1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25
26 /*
27 * Copyright 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T.
28 * All rights reserved.
29 */
30
31 /*
32 * Copyright 2018 Nexenta Systems, Inc.
33 */
34
35 #include <sys/types.h>
36 #include <sys/param.h>
37 #include <sys/time.h>
38 #include <sys/vfs.h>
39 #include <sys/vnode.h>
40 #include <sys/socket.h>
41 #include <sys/errno.h>
42 #include <sys/uio.h>
43 #include <sys/proc.h>
44 #include <sys/user.h>
45 #include <sys/file.h>
46 #include <sys/tiuser.h>
47 #include <sys/kmem.h>
48 #include <sys/pathname.h>
49 #include <sys/debug.h>
50 #include <sys/vtrace.h>
51 #include <sys/cmn_err.h>
52 #include <sys/acl.h>
53 #include <sys/utsname.h>
54 #include <sys/sdt.h>
55 #include <netinet/in.h>
56 #include <sys/avl.h>
57
58 #include <rpc/types.h>
59 #include <rpc/auth.h>
60 #include <rpc/svc.h>
61
62 #include <nfs/nfs.h>
63 #include <nfs/export.h>
64 #include <nfs/nfssys.h>
65 #include <nfs/nfs_clnt.h>
66 #include <nfs/nfs_acl.h>
67 #include <nfs/nfs_log.h>
68 #include <nfs/lm.h>
69 #include <sys/sunddi.h>
70
71 /*
72 * exi_id support
73 *
74 * exi_id_next The next exi_id available.
75 * exi_id_overflow The exi_id_next already overflowed, so we should
76 * thoroughly check for duplicates.
77 * exi_id_tree AVL tree indexed by exi_id.
78 * nfs_exi_id_lock Lock to protect the export ID list
79 *
80 * All exi_id_next, exi_id_overflow, and exi_id_tree are protected by
81 * nfs_exi_id_lock.
82 */
83 static int exi_id_next;
84 static bool_t exi_id_overflow;
85 avl_tree_t exi_id_tree;
86 kmutex_t nfs_exi_id_lock;
87
88 static int unexport(nfs_export_t *, exportinfo_t *);
89 static void exportfree(exportinfo_t *);
90 static int loadindex(exportdata_t *);
91
92 extern void nfsauth_cache_free(exportinfo_t *);
93 extern int sec_svc_loadrootnames(int, int, caddr_t **, model_t);
94 extern void sec_svc_freerootnames(int, int, caddr_t *);
95
96 static int build_seclist_nodups(exportdata_t *, secinfo_t *, int);
97 static void srv_secinfo_add(secinfo_t **, int *, secinfo_t *, int, int);
98 static void srv_secinfo_remove(secinfo_t **, int *, secinfo_t *, int);
99 static void srv_secinfo_treeclimb(nfs_export_t *, exportinfo_t *,
100 secinfo_t *, int, bool_t);
101
102 #ifdef VOLATILE_FH_TEST
103 static struct ex_vol_rename *find_volrnm_fh(exportinfo_t *, nfs_fh4 *);
104 static uint32_t find_volrnm_fh_id(exportinfo_t *, nfs_fh4 *);
105 static void free_volrnm_list(exportinfo_t *);
106 #endif /* VOLATILE_FH_TEST */
107
108 fhandle_t nullfh2; /* for comparing V2 filehandles */
109
110 /*
111 * macro for static dtrace probes to trace server namespace ref count mods.
112 */
113 #define SECREF_TRACE(seclist, tag, flav, aftcnt) \
114 DTRACE_PROBE4(nfss__i__nmspc__secref, struct secinfo *, (seclist), \
115 char *, (tag), int, (int)(flav), int, (int)(aftcnt))
116
117
118 #define exptablehash(fsid, fid) (nfs_fhhash((fsid), (fid)) & (EXPTABLESIZE - 1))
119
120 extern nfs_export_t *
121 nfs_get_export(void)
122 {
123 nfs_globals_t *ng = zone_getspecific(nfssrv_zone_key, curzone);
124 nfs_export_t *ne = ng->nfs_export;
125 ASSERT(ne != NULL);
126 return (ne);
127 }
128
129 static uint8_t
130 xor_hash(uint8_t *data, int len)
131 {
132 uint8_t h = 0;
133
134 while (len--)
135 h ^= *data++;
136
137 return (h);
138 }
139
140 /*
141 * File handle hash function, XOR over all bytes in fsid and fid.
142 */
143 static unsigned
144 nfs_fhhash(fsid_t *fsid, fid_t *fid)
145 {
146 int len;
147 uint8_t h;
148
149 h = xor_hash((uint8_t *)fsid, sizeof (fsid_t));
150
151 /*
152 * Sanity check the length before using it
153 * blindly in case the client trashed it.
154 */
155 len = fid->fid_len > NFS_FH4MAXDATA ? 0 : fid->fid_len;
156 h ^= xor_hash((uint8_t *)fid->fid_data, len);
157
158 return ((unsigned)h);
159 }
160
161 /*
162 * Free the memory allocated within a secinfo entry.
163 */
164 void
165 srv_secinfo_entry_free(struct secinfo *secp)
166 {
167 if (secp->s_rootcnt > 0 && secp->s_rootnames != NULL) {
168 sec_svc_freerootnames(secp->s_secinfo.sc_rpcnum,
169 secp->s_rootcnt, secp->s_rootnames);
170 secp->s_rootcnt = 0;
171 }
172
173 if ((secp->s_secinfo.sc_rpcnum == RPCSEC_GSS) &&
174 (secp->s_secinfo.sc_gss_mech_type)) {
175 kmem_free(secp->s_secinfo.sc_gss_mech_type->elements,
176 secp->s_secinfo.sc_gss_mech_type->length);
177 kmem_free(secp->s_secinfo.sc_gss_mech_type,
178 sizeof (rpc_gss_OID_desc));
179 secp->s_secinfo.sc_gss_mech_type = NULL;
180 }
181 }
182
183 /*
184 * Free a list of secinfo allocated in the exportdata structure.
185 */
186 void
187 srv_secinfo_list_free(struct secinfo *secinfo, int cnt)
188 {
189 int i;
190
191 if (cnt == 0)
192 return;
193
194 for (i = 0; i < cnt; i++)
195 srv_secinfo_entry_free(&secinfo[i]);
196
197 kmem_free(secinfo, cnt * sizeof (struct secinfo));
198 }
199
200 /*
201 * Allocate and copy a secinfo data from "from" to "to".
202 *
203 * This routine is used by srv_secinfo_add() to add a new flavor to an
204 * ancestor's export node. The rootnames are not copied because the
205 * allowable rootname access only applies to the explicit exported node,
206 * not its ancestor's.
207 *
208 * "to" should have already been allocated and zeroed before calling
209 * this routine.
210 *
211 * This routine is used under the protection of exported_lock (RW_WRITER).
212 */
213 void
214 srv_secinfo_copy(struct secinfo *from, struct secinfo *to)
215 {
216 to->s_secinfo.sc_nfsnum = from->s_secinfo.sc_nfsnum;
217 to->s_secinfo.sc_rpcnum = from->s_secinfo.sc_rpcnum;
218
219 if (from->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
220 to->s_secinfo.sc_service = from->s_secinfo.sc_service;
221 bcopy(from->s_secinfo.sc_name, to->s_secinfo.sc_name,
222 strlen(from->s_secinfo.sc_name));
223 bcopy(from->s_secinfo.sc_gss_mech, to->s_secinfo.sc_gss_mech,
224 strlen(from->s_secinfo.sc_gss_mech));
225
226 /* copy mechanism oid */
227 to->s_secinfo.sc_gss_mech_type =
228 kmem_alloc(sizeof (rpc_gss_OID_desc), KM_SLEEP);
229 to->s_secinfo.sc_gss_mech_type->length =
230 from->s_secinfo.sc_gss_mech_type->length;
231 to->s_secinfo.sc_gss_mech_type->elements =
232 kmem_alloc(from->s_secinfo.sc_gss_mech_type->length,
233 KM_SLEEP);
234 bcopy(from->s_secinfo.sc_gss_mech_type->elements,
235 to->s_secinfo.sc_gss_mech_type->elements,
236 from->s_secinfo.sc_gss_mech_type->length);
237 }
238
239 to->s_refcnt = from->s_refcnt;
240 to->s_window = from->s_window;
241 /* no need to copy the mode bits - s_flags */
242 }
243
244 /*
245 * Create a secinfo array without duplicates. The condensed
246 * flavor list is used to propagate flavor ref counts to an
247 * export's ancestor pseudonodes.
248 */
249 static int
250 build_seclist_nodups(exportdata_t *exd, secinfo_t *nodups, int exponly)
251 {
252 int ccnt, c;
253 int ncnt, n;
254 struct secinfo *cursec;
255
256 ncnt = 0;
257 ccnt = exd->ex_seccnt;
258 cursec = exd->ex_secinfo;
259
260 for (c = 0; c < ccnt; c++) {
261
262 if (exponly && ! SEC_REF_EXPORTED(&cursec[c]))
263 continue;
264
265 for (n = 0; n < ncnt; n++) {
266 if (nodups[n].s_secinfo.sc_nfsnum ==
267 cursec[c].s_secinfo.sc_nfsnum)
268 break;
269 }
270
271 /*
272 * The structure copy below also copys ptrs embedded
273 * within struct secinfo. The ptrs are copied but
274 * they are never freed from the nodups array. If
275 * an ancestor's secinfo array doesn't contain one
276 * of the nodups flavors, then the entry is properly
277 * copied into the ancestor's secinfo array.
278 * (see srv_secinfo_copy)
279 */
280 if (n == ncnt) {
281 nodups[n] = cursec[c];
282 ncnt++;
283 }
284 }
285 return (ncnt);
286 }
287
288 /*
289 * Add the new security flavors from newdata to the current list, pcursec.
290 * Upon return, *pcursec has the newly merged secinfo list.
291 *
292 * There should be at least 1 secinfo entry in newsec.
293 *
294 * This routine is used under the protection of exported_lock (RW_WRITER).
295 */
296 static void
297 srv_secinfo_add(secinfo_t **pcursec, int *pcurcnt, secinfo_t *newsec,
298 int newcnt, int is_pseudo)
299 {
300 int ccnt, c; /* sec count in current data - curdata */
301 int n; /* index for newsec - newsecinfo */
302 int tcnt; /* total sec count after merge */
303 int mcnt; /* total sec count after merge */
304 struct secinfo *msec; /* merged secinfo list */
305 struct secinfo *cursec;
306
307 cursec = *pcursec;
308 ccnt = *pcurcnt;
309
310 ASSERT(newcnt > 0);
311 tcnt = ccnt + newcnt;
312
313 for (n = 0; n < newcnt; n++) {
314 for (c = 0; c < ccnt; c++) {
315 if (newsec[n].s_secinfo.sc_nfsnum ==
316 cursec[c].s_secinfo.sc_nfsnum) {
317 cursec[c].s_refcnt += newsec[n].s_refcnt;
318 SECREF_TRACE(cursec, "add_ref",
319 cursec[c].s_secinfo.sc_nfsnum,
320 cursec[c].s_refcnt);
321 tcnt--;
322 break;
323 }
324 }
325 }
326
327 if (tcnt == ccnt)
328 return; /* no change; no new flavors */
329
330 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
331
332 /* move current secinfo list data to the new list */
333 for (c = 0; c < ccnt; c++)
334 msec[c] = cursec[c];
335
336 /* Add the flavor that's not in the current data */
337 mcnt = ccnt;
338 for (n = 0; n < newcnt; n++) {
339 for (c = 0; c < ccnt; c++) {
340 if (newsec[n].s_secinfo.sc_nfsnum ==
341 cursec[c].s_secinfo.sc_nfsnum)
342 break;
343 }
344
345 /* This is the one. Add it. */
346 if (c == ccnt) {
347 srv_secinfo_copy(&newsec[n], &msec[mcnt]);
348
349 if (is_pseudo)
350 msec[mcnt].s_flags = M_RO;
351
352 SECREF_TRACE(msec, "new_ref",
353 msec[mcnt].s_secinfo.sc_nfsnum,
354 msec[mcnt].s_refcnt);
355 mcnt++;
356 }
357 }
358
359 ASSERT(mcnt == tcnt);
360
361 /*
362 * Done. Update curdata. Free the old secinfo list in
363 * curdata and return the new sec array info
364 */
365 if (ccnt > 0)
366 kmem_free(cursec, ccnt * sizeof (struct secinfo));
367 *pcurcnt = tcnt;
368 *pcursec = msec;
369 }
370
371 /*
372 * For NFS V4.
373 * Remove the security data of the unexported node from its ancestors.
374 * Assume there is at least one flavor entry in the current sec list
375 * (pcursec).
376 *
377 * This routine is used under the protection of exported_lock (RW_WRITER).
378 *
379 * Every element of remsec is an explicitly exported flavor. If
380 * srv_secinfo_remove() is called fom an exportfs error path, then
381 * the flavor list was derived from the user's share cmdline,
382 * and all flavors are explicit. If it was called from the unshare path,
383 * build_seclist_nodups() was called with the exponly flag.
384 */
385 static void
386 srv_secinfo_remove(secinfo_t **pcursec, int *pcurcnt, secinfo_t *remsec,
387 int remcnt)
388 {
389 int ccnt, c; /* sec count in current data - cursec */
390 int r; /* sec count in removal data - remsec */
391 int tcnt, mcnt; /* total sec count after removing */
392 struct secinfo *msec; /* final secinfo list after removing */
393 struct secinfo *cursec;
394
395 cursec = *pcursec;
396 ccnt = *pcurcnt;
397 tcnt = ccnt;
398
399 for (r = 0; r < remcnt; r++) {
400 /*
401 * At unshare/reshare time, only explicitly shared flavor ref
402 * counts are decremented and propagated to ancestors.
403 * Implicit flavor refs came from shared descendants, and
404 * they must be kept.
405 */
406 if (! SEC_REF_EXPORTED(&remsec[r]))
407 continue;
408
409 for (c = 0; c < ccnt; c++) {
410 if (remsec[r].s_secinfo.sc_nfsnum ==
411 cursec[c].s_secinfo.sc_nfsnum) {
412
413 /*
414 * Decrement secinfo reference count by 1.
415 * If this entry is invalid after decrementing
416 * the count (i.e. count < 1), this entry will
417 * be removed.
418 */
419 cursec[c].s_refcnt--;
420
421 SECREF_TRACE(cursec, "del_ref",
422 cursec[c].s_secinfo.sc_nfsnum,
423 cursec[c].s_refcnt);
424
425 ASSERT(cursec[c].s_refcnt >= 0);
426
427 if (SEC_REF_INVALID(&cursec[c]))
428 tcnt--;
429 break;
430 }
431 }
432 }
433
434 ASSERT(tcnt >= 0);
435 if (tcnt == ccnt)
436 return; /* no change; no flavors to remove */
437
438 if (tcnt == 0) {
439 srv_secinfo_list_free(cursec, ccnt);
440 *pcurcnt = 0;
441 *pcursec = NULL;
442 return;
443 }
444
445 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
446
447 /* walk thru the given secinfo list to remove the flavors */
448 mcnt = 0;
449 for (c = 0; c < ccnt; c++) {
450 if (SEC_REF_INVALID(&cursec[c])) {
451 srv_secinfo_entry_free(&cursec[c]);
452 } else {
453 msec[mcnt] = cursec[c];
454 mcnt++;
455 }
456 }
457
458 ASSERT(mcnt == tcnt);
459 /*
460 * Done. Update curdata.
461 * Free the existing secinfo list in curdata. All pointers
462 * within the list have either been moved to msec or freed
463 * if it's invalid.
464 */
465 kmem_free(*pcursec, ccnt * sizeof (struct secinfo));
466 *pcursec = msec;
467 *pcurcnt = tcnt;
468 }
469
470
471 /*
472 * For the reshare case, sec flavor accounting happens in 3 steps:
473 * 1) propagate addition of new flavor refs up the ancestor tree
474 * 2) transfer flavor refs of descendants to new/reshared exportdata
475 * 3) propagate removal of old flavor refs up the ancestor tree
476 *
477 * srv_secinfo_exp2exp() implements step 2 of a reshare. At this point,
478 * the new flavor list has already been propagated up through the
479 * ancestor tree via srv_secinfo_treeclimb().
480 *
481 * If there is more than 1 export reference to an old flavor (i.e. some
482 * of its children shared with this flavor), this flavor information
483 * needs to be transferred to the new exportdata struct. A flavor in
484 * the old exportdata has descendant refs when its s_refcnt > 1 or it
485 * is implicitly shared (M_SEC4_EXPORTED not set in s_flags).
486 *
487 * SEC_REF_EXPORTED() is only true when M_SEC4_EXPORTED is set
488 * SEC_REF_SELF() is only true when both M_SEC4_EXPORTED is set and s_refcnt==1
489 *
490 * Transferring descendant flavor refcnts happens in 2 passes:
491 * a) flavors used before (oldsecinfo) and after (curdata->ex_secinfo) reshare
492 * b) flavors used before but not after reshare
493 *
494 * This routine is used under the protection of exported_lock (RW_WRITER).
495 */
496 void
497 srv_secinfo_exp2exp(exportdata_t *curdata, secinfo_t *oldsecinfo, int ocnt)
498 {
499 int ccnt, c; /* sec count in current data - curdata */
500 int o; /* sec count in old data - oldsecinfo */
501 int tcnt, mcnt; /* total sec count after the transfer */
502 struct secinfo *msec; /* merged secinfo list */
503
504 ccnt = curdata->ex_seccnt;
505
506 ASSERT(ocnt > 0);
507 ASSERT(!(curdata->ex_flags & EX_PSEUDO));
508
509 /*
510 * If the oldsecinfo has flavors with more than 1 reference count
511 * and the flavor is specified in the reshare, transfer the flavor
512 * refs to the new seclist (curdata.ex_secinfo).
513 */
514 tcnt = ccnt + ocnt;
515
516 for (o = 0; o < ocnt; o++) {
517
518 if (SEC_REF_SELF(&oldsecinfo[o])) {
519 tcnt--;
520 continue;
521 }
522
523 for (c = 0; c < ccnt; c++) {
524 if (oldsecinfo[o].s_secinfo.sc_nfsnum ==
525 curdata->ex_secinfo[c].s_secinfo.sc_nfsnum) {
526
527 /*
528 * add old reference to the current
529 * secinfo count
530 */
531 curdata->ex_secinfo[c].s_refcnt +=
532 oldsecinfo[o].s_refcnt;
533
534 /*
535 * Delete the old export flavor
536 * reference. The initial reference
537 * was created during srv_secinfo_add,
538 * and the count is decremented below
539 * to account for the initial reference.
540 */
541 if (SEC_REF_EXPORTED(&oldsecinfo[o]))
542 curdata->ex_secinfo[c].s_refcnt--;
543
544 SECREF_TRACE(curdata->ex_path,
545 "reshare_xfer_common_child_refs",
546 curdata->ex_secinfo[c].s_secinfo.sc_nfsnum,
547 curdata->ex_secinfo[c].s_refcnt);
548
549 ASSERT(curdata->ex_secinfo[c].s_refcnt >= 0);
550
551 tcnt--;
552 break;
553 }
554 }
555 }
556
557 if (tcnt == ccnt)
558 return; /* no more transfer to do */
559
560 /*
561 * oldsecinfo has flavors referenced by its children that are not
562 * in the current (new) export flavor list. Add these flavors.
563 */
564 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
565
566 /* move current secinfo list data to the new list */
567 for (c = 0; c < ccnt; c++)
568 msec[c] = curdata->ex_secinfo[c];
569
570 /*
571 * Add the flavor that's not in the new export, but still
572 * referenced by its children.
573 */
574 mcnt = ccnt;
575 for (o = 0; o < ocnt; o++) {
576 if (! SEC_REF_SELF(&oldsecinfo[o])) {
577 for (c = 0; c < ccnt; c++) {
578 if (oldsecinfo[o].s_secinfo.sc_nfsnum ==
579 curdata->ex_secinfo[c].s_secinfo.sc_nfsnum)
580 break;
581 }
582
583 /*
584 * This is the one. Add it. Decrement the ref count
585 * by 1 if the flavor is an explicitly shared flavor
586 * for the oldsecinfo export node.
587 */
588 if (c == ccnt) {
589 srv_secinfo_copy(&oldsecinfo[o], &msec[mcnt]);
590 if (SEC_REF_EXPORTED(&oldsecinfo[o]))
591 msec[mcnt].s_refcnt--;
592
593 SECREF_TRACE(curdata,
594 "reshare_xfer_implicit_child_refs",
595 msec[mcnt].s_secinfo.sc_nfsnum,
596 msec[mcnt].s_refcnt);
597
598 ASSERT(msec[mcnt].s_refcnt >= 0);
599 mcnt++;
600 }
601 }
602 }
603
604 ASSERT(mcnt == tcnt);
605 /*
606 * Done. Update curdata, free the existing secinfo list in
607 * curdata and set the new value.
608 */
609 if (ccnt > 0)
610 kmem_free(curdata->ex_secinfo, ccnt * sizeof (struct secinfo));
611 curdata->ex_seccnt = tcnt;
612 curdata->ex_secinfo = msec;
613 }
614
615 /*
616 * When unsharing an old export node and the old node becomes a pseudo node,
617 * if there is more than 1 export reference to an old flavor (i.e. some of
618 * its children shared with this flavor), this flavor information needs to
619 * be transferred to the new shared node.
620 *
621 * This routine is used under the protection of exported_lock (RW_WRITER).
622 */
623 void
624 srv_secinfo_exp2pseu(exportdata_t *curdata, exportdata_t *olddata)
625 {
626 int ocnt, o; /* sec count in transfer data - trandata */
627 int tcnt, mcnt; /* total sec count after transfer */
628 struct secinfo *msec; /* merged secinfo list */
629
630 ASSERT(curdata->ex_flags & EX_PSEUDO);
631 ASSERT(curdata->ex_seccnt == 0);
632
633 ocnt = olddata->ex_seccnt;
634
635 /*
636 * If the olddata has flavors with more than 1 reference count,
637 * transfer the information to the curdata.
638 */
639 tcnt = ocnt;
640
641 for (o = 0; o < ocnt; o++) {
642 if (SEC_REF_SELF(&olddata->ex_secinfo[o]))
643 tcnt--;
644 }
645
646 if (tcnt == 0)
647 return; /* no transfer to do */
648
649 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
650
651 mcnt = 0;
652 for (o = 0; o < ocnt; o++) {
653 if (! SEC_REF_SELF(&olddata->ex_secinfo[o])) {
654
655 /*
656 * Decrement the reference count by 1 if the flavor is
657 * an explicitly shared flavor for the olddata export
658 * node.
659 */
660 srv_secinfo_copy(&olddata->ex_secinfo[o], &msec[mcnt]);
661 msec[mcnt].s_flags = M_RO;
662 if (SEC_REF_EXPORTED(&olddata->ex_secinfo[o]))
663 msec[mcnt].s_refcnt--;
664
665 SECREF_TRACE(curdata, "unshare_morph_pseudo",
666 msec[mcnt].s_secinfo.sc_nfsnum,
667 msec[mcnt].s_refcnt);
668
669 ASSERT(msec[mcnt].s_refcnt >= 0);
670 mcnt++;
671 }
672 }
673
674 ASSERT(mcnt == tcnt);
675 /*
676 * Done. Update curdata.
677 * Free up the existing secinfo list in curdata and
678 * set the new value.
679 */
680 curdata->ex_seccnt = tcnt;
681 curdata->ex_secinfo = msec;
682 }
683
684 /*
685 * Find for given treenode the exportinfo which has its
686 * exp_visible linked on its exi_visible list.
687 *
688 * Note: We could add new pointer either to treenode or
689 * to exp_visible, which will point there directly.
690 * This would buy some speed for some memory.
691 */
692 exportinfo_t *
693 vis2exi(treenode_t *tnode)
694 {
695 exportinfo_t *exi_ret = NULL;
696
697 for (;;) {
698 tnode = tnode->tree_parent;
699 if (TREE_ROOT(tnode)) {
700 exi_ret = tnode->tree_exi;
701 break;
702 }
703 }
704
705 /* 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,
1015 "NFS: failed un-exports in zone %d",
1016 (int) ng->nfs_zoneid);
1017 }
1018
1019 rw_exit(&ne->exported_lock);
1020 }
1021
1022 void
1023 nfs_export_zone_fini(nfs_globals_t *ng)
1024 {
1025 int i;
1026 nfs_export_t *ne = ng->nfs_export;
1027 struct exportinfo *exi;
1028
1029 ng->nfs_export = NULL;
1030
1031 rw_enter(&ne->exported_lock, RW_WRITER);
1032
1033 mutex_enter(&nfs_exi_id_lock);
1034 avl_remove(&exi_id_tree, ne->exi_root);
1035 mutex_exit(&nfs_exi_id_lock);
1036
1037 export_unlink(ne, ne->exi_root);
1038
1039 rw_exit(&ne->exported_lock);
1040
1041 /* Deallocate the place holder for the public file handle */
1042 srv_secinfo_list_free(ne->exi_root->exi_export.ex_secinfo,
1043 ne->exi_root->exi_export.ex_seccnt);
1044 mutex_destroy(&ne->exi_root->exi_lock);
1045
1046 rw_destroy(&ne->exi_root->exi_cache_lock);
1047 for (i = 0; i < AUTH_TABLESIZE; i++) {
1048 avl_destroy(ne->exi_root->exi_cache[i]);
1049 kmem_free(ne->exi_root->exi_cache[i], sizeof (avl_tree_t));
1050 }
1051
1052 kmem_free(ne->exi_root->exi_export.ex_path,
1053 ne->exi_root->exi_export.ex_pathlen + 1);
1054 kmem_free(ne->exi_root, sizeof (*ne->exi_root));
1055
1056 /*
1057 * The shutdown hook should have left the exi_id_tree
1058 * with nothing belonging to this zone.
1059 */
1060 mutex_enter(&nfs_exi_id_lock);
1061 i = 0;
1062 exi = avl_first(&exi_id_tree);
1063 while (exi != NULL) {
1064 if (exi->exi_zoneid == ng->nfs_zoneid)
1065 i++;
1066 exi = AVL_NEXT(&exi_id_tree, exi);
1067 }
1068 mutex_exit(&nfs_exi_id_lock);
1069 if (i > 0) {
1070 cmn_err(CE_NOTE,
1071 "NFS: zone %d has %d export IDs left after shutdown",
1072 (int) ng->nfs_zoneid, i);
1073 }
1074 rw_destroy(&ne->exported_lock);
1075 kmem_free(ne, sizeof (*ne));
1076 }
1077
1078 /*
1079 * Initialization routine for export routines.
1080 * Should only be called once.
1081 */
1082 void
1083 nfs_exportinit(void)
1084 {
1085 mutex_init(&nfs_exi_id_lock, NULL, MUTEX_DEFAULT, NULL);
1086
1087 /* exi_id handling initialization */
1088 exi_id_next = 0;
1089 exi_id_overflow = FALSE;
1090 avl_create(&exi_id_tree, exi_id_compar, sizeof (struct exportinfo),
1091 offsetof(struct exportinfo, exi_id_link));
1092
1093 nfslog_init();
1094 }
1095
1096 /*
1097 * Finalization routine for export routines.
1098 */
1099 void
1100 nfs_exportfini(void)
1101 {
1102 avl_destroy(&exi_id_tree);
1103 mutex_destroy(&nfs_exi_id_lock);
1104 }
1105
1106 /*
1107 * Check if 2 gss mechanism identifiers are the same.
1108 *
1109 * return FALSE if not the same.
1110 * return TRUE if the same.
1111 */
1112 static bool_t
1113 nfs_mech_equal(rpc_gss_OID mech1, rpc_gss_OID mech2)
1114 {
1115 if ((mech1->length == 0) && (mech2->length == 0))
1116 return (TRUE);
1117
1118 if (mech1->length != mech2->length)
1119 return (FALSE);
1120
1121 return (bcmp(mech1->elements, mech2->elements, mech1->length) == 0);
1122 }
1123
1124 /*
1125 * This routine is used by rpc to map rpc security number
1126 * to nfs specific security flavor number.
1127 *
1128 * The gss callback prototype is
1129 * callback(struct svc_req *, gss_cred_id_t *, gss_ctx_id_t *,
1130 * rpc_gss_lock_t *, void **),
1131 * since nfs does not use the gss_cred_id_t/gss_ctx_id_t arguments
1132 * we cast them to void.
1133 */
1134 /*ARGSUSED*/
1135 bool_t
1136 rfs_gsscallback(struct svc_req *req, gss_cred_id_t deleg, void *gss_context,
1137 rpc_gss_lock_t *lock, void **cookie)
1138 {
1139 int i, j;
1140 rpc_gss_rawcred_t *raw_cred;
1141 struct exportinfo *exi;
1142 nfs_export_t *ne = nfs_get_export();
1143
1144 /*
1145 * We don't deal with delegated credentials.
1146 */
1147 if (deleg != GSS_C_NO_CREDENTIAL)
1148 return (FALSE);
1149
1150 raw_cred = lock->raw_cred;
1151 *cookie = NULL;
1152
1153 rw_enter(&ne->exported_lock, RW_READER);
1154
1155 for (i = 0; i < EXPTABLESIZE; i++) {
1156 exi = ne->exptable[i];
1157 while (exi) {
1158 if (exi->exi_export.ex_seccnt > 0) {
1159 struct secinfo *secp;
1160 seconfig_t *se;
1161 int seccnt;
1162
1163 secp = exi->exi_export.ex_secinfo;
1164 seccnt = exi->exi_export.ex_seccnt;
1165 for (j = 0; j < seccnt; j++) {
1166 /*
1167 * If there is a map of the triplet
1168 * (mechanism, service, qop) between
1169 * raw_cred and the exported flavor,
1170 * get the psudo flavor number.
1171 * Also qop should not be NULL, it
1172 * should be "default" or something
1173 * else.
1174 */
1175 se = &secp[j].s_secinfo;
1176 if ((se->sc_rpcnum == RPCSEC_GSS) &&
1177
1178 (nfs_mech_equal(
1179 se->sc_gss_mech_type,
1180 raw_cred->mechanism)) &&
1181
1182 (se->sc_service ==
1183 raw_cred->service) &&
1184 (raw_cred->qop == se->sc_qop)) {
1185
1186 *cookie = (void *)(uintptr_t)
1187 se->sc_nfsnum;
1188 goto done;
1189 }
1190 }
1191 }
1192 exi = exi->fid_hash.next;
1193 }
1194 }
1195 done:
1196 rw_exit(&ne->exported_lock);
1197
1198 /*
1199 * If no nfs pseudo number mapping can be found in the export
1200 * table, assign the nfsflavor to NFS_FLAVOR_NOMAP. In V4, we may
1201 * recover the flavor mismatch from NFS layer (NFS4ERR_WRONGSEC).
1202 *
1203 * For example:
1204 * server first shares with krb5i;
1205 * client mounts with krb5i;
1206 * server re-shares with krb5p;
1207 * client tries with krb5i, but no mapping can be found;
1208 * rpcsec_gss module calls this routine to do the mapping,
1209 * if this routine fails, request is rejected from
1210 * the rpc layer.
1211 * What we need is to let the nfs layer rejects the request.
1212 * For V4, we can reject with NFS4ERR_WRONGSEC and the client
1213 * may recover from it by getting the new flavor via SECINFO.
1214 *
1215 * nfs pseudo number for RPCSEC_GSS mapping (see nfssec.conf)
1216 * is owned by IANA (see RFC 2623).
1217 *
1218 * XXX NFS_FLAVOR_NOMAP is defined in Solaris to work around
1219 * the implementation issue. This number should not overlap with
1220 * any new IANA defined pseudo flavor numbers.
1221 */
1222 if (*cookie == NULL)
1223 *cookie = (void *)NFS_FLAVOR_NOMAP;
1224
1225 lock->locked = TRUE;
1226
1227 return (TRUE);
1228 }
1229
1230
1231 /*
1232 * Exportfs system call; credentials should be checked before
1233 * calling this function.
1234 */
1235 int
1236 exportfs(struct exportfs_args *args, model_t model, cred_t *cr)
1237 {
1238 vnode_t *vp;
1239 vnode_t *dvp;
1240 struct exportdata *kex;
1241 struct exportinfo *exi = NULL;
1242 struct exportinfo *ex, *ex1, *ex2;
1243 fid_t fid;
1244 fsid_t fsid;
1245 int error;
1246 size_t allocsize;
1247 struct secinfo *sp;
1248 struct secinfo *exs;
1249 rpc_gss_callback_t cb;
1250 char *pathbuf;
1251 char *log_buffer;
1252 char *tagbuf;
1253 int callback;
1254 int allocd_seccnt;
1255 STRUCT_HANDLE(exportfs_args, uap);
1256 STRUCT_DECL(exportdata, uexi);
1257 struct secinfo newsec[MAX_FLAVORS];
1258 int newcnt;
1259 struct secinfo oldsec[MAX_FLAVORS];
1260 int oldcnt;
1261 int i;
1262 struct pathname lookpn;
1263 nfs_export_t *ne = nfs_get_export();
1264
1265 STRUCT_SET_HANDLE(uap, model, args);
1266
1267 /* Read in pathname from userspace */
1268 if (error = pn_get(STRUCT_FGETP(uap, dname), UIO_USERSPACE, &lookpn))
1269 return (error);
1270
1271 /* Walk the export list looking for that pathname */
1272 rw_enter(&ne->exported_lock, RW_READER);
1273 DTRACE_PROBE(nfss__i__exported_lock1_start);
1274 for (ex1 = ne->exptable_path_hash[pkp_tab_hash(lookpn.pn_path,
1275 strlen(lookpn.pn_path))]; ex1; ex1 = ex1->path_hash.next) {
1276 if (ex1 != ne->exi_root && 0 ==
1277 strcmp(ex1->exi_export.ex_path, lookpn.pn_path)) {
1278 exi_hold(ex1);
1279 break;
1280 }
1281 }
1282 DTRACE_PROBE(nfss__i__exported_lock1_stop);
1283 rw_exit(&ne->exported_lock);
1284
1285 /* Is this an unshare? */
1286 if (STRUCT_FGETP(uap, uex) == NULL) {
1287 pn_free(&lookpn);
1288 if (ex1 == NULL)
1289 return (EINVAL);
1290 error = unexport(ne, ex1);
1291 exi_rele(ex1);
1292 return (error);
1293 }
1294
1295 /* It is a share or a re-share */
1296 error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1297 FOLLOW, &dvp, &vp);
1298 if (error == EINVAL) {
1299 /*
1300 * if fname resolves to / we get EINVAL error
1301 * since we wanted the parent vnode. Try again
1302 * with NULL dvp.
1303 */
1304 error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1305 FOLLOW, NULL, &vp);
1306 dvp = NULL;
1307 }
1308 if (!error && vp == NULL) {
1309 /* Last component of fname not found */
1310 if (dvp != NULL)
1311 VN_RELE(dvp);
1312 error = ENOENT;
1313 }
1314 if (error) {
1315 pn_free(&lookpn);
1316 if (ex1)
1317 exi_rele(ex1);
1318 return (error);
1319 }
1320
1321 /*
1322 * 'vp' may be an AUTOFS node, so we perform a
1323 * VOP_ACCESS() to trigger the mount of the
1324 * intended filesystem, so we can share the intended
1325 * filesystem instead of the AUTOFS filesystem.
1326 */
1327 (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
1328
1329 /*
1330 * We're interested in the top most filesystem.
1331 * This is specially important when uap->dname is a trigger
1332 * AUTOFS node, since we're really interested in sharing the
1333 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
1334 * call not the AUTOFS node itself.
1335 */
1336 if (vn_mountedvfs(vp) != NULL) {
1337 if (error = traverse(&vp)) {
1338 VN_RELE(vp);
1339 if (dvp != NULL)
1340 VN_RELE(dvp);
1341 pn_free(&lookpn);
1342 if (ex1)
1343 exi_rele(ex1);
1344 return (error);
1345 }
1346 }
1347
1348 /* Do not allow sharing another vnode for already shared path */
1349 if (ex1 && !PSEUDO(ex1) && !VN_CMP(ex1->exi_vp, vp)) {
1350 VN_RELE(vp);
1351 if (dvp != NULL)
1352 VN_RELE(dvp);
1353 pn_free(&lookpn);
1354 exi_rele(ex1);
1355 return (EEXIST);
1356 }
1357 if (ex1)
1358 exi_rele(ex1);
1359
1360 /*
1361 * Get the vfs id
1362 */
1363 bzero(&fid, sizeof (fid));
1364 fid.fid_len = MAXFIDSZ;
1365 error = VOP_FID(vp, &fid, NULL);
1366 fsid = vp->v_vfsp->vfs_fsid;
1367
1368 if (error) {
1369 VN_RELE(vp);
1370 if (dvp != NULL)
1371 VN_RELE(dvp);
1372 /*
1373 * If VOP_FID returns ENOSPC then the fid supplied
1374 * is too small. For now we simply return EREMOTE.
1375 */
1376 if (error == ENOSPC)
1377 error = EREMOTE;
1378 pn_free(&lookpn);
1379 return (error);
1380 }
1381
1382 /*
1383 * Do not allow re-sharing a shared vnode under a different path
1384 * PSEUDO export has ex_path fabricated, e.g. "/tmp (pseudo)", skip it.
1385 */
1386 rw_enter(&ne->exported_lock, RW_READER);
1387 DTRACE_PROBE(nfss__i__exported_lock2_start);
1388 for (ex2 = ne->exptable[exptablehash(&fsid, &fid)]; ex2;
1389 ex2 = ex2->fid_hash.next) {
1390 if (ex2 != ne->exi_root && !PSEUDO(ex2) &&
1391 VN_CMP(ex2->exi_vp, vp) &&
1392 strcmp(ex2->exi_export.ex_path, lookpn.pn_path) != 0) {
1393 DTRACE_PROBE(nfss__i__exported_lock2_stop);
1394 rw_exit(&ne->exported_lock);
1395 VN_RELE(vp);
1396 if (dvp != NULL)
1397 VN_RELE(dvp);
1398 pn_free(&lookpn);
1399 return (EEXIST);
1400 }
1401 }
1402 DTRACE_PROBE(nfss__i__exported_lock2_stop);
1403 rw_exit(&ne->exported_lock);
1404 pn_free(&lookpn);
1405
1406 exi = kmem_zalloc(sizeof (*exi), KM_SLEEP);
1407 exi->exi_fsid = fsid;
1408 exi->exi_fid = fid;
1409 exi->exi_vp = vp;
1410 exi->exi_count = 1;
1411 exi->exi_zoneid = crgetzoneid(cr);
1412 ASSERT3U(exi->exi_zoneid, ==, curzone->zone_id);
1413 exi->exi_volatile_dev = (vfssw[vp->v_vfsp->vfs_fstype].vsw_flag &
1414 VSW_VOLATILEDEV) ? 1 : 0;
1415 mutex_init(&exi->exi_lock, NULL, MUTEX_DEFAULT, NULL);
1416 exi->exi_dvp = dvp;
1417
1418 /*
1419 * Initialize auth cache and auth cache lock
1420 */
1421 for (i = 0; i < AUTH_TABLESIZE; i++) {
1422 exi->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1423 avl_create(exi->exi_cache[i], nfsauth_cache_clnt_compar,
1424 sizeof (struct auth_cache_clnt),
1425 offsetof(struct auth_cache_clnt, authc_link));
1426 }
1427 rw_init(&exi->exi_cache_lock, NULL, RW_DEFAULT, NULL);
1428
1429 /*
1430 * Build up the template fhandle
1431 */
1432 exi->exi_fh.fh_fsid = fsid;
1433 if (exi->exi_fid.fid_len > sizeof (exi->exi_fh.fh_xdata)) {
1434 error = EREMOTE;
1435 goto out1;
1436 }
1437 exi->exi_fh.fh_xlen = exi->exi_fid.fid_len;
1438 bcopy(exi->exi_fid.fid_data, exi->exi_fh.fh_xdata,
1439 exi->exi_fid.fid_len);
1440
1441 exi->exi_fh.fh_len = sizeof (exi->exi_fh.fh_data);
1442
1443 kex = &exi->exi_export;
1444
1445 /*
1446 * Load in everything, and do sanity checking
1447 */
1448 STRUCT_INIT(uexi, model);
1449 if (copyin(STRUCT_FGETP(uap, uex), STRUCT_BUF(uexi),
1450 STRUCT_SIZE(uexi))) {
1451 error = EFAULT;
1452 goto out1;
1453 }
1454
1455 kex->ex_version = STRUCT_FGET(uexi, ex_version);
1456 if (kex->ex_version != EX_CURRENT_VERSION) {
1457 error = EINVAL;
1458 cmn_err(CE_WARN,
1459 "NFS: exportfs requires export struct version 2 - got %d\n",
1460 kex->ex_version);
1461 goto out1;
1462 }
1463
1464 /*
1465 * Must have at least one security entry
1466 */
1467 kex->ex_seccnt = STRUCT_FGET(uexi, ex_seccnt);
1468 if (kex->ex_seccnt < 1) {
1469 error = EINVAL;
1470 goto out1;
1471 }
1472
1473 kex->ex_path = STRUCT_FGETP(uexi, ex_path);
1474 kex->ex_pathlen = STRUCT_FGET(uexi, ex_pathlen);
1475 kex->ex_flags = STRUCT_FGET(uexi, ex_flags);
1476 kex->ex_anon = STRUCT_FGET(uexi, ex_anon);
1477 kex->ex_secinfo = STRUCT_FGETP(uexi, ex_secinfo);
1478 kex->ex_index = STRUCT_FGETP(uexi, ex_index);
1479 kex->ex_log_buffer = STRUCT_FGETP(uexi, ex_log_buffer);
1480 kex->ex_log_bufferlen = STRUCT_FGET(uexi, ex_log_bufferlen);
1481 kex->ex_tag = STRUCT_FGETP(uexi, ex_tag);
1482 kex->ex_taglen = STRUCT_FGET(uexi, ex_taglen);
1483
1484 /*
1485 * Copy the exported pathname into
1486 * an appropriately sized buffer.
1487 */
1488 pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1489 if (copyinstr(kex->ex_path, pathbuf, MAXPATHLEN, &kex->ex_pathlen)) {
1490 kmem_free(pathbuf, MAXPATHLEN);
1491 error = EFAULT;
1492 goto out1;
1493 }
1494 kex->ex_path = kmem_alloc(kex->ex_pathlen + 1, KM_SLEEP);
1495 bcopy(pathbuf, kex->ex_path, kex->ex_pathlen);
1496 kex->ex_path[kex->ex_pathlen] = '\0';
1497 kmem_free(pathbuf, MAXPATHLEN);
1498
1499 /*
1500 * Get the path to the logging buffer and the tag
1501 */
1502 if (kex->ex_flags & EX_LOG) {
1503 log_buffer = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1504 if (copyinstr(kex->ex_log_buffer, log_buffer, MAXPATHLEN,
1505 &kex->ex_log_bufferlen)) {
1506 kmem_free(log_buffer, MAXPATHLEN);
1507 error = EFAULT;
1508 goto out2;
1509 }
1510 kex->ex_log_buffer =
1511 kmem_alloc(kex->ex_log_bufferlen + 1, KM_SLEEP);
1512 bcopy(log_buffer, kex->ex_log_buffer, kex->ex_log_bufferlen);
1513 kex->ex_log_buffer[kex->ex_log_bufferlen] = '\0';
1514 kmem_free(log_buffer, MAXPATHLEN);
1515
1516 tagbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1517 if (copyinstr(kex->ex_tag, tagbuf, MAXPATHLEN,
1518 &kex->ex_taglen)) {
1519 kmem_free(tagbuf, MAXPATHLEN);
1520 error = EFAULT;
1521 goto out3;
1522 }
1523 kex->ex_tag = kmem_alloc(kex->ex_taglen + 1, KM_SLEEP);
1524 bcopy(tagbuf, kex->ex_tag, kex->ex_taglen);
1525 kex->ex_tag[kex->ex_taglen] = '\0';
1526 kmem_free(tagbuf, MAXPATHLEN);
1527 }
1528
1529 /*
1530 * Load the security information for each flavor
1531 */
1532 allocsize = kex->ex_seccnt * SIZEOF_STRUCT(secinfo, model);
1533 sp = kmem_zalloc(allocsize, KM_SLEEP);
1534 if (copyin(kex->ex_secinfo, sp, allocsize)) {
1535 kmem_free(sp, allocsize);
1536 error = EFAULT;
1537 goto out4;
1538 }
1539
1540 /*
1541 * All of these nested structures need to be converted to
1542 * the kernel native format.
1543 */
1544 if (model != DATAMODEL_NATIVE) {
1545 size_t allocsize2;
1546 struct secinfo *sp2;
1547
1548 allocsize2 = kex->ex_seccnt * sizeof (struct secinfo);
1549 sp2 = kmem_zalloc(allocsize2, KM_SLEEP);
1550
1551 for (i = 0; i < kex->ex_seccnt; i++) {
1552 STRUCT_HANDLE(secinfo, usi);
1553
1554 STRUCT_SET_HANDLE(usi, model,
1555 (struct secinfo *)((caddr_t)sp +
1556 (i * SIZEOF_STRUCT(secinfo, model))));
1557 bcopy(STRUCT_FGET(usi, s_secinfo.sc_name),
1558 sp2[i].s_secinfo.sc_name, MAX_NAME_LEN);
1559 sp2[i].s_secinfo.sc_nfsnum =
1560 STRUCT_FGET(usi, s_secinfo.sc_nfsnum);
1561 sp2[i].s_secinfo.sc_rpcnum =
1562 STRUCT_FGET(usi, s_secinfo.sc_rpcnum);
1563 bcopy(STRUCT_FGET(usi, s_secinfo.sc_gss_mech),
1564 sp2[i].s_secinfo.sc_gss_mech, MAX_NAME_LEN);
1565 sp2[i].s_secinfo.sc_gss_mech_type =
1566 STRUCT_FGETP(usi, s_secinfo.sc_gss_mech_type);
1567 sp2[i].s_secinfo.sc_qop =
1568 STRUCT_FGET(usi, s_secinfo.sc_qop);
1569 sp2[i].s_secinfo.sc_service =
1570 STRUCT_FGET(usi, s_secinfo.sc_service);
1571
1572 sp2[i].s_flags = STRUCT_FGET(usi, s_flags);
1573 sp2[i].s_window = STRUCT_FGET(usi, s_window);
1574 sp2[i].s_rootid = STRUCT_FGET(usi, s_rootid);
1575 sp2[i].s_rootcnt = STRUCT_FGET(usi, s_rootcnt);
1576 sp2[i].s_rootnames = STRUCT_FGETP(usi, s_rootnames);
1577 }
1578 kmem_free(sp, allocsize);
1579 sp = sp2;
1580 allocsize = allocsize2;
1581 }
1582
1583 kex->ex_secinfo = sp;
1584
1585 /*
1586 * And now copy rootnames for each individual secinfo.
1587 */
1588 callback = 0;
1589 allocd_seccnt = 0;
1590 while (allocd_seccnt < kex->ex_seccnt) {
1591
1592 exs = &sp[allocd_seccnt];
1593 if (exs->s_rootcnt > 0) {
1594 if (!sec_svc_loadrootnames(exs->s_secinfo.sc_rpcnum,
1595 exs->s_rootcnt, &exs->s_rootnames, model)) {
1596 error = EFAULT;
1597 goto out5;
1598 }
1599 }
1600
1601 if (exs->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
1602 rpc_gss_OID mech_tmp;
1603 STRUCT_DECL(rpc_gss_OID_s, umech_tmp);
1604 caddr_t elements_tmp;
1605
1606 /* Copyin mechanism type */
1607 STRUCT_INIT(umech_tmp, model);
1608 mech_tmp = kmem_alloc(sizeof (*mech_tmp), KM_SLEEP);
1609 if (copyin(exs->s_secinfo.sc_gss_mech_type,
1610 STRUCT_BUF(umech_tmp), STRUCT_SIZE(umech_tmp))) {
1611 kmem_free(mech_tmp, sizeof (*mech_tmp));
1612 error = EFAULT;
1613 goto out5;
1614 }
1615 mech_tmp->length = STRUCT_FGET(umech_tmp, length);
1616 mech_tmp->elements = STRUCT_FGETP(umech_tmp, elements);
1617
1618 elements_tmp = kmem_alloc(mech_tmp->length, KM_SLEEP);
1619 if (copyin(mech_tmp->elements, elements_tmp,
1620 mech_tmp->length)) {
1621 kmem_free(elements_tmp, mech_tmp->length);
1622 kmem_free(mech_tmp, sizeof (*mech_tmp));
1623 error = EFAULT;
1624 goto out5;
1625 }
1626 mech_tmp->elements = elements_tmp;
1627 exs->s_secinfo.sc_gss_mech_type = mech_tmp;
1628 allocd_seccnt++;
1629
1630 callback = 1;
1631 } else
1632 allocd_seccnt++;
1633 }
1634
1635 /*
1636 * Init the secinfo reference count and mark these flavors
1637 * explicitly exported flavors.
1638 */
1639 for (i = 0; i < kex->ex_seccnt; i++) {
1640 kex->ex_secinfo[i].s_flags |= M_4SEC_EXPORTED;
1641 kex->ex_secinfo[i].s_refcnt = 1;
1642 }
1643
1644 /*
1645 * Set up rpcsec_gss callback routine entry if any.
1646 */
1647 if (callback) {
1648 cb.callback = rfs_gsscallback;
1649 cb.program = NFS_ACL_PROGRAM;
1650 for (cb.version = NFS_ACL_VERSMIN;
1651 cb.version <= NFS_ACL_VERSMAX; cb.version++) {
1652 (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1653 (void *)&cb);
1654 }
1655
1656 cb.program = NFS_PROGRAM;
1657 for (cb.version = NFS_VERSMIN;
1658 cb.version <= NFS_VERSMAX; cb.version++) {
1659 (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1660 (void *)&cb);
1661 }
1662 }
1663
1664 /*
1665 * Check the index flag. Do this here to avoid holding the
1666 * lock while dealing with the index option (as we do with
1667 * the public option).
1668 */
1669 if (kex->ex_flags & EX_INDEX) {
1670 if (!kex->ex_index) { /* sanity check */
1671 error = EINVAL;
1672 goto out5;
1673 }
1674 if (error = loadindex(kex))
1675 goto out5;
1676 }
1677
1678 if (kex->ex_flags & EX_LOG) {
1679 if (error = nfslog_setup(exi))
1680 goto out6;
1681 }
1682
1683 /*
1684 * Insert the new entry at the front of the export list
1685 */
1686 rw_enter(&ne->exported_lock, RW_WRITER);
1687 DTRACE_PROBE(nfss__i__exported_lock3_start);
1688
1689 export_link(ne, exi);
1690
1691 /*
1692 * Check the rest of the list for an old entry for the fs.
1693 * If one is found then unlink it, wait until this is the
1694 * only reference and then free it.
1695 */
1696 for (ex = exi->fid_hash.next; ex != NULL; ex = ex->fid_hash.next) {
1697 if (ex != ne->exi_root && VN_CMP(ex->exi_vp, vp)) {
1698 mutex_enter(&nfs_exi_id_lock);
1699 avl_remove(&exi_id_tree, ex);
1700 mutex_exit(&nfs_exi_id_lock);
1701 export_unlink(ne, ex);
1702 break;
1703 }
1704 }
1705
1706 /*
1707 * If the public filehandle is pointing at the
1708 * old entry, then point it back at the root.
1709 */
1710 if (ex != NULL && ex == ne->exi_public)
1711 ne->exi_public = ne->exi_root;
1712
1713 /*
1714 * If the public flag is on, make the global exi_public
1715 * point to this entry and turn off the public bit so that
1716 * we can distinguish it from the place holder export.
1717 */
1718 if (kex->ex_flags & EX_PUBLIC) {
1719 ne->exi_public = exi;
1720 kex->ex_flags &= ~EX_PUBLIC;
1721 }
1722
1723 #ifdef VOLATILE_FH_TEST
1724 /*
1725 * Set up the volatile_id value if volatile on share.
1726 * The list of volatile renamed filehandles is always destroyed,
1727 * if the fs was reshared.
1728 */
1729 if (kex->ex_flags & EX_VOLFH)
1730 exi->exi_volatile_id = gethrestime_sec();
1731
1732 mutex_init(&exi->exi_vol_rename_lock, NULL, MUTEX_DEFAULT, NULL);
1733 #endif /* VOLATILE_FH_TEST */
1734
1735 /*
1736 * If this is a new export, then climb up
1737 * the tree and check if any pseudo exports
1738 * need to be created to provide a path for
1739 * NFS v4 clients.
1740 */
1741 if (ex == NULL) {
1742 error = treeclimb_export(exi);
1743 if (error)
1744 goto out7;
1745 } else {
1746 /* If it's a re-export update namespace tree */
1747 exi->exi_tree = ex->exi_tree;
1748 exi->exi_tree->tree_exi = exi;
1749
1750 /* Update the change timestamp */
1751 tree_update_change(ne, exi->exi_tree, NULL);
1752 }
1753
1754 /*
1755 * build a unique flavor list from the flavors specified
1756 * in the share cmd. unique means that each flavor only
1757 * appears once in the secinfo list -- no duplicates allowed.
1758 */
1759 newcnt = build_seclist_nodups(&exi->exi_export, newsec, FALSE);
1760
1761 srv_secinfo_treeclimb(ne, exi, newsec, newcnt, TRUE);
1762
1763 /*
1764 * If re-sharing an old export entry, update the secinfo data
1765 * depending on if the old entry is a pseudo node or not.
1766 */
1767 if (ex != NULL) {
1768 oldcnt = build_seclist_nodups(&ex->exi_export, oldsec, FALSE);
1769 if (PSEUDO(ex)) {
1770 /*
1771 * The dir being shared is a pseudo export root (which
1772 * will be transformed into a real export root). The
1773 * flavor(s) of the new share were propagated to the
1774 * ancestors by srv_secinfo_treeclimb() above. Now
1775 * transfer the implicit flavor refs from the old
1776 * pseudo exprot root to the new (real) export root.
1777 */
1778 srv_secinfo_add(&exi->exi_export.ex_secinfo,
1779 &exi->exi_export.ex_seccnt, oldsec, oldcnt, TRUE);
1780 } else {
1781 /*
1782 * First transfer implicit flavor refs to new export.
1783 * Remove old flavor refs last.
1784 */
1785 srv_secinfo_exp2exp(&exi->exi_export, oldsec, oldcnt);
1786 srv_secinfo_treeclimb(ne, ex, oldsec, oldcnt, FALSE);
1787 }
1788 }
1789
1790 /*
1791 * If it's a re-export and the old entry has a pseudonode list,
1792 * transfer it to the new export.
1793 */
1794 if (ex != NULL && (ex->exi_visible != NULL)) {
1795 exi->exi_visible = ex->exi_visible;
1796 ex->exi_visible = NULL;
1797 }
1798
1799 /*
1800 * Initialize exi_id and exi_kstats
1801 */
1802 if (ex != NULL) {
1803 exi->exi_id = ex->exi_id;
1804 } else {
1805 mutex_enter(&nfs_exi_id_lock);
1806 exi->exi_id = exi_id_get_next();
1807 mutex_exit(&nfs_exi_id_lock);
1808 }
1809 mutex_enter(&nfs_exi_id_lock);
1810 avl_add(&exi_id_tree, exi);
1811 mutex_exit(&nfs_exi_id_lock);
1812
1813 DTRACE_PROBE(nfss__i__exported_lock3_stop);
1814 rw_exit(&ne->exported_lock);
1815
1816 if (ne->exi_public == exi || kex->ex_flags & EX_LOG) {
1817 /*
1818 * Log share operation to this buffer only.
1819 */
1820 nfslog_share_record(exi, cr);
1821 }
1822
1823 if (ex != NULL)
1824 exi_rele(ex);
1825
1826 return (0);
1827
1828 out7:
1829 /* Unlink the new export in exptable. */
1830 export_unlink(ne, exi);
1831 DTRACE_PROBE(nfss__i__exported_lock3_stop);
1832 rw_exit(&ne->exported_lock);
1833 out6:
1834 if (kex->ex_flags & EX_INDEX)
1835 kmem_free(kex->ex_index, strlen(kex->ex_index) + 1);
1836 out5:
1837 /* free partially completed allocation */
1838 while (--allocd_seccnt >= 0) {
1839 exs = &kex->ex_secinfo[allocd_seccnt];
1840 srv_secinfo_entry_free(exs);
1841 }
1842
1843 if (kex->ex_secinfo) {
1844 kmem_free(kex->ex_secinfo,
1845 kex->ex_seccnt * sizeof (struct secinfo));
1846 }
1847
1848 out4:
1849 if ((kex->ex_flags & EX_LOG) && kex->ex_tag != NULL)
1850 kmem_free(kex->ex_tag, kex->ex_taglen + 1);
1851 out3:
1852 if ((kex->ex_flags & EX_LOG) && kex->ex_log_buffer != NULL)
1853 kmem_free(kex->ex_log_buffer, kex->ex_log_bufferlen + 1);
1854 out2:
1855 kmem_free(kex->ex_path, kex->ex_pathlen + 1);
1856 out1:
1857 VN_RELE(vp);
1858 if (dvp != NULL)
1859 VN_RELE(dvp);
1860 mutex_destroy(&exi->exi_lock);
1861 rw_destroy(&exi->exi_cache_lock);
1862 for (i = 0; i < AUTH_TABLESIZE; i++) {
1863 avl_destroy(exi->exi_cache[i]);
1864 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
1865 }
1866
1867 kmem_free(exi, sizeof (*exi));
1868
1869 return (error);
1870 }
1871
1872 /*
1873 * Remove the exportinfo from the export list
1874 */
1875 void
1876 export_unlink(nfs_export_t *ne, struct exportinfo *exi)
1877 {
1878 ASSERT(RW_WRITE_HELD(&ne->exported_lock));
1879
1880 exp_hash_unlink(exi, fid_hash);
1881 exp_hash_unlink(exi, path_hash);
1882 ASSERT3P(exi->exi_ne, ==, ne);
1883 exi->exi_ne = NULL;
1884 }
1885
1886 /*
1887 * Unexport an exported filesystem
1888 */
1889 static int
1890 unexport(nfs_export_t *ne, struct exportinfo *exi)
1891 {
1892 struct secinfo cursec[MAX_FLAVORS];
1893 int curcnt;
1894
1895 rw_enter(&ne->exported_lock, RW_WRITER);
1896
1897 /* Check if exi is still linked in the export table */
1898 if (!EXP_LINKED(exi) || PSEUDO(exi)) {
1899 rw_exit(&ne->exported_lock);
1900 return (EINVAL);
1901 }
1902
1903 mutex_enter(&nfs_exi_id_lock);
1904 avl_remove(&exi_id_tree, exi);
1905 mutex_exit(&nfs_exi_id_lock);
1906 export_unlink(ne, exi);
1907
1908 /*
1909 * Remove security flavors before treeclimb_unexport() is called
1910 * because srv_secinfo_treeclimb needs the namespace tree
1911 */
1912 curcnt = build_seclist_nodups(&exi->exi_export, cursec, TRUE);
1913 srv_secinfo_treeclimb(ne, exi, cursec, curcnt, FALSE);
1914
1915 /*
1916 * If there's a visible list, then need to leave
1917 * a pseudo export here to retain the visible list
1918 * for paths to exports below.
1919 */
1920 if (exi->exi_visible != NULL) {
1921 struct exportinfo *newexi;
1922
1923 newexi = pseudo_exportfs(ne, exi->exi_vp, &exi->exi_fid,
1924 exi->exi_visible, &exi->exi_export);
1925 exi->exi_visible = NULL;
1926
1927 /* interconnect the existing treenode with the new exportinfo */
1928 newexi->exi_tree = exi->exi_tree;
1929 newexi->exi_tree->tree_exi = newexi;
1930
1931 /* Update the change timestamp */
1932 tree_update_change(ne, exi->exi_tree, NULL);
1933 } else {
1934 treeclimb_unexport(ne, exi);
1935 }
1936
1937 rw_exit(&ne->exported_lock);
1938
1939 /*
1940 * Need to call into the NFSv4 server and release all data
1941 * held on this particular export. This is important since
1942 * the v4 server may be holding file locks or vnodes under
1943 * this export.
1944 */
1945 rfs4_clean_state_exi(ne, exi);
1946
1947 /*
1948 * Notify the lock manager that the filesystem is being
1949 * unexported.
1950 */
1951 lm_unexport(exi);
1952
1953 /*
1954 * If this was a public export, restore
1955 * the public filehandle to the root.
1956 */
1957
1958 /*
1959 * XXX KEBE ASKS --> Should CRED() instead be
1960 * exi->exi_zone->zone_kcred?
1961 */
1962 if (exi == ne->exi_public) {
1963 ne->exi_public = ne->exi_root;
1964
1965 nfslog_share_record(ne->exi_public, CRED());
1966 }
1967
1968 if (exi->exi_export.ex_flags & EX_LOG)
1969 nfslog_unshare_record(exi, CRED());
1970
1971 exi_rele(exi);
1972 return (0);
1973 }
1974
1975 /*
1976 * Get file handle system call.
1977 * Takes file name and returns a file handle for it.
1978 * Credentials must be verified before calling.
1979 */
1980 int
1981 nfs_getfh(struct nfs_getfh_args *args, model_t model, cred_t *cr)
1982 {
1983 nfs_fh3 fh;
1984 char buf[NFS3_MAXFHSIZE];
1985 char *logptr, logbuf[NFS3_MAXFHSIZE];
1986 int l = NFS3_MAXFHSIZE;
1987 vnode_t *vp;
1988 vnode_t *dvp;
1989 struct exportinfo *exi;
1990 int error;
1991 int vers;
1992 STRUCT_HANDLE(nfs_getfh_args, uap);
1993
1994 #ifdef lint
1995 model = model; /* STRUCT macros don't always use it */
1996 #endif
1997
1998 STRUCT_SET_HANDLE(uap, model, args);
1999
2000 error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
2001 FOLLOW, &dvp, &vp);
2002 if (error == EINVAL) {
2003 /*
2004 * if fname resolves to / we get EINVAL error
2005 * since we wanted the parent vnode. Try again
2006 * with NULL dvp.
2007 */
2008 error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
2009 FOLLOW, NULL, &vp);
2010 dvp = NULL;
2011 }
2012 if (!error && vp == NULL) {
2013 /*
2014 * Last component of fname not found
2015 */
2016 if (dvp != NULL) {
2017 VN_RELE(dvp);
2018 }
2019 error = ENOENT;
2020 }
2021 if (error)
2022 return (error);
2023
2024 /*
2025 * 'vp' may be an AUTOFS node, so we perform a
2026 * VOP_ACCESS() to trigger the mount of the
2027 * intended filesystem, so we can share the intended
2028 * filesystem instead of the AUTOFS filesystem.
2029 */
2030 (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
2031
2032 /*
2033 * We're interested in the top most filesystem.
2034 * This is specially important when uap->dname is a trigger
2035 * AUTOFS node, since we're really interested in sharing the
2036 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
2037 * call not the AUTOFS node itself.
2038 */
2039 if (vn_mountedvfs(vp) != NULL) {
2040 if (error = traverse(&vp)) {
2041 VN_RELE(vp);
2042 if (dvp != NULL)
2043 VN_RELE(dvp);
2044 return (error);
2045 }
2046 }
2047
2048 vers = STRUCT_FGET(uap, vers);
2049 exi = nfs_vptoexi(dvp, vp, cr, NULL, &error, FALSE);
2050 if (!error) {
2051 if (vers == NFS_VERSION) {
2052 error = makefh((fhandle_t *)buf, vp, exi);
2053 l = NFS_FHSIZE;
2054 logptr = buf;
2055 } else if (vers == NFS_V3) {
2056 int i, sz, pad;
2057
2058 error = makefh3(&fh, vp, exi);
2059 l = RNDUP(fh.fh3_length);
2060 if (!error && (l > sizeof (fhandle3_t)))
2061 error = EREMOTE;
2062 logptr = logbuf;
2063 if (!error) {
2064 i = 0;
2065 sz = sizeof (fsid_t);
2066 bcopy(&fh.fh3_fsid, &buf[i], sz);
2067 i += sz;
2068
2069 /*
2070 * For backwards compatibility, the
2071 * fid length may be less than
2072 * NFS_FHMAXDATA, but it was always
2073 * encoded as NFS_FHMAXDATA bytes.
2074 */
2075
2076 sz = sizeof (ushort_t);
2077 bcopy(&fh.fh3_len, &buf[i], sz);
2078 i += sz;
2079 bcopy(fh.fh3_data, &buf[i], fh.fh3_len);
2080 i += fh.fh3_len;
2081 pad = (NFS_FHMAXDATA - fh.fh3_len);
2082 if (pad > 0) {
2083 bzero(&buf[i], pad);
2084 i += pad;
2085 l += pad;
2086 }
2087
2088 sz = sizeof (ushort_t);
2089 bcopy(&fh.fh3_xlen, &buf[i], sz);
2090 i += sz;
2091 bcopy(fh.fh3_xdata, &buf[i], fh.fh3_xlen);
2092 i += fh.fh3_xlen;
2093 pad = (NFS_FHMAXDATA - fh.fh3_xlen);
2094 if (pad > 0) {
2095 bzero(&buf[i], pad);
2096 i += pad;
2097 l += pad;
2098 }
2099 }
2100 /*
2101 * If we need to do NFS logging, the filehandle
2102 * must be downsized to 32 bytes.
2103 */
2104 if (!error && exi->exi_export.ex_flags & EX_LOG) {
2105 i = 0;
2106 sz = sizeof (fsid_t);
2107 bcopy(&fh.fh3_fsid, &logbuf[i], sz);
2108 i += sz;
2109 sz = sizeof (ushort_t);
2110 bcopy(&fh.fh3_len, &logbuf[i], sz);
2111 i += sz;
2112 sz = NFS_FHMAXDATA;
2113 bcopy(fh.fh3_data, &logbuf[i], sz);
2114 i += sz;
2115 sz = sizeof (ushort_t);
2116 bcopy(&fh.fh3_xlen, &logbuf[i], sz);
2117 i += sz;
2118 sz = NFS_FHMAXDATA;
2119 bcopy(fh.fh3_xdata, &logbuf[i], sz);
2120 i += sz;
2121 }
2122 }
2123 if (!error && exi->exi_export.ex_flags & EX_LOG) {
2124 nfslog_getfh(exi, (fhandle_t *)logptr,
2125 STRUCT_FGETP(uap, fname), UIO_USERSPACE, cr);
2126 }
2127 exi_rele(exi);
2128 if (!error) {
2129 if (copyout(&l, STRUCT_FGETP(uap, lenp), sizeof (int)))
2130 error = EFAULT;
2131 if (copyout(buf, STRUCT_FGETP(uap, fhp), l))
2132 error = EFAULT;
2133 }
2134 }
2135 VN_RELE(vp);
2136 if (dvp != NULL) {
2137 VN_RELE(dvp);
2138 }
2139 return (error);
2140 }
2141
2142 /*
2143 * Strategy: if vp is in the export list, then
2144 * return the associated file handle. Otherwise, ".."
2145 * once up the vp and try again, until the root of the
2146 * filesystem is reached.
2147 */
2148 struct exportinfo *
2149 nfs_vptoexi(vnode_t *dvp, vnode_t *vp, cred_t *cr, int *walk,
2150 int *err, bool_t v4srv)
2151 {
2152 fid_t fid;
2153 int error;
2154 struct exportinfo *exi;
2155
2156 ASSERT(vp);
2157 VN_HOLD(vp);
2158 if (dvp != NULL) {
2159 VN_HOLD(dvp);
2160 }
2161 if (walk != NULL)
2162 *walk = 0;
2163
2164 for (;;) {
2165 bzero(&fid, sizeof (fid));
2166 fid.fid_len = MAXFIDSZ;
2167 error = vop_fid_pseudo(vp, &fid);
2168 if (error) {
2169 /*
2170 * If vop_fid_pseudo returns ENOSPC then the fid
2171 * supplied is too small. For now we simply
2172 * return EREMOTE.
2173 */
2174 if (error == ENOSPC)
2175 error = EREMOTE;
2176 break;
2177 }
2178
2179 if (v4srv)
2180 exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
2181 else
2182 exi = checkexport(&vp->v_vfsp->vfs_fsid, &fid);
2183
2184 if (exi != NULL) {
2185 /*
2186 * Found the export info
2187 */
2188 break;
2189 }
2190
2191 /*
2192 * We have just failed finding a matching export.
2193 * If we're at the root of this filesystem, then
2194 * it's time to stop (with failure).
2195 */
2196 ASSERT3P(vp->v_vfsp->vfs_zone, ==, curzone);
2197 if ((vp->v_flag & VROOT) || VN_IS_CURZONEROOT(vp)) {
2198 error = EINVAL;
2199 break;
2200 }
2201
2202 if (walk != NULL)
2203 (*walk)++;
2204
2205 /*
2206 * Now, do a ".." up vp. If dvp is supplied, use it,
2207 * otherwise, look it up.
2208 */
2209 if (dvp == NULL) {
2210 error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, cr,
2211 NULL, NULL, NULL);
2212 if (error)
2213 break;
2214 }
2215 VN_RELE(vp);
2216 vp = dvp;
2217 dvp = NULL;
2218 }
2219 VN_RELE(vp);
2220 if (dvp != NULL) {
2221 VN_RELE(dvp);
2222 }
2223 if (error != 0) {
2224 if (err != NULL)
2225 *err = error;
2226 return (NULL);
2227 }
2228 return (exi);
2229 }
2230
2231 int
2232 chk_clnt_sec(exportinfo_t *exi, struct svc_req *req)
2233 {
2234 int i, nfsflavor;
2235 struct secinfo *sp;
2236
2237 /*
2238 * Get the nfs flavor number from xprt.
2239 */
2240 nfsflavor = (int)(uintptr_t)req->rq_xprt->xp_cookie;
2241
2242 sp = exi->exi_export.ex_secinfo;
2243 for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2244 if ((nfsflavor == sp[i].s_secinfo.sc_nfsnum) &&
2245 SEC_REF_EXPORTED(sp + i))
2246 return (TRUE);
2247 }
2248 return (FALSE);
2249 }
2250
2251 /*
2252 * Make an fhandle from a vnode
2253 */
2254 int
2255 makefh(fhandle_t *fh, vnode_t *vp, exportinfo_t *exi)
2256 {
2257 int error;
2258
2259 *fh = exi->exi_fh; /* struct copy */
2260
2261 error = VOP_FID(vp, (fid_t *)&fh->fh_len, NULL);
2262 if (error) {
2263 /*
2264 * Should be something other than EREMOTE
2265 */
2266 return (EREMOTE);
2267 }
2268 return (0);
2269 }
2270
2271 /*
2272 * This routine makes an overloaded V2 fhandle which contains
2273 * sec modes.
2274 *
2275 * Note that the first four octets contain the length octet,
2276 * the status octet, and two padded octets to make them XDR
2277 * four-octet aligned.
2278 *
2279 * 1 2 3 4 32
2280 * +---+---+---+---+---+---+---+---+ +---+---+---+---+ +---+
2281 * | l | s | | | sec_1 |...| sec_n |...| |
2282 * +---+---+---+---+---+---+---+---+ +---+---+---+---+ +---+
2283 *
2284 * where
2285 *
2286 * the status octet s indicates whether there are more security
2287 * flavors (1 means yes, 0 means no) that require the client to
2288 * perform another 0x81 LOOKUP to get them,
2289 *
2290 * the length octet l is the length describing the number of
2291 * valid octets that follow. (l = 4 * n, where n is the number
2292 * of security flavors sent in the current overloaded filehandle.)
2293 *
2294 * sec_index should always be in the inclusive range: [1 - ex_seccnt],
2295 * and it tells server where to start within the secinfo array.
2296 * Usually it will always be 1; however, if more flavors are used
2297 * for the public export than can be encoded in the overloaded FH
2298 * (7 for NFS2), subsequent SNEGO MCLs will have a larger index
2299 * so the server will pick up where it left off from the previous
2300 * MCL reply.
2301 *
2302 * With NFS4 support, implicitly allowed flavors are also in
2303 * the secinfo array; however, they should not be returned in
2304 * SNEGO MCL replies.
2305 */
2306 int
2307 makefh_ol(fhandle_t *fh, exportinfo_t *exi, uint_t sec_index)
2308 {
2309 secinfo_t sec[MAX_FLAVORS];
2310 int totalcnt, i, *ipt, cnt, seccnt, secidx, fh_max_cnt;
2311 char *c;
2312
2313 if (fh == NULL || exi == NULL || sec_index < 1)
2314 return (EREMOTE);
2315
2316 /*
2317 * WebNFS clients need to know the unique set of explicitly
2318 * shared flavors in used for the public export. When
2319 * "TRUE" is passed to build_seclist_nodups(), only explicitly
2320 * shared flavors are included in the list.
2321 */
2322 seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2323 if (sec_index > seccnt)
2324 return (EREMOTE);
2325
2326 fh_max_cnt = (NFS_FHSIZE / sizeof (int)) - 1;
2327 totalcnt = seccnt - sec_index + 1;
2328 cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2329
2330 c = (char *)fh;
2331 /*
2332 * Encode the length octet representing the number of
2333 * security flavors (in bytes) in this overloaded fh.
2334 */
2335 *c = cnt * sizeof (int);
2336
2337 /*
2338 * Encode the status octet that indicates whether there
2339 * are more security flavors the client needs to get.
2340 */
2341 *(c + 1) = totalcnt > fh_max_cnt;
2342
2343 /*
2344 * put security flavors in the overloaded fh
2345 */
2346 ipt = (int *)(c + sizeof (int32_t));
2347 secidx = sec_index - 1;
2348 for (i = 0; i < cnt; i++) {
2349 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2350 }
2351 return (0);
2352 }
2353
2354 /*
2355 * Make an nfs_fh3 from a vnode
2356 */
2357 int
2358 makefh3(nfs_fh3 *fh, vnode_t *vp, struct exportinfo *exi)
2359 {
2360 int error;
2361 fid_t fid;
2362
2363 bzero(&fid, sizeof (fid));
2364 fid.fid_len = sizeof (fh->fh3_data);
2365 error = VOP_FID(vp, &fid, NULL);
2366 if (error)
2367 return (EREMOTE);
2368
2369 bzero(fh, sizeof (nfs_fh3));
2370 fh->fh3_fsid = exi->exi_fsid;
2371 fh->fh3_len = fid.fid_len;
2372 bcopy(fid.fid_data, fh->fh3_data, fh->fh3_len);
2373
2374 fh->fh3_xlen = exi->exi_fid.fid_len;
2375 ASSERT(fh->fh3_xlen <= sizeof (fh->fh3_xdata));
2376 bcopy(exi->exi_fid.fid_data, fh->fh3_xdata, fh->fh3_xlen);
2377
2378 fh->fh3_length = sizeof (fh->fh3_fsid)
2379 + sizeof (fh->fh3_len) + fh->fh3_len
2380 + sizeof (fh->fh3_xlen) + fh->fh3_xlen;
2381 fh->fh3_flags = 0;
2382
2383 return (0);
2384 }
2385
2386 /*
2387 * This routine makes an overloaded V3 fhandle which contains
2388 * sec modes.
2389 *
2390 * 1 4
2391 * +--+--+--+--+
2392 * | len |
2393 * +--+--+--+--+
2394 * up to 64
2395 * +--+--+--+--+--+--+--+--+--+--+--+--+ +--+--+--+--+
2396 * |s | | | | sec_1 | sec_2 | ... | sec_n |
2397 * +--+--+--+--+--+--+--+--+--+--+--+--+ +--+--+--+--+
2398 *
2399 * len = 4 * (n+1), where n is the number of security flavors
2400 * sent in the current overloaded filehandle.
2401 *
2402 * the status octet s indicates whether there are more security
2403 * mechanisms (1 means yes, 0 means no) that require the client
2404 * to perform another 0x81 LOOKUP to get them.
2405 *
2406 * Three octets are padded after the status octet.
2407 */
2408 int
2409 makefh3_ol(nfs_fh3 *fh, struct exportinfo *exi, uint_t sec_index)
2410 {
2411 secinfo_t sec[MAX_FLAVORS];
2412 int totalcnt, cnt, *ipt, i, seccnt, fh_max_cnt, secidx;
2413 char *c;
2414
2415 if (fh == NULL || exi == NULL || sec_index < 1)
2416 return (EREMOTE);
2417
2418 /*
2419 * WebNFS clients need to know the unique set of explicitly
2420 * shared flavors in used for the public export. When
2421 * "TRUE" is passed to build_seclist_nodups(), only explicitly
2422 * shared flavors are included in the list.
2423 */
2424 seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2425
2426 if (sec_index > seccnt)
2427 return (EREMOTE);
2428
2429 fh_max_cnt = (NFS3_FHSIZE / sizeof (int)) - 1;
2430 totalcnt = seccnt - sec_index + 1;
2431 cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2432
2433 /*
2434 * Place the length in fh3_length representing the number
2435 * of security flavors (in bytes) in this overloaded fh.
2436 */
2437 fh->fh3_flags = FH_WEBNFS;
2438 fh->fh3_length = (cnt+1) * sizeof (int32_t);
2439
2440 c = (char *)&fh->fh3_u.nfs_fh3_i.fh3_i;
2441 /*
2442 * Encode the status octet that indicates whether there
2443 * are more security flavors the client needs to get.
2444 */
2445 *c = totalcnt > fh_max_cnt;
2446
2447 /*
2448 * put security flavors in the overloaded fh
2449 */
2450 secidx = sec_index - 1;
2451 ipt = (int *)(c + sizeof (int32_t));
2452 for (i = 0; i < cnt; i++) {
2453 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2454 }
2455 return (0);
2456 }
2457
2458 /*
2459 * Make an nfs_fh4 from a vnode
2460 */
2461 int
2462 makefh4(nfs_fh4 *fh, vnode_t *vp, struct exportinfo *exi)
2463 {
2464 int error;
2465 nfs_fh4_fmt_t *fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2466 fid_t fid;
2467
2468 bzero(&fid, sizeof (fid));
2469 fid.fid_len = MAXFIDSZ;
2470 /*
2471 * vop_fid_pseudo() is used to set up NFSv4 namespace, so
2472 * use vop_fid_pseudo() here to get the fid instead of VOP_FID.
2473 */
2474 error = vop_fid_pseudo(vp, &fid);
2475 if (error)
2476 return (error);
2477
2478 fh->nfs_fh4_len = NFS_FH4_LEN;
2479
2480 fh_fmtp->fh4_i.fhx_fsid = exi->exi_fh.fh_fsid;
2481 fh_fmtp->fh4_i.fhx_xlen = exi->exi_fh.fh_xlen;
2482
2483 bzero(fh_fmtp->fh4_i.fhx_data, sizeof (fh_fmtp->fh4_i.fhx_data));
2484 bzero(fh_fmtp->fh4_i.fhx_xdata, sizeof (fh_fmtp->fh4_i.fhx_xdata));
2485 ASSERT(exi->exi_fh.fh_xlen <= sizeof (fh_fmtp->fh4_i.fhx_xdata));
2486 bcopy(exi->exi_fh.fh_xdata, fh_fmtp->fh4_i.fhx_xdata,
2487 exi->exi_fh.fh_xlen);
2488
2489 fh_fmtp->fh4_len = fid.fid_len;
2490 ASSERT(fid.fid_len <= sizeof (fh_fmtp->fh4_data));
2491 bcopy(fid.fid_data, fh_fmtp->fh4_data, fid.fid_len);
2492 fh_fmtp->fh4_flag = 0;
2493
2494 #ifdef VOLATILE_FH_TEST
2495 /*
2496 * XXX (temporary?)
2497 * Use the rnode volatile_id value to add volatility to the fh.
2498 *
2499 * For testing purposes there are currently two scenarios, based
2500 * on whether the filesystem was shared with "volatile_fh"
2501 * or "expire_on_rename". In the first case, use the value of
2502 * export struct share_time as the volatile_id. In the second
2503 * case use the vnode volatile_id value (which is set to the
2504 * time in which the file was renamed).
2505 *
2506 * Note that the above are temporary constructs for testing only
2507 * XXX
2508 */
2509 if (exi->exi_export.ex_flags & EX_VOLRNM) {
2510 fh_fmtp->fh4_volatile_id = find_volrnm_fh_id(exi, fh);
2511 } else if (exi->exi_export.ex_flags & EX_VOLFH) {
2512 fh_fmtp->fh4_volatile_id = exi->exi_volatile_id;
2513 } else {
2514 fh_fmtp->fh4_volatile_id = 0;
2515 }
2516 #endif /* VOLATILE_FH_TEST */
2517
2518 return (0);
2519 }
2520
2521 /*
2522 * Convert an fhandle into a vnode.
2523 * Uses the file id (fh_len + fh_data) in the fhandle to get the vnode.
2524 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2525 * are done with it.
2526 */
2527 vnode_t *
2528 nfs_fhtovp(fhandle_t *fh, struct exportinfo *exi)
2529 {
2530 vfs_t *vfsp;
2531 vnode_t *vp;
2532 int error;
2533 fid_t *fidp;
2534
2535 TRACE_0(TR_FAC_NFS, TR_FHTOVP_START,
2536 "fhtovp_start");
2537
2538 if (exi == NULL) {
2539 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2540 "fhtovp_end:(%S)", "exi NULL");
2541 return (NULL); /* not exported */
2542 }
2543
2544 ASSERT(exi->exi_vp != NULL);
2545
2546 if (PUBLIC_FH2(fh)) {
2547 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2548 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2549 "fhtovp_end:(%S)", "root not exported");
2550 return (NULL);
2551 }
2552 vp = exi->exi_vp;
2553 VN_HOLD(vp);
2554 return (vp);
2555 }
2556
2557 vfsp = exi->exi_vp->v_vfsp;
2558 ASSERT(vfsp != NULL);
2559 fidp = (fid_t *)&fh->fh_len;
2560
2561 error = VFS_VGET(vfsp, &vp, fidp);
2562 if (error || vp == NULL) {
2563 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2564 "fhtovp_end:(%S)", "VFS_GET failed or vp NULL");
2565 return (NULL);
2566 }
2567 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2568 "fhtovp_end:(%S)", "end");
2569 return (vp);
2570 }
2571
2572 /*
2573 * Convert an nfs_fh3 into a vnode.
2574 * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2575 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2576 * are done with it.
2577 */
2578 vnode_t *
2579 nfs3_fhtovp(nfs_fh3 *fh, struct exportinfo *exi)
2580 {
2581 vfs_t *vfsp;
2582 vnode_t *vp;
2583 int error;
2584 fid_t *fidp;
2585
2586 if (exi == NULL)
2587 return (NULL); /* not exported */
2588
2589 ASSERT(exi->exi_vp != NULL);
2590
2591 if (PUBLIC_FH3(fh)) {
2592 if (exi->exi_export.ex_flags & EX_PUBLIC)
2593 return (NULL);
2594 vp = exi->exi_vp;
2595 VN_HOLD(vp);
2596 return (vp);
2597 }
2598
2599 if (fh->fh3_length < NFS3_OLDFHSIZE ||
2600 fh->fh3_length > NFS3_MAXFHSIZE)
2601 return (NULL);
2602
2603 vfsp = exi->exi_vp->v_vfsp;
2604 ASSERT(vfsp != NULL);
2605 fidp = FH3TOFIDP(fh);
2606
2607 error = VFS_VGET(vfsp, &vp, fidp);
2608 if (error || vp == NULL)
2609 return (NULL);
2610
2611 return (vp);
2612 }
2613
2614 /*
2615 * Convert an nfs_fh4 into a vnode.
2616 * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2617 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2618 * are done with it.
2619 */
2620 vnode_t *
2621 nfs4_fhtovp(nfs_fh4 *fh, struct exportinfo *exi, nfsstat4 *statp)
2622 {
2623 vfs_t *vfsp;
2624 vnode_t *vp = NULL;
2625 int error;
2626 fid_t *fidp;
2627 nfs_fh4_fmt_t *fh_fmtp;
2628 #ifdef VOLATILE_FH_TEST
2629 uint32_t volatile_id = 0;
2630 #endif /* VOLATILE_FH_TEST */
2631
2632 if (exi == NULL) {
2633 *statp = NFS4ERR_STALE;
2634 return (NULL); /* not exported */
2635 }
2636 ASSERT(exi->exi_vp != NULL);
2637
2638 /* caller should have checked this */
2639 ASSERT(fh->nfs_fh4_len >= NFS_FH4_LEN);
2640
2641 fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2642 vfsp = exi->exi_vp->v_vfsp;
2643 ASSERT(vfsp != NULL);
2644 fidp = (fid_t *)&fh_fmtp->fh4_len;
2645
2646 #ifdef VOLATILE_FH_TEST
2647 /* XXX check if volatile - should be changed later */
2648 if (exi->exi_export.ex_flags & (EX_VOLRNM | EX_VOLFH)) {
2649 /*
2650 * Filesystem is shared with volatile filehandles
2651 */
2652 if (exi->exi_export.ex_flags & EX_VOLRNM)
2653 volatile_id = find_volrnm_fh_id(exi, fh);
2654 else
2655 volatile_id = exi->exi_volatile_id;
2656
2657 if (fh_fmtp->fh4_volatile_id != volatile_id) {
2658 *statp = NFS4ERR_FHEXPIRED;
2659 return (NULL);
2660 }
2661 }
2662 /*
2663 * XXX even if test_volatile_fh false, the fh may contain a
2664 * volatile id if obtained when the test was set.
2665 */
2666 fh_fmtp->fh4_volatile_id = (uchar_t)0;
2667 #endif /* VOLATILE_FH_TEST */
2668
2669 error = VFS_VGET(vfsp, &vp, fidp);
2670 /*
2671 * If we can not get vp from VFS_VGET, perhaps this is
2672 * an nfs v2/v3/v4 node in an nfsv4 pseudo filesystem.
2673 * Check it out.
2674 */
2675 if (error && PSEUDO(exi))
2676 error = nfs4_vget_pseudo(exi, &vp, fidp);
2677
2678 if (error || vp == NULL) {
2679 *statp = NFS4ERR_STALE;
2680 return (NULL);
2681 }
2682 /* XXX - disgusting hack */
2683 if (vp->v_type == VNON && vp->v_flag & V_XATTRDIR)
2684 vp->v_type = VDIR;
2685 *statp = NFS4_OK;
2686 return (vp);
2687 }
2688
2689 /*
2690 * Find the export structure associated with the given filesystem.
2691 * If found, then increment the ref count (exi_count).
2692 */
2693 struct exportinfo *
2694 checkexport(fsid_t *fsid, fid_t *fid)
2695 {
2696 struct exportinfo *exi;
2697 nfs_export_t *ne = nfs_get_export();
2698
2699 rw_enter(&ne->exported_lock, RW_READER);
2700 for (exi = ne->exptable[exptablehash(fsid, fid)];
2701 exi != NULL;
2702 exi = exi->fid_hash.next) {
2703 if (exportmatch(exi, fsid, fid)) {
2704 /*
2705 * If this is the place holder for the
2706 * public file handle, then return the
2707 * real export entry for the public file
2708 * handle.
2709 */
2710 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2711 exi = ne->exi_public;
2712 }
2713
2714 exi_hold(exi);
2715 rw_exit(&ne->exported_lock);
2716 return (exi);
2717 }
2718 }
2719 rw_exit(&ne->exported_lock);
2720 return (NULL);
2721 }
2722
2723
2724 /*
2725 * "old school" version of checkexport() for NFS4. NFS4
2726 * rfs4_compound holds exported_lock for duration of compound
2727 * processing. This version doesn't manipulate exi_count
2728 * since NFS4 breaks fundamental assumptions in the exi_count
2729 * design.
2730 */
2731 struct exportinfo *
2732 checkexport4(fsid_t *fsid, fid_t *fid, vnode_t *vp)
2733 {
2734 struct exportinfo *exi;
2735 nfs_export_t *ne = nfs_get_export();
2736
2737 ASSERT(RW_LOCK_HELD(&ne->exported_lock));
2738
2739 for (exi = ne->exptable[exptablehash(fsid, fid)];
2740 exi != NULL;
2741 exi = exi->fid_hash.next) {
2742 if (exportmatch(exi, fsid, fid)) {
2743 /*
2744 * If this is the place holder for the
2745 * public file handle, then return the
2746 * real export entry for the public file
2747 * handle.
2748 */
2749 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2750 exi = ne->exi_public;
2751 }
2752
2753 /*
2754 * If vp is given, check if vp is the
2755 * same vnode as the exported node.
2756 *
2757 * Since VOP_FID of a lofs node returns the
2758 * fid of its real node (ufs), the exported
2759 * node for lofs and (pseudo) ufs may have
2760 * the same fsid and fid.
2761 */
2762 if (vp == NULL || vp == exi->exi_vp)
2763 return (exi);
2764 }
2765 }
2766
2767 return (NULL);
2768 }
2769
2770 /*
2771 * Free an entire export list node
2772 */
2773 void
2774 exportfree(struct exportinfo *exi)
2775 {
2776 struct exportdata *ex;
2777 struct charset_cache *cache;
2778 int i;
2779
2780 ex = &exi->exi_export;
2781
2782 ASSERT(exi->exi_vp != NULL && !(exi->exi_export.ex_flags & EX_PUBLIC));
2783 VN_RELE(exi->exi_vp);
2784 if (exi->exi_dvp != NULL)
2785 VN_RELE(exi->exi_dvp);
2786
2787 if (ex->ex_flags & EX_INDEX)
2788 kmem_free(ex->ex_index, strlen(ex->ex_index) + 1);
2789
2790 kmem_free(ex->ex_path, ex->ex_pathlen + 1);
2791 nfsauth_cache_free(exi);
2792
2793 /*
2794 * if there is a character set mapping cached, clean it up.
2795 */
2796 for (cache = exi->exi_charset; cache != NULL;
2797 cache = exi->exi_charset) {
2798 if (cache->inbound != (kiconv_t)-1)
2799 (void) kiconv_close(cache->inbound);
2800 if (cache->outbound != (kiconv_t)-1)
2801 (void) kiconv_close(cache->outbound);
2802 exi->exi_charset = cache->next;
2803 kmem_free(cache, sizeof (struct charset_cache));
2804 }
2805
2806 if (exi->exi_logbuffer != NULL)
2807 nfslog_disable(exi);
2808
2809 if (ex->ex_flags & EX_LOG) {
2810 kmem_free(ex->ex_log_buffer, ex->ex_log_bufferlen + 1);
2811 kmem_free(ex->ex_tag, ex->ex_taglen + 1);
2812 }
2813
2814 if (exi->exi_visible)
2815 free_visible(exi->exi_visible);
2816
2817 srv_secinfo_list_free(ex->ex_secinfo, ex->ex_seccnt);
2818
2819 #ifdef VOLATILE_FH_TEST
2820 free_volrnm_list(exi);
2821 mutex_destroy(&exi->exi_vol_rename_lock);
2822 #endif /* VOLATILE_FH_TEST */
2823
2824 mutex_destroy(&exi->exi_lock);
2825 rw_destroy(&exi->exi_cache_lock);
2826 /*
2827 * All nodes in the exi_cache AVL trees were removed and freed in the
2828 * nfsauth_cache_free() call above. We will just destroy and free the
2829 * empty AVL trees here.
2830 */
2831 for (i = 0; i < AUTH_TABLESIZE; i++) {
2832 avl_destroy(exi->exi_cache[i]);
2833 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
2834 }
2835
2836 kmem_free(exi, sizeof (*exi));
2837 }
2838
2839 /*
2840 * load the index file from user space into kernel space.
2841 */
2842 static int
2843 loadindex(struct exportdata *kex)
2844 {
2845 int error;
2846 char index[MAXNAMELEN+1];
2847 size_t len;
2848
2849 /*
2850 * copyinstr copies the complete string including the NULL and
2851 * returns the len with the NULL byte included in the calculation
2852 * as long as the max length is not exceeded.
2853 */
2854 if (error = copyinstr(kex->ex_index, index, sizeof (index), &len))
2855 return (error);
2856
2857 kex->ex_index = kmem_alloc(len, KM_SLEEP);
2858 bcopy(index, kex->ex_index, len);
2859
2860 return (0);
2861 }
2862
2863 void
2864 exi_hold(struct exportinfo *exi)
2865 {
2866 mutex_enter(&exi->exi_lock);
2867 exi->exi_count++;
2868 mutex_exit(&exi->exi_lock);
2869 }
2870
2871 /*
2872 * When a thread completes using exi, it should call exi_rele().
2873 * exi_rele() decrements exi_count. It releases exi if exi_count == 0, i.e.
2874 * if this is the last user of exi and exi is not on exportinfo list anymore
2875 */
2876 void
2877 exi_rele(struct exportinfo *exi)
2878 {
2879 mutex_enter(&exi->exi_lock);
2880 exi->exi_count--;
2881 if (exi->exi_count == 0) {
2882 mutex_exit(&exi->exi_lock);
2883 exportfree(exi);
2884 } else
2885 mutex_exit(&exi->exi_lock);
2886 }
2887
2888 #ifdef VOLATILE_FH_TEST
2889 /*
2890 * Test for volatile fh's - add file handle to list and set its volatile id
2891 * to time it was renamed. If EX_VOLFH is also on and the fs is reshared,
2892 * the vol_rename queue is purged.
2893 *
2894 * XXX This code is for unit testing purposes only... To correctly use it, it
2895 * needs to tie a rename list to the export struct and (more
2896 * important), protect access to the exi rename list using a write lock.
2897 */
2898
2899 /*
2900 * get the fh vol record if it's in the volatile on rename list. Don't check
2901 * volatile_id in the file handle - compare only the file handles.
2902 */
2903 static struct ex_vol_rename *
2904 find_volrnm_fh(struct exportinfo *exi, nfs_fh4 *fh4p)
2905 {
2906 struct ex_vol_rename *p = NULL;
2907 fhandle4_t *fhp;
2908
2909 /* XXX shouldn't we assert &exported_lock held? */
2910 ASSERT(MUTEX_HELD(&exi->exi_vol_rename_lock));
2911
2912 if (fh4p->nfs_fh4_len != NFS_FH4_LEN) {
2913 return (NULL);
2914 }
2915 fhp = &((nfs_fh4_fmt_t *)fh4p->nfs_fh4_val)->fh4_i;
2916 for (p = exi->exi_vol_rename; p != NULL; p = p->vrn_next) {
2917 if (bcmp(fhp, &p->vrn_fh_fmt.fh4_i,
2918 sizeof (fhandle4_t)) == 0)
2919 break;
2920 }
2921 return (p);
2922 }
2923
2924 /*
2925 * get the volatile id for the fh (if there is - else return 0). Ignore the
2926 * volatile_id in the file handle - compare only the file handles.
2927 */
2928 static uint32_t
2929 find_volrnm_fh_id(struct exportinfo *exi, nfs_fh4 *fh4p)
2930 {
2931 struct ex_vol_rename *p;
2932 uint32_t volatile_id;
2933
2934 mutex_enter(&exi->exi_vol_rename_lock);
2935 p = find_volrnm_fh(exi, fh4p);
2936 volatile_id = (p ? p->vrn_fh_fmt.fh4_volatile_id :
2937 exi->exi_volatile_id);
2938 mutex_exit(&exi->exi_vol_rename_lock);
2939 return (volatile_id);
2940 }
2941
2942 /*
2943 * Free the volatile on rename list - will be called if a filesystem is
2944 * unshared or reshared without EX_VOLRNM
2945 */
2946 static void
2947 free_volrnm_list(struct exportinfo *exi)
2948 {
2949 struct ex_vol_rename *p, *pnext;
2950
2951 /* no need to hold mutex lock - this one is called from exportfree */
2952 for (p = exi->exi_vol_rename; p != NULL; p = pnext) {
2953 pnext = p->vrn_next;
2954 kmem_free(p, sizeof (*p));
2955 }
2956 exi->exi_vol_rename = NULL;
2957 }
2958
2959 /*
2960 * Add a file handle to the volatile on rename list.
2961 */
2962 void
2963 add_volrnm_fh(struct exportinfo *exi, vnode_t *vp)
2964 {
2965 struct ex_vol_rename *p;
2966 char fhbuf[NFS4_FHSIZE];
2967 nfs_fh4 fh4;
2968 int error;
2969
2970 fh4.nfs_fh4_val = fhbuf;
2971 error = makefh4(&fh4, vp, exi);
2972 if ((error) || (fh4.nfs_fh4_len != sizeof (p->vrn_fh_fmt))) {
2973 return;
2974 }
2975
2976 mutex_enter(&exi->exi_vol_rename_lock);
2977
2978 p = find_volrnm_fh(exi, &fh4);
2979
2980 if (p == NULL) {
2981 p = kmem_alloc(sizeof (*p), KM_SLEEP);
2982 bcopy(fh4.nfs_fh4_val, &p->vrn_fh_fmt, sizeof (p->vrn_fh_fmt));
2983 p->vrn_next = exi->exi_vol_rename;
2984 exi->exi_vol_rename = p;
2985 }
2986
2987 p->vrn_fh_fmt.fh4_volatile_id = gethrestime_sec();
2988 mutex_exit(&exi->exi_vol_rename_lock);
2989 }
2990
2991 #endif /* VOLATILE_FH_TEST */