech2)
1113 {
1114 if ((mech1->length == 0) && (mech2->length == 0))
1115 return (TRUE);
1116
1117 if (mech1->length != mech2->length)
1118 return (FALSE);
1119
1120 return (bcmp(mech1->elements, mech2->elements, mech1->length) == 0);
1121 }
1122
1123 /*
1124 * This routine is used by rpc to map rpc security number
1125 * to nfs specific security flavor number.
1126 *
1127 * The gss callback prototype is
1128 * callback(struct svc_req *, gss_cred_id_t *, gss_ctx_id_t *,
1129 * rpc_gss_lock_t *, void **),
1130 * since nfs does not use the gss_cred_id_t/gss_ctx_id_t arguments
1131 * we cast them to void.
1132 */
1133 /*ARGSUSED*/
1134 bool_t
1135 rfs_gsscallback(struct svc_req *req, gss_cred_id_t deleg, void *gss_context,
1136 rpc_gss_lock_t *lock, void **cookie)
1137 {
1138 int i, j;
1139 rpc_gss_rawcred_t *raw_cred;
1140 struct exportinfo *exi;
1141 nfs_export_t *ne = nfs_get_export();
1142
1143 /*
1144 * We don't deal with delegated credentials.
1145 */
1146 if (deleg != GSS_C_NO_CREDENTIAL)
1147 return (FALSE);
1148
1149 raw_cred = lock->raw_cred;
1150 *cookie = NULL;
1151
1152 rw_enter(&ne->exported_lock, RW_READER);
1153
1154 for (i = 0; i < EXPTABLESIZE; i++) {
1155 exi = ne->exptable[i];
1156 while (exi) {
1157 if (exi->exi_export.ex_seccnt > 0) {
1158 struct secinfo *secp;
1159 seconfig_t *se;
1160 int seccnt;
1161
1162 secp = exi->exi_export.ex_secinfo;
1163 seccnt = exi->exi_export.ex_seccnt;
1164 for (j = 0; j < seccnt; j++) {
1165 /*
1166 * If there is a map of the triplet
1167 * (mechanism, service, qop) between
1168 * raw_cred and the exported flavor,
1169 * get the psudo flavor number.
1170 * Also qop should not be NULL, it
1171 * should be "default" or something
1172 * else.
1173 */
1174 se = &secp[j].s_secinfo;
1175 if ((se->sc_rpcnum == RPCSEC_GSS) &&
1176
1177 (nfs_mech_equal(
1178 se->sc_gss_mech_type,
1179 raw_cred->mechanism)) &&
1180
1181 (se->sc_service ==
1182 raw_cred->service) &&
1183 (raw_cred->qop == se->sc_qop)) {
1184
1185 *cookie = (void *)(uintptr_t)
1186 se->sc_nfsnum;
1187 goto done;
1188 }
1189 }
1190 }
1191 exi = exi->fid_hash.next;
1192 }
1193 }
1194 done:
1195 rw_exit(&ne->exported_lock);
1196
1197 /*
1198 * If no nfs pseudo number mapping can be found in the export
1199 * table, assign the nfsflavor to NFS_FLAVOR_NOMAP. In V4, we may
1200 * recover the flavor mismatch from NFS layer (NFS4ERR_WRONGSEC).
1201 *
1202 * For example:
1203 * server first shares with krb5i;
1204 * client mounts with krb5i;
1205 * server re-shares with krb5p;
1206 * client tries with krb5i, but no mapping can be found;
1207 * rpcsec_gss module calls this routine to do the mapping,
1208 * if this routine fails, request is rejected from
1209 * the rpc layer.
1210 * What we need is to let the nfs layer rejects the request.
1211 * For V4, we can reject with NFS4ERR_WRONGSEC and the client
1212 * may recover from it by getting the new flavor via SECINFO.
1213 *
1214 * nfs pseudo number for RPCSEC_GSS mapping (see nfssec.conf)
1215 * is owned by IANA (see RFC 2623).
1216 *
1217 * XXX NFS_FLAVOR_NOMAP is defined in Solaris to work around
1218 * the implementation issue. This number should not overlap with
1219 * any new IANA defined pseudo flavor numbers.
1220 */
1221 if (*cookie == NULL)
1222 *cookie = (void *)NFS_FLAVOR_NOMAP;
1223
1224 lock->locked = TRUE;
1225
1226 return (TRUE);
1227 }
1228
1229
1230 /*
1231 * Exportfs system call; credentials should be checked before
1232 * calling this function.
1233 */
1234 int
1235 exportfs(struct exportfs_args *args, model_t model, cred_t *cr)
1236 {
1237 vnode_t *vp;
1238 vnode_t *dvp;
1239 struct exportdata *kex;
1240 struct exportinfo *exi = NULL;
1241 struct exportinfo *ex, *ex1, *ex2;
1242 fid_t fid;
1243 fsid_t fsid;
1244 int error;
1245 size_t allocsize;
1246 struct secinfo *sp;
1247 struct secinfo *exs;
1248 rpc_gss_callback_t cb;
1249 char *pathbuf;
1250 char *log_buffer;
1251 char *tagbuf;
1252 int callback;
1253 int allocd_seccnt;
1254 STRUCT_HANDLE(exportfs_args, uap);
1255 STRUCT_DECL(exportdata, uexi);
1256 struct secinfo newsec[MAX_FLAVORS];
1257 int newcnt;
1258 struct secinfo oldsec[MAX_FLAVORS];
1259 int oldcnt;
1260 int i;
1261 struct pathname lookpn;
1262 nfs_export_t *ne = nfs_get_export();
1263
1264 STRUCT_SET_HANDLE(uap, model, args);
1265
1266 /* Read in pathname from userspace */
1267 if (error = pn_get(STRUCT_FGETP(uap, dname), UIO_USERSPACE, &lookpn))
1268 return (error);
1269
1270 /* Walk the export list looking for that pathname */
1271 rw_enter(&ne->exported_lock, RW_READER);
1272 DTRACE_PROBE(nfss__i__exported_lock1_start);
1273 for (ex1 = ne->exptable_path_hash[pkp_tab_hash(lookpn.pn_path,
1274 strlen(lookpn.pn_path))]; ex1; ex1 = ex1->path_hash.next) {
1275 if (ex1 != ne->exi_root && 0 ==
1276 strcmp(ex1->exi_export.ex_path, lookpn.pn_path)) {
1277 exi_hold(ex1);
1278 break;
1279 }
1280 }
1281 DTRACE_PROBE(nfss__i__exported_lock1_stop);
1282 rw_exit(&ne->exported_lock);
1283
1284 /* Is this an unshare? */
1285 if (STRUCT_FGETP(uap, uex) == NULL) {
1286 pn_free(&lookpn);
1287 if (ex1 == NULL)
1288 return (EINVAL);
1289 error = unexport(ne, ex1);
1290 exi_rele(ex1);
1291 return (error);
1292 }
1293
1294 /* It is a share or a re-share */
1295 error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1296 FOLLOW, &dvp, &vp);
1297 if (error == EINVAL) {
1298 /*
1299 * if fname resolves to / we get EINVAL error
1300 * since we wanted the parent vnode. Try again
1301 * with NULL dvp.
1302 */
1303 error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1304 FOLLOW, NULL, &vp);
1305 dvp = NULL;
1306 }
1307 if (!error && vp == NULL) {
1308 /* Last component of fname not found */
1309 if (dvp != NULL)
1310 VN_RELE(dvp);
1311 error = ENOENT;
1312 }
1313 if (error) {
1314 pn_free(&lookpn);
1315 if (ex1)
1316 exi_rele(ex1);
1317 return (error);
1318 }
1319
1320 /*
1321 * 'vp' may be an AUTOFS node, so we perform a
1322 * VOP_ACCESS() to trigger the mount of the
1323 * intended filesystem, so we can share the intended
1324 * filesystem instead of the AUTOFS filesystem.
1325 */
1326 (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
1327
1328 /*
1329 * We're interested in the top most filesystem.
1330 * This is specially important when uap->dname is a trigger
1331 * AUTOFS node, since we're really interested in sharing the
1332 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
1333 * call not the AUTOFS node itself.
1334 */
1335 if (vn_mountedvfs(vp) != NULL) {
1336 if (error = traverse(&vp)) {
1337 VN_RELE(vp);
1338 if (dvp != NULL)
1339 VN_RELE(dvp);
1340 pn_free(&lookpn);
1341 if (ex1)
1342 exi_rele(ex1);
1343 return (error);
1344 }
1345 }
1346
1347 /* Do not allow sharing another vnode for already shared path */
1348 if (ex1 && !PSEUDO(ex1) && !VN_CMP(ex1->exi_vp, vp)) {
1349 VN_RELE(vp);
1350 if (dvp != NULL)
1351 VN_RELE(dvp);
1352 pn_free(&lookpn);
1353 exi_rele(ex1);
1354 return (EEXIST);
1355 }
1356 if (ex1)
1357 exi_rele(ex1);
1358
1359 /*
1360 * Get the vfs id
1361 */
1362 bzero(&fid, sizeof (fid));
1363 fid.fid_len = MAXFIDSZ;
1364 error = VOP_FID(vp, &fid, NULL);
1365 fsid = vp->v_vfsp->vfs_fsid;
1366
1367 if (error) {
1368 VN_RELE(vp);
1369 if (dvp != NULL)
1370 VN_RELE(dvp);
1371 /*
1372 * If VOP_FID returns ENOSPC then the fid supplied
1373 * is too small. For now we simply return EREMOTE.
1374 */
1375 if (error == ENOSPC)
1376 error = EREMOTE;
1377 pn_free(&lookpn);
1378 return (error);
1379 }
1380
1381 /*
1382 * Do not allow re-sharing a shared vnode under a different path
1383 * PSEUDO export has ex_path fabricated, e.g. "/tmp (pseudo)", skip it.
1384 */
1385 rw_enter(&ne->exported_lock, RW_READER);
1386 DTRACE_PROBE(nfss__i__exported_lock2_start);
1387 for (ex2 = ne->exptable[exptablehash(&fsid, &fid)]; ex2;
1388 ex2 = ex2->fid_hash.next) {
1389 if (ex2 != ne->exi_root && !PSEUDO(ex2) &&
1390 VN_CMP(ex2->exi_vp, vp) &&
1391 strcmp(ex2->exi_export.ex_path, lookpn.pn_path) != 0) {
1392 DTRACE_PROBE(nfss__i__exported_lock2_stop);
1393 rw_exit(&ne->exported_lock);
1394 VN_RELE(vp);
1395 if (dvp != NULL)
1396 VN_RELE(dvp);
1397 pn_free(&lookpn);
1398 return (EEXIST);
1399 }
1400 }
1401 DTRACE_PROBE(nfss__i__exported_lock2_stop);
1402 rw_exit(&ne->exported_lock);
1403 pn_free(&lookpn);
1404
1405 exi = kmem_zalloc(sizeof (*exi), KM_SLEEP);
1406 exi->exi_fsid = fsid;
1407 exi->exi_fid = fid;
1408 exi->exi_vp = vp;
1409 exi->exi_count = 1;
1410 exi->exi_zoneid = crgetzoneid(cr);
1411 ASSERT3U(exi->exi_zoneid, ==, curzone->zone_id);
1412 exi->exi_volatile_dev = (vfssw[vp->v_vfsp->vfs_fstype].vsw_flag &
1413 VSW_VOLATILEDEV) ? 1 : 0;
1414 mutex_init(&exi->exi_lock, NULL, MUTEX_DEFAULT, NULL);
1415 exi->exi_dvp = dvp;
1416
1417 /*
1418 * Initialize auth cache and auth cache lock
1419 */
1420 for (i = 0; i < AUTH_TABLESIZE; i++) {
1421 exi->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1422 avl_create(exi->exi_cache[i], nfsauth_cache_clnt_compar,
1423 sizeof (struct auth_cache_clnt),
1424 offsetof(struct auth_cache_clnt, authc_link));
1425 }
1426 rw_init(&exi->exi_cache_lock, NULL, RW_DEFAULT, NULL);
1427
1428 /*
1429 * Build up the template fhandle
1430 */
1431 exi->exi_fh.fh_fsid = fsid;
1432 if (exi->exi_fid.fid_len > sizeof (exi->exi_fh.fh_xdata)) {
1433 error = EREMOTE;
1434 goto out1;
1435 }
1436 exi->exi_fh.fh_xlen = exi->exi_fid.fid_len;
1437 bcopy(exi->exi_fid.fid_data, exi->exi_fh.fh_xdata,
1438 exi->exi_fid.fid_len);
1439
1440 exi->exi_fh.fh_len = sizeof (exi->exi_fh.fh_data);
1441
1442 kex = &exi->exi_export;
1443
1444 /*
1445 * Load in everything, and do sanity checking
1446 */
1447 STRUCT_INIT(uexi, model);
1448 if (copyin(STRUCT_FGETP(uap, uex), STRUCT_BUF(uexi),
1449 STRUCT_SIZE(uexi))) {
1450 error = EFAULT;
1451 goto out1;
1452 }
1453
1454 kex->ex_version = STRUCT_FGET(uexi, ex_version);
1455 if (kex->ex_version != EX_CURRENT_VERSION) {
1456 error = EINVAL;
1457 cmn_err(CE_WARN,
1458 "NFS: exportfs requires export struct version 2 - got %d\n",
1459 kex->ex_version);
1460 goto out1;
1461 }
1462
1463 /*
1464 * Must have at least one security entry
1465 */
1466 kex->ex_seccnt = STRUCT_FGET(uexi, ex_seccnt);
1467 if (kex->ex_seccnt < 1) {
1468 error = EINVAL;
1469 goto out1;
1470 }
1471
1472 kex->ex_path = STRUCT_FGETP(uexi, ex_path);
1473 kex->ex_pathlen = STRUCT_FGET(uexi, ex_pathlen);
1474 kex->ex_flags = STRUCT_FGET(uexi, ex_flags);
1475 kex->ex_anon = STRUCT_FGET(uexi, ex_anon);
1476 kex->ex_secinfo = STRUCT_FGETP(uexi, ex_secinfo);
1477 kex->ex_index = STRUCT_FGETP(uexi, ex_index);
1478 kex->ex_log_buffer = STRUCT_FGETP(uexi, ex_log_buffer);
1479 kex->ex_log_bufferlen = STRUCT_FGET(uexi, ex_log_bufferlen);
1480 kex->ex_tag = STRUCT_FGETP(uexi, ex_tag);
1481 kex->ex_taglen = STRUCT_FGET(uexi, ex_taglen);
1482
1483 /*
1484 * Copy the exported pathname into
1485 * an appropriately sized buffer.
1486 */
1487 pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1488 if (copyinstr(kex->ex_path, pathbuf, MAXPATHLEN, &kex->ex_pathlen)) {
1489 kmem_free(pathbuf, MAXPATHLEN);
1490 error = EFAULT;
1491 goto out1;
1492 }
1493 kex->ex_path = kmem_alloc(kex->ex_pathlen + 1, KM_SLEEP);
1494 bcopy(pathbuf, kex->ex_path, kex->ex_pathlen);
1495 kex->ex_path[kex->ex_pathlen] = '\0';
1496 kmem_free(pathbuf, MAXPATHLEN);
1497
1498 /*
1499 * Get the path to the logging buffer and the tag
1500 */
1501 if (kex->ex_flags & EX_LOG) {
1502 log_buffer = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1503 if (copyinstr(kex->ex_log_buffer, log_buffer, MAXPATHLEN,
1504 &kex->ex_log_bufferlen)) {
1505 kmem_free(log_buffer, MAXPATHLEN);
1506 error = EFAULT;
1507 goto out2;
1508 }
1509 kex->ex_log_buffer =
1510 kmem_alloc(kex->ex_log_bufferlen + 1, KM_SLEEP);
1511 bcopy(log_buffer, kex->ex_log_buffer, kex->ex_log_bufferlen);
1512 kex->ex_log_buffer[kex->ex_log_bufferlen] = '\0';
1513 kmem_free(log_buffer, MAXPATHLEN);
1514
1515 tagbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1516 if (copyinstr(kex->ex_tag, tagbuf, MAXPATHLEN,
1517 &kex->ex_taglen)) {
1518 kmem_free(tagbuf, MAXPATHLEN);
1519 error = EFAULT;
1520 goto out3;
1521 }
1522 kex->ex_tag = kmem_alloc(kex->ex_taglen + 1, KM_SLEEP);
1523 bcopy(tagbuf, kex->ex_tag, kex->ex_taglen);
1524 kex->ex_tag[kex->ex_taglen] = '\0';
1525 kmem_free(tagbuf, MAXPATHLEN);
1526 }
1527
1528 /*
1529 * Load the security information for each flavor
1530 */
1531 allocsize = kex->ex_seccnt * SIZEOF_STRUCT(secinfo, model);
1532 sp = kmem_zalloc(allocsize, KM_SLEEP);
1533 if (copyin(kex->ex_secinfo, sp, allocsize)) {
1534 kmem_free(sp, allocsize);
1535 error = EFAULT;
1536 goto out4;
1537 }
1538
1539 /*
1540 * All of these nested structures need to be converted to
1541 * the kernel native format.
1542 */
1543 if (model != DATAMODEL_NATIVE) {
1544 size_t allocsize2;
1545 struct secinfo *sp2;
1546
1547 allocsize2 = kex->ex_seccnt * sizeof (struct secinfo);
1548 sp2 = kmem_zalloc(allocsize2, KM_SLEEP);
1549
1550 for (i = 0; i < kex->ex_seccnt; i++) {
1551 STRUCT_HANDLE(secinfo, usi);
1552
1553 STRUCT_SET_HANDLE(usi, model,
1554 (struct secinfo *)((caddr_t)sp +
1555 (i * SIZEOF_STRUCT(secinfo, model))));
1556 bcopy(STRUCT_FGET(usi, s_secinfo.sc_name),
1557 sp2[i].s_secinfo.sc_name, MAX_NAME_LEN);
1558 sp2[i].s_secinfo.sc_nfsnum =
1559 STRUCT_FGET(usi, s_secinfo.sc_nfsnum);
1560 sp2[i].s_secinfo.sc_rpcnum =
1561 STRUCT_FGET(usi, s_secinfo.sc_rpcnum);
1562 bcopy(STRUCT_FGET(usi, s_secinfo.sc_gss_mech),
1563 sp2[i].s_secinfo.sc_gss_mech, MAX_NAME_LEN);
1564 sp2[i].s_secinfo.sc_gss_mech_type =
1565 STRUCT_FGETP(usi, s_secinfo.sc_gss_mech_type);
1566 sp2[i].s_secinfo.sc_qop =
1567 STRUCT_FGET(usi, s_secinfo.sc_qop);
1568 sp2[i].s_secinfo.sc_service =
1569 STRUCT_FGET(usi, s_secinfo.sc_service);
1570
1571 sp2[i].s_flags = STRUCT_FGET(usi, s_flags);
1572 sp2[i].s_window = STRUCT_FGET(usi, s_window);
1573 sp2[i].s_rootid = STRUCT_FGET(usi, s_rootid);
1574 sp2[i].s_rootcnt = STRUCT_FGET(usi, s_rootcnt);
1575 sp2[i].s_rootnames = STRUCT_FGETP(usi, s_rootnames);
1576 }
1577 kmem_free(sp, allocsize);
1578 sp = sp2;
1579 allocsize = allocsize2;
1580 }
1581
1582 kex->ex_secinfo = sp;
1583
1584 /*
1585 * And now copy rootnames for each individual secinfo.
1586 */
1587 callback = 0;
1588 allocd_seccnt = 0;
1589 while (allocd_seccnt < kex->ex_seccnt) {
1590
1591 exs = &sp[allocd_seccnt];
1592 if (exs->s_rootcnt > 0) {
1593 if (!sec_svc_loadrootnames(exs->s_secinfo.sc_rpcnum,
1594 exs->s_rootcnt, &exs->s_rootnames, model)) {
1595 error = EFAULT;
1596 goto out5;
1597 }
1598 }
1599
1600 if (exs->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
1601 rpc_gss_OID mech_tmp;
1602 STRUCT_DECL(rpc_gss_OID_s, umech_tmp);
1603 caddr_t elements_tmp;
1604
1605 /* Copyin mechanism type */
1606 STRUCT_INIT(umech_tmp, model);
1607 mech_tmp = kmem_alloc(sizeof (*mech_tmp), KM_SLEEP);
1608 if (copyin(exs->s_secinfo.sc_gss_mech_type,
1609 STRUCT_BUF(umech_tmp), STRUCT_SIZE(umech_tmp))) {
1610 kmem_free(mech_tmp, sizeof (*mech_tmp));
1611 error = EFAULT;
1612 goto out5;
1613 }
1614 mech_tmp->length = STRUCT_FGET(umech_tmp, length);
1615 mech_tmp->elements = STRUCT_FGETP(umech_tmp, elements);
1616
1617 elements_tmp = kmem_alloc(mech_tmp->length, KM_SLEEP);
1618 if (copyin(mech_tmp->elements, elements_tmp,
1619 mech_tmp->length)) {
1620 kmem_free(elements_tmp, mech_tmp->length);
1621 kmem_free(mech_tmp, sizeof (*mech_tmp));
1622 error = EFAULT;
1623 goto out5;
1624 }
1625 mech_tmp->elements = elements_tmp;
1626 exs->s_secinfo.sc_gss_mech_type = mech_tmp;
1627 allocd_seccnt++;
1628
1629 callback = 1;
1630 } else
1631 allocd_seccnt++;
1632 }
1633
1634 /*
1635 * Init the secinfo reference count and mark these flavors
1636 * explicitly exported flavors.
1637 */
1638 for (i = 0; i < kex->ex_seccnt; i++) {
1639 kex->ex_secinfo[i].s_flags |= M_4SEC_EXPORTED;
1640 kex->ex_secinfo[i].s_refcnt = 1;
1641 }
1642
1643 /*
1644 * Set up rpcsec_gss callback routine entry if any.
1645 */
1646 if (callback) {
1647 cb.callback = rfs_gsscallback;
1648 cb.program = NFS_ACL_PROGRAM;
1649 for (cb.version = NFS_ACL_VERSMIN;
1650 cb.version <= NFS_ACL_VERSMAX; cb.version++) {
1651 (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1652 (void *)&cb);
1653 }
1654
1655 cb.program = NFS_PROGRAM;
1656 for (cb.version = NFS_VERSMIN;
1657 cb.version <= NFS_VERSMAX; cb.version++) {
1658 (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1659 (void *)&cb);
1660 }
1661 }
1662
1663 /*
1664 * Check the index flag. Do this here to avoid holding the
1665 * lock while dealing with the index option (as we do with
1666 * the public option).
1667 */
1668 if (kex->ex_flags & EX_INDEX) {
1669 if (!kex->ex_index) { /* sanity check */
1670 error = EINVAL;
1671 goto out5;
1672 }
1673 if (error = loadindex(kex))
1674 goto out5;
1675 }
1676
1677 if (kex->ex_flags & EX_LOG) {
1678 if (error = nfslog_setup(exi))
1679 goto out6;
1680 }
1681
1682 /*
1683 * Insert the new entry at the front of the export list
1684 */
1685 rw_enter(&ne->exported_lock, RW_WRITER);
1686 DTRACE_PROBE(nfss__i__exported_lock3_start);
1687
1688 export_link(ne, exi);
1689
1690 /*
1691 * Check the rest of the list for an old entry for the fs.
1692 * If one is found then unlink it, wait until this is the
1693 * only reference and then free it.
1694 */
1695 for (ex = exi->fid_hash.next; ex != NULL; ex = ex->fid_hash.next) {
1696 if (ex != ne->exi_root && VN_CMP(ex->exi_vp, vp)) {
1697 mutex_enter(&nfs_exi_id_lock);
1698 avl_remove(&exi_id_tree, ex);
1699 mutex_exit(&nfs_exi_id_lock);
1700 export_unlink(ne, ex);
1701 break;
1702 }
1703 }
1704
1705 /*
1706 * If the public filehandle is pointing at the
1707 * old entry, then point it back at the root.
1708 */
1709 if (ex != NULL && ex == ne->exi_public)
1710 ne->exi_public = ne->exi_root;
1711
1712 /*
1713 * If the public flag is on, make the global exi_public
1714 * point to this entry and turn off the public bit so that
1715 * we can distinguish it from the place holder export.
1716 */
1717 if (kex->ex_flags & EX_PUBLIC) {
1718 ne->exi_public = exi;
1719 kex->ex_flags &= ~EX_PUBLIC;
1720 }
1721
1722 #ifdef VOLATILE_FH_TEST
1723 /*
1724 * Set up the volatile_id value if volatile on share.
1725 * The list of volatile renamed filehandles is always destroyed,
1726 * if the fs was reshared.
1727 */
1728 if (kex->ex_flags & EX_VOLFH)
1729 exi->exi_volatile_id = gethrestime_sec();
1730
1731 mutex_init(&exi->exi_vol_rename_lock, NULL, MUTEX_DEFAULT, NULL);
1732 #endif /* VOLATILE_FH_TEST */
1733
1734 /*
1735 * If this is a new export, then climb up
1736 * the tree and check if any pseudo exports
1737 * need to be created to provide a path for
1738 * NFS v4 clients.
1739 */
1740 if (ex == NULL) {
1741 error = treeclimb_export(exi);
1742 if (error)
1743 goto out7;
1744 } else {
1745 /* If it's a re-export update namespace tree */
1746 exi->exi_tree = ex->exi_tree;
1747 exi->exi_tree->tree_exi = exi;
1748
1749 /* Update the change timestamp */
1750 tree_update_change(ne, exi->exi_tree, NULL);
1751 }
1752
1753 /*
1754 * build a unique flavor list from the flavors specified
1755 * in the share cmd. unique means that each flavor only
1756 * appears once in the secinfo list -- no duplicates allowed.
1757 */
1758 newcnt = build_seclist_nodups(&exi->exi_export, newsec, FALSE);
1759
1760 srv_secinfo_treeclimb(ne, exi, newsec, newcnt, TRUE);
1761
1762 /*
1763 * If re-sharing an old export entry, update the secinfo data
1764 * depending on if the old entry is a pseudo node or not.
1765 */
1766 if (ex != NULL) {
1767 oldcnt = build_seclist_nodups(&ex->exi_export, oldsec, FALSE);
1768 if (PSEUDO(ex)) {
1769 /*
1770 * The dir being shared is a pseudo export root (which
1771 * will be transformed into a real export root). The
1772 * flavor(s) of the new share were propagated to the
1773 * ancestors by srv_secinfo_treeclimb() above. Now
1774 * transfer the implicit flavor refs from the old
1775 * pseudo exprot root to the new (real) export root.
1776 */
1777 srv_secinfo_add(&exi->exi_export.ex_secinfo,
1778 &exi->exi_export.ex_seccnt, oldsec, oldcnt, TRUE);
1779 } else {
1780 /*
1781 * First transfer implicit flavor refs to new export.
1782 * Remove old flavor refs last.
1783 */
1784 srv_secinfo_exp2exp(&exi->exi_export, oldsec, oldcnt);
1785 srv_secinfo_treeclimb(ne, ex, oldsec, oldcnt, FALSE);
1786 }
1787 }
1788
1789 /*
1790 * If it's a re-export and the old entry has a pseudonode list,
1791 * transfer it to the new export.
1792 */
1793 if (ex != NULL && (ex->exi_visible != NULL)) {
1794 exi->exi_visible = ex->exi_visible;
1795 ex->exi_visible = NULL;
1796 }
1797
1798 /*
1799 * Initialize exi_id and exi_kstats
1800 */
1801 if (ex != NULL) {
1802 exi->exi_id = ex->exi_id;
1803 } else {
1804 mutex_enter(&nfs_exi_id_lock);
1805 exi->exi_id = exi_id_get_next();
1806 mutex_exit(&nfs_exi_id_lock);
1807 }
1808 mutex_enter(&nfs_exi_id_lock);
1809 avl_add(&exi_id_tree, exi);
1810 mutex_exit(&nfs_exi_id_lock);
1811
1812 DTRACE_PROBE(nfss__i__exported_lock3_stop);
1813 rw_exit(&ne->exported_lock);
1814
1815 if (ne->exi_public == exi || kex->ex_flags & EX_LOG) {
1816 /*
1817 * Log share operation to this buffer only.
1818 */
1819 nfslog_share_record(exi, cr);
1820 }
1821
1822 if (ex != NULL)
1823 exi_rele(ex);
1824
1825 return (0);
1826
1827 out7:
1828 /* Unlink the new export in exptable. */
1829 export_unlink(ne, exi);
1830 DTRACE_PROBE(nfss__i__exported_lock3_stop);
1831 rw_exit(&ne->exported_lock);
1832 out6:
1833 if (kex->ex_flags & EX_INDEX)
1834 kmem_free(kex->ex_index, strlen(kex->ex_index) + 1);
1835 out5:
1836 /* free partially completed allocation */
1837 while (--allocd_seccnt >= 0) {
1838 exs = &kex->ex_secinfo[allocd_seccnt];
1839 srv_secinfo_entry_free(exs);
1840 }
1841
1842 if (kex->ex_secinfo) {
1843 kmem_free(kex->ex_secinfo,
1844 kex->ex_seccnt * sizeof (struct secinfo));
1845 }
1846
1847 out4:
1848 if ((kex->ex_flags & EX_LOG) && kex->ex_tag != NULL)
1849 kmem_free(kex->ex_tag, kex->ex_taglen + 1);
1850 out3:
1851 if ((kex->ex_flags & EX_LOG) && kex->ex_log_buffer != NULL)
1852 kmem_free(kex->ex_log_buffer, kex->ex_log_bufferlen + 1);
1853 out2:
1854 kmem_free(kex->ex_path, kex->ex_pathlen + 1);
1855 out1:
1856 VN_RELE(vp);
1857 if (dvp != NULL)
1858 VN_RELE(dvp);
1859 mutex_destroy(&exi->exi_lock);
1860 rw_destroy(&exi->exi_cache_lock);
1861 for (i = 0; i < AUTH_TABLESIZE; i++) {
1862 avl_destroy(exi->exi_cache[i]);
1863 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
1864 }
1865
1866 kmem_free(exi, sizeof (*exi));
1867
1868 return (error);
1869 }
1870
1871 /*
1872 * Remove the exportinfo from the export list
1873 */
1874 void
1875 export_unlink(nfs_export_t *ne, struct exportinfo *exi)
1876 {
1877 ASSERT(RW_WRITE_HELD(&ne->exported_lock));
1878
1879 exp_hash_unlink(exi, fid_hash);
1880 exp_hash_unlink(exi, path_hash);
1881 ASSERT3P(exi->exi_ne, ==, ne);
1882 exi->exi_ne = NULL;
1883 }
1884
1885 /*
1886 * Unexport an exported filesystem
1887 */
1888 static int
1889 unexport(nfs_export_t *ne, struct exportinfo *exi)
1890 {
1891 struct secinfo cursec[MAX_FLAVORS];
1892 int curcnt;
1893
1894 rw_enter(&ne->exported_lock, RW_WRITER);
1895
1896 /* Check if exi is still linked in the export table */
1897 if (!EXP_LINKED(exi) || PSEUDO(exi)) {
1898 rw_exit(&ne->exported_lock);
1899 return (EINVAL);
1900 }
1901
1902 mutex_enter(&nfs_exi_id_lock);
1903 avl_remove(&exi_id_tree, exi);
1904 mutex_exit(&nfs_exi_id_lock);
1905 export_unlink(ne, exi);
1906
1907 /*
1908 * Remove security flavors before treeclimb_unexport() is called
1909 * because srv_secinfo_treeclimb needs the namespace tree
1910 */
1911 curcnt = build_seclist_nodups(&exi->exi_export, cursec, TRUE);
1912 srv_secinfo_treeclimb(ne, exi, cursec, curcnt, FALSE);
1913
1914 /*
1915 * If there's a visible list, then need to leave
1916 * a pseudo export here to retain the visible list
1917 * for paths to exports below.
1918 */
1919 if (exi->exi_visible != NULL) {
1920 struct exportinfo *newexi;
1921
1922 newexi = pseudo_exportfs(ne, exi->exi_vp, &exi->exi_fid,
1923 exi->exi_visible, &exi->exi_export);
1924 exi->exi_visible = NULL;
1925
1926 /* interconnect the existing treenode with the new exportinfo */
1927 newexi->exi_tree = exi->exi_tree;
1928 newexi->exi_tree->tree_exi = newexi;
1929
1930 /* Update the change timestamp */
1931 tree_update_change(ne, exi->exi_tree, NULL);
1932 } else {
1933 treeclimb_unexport(ne, exi);
1934 }
1935
1936 rw_exit(&ne->exported_lock);
1937
1938 /*
1939 * Need to call into the NFSv4 server and release all data
1940 * held on this particular export. This is important since
1941 * the v4 server may be holding file locks or vnodes under
1942 * this export.
1943 */
1944 rfs4_clean_state_exi(ne, exi);
1945
1946 /*
1947 * Notify the lock manager that the filesystem is being
1948 * unexported.
1949 */
1950 lm_unexport(exi);
1951
1952 /*
1953 * If this was a public export, restore
1954 * the public filehandle to the root.
1955 */
1956
1957 /*
1958 * XXX KEBE ASKS --> Should CRED() instead be
1959 * exi->exi_zone->zone_kcred?
1960 */
1961 if (exi == ne->exi_public) {
1962 ne->exi_public = ne->exi_root;
1963
1964 nfslog_share_record(ne->exi_public, CRED());
1965 }
1966
1967 if (exi->exi_export.ex_flags & EX_LOG)
1968 nfslog_unshare_record(exi, CRED());
1969
1970 exi_rele(exi);
1971 return (0);
1972 }
1973
1974 /*
1975 * Get file handle system call.
1976 * Takes file name and returns a file handle for it.
1977 * Credentials must be verified before calling.
1978 */
1979 int
1980 nfs_getfh(struct nfs_getfh_args *args, model_t model, cred_t *cr)
1981 {
1982 nfs_fh3 fh;
1983 char buf[NFS3_MAXFHSIZE];
1984 char *logptr, logbuf[NFS3_MAXFHSIZE];
1985 int l = NFS3_MAXFHSIZE;
1986 vnode_t *vp;
1987 vnode_t *dvp;
1988 struct exportinfo *exi;
1989 int error;
1990 int vers;
1991 STRUCT_HANDLE(nfs_getfh_args, uap);
1992
1993 #ifdef lint
1994 model = model; /* STRUCT macros don't always use it */
1995 #endif
1996
1997 STRUCT_SET_HANDLE(uap, model, args);
1998
1999 error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
2000 FOLLOW, &dvp, &vp);
2001 if (error == EINVAL) {
2002 /*
2003 * if fname resolves to / we get EINVAL error
2004 * since we wanted the parent vnode. Try again
2005 * with NULL dvp.
2006 */
2007 error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
2008 FOLLOW, NULL, &vp);
2009 dvp = NULL;
2010 }
2011 if (!error && vp == NULL) {
2012 /*
2013 * Last component of fname not found
2014 */
2015 if (dvp != NULL) {
2016 VN_RELE(dvp);
2017 }
2018 error = ENOENT;
2019 }
2020 if (error)
2021 return (error);
2022
2023 /*
2024 * 'vp' may be an AUTOFS node, so we perform a
2025 * VOP_ACCESS() to trigger the mount of the
2026 * intended filesystem, so we can share the intended
2027 * filesystem instead of the AUTOFS filesystem.
2028 */
2029 (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
2030
2031 /*
2032 * We're interested in the top most filesystem.
2033 * This is specially important when uap->dname is a trigger
2034 * AUTOFS node, since we're really interested in sharing the
2035 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
2036 * call not the AUTOFS node itself.
2037 */
2038 if (vn_mountedvfs(vp) != NULL) {
2039 if (error = traverse(&vp)) {
2040 VN_RELE(vp);
2041 if (dvp != NULL)
2042 VN_RELE(dvp);
2043 return (error);
2044 }
2045 }
2046
2047 vers = STRUCT_FGET(uap, vers);
2048 exi = nfs_vptoexi(dvp, vp, cr, NULL, &error, FALSE);
2049 if (!error) {
2050 if (vers == NFS_VERSION) {
2051 error = makefh((fhandle_t *)buf, vp, exi);
2052 l = NFS_FHSIZE;
2053 logptr = buf;
2054 } else if (vers == NFS_V3) {
2055 int i, sz, pad;
2056
2057 error = makefh3(&fh, vp, exi);
2058 l = RNDUP(fh.fh3_length);
2059 if (!error && (l > sizeof (fhandle3_t)))
2060 error = EREMOTE;
2061 logptr = logbuf;
2062 if (!error) {
2063 i = 0;
2064 sz = sizeof (fsid_t);
2065 bcopy(&fh.fh3_fsid, &buf[i], sz);
2066 i += sz;
2067
2068 /*
2069 * For backwards compatibility, the
2070 * fid length may be less than
2071 * NFS_FHMAXDATA, but it was always
2072 * encoded as NFS_FHMAXDATA bytes.
2073 */
2074
2075 sz = sizeof (ushort_t);
2076 bcopy(&fh.fh3_len, &buf[i], sz);
2077 i += sz;
2078 bcopy(fh.fh3_data, &buf[i], fh.fh3_len);
2079 i += fh.fh3_len;
2080 pad = (NFS_FHMAXDATA - fh.fh3_len);
2081 if (pad > 0) {
2082 bzero(&buf[i], pad);
2083 i += pad;
2084 l += pad;
2085 }
2086
2087 sz = sizeof (ushort_t);
2088 bcopy(&fh.fh3_xlen, &buf[i], sz);
2089 i += sz;
2090 bcopy(fh.fh3_xdata, &buf[i], fh.fh3_xlen);
2091 i += fh.fh3_xlen;
2092 pad = (NFS_FHMAXDATA - fh.fh3_xlen);
2093 if (pad > 0) {
2094 bzero(&buf[i], pad);
2095 i += pad;
2096 l += pad;
2097 }
2098 }
2099 /*
2100 * If we need to do NFS logging, the filehandle
2101 * must be downsized to 32 bytes.
2102 */
2103 if (!error && exi->exi_export.ex_flags & EX_LOG) {
2104 i = 0;
2105 sz = sizeof (fsid_t);
2106 bcopy(&fh.fh3_fsid, &logbuf[i], sz);
2107 i += sz;
2108 sz = sizeof (ushort_t);
2109 bcopy(&fh.fh3_len, &logbuf[i], sz);
2110 i += sz;
2111 sz = NFS_FHMAXDATA;
2112 bcopy(fh.fh3_data, &logbuf[i], sz);
2113 i += sz;
2114 sz = sizeof (ushort_t);
2115 bcopy(&fh.fh3_xlen, &logbuf[i], sz);
2116 i += sz;
2117 sz = NFS_FHMAXDATA;
2118 bcopy(fh.fh3_xdata, &logbuf[i], sz);
2119 i += sz;
2120 }
2121 }
2122 if (!error && exi->exi_export.ex_flags & EX_LOG) {
2123 nfslog_getfh(exi, (fhandle_t *)logptr,
2124 STRUCT_FGETP(uap, fname), UIO_USERSPACE, cr);
2125 }
2126 exi_rele(exi);
2127 if (!error) {
2128 if (copyout(&l, STRUCT_FGETP(uap, lenp), sizeof (int)))
2129 error = EFAULT;
2130 if (copyout(buf, STRUCT_FGETP(uap, fhp), l))
2131 error = EFAULT;
2132 }
2133 }
2134 VN_RELE(vp);
2135 if (dvp != NULL) {
2136 VN_RELE(dvp);
2137 }
2138 return (error);
2139 }
2140
2141 /*
2142 * Strategy: if vp is in the export list, then
2143 * return the associated file handle. Otherwise, ".."
2144 * once up the vp and try again, until the root of the
2145 * filesystem is reached.
2146 */
2147 struct exportinfo *
2148 nfs_vptoexi(vnode_t *dvp, vnode_t *vp, cred_t *cr, int *walk,
2149 int *err, bool_t v4srv)
2150 {
2151 fid_t fid;
2152 int error;
2153 struct exportinfo *exi;
2154
2155 ASSERT(vp);
2156 VN_HOLD(vp);
2157 if (dvp != NULL) {
2158 VN_HOLD(dvp);
2159 }
2160 if (walk != NULL)
2161 *walk = 0;
2162
2163 for (;;) {
2164 bzero(&fid, sizeof (fid));
2165 fid.fid_len = MAXFIDSZ;
2166 error = vop_fid_pseudo(vp, &fid);
2167 if (error) {
2168 /*
2169 * If vop_fid_pseudo returns ENOSPC then the fid
2170 * supplied is too small. For now we simply
2171 * return EREMOTE.
2172 */
2173 if (error == ENOSPC)
2174 error = EREMOTE;
2175 break;
2176 }
2177
2178 if (v4srv)
2179 exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
2180 else
2181 exi = checkexport(&vp->v_vfsp->vfs_fsid, &fid);
2182
2183 if (exi != NULL) {
2184 /*
2185 * Found the export info
2186 */
2187 break;
2188 }
2189
2190 /*
2191 * We have just failed finding a matching export.
2192 * If we're at the root of this filesystem, then
2193 * it's time to stop (with failure).
2194 */
2195 ASSERT3P(vp->v_vfsp->vfs_zone, ==, curzone);
2196 if ((vp->v_flag & VROOT) || VN_IS_CURZONEROOT(vp)) {
2197 error = EINVAL;
2198 break;
2199 }
2200
2201 if (walk != NULL)
2202 (*walk)++;
2203
2204 /*
2205 * Now, do a ".." up vp. If dvp is supplied, use it,
2206 * otherwise, look it up.
2207 */
2208 if (dvp == NULL) {
2209 error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, cr,
2210 NULL, NULL, NULL);
2211 if (error)
2212 break;
2213 }
2214 VN_RELE(vp);
2215 vp = dvp;
2216 dvp = NULL;
2217 }
2218 VN_RELE(vp);
2219 if (dvp != NULL) {
2220 VN_RELE(dvp);
2221 }
2222 if (error != 0) {
2223 if (err != NULL)
2224 *err = error;
2225 return (NULL);
2226 }
2227 return (exi);
2228 }
2229
2230 int
2231 chk_clnt_sec(exportinfo_t *exi, struct svc_req *req)
2232 {
2233 int i, nfsflavor;
2234 struct secinfo *sp;
2235
2236 /*
2237 * Get the nfs flavor number from xprt.
2238 */
2239 nfsflavor = (int)(uintptr_t)req->rq_xprt->xp_cookie;
2240
2241 sp = exi->exi_export.ex_secinfo;
2242 for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2243 if ((nfsflavor == sp[i].s_secinfo.sc_nfsnum) &&
2244 SEC_REF_EXPORTED(sp + i))
2245 return (TRUE);
2246 }
2247 return (FALSE);
2248 }
2249
2250 /*
2251 * Make an fhandle from a vnode
2252 */
2253 int
2254 makefh(fhandle_t *fh, vnode_t *vp, exportinfo_t *exi)
2255 {
2256 int error;
2257
2258 *fh = exi->exi_fh; /* struct copy */
2259
2260 error = VOP_FID(vp, (fid_t *)&fh->fh_len, NULL);
2261 if (error) {
2262 /*
2263 * Should be something other than EREMOTE
2264 */
2265 return (EREMOTE);
2266 }
2267 return (0);
2268 }
2269
2270 /*
2271 * This routine makes an overloaded V2 fhandle which contains
2272 * sec modes.
2273 *
2274 * Note that the first four octets contain the length octet,
2275 * the status octet, and two padded octets to make them XDR
2276 * four-octet aligned.
2277 *
2278 * 1 2 3 4 32
2279 * +---+---+---+---+---+---+---+---+ +---+---+---+---+ +---+
2280 * | l | s | | | sec_1 |...| sec_n |...| |
2281 * +---+---+---+---+---+---+---+---+ +---+---+---+---+ +---+
2282 *
2283 * where
2284 *
2285 * the status octet s indicates whether there are more security
2286 * flavors (1 means yes, 0 means no) that require the client to
2287 * perform another 0x81 LOOKUP to get them,
2288 *
2289 * the length octet l is the length describing the number of
2290 * valid octets that follow. (l = 4 * n, where n is the number
2291 * of security flavors sent in the current overloaded filehandle.)
2292 *
2293 * sec_index should always be in the inclusive range: [1 - ex_seccnt],
2294 * and it tells server where to start within the secinfo array.
2295 * Usually it will always be 1; however, if more flavors are used
2296 * for the public export than can be encoded in the overloaded FH
2297 * (7 for NFS2), subsequent SNEGO MCLs will have a larger index
2298 * so the server will pick up where it left off from the previous
2299 * MCL reply.
2300 *
2301 * With NFS4 support, implicitly allowed flavors are also in
2302 * the secinfo array; however, they should not be returned in
2303 * SNEGO MCL replies.
2304 */
2305 int
2306 makefh_ol(fhandle_t *fh, exportinfo_t *exi, uint_t sec_index)
2307 {
2308 secinfo_t sec[MAX_FLAVORS];
2309 int totalcnt, i, *ipt, cnt, seccnt, secidx, fh_max_cnt;
2310 char *c;
2311
2312 if (fh == NULL || exi == NULL || sec_index < 1)
2313 return (EREMOTE);
2314
2315 /*
2316 * WebNFS clients need to know the unique set of explicitly
2317 * shared flavors in used for the public export. When
2318 * "TRUE" is passed to build_seclist_nodups(), only explicitly
2319 * shared flavors are included in the list.
2320 */
2321 seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2322 if (sec_index > seccnt)
2323 return (EREMOTE);
2324
2325 fh_max_cnt = (NFS_FHSIZE / sizeof (int)) - 1;
2326 totalcnt = seccnt - sec_index + 1;
2327 cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2328
2329 c = (char *)fh;
2330 /*
2331 * Encode the length octet representing the number of
2332 * security flavors (in bytes) in this overloaded fh.
2333 */
2334 *c = cnt * sizeof (int);
2335
2336 /*
2337 * Encode the status octet that indicates whether there
2338 * are more security flavors the client needs to get.
2339 */
2340 *(c + 1) = totalcnt > fh_max_cnt;
2341
2342 /*
2343 * put security flavors in the overloaded fh
2344 */
2345 ipt = (int *)(c + sizeof (int32_t));
2346 secidx = sec_index - 1;
2347 for (i = 0; i < cnt; i++) {
2348 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2349 }
2350 return (0);
2351 }
2352
2353 /*
2354 * Make an nfs_fh3 from a vnode
2355 */
2356 int
2357 makefh3(nfs_fh3 *fh, vnode_t *vp, struct exportinfo *exi)
2358 {
2359 int error;
2360 fid_t fid;
2361
2362 bzero(&fid, sizeof (fid));
2363 fid.fid_len = sizeof (fh->fh3_data);
2364 error = VOP_FID(vp, &fid, NULL);
2365 if (error)
2366 return (EREMOTE);
2367
2368 bzero(fh, sizeof (nfs_fh3));
2369 fh->fh3_fsid = exi->exi_fsid;
2370 fh->fh3_len = fid.fid_len;
2371 bcopy(fid.fid_data, fh->fh3_data, fh->fh3_len);
2372
2373 fh->fh3_xlen = exi->exi_fid.fid_len;
2374 ASSERT(fh->fh3_xlen <= sizeof (fh->fh3_xdata));
2375 bcopy(exi->exi_fid.fid_data, fh->fh3_xdata, fh->fh3_xlen);
2376
2377 fh->fh3_length = sizeof (fh->fh3_fsid)
2378 + sizeof (fh->fh3_len) + fh->fh3_len
2379 + sizeof (fh->fh3_xlen) + fh->fh3_xlen;
2380 fh->fh3_flags = 0;
2381
2382 return (0);
2383 }
2384
2385 /*
2386 * This routine makes an overloaded V3 fhandle which contains
2387 * sec modes.
2388 *
2389 * 1 4
2390 * +--+--+--+--+
2391 * | len |
2392 * +--+--+--+--+
2393 * up to 64
2394 * +--+--+--+--+--+--+--+--+--+--+--+--+ +--+--+--+--+
2395 * |s | | | | sec_1 | sec_2 | ... | sec_n |
2396 * +--+--+--+--+--+--+--+--+--+--+--+--+ +--+--+--+--+
2397 *
2398 * len = 4 * (n+1), where n is the number of security flavors
2399 * sent in the current overloaded filehandle.
2400 *
2401 * the status octet s indicates whether there are more security
2402 * mechanisms (1 means yes, 0 means no) that require the client
2403 * to perform another 0x81 LOOKUP to get them.
2404 *
2405 * Three octets are padded after the status octet.
2406 */
2407 int
2408 makefh3_ol(nfs_fh3 *fh, struct exportinfo *exi, uint_t sec_index)
2409 {
2410 secinfo_t sec[MAX_FLAVORS];
2411 int totalcnt, cnt, *ipt, i, seccnt, fh_max_cnt, secidx;
2412 char *c;
2413
2414 if (fh == NULL || exi == NULL || sec_index < 1)
2415 return (EREMOTE);
2416
2417 /*
2418 * WebNFS clients need to know the unique set of explicitly
2419 * shared flavors in used for the public export. When
2420 * "TRUE" is passed to build_seclist_nodups(), only explicitly
2421 * shared flavors are included in the list.
2422 */
2423 seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2424
2425 if (sec_index > seccnt)
2426 return (EREMOTE);
2427
2428 fh_max_cnt = (NFS3_FHSIZE / sizeof (int)) - 1;
2429 totalcnt = seccnt - sec_index + 1;
2430 cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2431
2432 /*
2433 * Place the length in fh3_length representing the number
2434 * of security flavors (in bytes) in this overloaded fh.
2435 */
2436 fh->fh3_flags = FH_WEBNFS;
2437 fh->fh3_length = (cnt+1) * sizeof (int32_t);
2438
2439 c = (char *)&fh->fh3_u.nfs_fh3_i.fh3_i;
2440 /*
2441 * Encode the status octet that indicates whether there
2442 * are more security flavors the client needs to get.
2443 */
2444 *c = totalcnt > fh_max_cnt;
2445
2446 /*
2447 * put security flavors in the overloaded fh
2448 */
2449 secidx = sec_index - 1;
2450 ipt = (int *)(c + sizeof (int32_t));
2451 for (i = 0; i < cnt; i++) {
2452 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2453 }
2454 return (0);
2455 }
2456
2457 /*
2458 * Make an nfs_fh4 from a vnode
2459 */
2460 int
2461 makefh4(nfs_fh4 *fh, vnode_t *vp, struct exportinfo *exi)
2462 {
2463 int error;
2464 nfs_fh4_fmt_t *fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2465 fid_t fid;
2466
2467 bzero(&fid, sizeof (fid));
2468 fid.fid_len = MAXFIDSZ;
2469 /*
2470 * vop_fid_pseudo() is used to set up NFSv4 namespace, so
2471 * use vop_fid_pseudo() here to get the fid instead of VOP_FID.
2472 */
2473 error = vop_fid_pseudo(vp, &fid);
2474 if (error)
2475 return (error);
2476
2477 fh->nfs_fh4_len = NFS_FH4_LEN;
2478
2479 fh_fmtp->fh4_i.fhx_fsid = exi->exi_fh.fh_fsid;
2480 fh_fmtp->fh4_i.fhx_xlen = exi->exi_fh.fh_xlen;
2481
2482 bzero(fh_fmtp->fh4_i.fhx_data, sizeof (fh_fmtp->fh4_i.fhx_data));
2483 bzero(fh_fmtp->fh4_i.fhx_xdata, sizeof (fh_fmtp->fh4_i.fhx_xdata));
2484 ASSERT(exi->exi_fh.fh_xlen <= sizeof (fh_fmtp->fh4_i.fhx_xdata));
2485 bcopy(exi->exi_fh.fh_xdata, fh_fmtp->fh4_i.fhx_xdata,
2486 exi->exi_fh.fh_xlen);
2487
2488 fh_fmtp->fh4_len = fid.fid_len;
2489 ASSERT(fid.fid_len <= sizeof (fh_fmtp->fh4_data));
2490 bcopy(fid.fid_data, fh_fmtp->fh4_data, fid.fid_len);
2491 fh_fmtp->fh4_flag = 0;
2492
2493 #ifdef VOLATILE_FH_TEST
2494 /*
2495 * XXX (temporary?)
2496 * Use the rnode volatile_id value to add volatility to the fh.
2497 *
2498 * For testing purposes there are currently two scenarios, based
2499 * on whether the filesystem was shared with "volatile_fh"
2500 * or "expire_on_rename". In the first case, use the value of
2501 * export struct share_time as the volatile_id. In the second
2502 * case use the vnode volatile_id value (which is set to the
2503 * time in which the file was renamed).
2504 *
2505 * Note that the above are temporary constructs for testing only
2506 * XXX
2507 */
2508 if (exi->exi_export.ex_flags & EX_VOLRNM) {
2509 fh_fmtp->fh4_volatile_id = find_volrnm_fh_id(exi, fh);
2510 } else if (exi->exi_export.ex_flags & EX_VOLFH) {
2511 fh_fmtp->fh4_volatile_id = exi->exi_volatile_id;
2512 } else {
2513 fh_fmtp->fh4_volatile_id = 0;
2514 }
2515 #endif /* VOLATILE_FH_TEST */
2516
2517 return (0);
2518 }
2519
2520 /*
2521 * Convert an fhandle into a vnode.
2522 * Uses the file id (fh_len + fh_data) in the fhandle to get the vnode.
2523 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2524 * are done with it.
2525 */
2526 vnode_t *
2527 nfs_fhtovp(fhandle_t *fh, struct exportinfo *exi)
2528 {
2529 vfs_t *vfsp;
2530 vnode_t *vp;
2531 int error;
2532 fid_t *fidp;
2533
2534 TRACE_0(TR_FAC_NFS, TR_FHTOVP_START,
2535 "fhtovp_start");
2536
2537 if (exi == NULL) {
2538 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2539 "fhtovp_end:(%S)", "exi NULL");
2540 return (NULL); /* not exported */
2541 }
2542
2543 ASSERT(exi->exi_vp != NULL);
2544
2545 if (PUBLIC_FH2(fh)) {
2546 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2547 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2548 "fhtovp_end:(%S)", "root not exported");
2549 return (NULL);
2550 }
2551 vp = exi->exi_vp;
2552 VN_HOLD(vp);
2553 return (vp);
2554 }
2555
2556 vfsp = exi->exi_vp->v_vfsp;
2557 ASSERT(vfsp != NULL);
2558 fidp = (fid_t *)&fh->fh_len;
2559
2560 error = VFS_VGET(vfsp, &vp, fidp);
2561 if (error || vp == NULL) {
2562 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2563 "fhtovp_end:(%S)", "VFS_GET failed or vp NULL");
2564 return (NULL);
2565 }
2566 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2567 "fhtovp_end:(%S)", "end");
2568 return (vp);
2569 }
2570
2571 /*
2572 * Convert an nfs_fh3 into a vnode.
2573 * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2574 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2575 * are done with it.
2576 */
2577 vnode_t *
2578 nfs3_fhtovp(nfs_fh3 *fh, struct exportinfo *exi)
2579 {
2580 vfs_t *vfsp;
2581 vnode_t *vp;
2582 int error;
2583 fid_t *fidp;
2584
2585 if (exi == NULL)
2586 return (NULL); /* not exported */
2587
2588 ASSERT(exi->exi_vp != NULL);
2589
2590 if (PUBLIC_FH3(fh)) {
2591 if (exi->exi_export.ex_flags & EX_PUBLIC)
2592 return (NULL);
2593 vp = exi->exi_vp;
2594 VN_HOLD(vp);
2595 return (vp);
2596 }
2597
2598 if (fh->fh3_length < NFS3_OLDFHSIZE ||
2599 fh->fh3_length > NFS3_MAXFHSIZE)
2600 return (NULL);
2601
2602 vfsp = exi->exi_vp->v_vfsp;
2603 ASSERT(vfsp != NULL);
2604 fidp = FH3TOFIDP(fh);
2605
2606 error = VFS_VGET(vfsp, &vp, fidp);
2607 if (error || vp == NULL)
2608 return (NULL);
2609
2610 return (vp);
2611 }
2612
2613 /*
2614 * Convert an nfs_fh4 into a vnode.
2615 * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2616 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2617 * are done with it.
2618 */
2619 vnode_t *
2620 nfs4_fhtovp(nfs_fh4 *fh, struct exportinfo *exi, nfsstat4 *statp)
2621 {
2622 vfs_t *vfsp;
2623 vnode_t *vp = NULL;
2624 int error;
2625 fid_t *fidp;
2626 nfs_fh4_fmt_t *fh_fmtp;
2627 #ifdef VOLATILE_FH_TEST
2628 uint32_t volatile_id = 0;
2629 #endif /* VOLATILE_FH_TEST */
2630
2631 if (exi == NULL) {
2632 *statp = NFS4ERR_STALE;
2633 return (NULL); /* not exported */
2634 }
2635 ASSERT(exi->exi_vp != NULL);
2636
2637 /* caller should have checked this */
2638 ASSERT(fh->nfs_fh4_len >= NFS_FH4_LEN);
2639
2640 fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2641 vfsp = exi->exi_vp->v_vfsp;
2642 ASSERT(vfsp != NULL);
2643 fidp = (fid_t *)&fh_fmtp->fh4_len;
2644
2645 #ifdef VOLATILE_FH_TEST
2646 /* XXX check if volatile - should be changed later */
2647 if (exi->exi_export.ex_flags & (EX_VOLRNM | EX_VOLFH)) {
2648 /*
2649 * Filesystem is shared with volatile filehandles
2650 */
2651 if (exi->exi_export.ex_flags & EX_VOLRNM)
2652 volatile_id = find_volrnm_fh_id(exi, fh);
2653 else
2654 volatile_id = exi->exi_volatile_id;
2655
2656 if (fh_fmtp->fh4_volatile_id != volatile_id) {
2657 *statp = NFS4ERR_FHEXPIRED;
2658 return (NULL);
2659 }
2660 }
2661 /*
2662 * XXX even if test_volatile_fh false, the fh may contain a
2663 * volatile id if obtained when the test was set.
2664 */
2665 fh_fmtp->fh4_volatile_id = (uchar_t)0;
2666 #endif /* VOLATILE_FH_TEST */
2667
2668 error = VFS_VGET(vfsp, &vp, fidp);
2669 /*
2670 * If we can not get vp from VFS_VGET, perhaps this is
2671 * an nfs v2/v3/v4 node in an nfsv4 pseudo filesystem.
2672 * Check it out.
2673 */
2674 if (error && PSEUDO(exi))
2675 error = nfs4_vget_pseudo(exi, &vp, fidp);
2676
2677 if (error || vp == NULL) {
2678 *statp = NFS4ERR_STALE;
2679 return (NULL);
2680 }
2681 /* XXX - disgusting hack */
2682 if (vp->v_type == VNON && vp->v_flag & V_XATTRDIR)
2683 vp->v_type = VDIR;
2684 *statp = NFS4_OK;
2685 return (vp);
2686 }
2687
2688 /*
2689 * Find the export structure associated with the given filesystem.
2690 * If found, then increment the ref count (exi_count).
2691 */
2692 struct exportinfo *
2693 checkexport(fsid_t *fsid, fid_t *fid)
2694 {
2695 struct exportinfo *exi;
2696 nfs_export_t *ne = nfs_get_export();
2697
2698 rw_enter(&ne->exported_lock, RW_READER);
2699 for (exi = ne->exptable[exptablehash(fsid, fid)];
2700 exi != NULL;
2701 exi = exi->fid_hash.next) {
2702 if (exportmatch(exi, fsid, fid)) {
2703 /*
2704 * If this is the place holder for the
2705 * public file handle, then return the
2706 * real export entry for the public file
2707 * handle.
2708 */
2709 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2710 exi = ne->exi_public;
2711 }
2712
2713 exi_hold(exi);
2714 rw_exit(&ne->exported_lock);
2715 return (exi);
2716 }
2717 }
2718 rw_exit(&ne->exported_lock);
2719 return (NULL);
2720 }
2721
2722
2723 /*
2724 * "old school" version of checkexport() for NFS4. NFS4
2725 * rfs4_compound holds exported_lock for duration of compound
2726 * processing. This version doesn't manipulate exi_count
2727 * since NFS4 breaks fundamental assumptions in the exi_count
2728 * design.
2729 */
2730 struct exportinfo *
2731 checkexport4(fsid_t *fsid, fid_t *fid, vnode_t *vp)
2732 {
2733 struct exportinfo *exi;
2734 nfs_export_t *ne = nfs_get_export();
2735
2736 ASSERT(RW_LOCK_HELD(&ne->exported_lock));
2737
2738 for (exi = ne->exptable[exptablehash(fsid, fid)];
2739 exi != NULL;
2740 exi = exi->fid_hash.next) {
2741 if (exportmatch(exi, fsid, fid)) {
2742 /*
2743 * If this is the place holder for the
2744 * public file handle, then return the
2745 * real export entry for the public file
2746 * handle.
2747 */
2748 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2749 exi = ne->exi_public;
2750 }
2751
2752 /*
2753 * If vp is given, check if vp is the
2754 * same vnode as the exported node.
2755 *
2756 * Since VOP_FID of a lofs node returns the
2757 * fid of its real node (ufs), the exported
2758 * node for lofs and (pseudo) ufs may have
2759 * the same fsid and fid.
2760 */
2761 if (vp == NULL || vp == exi->exi_vp)
2762 return (exi);
2763 }
2764 }
2765
2766 return (NULL);
2767 }
2768
2769 /*
2770 * Free an entire export list node
2771 */
2772 void
2773 exportfree(struct exportinfo *exi)
2774 {
2775 struct exportdata *ex;
2776 struct charset_cache *cache;
2777 int i;
2778
2779 ex = &exi->exi_export;
2780
2781 ASSERT(exi->exi_vp != NULL && !(exi->exi_export.ex_flags & EX_PUBLIC));
2782 VN_RELE(exi->exi_vp);
2783 if (exi->exi_dvp != NULL)
2784 VN_RELE(exi->exi_dvp);
2785
2786 if (ex->ex_flags & EX_INDEX)
2787 kmem_free(ex->ex_index, strlen(ex->ex_index) + 1);
2788
2789 kmem_free(ex->ex_path, ex->ex_pathlen + 1);
2790 nfsauth_cache_free(exi);
2791
2792 /*
2793 * if there is a character set mapping cached, clean it up.
2794 */
2795 for (cache = exi->exi_charset; cache != NULL;
2796 cache = exi->exi_charset) {
2797 if (cache->inbound != (kiconv_t)-1)
2798 (void) kiconv_close(cache->inbound);
2799 if (cache->outbound != (kiconv_t)-1)
2800 (void) kiconv_close(cache->outbound);
2801 exi->exi_charset = cache->next;
2802 kmem_free(cache, sizeof (struct charset_cache));
2803 }
2804
2805 if (exi->exi_logbuffer != NULL)
2806 nfslog_disable(exi);
2807
2808 if (ex->ex_flags & EX_LOG) {
2809 kmem_free(ex->ex_log_buffer, ex->ex_log_bufferlen + 1);
2810 kmem_free(ex->ex_tag, ex->ex_taglen + 1);
2811 }
2812
2813 if (exi->exi_visible)
2814 free_visible(exi->exi_visible);
2815
2816 srv_secinfo_list_free(ex->ex_secinfo, ex->ex_seccnt);
2817
2818 #ifdef VOLATILE_FH_TEST
2819 free_volrnm_list(exi);
2820 mutex_destroy(&exi->exi_vol_rename_lock);
2821 #endif /* VOLATILE_FH_TEST */
2822
2823 mutex_destroy(&exi->exi_lock);
2824 rw_destroy(&exi->exi_cache_lock);
2825 /*
2826 * All nodes in the exi_cache AVL trees were removed and freed in the
2827 * nfsauth_cache_free() call above. We will just destroy and free the
2828 * empty AVL trees here.
2829 */
2830 for (i = 0; i < AUTH_TABLESIZE; i++) {
2831 avl_destroy(exi->exi_cache[i]);
2832 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
2833 }
2834
2835 kmem_free(exi, sizeof (*exi));
2836 }
2837
2838 /*
2839 * load the index file from user space into kernel space.
2840 */
2841 static int
2842 loadindex(struct exportdata *kex)
2843 {
2844 int error;
2845 char index[MAXNAMELEN+1];
2846 size_t len;
2847
2848 /*
2849 * copyinstr copies the complete string including the NULL and
2850 * returns the len with the NULL byte included in the calculation
2851 * as long as the max length is not exceeded.
2852 */
2853 if (error = copyinstr(kex->ex_index, index, sizeof (index), &len))
2854 return (error);
2855
2856 kex->ex_index = kmem_alloc(len, KM_SLEEP);
2857 bcopy(index, kex->ex_index, len);
2858
2859 return (0);
2860 }
2861
2862 void
2863 exi_hold(struct exportinfo *exi)
2864 {
2865 mutex_enter(&exi->exi_lock);
2866 exi->exi_count++;
2867 mutex_exit(&exi->exi_lock);
2868 }
2869
2870 /*
2871 * When a thread completes using exi, it should call exi_rele().
2872 * exi_rele() decrements exi_count. It releases exi if exi_count == 0, i.e.
2873 * if this is the last user of exi and exi is not on exportinfo list anymore
2874 */
2875 void
2876 exi_rele(struct exportinfo *exi)
2877 {
2878 mutex_enter(&exi->exi_lock);
2879 exi->exi_count--;
2880 if (exi->exi_count == 0) {
2881 mutex_exit(&exi->exi_lock);
2882 exportfree(exi);
2883 } else
2884 mutex_exit(&exi->exi_lock);
2885 }
2886
2887 #ifdef VOLATILE_FH_TEST
2888 /*
2889 * Test for volatile fh's - add file handle to list and set its volatile id
2890 * to time it was renamed. If EX_VOLFH is also on and the fs is reshared,
2891 * the vol_rename queue is purged.
2892 *
2893 * XXX This code is for unit testing purposes only... To correctly use it, it
2894 * needs to tie a rename list to the export struct and (more
2895 * important), protect access to the exi rename list using a write lock.
2896 */
2897
2898 /*
2899 * get the fh vol record if it's in the volatile on rename list. Don't check
2900 * volatile_id in the file handle - compare only the file handles.
2901 */
2902 static struct ex_vol_rename *
2903 find_volrnm_fh(struct exportinfo *exi, nfs_fh4 *fh4p)
2904 {
2905 struct ex_vol_rename *p = NULL;
2906 fhandle4_t *fhp;
2907
2908 /* XXX shouldn't we assert &exported_lock held? */
2909 ASSERT(MUTEX_HELD(&exi->exi_vol_rename_lock));
2910
2911 if (fh4p->nfs_fh4_len != NFS_FH4_LEN) {
2912 return (NULL);
2913 }
2914 fhp = &((nfs_fh4_fmt_t *)fh4p->nfs_fh4_val)->fh4_i;
2915 for (p = exi->exi_vol_rename; p != NULL; p = p->vrn_next) {
2916 if (bcmp(fhp, &p->vrn_fh_fmt.fh4_i,
2917 sizeof (fhandle4_t)) == 0)
2918 break;
2919 }
2920 return (p);
2921 }
2922
2923 /*
2924 * get the volatile id for the fh (if there is - else return 0). Ignore the
2925 * volatile_id in the file handle - compare only the file handles.
2926 */
2927 static uint32_t
2928 find_volrnm_fh_id(struct exportinfo *exi, nfs_fh4 *fh4p)
2929 {
2930 struct ex_vol_rename *p;
2931 uint32_t volatile_id;
2932
2933 mutex_enter(&exi->exi_vol_rename_lock);
2934 p = find_volrnm_fh(exi, fh4p);
2935 volatile_id = (p ? p->vrn_fh_fmt.fh4_volatile_id :
2936 exi->exi_volatile_id);
2937 mutex_exit(&exi->exi_vol_rename_lock);
2938 return (volatile_id);
2939 }
2940
2941 /*
2942 * Free the volatile on rename list - will be called if a filesystem is
2943 * unshared or reshared without EX_VOLRNM
2944 */
2945 static void
2946 free_volrnm_list(struct exportinfo *exi)
2947 {
2948 struct ex_vol_rename *p, *pnext;
2949
2950 /* no need to hold mutex lock - this one is called from exportfree */
2951 for (p = exi->exi_vol_rename; p != NULL; p = pnext) {
2952 pnext = p->vrn_next;
2953 kmem_free(p, sizeof (*p));
2954 }
2955 exi->exi_vol_rename = NULL;
2956 }
2957
2958 /*
2959 * Add a file handle to the volatile on rename list.
2960 */
2961 void
2962 add_volrnm_fh(struct exportinfo *exi, vnode_t *vp)
2963 {
2964 struct ex_vol_rename *p;
2965 char fhbuf[NFS4_FHSIZE];
2966 nfs_fh4 fh4;
2967 int error;
2968
2969 fh4.nfs_fh4_val = fhbuf;
2970 error = makefh4(&fh4, vp, exi);
2971 if ((error) || (fh4.nfs_fh4_len != sizeof (p->vrn_fh_fmt))) {
2972 return;
2973 }
2974
2975 mutex_enter(&exi->exi_vol_rename_lock);
2976
2977 p = find_volrnm_fh(exi, &fh4);
2978
2979 if (p == NULL) {
2980 p = kmem_alloc(sizeof (*p), KM_SLEEP);
2981 bcopy(fh4.nfs_fh4_val, &p->vrn_fh_fmt, sizeof (p->vrn_fh_fmt));
2982 p->vrn_next = exi->exi_vol_rename;
2983 exi->exi_vol_rename = p;
2984 }
2985
2986 p->vrn_fh_fmt.fh4_volatile_id = gethrestime_sec();
2987 mutex_exit(&exi->exi_vol_rename_lock);
2988 }
2989
2990 #endif /* VOLATILE_FH_TEST */