Print this page
Hyperaggressive asserts pt 1/N
Ooops exi_zoneid isn't a variable again yet
Be far more judicious in the use of curzone-using macros.
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/nfs/nfs_export.c
+++ new/usr/src/uts/common/fs/nfs/nfs_export.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 */
25 25
26 26 /*
27 27 * Copyright 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T.
28 28 * All rights reserved.
29 29 */
30 30
31 31 /*
32 32 * Copyright 2018 Nexenta Systems, Inc.
33 33 */
34 34
35 35 #include <sys/types.h>
36 36 #include <sys/param.h>
37 37 #include <sys/time.h>
38 38 #include <sys/vfs.h>
39 39 #include <sys/vnode.h>
40 40 #include <sys/socket.h>
41 41 #include <sys/errno.h>
42 42 #include <sys/uio.h>
43 43 #include <sys/proc.h>
44 44 #include <sys/user.h>
45 45 #include <sys/file.h>
46 46 #include <sys/tiuser.h>
47 47 #include <sys/kmem.h>
48 48 #include <sys/pathname.h>
49 49 #include <sys/debug.h>
50 50 #include <sys/vtrace.h>
51 51 #include <sys/cmn_err.h>
52 52 #include <sys/acl.h>
53 53 #include <sys/utsname.h>
54 54 #include <sys/sdt.h>
55 55 #include <netinet/in.h>
56 56 #include <sys/avl.h>
57 57
58 58 #include <rpc/types.h>
59 59 #include <rpc/auth.h>
60 60 #include <rpc/svc.h>
61 61
62 62 #include <nfs/nfs.h>
63 63 #include <nfs/export.h>
64 64 #include <nfs/nfssys.h>
65 65 #include <nfs/nfs_clnt.h>
66 66 #include <nfs/nfs_acl.h>
67 67 #include <nfs/nfs_log.h>
68 68 #include <nfs/lm.h>
69 69 #include <sys/sunddi.h>
70 70
71 71 /*
72 72 * exi_id support
73 73 *
74 74 * exi_id_next The next exi_id available.
75 75 * exi_id_overflow The exi_id_next already overflowed, so we should
76 76 * thoroughly check for duplicates.
77 77 * exi_id_tree AVL tree indexed by exi_id.
78 78 * nfs_exi_id_lock Lock to protect the export ID list
79 79 *
80 80 * All exi_id_next, exi_id_overflow, and exi_id_tree are protected by
81 81 * nfs_exi_id_lock.
82 82 */
83 83 static int exi_id_next;
84 84 static bool_t exi_id_overflow;
85 85 avl_tree_t exi_id_tree;
86 86 kmutex_t nfs_exi_id_lock;
87 87
88 88 static int unexport(nfs_export_t *, exportinfo_t *);
89 89 static void exportfree(exportinfo_t *);
90 90 static int loadindex(exportdata_t *);
91 91
92 92 extern void nfsauth_cache_free(exportinfo_t *);
93 93 extern int sec_svc_loadrootnames(int, int, caddr_t **, model_t);
94 94 extern void sec_svc_freerootnames(int, int, caddr_t *);
95 95
96 96 static int build_seclist_nodups(exportdata_t *, secinfo_t *, int);
97 97 static void srv_secinfo_add(secinfo_t **, int *, secinfo_t *, int, int);
98 98 static void srv_secinfo_remove(secinfo_t **, int *, secinfo_t *, int);
99 99 static void srv_secinfo_treeclimb(nfs_export_t *, exportinfo_t *,
100 100 secinfo_t *, int, bool_t);
101 101
102 102 #ifdef VOLATILE_FH_TEST
103 103 static struct ex_vol_rename *find_volrnm_fh(exportinfo_t *, nfs_fh4 *);
104 104 static uint32_t find_volrnm_fh_id(exportinfo_t *, nfs_fh4 *);
105 105 static void free_volrnm_list(exportinfo_t *);
106 106 #endif /* VOLATILE_FH_TEST */
107 107
108 108 fhandle_t nullfh2; /* for comparing V2 filehandles */
109 109
110 110 /*
111 111 * macro for static dtrace probes to trace server namespace ref count mods.
112 112 */
113 113 #define SECREF_TRACE(seclist, tag, flav, aftcnt) \
114 114 DTRACE_PROBE4(nfss__i__nmspc__secref, struct secinfo *, (seclist), \
115 115 char *, (tag), int, (int)(flav), int, (int)(aftcnt))
116 116
117 117
118 118 #define exptablehash(fsid, fid) (nfs_fhhash((fsid), (fid)) & (EXPTABLESIZE - 1))
119 119
120 120 extern nfs_export_t *
121 121 nfs_get_export(void)
122 122 {
123 123 nfs_globals_t *ng = zone_getspecific(nfssrv_zone_key, curzone);
124 124 nfs_export_t *ne = ng->nfs_export;
125 125 ASSERT(ne != NULL);
126 126 return (ne);
127 127 }
128 128
129 129 static uint8_t
130 130 xor_hash(uint8_t *data, int len)
131 131 {
132 132 uint8_t h = 0;
133 133
134 134 while (len--)
135 135 h ^= *data++;
136 136
137 137 return (h);
138 138 }
139 139
140 140 /*
141 141 * File handle hash function, XOR over all bytes in fsid and fid.
142 142 */
143 143 static unsigned
144 144 nfs_fhhash(fsid_t *fsid, fid_t *fid)
145 145 {
146 146 int len;
147 147 uint8_t h;
148 148
149 149 h = xor_hash((uint8_t *)fsid, sizeof (fsid_t));
150 150
151 151 /*
152 152 * Sanity check the length before using it
153 153 * blindly in case the client trashed it.
154 154 */
155 155 len = fid->fid_len > NFS_FH4MAXDATA ? 0 : fid->fid_len;
156 156 h ^= xor_hash((uint8_t *)fid->fid_data, len);
157 157
158 158 return ((unsigned)h);
159 159 }
160 160
161 161 /*
162 162 * Free the memory allocated within a secinfo entry.
163 163 */
164 164 void
165 165 srv_secinfo_entry_free(struct secinfo *secp)
166 166 {
167 167 if (secp->s_rootcnt > 0 && secp->s_rootnames != NULL) {
168 168 sec_svc_freerootnames(secp->s_secinfo.sc_rpcnum,
169 169 secp->s_rootcnt, secp->s_rootnames);
170 170 secp->s_rootcnt = 0;
171 171 }
172 172
173 173 if ((secp->s_secinfo.sc_rpcnum == RPCSEC_GSS) &&
174 174 (secp->s_secinfo.sc_gss_mech_type)) {
175 175 kmem_free(secp->s_secinfo.sc_gss_mech_type->elements,
176 176 secp->s_secinfo.sc_gss_mech_type->length);
177 177 kmem_free(secp->s_secinfo.sc_gss_mech_type,
178 178 sizeof (rpc_gss_OID_desc));
179 179 secp->s_secinfo.sc_gss_mech_type = NULL;
180 180 }
181 181 }
182 182
183 183 /*
184 184 * Free a list of secinfo allocated in the exportdata structure.
185 185 */
186 186 void
187 187 srv_secinfo_list_free(struct secinfo *secinfo, int cnt)
188 188 {
189 189 int i;
190 190
191 191 if (cnt == 0)
192 192 return;
193 193
194 194 for (i = 0; i < cnt; i++)
195 195 srv_secinfo_entry_free(&secinfo[i]);
196 196
197 197 kmem_free(secinfo, cnt * sizeof (struct secinfo));
198 198 }
199 199
200 200 /*
201 201 * Allocate and copy a secinfo data from "from" to "to".
202 202 *
203 203 * This routine is used by srv_secinfo_add() to add a new flavor to an
204 204 * ancestor's export node. The rootnames are not copied because the
205 205 * allowable rootname access only applies to the explicit exported node,
206 206 * not its ancestor's.
207 207 *
208 208 * "to" should have already been allocated and zeroed before calling
209 209 * this routine.
210 210 *
211 211 * This routine is used under the protection of exported_lock (RW_WRITER).
212 212 */
213 213 void
214 214 srv_secinfo_copy(struct secinfo *from, struct secinfo *to)
215 215 {
216 216 to->s_secinfo.sc_nfsnum = from->s_secinfo.sc_nfsnum;
217 217 to->s_secinfo.sc_rpcnum = from->s_secinfo.sc_rpcnum;
218 218
219 219 if (from->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
220 220 to->s_secinfo.sc_service = from->s_secinfo.sc_service;
221 221 bcopy(from->s_secinfo.sc_name, to->s_secinfo.sc_name,
222 222 strlen(from->s_secinfo.sc_name));
223 223 bcopy(from->s_secinfo.sc_gss_mech, to->s_secinfo.sc_gss_mech,
224 224 strlen(from->s_secinfo.sc_gss_mech));
225 225
226 226 /* copy mechanism oid */
227 227 to->s_secinfo.sc_gss_mech_type =
228 228 kmem_alloc(sizeof (rpc_gss_OID_desc), KM_SLEEP);
229 229 to->s_secinfo.sc_gss_mech_type->length =
230 230 from->s_secinfo.sc_gss_mech_type->length;
231 231 to->s_secinfo.sc_gss_mech_type->elements =
232 232 kmem_alloc(from->s_secinfo.sc_gss_mech_type->length,
233 233 KM_SLEEP);
234 234 bcopy(from->s_secinfo.sc_gss_mech_type->elements,
235 235 to->s_secinfo.sc_gss_mech_type->elements,
236 236 from->s_secinfo.sc_gss_mech_type->length);
237 237 }
238 238
239 239 to->s_refcnt = from->s_refcnt;
240 240 to->s_window = from->s_window;
241 241 /* no need to copy the mode bits - s_flags */
242 242 }
243 243
244 244 /*
245 245 * Create a secinfo array without duplicates. The condensed
246 246 * flavor list is used to propagate flavor ref counts to an
247 247 * export's ancestor pseudonodes.
248 248 */
249 249 static int
250 250 build_seclist_nodups(exportdata_t *exd, secinfo_t *nodups, int exponly)
251 251 {
252 252 int ccnt, c;
253 253 int ncnt, n;
254 254 struct secinfo *cursec;
255 255
256 256 ncnt = 0;
257 257 ccnt = exd->ex_seccnt;
258 258 cursec = exd->ex_secinfo;
259 259
260 260 for (c = 0; c < ccnt; c++) {
261 261
262 262 if (exponly && ! SEC_REF_EXPORTED(&cursec[c]))
263 263 continue;
264 264
265 265 for (n = 0; n < ncnt; n++) {
266 266 if (nodups[n].s_secinfo.sc_nfsnum ==
267 267 cursec[c].s_secinfo.sc_nfsnum)
268 268 break;
269 269 }
270 270
271 271 /*
272 272 * The structure copy below also copys ptrs embedded
273 273 * within struct secinfo. The ptrs are copied but
274 274 * they are never freed from the nodups array. If
275 275 * an ancestor's secinfo array doesn't contain one
276 276 * of the nodups flavors, then the entry is properly
277 277 * copied into the ancestor's secinfo array.
278 278 * (see srv_secinfo_copy)
279 279 */
280 280 if (n == ncnt) {
281 281 nodups[n] = cursec[c];
282 282 ncnt++;
283 283 }
284 284 }
285 285 return (ncnt);
286 286 }
287 287
288 288 /*
289 289 * Add the new security flavors from newdata to the current list, pcursec.
290 290 * Upon return, *pcursec has the newly merged secinfo list.
291 291 *
292 292 * There should be at least 1 secinfo entry in newsec.
293 293 *
294 294 * This routine is used under the protection of exported_lock (RW_WRITER).
295 295 */
296 296 static void
297 297 srv_secinfo_add(secinfo_t **pcursec, int *pcurcnt, secinfo_t *newsec,
298 298 int newcnt, int is_pseudo)
299 299 {
300 300 int ccnt, c; /* sec count in current data - curdata */
301 301 int n; /* index for newsec - newsecinfo */
302 302 int tcnt; /* total sec count after merge */
303 303 int mcnt; /* total sec count after merge */
304 304 struct secinfo *msec; /* merged secinfo list */
305 305 struct secinfo *cursec;
306 306
307 307 cursec = *pcursec;
308 308 ccnt = *pcurcnt;
309 309
310 310 ASSERT(newcnt > 0);
311 311 tcnt = ccnt + newcnt;
312 312
313 313 for (n = 0; n < newcnt; n++) {
314 314 for (c = 0; c < ccnt; c++) {
315 315 if (newsec[n].s_secinfo.sc_nfsnum ==
316 316 cursec[c].s_secinfo.sc_nfsnum) {
317 317 cursec[c].s_refcnt += newsec[n].s_refcnt;
318 318 SECREF_TRACE(cursec, "add_ref",
319 319 cursec[c].s_secinfo.sc_nfsnum,
320 320 cursec[c].s_refcnt);
321 321 tcnt--;
322 322 break;
323 323 }
324 324 }
325 325 }
326 326
327 327 if (tcnt == ccnt)
328 328 return; /* no change; no new flavors */
329 329
330 330 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
331 331
332 332 /* move current secinfo list data to the new list */
333 333 for (c = 0; c < ccnt; c++)
334 334 msec[c] = cursec[c];
335 335
336 336 /* Add the flavor that's not in the current data */
337 337 mcnt = ccnt;
338 338 for (n = 0; n < newcnt; n++) {
339 339 for (c = 0; c < ccnt; c++) {
340 340 if (newsec[n].s_secinfo.sc_nfsnum ==
341 341 cursec[c].s_secinfo.sc_nfsnum)
342 342 break;
343 343 }
344 344
345 345 /* This is the one. Add it. */
346 346 if (c == ccnt) {
347 347 srv_secinfo_copy(&newsec[n], &msec[mcnt]);
348 348
349 349 if (is_pseudo)
350 350 msec[mcnt].s_flags = M_RO;
351 351
352 352 SECREF_TRACE(msec, "new_ref",
353 353 msec[mcnt].s_secinfo.sc_nfsnum,
354 354 msec[mcnt].s_refcnt);
355 355 mcnt++;
356 356 }
357 357 }
358 358
359 359 ASSERT(mcnt == tcnt);
360 360
361 361 /*
362 362 * Done. Update curdata. Free the old secinfo list in
363 363 * curdata and return the new sec array info
364 364 */
365 365 if (ccnt > 0)
366 366 kmem_free(cursec, ccnt * sizeof (struct secinfo));
367 367 *pcurcnt = tcnt;
368 368 *pcursec = msec;
369 369 }
370 370
371 371 /*
372 372 * For NFS V4.
373 373 * Remove the security data of the unexported node from its ancestors.
374 374 * Assume there is at least one flavor entry in the current sec list
375 375 * (pcursec).
376 376 *
377 377 * This routine is used under the protection of exported_lock (RW_WRITER).
378 378 *
379 379 * Every element of remsec is an explicitly exported flavor. If
380 380 * srv_secinfo_remove() is called fom an exportfs error path, then
381 381 * the flavor list was derived from the user's share cmdline,
382 382 * and all flavors are explicit. If it was called from the unshare path,
383 383 * build_seclist_nodups() was called with the exponly flag.
384 384 */
385 385 static void
386 386 srv_secinfo_remove(secinfo_t **pcursec, int *pcurcnt, secinfo_t *remsec,
387 387 int remcnt)
388 388 {
389 389 int ccnt, c; /* sec count in current data - cursec */
390 390 int r; /* sec count in removal data - remsec */
391 391 int tcnt, mcnt; /* total sec count after removing */
392 392 struct secinfo *msec; /* final secinfo list after removing */
393 393 struct secinfo *cursec;
394 394
395 395 cursec = *pcursec;
396 396 ccnt = *pcurcnt;
397 397 tcnt = ccnt;
398 398
399 399 for (r = 0; r < remcnt; r++) {
400 400 /*
401 401 * At unshare/reshare time, only explicitly shared flavor ref
402 402 * counts are decremented and propagated to ancestors.
403 403 * Implicit flavor refs came from shared descendants, and
404 404 * they must be kept.
405 405 */
406 406 if (! SEC_REF_EXPORTED(&remsec[r]))
407 407 continue;
408 408
409 409 for (c = 0; c < ccnt; c++) {
410 410 if (remsec[r].s_secinfo.sc_nfsnum ==
411 411 cursec[c].s_secinfo.sc_nfsnum) {
412 412
413 413 /*
414 414 * Decrement secinfo reference count by 1.
415 415 * If this entry is invalid after decrementing
416 416 * the count (i.e. count < 1), this entry will
417 417 * be removed.
418 418 */
419 419 cursec[c].s_refcnt--;
420 420
421 421 SECREF_TRACE(cursec, "del_ref",
422 422 cursec[c].s_secinfo.sc_nfsnum,
423 423 cursec[c].s_refcnt);
424 424
425 425 ASSERT(cursec[c].s_refcnt >= 0);
426 426
427 427 if (SEC_REF_INVALID(&cursec[c]))
428 428 tcnt--;
429 429 break;
430 430 }
431 431 }
432 432 }
433 433
434 434 ASSERT(tcnt >= 0);
435 435 if (tcnt == ccnt)
436 436 return; /* no change; no flavors to remove */
437 437
438 438 if (tcnt == 0) {
439 439 srv_secinfo_list_free(cursec, ccnt);
440 440 *pcurcnt = 0;
441 441 *pcursec = NULL;
442 442 return;
443 443 }
444 444
445 445 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
446 446
447 447 /* walk thru the given secinfo list to remove the flavors */
448 448 mcnt = 0;
449 449 for (c = 0; c < ccnt; c++) {
450 450 if (SEC_REF_INVALID(&cursec[c])) {
451 451 srv_secinfo_entry_free(&cursec[c]);
452 452 } else {
453 453 msec[mcnt] = cursec[c];
454 454 mcnt++;
455 455 }
456 456 }
457 457
458 458 ASSERT(mcnt == tcnt);
459 459 /*
460 460 * Done. Update curdata.
461 461 * Free the existing secinfo list in curdata. All pointers
462 462 * within the list have either been moved to msec or freed
463 463 * if it's invalid.
464 464 */
465 465 kmem_free(*pcursec, ccnt * sizeof (struct secinfo));
466 466 *pcursec = msec;
467 467 *pcurcnt = tcnt;
468 468 }
469 469
470 470
471 471 /*
472 472 * For the reshare case, sec flavor accounting happens in 3 steps:
473 473 * 1) propagate addition of new flavor refs up the ancestor tree
474 474 * 2) transfer flavor refs of descendants to new/reshared exportdata
475 475 * 3) propagate removal of old flavor refs up the ancestor tree
476 476 *
477 477 * srv_secinfo_exp2exp() implements step 2 of a reshare. At this point,
478 478 * the new flavor list has already been propagated up through the
479 479 * ancestor tree via srv_secinfo_treeclimb().
480 480 *
481 481 * If there is more than 1 export reference to an old flavor (i.e. some
482 482 * of its children shared with this flavor), this flavor information
483 483 * needs to be transferred to the new exportdata struct. A flavor in
484 484 * the old exportdata has descendant refs when its s_refcnt > 1 or it
485 485 * is implicitly shared (M_SEC4_EXPORTED not set in s_flags).
486 486 *
487 487 * SEC_REF_EXPORTED() is only true when M_SEC4_EXPORTED is set
488 488 * SEC_REF_SELF() is only true when both M_SEC4_EXPORTED is set and s_refcnt==1
489 489 *
490 490 * Transferring descendant flavor refcnts happens in 2 passes:
491 491 * a) flavors used before (oldsecinfo) and after (curdata->ex_secinfo) reshare
492 492 * b) flavors used before but not after reshare
493 493 *
494 494 * This routine is used under the protection of exported_lock (RW_WRITER).
495 495 */
496 496 void
497 497 srv_secinfo_exp2exp(exportdata_t *curdata, secinfo_t *oldsecinfo, int ocnt)
498 498 {
499 499 int ccnt, c; /* sec count in current data - curdata */
500 500 int o; /* sec count in old data - oldsecinfo */
501 501 int tcnt, mcnt; /* total sec count after the transfer */
502 502 struct secinfo *msec; /* merged secinfo list */
503 503
504 504 ccnt = curdata->ex_seccnt;
505 505
506 506 ASSERT(ocnt > 0);
507 507 ASSERT(!(curdata->ex_flags & EX_PSEUDO));
508 508
509 509 /*
510 510 * If the oldsecinfo has flavors with more than 1 reference count
511 511 * and the flavor is specified in the reshare, transfer the flavor
512 512 * refs to the new seclist (curdata.ex_secinfo).
513 513 */
514 514 tcnt = ccnt + ocnt;
515 515
516 516 for (o = 0; o < ocnt; o++) {
517 517
518 518 if (SEC_REF_SELF(&oldsecinfo[o])) {
519 519 tcnt--;
520 520 continue;
521 521 }
522 522
523 523 for (c = 0; c < ccnt; c++) {
524 524 if (oldsecinfo[o].s_secinfo.sc_nfsnum ==
525 525 curdata->ex_secinfo[c].s_secinfo.sc_nfsnum) {
526 526
527 527 /*
528 528 * add old reference to the current
529 529 * secinfo count
530 530 */
531 531 curdata->ex_secinfo[c].s_refcnt +=
532 532 oldsecinfo[o].s_refcnt;
533 533
534 534 /*
535 535 * Delete the old export flavor
536 536 * reference. The initial reference
537 537 * was created during srv_secinfo_add,
538 538 * and the count is decremented below
539 539 * to account for the initial reference.
540 540 */
541 541 if (SEC_REF_EXPORTED(&oldsecinfo[o]))
542 542 curdata->ex_secinfo[c].s_refcnt--;
543 543
544 544 SECREF_TRACE(curdata->ex_path,
545 545 "reshare_xfer_common_child_refs",
546 546 curdata->ex_secinfo[c].s_secinfo.sc_nfsnum,
547 547 curdata->ex_secinfo[c].s_refcnt);
548 548
549 549 ASSERT(curdata->ex_secinfo[c].s_refcnt >= 0);
550 550
551 551 tcnt--;
552 552 break;
553 553 }
554 554 }
555 555 }
556 556
557 557 if (tcnt == ccnt)
558 558 return; /* no more transfer to do */
559 559
560 560 /*
561 561 * oldsecinfo has flavors referenced by its children that are not
562 562 * in the current (new) export flavor list. Add these flavors.
563 563 */
564 564 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
565 565
566 566 /* move current secinfo list data to the new list */
567 567 for (c = 0; c < ccnt; c++)
568 568 msec[c] = curdata->ex_secinfo[c];
569 569
570 570 /*
571 571 * Add the flavor that's not in the new export, but still
572 572 * referenced by its children.
573 573 */
574 574 mcnt = ccnt;
575 575 for (o = 0; o < ocnt; o++) {
576 576 if (! SEC_REF_SELF(&oldsecinfo[o])) {
577 577 for (c = 0; c < ccnt; c++) {
578 578 if (oldsecinfo[o].s_secinfo.sc_nfsnum ==
579 579 curdata->ex_secinfo[c].s_secinfo.sc_nfsnum)
580 580 break;
581 581 }
582 582
583 583 /*
584 584 * This is the one. Add it. Decrement the ref count
585 585 * by 1 if the flavor is an explicitly shared flavor
586 586 * for the oldsecinfo export node.
587 587 */
588 588 if (c == ccnt) {
589 589 srv_secinfo_copy(&oldsecinfo[o], &msec[mcnt]);
590 590 if (SEC_REF_EXPORTED(&oldsecinfo[o]))
591 591 msec[mcnt].s_refcnt--;
592 592
593 593 SECREF_TRACE(curdata,
594 594 "reshare_xfer_implicit_child_refs",
595 595 msec[mcnt].s_secinfo.sc_nfsnum,
596 596 msec[mcnt].s_refcnt);
597 597
598 598 ASSERT(msec[mcnt].s_refcnt >= 0);
599 599 mcnt++;
600 600 }
601 601 }
602 602 }
603 603
604 604 ASSERT(mcnt == tcnt);
605 605 /*
606 606 * Done. Update curdata, free the existing secinfo list in
607 607 * curdata and set the new value.
608 608 */
609 609 if (ccnt > 0)
610 610 kmem_free(curdata->ex_secinfo, ccnt * sizeof (struct secinfo));
611 611 curdata->ex_seccnt = tcnt;
612 612 curdata->ex_secinfo = msec;
613 613 }
614 614
615 615 /*
616 616 * When unsharing an old export node and the old node becomes a pseudo node,
617 617 * if there is more than 1 export reference to an old flavor (i.e. some of
618 618 * its children shared with this flavor), this flavor information needs to
619 619 * be transferred to the new shared node.
620 620 *
621 621 * This routine is used under the protection of exported_lock (RW_WRITER).
622 622 */
623 623 void
624 624 srv_secinfo_exp2pseu(exportdata_t *curdata, exportdata_t *olddata)
625 625 {
626 626 int ocnt, o; /* sec count in transfer data - trandata */
627 627 int tcnt, mcnt; /* total sec count after transfer */
628 628 struct secinfo *msec; /* merged secinfo list */
629 629
630 630 ASSERT(curdata->ex_flags & EX_PSEUDO);
631 631 ASSERT(curdata->ex_seccnt == 0);
632 632
633 633 ocnt = olddata->ex_seccnt;
634 634
635 635 /*
636 636 * If the olddata has flavors with more than 1 reference count,
637 637 * transfer the information to the curdata.
638 638 */
639 639 tcnt = ocnt;
640 640
641 641 for (o = 0; o < ocnt; o++) {
642 642 if (SEC_REF_SELF(&olddata->ex_secinfo[o]))
643 643 tcnt--;
644 644 }
645 645
646 646 if (tcnt == 0)
647 647 return; /* no transfer to do */
648 648
649 649 msec = kmem_zalloc(tcnt * sizeof (struct secinfo), KM_SLEEP);
650 650
651 651 mcnt = 0;
652 652 for (o = 0; o < ocnt; o++) {
653 653 if (! SEC_REF_SELF(&olddata->ex_secinfo[o])) {
654 654
655 655 /*
656 656 * Decrement the reference count by 1 if the flavor is
657 657 * an explicitly shared flavor for the olddata export
658 658 * node.
659 659 */
660 660 srv_secinfo_copy(&olddata->ex_secinfo[o], &msec[mcnt]);
661 661 msec[mcnt].s_flags = M_RO;
662 662 if (SEC_REF_EXPORTED(&olddata->ex_secinfo[o]))
663 663 msec[mcnt].s_refcnt--;
664 664
665 665 SECREF_TRACE(curdata, "unshare_morph_pseudo",
666 666 msec[mcnt].s_secinfo.sc_nfsnum,
667 667 msec[mcnt].s_refcnt);
668 668
669 669 ASSERT(msec[mcnt].s_refcnt >= 0);
670 670 mcnt++;
671 671 }
672 672 }
673 673
674 674 ASSERT(mcnt == tcnt);
675 675 /*
676 676 * Done. Update curdata.
677 677 * Free up the existing secinfo list in curdata and
678 678 * set the new value.
679 679 */
680 680 curdata->ex_seccnt = tcnt;
681 681 curdata->ex_secinfo = msec;
682 682 }
683 683
684 684 /*
685 685 * Find for given treenode the exportinfo which has its
686 686 * exp_visible linked on its exi_visible list.
687 687 *
688 688 * Note: We could add new pointer either to treenode or
689 689 * to exp_visible, which will point there directly.
690 690 * This would buy some speed for some memory.
691 691 */
692 692 exportinfo_t *
693 693 vis2exi(treenode_t *tnode)
694 694 {
695 695 exportinfo_t *exi_ret = NULL;
696 696 #ifdef DEBUG
697 697 zone_t *zone = NULL;
698 698 #endif
699 699
700 700 for (;;) {
701 701 tnode = tnode->tree_parent;
702 702 #ifdef DEBUG
703 703 if (zone == NULL && tnode->tree_exi != NULL)
704 704 zone = tnode->tree_exi->exi_zone;
705 705 #endif
706 706 if (TREE_ROOT(tnode)) {
707 707 ASSERT3P(zone, ==, tnode->tree_exi->exi_zone);
708 708 exi_ret = tnode->tree_exi;
709 709 break;
710 710 }
711 711 }
|
↓ open down ↓ |
711 lines elided |
↑ open up ↑ |
712 712
713 713 ASSERT(exi_ret); /* Every visible should have its home exportinfo */
714 714 return (exi_ret);
715 715 }
716 716
717 717 /*
718 718 * For NFS V4.
719 719 * Add or remove the newly exported or unexported security flavors of the
720 720 * given exportinfo from its ancestors upto the system root.
721 721 */
722 -void
722 +static void
723 723 srv_secinfo_treeclimb(nfs_export_t *ne, exportinfo_t *exip, secinfo_t *sec,
724 724 int seccnt, bool_t isadd)
725 725 {
726 726 treenode_t *tnode;
727 727
728 728 ASSERT(RW_WRITE_HELD(&ne->exported_lock));
729 + ASSERT3U(exip->exi_zoneid, ==, curzone->zone_id);
729 730
730 731 /*
731 732 * exi_tree can be null for the zone root
732 733 * which means we're already at the "top"
733 734 * and there's nothing more to "climb".
734 735 */
735 736 tnode = exip->exi_tree;
736 737 if (tnode == NULL) {
737 738 /* Should only happen for... */
738 739 ASSERT(exip == ne->exi_root);
739 740 return;
740 741 }
741 742
742 743 if (seccnt == 0)
743 744 return;
744 745
745 746 /*
746 747 * If flavors are being added and the new export root isn't
747 748 * also VROOT, its implicitly allowed flavors are inherited from
748 749 * its pseudonode.
749 750 * Note - for VROOT exports the implicitly allowed flavors were
750 751 * transferred from the PSEUDO export in exportfs()
751 752 */
752 753 if (isadd && !(exip->exi_vp->v_flag & VROOT) &&
753 754 !VN_IS_CURZONEROOT(exip->exi_vp) &&
754 755 tnode->tree_vis->vis_seccnt > 0) {
755 756 srv_secinfo_add(&exip->exi_export.ex_secinfo,
756 757 &exip->exi_export.ex_seccnt, tnode->tree_vis->vis_secinfo,
757 758 tnode->tree_vis->vis_seccnt, FALSE);
758 759 }
759 760
760 761 /*
761 762 * Move to parent node and propagate sec flavor
762 763 * to exportinfo and to visible structures.
763 764 */
764 765 tnode = tnode->tree_parent;
765 766
766 767 while (tnode != NULL) {
767 768
768 769 /* If there is exportinfo, update it */
769 770 if (tnode->tree_exi != NULL) {
770 771 secinfo_t **pxsec =
771 772 &tnode->tree_exi->exi_export.ex_secinfo;
772 773 int *pxcnt = &tnode->tree_exi->exi_export.ex_seccnt;
773 774 int is_pseudo = PSEUDO(tnode->tree_exi);
774 775 if (isadd)
775 776 srv_secinfo_add(pxsec, pxcnt, sec, seccnt,
776 777 is_pseudo);
777 778 else
778 779 srv_secinfo_remove(pxsec, pxcnt, sec, seccnt);
779 780 }
780 781
781 782 /* Update every visible - only root node has no visible */
782 783 if (tnode->tree_vis != NULL) {
783 784 secinfo_t **pxsec = &tnode->tree_vis->vis_secinfo;
784 785 int *pxcnt = &tnode->tree_vis->vis_seccnt;
785 786 if (isadd)
786 787 srv_secinfo_add(pxsec, pxcnt, sec, seccnt,
787 788 FALSE);
788 789 else
789 790 srv_secinfo_remove(pxsec, pxcnt, sec, seccnt);
790 791 }
791 792 tnode = tnode->tree_parent;
792 793 }
793 794 }
794 795
795 796 /* hash_name is a text substitution for either fid_hash or path_hash */
796 797 #define exp_hash_unlink(exi, hash_name) \
797 798 if (*(exi)->hash_name.bckt == (exi)) \
798 799 *(exi)->hash_name.bckt = (exi)->hash_name.next; \
799 800 if ((exi)->hash_name.prev) \
800 801 (exi)->hash_name.prev->hash_name.next = (exi)->hash_name.next; \
801 802 if ((exi)->hash_name.next) \
802 803 (exi)->hash_name.next->hash_name.prev = (exi)->hash_name.prev; \
803 804 (exi)->hash_name.bckt = NULL;
804 805
805 806 #define exp_hash_link(exi, hash_name, bucket) \
806 807 (exi)->hash_name.bckt = (bucket); \
807 808 (exi)->hash_name.prev = NULL; \
808 809 (exi)->hash_name.next = *(bucket); \
|
↓ open down ↓ |
70 lines elided |
↑ open up ↑ |
809 810 if ((exi)->hash_name.next) \
810 811 (exi)->hash_name.next->hash_name.prev = (exi); \
811 812 *(bucket) = (exi);
812 813
813 814 void
814 815 export_link(nfs_export_t *ne, exportinfo_t *exi)
815 816 {
816 817 exportinfo_t **bckt;
817 818
818 819 ASSERT(RW_WRITE_HELD(&ne->exported_lock));
819 - ASSERT(exi->exi_zoneid == ne->ne_globals->nfs_zoneid);
820 820
821 821 bckt = &ne->exptable[exptablehash(&exi->exi_fsid, &exi->exi_fid)];
822 822 exp_hash_link(exi, fid_hash, bckt);
823 823
824 824 bckt = &ne->exptable_path_hash[pkp_tab_hash(exi->exi_export.ex_path,
825 825 strlen(exi->exi_export.ex_path))];
826 826 exp_hash_link(exi, path_hash, bckt);
827 827 }
828 828
829 829 /*
830 830 * Helper functions for exi_id handling
831 831 */
832 832 static int
833 833 exi_id_compar(const void *v1, const void *v2)
834 834 {
835 835 const struct exportinfo *e1 = v1;
836 836 const struct exportinfo *e2 = v2;
837 837
838 838 if (e1->exi_id < e2->exi_id)
839 839 return (-1);
840 840 if (e1->exi_id > e2->exi_id)
841 841 return (1);
842 842
843 843 return (0);
844 844 }
845 845
846 846 int
847 847 exi_id_get_next()
848 848 {
849 849 struct exportinfo e;
850 850 int ret = exi_id_next;
851 851
852 852 ASSERT(MUTEX_HELD(&nfs_exi_id_lock));
853 853
854 854 do {
855 855 exi_id_next++;
856 856 if (exi_id_next == 0)
857 857 exi_id_overflow = TRUE;
858 858
859 859 if (!exi_id_overflow)
860 860 break;
861 861
862 862 if (exi_id_next == ret)
863 863 cmn_err(CE_PANIC, "exi_id exhausted");
864 864
865 865 e.exi_id = exi_id_next;
866 866 } while (avl_find(&exi_id_tree, &e, NULL) != NULL);
867 867
868 868 return (ret);
869 869 }
870 870
871 871 /*
872 872 * Get the root file handle for this zone.
873 873 * Called when nfs_svc() starts
874 874 */
875 875 int
876 876 nfs_export_get_rootfh(nfs_globals_t *g)
877 877 {
878 878 nfs_export_t *ne = g->nfs_export;
879 879 int err;
880 880
881 881 ne->exi_rootfid.fid_len = MAXFIDSZ;
882 882 err = vop_fid_pseudo(ne->exi_root->exi_vp, &ne->exi_rootfid);
883 883 if (err != 0) {
884 884 ne->exi_rootfid.fid_len = 0;
885 885 return (err);
886 886 }
887 887
888 888 /* Setup the fhandle template exi_fh */
889 889 ne->exi_root->exi_fh.fh_fsid = rootdir->v_vfsp->vfs_fsid;
890 890 ne->exi_root->exi_fh.fh_xlen = ne->exi_rootfid.fid_len;
891 891 bcopy(ne->exi_rootfid.fid_data, ne->exi_root->exi_fh.fh_xdata,
892 892 ne->exi_rootfid.fid_len);
|
↓ open down ↓ |
63 lines elided |
↑ open up ↑ |
893 893 ne->exi_root->exi_fh.fh_len = sizeof (ne->exi_root->exi_fh.fh_data);
894 894
895 895 return (0);
896 896 }
897 897
898 898 void
899 899 nfs_export_zone_init(nfs_globals_t *ng)
900 900 {
901 901 int i;
902 902 nfs_export_t *ne;
903 + zone_t *zone;
903 904
904 905 ne = kmem_zalloc(sizeof (*ne), KM_SLEEP);
905 906
906 907 rw_init(&ne->exported_lock, NULL, RW_DEFAULT, NULL);
907 908
908 909 ne->ne_globals = ng; /* "up" pointer */
909 910
910 911 /*
911 912 * Allocate the place holder for the public file handle, which
912 913 * is all zeroes. It is initially set to the root filesystem.
913 914 */
914 915 ne->exi_root = kmem_zalloc(sizeof (*ne->exi_root), KM_SLEEP);
915 916 ne->exi_public = ne->exi_root;
916 917
|
↓ open down ↓ |
4 lines elided |
↑ open up ↑ |
917 918 ne->exi_root->exi_export.ex_flags = EX_PUBLIC;
918 919 ne->exi_root->exi_export.ex_pathlen = 1; /* length of "/" */
919 920 ne->exi_root->exi_export.ex_path =
920 921 kmem_alloc(ne->exi_root->exi_export.ex_pathlen + 1, KM_SLEEP);
921 922 ne->exi_root->exi_export.ex_path[0] = '/';
922 923 ne->exi_root->exi_export.ex_path[1] = '\0';
923 924
924 925 ne->exi_root->exi_count = 1;
925 926 mutex_init(&ne->exi_root->exi_lock, NULL, MUTEX_DEFAULT, NULL);
926 927
927 - ASSERT(curzone->zone_id == ng->nfs_zoneid);
928 - ne->exi_root->exi_vp = ZONE_ROOTVP();
929 - ne->exi_root->exi_zoneid = ng->nfs_zoneid;
928 + /*
929 + * Because we cannot:
930 + * ASSERT(curzone->zone_id == ng->nfs_zoneid);
931 + * We grab the zone pointer explicitly (like netstacks do) and
932 + * set the rootvp here.
933 + */
934 + zone = zone_find_by_id_nolock(ng->nfs_zoneid);
935 + ne->exi_root->exi_vp = zone->zone_rootvp;
936 + ne->exi_root->exi_zone = zone; /* XXX KEBE SAYS lose me, and... */
937 + /* ne->exi_root->exi_zoneid = ng->nfs_zoneid; */ /* use me instead! */
930 938
931 939 /*
932 940 * Fill in ne->exi_rootfid later, in nfs_export_get_rootfid
933 941 * because we can't correctly return errors here.
934 942 */
935 943
936 944 /* Initialize auth cache and auth cache lock */
937 945 for (i = 0; i < AUTH_TABLESIZE; i++) {
938 946 ne->exi_root->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t),
939 947 KM_SLEEP);
940 948 avl_create(ne->exi_root->exi_cache[i],
941 949 nfsauth_cache_clnt_compar, sizeof (struct auth_cache_clnt),
942 950 offsetof(struct auth_cache_clnt, authc_link));
943 951 }
944 952 rw_init(&ne->exi_root->exi_cache_lock, NULL, RW_DEFAULT, NULL);
945 953
946 954 /* setup exi_fh later, in nfs_export_get_rootfid */
947 955
948 956 rw_enter(&ne->exported_lock, RW_WRITER);
949 957
950 958 /* Publish the exportinfo in the hash table */
951 959 export_link(ne, ne->exi_root);
952 960
953 961 /* Initialize exi_id and exi_kstats */
954 962 mutex_enter(&nfs_exi_id_lock);
955 963 ne->exi_root->exi_id = exi_id_get_next();
956 964 avl_add(&exi_id_tree, ne->exi_root);
957 965 mutex_exit(&nfs_exi_id_lock);
958 966
959 967 rw_exit(&ne->exported_lock);
960 968 ne->ns_root = NULL;
961 969
962 970 ng->nfs_export = ne;
963 971 }
964 972
965 973 /*
966 974 * During zone shutdown, remove exports
967 975 */
968 976 void
969 977 nfs_export_zone_shutdown(nfs_globals_t *ng)
970 978 {
971 979 nfs_export_t *ne = ng->nfs_export;
972 980 struct exportinfo *exi, *nexi;
973 981 int i, errors;
974 982
975 983 rw_enter(&ne->exported_lock, RW_READER);
976 984
977 985 errors = 0;
978 986 for (i = 0; i < EXPTABLESIZE; i++) {
979 987
980 988 exi = ne->exptable[i];
981 989 if (exi != NULL)
982 990 exi_hold(exi);
983 991
984 992 while (exi != NULL) {
985 993
986 994 /*
987 995 * Get and hold next export before
988 996 * dropping the rwlock and unexport
989 997 */
990 998 nexi = exi->fid_hash.next;
991 999 if (nexi != NULL)
992 1000 exi_hold(nexi);
993 1001
994 1002 rw_exit(&ne->exported_lock);
995 1003
996 1004 /*
997 1005 * Skip ne->exi_root which gets special
998 1006 * create/destroy handling.
999 1007 */
1000 1008 if (exi != ne->exi_root &&
1001 1009 unexport(ne, exi) != 0)
1002 1010 errors++;
1003 1011 exi_rele(exi);
1004 1012
1005 1013 rw_enter(&ne->exported_lock, RW_READER);
1006 1014 exi = nexi;
1007 1015 }
1008 1016 }
1009 1017 if (errors > 0) {
1010 1018 cmn_err(CE_NOTE,
1011 1019 "NFS: failed un-exports in zone %d",
1012 1020 (int) ng->nfs_zoneid);
1013 1021 }
1014 1022
1015 1023 rw_exit(&ne->exported_lock);
1016 1024 }
1017 1025
1018 1026 void
1019 1027 nfs_export_zone_fini(nfs_globals_t *ng)
1020 1028 {
1021 1029 int i;
1022 1030 nfs_export_t *ne = ng->nfs_export;
1023 1031 struct exportinfo *exi;
1024 1032
1025 1033 ng->nfs_export = NULL;
1026 1034
1027 1035 rw_enter(&ne->exported_lock, RW_WRITER);
1028 1036
1029 1037 mutex_enter(&nfs_exi_id_lock);
1030 1038 avl_remove(&exi_id_tree, ne->exi_root);
1031 1039 mutex_exit(&nfs_exi_id_lock);
1032 1040
1033 1041 export_unlink(ne, ne->exi_root);
1034 1042
1035 1043 rw_exit(&ne->exported_lock);
1036 1044
1037 1045 /* Deallocate the place holder for the public file handle */
1038 1046 srv_secinfo_list_free(ne->exi_root->exi_export.ex_secinfo,
1039 1047 ne->exi_root->exi_export.ex_seccnt);
1040 1048 mutex_destroy(&ne->exi_root->exi_lock);
1041 1049
1042 1050 rw_destroy(&ne->exi_root->exi_cache_lock);
1043 1051 for (i = 0; i < AUTH_TABLESIZE; i++) {
1044 1052 avl_destroy(ne->exi_root->exi_cache[i]);
1045 1053 kmem_free(ne->exi_root->exi_cache[i], sizeof (avl_tree_t));
1046 1054 }
1047 1055
1048 1056 kmem_free(ne->exi_root->exi_export.ex_path,
1049 1057 ne->exi_root->exi_export.ex_pathlen + 1);
1050 1058 kmem_free(ne->exi_root, sizeof (*ne->exi_root));
1051 1059
1052 1060 /*
1053 1061 * The shutdown hook should have left the exi_id_tree
1054 1062 * with nothing belonging to this zone.
1055 1063 */
1056 1064 mutex_enter(&nfs_exi_id_lock);
1057 1065 i = 0;
1058 1066 exi = avl_first(&exi_id_tree);
1059 1067 while (exi != NULL) {
1060 1068 if (exi->exi_zoneid == ng->nfs_zoneid)
1061 1069 i++;
1062 1070 exi = AVL_NEXT(&exi_id_tree, exi);
1063 1071 }
1064 1072 mutex_exit(&nfs_exi_id_lock);
1065 1073 if (i > 0) {
1066 1074 cmn_err(CE_NOTE,
1067 1075 "NFS: zone %d has %d export IDs left after shutdown",
1068 1076 (int) ng->nfs_zoneid, i);
1069 1077 }
1070 1078 rw_destroy(&ne->exported_lock);
1071 1079 kmem_free(ne, sizeof (*ne));
1072 1080 }
1073 1081
1074 1082 /*
1075 1083 * Initialization routine for export routines.
1076 1084 * Should only be called once.
1077 1085 */
1078 1086 void
1079 1087 nfs_exportinit(void)
1080 1088 {
1081 1089 mutex_init(&nfs_exi_id_lock, NULL, MUTEX_DEFAULT, NULL);
1082 1090
1083 1091 /* exi_id handling initialization */
1084 1092 exi_id_next = 0;
1085 1093 exi_id_overflow = FALSE;
1086 1094 avl_create(&exi_id_tree, exi_id_compar, sizeof (struct exportinfo),
1087 1095 offsetof(struct exportinfo, exi_id_link));
1088 1096
1089 1097 nfslog_init();
1090 1098 }
1091 1099
1092 1100 /*
1093 1101 * Finalization routine for export routines.
1094 1102 */
1095 1103 void
1096 1104 nfs_exportfini(void)
1097 1105 {
1098 1106 avl_destroy(&exi_id_tree);
1099 1107 mutex_destroy(&nfs_exi_id_lock);
1100 1108 }
1101 1109
1102 1110 /*
1103 1111 * Check if 2 gss mechanism identifiers are the same.
1104 1112 *
1105 1113 * return FALSE if not the same.
1106 1114 * return TRUE if the same.
1107 1115 */
1108 1116 static bool_t
1109 1117 nfs_mech_equal(rpc_gss_OID mech1, rpc_gss_OID mech2)
1110 1118 {
1111 1119 if ((mech1->length == 0) && (mech2->length == 0))
1112 1120 return (TRUE);
1113 1121
1114 1122 if (mech1->length != mech2->length)
1115 1123 return (FALSE);
1116 1124
1117 1125 return (bcmp(mech1->elements, mech2->elements, mech1->length) == 0);
1118 1126 }
1119 1127
1120 1128 /*
1121 1129 * This routine is used by rpc to map rpc security number
1122 1130 * to nfs specific security flavor number.
1123 1131 *
1124 1132 * The gss callback prototype is
1125 1133 * callback(struct svc_req *, gss_cred_id_t *, gss_ctx_id_t *,
1126 1134 * rpc_gss_lock_t *, void **),
1127 1135 * since nfs does not use the gss_cred_id_t/gss_ctx_id_t arguments
1128 1136 * we cast them to void.
1129 1137 */
1130 1138 /*ARGSUSED*/
1131 1139 bool_t
1132 1140 rfs_gsscallback(struct svc_req *req, gss_cred_id_t deleg, void *gss_context,
1133 1141 rpc_gss_lock_t *lock, void **cookie)
1134 1142 {
1135 1143 int i, j;
1136 1144 rpc_gss_rawcred_t *raw_cred;
1137 1145 struct exportinfo *exi;
1138 1146 nfs_export_t *ne = nfs_get_export();
1139 1147
1140 1148 /*
1141 1149 * We don't deal with delegated credentials.
1142 1150 */
1143 1151 if (deleg != GSS_C_NO_CREDENTIAL)
1144 1152 return (FALSE);
1145 1153
1146 1154 raw_cred = lock->raw_cred;
1147 1155 *cookie = NULL;
1148 1156
1149 1157 rw_enter(&ne->exported_lock, RW_READER);
1150 1158
1151 1159 for (i = 0; i < EXPTABLESIZE; i++) {
1152 1160 exi = ne->exptable[i];
1153 1161 while (exi) {
1154 1162 if (exi->exi_export.ex_seccnt > 0) {
1155 1163 struct secinfo *secp;
1156 1164 seconfig_t *se;
1157 1165 int seccnt;
1158 1166
1159 1167 secp = exi->exi_export.ex_secinfo;
1160 1168 seccnt = exi->exi_export.ex_seccnt;
1161 1169 for (j = 0; j < seccnt; j++) {
1162 1170 /*
1163 1171 * If there is a map of the triplet
1164 1172 * (mechanism, service, qop) between
1165 1173 * raw_cred and the exported flavor,
1166 1174 * get the psudo flavor number.
1167 1175 * Also qop should not be NULL, it
1168 1176 * should be "default" or something
1169 1177 * else.
1170 1178 */
1171 1179 se = &secp[j].s_secinfo;
1172 1180 if ((se->sc_rpcnum == RPCSEC_GSS) &&
1173 1181
1174 1182 (nfs_mech_equal(
1175 1183 se->sc_gss_mech_type,
1176 1184 raw_cred->mechanism)) &&
1177 1185
1178 1186 (se->sc_service ==
1179 1187 raw_cred->service) &&
1180 1188 (raw_cred->qop == se->sc_qop)) {
1181 1189
1182 1190 *cookie = (void *)(uintptr_t)
1183 1191 se->sc_nfsnum;
1184 1192 goto done;
1185 1193 }
1186 1194 }
1187 1195 }
1188 1196 exi = exi->fid_hash.next;
1189 1197 }
1190 1198 }
1191 1199 done:
1192 1200 rw_exit(&ne->exported_lock);
1193 1201
1194 1202 /*
1195 1203 * If no nfs pseudo number mapping can be found in the export
1196 1204 * table, assign the nfsflavor to NFS_FLAVOR_NOMAP. In V4, we may
1197 1205 * recover the flavor mismatch from NFS layer (NFS4ERR_WRONGSEC).
1198 1206 *
1199 1207 * For example:
1200 1208 * server first shares with krb5i;
1201 1209 * client mounts with krb5i;
1202 1210 * server re-shares with krb5p;
1203 1211 * client tries with krb5i, but no mapping can be found;
1204 1212 * rpcsec_gss module calls this routine to do the mapping,
1205 1213 * if this routine fails, request is rejected from
1206 1214 * the rpc layer.
1207 1215 * What we need is to let the nfs layer rejects the request.
1208 1216 * For V4, we can reject with NFS4ERR_WRONGSEC and the client
1209 1217 * may recover from it by getting the new flavor via SECINFO.
1210 1218 *
1211 1219 * nfs pseudo number for RPCSEC_GSS mapping (see nfssec.conf)
1212 1220 * is owned by IANA (see RFC 2623).
1213 1221 *
1214 1222 * XXX NFS_FLAVOR_NOMAP is defined in Solaris to work around
1215 1223 * the implementation issue. This number should not overlap with
1216 1224 * any new IANA defined pseudo flavor numbers.
1217 1225 */
1218 1226 if (*cookie == NULL)
1219 1227 *cookie = (void *)NFS_FLAVOR_NOMAP;
1220 1228
1221 1229 lock->locked = TRUE;
1222 1230
1223 1231 return (TRUE);
1224 1232 }
1225 1233
1226 1234
1227 1235 /*
1228 1236 * Exportfs system call; credentials should be checked before
1229 1237 * calling this function.
1230 1238 */
1231 1239 int
1232 1240 exportfs(struct exportfs_args *args, model_t model, cred_t *cr)
1233 1241 {
1234 1242 vnode_t *vp;
1235 1243 vnode_t *dvp;
1236 1244 struct exportdata *kex;
1237 1245 struct exportinfo *exi = NULL;
1238 1246 struct exportinfo *ex, *ex1, *ex2;
1239 1247 fid_t fid;
1240 1248 fsid_t fsid;
1241 1249 int error;
1242 1250 size_t allocsize;
1243 1251 struct secinfo *sp;
1244 1252 struct secinfo *exs;
1245 1253 rpc_gss_callback_t cb;
1246 1254 char *pathbuf;
1247 1255 char *log_buffer;
1248 1256 char *tagbuf;
1249 1257 int callback;
1250 1258 int allocd_seccnt;
1251 1259 STRUCT_HANDLE(exportfs_args, uap);
1252 1260 STRUCT_DECL(exportdata, uexi);
1253 1261 struct secinfo newsec[MAX_FLAVORS];
1254 1262 int newcnt;
1255 1263 struct secinfo oldsec[MAX_FLAVORS];
1256 1264 int oldcnt;
1257 1265 int i;
1258 1266 struct pathname lookpn;
1259 1267 nfs_export_t *ne = nfs_get_export();
1260 1268
1261 1269 STRUCT_SET_HANDLE(uap, model, args);
1262 1270
1263 1271 /* Read in pathname from userspace */
1264 1272 if (error = pn_get(STRUCT_FGETP(uap, dname), UIO_USERSPACE, &lookpn))
1265 1273 return (error);
1266 1274
1267 1275 /* Walk the export list looking for that pathname */
1268 1276 rw_enter(&ne->exported_lock, RW_READER);
1269 1277 DTRACE_PROBE(nfss__i__exported_lock1_start);
1270 1278 for (ex1 = ne->exptable_path_hash[pkp_tab_hash(lookpn.pn_path,
1271 1279 strlen(lookpn.pn_path))]; ex1; ex1 = ex1->path_hash.next) {
1272 1280 if (ex1 != ne->exi_root && 0 ==
1273 1281 strcmp(ex1->exi_export.ex_path, lookpn.pn_path)) {
1274 1282 exi_hold(ex1);
1275 1283 break;
1276 1284 }
1277 1285 }
1278 1286 DTRACE_PROBE(nfss__i__exported_lock1_stop);
1279 1287 rw_exit(&ne->exported_lock);
1280 1288
1281 1289 /* Is this an unshare? */
1282 1290 if (STRUCT_FGETP(uap, uex) == NULL) {
1283 1291 pn_free(&lookpn);
1284 1292 if (ex1 == NULL)
1285 1293 return (EINVAL);
1286 1294 error = unexport(ne, ex1);
1287 1295 exi_rele(ex1);
1288 1296 return (error);
1289 1297 }
1290 1298
1291 1299 /* It is a share or a re-share */
1292 1300 error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1293 1301 FOLLOW, &dvp, &vp);
1294 1302 if (error == EINVAL) {
1295 1303 /*
1296 1304 * if fname resolves to / we get EINVAL error
1297 1305 * since we wanted the parent vnode. Try again
1298 1306 * with NULL dvp.
1299 1307 */
1300 1308 error = lookupname(STRUCT_FGETP(uap, dname), UIO_USERSPACE,
1301 1309 FOLLOW, NULL, &vp);
1302 1310 dvp = NULL;
1303 1311 }
1304 1312 if (!error && vp == NULL) {
1305 1313 /* Last component of fname not found */
1306 1314 if (dvp != NULL)
1307 1315 VN_RELE(dvp);
1308 1316 error = ENOENT;
1309 1317 }
1310 1318 if (error) {
1311 1319 pn_free(&lookpn);
1312 1320 if (ex1)
1313 1321 exi_rele(ex1);
1314 1322 return (error);
1315 1323 }
1316 1324
1317 1325 /*
1318 1326 * 'vp' may be an AUTOFS node, so we perform a
1319 1327 * VOP_ACCESS() to trigger the mount of the
1320 1328 * intended filesystem, so we can share the intended
1321 1329 * filesystem instead of the AUTOFS filesystem.
1322 1330 */
1323 1331 (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
1324 1332
1325 1333 /*
1326 1334 * We're interested in the top most filesystem.
1327 1335 * This is specially important when uap->dname is a trigger
1328 1336 * AUTOFS node, since we're really interested in sharing the
1329 1337 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
1330 1338 * call not the AUTOFS node itself.
1331 1339 */
1332 1340 if (vn_mountedvfs(vp) != NULL) {
1333 1341 if (error = traverse(&vp)) {
1334 1342 VN_RELE(vp);
1335 1343 if (dvp != NULL)
1336 1344 VN_RELE(dvp);
1337 1345 pn_free(&lookpn);
1338 1346 if (ex1)
1339 1347 exi_rele(ex1);
1340 1348 return (error);
1341 1349 }
1342 1350 }
1343 1351
1344 1352 /* Do not allow sharing another vnode for already shared path */
1345 1353 if (ex1 && !PSEUDO(ex1) && !VN_CMP(ex1->exi_vp, vp)) {
1346 1354 VN_RELE(vp);
1347 1355 if (dvp != NULL)
1348 1356 VN_RELE(dvp);
1349 1357 pn_free(&lookpn);
1350 1358 exi_rele(ex1);
1351 1359 return (EEXIST);
1352 1360 }
1353 1361 if (ex1)
1354 1362 exi_rele(ex1);
1355 1363
1356 1364 /*
1357 1365 * Get the vfs id
1358 1366 */
1359 1367 bzero(&fid, sizeof (fid));
1360 1368 fid.fid_len = MAXFIDSZ;
1361 1369 error = VOP_FID(vp, &fid, NULL);
1362 1370 fsid = vp->v_vfsp->vfs_fsid;
1363 1371
1364 1372 if (error) {
1365 1373 VN_RELE(vp);
1366 1374 if (dvp != NULL)
1367 1375 VN_RELE(dvp);
1368 1376 /*
1369 1377 * If VOP_FID returns ENOSPC then the fid supplied
1370 1378 * is too small. For now we simply return EREMOTE.
1371 1379 */
1372 1380 if (error == ENOSPC)
1373 1381 error = EREMOTE;
1374 1382 pn_free(&lookpn);
1375 1383 return (error);
1376 1384 }
1377 1385
1378 1386 /*
1379 1387 * Do not allow re-sharing a shared vnode under a different path
1380 1388 * PSEUDO export has ex_path fabricated, e.g. "/tmp (pseudo)", skip it.
1381 1389 */
1382 1390 rw_enter(&ne->exported_lock, RW_READER);
1383 1391 DTRACE_PROBE(nfss__i__exported_lock2_start);
1384 1392 for (ex2 = ne->exptable[exptablehash(&fsid, &fid)]; ex2;
1385 1393 ex2 = ex2->fid_hash.next) {
1386 1394 if (ex2 != ne->exi_root && !PSEUDO(ex2) &&
1387 1395 VN_CMP(ex2->exi_vp, vp) &&
1388 1396 strcmp(ex2->exi_export.ex_path, lookpn.pn_path) != 0) {
1389 1397 DTRACE_PROBE(nfss__i__exported_lock2_stop);
1390 1398 rw_exit(&ne->exported_lock);
1391 1399 VN_RELE(vp);
1392 1400 if (dvp != NULL)
1393 1401 VN_RELE(dvp);
1394 1402 pn_free(&lookpn);
1395 1403 return (EEXIST);
1396 1404 }
1397 1405 }
1398 1406 DTRACE_PROBE(nfss__i__exported_lock2_stop);
1399 1407 rw_exit(&ne->exported_lock);
1400 1408 pn_free(&lookpn);
1401 1409
1402 1410 exi = kmem_zalloc(sizeof (*exi), KM_SLEEP);
1403 1411 exi->exi_fsid = fsid;
1404 1412 exi->exi_fid = fid;
1405 1413 exi->exi_vp = vp;
1406 1414 exi->exi_count = 1;
1407 1415 exi->exi_zone = crgetzone(cr);
1408 1416 ASSERT(exi->exi_zone != NULL); /* XXX KEBE ASKS... */
1409 1417 ASSERT3P(exi->exi_zone, ==, curzone); /* ... are these legit? */
1410 1418 exi->exi_volatile_dev = (vfssw[vp->v_vfsp->vfs_fstype].vsw_flag &
1411 1419 VSW_VOLATILEDEV) ? 1 : 0;
1412 1420 mutex_init(&exi->exi_lock, NULL, MUTEX_DEFAULT, NULL);
1413 1421 exi->exi_dvp = dvp;
1414 1422
1415 1423 /*
1416 1424 * Initialize auth cache and auth cache lock
1417 1425 */
1418 1426 for (i = 0; i < AUTH_TABLESIZE; i++) {
1419 1427 exi->exi_cache[i] = kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1420 1428 avl_create(exi->exi_cache[i], nfsauth_cache_clnt_compar,
1421 1429 sizeof (struct auth_cache_clnt),
1422 1430 offsetof(struct auth_cache_clnt, authc_link));
1423 1431 }
1424 1432 rw_init(&exi->exi_cache_lock, NULL, RW_DEFAULT, NULL);
1425 1433
1426 1434 /*
1427 1435 * Build up the template fhandle
1428 1436 */
1429 1437 exi->exi_fh.fh_fsid = fsid;
1430 1438 if (exi->exi_fid.fid_len > sizeof (exi->exi_fh.fh_xdata)) {
1431 1439 error = EREMOTE;
1432 1440 goto out1;
1433 1441 }
1434 1442 exi->exi_fh.fh_xlen = exi->exi_fid.fid_len;
1435 1443 bcopy(exi->exi_fid.fid_data, exi->exi_fh.fh_xdata,
1436 1444 exi->exi_fid.fid_len);
1437 1445
1438 1446 exi->exi_fh.fh_len = sizeof (exi->exi_fh.fh_data);
1439 1447
1440 1448 kex = &exi->exi_export;
1441 1449
1442 1450 /*
1443 1451 * Load in everything, and do sanity checking
1444 1452 */
1445 1453 STRUCT_INIT(uexi, model);
1446 1454 if (copyin(STRUCT_FGETP(uap, uex), STRUCT_BUF(uexi),
1447 1455 STRUCT_SIZE(uexi))) {
1448 1456 error = EFAULT;
1449 1457 goto out1;
1450 1458 }
1451 1459
1452 1460 kex->ex_version = STRUCT_FGET(uexi, ex_version);
1453 1461 if (kex->ex_version != EX_CURRENT_VERSION) {
1454 1462 error = EINVAL;
1455 1463 cmn_err(CE_WARN,
1456 1464 "NFS: exportfs requires export struct version 2 - got %d\n",
1457 1465 kex->ex_version);
1458 1466 goto out1;
1459 1467 }
1460 1468
1461 1469 /*
1462 1470 * Must have at least one security entry
1463 1471 */
1464 1472 kex->ex_seccnt = STRUCT_FGET(uexi, ex_seccnt);
1465 1473 if (kex->ex_seccnt < 1) {
1466 1474 error = EINVAL;
1467 1475 goto out1;
1468 1476 }
1469 1477
1470 1478 kex->ex_path = STRUCT_FGETP(uexi, ex_path);
1471 1479 kex->ex_pathlen = STRUCT_FGET(uexi, ex_pathlen);
1472 1480 kex->ex_flags = STRUCT_FGET(uexi, ex_flags);
1473 1481 kex->ex_anon = STRUCT_FGET(uexi, ex_anon);
1474 1482 kex->ex_secinfo = STRUCT_FGETP(uexi, ex_secinfo);
1475 1483 kex->ex_index = STRUCT_FGETP(uexi, ex_index);
1476 1484 kex->ex_log_buffer = STRUCT_FGETP(uexi, ex_log_buffer);
1477 1485 kex->ex_log_bufferlen = STRUCT_FGET(uexi, ex_log_bufferlen);
1478 1486 kex->ex_tag = STRUCT_FGETP(uexi, ex_tag);
1479 1487 kex->ex_taglen = STRUCT_FGET(uexi, ex_taglen);
1480 1488
1481 1489 /*
1482 1490 * Copy the exported pathname into
1483 1491 * an appropriately sized buffer.
1484 1492 */
1485 1493 pathbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1486 1494 if (copyinstr(kex->ex_path, pathbuf, MAXPATHLEN, &kex->ex_pathlen)) {
1487 1495 kmem_free(pathbuf, MAXPATHLEN);
1488 1496 error = EFAULT;
1489 1497 goto out1;
1490 1498 }
1491 1499 kex->ex_path = kmem_alloc(kex->ex_pathlen + 1, KM_SLEEP);
1492 1500 bcopy(pathbuf, kex->ex_path, kex->ex_pathlen);
1493 1501 kex->ex_path[kex->ex_pathlen] = '\0';
1494 1502 kmem_free(pathbuf, MAXPATHLEN);
1495 1503
1496 1504 /*
1497 1505 * Get the path to the logging buffer and the tag
1498 1506 */
1499 1507 if (kex->ex_flags & EX_LOG) {
1500 1508 log_buffer = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1501 1509 if (copyinstr(kex->ex_log_buffer, log_buffer, MAXPATHLEN,
1502 1510 &kex->ex_log_bufferlen)) {
1503 1511 kmem_free(log_buffer, MAXPATHLEN);
1504 1512 error = EFAULT;
1505 1513 goto out2;
1506 1514 }
1507 1515 kex->ex_log_buffer =
1508 1516 kmem_alloc(kex->ex_log_bufferlen + 1, KM_SLEEP);
1509 1517 bcopy(log_buffer, kex->ex_log_buffer, kex->ex_log_bufferlen);
1510 1518 kex->ex_log_buffer[kex->ex_log_bufferlen] = '\0';
1511 1519 kmem_free(log_buffer, MAXPATHLEN);
1512 1520
1513 1521 tagbuf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1514 1522 if (copyinstr(kex->ex_tag, tagbuf, MAXPATHLEN,
1515 1523 &kex->ex_taglen)) {
1516 1524 kmem_free(tagbuf, MAXPATHLEN);
1517 1525 error = EFAULT;
1518 1526 goto out3;
1519 1527 }
1520 1528 kex->ex_tag = kmem_alloc(kex->ex_taglen + 1, KM_SLEEP);
1521 1529 bcopy(tagbuf, kex->ex_tag, kex->ex_taglen);
1522 1530 kex->ex_tag[kex->ex_taglen] = '\0';
1523 1531 kmem_free(tagbuf, MAXPATHLEN);
1524 1532 }
1525 1533
1526 1534 /*
1527 1535 * Load the security information for each flavor
1528 1536 */
1529 1537 allocsize = kex->ex_seccnt * SIZEOF_STRUCT(secinfo, model);
1530 1538 sp = kmem_zalloc(allocsize, KM_SLEEP);
1531 1539 if (copyin(kex->ex_secinfo, sp, allocsize)) {
1532 1540 kmem_free(sp, allocsize);
1533 1541 error = EFAULT;
1534 1542 goto out4;
1535 1543 }
1536 1544
1537 1545 /*
1538 1546 * All of these nested structures need to be converted to
1539 1547 * the kernel native format.
1540 1548 */
1541 1549 if (model != DATAMODEL_NATIVE) {
1542 1550 size_t allocsize2;
1543 1551 struct secinfo *sp2;
1544 1552
1545 1553 allocsize2 = kex->ex_seccnt * sizeof (struct secinfo);
1546 1554 sp2 = kmem_zalloc(allocsize2, KM_SLEEP);
1547 1555
1548 1556 for (i = 0; i < kex->ex_seccnt; i++) {
1549 1557 STRUCT_HANDLE(secinfo, usi);
1550 1558
1551 1559 STRUCT_SET_HANDLE(usi, model,
1552 1560 (struct secinfo *)((caddr_t)sp +
1553 1561 (i * SIZEOF_STRUCT(secinfo, model))));
1554 1562 bcopy(STRUCT_FGET(usi, s_secinfo.sc_name),
1555 1563 sp2[i].s_secinfo.sc_name, MAX_NAME_LEN);
1556 1564 sp2[i].s_secinfo.sc_nfsnum =
1557 1565 STRUCT_FGET(usi, s_secinfo.sc_nfsnum);
1558 1566 sp2[i].s_secinfo.sc_rpcnum =
1559 1567 STRUCT_FGET(usi, s_secinfo.sc_rpcnum);
1560 1568 bcopy(STRUCT_FGET(usi, s_secinfo.sc_gss_mech),
1561 1569 sp2[i].s_secinfo.sc_gss_mech, MAX_NAME_LEN);
1562 1570 sp2[i].s_secinfo.sc_gss_mech_type =
1563 1571 STRUCT_FGETP(usi, s_secinfo.sc_gss_mech_type);
1564 1572 sp2[i].s_secinfo.sc_qop =
1565 1573 STRUCT_FGET(usi, s_secinfo.sc_qop);
1566 1574 sp2[i].s_secinfo.sc_service =
1567 1575 STRUCT_FGET(usi, s_secinfo.sc_service);
1568 1576
1569 1577 sp2[i].s_flags = STRUCT_FGET(usi, s_flags);
1570 1578 sp2[i].s_window = STRUCT_FGET(usi, s_window);
1571 1579 sp2[i].s_rootid = STRUCT_FGET(usi, s_rootid);
1572 1580 sp2[i].s_rootcnt = STRUCT_FGET(usi, s_rootcnt);
1573 1581 sp2[i].s_rootnames = STRUCT_FGETP(usi, s_rootnames);
1574 1582 }
1575 1583 kmem_free(sp, allocsize);
1576 1584 sp = sp2;
1577 1585 allocsize = allocsize2;
1578 1586 }
1579 1587
1580 1588 kex->ex_secinfo = sp;
1581 1589
1582 1590 /*
1583 1591 * And now copy rootnames for each individual secinfo.
1584 1592 */
1585 1593 callback = 0;
1586 1594 allocd_seccnt = 0;
1587 1595 while (allocd_seccnt < kex->ex_seccnt) {
1588 1596
1589 1597 exs = &sp[allocd_seccnt];
1590 1598 if (exs->s_rootcnt > 0) {
1591 1599 if (!sec_svc_loadrootnames(exs->s_secinfo.sc_rpcnum,
1592 1600 exs->s_rootcnt, &exs->s_rootnames, model)) {
1593 1601 error = EFAULT;
1594 1602 goto out5;
1595 1603 }
1596 1604 }
1597 1605
1598 1606 if (exs->s_secinfo.sc_rpcnum == RPCSEC_GSS) {
1599 1607 rpc_gss_OID mech_tmp;
1600 1608 STRUCT_DECL(rpc_gss_OID_s, umech_tmp);
1601 1609 caddr_t elements_tmp;
1602 1610
1603 1611 /* Copyin mechanism type */
1604 1612 STRUCT_INIT(umech_tmp, model);
1605 1613 mech_tmp = kmem_alloc(sizeof (*mech_tmp), KM_SLEEP);
1606 1614 if (copyin(exs->s_secinfo.sc_gss_mech_type,
1607 1615 STRUCT_BUF(umech_tmp), STRUCT_SIZE(umech_tmp))) {
1608 1616 kmem_free(mech_tmp, sizeof (*mech_tmp));
1609 1617 error = EFAULT;
1610 1618 goto out5;
1611 1619 }
1612 1620 mech_tmp->length = STRUCT_FGET(umech_tmp, length);
1613 1621 mech_tmp->elements = STRUCT_FGETP(umech_tmp, elements);
1614 1622
1615 1623 elements_tmp = kmem_alloc(mech_tmp->length, KM_SLEEP);
1616 1624 if (copyin(mech_tmp->elements, elements_tmp,
1617 1625 mech_tmp->length)) {
1618 1626 kmem_free(elements_tmp, mech_tmp->length);
1619 1627 kmem_free(mech_tmp, sizeof (*mech_tmp));
1620 1628 error = EFAULT;
1621 1629 goto out5;
1622 1630 }
1623 1631 mech_tmp->elements = elements_tmp;
1624 1632 exs->s_secinfo.sc_gss_mech_type = mech_tmp;
1625 1633 allocd_seccnt++;
1626 1634
1627 1635 callback = 1;
1628 1636 } else
1629 1637 allocd_seccnt++;
1630 1638 }
1631 1639
1632 1640 /*
1633 1641 * Init the secinfo reference count and mark these flavors
1634 1642 * explicitly exported flavors.
1635 1643 */
1636 1644 for (i = 0; i < kex->ex_seccnt; i++) {
1637 1645 kex->ex_secinfo[i].s_flags |= M_4SEC_EXPORTED;
1638 1646 kex->ex_secinfo[i].s_refcnt = 1;
1639 1647 }
1640 1648
1641 1649 /*
1642 1650 * Set up rpcsec_gss callback routine entry if any.
1643 1651 */
1644 1652 if (callback) {
1645 1653 cb.callback = rfs_gsscallback;
1646 1654 cb.program = NFS_ACL_PROGRAM;
1647 1655 for (cb.version = NFS_ACL_VERSMIN;
1648 1656 cb.version <= NFS_ACL_VERSMAX; cb.version++) {
1649 1657 (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1650 1658 (void *)&cb);
1651 1659 }
1652 1660
1653 1661 cb.program = NFS_PROGRAM;
1654 1662 for (cb.version = NFS_VERSMIN;
1655 1663 cb.version <= NFS_VERSMAX; cb.version++) {
1656 1664 (void) sec_svc_control(RPC_SVC_SET_GSS_CALLBACK,
1657 1665 (void *)&cb);
1658 1666 }
1659 1667 }
1660 1668
1661 1669 /*
1662 1670 * Check the index flag. Do this here to avoid holding the
1663 1671 * lock while dealing with the index option (as we do with
1664 1672 * the public option).
1665 1673 */
1666 1674 if (kex->ex_flags & EX_INDEX) {
1667 1675 if (!kex->ex_index) { /* sanity check */
1668 1676 error = EINVAL;
1669 1677 goto out5;
1670 1678 }
1671 1679 if (error = loadindex(kex))
1672 1680 goto out5;
1673 1681 }
1674 1682
1675 1683 if (kex->ex_flags & EX_LOG) {
1676 1684 if (error = nfslog_setup(exi))
1677 1685 goto out6;
1678 1686 }
1679 1687
1680 1688 /*
1681 1689 * Insert the new entry at the front of the export list
1682 1690 */
1683 1691 rw_enter(&ne->exported_lock, RW_WRITER);
1684 1692 DTRACE_PROBE(nfss__i__exported_lock3_start);
1685 1693
1686 1694 export_link(ne, exi);
1687 1695
1688 1696 /*
1689 1697 * Check the rest of the list for an old entry for the fs.
1690 1698 * If one is found then unlink it, wait until this is the
1691 1699 * only reference and then free it.
1692 1700 */
1693 1701 for (ex = exi->fid_hash.next; ex != NULL; ex = ex->fid_hash.next) {
1694 1702 if (ex != ne->exi_root && VN_CMP(ex->exi_vp, vp)) {
1695 1703 mutex_enter(&nfs_exi_id_lock);
1696 1704 avl_remove(&exi_id_tree, ex);
1697 1705 mutex_exit(&nfs_exi_id_lock);
1698 1706 export_unlink(ne, ex);
1699 1707 break;
1700 1708 }
1701 1709 }
1702 1710
1703 1711 /*
1704 1712 * If the public filehandle is pointing at the
1705 1713 * old entry, then point it back at the root.
1706 1714 */
1707 1715 if (ex != NULL && ex == ne->exi_public)
1708 1716 ne->exi_public = ne->exi_root;
1709 1717
1710 1718 /*
1711 1719 * If the public flag is on, make the global exi_public
1712 1720 * point to this entry and turn off the public bit so that
1713 1721 * we can distinguish it from the place holder export.
1714 1722 */
1715 1723 if (kex->ex_flags & EX_PUBLIC) {
1716 1724 ne->exi_public = exi;
1717 1725 kex->ex_flags &= ~EX_PUBLIC;
1718 1726 }
1719 1727
1720 1728 #ifdef VOLATILE_FH_TEST
1721 1729 /*
1722 1730 * Set up the volatile_id value if volatile on share.
1723 1731 * The list of volatile renamed filehandles is always destroyed,
1724 1732 * if the fs was reshared.
1725 1733 */
1726 1734 if (kex->ex_flags & EX_VOLFH)
1727 1735 exi->exi_volatile_id = gethrestime_sec();
1728 1736
1729 1737 mutex_init(&exi->exi_vol_rename_lock, NULL, MUTEX_DEFAULT, NULL);
1730 1738 #endif /* VOLATILE_FH_TEST */
1731 1739
1732 1740 /*
1733 1741 * If this is a new export, then climb up
1734 1742 * the tree and check if any pseudo exports
1735 1743 * need to be created to provide a path for
1736 1744 * NFS v4 clients.
1737 1745 */
1738 1746 if (ex == NULL) {
1739 1747 error = treeclimb_export(exi);
1740 1748 if (error)
1741 1749 goto out7;
1742 1750 } else {
1743 1751 /* If it's a re-export update namespace tree */
1744 1752 exi->exi_tree = ex->exi_tree;
1745 1753 exi->exi_tree->tree_exi = exi;
1746 1754
1747 1755 /* Update the change timestamp */
1748 1756 tree_update_change(ne, exi->exi_tree, NULL);
1749 1757 }
1750 1758
1751 1759 /*
1752 1760 * build a unique flavor list from the flavors specified
1753 1761 * in the share cmd. unique means that each flavor only
1754 1762 * appears once in the secinfo list -- no duplicates allowed.
1755 1763 */
1756 1764 newcnt = build_seclist_nodups(&exi->exi_export, newsec, FALSE);
1757 1765
1758 1766 srv_secinfo_treeclimb(ne, exi, newsec, newcnt, TRUE);
1759 1767
1760 1768 /*
1761 1769 * If re-sharing an old export entry, update the secinfo data
1762 1770 * depending on if the old entry is a pseudo node or not.
1763 1771 */
1764 1772 if (ex != NULL) {
1765 1773 oldcnt = build_seclist_nodups(&ex->exi_export, oldsec, FALSE);
1766 1774 if (PSEUDO(ex)) {
1767 1775 /*
1768 1776 * The dir being shared is a pseudo export root (which
1769 1777 * will be transformed into a real export root). The
1770 1778 * flavor(s) of the new share were propagated to the
1771 1779 * ancestors by srv_secinfo_treeclimb() above. Now
1772 1780 * transfer the implicit flavor refs from the old
1773 1781 * pseudo exprot root to the new (real) export root.
1774 1782 */
1775 1783 srv_secinfo_add(&exi->exi_export.ex_secinfo,
1776 1784 &exi->exi_export.ex_seccnt, oldsec, oldcnt, TRUE);
1777 1785 } else {
1778 1786 /*
1779 1787 * First transfer implicit flavor refs to new export.
1780 1788 * Remove old flavor refs last.
1781 1789 */
1782 1790 srv_secinfo_exp2exp(&exi->exi_export, oldsec, oldcnt);
1783 1791 srv_secinfo_treeclimb(ne, ex, oldsec, oldcnt, FALSE);
1784 1792 }
1785 1793 }
1786 1794
1787 1795 /*
1788 1796 * If it's a re-export and the old entry has a pseudonode list,
1789 1797 * transfer it to the new export.
1790 1798 */
1791 1799 if (ex != NULL && (ex->exi_visible != NULL)) {
1792 1800 exi->exi_visible = ex->exi_visible;
1793 1801 ex->exi_visible = NULL;
1794 1802 }
1795 1803
1796 1804 /*
1797 1805 * Initialize exi_id and exi_kstats
1798 1806 */
1799 1807 if (ex != NULL) {
1800 1808 exi->exi_id = ex->exi_id;
1801 1809 } else {
1802 1810 mutex_enter(&nfs_exi_id_lock);
1803 1811 exi->exi_id = exi_id_get_next();
1804 1812 mutex_exit(&nfs_exi_id_lock);
1805 1813 }
1806 1814 mutex_enter(&nfs_exi_id_lock);
1807 1815 avl_add(&exi_id_tree, exi);
1808 1816 mutex_exit(&nfs_exi_id_lock);
1809 1817
1810 1818 DTRACE_PROBE(nfss__i__exported_lock3_stop);
1811 1819 rw_exit(&ne->exported_lock);
1812 1820
1813 1821 if (ne->exi_public == exi || kex->ex_flags & EX_LOG) {
1814 1822 /*
1815 1823 * Log share operation to this buffer only.
1816 1824 */
1817 1825 nfslog_share_record(exi, cr);
1818 1826 }
1819 1827
1820 1828 if (ex != NULL)
1821 1829 exi_rele(ex);
1822 1830
1823 1831 return (0);
1824 1832
1825 1833 out7:
1826 1834 /* Unlink the new export in exptable. */
1827 1835 export_unlink(ne, exi);
1828 1836 DTRACE_PROBE(nfss__i__exported_lock3_stop);
1829 1837 rw_exit(&ne->exported_lock);
1830 1838 out6:
1831 1839 if (kex->ex_flags & EX_INDEX)
1832 1840 kmem_free(kex->ex_index, strlen(kex->ex_index) + 1);
1833 1841 out5:
1834 1842 /* free partially completed allocation */
1835 1843 while (--allocd_seccnt >= 0) {
1836 1844 exs = &kex->ex_secinfo[allocd_seccnt];
1837 1845 srv_secinfo_entry_free(exs);
1838 1846 }
1839 1847
1840 1848 if (kex->ex_secinfo) {
1841 1849 kmem_free(kex->ex_secinfo,
1842 1850 kex->ex_seccnt * sizeof (struct secinfo));
1843 1851 }
1844 1852
1845 1853 out4:
1846 1854 if ((kex->ex_flags & EX_LOG) && kex->ex_tag != NULL)
1847 1855 kmem_free(kex->ex_tag, kex->ex_taglen + 1);
1848 1856 out3:
1849 1857 if ((kex->ex_flags & EX_LOG) && kex->ex_log_buffer != NULL)
1850 1858 kmem_free(kex->ex_log_buffer, kex->ex_log_bufferlen + 1);
1851 1859 out2:
1852 1860 kmem_free(kex->ex_path, kex->ex_pathlen + 1);
1853 1861 out1:
1854 1862 VN_RELE(vp);
1855 1863 if (dvp != NULL)
1856 1864 VN_RELE(dvp);
1857 1865 mutex_destroy(&exi->exi_lock);
1858 1866 rw_destroy(&exi->exi_cache_lock);
1859 1867 for (i = 0; i < AUTH_TABLESIZE; i++) {
1860 1868 avl_destroy(exi->exi_cache[i]);
1861 1869 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
1862 1870 }
1863 1871
1864 1872 kmem_free(exi, sizeof (*exi));
1865 1873
1866 1874 return (error);
1867 1875 }
1868 1876
1869 1877 /*
1870 1878 * Remove the exportinfo from the export list
1871 1879 */
1872 1880 void
1873 1881 export_unlink(nfs_export_t *ne, struct exportinfo *exi)
1874 1882 {
1875 1883 ASSERT(RW_WRITE_HELD(&ne->exported_lock));
1876 1884
1877 1885 exp_hash_unlink(exi, fid_hash);
1878 1886 exp_hash_unlink(exi, path_hash);
1879 1887 }
1880 1888
1881 1889 /*
1882 1890 * Unexport an exported filesystem
1883 1891 */
1884 1892 static int
1885 1893 unexport(nfs_export_t *ne, struct exportinfo *exi)
1886 1894 {
1887 1895 struct secinfo cursec[MAX_FLAVORS];
1888 1896 int curcnt;
1889 1897
1890 1898 rw_enter(&ne->exported_lock, RW_WRITER);
1891 1899
1892 1900 /* Check if exi is still linked in the export table */
1893 1901 if (!EXP_LINKED(exi) || PSEUDO(exi)) {
1894 1902 rw_exit(&ne->exported_lock);
1895 1903 return (EINVAL);
1896 1904 }
1897 1905
1898 1906 mutex_enter(&nfs_exi_id_lock);
1899 1907 avl_remove(&exi_id_tree, exi);
1900 1908 mutex_exit(&nfs_exi_id_lock);
1901 1909 export_unlink(ne, exi);
1902 1910
1903 1911 /*
1904 1912 * Remove security flavors before treeclimb_unexport() is called
1905 1913 * because srv_secinfo_treeclimb needs the namespace tree
1906 1914 */
1907 1915 curcnt = build_seclist_nodups(&exi->exi_export, cursec, TRUE);
1908 1916 srv_secinfo_treeclimb(ne, exi, cursec, curcnt, FALSE);
1909 1917
1910 1918 /*
1911 1919 * If there's a visible list, then need to leave
1912 1920 * a pseudo export here to retain the visible list
|
↓ open down ↓ |
973 lines elided |
↑ open up ↑ |
1913 1921 * for paths to exports below.
1914 1922 */
1915 1923 if (exi->exi_visible != NULL) {
1916 1924 struct exportinfo *newexi;
1917 1925
1918 1926 newexi = pseudo_exportfs(ne, exi->exi_vp, &exi->exi_fid,
1919 1927 exi->exi_visible, &exi->exi_export);
1920 1928 exi->exi_visible = NULL;
1921 1929
1922 1930 /* interconnect the existing treenode with the new exportinfo */
1923 - newexi->exi_zone = exi->exi_zone;
1931 + newexi->exi_zone = exi->exi_zone; /* XXX KEBE SAYS LOSE ME */
1924 1932 newexi->exi_tree = exi->exi_tree;
1925 1933 newexi->exi_tree->tree_exi = newexi;
1926 1934
1927 1935 /* Update the change timestamp */
1928 1936 tree_update_change(ne, exi->exi_tree, NULL);
1929 1937 } else {
1930 1938 treeclimb_unexport(ne, exi);
1931 1939 }
1932 1940
1933 1941 rw_exit(&ne->exported_lock);
1934 1942
1935 1943 /*
1936 1944 * Need to call into the NFSv4 server and release all data
1937 1945 * held on this particular export. This is important since
1938 1946 * the v4 server may be holding file locks or vnodes under
1939 1947 * this export.
1940 1948 */
1941 1949 rfs4_clean_state_exi(ne, exi);
1942 1950
1943 1951 /*
1944 1952 * Notify the lock manager that the filesystem is being
1945 1953 * unexported.
1946 1954 */
1947 1955 lm_unexport(exi);
1948 1956
1949 1957 /*
1950 1958 * If this was a public export, restore
1951 1959 * the public filehandle to the root.
1952 1960 */
1953 1961
1954 1962 /*
1955 1963 * XXX KEBE ASKS --> Should CRED() instead be
1956 1964 * exi->exi_zone->zone_kcred?
1957 1965 */
1958 1966 if (exi == ne->exi_public) {
1959 1967 ne->exi_public = ne->exi_root;
1960 1968
1961 1969 nfslog_share_record(ne->exi_public, CRED());
1962 1970 }
1963 1971
1964 1972 if (exi->exi_export.ex_flags & EX_LOG)
1965 1973 nfslog_unshare_record(exi, CRED());
1966 1974
1967 1975 exi_rele(exi);
1968 1976 return (0);
1969 1977 }
1970 1978
1971 1979 /*
1972 1980 * Get file handle system call.
1973 1981 * Takes file name and returns a file handle for it.
1974 1982 * Credentials must be verified before calling.
1975 1983 */
1976 1984 int
1977 1985 nfs_getfh(struct nfs_getfh_args *args, model_t model, cred_t *cr)
1978 1986 {
1979 1987 nfs_fh3 fh;
1980 1988 char buf[NFS3_MAXFHSIZE];
1981 1989 char *logptr, logbuf[NFS3_MAXFHSIZE];
1982 1990 int l = NFS3_MAXFHSIZE;
1983 1991 vnode_t *vp;
1984 1992 vnode_t *dvp;
1985 1993 struct exportinfo *exi;
1986 1994 int error;
1987 1995 int vers;
1988 1996 STRUCT_HANDLE(nfs_getfh_args, uap);
1989 1997
1990 1998 #ifdef lint
1991 1999 model = model; /* STRUCT macros don't always use it */
1992 2000 #endif
1993 2001
1994 2002 STRUCT_SET_HANDLE(uap, model, args);
1995 2003
1996 2004 error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
1997 2005 FOLLOW, &dvp, &vp);
1998 2006 if (error == EINVAL) {
1999 2007 /*
2000 2008 * if fname resolves to / we get EINVAL error
2001 2009 * since we wanted the parent vnode. Try again
2002 2010 * with NULL dvp.
2003 2011 */
2004 2012 error = lookupname(STRUCT_FGETP(uap, fname), UIO_USERSPACE,
2005 2013 FOLLOW, NULL, &vp);
2006 2014 dvp = NULL;
2007 2015 }
2008 2016 if (!error && vp == NULL) {
2009 2017 /*
2010 2018 * Last component of fname not found
2011 2019 */
2012 2020 if (dvp != NULL) {
2013 2021 VN_RELE(dvp);
2014 2022 }
2015 2023 error = ENOENT;
2016 2024 }
2017 2025 if (error)
2018 2026 return (error);
2019 2027
2020 2028 /*
2021 2029 * 'vp' may be an AUTOFS node, so we perform a
2022 2030 * VOP_ACCESS() to trigger the mount of the
2023 2031 * intended filesystem, so we can share the intended
2024 2032 * filesystem instead of the AUTOFS filesystem.
2025 2033 */
2026 2034 (void) VOP_ACCESS(vp, 0, 0, cr, NULL);
2027 2035
2028 2036 /*
2029 2037 * We're interested in the top most filesystem.
2030 2038 * This is specially important when uap->dname is a trigger
2031 2039 * AUTOFS node, since we're really interested in sharing the
2032 2040 * filesystem AUTOFS mounted as result of the VOP_ACCESS()
2033 2041 * call not the AUTOFS node itself.
2034 2042 */
2035 2043 if (vn_mountedvfs(vp) != NULL) {
2036 2044 if (error = traverse(&vp)) {
2037 2045 VN_RELE(vp);
2038 2046 if (dvp != NULL)
2039 2047 VN_RELE(dvp);
2040 2048 return (error);
2041 2049 }
2042 2050 }
2043 2051
2044 2052 vers = STRUCT_FGET(uap, vers);
2045 2053 exi = nfs_vptoexi(dvp, vp, cr, NULL, &error, FALSE);
2046 2054 if (!error) {
2047 2055 if (vers == NFS_VERSION) {
2048 2056 error = makefh((fhandle_t *)buf, vp, exi);
2049 2057 l = NFS_FHSIZE;
2050 2058 logptr = buf;
2051 2059 } else if (vers == NFS_V3) {
2052 2060 int i, sz, pad;
2053 2061
2054 2062 error = makefh3(&fh, vp, exi);
2055 2063 l = RNDUP(fh.fh3_length);
2056 2064 if (!error && (l > sizeof (fhandle3_t)))
2057 2065 error = EREMOTE;
2058 2066 logptr = logbuf;
2059 2067 if (!error) {
2060 2068 i = 0;
2061 2069 sz = sizeof (fsid_t);
2062 2070 bcopy(&fh.fh3_fsid, &buf[i], sz);
2063 2071 i += sz;
2064 2072
2065 2073 /*
2066 2074 * For backwards compatibility, the
2067 2075 * fid length may be less than
2068 2076 * NFS_FHMAXDATA, but it was always
2069 2077 * encoded as NFS_FHMAXDATA bytes.
2070 2078 */
2071 2079
2072 2080 sz = sizeof (ushort_t);
2073 2081 bcopy(&fh.fh3_len, &buf[i], sz);
2074 2082 i += sz;
2075 2083 bcopy(fh.fh3_data, &buf[i], fh.fh3_len);
2076 2084 i += fh.fh3_len;
2077 2085 pad = (NFS_FHMAXDATA - fh.fh3_len);
2078 2086 if (pad > 0) {
2079 2087 bzero(&buf[i], pad);
2080 2088 i += pad;
2081 2089 l += pad;
2082 2090 }
2083 2091
2084 2092 sz = sizeof (ushort_t);
2085 2093 bcopy(&fh.fh3_xlen, &buf[i], sz);
2086 2094 i += sz;
2087 2095 bcopy(fh.fh3_xdata, &buf[i], fh.fh3_xlen);
2088 2096 i += fh.fh3_xlen;
2089 2097 pad = (NFS_FHMAXDATA - fh.fh3_xlen);
2090 2098 if (pad > 0) {
2091 2099 bzero(&buf[i], pad);
2092 2100 i += pad;
2093 2101 l += pad;
2094 2102 }
2095 2103 }
2096 2104 /*
2097 2105 * If we need to do NFS logging, the filehandle
2098 2106 * must be downsized to 32 bytes.
2099 2107 */
2100 2108 if (!error && exi->exi_export.ex_flags & EX_LOG) {
2101 2109 i = 0;
2102 2110 sz = sizeof (fsid_t);
2103 2111 bcopy(&fh.fh3_fsid, &logbuf[i], sz);
2104 2112 i += sz;
2105 2113 sz = sizeof (ushort_t);
2106 2114 bcopy(&fh.fh3_len, &logbuf[i], sz);
2107 2115 i += sz;
2108 2116 sz = NFS_FHMAXDATA;
2109 2117 bcopy(fh.fh3_data, &logbuf[i], sz);
2110 2118 i += sz;
2111 2119 sz = sizeof (ushort_t);
2112 2120 bcopy(&fh.fh3_xlen, &logbuf[i], sz);
2113 2121 i += sz;
2114 2122 sz = NFS_FHMAXDATA;
2115 2123 bcopy(fh.fh3_xdata, &logbuf[i], sz);
2116 2124 i += sz;
2117 2125 }
2118 2126 }
2119 2127 if (!error && exi->exi_export.ex_flags & EX_LOG) {
2120 2128 nfslog_getfh(exi, (fhandle_t *)logptr,
2121 2129 STRUCT_FGETP(uap, fname), UIO_USERSPACE, cr);
2122 2130 }
2123 2131 exi_rele(exi);
2124 2132 if (!error) {
2125 2133 if (copyout(&l, STRUCT_FGETP(uap, lenp), sizeof (int)))
2126 2134 error = EFAULT;
2127 2135 if (copyout(buf, STRUCT_FGETP(uap, fhp), l))
2128 2136 error = EFAULT;
2129 2137 }
2130 2138 }
2131 2139 VN_RELE(vp);
2132 2140 if (dvp != NULL) {
2133 2141 VN_RELE(dvp);
2134 2142 }
2135 2143 return (error);
2136 2144 }
2137 2145
2138 2146 /*
2139 2147 * Strategy: if vp is in the export list, then
2140 2148 * return the associated file handle. Otherwise, ".."
2141 2149 * once up the vp and try again, until the root of the
2142 2150 * filesystem is reached.
2143 2151 */
2144 2152 struct exportinfo *
2145 2153 nfs_vptoexi(vnode_t *dvp, vnode_t *vp, cred_t *cr, int *walk,
2146 2154 int *err, bool_t v4srv)
2147 2155 {
2148 2156 fid_t fid;
2149 2157 int error;
2150 2158 struct exportinfo *exi;
2151 2159
2152 2160 ASSERT(vp);
2153 2161 VN_HOLD(vp);
2154 2162 if (dvp != NULL) {
2155 2163 VN_HOLD(dvp);
2156 2164 }
2157 2165 if (walk != NULL)
2158 2166 *walk = 0;
2159 2167
2160 2168 for (;;) {
2161 2169 bzero(&fid, sizeof (fid));
2162 2170 fid.fid_len = MAXFIDSZ;
2163 2171 error = vop_fid_pseudo(vp, &fid);
2164 2172 if (error) {
2165 2173 /*
2166 2174 * If vop_fid_pseudo returns ENOSPC then the fid
2167 2175 * supplied is too small. For now we simply
2168 2176 * return EREMOTE.
2169 2177 */
2170 2178 if (error == ENOSPC)
2171 2179 error = EREMOTE;
2172 2180 break;
2173 2181 }
2174 2182
2175 2183 if (v4srv)
2176 2184 exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
2177 2185 else
2178 2186 exi = checkexport(&vp->v_vfsp->vfs_fsid, &fid);
2179 2187
2180 2188 if (exi != NULL) {
2181 2189 /*
|
↓ open down ↓ |
248 lines elided |
↑ open up ↑ |
2182 2190 * Found the export info
2183 2191 */
2184 2192 break;
2185 2193 }
2186 2194
2187 2195 /*
2188 2196 * We have just failed finding a matching export.
2189 2197 * If we're at the root of this filesystem, then
2190 2198 * it's time to stop (with failure).
2191 2199 */
2200 + ASSERT3P(vp->v_vfsp->vfs_zone, ==, curzone);
2192 2201 if ((vp->v_flag & VROOT) || VN_IS_CURZONEROOT(vp)) {
2193 2202 error = EINVAL;
2194 2203 break;
2195 2204 }
2196 2205
2197 2206 if (walk != NULL)
2198 2207 (*walk)++;
2199 2208
2200 2209 /*
2201 2210 * Now, do a ".." up vp. If dvp is supplied, use it,
2202 2211 * otherwise, look it up.
2203 2212 */
2204 2213 if (dvp == NULL) {
2205 2214 error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, cr,
2206 2215 NULL, NULL, NULL);
2207 2216 if (error)
2208 2217 break;
2209 2218 }
2210 2219 VN_RELE(vp);
2211 2220 vp = dvp;
2212 2221 dvp = NULL;
2213 2222 }
2214 2223 VN_RELE(vp);
2215 2224 if (dvp != NULL) {
2216 2225 VN_RELE(dvp);
2217 2226 }
2218 2227 if (error != 0) {
2219 2228 if (err != NULL)
2220 2229 *err = error;
2221 2230 return (NULL);
2222 2231 }
2223 2232 return (exi);
2224 2233 }
2225 2234
2226 2235 int
2227 2236 chk_clnt_sec(exportinfo_t *exi, struct svc_req *req)
2228 2237 {
2229 2238 int i, nfsflavor;
2230 2239 struct secinfo *sp;
2231 2240
2232 2241 /*
2233 2242 * Get the nfs flavor number from xprt.
2234 2243 */
2235 2244 nfsflavor = (int)(uintptr_t)req->rq_xprt->xp_cookie;
2236 2245
2237 2246 sp = exi->exi_export.ex_secinfo;
2238 2247 for (i = 0; i < exi->exi_export.ex_seccnt; i++) {
2239 2248 if ((nfsflavor == sp[i].s_secinfo.sc_nfsnum) &&
2240 2249 SEC_REF_EXPORTED(sp + i))
2241 2250 return (TRUE);
2242 2251 }
2243 2252 return (FALSE);
2244 2253 }
2245 2254
2246 2255 /*
2247 2256 * Make an fhandle from a vnode
2248 2257 */
2249 2258 int
2250 2259 makefh(fhandle_t *fh, vnode_t *vp, exportinfo_t *exi)
2251 2260 {
2252 2261 int error;
2253 2262
2254 2263 *fh = exi->exi_fh; /* struct copy */
2255 2264
2256 2265 error = VOP_FID(vp, (fid_t *)&fh->fh_len, NULL);
2257 2266 if (error) {
2258 2267 /*
2259 2268 * Should be something other than EREMOTE
2260 2269 */
2261 2270 return (EREMOTE);
2262 2271 }
2263 2272 return (0);
2264 2273 }
2265 2274
2266 2275 /*
2267 2276 * This routine makes an overloaded V2 fhandle which contains
2268 2277 * sec modes.
2269 2278 *
2270 2279 * Note that the first four octets contain the length octet,
2271 2280 * the status octet, and two padded octets to make them XDR
2272 2281 * four-octet aligned.
2273 2282 *
2274 2283 * 1 2 3 4 32
2275 2284 * +---+---+---+---+---+---+---+---+ +---+---+---+---+ +---+
2276 2285 * | l | s | | | sec_1 |...| sec_n |...| |
2277 2286 * +---+---+---+---+---+---+---+---+ +---+---+---+---+ +---+
2278 2287 *
2279 2288 * where
2280 2289 *
2281 2290 * the status octet s indicates whether there are more security
2282 2291 * flavors (1 means yes, 0 means no) that require the client to
2283 2292 * perform another 0x81 LOOKUP to get them,
2284 2293 *
2285 2294 * the length octet l is the length describing the number of
2286 2295 * valid octets that follow. (l = 4 * n, where n is the number
2287 2296 * of security flavors sent in the current overloaded filehandle.)
2288 2297 *
2289 2298 * sec_index should always be in the inclusive range: [1 - ex_seccnt],
2290 2299 * and it tells server where to start within the secinfo array.
2291 2300 * Usually it will always be 1; however, if more flavors are used
2292 2301 * for the public export than can be encoded in the overloaded FH
2293 2302 * (7 for NFS2), subsequent SNEGO MCLs will have a larger index
2294 2303 * so the server will pick up where it left off from the previous
2295 2304 * MCL reply.
2296 2305 *
2297 2306 * With NFS4 support, implicitly allowed flavors are also in
2298 2307 * the secinfo array; however, they should not be returned in
2299 2308 * SNEGO MCL replies.
2300 2309 */
2301 2310 int
2302 2311 makefh_ol(fhandle_t *fh, exportinfo_t *exi, uint_t sec_index)
2303 2312 {
2304 2313 secinfo_t sec[MAX_FLAVORS];
2305 2314 int totalcnt, i, *ipt, cnt, seccnt, secidx, fh_max_cnt;
2306 2315 char *c;
2307 2316
2308 2317 if (fh == NULL || exi == NULL || sec_index < 1)
2309 2318 return (EREMOTE);
2310 2319
2311 2320 /*
2312 2321 * WebNFS clients need to know the unique set of explicitly
2313 2322 * shared flavors in used for the public export. When
2314 2323 * "TRUE" is passed to build_seclist_nodups(), only explicitly
2315 2324 * shared flavors are included in the list.
2316 2325 */
2317 2326 seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2318 2327 if (sec_index > seccnt)
2319 2328 return (EREMOTE);
2320 2329
2321 2330 fh_max_cnt = (NFS_FHSIZE / sizeof (int)) - 1;
2322 2331 totalcnt = seccnt - sec_index + 1;
2323 2332 cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2324 2333
2325 2334 c = (char *)fh;
2326 2335 /*
2327 2336 * Encode the length octet representing the number of
2328 2337 * security flavors (in bytes) in this overloaded fh.
2329 2338 */
2330 2339 *c = cnt * sizeof (int);
2331 2340
2332 2341 /*
2333 2342 * Encode the status octet that indicates whether there
2334 2343 * are more security flavors the client needs to get.
2335 2344 */
2336 2345 *(c + 1) = totalcnt > fh_max_cnt;
2337 2346
2338 2347 /*
2339 2348 * put security flavors in the overloaded fh
2340 2349 */
2341 2350 ipt = (int *)(c + sizeof (int32_t));
2342 2351 secidx = sec_index - 1;
2343 2352 for (i = 0; i < cnt; i++) {
2344 2353 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2345 2354 }
2346 2355 return (0);
2347 2356 }
2348 2357
2349 2358 /*
2350 2359 * Make an nfs_fh3 from a vnode
2351 2360 */
2352 2361 int
2353 2362 makefh3(nfs_fh3 *fh, vnode_t *vp, struct exportinfo *exi)
2354 2363 {
2355 2364 int error;
2356 2365 fid_t fid;
2357 2366
2358 2367 bzero(&fid, sizeof (fid));
2359 2368 fid.fid_len = sizeof (fh->fh3_data);
2360 2369 error = VOP_FID(vp, &fid, NULL);
2361 2370 if (error)
2362 2371 return (EREMOTE);
2363 2372
2364 2373 bzero(fh, sizeof (nfs_fh3));
2365 2374 fh->fh3_fsid = exi->exi_fsid;
2366 2375 fh->fh3_len = fid.fid_len;
2367 2376 bcopy(fid.fid_data, fh->fh3_data, fh->fh3_len);
2368 2377
2369 2378 fh->fh3_xlen = exi->exi_fid.fid_len;
2370 2379 ASSERT(fh->fh3_xlen <= sizeof (fh->fh3_xdata));
2371 2380 bcopy(exi->exi_fid.fid_data, fh->fh3_xdata, fh->fh3_xlen);
2372 2381
2373 2382 fh->fh3_length = sizeof (fh->fh3_fsid)
2374 2383 + sizeof (fh->fh3_len) + fh->fh3_len
2375 2384 + sizeof (fh->fh3_xlen) + fh->fh3_xlen;
2376 2385 fh->fh3_flags = 0;
2377 2386
2378 2387 return (0);
2379 2388 }
2380 2389
2381 2390 /*
2382 2391 * This routine makes an overloaded V3 fhandle which contains
2383 2392 * sec modes.
2384 2393 *
2385 2394 * 1 4
2386 2395 * +--+--+--+--+
2387 2396 * | len |
2388 2397 * +--+--+--+--+
2389 2398 * up to 64
2390 2399 * +--+--+--+--+--+--+--+--+--+--+--+--+ +--+--+--+--+
2391 2400 * |s | | | | sec_1 | sec_2 | ... | sec_n |
2392 2401 * +--+--+--+--+--+--+--+--+--+--+--+--+ +--+--+--+--+
2393 2402 *
2394 2403 * len = 4 * (n+1), where n is the number of security flavors
2395 2404 * sent in the current overloaded filehandle.
2396 2405 *
2397 2406 * the status octet s indicates whether there are more security
2398 2407 * mechanisms (1 means yes, 0 means no) that require the client
2399 2408 * to perform another 0x81 LOOKUP to get them.
2400 2409 *
2401 2410 * Three octets are padded after the status octet.
2402 2411 */
2403 2412 int
2404 2413 makefh3_ol(nfs_fh3 *fh, struct exportinfo *exi, uint_t sec_index)
2405 2414 {
2406 2415 secinfo_t sec[MAX_FLAVORS];
2407 2416 int totalcnt, cnt, *ipt, i, seccnt, fh_max_cnt, secidx;
2408 2417 char *c;
2409 2418
2410 2419 if (fh == NULL || exi == NULL || sec_index < 1)
2411 2420 return (EREMOTE);
2412 2421
2413 2422 /*
2414 2423 * WebNFS clients need to know the unique set of explicitly
2415 2424 * shared flavors in used for the public export. When
2416 2425 * "TRUE" is passed to build_seclist_nodups(), only explicitly
2417 2426 * shared flavors are included in the list.
2418 2427 */
2419 2428 seccnt = build_seclist_nodups(&exi->exi_export, sec, TRUE);
2420 2429
2421 2430 if (sec_index > seccnt)
2422 2431 return (EREMOTE);
2423 2432
2424 2433 fh_max_cnt = (NFS3_FHSIZE / sizeof (int)) - 1;
2425 2434 totalcnt = seccnt - sec_index + 1;
2426 2435 cnt = totalcnt > fh_max_cnt ? fh_max_cnt : totalcnt;
2427 2436
2428 2437 /*
2429 2438 * Place the length in fh3_length representing the number
2430 2439 * of security flavors (in bytes) in this overloaded fh.
2431 2440 */
2432 2441 fh->fh3_flags = FH_WEBNFS;
2433 2442 fh->fh3_length = (cnt+1) * sizeof (int32_t);
2434 2443
2435 2444 c = (char *)&fh->fh3_u.nfs_fh3_i.fh3_i;
2436 2445 /*
2437 2446 * Encode the status octet that indicates whether there
2438 2447 * are more security flavors the client needs to get.
2439 2448 */
2440 2449 *c = totalcnt > fh_max_cnt;
2441 2450
2442 2451 /*
2443 2452 * put security flavors in the overloaded fh
2444 2453 */
2445 2454 secidx = sec_index - 1;
2446 2455 ipt = (int *)(c + sizeof (int32_t));
2447 2456 for (i = 0; i < cnt; i++) {
2448 2457 ipt[i] = htonl(sec[i + secidx].s_secinfo.sc_nfsnum);
2449 2458 }
2450 2459 return (0);
2451 2460 }
2452 2461
2453 2462 /*
2454 2463 * Make an nfs_fh4 from a vnode
2455 2464 */
2456 2465 int
2457 2466 makefh4(nfs_fh4 *fh, vnode_t *vp, struct exportinfo *exi)
2458 2467 {
2459 2468 int error;
2460 2469 nfs_fh4_fmt_t *fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2461 2470 fid_t fid;
2462 2471
2463 2472 bzero(&fid, sizeof (fid));
2464 2473 fid.fid_len = MAXFIDSZ;
2465 2474 /*
2466 2475 * vop_fid_pseudo() is used to set up NFSv4 namespace, so
2467 2476 * use vop_fid_pseudo() here to get the fid instead of VOP_FID.
2468 2477 */
2469 2478 error = vop_fid_pseudo(vp, &fid);
2470 2479 if (error)
2471 2480 return (error);
2472 2481
2473 2482 fh->nfs_fh4_len = NFS_FH4_LEN;
2474 2483
2475 2484 fh_fmtp->fh4_i.fhx_fsid = exi->exi_fh.fh_fsid;
2476 2485 fh_fmtp->fh4_i.fhx_xlen = exi->exi_fh.fh_xlen;
2477 2486
2478 2487 bzero(fh_fmtp->fh4_i.fhx_data, sizeof (fh_fmtp->fh4_i.fhx_data));
2479 2488 bzero(fh_fmtp->fh4_i.fhx_xdata, sizeof (fh_fmtp->fh4_i.fhx_xdata));
2480 2489 ASSERT(exi->exi_fh.fh_xlen <= sizeof (fh_fmtp->fh4_i.fhx_xdata));
2481 2490 bcopy(exi->exi_fh.fh_xdata, fh_fmtp->fh4_i.fhx_xdata,
2482 2491 exi->exi_fh.fh_xlen);
2483 2492
2484 2493 fh_fmtp->fh4_len = fid.fid_len;
2485 2494 ASSERT(fid.fid_len <= sizeof (fh_fmtp->fh4_data));
2486 2495 bcopy(fid.fid_data, fh_fmtp->fh4_data, fid.fid_len);
2487 2496 fh_fmtp->fh4_flag = 0;
2488 2497
2489 2498 #ifdef VOLATILE_FH_TEST
2490 2499 /*
2491 2500 * XXX (temporary?)
2492 2501 * Use the rnode volatile_id value to add volatility to the fh.
2493 2502 *
2494 2503 * For testing purposes there are currently two scenarios, based
2495 2504 * on whether the filesystem was shared with "volatile_fh"
2496 2505 * or "expire_on_rename". In the first case, use the value of
2497 2506 * export struct share_time as the volatile_id. In the second
2498 2507 * case use the vnode volatile_id value (which is set to the
2499 2508 * time in which the file was renamed).
2500 2509 *
2501 2510 * Note that the above are temporary constructs for testing only
2502 2511 * XXX
2503 2512 */
2504 2513 if (exi->exi_export.ex_flags & EX_VOLRNM) {
2505 2514 fh_fmtp->fh4_volatile_id = find_volrnm_fh_id(exi, fh);
2506 2515 } else if (exi->exi_export.ex_flags & EX_VOLFH) {
2507 2516 fh_fmtp->fh4_volatile_id = exi->exi_volatile_id;
2508 2517 } else {
2509 2518 fh_fmtp->fh4_volatile_id = 0;
2510 2519 }
2511 2520 #endif /* VOLATILE_FH_TEST */
2512 2521
2513 2522 return (0);
2514 2523 }
2515 2524
2516 2525 /*
2517 2526 * Convert an fhandle into a vnode.
2518 2527 * Uses the file id (fh_len + fh_data) in the fhandle to get the vnode.
2519 2528 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2520 2529 * are done with it.
2521 2530 */
2522 2531 vnode_t *
2523 2532 nfs_fhtovp(fhandle_t *fh, struct exportinfo *exi)
2524 2533 {
2525 2534 vfs_t *vfsp;
2526 2535 vnode_t *vp;
2527 2536 int error;
2528 2537 fid_t *fidp;
2529 2538
2530 2539 TRACE_0(TR_FAC_NFS, TR_FHTOVP_START,
2531 2540 "fhtovp_start");
2532 2541
2533 2542 if (exi == NULL) {
2534 2543 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2535 2544 "fhtovp_end:(%S)", "exi NULL");
2536 2545 return (NULL); /* not exported */
2537 2546 }
2538 2547
2539 2548 ASSERT(exi->exi_vp != NULL);
2540 2549
2541 2550 if (PUBLIC_FH2(fh)) {
2542 2551 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2543 2552 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2544 2553 "fhtovp_end:(%S)", "root not exported");
2545 2554 return (NULL);
2546 2555 }
2547 2556 vp = exi->exi_vp;
2548 2557 VN_HOLD(vp);
2549 2558 return (vp);
2550 2559 }
2551 2560
2552 2561 vfsp = exi->exi_vp->v_vfsp;
2553 2562 ASSERT(vfsp != NULL);
2554 2563 fidp = (fid_t *)&fh->fh_len;
2555 2564
2556 2565 error = VFS_VGET(vfsp, &vp, fidp);
2557 2566 if (error || vp == NULL) {
2558 2567 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2559 2568 "fhtovp_end:(%S)", "VFS_GET failed or vp NULL");
2560 2569 return (NULL);
2561 2570 }
2562 2571 TRACE_1(TR_FAC_NFS, TR_FHTOVP_END,
2563 2572 "fhtovp_end:(%S)", "end");
2564 2573 return (vp);
2565 2574 }
2566 2575
2567 2576 /*
2568 2577 * Convert an nfs_fh3 into a vnode.
2569 2578 * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2570 2579 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2571 2580 * are done with it.
2572 2581 */
2573 2582 vnode_t *
2574 2583 nfs3_fhtovp(nfs_fh3 *fh, struct exportinfo *exi)
2575 2584 {
2576 2585 vfs_t *vfsp;
2577 2586 vnode_t *vp;
2578 2587 int error;
2579 2588 fid_t *fidp;
2580 2589
2581 2590 if (exi == NULL)
2582 2591 return (NULL); /* not exported */
2583 2592
2584 2593 ASSERT(exi->exi_vp != NULL);
2585 2594
2586 2595 if (PUBLIC_FH3(fh)) {
2587 2596 if (exi->exi_export.ex_flags & EX_PUBLIC)
2588 2597 return (NULL);
2589 2598 vp = exi->exi_vp;
2590 2599 VN_HOLD(vp);
2591 2600 return (vp);
2592 2601 }
2593 2602
2594 2603 if (fh->fh3_length < NFS3_OLDFHSIZE ||
2595 2604 fh->fh3_length > NFS3_MAXFHSIZE)
2596 2605 return (NULL);
2597 2606
2598 2607 vfsp = exi->exi_vp->v_vfsp;
2599 2608 ASSERT(vfsp != NULL);
2600 2609 fidp = FH3TOFIDP(fh);
2601 2610
2602 2611 error = VFS_VGET(vfsp, &vp, fidp);
2603 2612 if (error || vp == NULL)
2604 2613 return (NULL);
2605 2614
2606 2615 return (vp);
2607 2616 }
2608 2617
2609 2618 /*
2610 2619 * Convert an nfs_fh4 into a vnode.
2611 2620 * Uses the file id (fh_len + fh_data) in the file handle to get the vnode.
2612 2621 * WARNING: users of this routine must do a VN_RELE on the vnode when they
2613 2622 * are done with it.
2614 2623 */
2615 2624 vnode_t *
2616 2625 nfs4_fhtovp(nfs_fh4 *fh, struct exportinfo *exi, nfsstat4 *statp)
2617 2626 {
2618 2627 vfs_t *vfsp;
2619 2628 vnode_t *vp = NULL;
2620 2629 int error;
2621 2630 fid_t *fidp;
2622 2631 nfs_fh4_fmt_t *fh_fmtp;
2623 2632 #ifdef VOLATILE_FH_TEST
2624 2633 uint32_t volatile_id = 0;
2625 2634 #endif /* VOLATILE_FH_TEST */
2626 2635
2627 2636 if (exi == NULL) {
2628 2637 *statp = NFS4ERR_STALE;
2629 2638 return (NULL); /* not exported */
2630 2639 }
2631 2640 ASSERT(exi->exi_vp != NULL);
2632 2641
2633 2642 /* caller should have checked this */
2634 2643 ASSERT(fh->nfs_fh4_len >= NFS_FH4_LEN);
2635 2644
2636 2645 fh_fmtp = (nfs_fh4_fmt_t *)fh->nfs_fh4_val;
2637 2646 vfsp = exi->exi_vp->v_vfsp;
2638 2647 ASSERT(vfsp != NULL);
2639 2648 fidp = (fid_t *)&fh_fmtp->fh4_len;
2640 2649
2641 2650 #ifdef VOLATILE_FH_TEST
2642 2651 /* XXX check if volatile - should be changed later */
2643 2652 if (exi->exi_export.ex_flags & (EX_VOLRNM | EX_VOLFH)) {
2644 2653 /*
2645 2654 * Filesystem is shared with volatile filehandles
2646 2655 */
2647 2656 if (exi->exi_export.ex_flags & EX_VOLRNM)
2648 2657 volatile_id = find_volrnm_fh_id(exi, fh);
2649 2658 else
2650 2659 volatile_id = exi->exi_volatile_id;
2651 2660
2652 2661 if (fh_fmtp->fh4_volatile_id != volatile_id) {
2653 2662 *statp = NFS4ERR_FHEXPIRED;
2654 2663 return (NULL);
2655 2664 }
2656 2665 }
2657 2666 /*
2658 2667 * XXX even if test_volatile_fh false, the fh may contain a
2659 2668 * volatile id if obtained when the test was set.
2660 2669 */
2661 2670 fh_fmtp->fh4_volatile_id = (uchar_t)0;
2662 2671 #endif /* VOLATILE_FH_TEST */
2663 2672
2664 2673 error = VFS_VGET(vfsp, &vp, fidp);
2665 2674 /*
2666 2675 * If we can not get vp from VFS_VGET, perhaps this is
2667 2676 * an nfs v2/v3/v4 node in an nfsv4 pseudo filesystem.
2668 2677 * Check it out.
2669 2678 */
2670 2679 if (error && PSEUDO(exi))
2671 2680 error = nfs4_vget_pseudo(exi, &vp, fidp);
2672 2681
2673 2682 if (error || vp == NULL) {
2674 2683 *statp = NFS4ERR_STALE;
2675 2684 return (NULL);
2676 2685 }
2677 2686 /* XXX - disgusting hack */
2678 2687 if (vp->v_type == VNON && vp->v_flag & V_XATTRDIR)
2679 2688 vp->v_type = VDIR;
2680 2689 *statp = NFS4_OK;
2681 2690 return (vp);
2682 2691 }
2683 2692
2684 2693 /*
2685 2694 * Find the export structure associated with the given filesystem.
2686 2695 * If found, then increment the ref count (exi_count).
2687 2696 */
2688 2697 struct exportinfo *
2689 2698 checkexport(fsid_t *fsid, fid_t *fid)
2690 2699 {
2691 2700 struct exportinfo *exi;
2692 2701 nfs_export_t *ne = nfs_get_export();
2693 2702
2694 2703 rw_enter(&ne->exported_lock, RW_READER);
2695 2704 for (exi = ne->exptable[exptablehash(fsid, fid)];
2696 2705 exi != NULL;
2697 2706 exi = exi->fid_hash.next) {
2698 2707 if (exportmatch(exi, fsid, fid)) {
2699 2708 /*
2700 2709 * If this is the place holder for the
2701 2710 * public file handle, then return the
2702 2711 * real export entry for the public file
2703 2712 * handle.
2704 2713 */
2705 2714 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2706 2715 exi = ne->exi_public;
2707 2716 }
2708 2717
2709 2718 exi_hold(exi);
2710 2719 rw_exit(&ne->exported_lock);
2711 2720 return (exi);
2712 2721 }
2713 2722 }
2714 2723 rw_exit(&ne->exported_lock);
2715 2724 return (NULL);
2716 2725 }
2717 2726
2718 2727
2719 2728 /*
2720 2729 * "old school" version of checkexport() for NFS4. NFS4
2721 2730 * rfs4_compound holds exported_lock for duration of compound
2722 2731 * processing. This version doesn't manipulate exi_count
2723 2732 * since NFS4 breaks fundamental assumptions in the exi_count
2724 2733 * design.
2725 2734 */
2726 2735 struct exportinfo *
2727 2736 checkexport4(fsid_t *fsid, fid_t *fid, vnode_t *vp)
2728 2737 {
2729 2738 struct exportinfo *exi;
2730 2739 nfs_export_t *ne = nfs_get_export();
2731 2740
2732 2741 ASSERT(RW_LOCK_HELD(&ne->exported_lock));
2733 2742
2734 2743 for (exi = ne->exptable[exptablehash(fsid, fid)];
2735 2744 exi != NULL;
2736 2745 exi = exi->fid_hash.next) {
2737 2746 if (exportmatch(exi, fsid, fid)) {
2738 2747 /*
2739 2748 * If this is the place holder for the
2740 2749 * public file handle, then return the
2741 2750 * real export entry for the public file
2742 2751 * handle.
2743 2752 */
2744 2753 if (exi->exi_export.ex_flags & EX_PUBLIC) {
2745 2754 exi = ne->exi_public;
2746 2755 }
2747 2756
2748 2757 /*
2749 2758 * If vp is given, check if vp is the
2750 2759 * same vnode as the exported node.
2751 2760 *
2752 2761 * Since VOP_FID of a lofs node returns the
2753 2762 * fid of its real node (ufs), the exported
2754 2763 * node for lofs and (pseudo) ufs may have
2755 2764 * the same fsid and fid.
2756 2765 */
2757 2766 if (vp == NULL || vp == exi->exi_vp)
2758 2767 return (exi);
2759 2768 }
2760 2769 }
2761 2770
2762 2771 return (NULL);
2763 2772 }
2764 2773
2765 2774 /*
2766 2775 * Free an entire export list node
2767 2776 */
2768 2777 void
2769 2778 exportfree(struct exportinfo *exi)
2770 2779 {
2771 2780 struct exportdata *ex;
2772 2781 struct charset_cache *cache;
2773 2782 int i;
2774 2783
2775 2784 ex = &exi->exi_export;
2776 2785
2777 2786 ASSERT(exi->exi_vp != NULL && !(exi->exi_export.ex_flags & EX_PUBLIC));
2778 2787 VN_RELE(exi->exi_vp);
2779 2788 if (exi->exi_dvp != NULL)
2780 2789 VN_RELE(exi->exi_dvp);
2781 2790
2782 2791 if (ex->ex_flags & EX_INDEX)
2783 2792 kmem_free(ex->ex_index, strlen(ex->ex_index) + 1);
2784 2793
2785 2794 kmem_free(ex->ex_path, ex->ex_pathlen + 1);
2786 2795 nfsauth_cache_free(exi);
2787 2796
2788 2797 /*
2789 2798 * if there is a character set mapping cached, clean it up.
2790 2799 */
2791 2800 for (cache = exi->exi_charset; cache != NULL;
2792 2801 cache = exi->exi_charset) {
2793 2802 if (cache->inbound != (kiconv_t)-1)
2794 2803 (void) kiconv_close(cache->inbound);
2795 2804 if (cache->outbound != (kiconv_t)-1)
2796 2805 (void) kiconv_close(cache->outbound);
2797 2806 exi->exi_charset = cache->next;
2798 2807 kmem_free(cache, sizeof (struct charset_cache));
2799 2808 }
2800 2809
2801 2810 if (exi->exi_logbuffer != NULL)
2802 2811 nfslog_disable(exi);
2803 2812
2804 2813 if (ex->ex_flags & EX_LOG) {
2805 2814 kmem_free(ex->ex_log_buffer, ex->ex_log_bufferlen + 1);
2806 2815 kmem_free(ex->ex_tag, ex->ex_taglen + 1);
2807 2816 }
2808 2817
2809 2818 if (exi->exi_visible)
2810 2819 free_visible(exi->exi_visible);
2811 2820
2812 2821 srv_secinfo_list_free(ex->ex_secinfo, ex->ex_seccnt);
2813 2822
2814 2823 #ifdef VOLATILE_FH_TEST
2815 2824 free_volrnm_list(exi);
2816 2825 mutex_destroy(&exi->exi_vol_rename_lock);
2817 2826 #endif /* VOLATILE_FH_TEST */
2818 2827
2819 2828 mutex_destroy(&exi->exi_lock);
2820 2829 rw_destroy(&exi->exi_cache_lock);
2821 2830 /*
2822 2831 * All nodes in the exi_cache AVL trees were removed and freed in the
2823 2832 * nfsauth_cache_free() call above. We will just destroy and free the
2824 2833 * empty AVL trees here.
2825 2834 */
2826 2835 for (i = 0; i < AUTH_TABLESIZE; i++) {
2827 2836 avl_destroy(exi->exi_cache[i]);
2828 2837 kmem_free(exi->exi_cache[i], sizeof (avl_tree_t));
2829 2838 }
2830 2839
2831 2840 kmem_free(exi, sizeof (*exi));
2832 2841 }
2833 2842
2834 2843 /*
2835 2844 * load the index file from user space into kernel space.
2836 2845 */
2837 2846 static int
2838 2847 loadindex(struct exportdata *kex)
2839 2848 {
2840 2849 int error;
2841 2850 char index[MAXNAMELEN+1];
2842 2851 size_t len;
2843 2852
2844 2853 /*
2845 2854 * copyinstr copies the complete string including the NULL and
2846 2855 * returns the len with the NULL byte included in the calculation
2847 2856 * as long as the max length is not exceeded.
2848 2857 */
2849 2858 if (error = copyinstr(kex->ex_index, index, sizeof (index), &len))
2850 2859 return (error);
2851 2860
2852 2861 kex->ex_index = kmem_alloc(len, KM_SLEEP);
2853 2862 bcopy(index, kex->ex_index, len);
2854 2863
2855 2864 return (0);
2856 2865 }
2857 2866
2858 2867 void
2859 2868 exi_hold(struct exportinfo *exi)
2860 2869 {
2861 2870 mutex_enter(&exi->exi_lock);
2862 2871 exi->exi_count++;
2863 2872 mutex_exit(&exi->exi_lock);
2864 2873 }
2865 2874
2866 2875 /*
2867 2876 * When a thread completes using exi, it should call exi_rele().
2868 2877 * exi_rele() decrements exi_count. It releases exi if exi_count == 0, i.e.
2869 2878 * if this is the last user of exi and exi is not on exportinfo list anymore
2870 2879 */
2871 2880 void
2872 2881 exi_rele(struct exportinfo *exi)
2873 2882 {
2874 2883 mutex_enter(&exi->exi_lock);
2875 2884 exi->exi_count--;
2876 2885 if (exi->exi_count == 0) {
2877 2886 mutex_exit(&exi->exi_lock);
2878 2887 exportfree(exi);
2879 2888 } else
2880 2889 mutex_exit(&exi->exi_lock);
2881 2890 }
2882 2891
2883 2892 #ifdef VOLATILE_FH_TEST
2884 2893 /*
2885 2894 * Test for volatile fh's - add file handle to list and set its volatile id
2886 2895 * to time it was renamed. If EX_VOLFH is also on and the fs is reshared,
2887 2896 * the vol_rename queue is purged.
2888 2897 *
2889 2898 * XXX This code is for unit testing purposes only... To correctly use it, it
2890 2899 * needs to tie a rename list to the export struct and (more
2891 2900 * important), protect access to the exi rename list using a write lock.
2892 2901 */
2893 2902
2894 2903 /*
2895 2904 * get the fh vol record if it's in the volatile on rename list. Don't check
2896 2905 * volatile_id in the file handle - compare only the file handles.
2897 2906 */
2898 2907 static struct ex_vol_rename *
2899 2908 find_volrnm_fh(struct exportinfo *exi, nfs_fh4 *fh4p)
2900 2909 {
2901 2910 struct ex_vol_rename *p = NULL;
2902 2911 fhandle4_t *fhp;
2903 2912
2904 2913 /* XXX shouldn't we assert &exported_lock held? */
2905 2914 ASSERT(MUTEX_HELD(&exi->exi_vol_rename_lock));
2906 2915
2907 2916 if (fh4p->nfs_fh4_len != NFS_FH4_LEN) {
2908 2917 return (NULL);
2909 2918 }
2910 2919 fhp = &((nfs_fh4_fmt_t *)fh4p->nfs_fh4_val)->fh4_i;
2911 2920 for (p = exi->exi_vol_rename; p != NULL; p = p->vrn_next) {
2912 2921 if (bcmp(fhp, &p->vrn_fh_fmt.fh4_i,
2913 2922 sizeof (fhandle4_t)) == 0)
2914 2923 break;
2915 2924 }
2916 2925 return (p);
2917 2926 }
2918 2927
2919 2928 /*
2920 2929 * get the volatile id for the fh (if there is - else return 0). Ignore the
2921 2930 * volatile_id in the file handle - compare only the file handles.
2922 2931 */
2923 2932 static uint32_t
2924 2933 find_volrnm_fh_id(struct exportinfo *exi, nfs_fh4 *fh4p)
2925 2934 {
2926 2935 struct ex_vol_rename *p;
2927 2936 uint32_t volatile_id;
2928 2937
2929 2938 mutex_enter(&exi->exi_vol_rename_lock);
2930 2939 p = find_volrnm_fh(exi, fh4p);
2931 2940 volatile_id = (p ? p->vrn_fh_fmt.fh4_volatile_id :
2932 2941 exi->exi_volatile_id);
2933 2942 mutex_exit(&exi->exi_vol_rename_lock);
2934 2943 return (volatile_id);
2935 2944 }
2936 2945
2937 2946 /*
2938 2947 * Free the volatile on rename list - will be called if a filesystem is
2939 2948 * unshared or reshared without EX_VOLRNM
2940 2949 */
2941 2950 static void
2942 2951 free_volrnm_list(struct exportinfo *exi)
2943 2952 {
2944 2953 struct ex_vol_rename *p, *pnext;
2945 2954
2946 2955 /* no need to hold mutex lock - this one is called from exportfree */
2947 2956 for (p = exi->exi_vol_rename; p != NULL; p = pnext) {
2948 2957 pnext = p->vrn_next;
2949 2958 kmem_free(p, sizeof (*p));
2950 2959 }
2951 2960 exi->exi_vol_rename = NULL;
2952 2961 }
2953 2962
2954 2963 /*
2955 2964 * Add a file handle to the volatile on rename list.
2956 2965 */
2957 2966 void
2958 2967 add_volrnm_fh(struct exportinfo *exi, vnode_t *vp)
2959 2968 {
2960 2969 struct ex_vol_rename *p;
2961 2970 char fhbuf[NFS4_FHSIZE];
2962 2971 nfs_fh4 fh4;
2963 2972 int error;
2964 2973
2965 2974 fh4.nfs_fh4_val = fhbuf;
2966 2975 error = makefh4(&fh4, vp, exi);
2967 2976 if ((error) || (fh4.nfs_fh4_len != sizeof (p->vrn_fh_fmt))) {
2968 2977 return;
2969 2978 }
2970 2979
2971 2980 mutex_enter(&exi->exi_vol_rename_lock);
2972 2981
2973 2982 p = find_volrnm_fh(exi, &fh4);
2974 2983
2975 2984 if (p == NULL) {
2976 2985 p = kmem_alloc(sizeof (*p), KM_SLEEP);
2977 2986 bcopy(fh4.nfs_fh4_val, &p->vrn_fh_fmt, sizeof (p->vrn_fh_fmt));
2978 2987 p->vrn_next = exi->exi_vol_rename;
2979 2988 exi->exi_vol_rename = p;
2980 2989 }
2981 2990
2982 2991 p->vrn_fh_fmt.fh4_volatile_id = gethrestime_sec();
2983 2992 mutex_exit(&exi->exi_vol_rename_lock);
2984 2993 }
2985 2994
2986 2995 #endif /* VOLATILE_FH_TEST */
|
↓ open down ↓ |
785 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX