xref: /netbsd-src/sys/nfs/nfs_subs.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: nfs_subs.c,v 1.138 2004/10/26 04:34:46 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_subs.c	8.8 (Berkeley) 5/22/95
35  */
36 
37 /*
38  * Copyright 2000 Wasabi Systems, Inc.
39  * All rights reserved.
40  *
41  * Written by Frank van der Linden for Wasabi Systems, Inc.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. All advertising materials mentioning features or use of this software
52  *    must display the following acknowledgement:
53  *      This product includes software developed for the NetBSD Project by
54  *      Wasabi Systems, Inc.
55  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
56  *    or promote products derived from this software without specific prior
57  *    written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
61  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
62  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
63  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
64  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
65  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
66  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
67  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
68  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
69  * POSSIBILITY OF SUCH DAMAGE.
70  */
71 
72 #include <sys/cdefs.h>
73 __KERNEL_RCSID(0, "$NetBSD: nfs_subs.c,v 1.138 2004/10/26 04:34:46 yamt Exp $");
74 
75 #include "fs_nfs.h"
76 #include "opt_nfs.h"
77 #include "opt_nfsserver.h"
78 #include "opt_iso.h"
79 #include "opt_inet.h"
80 
81 /*
82  * These functions support the macros and help fiddle mbuf chains for
83  * the nfs op functions. They do things like create the rpc header and
84  * copy data between mbuf chains and uio lists.
85  */
86 #include <sys/param.h>
87 #include <sys/proc.h>
88 #include <sys/systm.h>
89 #include <sys/kernel.h>
90 #include <sys/mount.h>
91 #include <sys/vnode.h>
92 #include <sys/namei.h>
93 #include <sys/mbuf.h>
94 #include <sys/socket.h>
95 #include <sys/stat.h>
96 #include <sys/malloc.h>
97 #include <sys/filedesc.h>
98 #include <sys/time.h>
99 #include <sys/dirent.h>
100 
101 #include <uvm/uvm_extern.h>
102 
103 #include <nfs/rpcv2.h>
104 #include <nfs/nfsproto.h>
105 #include <nfs/nfsnode.h>
106 #include <nfs/nfs.h>
107 #include <nfs/xdr_subs.h>
108 #include <nfs/nfsm_subs.h>
109 #include <nfs/nfsmount.h>
110 #include <nfs/nqnfs.h>
111 #include <nfs/nfsrtt.h>
112 #include <nfs/nfs_var.h>
113 
114 #include <miscfs/specfs/specdev.h>
115 
116 #include <netinet/in.h>
117 #ifdef ISO
118 #include <netiso/iso.h>
119 #endif
120 
121 /*
122  * Data items converted to xdr at startup, since they are constant
123  * This is kinda hokey, but may save a little time doing byte swaps
124  */
125 u_int32_t nfs_xdrneg1;
126 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr,
127 	rpc_mismatch, rpc_auth_unix, rpc_msgaccepted,
128 	rpc_auth_kerb;
129 u_int32_t nfs_prog, nqnfs_prog, nfs_true, nfs_false;
130 
131 /* And other global data */
132 const nfstype nfsv2_type[9] =
133 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON };
134 const nfstype nfsv3_type[9] =
135 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON };
136 const enum vtype nv2tov_type[8] =
137 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON };
138 const enum vtype nv3tov_type[8] =
139 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO };
140 int nfs_ticks;
141 int nfs_commitsize;
142 
143 MALLOC_DEFINE(M_NFSDIROFF, "NFS diroff", "NFS directory cookies");
144 
145 /* NFS client/server stats. */
146 struct nfsstats nfsstats;
147 
148 /*
149  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
150  */
151 const int nfsv3_procid[NFS_NPROCS] = {
152 	NFSPROC_NULL,
153 	NFSPROC_GETATTR,
154 	NFSPROC_SETATTR,
155 	NFSPROC_NOOP,
156 	NFSPROC_LOOKUP,
157 	NFSPROC_READLINK,
158 	NFSPROC_READ,
159 	NFSPROC_NOOP,
160 	NFSPROC_WRITE,
161 	NFSPROC_CREATE,
162 	NFSPROC_REMOVE,
163 	NFSPROC_RENAME,
164 	NFSPROC_LINK,
165 	NFSPROC_SYMLINK,
166 	NFSPROC_MKDIR,
167 	NFSPROC_RMDIR,
168 	NFSPROC_READDIR,
169 	NFSPROC_FSSTAT,
170 	NFSPROC_NOOP,
171 	NFSPROC_NOOP,
172 	NFSPROC_NOOP,
173 	NFSPROC_NOOP,
174 	NFSPROC_NOOP,
175 	NFSPROC_NOOP,
176 	NFSPROC_NOOP,
177 	NFSPROC_NOOP
178 };
179 
180 /*
181  * and the reverse mapping from generic to Version 2 procedure numbers
182  */
183 const int nfsv2_procid[NFS_NPROCS] = {
184 	NFSV2PROC_NULL,
185 	NFSV2PROC_GETATTR,
186 	NFSV2PROC_SETATTR,
187 	NFSV2PROC_LOOKUP,
188 	NFSV2PROC_NOOP,
189 	NFSV2PROC_READLINK,
190 	NFSV2PROC_READ,
191 	NFSV2PROC_WRITE,
192 	NFSV2PROC_CREATE,
193 	NFSV2PROC_MKDIR,
194 	NFSV2PROC_SYMLINK,
195 	NFSV2PROC_CREATE,
196 	NFSV2PROC_REMOVE,
197 	NFSV2PROC_RMDIR,
198 	NFSV2PROC_RENAME,
199 	NFSV2PROC_LINK,
200 	NFSV2PROC_READDIR,
201 	NFSV2PROC_NOOP,
202 	NFSV2PROC_STATFS,
203 	NFSV2PROC_NOOP,
204 	NFSV2PROC_NOOP,
205 	NFSV2PROC_NOOP,
206 	NFSV2PROC_NOOP,
207 	NFSV2PROC_NOOP,
208 	NFSV2PROC_NOOP,
209 	NFSV2PROC_NOOP,
210 };
211 
212 /*
213  * Maps errno values to nfs error numbers.
214  * Use NFSERR_IO as the catch all for ones not specifically defined in
215  * RFC 1094.
216  */
217 static const u_char nfsrv_v2errmap[ELAST] = {
218   NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
219   NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
220   NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
221   NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
222   NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
223   NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
224   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
225   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
226   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
227   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
228   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
229   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
230   NFSERR_IO,	NFSERR_IO,	NFSERR_NAMETOL,	NFSERR_IO,	NFSERR_IO,
231   NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
232   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
233   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
234   NFSERR_IO,	NFSERR_IO,
235 };
236 
237 /*
238  * Maps errno values to nfs error numbers.
239  * Although it is not obvious whether or not NFS clients really care if
240  * a returned error value is in the specified list for the procedure, the
241  * safest thing to do is filter them appropriately. For Version 2, the
242  * X/Open XNFS document is the only specification that defines error values
243  * for each RPC (The RFC simply lists all possible error values for all RPCs),
244  * so I have decided to not do this for Version 2.
245  * The first entry is the default error return and the rest are the valid
246  * errors for that RPC in increasing numeric order.
247  */
248 static const short nfsv3err_null[] = {
249 	0,
250 	0,
251 };
252 
253 static const short nfsv3err_getattr[] = {
254 	NFSERR_IO,
255 	NFSERR_IO,
256 	NFSERR_STALE,
257 	NFSERR_BADHANDLE,
258 	NFSERR_SERVERFAULT,
259 	0,
260 };
261 
262 static const short nfsv3err_setattr[] = {
263 	NFSERR_IO,
264 	NFSERR_PERM,
265 	NFSERR_IO,
266 	NFSERR_ACCES,
267 	NFSERR_INVAL,
268 	NFSERR_NOSPC,
269 	NFSERR_ROFS,
270 	NFSERR_DQUOT,
271 	NFSERR_STALE,
272 	NFSERR_BADHANDLE,
273 	NFSERR_NOT_SYNC,
274 	NFSERR_SERVERFAULT,
275 	0,
276 };
277 
278 static const short nfsv3err_lookup[] = {
279 	NFSERR_IO,
280 	NFSERR_NOENT,
281 	NFSERR_IO,
282 	NFSERR_ACCES,
283 	NFSERR_NOTDIR,
284 	NFSERR_NAMETOL,
285 	NFSERR_STALE,
286 	NFSERR_BADHANDLE,
287 	NFSERR_SERVERFAULT,
288 	0,
289 };
290 
291 static const short nfsv3err_access[] = {
292 	NFSERR_IO,
293 	NFSERR_IO,
294 	NFSERR_STALE,
295 	NFSERR_BADHANDLE,
296 	NFSERR_SERVERFAULT,
297 	0,
298 };
299 
300 static const short nfsv3err_readlink[] = {
301 	NFSERR_IO,
302 	NFSERR_IO,
303 	NFSERR_ACCES,
304 	NFSERR_INVAL,
305 	NFSERR_STALE,
306 	NFSERR_BADHANDLE,
307 	NFSERR_NOTSUPP,
308 	NFSERR_SERVERFAULT,
309 	0,
310 };
311 
312 static const short nfsv3err_read[] = {
313 	NFSERR_IO,
314 	NFSERR_IO,
315 	NFSERR_NXIO,
316 	NFSERR_ACCES,
317 	NFSERR_INVAL,
318 	NFSERR_STALE,
319 	NFSERR_BADHANDLE,
320 	NFSERR_SERVERFAULT,
321 	NFSERR_JUKEBOX,
322 	0,
323 };
324 
325 static const short nfsv3err_write[] = {
326 	NFSERR_IO,
327 	NFSERR_IO,
328 	NFSERR_ACCES,
329 	NFSERR_INVAL,
330 	NFSERR_FBIG,
331 	NFSERR_NOSPC,
332 	NFSERR_ROFS,
333 	NFSERR_DQUOT,
334 	NFSERR_STALE,
335 	NFSERR_BADHANDLE,
336 	NFSERR_SERVERFAULT,
337 	NFSERR_JUKEBOX,
338 	0,
339 };
340 
341 static const short nfsv3err_create[] = {
342 	NFSERR_IO,
343 	NFSERR_IO,
344 	NFSERR_ACCES,
345 	NFSERR_EXIST,
346 	NFSERR_NOTDIR,
347 	NFSERR_NOSPC,
348 	NFSERR_ROFS,
349 	NFSERR_NAMETOL,
350 	NFSERR_DQUOT,
351 	NFSERR_STALE,
352 	NFSERR_BADHANDLE,
353 	NFSERR_NOTSUPP,
354 	NFSERR_SERVERFAULT,
355 	0,
356 };
357 
358 static const short nfsv3err_mkdir[] = {
359 	NFSERR_IO,
360 	NFSERR_IO,
361 	NFSERR_ACCES,
362 	NFSERR_EXIST,
363 	NFSERR_NOTDIR,
364 	NFSERR_NOSPC,
365 	NFSERR_ROFS,
366 	NFSERR_NAMETOL,
367 	NFSERR_DQUOT,
368 	NFSERR_STALE,
369 	NFSERR_BADHANDLE,
370 	NFSERR_NOTSUPP,
371 	NFSERR_SERVERFAULT,
372 	0,
373 };
374 
375 static const short nfsv3err_symlink[] = {
376 	NFSERR_IO,
377 	NFSERR_IO,
378 	NFSERR_ACCES,
379 	NFSERR_EXIST,
380 	NFSERR_NOTDIR,
381 	NFSERR_NOSPC,
382 	NFSERR_ROFS,
383 	NFSERR_NAMETOL,
384 	NFSERR_DQUOT,
385 	NFSERR_STALE,
386 	NFSERR_BADHANDLE,
387 	NFSERR_NOTSUPP,
388 	NFSERR_SERVERFAULT,
389 	0,
390 };
391 
392 static const short nfsv3err_mknod[] = {
393 	NFSERR_IO,
394 	NFSERR_IO,
395 	NFSERR_ACCES,
396 	NFSERR_EXIST,
397 	NFSERR_NOTDIR,
398 	NFSERR_NOSPC,
399 	NFSERR_ROFS,
400 	NFSERR_NAMETOL,
401 	NFSERR_DQUOT,
402 	NFSERR_STALE,
403 	NFSERR_BADHANDLE,
404 	NFSERR_NOTSUPP,
405 	NFSERR_SERVERFAULT,
406 	NFSERR_BADTYPE,
407 	0,
408 };
409 
410 static const short nfsv3err_remove[] = {
411 	NFSERR_IO,
412 	NFSERR_NOENT,
413 	NFSERR_IO,
414 	NFSERR_ACCES,
415 	NFSERR_NOTDIR,
416 	NFSERR_ROFS,
417 	NFSERR_NAMETOL,
418 	NFSERR_STALE,
419 	NFSERR_BADHANDLE,
420 	NFSERR_SERVERFAULT,
421 	0,
422 };
423 
424 static const short nfsv3err_rmdir[] = {
425 	NFSERR_IO,
426 	NFSERR_NOENT,
427 	NFSERR_IO,
428 	NFSERR_ACCES,
429 	NFSERR_EXIST,
430 	NFSERR_NOTDIR,
431 	NFSERR_INVAL,
432 	NFSERR_ROFS,
433 	NFSERR_NAMETOL,
434 	NFSERR_NOTEMPTY,
435 	NFSERR_STALE,
436 	NFSERR_BADHANDLE,
437 	NFSERR_NOTSUPP,
438 	NFSERR_SERVERFAULT,
439 	0,
440 };
441 
442 static const short nfsv3err_rename[] = {
443 	NFSERR_IO,
444 	NFSERR_NOENT,
445 	NFSERR_IO,
446 	NFSERR_ACCES,
447 	NFSERR_EXIST,
448 	NFSERR_XDEV,
449 	NFSERR_NOTDIR,
450 	NFSERR_ISDIR,
451 	NFSERR_INVAL,
452 	NFSERR_NOSPC,
453 	NFSERR_ROFS,
454 	NFSERR_MLINK,
455 	NFSERR_NAMETOL,
456 	NFSERR_NOTEMPTY,
457 	NFSERR_DQUOT,
458 	NFSERR_STALE,
459 	NFSERR_BADHANDLE,
460 	NFSERR_NOTSUPP,
461 	NFSERR_SERVERFAULT,
462 	0,
463 };
464 
465 static const short nfsv3err_link[] = {
466 	NFSERR_IO,
467 	NFSERR_IO,
468 	NFSERR_ACCES,
469 	NFSERR_EXIST,
470 	NFSERR_XDEV,
471 	NFSERR_NOTDIR,
472 	NFSERR_INVAL,
473 	NFSERR_NOSPC,
474 	NFSERR_ROFS,
475 	NFSERR_MLINK,
476 	NFSERR_NAMETOL,
477 	NFSERR_DQUOT,
478 	NFSERR_STALE,
479 	NFSERR_BADHANDLE,
480 	NFSERR_NOTSUPP,
481 	NFSERR_SERVERFAULT,
482 	0,
483 };
484 
485 static const short nfsv3err_readdir[] = {
486 	NFSERR_IO,
487 	NFSERR_IO,
488 	NFSERR_ACCES,
489 	NFSERR_NOTDIR,
490 	NFSERR_STALE,
491 	NFSERR_BADHANDLE,
492 	NFSERR_BAD_COOKIE,
493 	NFSERR_TOOSMALL,
494 	NFSERR_SERVERFAULT,
495 	0,
496 };
497 
498 static const short nfsv3err_readdirplus[] = {
499 	NFSERR_IO,
500 	NFSERR_IO,
501 	NFSERR_ACCES,
502 	NFSERR_NOTDIR,
503 	NFSERR_STALE,
504 	NFSERR_BADHANDLE,
505 	NFSERR_BAD_COOKIE,
506 	NFSERR_NOTSUPP,
507 	NFSERR_TOOSMALL,
508 	NFSERR_SERVERFAULT,
509 	0,
510 };
511 
512 static const short nfsv3err_fsstat[] = {
513 	NFSERR_IO,
514 	NFSERR_IO,
515 	NFSERR_STALE,
516 	NFSERR_BADHANDLE,
517 	NFSERR_SERVERFAULT,
518 	0,
519 };
520 
521 static const short nfsv3err_fsinfo[] = {
522 	NFSERR_STALE,
523 	NFSERR_STALE,
524 	NFSERR_BADHANDLE,
525 	NFSERR_SERVERFAULT,
526 	0,
527 };
528 
529 static const short nfsv3err_pathconf[] = {
530 	NFSERR_STALE,
531 	NFSERR_STALE,
532 	NFSERR_BADHANDLE,
533 	NFSERR_SERVERFAULT,
534 	0,
535 };
536 
537 static const short nfsv3err_commit[] = {
538 	NFSERR_IO,
539 	NFSERR_IO,
540 	NFSERR_STALE,
541 	NFSERR_BADHANDLE,
542 	NFSERR_SERVERFAULT,
543 	0,
544 };
545 
546 static const short * const nfsrv_v3errmap[] = {
547 	nfsv3err_null,
548 	nfsv3err_getattr,
549 	nfsv3err_setattr,
550 	nfsv3err_lookup,
551 	nfsv3err_access,
552 	nfsv3err_readlink,
553 	nfsv3err_read,
554 	nfsv3err_write,
555 	nfsv3err_create,
556 	nfsv3err_mkdir,
557 	nfsv3err_symlink,
558 	nfsv3err_mknod,
559 	nfsv3err_remove,
560 	nfsv3err_rmdir,
561 	nfsv3err_rename,
562 	nfsv3err_link,
563 	nfsv3err_readdir,
564 	nfsv3err_readdirplus,
565 	nfsv3err_fsstat,
566 	nfsv3err_fsinfo,
567 	nfsv3err_pathconf,
568 	nfsv3err_commit,
569 };
570 
571 extern struct nfsrtt nfsrtt;
572 extern time_t nqnfsstarttime;
573 extern int nqsrv_clockskew;
574 extern int nqsrv_writeslack;
575 extern int nqsrv_maxlease;
576 extern const int nqnfs_piggy[NFS_NPROCS];
577 extern struct nfsnodehashhead *nfsnodehashtbl;
578 extern u_long nfsnodehash;
579 
580 u_long nfsdirhashmask;
581 
582 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *));
583 
584 /*
585  * Create the header for an rpc request packet
586  * The hsiz is the size of the rest of the nfs request header.
587  * (just used to decide if a cluster is a good idea)
588  */
589 struct mbuf *
590 nfsm_reqh(np, procid, hsiz, bposp)
591 	struct nfsnode *np;
592 	u_long procid;
593 	int hsiz;
594 	caddr_t *bposp;
595 {
596 	struct mbuf *mb;
597 	caddr_t bpos;
598 #ifndef NFS_V2_ONLY
599 	struct nfsmount *nmp;
600 	u_int32_t *tl;
601 	int nqflag;
602 #endif
603 
604 	mb = m_get(M_WAIT, MT_DATA);
605 	MCLAIM(mb, &nfs_mowner);
606 	if (hsiz >= MINCLSIZE)
607 		m_clget(mb, M_WAIT);
608 	mb->m_len = 0;
609 	bpos = mtod(mb, caddr_t);
610 
611 #ifndef NFS_V2_ONLY
612 	/*
613 	 * For NQNFS, add lease request.
614 	 */
615 	if (np) {
616 		nmp = VFSTONFS(np->n_vnode->v_mount);
617 		if (nmp->nm_flag & NFSMNT_NQNFS) {
618 			nqflag = NQNFS_NEEDLEASE(np, procid);
619 			if (nqflag) {
620 				nfsm_build(tl, u_int32_t *, 2*NFSX_UNSIGNED);
621 				*tl++ = txdr_unsigned(nqflag);
622 				*tl = txdr_unsigned(nmp->nm_leaseterm);
623 			} else {
624 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
625 				*tl = 0;
626 			}
627 		}
628 	}
629 #endif
630 	/* Finally, return values */
631 	*bposp = bpos;
632 	return (mb);
633 }
634 
635 /*
636  * Build the RPC header and fill in the authorization info.
637  * The authorization string argument is only used when the credentials
638  * come from outside of the kernel.
639  * Returns the head of the mbuf list.
640  */
641 struct mbuf *
642 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len,
643 	verf_str, mrest, mrest_len, mbp, xidp)
644 	struct ucred *cr;
645 	int nmflag;
646 	int procid;
647 	int auth_type;
648 	int auth_len;
649 	char *auth_str;
650 	int verf_len;
651 	char *verf_str;
652 	struct mbuf *mrest;
653 	int mrest_len;
654 	struct mbuf **mbp;
655 	u_int32_t *xidp;
656 {
657 	struct mbuf *mb;
658 	u_int32_t *tl;
659 	caddr_t bpos;
660 	int i;
661 	struct mbuf *mreq;
662 	int siz, grpsiz, authsiz;
663 
664 	authsiz = nfsm_rndup(auth_len);
665 	mb = m_gethdr(M_WAIT, MT_DATA);
666 	MCLAIM(mb, &nfs_mowner);
667 	if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
668 		m_clget(mb, M_WAIT);
669 	} else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
670 		MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED);
671 	} else {
672 		MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
673 	}
674 	mb->m_len = 0;
675 	mreq = mb;
676 	bpos = mtod(mb, caddr_t);
677 
678 	/*
679 	 * First the RPC header.
680 	 */
681 	nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
682 
683 	*tl++ = *xidp = nfs_getxid();
684 	*tl++ = rpc_call;
685 	*tl++ = rpc_vers;
686 	if (nmflag & NFSMNT_NQNFS) {
687 		*tl++ = txdr_unsigned(NQNFS_PROG);
688 		*tl++ = txdr_unsigned(NQNFS_VER3);
689 	} else {
690 		*tl++ = txdr_unsigned(NFS_PROG);
691 		if (nmflag & NFSMNT_NFSV3)
692 			*tl++ = txdr_unsigned(NFS_VER3);
693 		else
694 			*tl++ = txdr_unsigned(NFS_VER2);
695 	}
696 	if (nmflag & NFSMNT_NFSV3)
697 		*tl++ = txdr_unsigned(procid);
698 	else
699 		*tl++ = txdr_unsigned(nfsv2_procid[procid]);
700 
701 	/*
702 	 * And then the authorization cred.
703 	 */
704 	*tl++ = txdr_unsigned(auth_type);
705 	*tl = txdr_unsigned(authsiz);
706 	switch (auth_type) {
707 	case RPCAUTH_UNIX:
708 		nfsm_build(tl, u_int32_t *, auth_len);
709 		*tl++ = 0;		/* stamp ?? */
710 		*tl++ = 0;		/* NULL hostname */
711 		*tl++ = txdr_unsigned(cr->cr_uid);
712 		*tl++ = txdr_unsigned(cr->cr_gid);
713 		grpsiz = (auth_len >> 2) - 5;
714 		*tl++ = txdr_unsigned(grpsiz);
715 		for (i = 0; i < grpsiz; i++)
716 			*tl++ = txdr_unsigned(cr->cr_groups[i]);
717 		break;
718 	case RPCAUTH_KERB4:
719 		siz = auth_len;
720 		while (siz > 0) {
721 			if (M_TRAILINGSPACE(mb) == 0) {
722 				struct mbuf *mb2;
723 				mb2 = m_get(M_WAIT, MT_DATA);
724 				MCLAIM(mb2, &nfs_mowner);
725 				if (siz >= MINCLSIZE)
726 					m_clget(mb2, M_WAIT);
727 				mb->m_next = mb2;
728 				mb = mb2;
729 				mb->m_len = 0;
730 				bpos = mtod(mb, caddr_t);
731 			}
732 			i = min(siz, M_TRAILINGSPACE(mb));
733 			memcpy(bpos, auth_str, i);
734 			mb->m_len += i;
735 			auth_str += i;
736 			bpos += i;
737 			siz -= i;
738 		}
739 		if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
740 			for (i = 0; i < siz; i++)
741 				*bpos++ = '\0';
742 			mb->m_len += siz;
743 		}
744 		break;
745 	};
746 
747 	/*
748 	 * And the verifier...
749 	 */
750 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
751 	if (verf_str) {
752 		*tl++ = txdr_unsigned(RPCAUTH_KERB4);
753 		*tl = txdr_unsigned(verf_len);
754 		siz = verf_len;
755 		while (siz > 0) {
756 			if (M_TRAILINGSPACE(mb) == 0) {
757 				struct mbuf *mb2;
758 				mb2 = m_get(M_WAIT, MT_DATA);
759 				MCLAIM(mb2, &nfs_mowner);
760 				if (siz >= MINCLSIZE)
761 					m_clget(mb2, M_WAIT);
762 				mb->m_next = mb2;
763 				mb = mb2;
764 				mb->m_len = 0;
765 				bpos = mtod(mb, caddr_t);
766 			}
767 			i = min(siz, M_TRAILINGSPACE(mb));
768 			memcpy(bpos, verf_str, i);
769 			mb->m_len += i;
770 			verf_str += i;
771 			bpos += i;
772 			siz -= i;
773 		}
774 		if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
775 			for (i = 0; i < siz; i++)
776 				*bpos++ = '\0';
777 			mb->m_len += siz;
778 		}
779 	} else {
780 		*tl++ = txdr_unsigned(RPCAUTH_NULL);
781 		*tl = 0;
782 	}
783 	mb->m_next = mrest;
784 	mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
785 	mreq->m_pkthdr.rcvif = (struct ifnet *)0;
786 	*mbp = mb;
787 	return (mreq);
788 }
789 
790 /*
791  * copies mbuf chain to the uio scatter/gather list
792  */
793 int
794 nfsm_mbuftouio(mrep, uiop, siz, dpos)
795 	struct mbuf **mrep;
796 	struct uio *uiop;
797 	int siz;
798 	caddr_t *dpos;
799 {
800 	char *mbufcp, *uiocp;
801 	int xfer, left, len;
802 	struct mbuf *mp;
803 	long uiosiz, rem;
804 	int error = 0;
805 
806 	mp = *mrep;
807 	mbufcp = *dpos;
808 	len = mtod(mp, caddr_t)+mp->m_len-mbufcp;
809 	rem = nfsm_rndup(siz)-siz;
810 	while (siz > 0) {
811 		if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
812 			return (EFBIG);
813 		left = uiop->uio_iov->iov_len;
814 		uiocp = uiop->uio_iov->iov_base;
815 		if (left > siz)
816 			left = siz;
817 		uiosiz = left;
818 		while (left > 0) {
819 			while (len == 0) {
820 				mp = mp->m_next;
821 				if (mp == NULL)
822 					return (EBADRPC);
823 				mbufcp = mtod(mp, caddr_t);
824 				len = mp->m_len;
825 			}
826 			xfer = (left > len) ? len : left;
827 #ifdef notdef
828 			/* Not Yet.. */
829 			if (uiop->uio_iov->iov_op != NULL)
830 				(*(uiop->uio_iov->iov_op))
831 				(mbufcp, uiocp, xfer);
832 			else
833 #endif
834 			if (uiop->uio_segflg == UIO_SYSSPACE)
835 				memcpy(uiocp, mbufcp, xfer);
836 			else
837 				copyout(mbufcp, uiocp, xfer);
838 			left -= xfer;
839 			len -= xfer;
840 			mbufcp += xfer;
841 			uiocp += xfer;
842 			uiop->uio_offset += xfer;
843 			uiop->uio_resid -= xfer;
844 		}
845 		if (uiop->uio_iov->iov_len <= siz) {
846 			uiop->uio_iovcnt--;
847 			uiop->uio_iov++;
848 		} else {
849 			uiop->uio_iov->iov_base =
850 			    (caddr_t)uiop->uio_iov->iov_base + uiosiz;
851 			uiop->uio_iov->iov_len -= uiosiz;
852 		}
853 		siz -= uiosiz;
854 	}
855 	*dpos = mbufcp;
856 	*mrep = mp;
857 	if (rem > 0) {
858 		if (len < rem)
859 			error = nfs_adv(mrep, dpos, rem, len);
860 		else
861 			*dpos += rem;
862 	}
863 	return (error);
864 }
865 
866 /*
867  * copies a uio scatter/gather list to an mbuf chain.
868  * NOTE: can ony handle iovcnt == 1
869  */
870 int
871 nfsm_uiotombuf(uiop, mq, siz, bpos)
872 	struct uio *uiop;
873 	struct mbuf **mq;
874 	int siz;
875 	caddr_t *bpos;
876 {
877 	char *uiocp;
878 	struct mbuf *mp, *mp2;
879 	int xfer, left, mlen;
880 	int uiosiz, clflg, rem;
881 	char *cp;
882 
883 #ifdef DIAGNOSTIC
884 	if (uiop->uio_iovcnt != 1)
885 		panic("nfsm_uiotombuf: iovcnt != 1");
886 #endif
887 
888 	if (siz > MLEN)		/* or should it >= MCLBYTES ?? */
889 		clflg = 1;
890 	else
891 		clflg = 0;
892 	rem = nfsm_rndup(siz)-siz;
893 	mp = mp2 = *mq;
894 	while (siz > 0) {
895 		left = uiop->uio_iov->iov_len;
896 		uiocp = uiop->uio_iov->iov_base;
897 		if (left > siz)
898 			left = siz;
899 		uiosiz = left;
900 		while (left > 0) {
901 			mlen = M_TRAILINGSPACE(mp);
902 			if (mlen == 0) {
903 				mp = m_get(M_WAIT, MT_DATA);
904 				MCLAIM(mp, &nfs_mowner);
905 				if (clflg)
906 					m_clget(mp, M_WAIT);
907 				mp->m_len = 0;
908 				mp2->m_next = mp;
909 				mp2 = mp;
910 				mlen = M_TRAILINGSPACE(mp);
911 			}
912 			xfer = (left > mlen) ? mlen : left;
913 #ifdef notdef
914 			/* Not Yet.. */
915 			if (uiop->uio_iov->iov_op != NULL)
916 				(*(uiop->uio_iov->iov_op))
917 				(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
918 			else
919 #endif
920 			if (uiop->uio_segflg == UIO_SYSSPACE)
921 				memcpy(mtod(mp, caddr_t)+mp->m_len, uiocp, xfer);
922 			else
923 				copyin(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
924 			mp->m_len += xfer;
925 			left -= xfer;
926 			uiocp += xfer;
927 			uiop->uio_offset += xfer;
928 			uiop->uio_resid -= xfer;
929 		}
930 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
931 		    uiosiz;
932 		uiop->uio_iov->iov_len -= uiosiz;
933 		siz -= uiosiz;
934 	}
935 	if (rem > 0) {
936 		if (rem > M_TRAILINGSPACE(mp)) {
937 			mp = m_get(M_WAIT, MT_DATA);
938 			MCLAIM(mp, &nfs_mowner);
939 			mp->m_len = 0;
940 			mp2->m_next = mp;
941 		}
942 		cp = mtod(mp, caddr_t)+mp->m_len;
943 		for (left = 0; left < rem; left++)
944 			*cp++ = '\0';
945 		mp->m_len += rem;
946 		*bpos = cp;
947 	} else
948 		*bpos = mtod(mp, caddr_t)+mp->m_len;
949 	*mq = mp;
950 	return (0);
951 }
952 
953 /*
954  * Get at least "siz" bytes of correctly aligned data.
955  * When called the mbuf pointers are not necessarily correct,
956  * dsosp points to what ought to be in m_data and left contains
957  * what ought to be in m_len.
958  * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
959  * cases. (The macros use the vars. dpos and dpos2)
960  */
961 int
962 nfsm_disct(mdp, dposp, siz, left, cp2)
963 	struct mbuf **mdp;
964 	caddr_t *dposp;
965 	int siz;
966 	int left;
967 	caddr_t *cp2;
968 {
969 	struct mbuf *m1, *m2;
970 	struct mbuf *havebuf = NULL;
971 	caddr_t src = *dposp;
972 	caddr_t dst;
973 	int len;
974 
975 #ifdef DEBUG
976 	if (left < 0)
977 		panic("nfsm_disct: left < 0");
978 #endif
979 	m1 = *mdp;
980 	/*
981 	 * Skip through the mbuf chain looking for an mbuf with
982 	 * some data. If the first mbuf found has enough data
983 	 * and it is correctly aligned return it.
984 	 */
985 	while (left == 0) {
986 		havebuf = m1;
987 		*mdp = m1 = m1->m_next;
988 		if (m1 == NULL)
989 			return (EBADRPC);
990 		src = mtod(m1, caddr_t);
991 		left = m1->m_len;
992 		/*
993 		 * If we start a new mbuf and it is big enough
994 		 * and correctly aligned just return it, don't
995 		 * do any pull up.
996 		 */
997 		if (left >= siz && nfsm_aligned(src)) {
998 			*cp2 = src;
999 			*dposp = src + siz;
1000 			return (0);
1001 		}
1002 	}
1003 	if (m1->m_flags & M_EXT) {
1004 		if (havebuf) {
1005 			/* If the first mbuf with data has external data
1006 			 * and there is a previous empty mbuf use it
1007 			 * to move the data into.
1008 			 */
1009 			m2 = m1;
1010 			*mdp = m1 = havebuf;
1011 			if (m1->m_flags & M_EXT) {
1012 				MEXTREMOVE(m1);
1013 			}
1014 		} else {
1015 			/*
1016 			 * If the first mbuf has a external data
1017 			 * and there is no previous empty mbuf
1018 			 * allocate a new mbuf and move the external
1019 			 * data to the new mbuf. Also make the first
1020 			 * mbuf look empty.
1021 			 */
1022 			m2 = m_get(M_WAIT, MT_DATA);
1023 			m2->m_ext = m1->m_ext;
1024 			m2->m_data = src;
1025 			m2->m_len = left;
1026 			MCLADDREFERENCE(m1, m2);
1027 			MEXTREMOVE(m1);
1028 			m2->m_next = m1->m_next;
1029 			m1->m_next = m2;
1030 		}
1031 		m1->m_len = 0;
1032 		if (m1->m_flags & M_PKTHDR)
1033 			dst = m1->m_pktdat;
1034 		else
1035 			dst = m1->m_dat;
1036 		m1->m_data = dst;
1037 	} else {
1038 		/*
1039 		 * If the first mbuf has no external data
1040 		 * move the data to the front of the mbuf.
1041 		 */
1042 		if (m1->m_flags & M_PKTHDR)
1043 			dst = m1->m_pktdat;
1044 		else
1045 			dst = m1->m_dat;
1046 		m1->m_data = dst;
1047 		if (dst != src)
1048 			memmove(dst, src, left);
1049 		dst += left;
1050 		m1->m_len = left;
1051 		m2 = m1->m_next;
1052 	}
1053 	*cp2 = m1->m_data;
1054 	*dposp = mtod(m1, caddr_t) + siz;
1055 	/*
1056 	 * Loop through mbufs pulling data up into first mbuf until
1057 	 * the first mbuf is full or there is no more data to
1058 	 * pullup.
1059 	 */
1060 	while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) {
1061 		if ((len = min(len, m2->m_len)) != 0)
1062 			memcpy(dst, m2->m_data, len);
1063 		m1->m_len += len;
1064 		dst += len;
1065 		m2->m_data += len;
1066 		m2->m_len -= len;
1067 		m2 = m2->m_next;
1068 	}
1069 	if (m1->m_len < siz)
1070 		return (EBADRPC);
1071 	return (0);
1072 }
1073 
1074 /*
1075  * Advance the position in the mbuf chain.
1076  */
1077 int
1078 nfs_adv(mdp, dposp, offs, left)
1079 	struct mbuf **mdp;
1080 	caddr_t *dposp;
1081 	int offs;
1082 	int left;
1083 {
1084 	struct mbuf *m;
1085 	int s;
1086 
1087 	m = *mdp;
1088 	s = left;
1089 	while (s < offs) {
1090 		offs -= s;
1091 		m = m->m_next;
1092 		if (m == NULL)
1093 			return (EBADRPC);
1094 		s = m->m_len;
1095 	}
1096 	*mdp = m;
1097 	*dposp = mtod(m, caddr_t)+offs;
1098 	return (0);
1099 }
1100 
1101 /*
1102  * Copy a string into mbufs for the hard cases...
1103  */
1104 int
1105 nfsm_strtmbuf(mb, bpos, cp, siz)
1106 	struct mbuf **mb;
1107 	char **bpos;
1108 	const char *cp;
1109 	long siz;
1110 {
1111 	struct mbuf *m1 = NULL, *m2;
1112 	long left, xfer, len, tlen;
1113 	u_int32_t *tl;
1114 	int putsize;
1115 
1116 	putsize = 1;
1117 	m2 = *mb;
1118 	left = M_TRAILINGSPACE(m2);
1119 	if (left > 0) {
1120 		tl = ((u_int32_t *)(*bpos));
1121 		*tl++ = txdr_unsigned(siz);
1122 		putsize = 0;
1123 		left -= NFSX_UNSIGNED;
1124 		m2->m_len += NFSX_UNSIGNED;
1125 		if (left > 0) {
1126 			memcpy((caddr_t) tl, cp, left);
1127 			siz -= left;
1128 			cp += left;
1129 			m2->m_len += left;
1130 			left = 0;
1131 		}
1132 	}
1133 	/* Loop around adding mbufs */
1134 	while (siz > 0) {
1135 		m1 = m_get(M_WAIT, MT_DATA);
1136 		MCLAIM(m1, &nfs_mowner);
1137 		if (siz > MLEN)
1138 			m_clget(m1, M_WAIT);
1139 		m1->m_len = NFSMSIZ(m1);
1140 		m2->m_next = m1;
1141 		m2 = m1;
1142 		tl = mtod(m1, u_int32_t *);
1143 		tlen = 0;
1144 		if (putsize) {
1145 			*tl++ = txdr_unsigned(siz);
1146 			m1->m_len -= NFSX_UNSIGNED;
1147 			tlen = NFSX_UNSIGNED;
1148 			putsize = 0;
1149 		}
1150 		if (siz < m1->m_len) {
1151 			len = nfsm_rndup(siz);
1152 			xfer = siz;
1153 			if (xfer < len)
1154 				*(tl+(xfer>>2)) = 0;
1155 		} else {
1156 			xfer = len = m1->m_len;
1157 		}
1158 		memcpy((caddr_t) tl, cp, xfer);
1159 		m1->m_len = len+tlen;
1160 		siz -= xfer;
1161 		cp += xfer;
1162 	}
1163 	*mb = m1;
1164 	*bpos = mtod(m1, caddr_t)+m1->m_len;
1165 	return (0);
1166 }
1167 
1168 /*
1169  * Directory caching routines. They work as follows:
1170  * - a cache is maintained per VDIR nfsnode.
1171  * - for each offset cookie that is exported to userspace, and can
1172  *   thus be thrown back at us as an offset to VOP_READDIR, store
1173  *   information in the cache.
1174  * - cached are:
1175  *   - cookie itself
1176  *   - blocknumber (essentially just a search key in the buffer cache)
1177  *   - entry number in block.
1178  *   - offset cookie of block in which this entry is stored
1179  *   - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
1180  * - entries are looked up in a hash table
1181  * - also maintained is an LRU list of entries, used to determine
1182  *   which ones to delete if the cache grows too large.
1183  * - if 32 <-> 64 translation mode is requested for a filesystem,
1184  *   the cache also functions as a translation table
1185  * - in the translation case, invalidating the cache does not mean
1186  *   flushing it, but just marking entries as invalid, except for
1187  *   the <64bit cookie, 32bitcookie> pair which is still valid, to
1188  *   still be able to use the cache as a translation table.
1189  * - 32 bit cookies are uniquely created by combining the hash table
1190  *   entry value, and one generation count per hash table entry,
1191  *   incremented each time an entry is appended to the chain.
1192  * - the cache is invalidated each time a direcory is modified
1193  * - sanity checks are also done; if an entry in a block turns
1194  *   out not to have a matching cookie, the cache is invalidated
1195  *   and a new block starting from the wanted offset is fetched from
1196  *   the server.
1197  * - directory entries as read from the server are extended to contain
1198  *   the 64bit and, optionally, the 32bit cookies, for sanity checking
1199  *   the cache and exporting them to userspace through the cookie
1200  *   argument to VOP_READDIR.
1201  */
1202 
1203 u_long
1204 nfs_dirhash(off)
1205 	off_t off;
1206 {
1207 	int i;
1208 	char *cp = (char *)&off;
1209 	u_long sum = 0L;
1210 
1211 	for (i = 0 ; i < sizeof (off); i++)
1212 		sum += *cp++;
1213 
1214 	return sum;
1215 }
1216 
1217 #define	_NFSDC_MTX(np)		(&NFSTOV(np)->v_interlock)
1218 #define	NFSDC_LOCK(np)		simple_lock(_NFSDC_MTX(np))
1219 #define	NFSDC_UNLOCK(np)	simple_unlock(_NFSDC_MTX(np))
1220 #define	NFSDC_ASSERT_LOCKED(np) LOCK_ASSERT(simple_lock_held(_NFSDC_MTX(np)))
1221 
1222 void
1223 nfs_initdircache(vp)
1224 	struct vnode *vp;
1225 {
1226 	struct nfsnode *np = VTONFS(vp);
1227 	struct nfsdirhashhead *dircache;
1228 
1229 	dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF,
1230 	    M_WAITOK, &nfsdirhashmask);
1231 
1232 	NFSDC_LOCK(np);
1233 	if (np->n_dircache == NULL) {
1234 		np->n_dircachesize = 0;
1235 		np->n_dircache = dircache;
1236 		dircache = NULL;
1237 		TAILQ_INIT(&np->n_dirchain);
1238 	}
1239 	NFSDC_UNLOCK(np);
1240 	if (dircache)
1241 		hashdone(dircache, M_NFSDIROFF);
1242 }
1243 
1244 void
1245 nfs_initdirxlatecookie(vp)
1246 	struct vnode *vp;
1247 {
1248 	struct nfsnode *np = VTONFS(vp);
1249 	unsigned *dirgens;
1250 
1251 	KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE);
1252 
1253 	dirgens = malloc(NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF,
1254 	    M_WAITOK|M_ZERO);
1255 	NFSDC_LOCK(np);
1256 	if (np->n_dirgens == NULL) {
1257 		np->n_dirgens = dirgens;
1258 		dirgens = NULL;
1259 	}
1260 	NFSDC_UNLOCK(np);
1261 	if (dirgens)
1262 		free(dirgens, M_NFSDIROFF);
1263 }
1264 
1265 static const struct nfsdircache dzero;
1266 
1267 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *));
1268 static void nfs_putdircache_unlocked __P((struct nfsnode *,
1269     struct nfsdircache *));
1270 
1271 static void
1272 nfs_unlinkdircache(np, ndp)
1273 	struct nfsnode *np;
1274 	struct nfsdircache *ndp;
1275 {
1276 
1277 	NFSDC_ASSERT_LOCKED(np);
1278 	KASSERT(ndp != &dzero);
1279 
1280 	if (LIST_NEXT(ndp, dc_hash) == (void *)-1)
1281 		return;
1282 
1283 	TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1284 	LIST_REMOVE(ndp, dc_hash);
1285 	LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */
1286 
1287 	nfs_putdircache_unlocked(np, ndp);
1288 }
1289 
1290 void
1291 nfs_putdircache(np, ndp)
1292 	struct nfsnode *np;
1293 	struct nfsdircache *ndp;
1294 {
1295 	int ref;
1296 
1297 	if (ndp == &dzero)
1298 		return;
1299 
1300 	KASSERT(ndp->dc_refcnt > 0);
1301 	NFSDC_LOCK(np);
1302 	ref = --ndp->dc_refcnt;
1303 	NFSDC_UNLOCK(np);
1304 
1305 	if (ref == 0)
1306 		free(ndp, M_NFSDIROFF);
1307 }
1308 
1309 static void
1310 nfs_putdircache_unlocked(np, ndp)
1311 	struct nfsnode *np;
1312 	struct nfsdircache *ndp;
1313 {
1314 	int ref;
1315 
1316 	NFSDC_ASSERT_LOCKED(np);
1317 
1318 	if (ndp == &dzero)
1319 		return;
1320 
1321 	KASSERT(ndp->dc_refcnt > 0);
1322 	ref = --ndp->dc_refcnt;
1323 	if (ref == 0)
1324 		free(ndp, M_NFSDIROFF);
1325 }
1326 
1327 struct nfsdircache *
1328 nfs_searchdircache(vp, off, do32, hashent)
1329 	struct vnode *vp;
1330 	off_t off;
1331 	int do32;
1332 	int *hashent;
1333 {
1334 	struct nfsdirhashhead *ndhp;
1335 	struct nfsdircache *ndp = NULL;
1336 	struct nfsnode *np = VTONFS(vp);
1337 	unsigned ent;
1338 
1339 	/*
1340 	 * Zero is always a valid cookie.
1341 	 */
1342 	if (off == 0)
1343 		/* LINTED const cast away */
1344 		return (struct nfsdircache *)&dzero;
1345 
1346 	if (!np->n_dircache)
1347 		return NULL;
1348 
1349 	/*
1350 	 * We use a 32bit cookie as search key, directly reconstruct
1351 	 * the hashentry. Else use the hashfunction.
1352 	 */
1353 	if (do32) {
1354 		ent = (u_int32_t)off >> 24;
1355 		if (ent >= NFS_DIRHASHSIZ)
1356 			return NULL;
1357 		ndhp = &np->n_dircache[ent];
1358 	} else {
1359 		ndhp = NFSDIRHASH(np, off);
1360 	}
1361 
1362 	if (hashent)
1363 		*hashent = (int)(ndhp - np->n_dircache);
1364 
1365 	NFSDC_LOCK(np);
1366 	if (do32) {
1367 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1368 			if (ndp->dc_cookie32 == (u_int32_t)off) {
1369 				/*
1370 				 * An invalidated entry will become the
1371 				 * start of a new block fetched from
1372 				 * the server.
1373 				 */
1374 				if (ndp->dc_flags & NFSDC_INVALID) {
1375 					ndp->dc_blkcookie = ndp->dc_cookie;
1376 					ndp->dc_entry = 0;
1377 					ndp->dc_flags &= ~NFSDC_INVALID;
1378 				}
1379 				break;
1380 			}
1381 		}
1382 	} else {
1383 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1384 			if (ndp->dc_cookie == off)
1385 				break;
1386 		}
1387 	}
1388 	if (ndp != NULL)
1389 		ndp->dc_refcnt++;
1390 	NFSDC_UNLOCK(np);
1391 	return ndp;
1392 }
1393 
1394 
1395 struct nfsdircache *
1396 nfs_enterdircache(vp, off, blkoff, en, blkno)
1397 	struct vnode *vp;
1398 	off_t off, blkoff;
1399 	int en;
1400 	daddr_t blkno;
1401 {
1402 	struct nfsnode *np = VTONFS(vp);
1403 	struct nfsdirhashhead *ndhp;
1404 	struct nfsdircache *ndp = NULL;
1405 	struct nfsdircache *newndp = NULL;
1406 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1407 	int hashent, gen, overwrite;
1408 
1409 	/*
1410 	 * XXX refuse entries for offset 0. amd(8) erroneously sets
1411 	 * cookie 0 for the '.' entry, making this necessary. This
1412 	 * isn't so bad, as 0 is a special case anyway.
1413 	 */
1414 	if (off == 0)
1415 		/* LINTED const cast away */
1416 		return (struct nfsdircache *)&dzero;
1417 
1418 	if (!np->n_dircache)
1419 		/*
1420 		 * XXX would like to do this in nfs_nget but vtype
1421 		 * isn't known at that time.
1422 		 */
1423 		nfs_initdircache(vp);
1424 
1425 	if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens)
1426 		nfs_initdirxlatecookie(vp);
1427 
1428 retry:
1429 	ndp = nfs_searchdircache(vp, off, 0, &hashent);
1430 
1431 	NFSDC_LOCK(np);
1432 	if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) {
1433 		/*
1434 		 * Overwriting an old entry. Check if it's the same.
1435 		 * If so, just return. If not, remove the old entry.
1436 		 */
1437 		if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
1438 			goto done;
1439 		nfs_unlinkdircache(np, ndp);
1440 		nfs_putdircache_unlocked(np, ndp);
1441 		ndp = NULL;
1442 	}
1443 
1444 	ndhp = &np->n_dircache[hashent];
1445 
1446 	if (!ndp) {
1447 		if (newndp == NULL) {
1448 			NFSDC_UNLOCK(np);
1449 			newndp = malloc(sizeof(*ndp), M_NFSDIROFF, M_WAITOK);
1450 			newndp->dc_refcnt = 1;
1451 			LIST_NEXT(newndp, dc_hash) = (void *)-1;
1452 			goto retry;
1453 		}
1454 		ndp = newndp;
1455 		newndp = NULL;
1456 		overwrite = 0;
1457 		if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1458 			/*
1459 			 * We're allocating a new entry, so bump the
1460 			 * generation number.
1461 			 */
1462 			gen = ++np->n_dirgens[hashent];
1463 			if (gen == 0) {
1464 				np->n_dirgens[hashent]++;
1465 				gen++;
1466 			}
1467 			ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
1468 		}
1469 	} else
1470 		overwrite = 1;
1471 
1472 	ndp->dc_cookie = off;
1473 	ndp->dc_blkcookie = blkoff;
1474 	ndp->dc_entry = en;
1475 	ndp->dc_flags = 0;
1476 
1477 	if (overwrite)
1478 		goto done;
1479 
1480 	/*
1481 	 * If the maximum directory cookie cache size has been reached
1482 	 * for this node, take one off the front. The idea is that
1483 	 * directories are typically read front-to-back once, so that
1484 	 * the oldest entries can be thrown away without much performance
1485 	 * loss.
1486 	 */
1487 	if (np->n_dircachesize == NFS_MAXDIRCACHE) {
1488 		nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain));
1489 	} else
1490 		np->n_dircachesize++;
1491 
1492 	KASSERT(ndp->dc_refcnt == 1);
1493 	LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
1494 	TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
1495 	ndp->dc_refcnt++;
1496 done:
1497 	KASSERT(ndp->dc_refcnt > 0);
1498 	NFSDC_UNLOCK(np);
1499 	if (newndp)
1500 		nfs_putdircache(np, newndp);
1501 	return ndp;
1502 }
1503 
1504 void
1505 nfs_invaldircache(vp, forcefree)
1506 	struct vnode *vp;
1507 	int forcefree;
1508 {
1509 	struct nfsnode *np = VTONFS(vp);
1510 	struct nfsdircache *ndp = NULL;
1511 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1512 
1513 #ifdef DIAGNOSTIC
1514 	if (vp->v_type != VDIR)
1515 		panic("nfs: invaldircache: not dir");
1516 #endif
1517 
1518 	if (!np->n_dircache)
1519 		return;
1520 
1521 	NFSDC_LOCK(np);
1522 	if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
1523 		while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) {
1524 			KASSERT(!forcefree || ndp->dc_refcnt == 1);
1525 			nfs_unlinkdircache(np, ndp);
1526 		}
1527 		np->n_dircachesize = 0;
1528 		if (forcefree && np->n_dirgens) {
1529 			FREE(np->n_dirgens, M_NFSDIROFF);
1530 			np->n_dirgens = NULL;
1531 		}
1532 	} else {
1533 		TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain)
1534 			ndp->dc_flags |= NFSDC_INVALID;
1535 	}
1536 
1537 	NFSDC_UNLOCK(np);
1538 }
1539 
1540 /*
1541  * Called once before VFS init to initialize shared and
1542  * server-specific data structures.
1543  */
1544 void
1545 nfs_init()
1546 {
1547 	nfsrtt.pos = 0;
1548 	rpc_vers = txdr_unsigned(RPC_VER2);
1549 	rpc_call = txdr_unsigned(RPC_CALL);
1550 	rpc_reply = txdr_unsigned(RPC_REPLY);
1551 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1552 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1553 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1554 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1555 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1556 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1557 	nfs_prog = txdr_unsigned(NFS_PROG);
1558 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
1559 	nfs_true = txdr_unsigned(TRUE);
1560 	nfs_false = txdr_unsigned(FALSE);
1561 	nfs_xdrneg1 = txdr_unsigned(-1);
1562 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1563 	if (nfs_ticks < 1)
1564 		nfs_ticks = 1;
1565 #ifdef NFSSERVER
1566 	nfsrv_init(0);			/* Init server data structures */
1567 	nfsrv_initcache();		/* Init the server request cache */
1568 	pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript),
1569 	    0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr);
1570 #endif /* NFSSERVER */
1571 
1572 #if defined(NFSSERVER) || !defined(NFS_V2_ONLY)
1573 	/*
1574 	 * Initialize the nqnfs data structures.
1575 	 */
1576 	if (nqnfsstarttime == 0) {
1577 		nqnfsstarttime = boottime.tv_sec + nqsrv_maxlease
1578 			+ nqsrv_clockskew + nqsrv_writeslack;
1579 		NQLOADNOVRAM(nqnfsstarttime);
1580 		CIRCLEQ_INIT(&nqtimerhead);
1581 		nqfhhashtbl = hashinit(NQLCHSZ, HASH_LIST, M_NQLEASE,
1582 		    M_WAITOK, &nqfhhash);
1583 	}
1584 #endif
1585 
1586 	exithook_establish(nfs_exit, NULL);
1587 
1588 	/*
1589 	 * Initialize reply list and start timer
1590 	 */
1591 	TAILQ_INIT(&nfs_reqq);
1592 	nfs_timer(NULL);
1593 	MOWNER_ATTACH(&nfs_mowner);
1594 
1595 #ifdef NFS
1596 	/* Initialize the kqueue structures */
1597 	nfs_kqinit();
1598 	/* Initialize the iod structures */
1599 	nfs_iodinit();
1600 #endif
1601 }
1602 
1603 #ifdef NFS
1604 /*
1605  * Called once at VFS init to initialize client-specific data structures.
1606  */
1607 void
1608 nfs_vfs_init()
1609 {
1610 	nfs_nhinit();			/* Init the nfsnode table */
1611 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1612 }
1613 
1614 void
1615 nfs_vfs_reinit()
1616 {
1617 	nfs_nhreinit();
1618 }
1619 
1620 void
1621 nfs_vfs_done()
1622 {
1623 	nfs_nhdone();
1624 }
1625 
1626 /*
1627  * Attribute cache routines.
1628  * nfs_loadattrcache() - loads or updates the cache contents from attributes
1629  *	that are on the mbuf list
1630  * nfs_getattrcache() - returns valid attributes if found in cache, returns
1631  *	error otherwise
1632  */
1633 
1634 /*
1635  * Load the attribute cache (that lives in the nfsnode entry) with
1636  * the values on the mbuf list and
1637  * Iff vap not NULL
1638  *    copy the attributes to *vaper
1639  */
1640 int
1641 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags)
1642 	struct vnode **vpp;
1643 	struct mbuf **mdp;
1644 	caddr_t *dposp;
1645 	struct vattr *vaper;
1646 	int flags;
1647 {
1648 	int32_t t1;
1649 	caddr_t cp2;
1650 	int error = 0;
1651 	struct mbuf *md;
1652 	int v3 = NFS_ISV3(*vpp);
1653 
1654 	md = *mdp;
1655 	t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
1656 	error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2);
1657 	if (error)
1658 		return (error);
1659 	return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags);
1660 }
1661 
1662 int
1663 nfs_loadattrcache(vpp, fp, vaper, flags)
1664 	struct vnode **vpp;
1665 	struct nfs_fattr *fp;
1666 	struct vattr *vaper;
1667 	int flags;
1668 {
1669 	struct vnode *vp = *vpp;
1670 	struct vattr *vap;
1671 	int v3 = NFS_ISV3(vp);
1672 	enum vtype vtyp;
1673 	u_short vmode;
1674 	struct timespec mtime;
1675 	struct vnode *nvp;
1676 	int32_t rdev;
1677 	struct nfsnode *np;
1678 	extern int (**spec_nfsv2nodeop_p) __P((void *));
1679 	uid_t uid;
1680 	gid_t gid;
1681 
1682 	if (v3) {
1683 		vtyp = nfsv3tov_type(fp->fa_type);
1684 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1685 		rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1),
1686 			fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2));
1687 		fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
1688 	} else {
1689 		vtyp = nfsv2tov_type(fp->fa_type);
1690 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1691 		if (vtyp == VNON || vtyp == VREG)
1692 			vtyp = IFTOVT(vmode);
1693 		rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
1694 		fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
1695 
1696 		/*
1697 		 * Really ugly NFSv2 kludge.
1698 		 */
1699 		if (vtyp == VCHR && rdev == 0xffffffff)
1700 			vtyp = VFIFO;
1701 	}
1702 
1703 	vmode &= ALLPERMS;
1704 
1705 	/*
1706 	 * If v_type == VNON it is a new node, so fill in the v_type,
1707 	 * n_mtime fields. Check to see if it represents a special
1708 	 * device, and if so, check for a possible alias. Once the
1709 	 * correct vnode has been obtained, fill in the rest of the
1710 	 * information.
1711 	 */
1712 	np = VTONFS(vp);
1713 	if (vp->v_type == VNON) {
1714 		vp->v_type = vtyp;
1715 		if (vp->v_type == VFIFO) {
1716 			extern int (**fifo_nfsv2nodeop_p) __P((void *));
1717 			vp->v_op = fifo_nfsv2nodeop_p;
1718 		} else if (vp->v_type == VREG) {
1719 			lockinit(&np->n_commitlock, PINOD, "nfsclock", 0, 0);
1720 		} else if (vp->v_type == VCHR || vp->v_type == VBLK) {
1721 			vp->v_op = spec_nfsv2nodeop_p;
1722 			nvp = checkalias(vp, (dev_t)rdev, vp->v_mount);
1723 			if (nvp) {
1724 				/*
1725 				 * Discard unneeded vnode, but save its nfsnode.
1726 				 * Since the nfsnode does not have a lock, its
1727 				 * vnode lock has to be carried over.
1728 				 */
1729 				/*
1730 				 * XXX is the old node sure to be locked here?
1731 				 */
1732 				KASSERT(lockstatus(&vp->v_lock) ==
1733 				    LK_EXCLUSIVE);
1734 				nvp->v_data = vp->v_data;
1735 				vp->v_data = NULL;
1736 				VOP_UNLOCK(vp, 0);
1737 				vp->v_op = spec_vnodeop_p;
1738 				vrele(vp);
1739 				vgone(vp);
1740 				lockmgr(&nvp->v_lock, LK_EXCLUSIVE,
1741 				    &nvp->v_interlock);
1742 				/*
1743 				 * Reinitialize aliased node.
1744 				 */
1745 				np->n_vnode = nvp;
1746 				*vpp = vp = nvp;
1747 			}
1748 		}
1749 		np->n_mtime = mtime;
1750 	}
1751 	uid = fxdr_unsigned(uid_t, fp->fa_uid);
1752 	gid = fxdr_unsigned(gid_t, fp->fa_gid);
1753 	vap = np->n_vattr;
1754 
1755 	/*
1756 	 * Invalidate access cache if uid, gid or mode changed.
1757 	 */
1758 	if (np->n_accstamp != -1 &&
1759 	    (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode))
1760 		np->n_accstamp = -1;
1761 
1762 	vap->va_type = vtyp;
1763 	vap->va_mode = vmode;
1764 	vap->va_rdev = (dev_t)rdev;
1765 	vap->va_mtime = mtime;
1766 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1767 	switch (vtyp) {
1768 	case VDIR:
1769 		vap->va_blocksize = NFS_DIRFRAGSIZ;
1770 		break;
1771 	case VBLK:
1772 		vap->va_blocksize = BLKDEV_IOSIZE;
1773 		break;
1774 	case VCHR:
1775 		vap->va_blocksize = MAXBSIZE;
1776 		break;
1777 	default:
1778 		vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize :
1779 		    fxdr_unsigned(int32_t, fp->fa2_blocksize);
1780 		break;
1781 	}
1782 	if (v3) {
1783 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1784 		vap->va_uid = uid;
1785 		vap->va_gid = gid;
1786 		vap->va_size = fxdr_hyper(&fp->fa3_size);
1787 		vap->va_bytes = fxdr_hyper(&fp->fa3_used);
1788 		vap->va_fileid = fxdr_unsigned(int32_t,
1789 		    fp->fa3_fileid.nfsuquad[1]);
1790 		fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
1791 		fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime);
1792 		vap->va_flags = 0;
1793 		vap->va_filerev = 0;
1794 	} else {
1795 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1796 		vap->va_uid = uid;
1797 		vap->va_gid = gid;
1798 		vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
1799 		vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks)
1800 		    * NFS_FABLKSIZE;
1801 		vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
1802 		fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
1803 		vap->va_flags = 0;
1804 		vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t,
1805 		    fp->fa2_ctime.nfsv2_sec);
1806 		vap->va_ctime.tv_nsec = 0;
1807 		vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
1808 		vap->va_filerev = 0;
1809 	}
1810 	if (vap->va_size != np->n_size) {
1811 		if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) {
1812 			vap->va_size = np->n_size;
1813 		} else {
1814 			np->n_size = vap->va_size;
1815 			if (vap->va_type == VREG) {
1816 				if ((flags & NAC_NOTRUNC)
1817 				    && np->n_size < vp->v_size) {
1818 					/*
1819 					 * we can't free pages now because
1820 					 * the pages can be owned by ourselves.
1821 					 */
1822 					np->n_flag |= NTRUNCDELAYED;
1823 				}
1824 				else {
1825 					uvm_vnp_setsize(vp, np->n_size);
1826 				}
1827 			}
1828 		}
1829 	}
1830 	np->n_attrstamp = time.tv_sec;
1831 	if (vaper != NULL) {
1832 		memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(*vap));
1833 		if (np->n_flag & NCHG) {
1834 			if (np->n_flag & NACC)
1835 				vaper->va_atime = np->n_atim;
1836 			if (np->n_flag & NUPD)
1837 				vaper->va_mtime = np->n_mtim;
1838 		}
1839 	}
1840 	return (0);
1841 }
1842 
1843 /*
1844  * Check the time stamp
1845  * If the cache is valid, copy contents to *vap and return 0
1846  * otherwise return an error
1847  */
1848 int
1849 nfs_getattrcache(vp, vaper)
1850 	struct vnode *vp;
1851 	struct vattr *vaper;
1852 {
1853 	struct nfsnode *np = VTONFS(vp);
1854 	struct vattr *vap;
1855 
1856 	if (np->n_attrstamp == 0 ||
1857 	    (time.tv_sec - np->n_attrstamp) >= NFS_ATTRTIMEO(np)) {
1858 		nfsstats.attrcache_misses++;
1859 		return (ENOENT);
1860 	}
1861 	nfsstats.attrcache_hits++;
1862 	vap = np->n_vattr;
1863 	if (vap->va_size != np->n_size) {
1864 		if (vap->va_type == VREG) {
1865 			if (np->n_flag & NMODIFIED) {
1866 				if (vap->va_size < np->n_size)
1867 					vap->va_size = np->n_size;
1868 				else
1869 					np->n_size = vap->va_size;
1870 			} else
1871 				np->n_size = vap->va_size;
1872 			uvm_vnp_setsize(vp, np->n_size);
1873 		} else
1874 			np->n_size = vap->va_size;
1875 	}
1876 	memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(struct vattr));
1877 	if (np->n_flag & NCHG) {
1878 		if (np->n_flag & NACC)
1879 			vaper->va_atime = np->n_atim;
1880 		if (np->n_flag & NUPD)
1881 			vaper->va_mtime = np->n_mtim;
1882 	}
1883 	return (0);
1884 }
1885 
1886 void
1887 nfs_delayedtruncate(vp)
1888 	struct vnode *vp;
1889 {
1890 	struct nfsnode *np = VTONFS(vp);
1891 
1892 	if (np->n_flag & NTRUNCDELAYED) {
1893 		np->n_flag &= ~NTRUNCDELAYED;
1894 		uvm_vnp_setsize(vp, np->n_size);
1895 	}
1896 }
1897 
1898 /*
1899  * Heuristic to see if the server XDR encodes directory cookies or not.
1900  * it is not supposed to, but a lot of servers may do this. Also, since
1901  * most/all servers will implement V2 as well, it is expected that they
1902  * may return just 32 bits worth of cookie information, so we need to
1903  * find out in which 32 bits this information is available. We do this
1904  * to avoid trouble with emulated binaries that can't handle 64 bit
1905  * directory offsets.
1906  */
1907 
1908 void
1909 nfs_cookieheuristic(vp, flagp, p, cred)
1910 	struct vnode *vp;
1911 	int *flagp;
1912 	struct proc *p;
1913 	struct ucred *cred;
1914 {
1915 	struct uio auio;
1916 	struct iovec aiov;
1917 	caddr_t buf, cp;
1918 	struct dirent *dp;
1919 	off_t *cookies = NULL, *cop;
1920 	int error, eof, nc, len;
1921 
1922 	MALLOC(buf, caddr_t, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK);
1923 
1924 	aiov.iov_base = buf;
1925 	aiov.iov_len = NFS_DIRFRAGSIZ;
1926 	auio.uio_iov = &aiov;
1927 	auio.uio_iovcnt = 1;
1928 	auio.uio_rw = UIO_READ;
1929 	auio.uio_segflg = UIO_SYSSPACE;
1930 	auio.uio_procp = NULL;
1931 	auio.uio_resid = NFS_DIRFRAGSIZ;
1932 	auio.uio_offset = 0;
1933 
1934 	error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc);
1935 
1936 	len = NFS_DIRFRAGSIZ - auio.uio_resid;
1937 	if (error || len == 0) {
1938 		FREE(buf, M_TEMP);
1939 		if (cookies)
1940 			free(cookies, M_TEMP);
1941 		return;
1942 	}
1943 
1944 	/*
1945 	 * Find the first valid entry and look at its offset cookie.
1946 	 */
1947 
1948 	cp = buf;
1949 	for (cop = cookies; len > 0; len -= dp->d_reclen) {
1950 		dp = (struct dirent *)cp;
1951 		if (dp->d_fileno != 0 && len >= dp->d_reclen) {
1952 			if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) {
1953 				*flagp |= NFSMNT_SWAPCOOKIE;
1954 				nfs_invaldircache(vp, 0);
1955 				nfs_vinvalbuf(vp, 0, cred, p, 1);
1956 			}
1957 			break;
1958 		}
1959 		cop++;
1960 		cp += dp->d_reclen;
1961 	}
1962 
1963 	FREE(buf, M_TEMP);
1964 	free(cookies, M_TEMP);
1965 }
1966 #endif /* NFS */
1967 
1968 /*
1969  * Set up nameidata for a lookup() call and do it.
1970  *
1971  * If pubflag is set, this call is done for a lookup operation on the
1972  * public filehandle. In that case we allow crossing mountpoints and
1973  * absolute pathnames. However, the caller is expected to check that
1974  * the lookup result is within the public fs, and deny access if
1975  * it is not.
1976  */
1977 int
1978 nfs_namei(ndp, fhp, len, slp, nam, mdp, dposp, retdirp, p, kerbflag, pubflag)
1979 	struct nameidata *ndp;
1980 	fhandle_t *fhp;
1981 	uint32_t len;
1982 	struct nfssvc_sock *slp;
1983 	struct mbuf *nam;
1984 	struct mbuf **mdp;
1985 	caddr_t *dposp;
1986 	struct vnode **retdirp;
1987 	struct proc *p;
1988 	int kerbflag, pubflag;
1989 {
1990 	int i, rem;
1991 	struct mbuf *md;
1992 	char *fromcp, *tocp, *cp;
1993 	struct iovec aiov;
1994 	struct uio auio;
1995 	struct vnode *dp;
1996 	int error, rdonly, linklen;
1997 	struct componentname *cnp = &ndp->ni_cnd;
1998 
1999 	*retdirp = (struct vnode *)0;
2000 
2001 	if ((len + 1) > MAXPATHLEN)
2002 		return (ENAMETOOLONG);
2003 	cnp->cn_pnbuf = PNBUF_GET();
2004 
2005 	/*
2006 	 * Copy the name from the mbuf list to ndp->ni_pnbuf
2007 	 * and set the various ndp fields appropriately.
2008 	 */
2009 	fromcp = *dposp;
2010 	tocp = cnp->cn_pnbuf;
2011 	md = *mdp;
2012 	rem = mtod(md, caddr_t) + md->m_len - fromcp;
2013 	for (i = 0; i < len; i++) {
2014 		while (rem == 0) {
2015 			md = md->m_next;
2016 			if (md == NULL) {
2017 				error = EBADRPC;
2018 				goto out;
2019 			}
2020 			fromcp = mtod(md, caddr_t);
2021 			rem = md->m_len;
2022 		}
2023 		if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
2024 			error = EACCES;
2025 			goto out;
2026 		}
2027 		*tocp++ = *fromcp++;
2028 		rem--;
2029 	}
2030 	*tocp = '\0';
2031 	*mdp = md;
2032 	*dposp = fromcp;
2033 	len = nfsm_rndup(len)-len;
2034 	if (len > 0) {
2035 		if (rem >= len)
2036 			*dposp += len;
2037 		else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
2038 			goto out;
2039 	}
2040 
2041 	/*
2042 	 * Extract and set starting directory.
2043 	 */
2044 	error = nfsrv_fhtovp(fhp, FALSE, &dp, ndp->ni_cnd.cn_cred, slp,
2045 	    nam, &rdonly, kerbflag, pubflag);
2046 	if (error)
2047 		goto out;
2048 	if (dp->v_type != VDIR) {
2049 		vrele(dp);
2050 		error = ENOTDIR;
2051 		goto out;
2052 	}
2053 
2054 	if (rdonly)
2055 		cnp->cn_flags |= RDONLY;
2056 
2057 	*retdirp = dp;
2058 
2059 	if (pubflag) {
2060 		/*
2061 		 * Oh joy. For WebNFS, handle those pesky '%' escapes,
2062 		 * and the 'native path' indicator.
2063 		 */
2064 		cp = PNBUF_GET();
2065 		fromcp = cnp->cn_pnbuf;
2066 		tocp = cp;
2067 		if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
2068 			switch ((unsigned char)*fromcp) {
2069 			case WEBNFS_NATIVE_CHAR:
2070 				/*
2071 				 * 'Native' path for us is the same
2072 				 * as a path according to the NFS spec,
2073 				 * just skip the escape char.
2074 				 */
2075 				fromcp++;
2076 				break;
2077 			/*
2078 			 * More may be added in the future, range 0x80-0xff
2079 			 */
2080 			default:
2081 				error = EIO;
2082 				PNBUF_PUT(cp);
2083 				goto out;
2084 			}
2085 		}
2086 		/*
2087 		 * Translate the '%' escapes, URL-style.
2088 		 */
2089 		while (*fromcp != '\0') {
2090 			if (*fromcp == WEBNFS_ESC_CHAR) {
2091 				if (fromcp[1] != '\0' && fromcp[2] != '\0') {
2092 					fromcp++;
2093 					*tocp++ = HEXSTRTOI(fromcp);
2094 					fromcp += 2;
2095 					continue;
2096 				} else {
2097 					error = ENOENT;
2098 					PNBUF_PUT(cp);
2099 					goto out;
2100 				}
2101 			} else
2102 				*tocp++ = *fromcp++;
2103 		}
2104 		*tocp = '\0';
2105 		PNBUF_PUT(cnp->cn_pnbuf);
2106 		cnp->cn_pnbuf = cp;
2107 	}
2108 
2109 	ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1;
2110 	ndp->ni_segflg = UIO_SYSSPACE;
2111 	ndp->ni_rootdir = rootvnode;
2112 
2113 	if (pubflag) {
2114 		ndp->ni_loopcnt = 0;
2115 		if (cnp->cn_pnbuf[0] == '/')
2116 			dp = rootvnode;
2117 	} else {
2118 		cnp->cn_flags |= NOCROSSMOUNT;
2119 	}
2120 
2121 	cnp->cn_proc = p;
2122 	VREF(dp);
2123 
2124     for (;;) {
2125 	cnp->cn_nameptr = cnp->cn_pnbuf;
2126 	ndp->ni_startdir = dp;
2127 	/*
2128 	 * And call lookup() to do the real work
2129 	 */
2130 	error = lookup(ndp);
2131 	if (error) {
2132 		PNBUF_PUT(cnp->cn_pnbuf);
2133 		return (error);
2134 	}
2135 	/*
2136 	 * Check for encountering a symbolic link
2137 	 */
2138 	if ((cnp->cn_flags & ISSYMLINK) == 0) {
2139 		if (cnp->cn_flags & (SAVENAME | SAVESTART))
2140 			cnp->cn_flags |= HASBUF;
2141 		else
2142 			PNBUF_PUT(cnp->cn_pnbuf);
2143 		return (0);
2144 	} else {
2145 		if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
2146 			VOP_UNLOCK(ndp->ni_dvp, 0);
2147 		if (!pubflag) {
2148 			error = EINVAL;
2149 			break;
2150 		}
2151 
2152 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
2153 			error = ELOOP;
2154 			break;
2155 		}
2156 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
2157 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred,
2158 			    cnp->cn_proc);
2159 			if (error != 0)
2160 				break;
2161 		}
2162 		if (ndp->ni_pathlen > 1)
2163 			cp = PNBUF_GET();
2164 		else
2165 			cp = cnp->cn_pnbuf;
2166 		aiov.iov_base = cp;
2167 		aiov.iov_len = MAXPATHLEN;
2168 		auio.uio_iov = &aiov;
2169 		auio.uio_iovcnt = 1;
2170 		auio.uio_offset = 0;
2171 		auio.uio_rw = UIO_READ;
2172 		auio.uio_segflg = UIO_SYSSPACE;
2173 		auio.uio_procp = NULL;
2174 		auio.uio_resid = MAXPATHLEN;
2175 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
2176 		if (error) {
2177 		badlink:
2178 			if (ndp->ni_pathlen > 1)
2179 				PNBUF_PUT(cp);
2180 			break;
2181 		}
2182 		linklen = MAXPATHLEN - auio.uio_resid;
2183 		if (linklen == 0) {
2184 			error = ENOENT;
2185 			goto badlink;
2186 		}
2187 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
2188 			error = ENAMETOOLONG;
2189 			goto badlink;
2190 		}
2191 		if (ndp->ni_pathlen > 1) {
2192 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
2193 			PNBUF_PUT(cnp->cn_pnbuf);
2194 			cnp->cn_pnbuf = cp;
2195 		} else
2196 			cnp->cn_pnbuf[linklen] = '\0';
2197 		ndp->ni_pathlen += linklen;
2198 		vput(ndp->ni_vp);
2199 		dp = ndp->ni_dvp;
2200 		/*
2201 		 * Check if root directory should replace current directory.
2202 		 */
2203 		if (cnp->cn_pnbuf[0] == '/') {
2204 			vrele(dp);
2205 			dp = ndp->ni_rootdir;
2206 			VREF(dp);
2207 		}
2208 	}
2209    }
2210 	vrele(ndp->ni_dvp);
2211 	vput(ndp->ni_vp);
2212 	ndp->ni_vp = NULL;
2213 out:
2214 	PNBUF_PUT(cnp->cn_pnbuf);
2215 	return (error);
2216 }
2217 
2218 /*
2219  * A fiddled version of m_adj() that ensures null fill to a 32-bit
2220  * boundary and only trims off the back end
2221  *
2222  * 1. trim off 'len' bytes as m_adj(mp, -len).
2223  * 2. add zero-padding 'nul' bytes at the end of the mbuf chain.
2224  */
2225 void
2226 nfs_zeropad(mp, len, nul)
2227 	struct mbuf *mp;
2228 	int len;
2229 	int nul;
2230 {
2231 	struct mbuf *m;
2232 	int count;
2233 
2234 	/*
2235 	 * Trim from tail.  Scan the mbuf chain,
2236 	 * calculating its length and finding the last mbuf.
2237 	 * If the adjustment only affects this mbuf, then just
2238 	 * adjust and return.  Otherwise, rescan and truncate
2239 	 * after the remaining size.
2240 	 */
2241 	count = 0;
2242 	m = mp;
2243 	for (;;) {
2244 		count += m->m_len;
2245 		if (m->m_next == NULL)
2246 			break;
2247 		m = m->m_next;
2248 	}
2249 
2250 	KDASSERT(count >= len);
2251 
2252 	if (m->m_len >= len) {
2253 		m->m_len -= len;
2254 	} else {
2255 		count -= len;
2256 		/*
2257 		 * Correct length for chain is "count".
2258 		 * Find the mbuf with last data, adjust its length,
2259 		 * and toss data from remaining mbufs on chain.
2260 		 */
2261 		for (m = mp; m; m = m->m_next) {
2262 			if (m->m_len >= count) {
2263 				m->m_len = count;
2264 				break;
2265 			}
2266 			count -= m->m_len;
2267 		}
2268 		m_freem(m->m_next);
2269 		m->m_next = NULL;
2270 	}
2271 
2272 	KDASSERT(m->m_next == NULL);
2273 
2274 	/*
2275 	 * zero-padding.
2276 	 */
2277 	if (nul > 0) {
2278 		char *cp;
2279 		int i;
2280 
2281 		if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) {
2282 			struct mbuf *n;
2283 
2284 			KDASSERT(MLEN >= nul);
2285 			n = m_get(M_WAIT, MT_DATA);
2286 			MCLAIM(n, &nfs_mowner);
2287 			n->m_len = nul;
2288 			n->m_next = NULL;
2289 			m->m_next = n;
2290 			cp = mtod(n, caddr_t);
2291 		} else {
2292 			cp = mtod(m, caddr_t) + m->m_len;
2293 			m->m_len += nul;
2294 		}
2295 		for (i = 0; i < nul; i++)
2296 			*cp++ = '\0';
2297 	}
2298 	return;
2299 }
2300 
2301 /*
2302  * Make these functions instead of macros, so that the kernel text size
2303  * doesn't get too big...
2304  */
2305 void
2306 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp)
2307 	struct nfsrv_descript *nfsd;
2308 	int before_ret;
2309 	struct vattr *before_vap;
2310 	int after_ret;
2311 	struct vattr *after_vap;
2312 	struct mbuf **mbp;
2313 	char **bposp;
2314 {
2315 	struct mbuf *mb = *mbp;
2316 	char *bpos = *bposp;
2317 	u_int32_t *tl;
2318 
2319 	if (before_ret) {
2320 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2321 		*tl = nfs_false;
2322 	} else {
2323 		nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
2324 		*tl++ = nfs_true;
2325 		txdr_hyper(before_vap->va_size, tl);
2326 		tl += 2;
2327 		txdr_nfsv3time(&(before_vap->va_mtime), tl);
2328 		tl += 2;
2329 		txdr_nfsv3time(&(before_vap->va_ctime), tl);
2330 	}
2331 	*bposp = bpos;
2332 	*mbp = mb;
2333 	nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
2334 }
2335 
2336 void
2337 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp)
2338 	struct nfsrv_descript *nfsd;
2339 	int after_ret;
2340 	struct vattr *after_vap;
2341 	struct mbuf **mbp;
2342 	char **bposp;
2343 {
2344 	struct mbuf *mb = *mbp;
2345 	char *bpos = *bposp;
2346 	u_int32_t *tl;
2347 	struct nfs_fattr *fp;
2348 
2349 	if (after_ret) {
2350 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2351 		*tl = nfs_false;
2352 	} else {
2353 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
2354 		*tl++ = nfs_true;
2355 		fp = (struct nfs_fattr *)tl;
2356 		nfsm_srvfattr(nfsd, after_vap, fp);
2357 	}
2358 	*mbp = mb;
2359 	*bposp = bpos;
2360 }
2361 
2362 void
2363 nfsm_srvfattr(nfsd, vap, fp)
2364 	struct nfsrv_descript *nfsd;
2365 	struct vattr *vap;
2366 	struct nfs_fattr *fp;
2367 {
2368 
2369 	fp->fa_nlink = txdr_unsigned(vap->va_nlink);
2370 	fp->fa_uid = txdr_unsigned(vap->va_uid);
2371 	fp->fa_gid = txdr_unsigned(vap->va_gid);
2372 	if (nfsd->nd_flag & ND_NFSV3) {
2373 		fp->fa_type = vtonfsv3_type(vap->va_type);
2374 		fp->fa_mode = vtonfsv3_mode(vap->va_mode);
2375 		txdr_hyper(vap->va_size, &fp->fa3_size);
2376 		txdr_hyper(vap->va_bytes, &fp->fa3_used);
2377 		fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
2378 		fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
2379 		fp->fa3_fsid.nfsuquad[0] = 0;
2380 		fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
2381 		fp->fa3_fileid.nfsuquad[0] = 0;
2382 		fp->fa3_fileid.nfsuquad[1] = txdr_unsigned(vap->va_fileid);
2383 		txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
2384 		txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
2385 		txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
2386 	} else {
2387 		fp->fa_type = vtonfsv2_type(vap->va_type);
2388 		fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
2389 		fp->fa2_size = txdr_unsigned(vap->va_size);
2390 		fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
2391 		if (vap->va_type == VFIFO)
2392 			fp->fa2_rdev = 0xffffffff;
2393 		else
2394 			fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
2395 		fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
2396 		fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
2397 		fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
2398 		txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
2399 		txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
2400 		txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
2401 	}
2402 }
2403 
2404 /*
2405  * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
2406  * 	- look up fsid in mount list (if not found ret error)
2407  *	- get vp and export rights by calling VFS_FHTOVP()
2408  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2409  *	- if not lockflag unlock it with VOP_UNLOCK()
2410  */
2411 int
2412 nfsrv_fhtovp(fhp, lockflag, vpp, cred, slp, nam, rdonlyp, kerbflag, pubflag)
2413 	fhandle_t *fhp;
2414 	int lockflag;
2415 	struct vnode **vpp;
2416 	struct ucred *cred;
2417 	struct nfssvc_sock *slp;
2418 	struct mbuf *nam;
2419 	int *rdonlyp;
2420 	int kerbflag;
2421 {
2422 	struct mount *mp;
2423 	int i;
2424 	struct ucred *credanon;
2425 	int error, exflags;
2426 	struct sockaddr_in *saddr;
2427 
2428 	*vpp = (struct vnode *)0;
2429 
2430 	if (nfs_ispublicfh(fhp)) {
2431 		if (!pubflag || !nfs_pub.np_valid)
2432 			return (ESTALE);
2433 		fhp = &nfs_pub.np_handle;
2434 	}
2435 
2436 	mp = vfs_getvfs(&fhp->fh_fsid);
2437 	if (!mp)
2438 		return (ESTALE);
2439 	error = VFS_CHECKEXP(mp, nam, &exflags, &credanon);
2440 	if (error)
2441 		return (error);
2442 	error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
2443 	if (error)
2444 		return (error);
2445 
2446 	if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
2447 		saddr = mtod(nam, struct sockaddr_in *);
2448 		if ((saddr->sin_family == AF_INET) &&
2449 		    ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
2450 			vput(*vpp);
2451 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2452 		}
2453 #ifdef INET6
2454 		if ((saddr->sin_family == AF_INET6) &&
2455 		    ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) {
2456 			vput(*vpp);
2457 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2458 		}
2459 #endif
2460 	}
2461 	/*
2462 	 * Check/setup credentials.
2463 	 */
2464 	if (exflags & MNT_EXKERB) {
2465 		if (!kerbflag) {
2466 			vput(*vpp);
2467 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2468 		}
2469 	} else if (kerbflag) {
2470 		vput(*vpp);
2471 		return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2472 	} else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
2473 		cred->cr_uid = credanon->cr_uid;
2474 		cred->cr_gid = credanon->cr_gid;
2475 		for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
2476 			cred->cr_groups[i] = credanon->cr_groups[i];
2477 		cred->cr_ngroups = i;
2478 	}
2479 	if (exflags & MNT_EXRDONLY)
2480 		*rdonlyp = 1;
2481 	else
2482 		*rdonlyp = 0;
2483 	if (!lockflag)
2484 		VOP_UNLOCK(*vpp, 0);
2485 	return (0);
2486 }
2487 
2488 /*
2489  * WebNFS: check if a filehandle is a public filehandle. For v3, this
2490  * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
2491  * transformed this to all zeroes in both cases, so check for it.
2492  */
2493 int
2494 nfs_ispublicfh(fhp)
2495 	fhandle_t *fhp;
2496 {
2497 	char *cp = (char *)fhp;
2498 	int i;
2499 
2500 	for (i = 0; i < NFSX_V3FH; i++)
2501 		if (*cp++ != 0)
2502 			return (FALSE);
2503 	return (TRUE);
2504 }
2505 
2506 /*
2507  * This function compares two net addresses by family and returns TRUE
2508  * if they are the same host.
2509  * If there is any doubt, return FALSE.
2510  * The AF_INET family is handled as a special case so that address mbufs
2511  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
2512  */
2513 int
2514 netaddr_match(family, haddr, nam)
2515 	int family;
2516 	union nethostaddr *haddr;
2517 	struct mbuf *nam;
2518 {
2519 	struct sockaddr_in *inetaddr;
2520 
2521 	switch (family) {
2522 	case AF_INET:
2523 		inetaddr = mtod(nam, struct sockaddr_in *);
2524 		if (inetaddr->sin_family == AF_INET &&
2525 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
2526 			return (1);
2527 		break;
2528 #ifdef INET6
2529 	case AF_INET6:
2530 	    {
2531 		struct sockaddr_in6 *sin6_1, *sin6_2;
2532 
2533 		sin6_1 = mtod(nam, struct sockaddr_in6 *);
2534 		sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
2535 		if (sin6_1->sin6_family == AF_INET6 &&
2536 		    IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
2537 			return 1;
2538 	    }
2539 #endif
2540 #ifdef ISO
2541 	case AF_ISO:
2542 	    {
2543 		struct sockaddr_iso *isoaddr1, *isoaddr2;
2544 
2545 		isoaddr1 = mtod(nam, struct sockaddr_iso *);
2546 		isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *);
2547 		if (isoaddr1->siso_family == AF_ISO &&
2548 		    isoaddr1->siso_nlen > 0 &&
2549 		    isoaddr1->siso_nlen == isoaddr2->siso_nlen &&
2550 		    SAME_ISOADDR(isoaddr1, isoaddr2))
2551 			return (1);
2552 		break;
2553 	    }
2554 #endif	/* ISO */
2555 	default:
2556 		break;
2557 	};
2558 	return (0);
2559 }
2560 
2561 /*
2562  * The write verifier has changed (probably due to a server reboot), so all
2563  * PG_NEEDCOMMIT pages will have to be written again. Since they are marked
2564  * as dirty or are being written out just now, all this takes is clearing
2565  * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for
2566  * the mount point.
2567  */
2568 void
2569 nfs_clearcommit(mp)
2570 	struct mount *mp;
2571 {
2572 	struct vnode *vp;
2573 	struct nfsnode *np;
2574 	struct vm_page *pg;
2575 	struct nfsmount *nmp = VFSTONFS(mp);
2576 
2577 	lockmgr(&nmp->nm_writeverflock, LK_EXCLUSIVE, NULL);
2578 
2579 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2580 		KASSERT(vp->v_mount == mp);
2581 		if (vp->v_type == VNON)
2582 			continue;
2583 		np = VTONFS(vp);
2584 		np->n_pushlo = np->n_pushhi = np->n_pushedlo =
2585 		    np->n_pushedhi = 0;
2586 		np->n_commitflags &=
2587 		    ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
2588 		simple_lock(&vp->v_uobj.vmobjlock);
2589 		TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) {
2590 			pg->flags &= ~PG_NEEDCOMMIT;
2591 		}
2592 		simple_unlock(&vp->v_uobj.vmobjlock);
2593 	}
2594 	simple_lock(&nmp->nm_slock);
2595 	nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF;
2596 	simple_unlock(&nmp->nm_slock);
2597 	lockmgr(&nmp->nm_writeverflock, LK_RELEASE, NULL);
2598 }
2599 
2600 void
2601 nfs_merge_commit_ranges(vp)
2602 	struct vnode *vp;
2603 {
2604 	struct nfsnode *np = VTONFS(vp);
2605 
2606 	KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID);
2607 
2608 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2609 		np->n_pushedlo = np->n_pushlo;
2610 		np->n_pushedhi = np->n_pushhi;
2611 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2612 	} else {
2613 		if (np->n_pushlo < np->n_pushedlo)
2614 			np->n_pushedlo = np->n_pushlo;
2615 		if (np->n_pushhi > np->n_pushedhi)
2616 			np->n_pushedhi = np->n_pushhi;
2617 	}
2618 
2619 	np->n_pushlo = np->n_pushhi = 0;
2620 	np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
2621 
2622 #ifdef NFS_DEBUG_COMMIT
2623 	printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2624 	    (unsigned)np->n_pushedhi);
2625 #endif
2626 }
2627 
2628 int
2629 nfs_in_committed_range(vp, off, len)
2630 	struct vnode *vp;
2631 	off_t off, len;
2632 {
2633 	struct nfsnode *np = VTONFS(vp);
2634 	off_t lo, hi;
2635 
2636 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2637 		return 0;
2638 	lo = off;
2639 	hi = lo + len;
2640 
2641 	return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
2642 }
2643 
2644 int
2645 nfs_in_tobecommitted_range(vp, off, len)
2646 	struct vnode *vp;
2647 	off_t off, len;
2648 {
2649 	struct nfsnode *np = VTONFS(vp);
2650 	off_t lo, hi;
2651 
2652 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2653 		return 0;
2654 	lo = off;
2655 	hi = lo + len;
2656 
2657 	return (lo >= np->n_pushlo && hi <= np->n_pushhi);
2658 }
2659 
2660 void
2661 nfs_add_committed_range(vp, off, len)
2662 	struct vnode *vp;
2663 	off_t off, len;
2664 {
2665 	struct nfsnode *np = VTONFS(vp);
2666 	off_t lo, hi;
2667 
2668 	lo = off;
2669 	hi = lo + len;
2670 
2671 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2672 		np->n_pushedlo = lo;
2673 		np->n_pushedhi = hi;
2674 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2675 	} else {
2676 		if (hi > np->n_pushedhi)
2677 			np->n_pushedhi = hi;
2678 		if (lo < np->n_pushedlo)
2679 			np->n_pushedlo = lo;
2680 	}
2681 #ifdef NFS_DEBUG_COMMIT
2682 	printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2683 	    (unsigned)np->n_pushedhi);
2684 #endif
2685 }
2686 
2687 void
2688 nfs_del_committed_range(vp, off, len)
2689 	struct vnode *vp;
2690 	off_t off, len;
2691 {
2692 	struct nfsnode *np = VTONFS(vp);
2693 	off_t lo, hi;
2694 
2695 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2696 		return;
2697 
2698 	lo = off;
2699 	hi = lo + len;
2700 
2701 	if (lo > np->n_pushedhi || hi < np->n_pushedlo)
2702 		return;
2703 	if (lo <= np->n_pushedlo)
2704 		np->n_pushedlo = hi;
2705 	else if (hi >= np->n_pushedhi)
2706 		np->n_pushedhi = lo;
2707 	else {
2708 		/*
2709 		 * XXX There's only one range. If the deleted range
2710 		 * is in the middle, pick the largest of the
2711 		 * contiguous ranges that it leaves.
2712 		 */
2713 		if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
2714 			np->n_pushedhi = lo;
2715 		else
2716 			np->n_pushedlo = hi;
2717 	}
2718 #ifdef NFS_DEBUG_COMMIT
2719 	printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2720 	    (unsigned)np->n_pushedhi);
2721 #endif
2722 }
2723 
2724 void
2725 nfs_add_tobecommitted_range(vp, off, len)
2726 	struct vnode *vp;
2727 	off_t off, len;
2728 {
2729 	struct nfsnode *np = VTONFS(vp);
2730 	off_t lo, hi;
2731 
2732 	lo = off;
2733 	hi = lo + len;
2734 
2735 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
2736 		np->n_pushlo = lo;
2737 		np->n_pushhi = hi;
2738 		np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
2739 	} else {
2740 		if (lo < np->n_pushlo)
2741 			np->n_pushlo = lo;
2742 		if (hi > np->n_pushhi)
2743 			np->n_pushhi = hi;
2744 	}
2745 #ifdef NFS_DEBUG_COMMIT
2746 	printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2747 	    (unsigned)np->n_pushhi);
2748 #endif
2749 }
2750 
2751 void
2752 nfs_del_tobecommitted_range(vp, off, len)
2753 	struct vnode *vp;
2754 	off_t off, len;
2755 {
2756 	struct nfsnode *np = VTONFS(vp);
2757 	off_t lo, hi;
2758 
2759 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2760 		return;
2761 
2762 	lo = off;
2763 	hi = lo + len;
2764 
2765 	if (lo > np->n_pushhi || hi < np->n_pushlo)
2766 		return;
2767 
2768 	if (lo <= np->n_pushlo)
2769 		np->n_pushlo = hi;
2770 	else if (hi >= np->n_pushhi)
2771 		np->n_pushhi = lo;
2772 	else {
2773 		/*
2774 		 * XXX There's only one range. If the deleted range
2775 		 * is in the middle, pick the largest of the
2776 		 * contiguous ranges that it leaves.
2777 		 */
2778 		if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
2779 			np->n_pushhi = lo;
2780 		else
2781 			np->n_pushlo = hi;
2782 	}
2783 #ifdef NFS_DEBUG_COMMIT
2784 	printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2785 	    (unsigned)np->n_pushhi);
2786 #endif
2787 }
2788 
2789 /*
2790  * Map errnos to NFS error numbers. For Version 3 also filter out error
2791  * numbers not specified for the associated procedure.
2792  */
2793 int
2794 nfsrv_errmap(nd, err)
2795 	struct nfsrv_descript *nd;
2796 	int err;
2797 {
2798 	const short *defaulterrp, *errp;
2799 
2800 	if (nd->nd_flag & ND_NFSV3) {
2801 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
2802 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2803 		while (*++errp) {
2804 			if (*errp == err)
2805 				return (err);
2806 			else if (*errp > err)
2807 				break;
2808 		}
2809 		return ((int)*defaulterrp);
2810 	    } else
2811 		return (err & 0xffff);
2812 	}
2813 	if (err <= ELAST)
2814 		return ((int)nfsrv_v2errmap[err - 1]);
2815 	return (NFSERR_IO);
2816 }
2817 
2818 /*
2819  * Sort the group list in increasing numerical order.
2820  * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
2821  *  that used to be here.)
2822  */
2823 void
2824 nfsrvw_sort(list, num)
2825         gid_t *list;
2826         int num;
2827 {
2828 	int i, j;
2829 	gid_t v;
2830 
2831 	/* Insertion sort. */
2832 	for (i = 1; i < num; i++) {
2833 		v = list[i];
2834 		/* find correct slot for value v, moving others up */
2835 		for (j = i; --j >= 0 && v < list[j];)
2836 			list[j + 1] = list[j];
2837 		list[j + 1] = v;
2838 	}
2839 }
2840 
2841 /*
2842  * copy credentials making sure that the result can be compared with memcmp().
2843  */
2844 void
2845 nfsrv_setcred(incred, outcred)
2846 	struct ucred *incred, *outcred;
2847 {
2848 	int i;
2849 
2850 	memset((caddr_t)outcred, 0, sizeof (struct ucred));
2851 	outcred->cr_ref = 1;
2852 	outcred->cr_uid = incred->cr_uid;
2853 	outcred->cr_gid = incred->cr_gid;
2854 	outcred->cr_ngroups = incred->cr_ngroups;
2855 	for (i = 0; i < incred->cr_ngroups; i++)
2856 		outcred->cr_groups[i] = incred->cr_groups[i];
2857 	nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
2858 }
2859 
2860 u_int32_t
2861 nfs_getxid()
2862 {
2863 	static u_int32_t base;
2864 	static u_int32_t nfs_xid = 0;
2865 	static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER;
2866 	u_int32_t newxid;
2867 
2868 	simple_lock(&nfs_xidlock);
2869 	/*
2870 	 * derive initial xid from system time
2871 	 * XXX time is invalid if root not yet mounted
2872 	 */
2873 	if (__predict_false(!base && (rootvp))) {
2874 		struct timeval tv;
2875 
2876 		microtime(&tv);
2877 		base = tv.tv_sec << 12;
2878 		nfs_xid = base;
2879 	}
2880 
2881 	/*
2882 	 * Skip zero xid if it should ever happen.
2883 	 */
2884 	if (__predict_false(++nfs_xid == 0))
2885 		nfs_xid++;
2886 	newxid = nfs_xid;
2887 	simple_unlock(&nfs_xidlock);
2888 
2889 	return txdr_unsigned(newxid);
2890 }
2891 
2892 /*
2893  * assign a new xid for existing request.
2894  * used for NFSERR_JUKEBOX handling.
2895  */
2896 void
2897 nfs_renewxid(struct nfsreq *req)
2898 {
2899 	u_int32_t xid;
2900 	int off;
2901 
2902 	xid = nfs_getxid();
2903 	if (req->r_nmp->nm_sotype == SOCK_STREAM)
2904 		off = sizeof(u_int32_t); /* RPC record mark */
2905 	else
2906 		off = 0;
2907 
2908 	m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid);
2909 	req->r_xid = xid;
2910 }
2911