xref: /netbsd-src/sys/nfs/nfs_subs.c (revision 5b84b3983f71fd20a534cfa5d1556623a8aaa717)
1 /*	$NetBSD: nfs_subs.c,v 1.151 2005/08/19 12:47:23 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 1.151 2005/08/19 12:47:23 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 				if ((error = copyout_proc(uiop->uio_procp,
838 				    mbufcp, uiocp, xfer)) != 0)
839 					return error;
840 			left -= xfer;
841 			len -= xfer;
842 			mbufcp += xfer;
843 			uiocp += xfer;
844 			uiop->uio_offset += xfer;
845 			uiop->uio_resid -= xfer;
846 		}
847 		if (uiop->uio_iov->iov_len <= siz) {
848 			uiop->uio_iovcnt--;
849 			uiop->uio_iov++;
850 		} else {
851 			uiop->uio_iov->iov_base =
852 			    (caddr_t)uiop->uio_iov->iov_base + uiosiz;
853 			uiop->uio_iov->iov_len -= uiosiz;
854 		}
855 		siz -= uiosiz;
856 	}
857 	*dpos = mbufcp;
858 	*mrep = mp;
859 	if (rem > 0) {
860 		if (len < rem)
861 			error = nfs_adv(mrep, dpos, rem, len);
862 		else
863 			*dpos += rem;
864 	}
865 	return (error);
866 }
867 
868 /*
869  * copies a uio scatter/gather list to an mbuf chain.
870  * NOTE: can ony handle iovcnt == 1
871  */
872 int
873 nfsm_uiotombuf(uiop, mq, siz, bpos)
874 	struct uio *uiop;
875 	struct mbuf **mq;
876 	int siz;
877 	caddr_t *bpos;
878 {
879 	char *uiocp;
880 	struct mbuf *mp, *mp2;
881 	int xfer, left, mlen;
882 	int uiosiz, clflg, rem;
883 	char *cp;
884 
885 #ifdef DIAGNOSTIC
886 	if (uiop->uio_iovcnt != 1)
887 		panic("nfsm_uiotombuf: iovcnt != 1");
888 #endif
889 
890 	if (siz > MLEN)		/* or should it >= MCLBYTES ?? */
891 		clflg = 1;
892 	else
893 		clflg = 0;
894 	rem = nfsm_rndup(siz)-siz;
895 	mp = mp2 = *mq;
896 	while (siz > 0) {
897 		left = uiop->uio_iov->iov_len;
898 		uiocp = uiop->uio_iov->iov_base;
899 		if (left > siz)
900 			left = siz;
901 		uiosiz = left;
902 		while (left > 0) {
903 			mlen = M_TRAILINGSPACE(mp);
904 			if (mlen == 0) {
905 				mp = m_get(M_WAIT, MT_DATA);
906 				MCLAIM(mp, &nfs_mowner);
907 				if (clflg)
908 					m_clget(mp, M_WAIT);
909 				mp->m_len = 0;
910 				mp2->m_next = mp;
911 				mp2 = mp;
912 				mlen = M_TRAILINGSPACE(mp);
913 			}
914 			xfer = (left > mlen) ? mlen : left;
915 			cp = mtod(mp, caddr_t) + mp->m_len;
916 #ifdef notdef
917 			/* Not Yet.. */
918 			if (uiop->uio_iov->iov_op != NULL)
919 				(*(uiop->uio_iov->iov_op))(uiocp, cp, xfer);
920 			else
921 #endif
922 			if (uiop->uio_segflg == UIO_SYSSPACE)
923 				(void)memcpy(cp, uiocp, xfer);
924 			else
925 				/*XXX: Check error */
926 				(void)copyin_proc(uiop->uio_procp, uiocp,
927 				    cp, xfer);
928 			mp->m_len += xfer;
929 			left -= xfer;
930 			uiocp += xfer;
931 			uiop->uio_offset += xfer;
932 			uiop->uio_resid -= xfer;
933 		}
934 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
935 		    uiosiz;
936 		uiop->uio_iov->iov_len -= uiosiz;
937 		siz -= uiosiz;
938 	}
939 	if (rem > 0) {
940 		if (rem > M_TRAILINGSPACE(mp)) {
941 			mp = m_get(M_WAIT, MT_DATA);
942 			MCLAIM(mp, &nfs_mowner);
943 			mp->m_len = 0;
944 			mp2->m_next = mp;
945 		}
946 		cp = mtod(mp, caddr_t) + mp->m_len;
947 		for (left = 0; left < rem; left++)
948 			*cp++ = '\0';
949 		mp->m_len += rem;
950 		*bpos = cp;
951 	} else
952 		*bpos = mtod(mp, caddr_t)+mp->m_len;
953 	*mq = mp;
954 	return (0);
955 }
956 
957 /*
958  * Get at least "siz" bytes of correctly aligned data.
959  * When called the mbuf pointers are not necessarily correct,
960  * dsosp points to what ought to be in m_data and left contains
961  * what ought to be in m_len.
962  * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
963  * cases. (The macros use the vars. dpos and dpos2)
964  */
965 int
966 nfsm_disct(mdp, dposp, siz, left, cp2)
967 	struct mbuf **mdp;
968 	caddr_t *dposp;
969 	int siz;
970 	int left;
971 	caddr_t *cp2;
972 {
973 	struct mbuf *m1, *m2;
974 	struct mbuf *havebuf = NULL;
975 	caddr_t src = *dposp;
976 	caddr_t dst;
977 	int len;
978 
979 #ifdef DEBUG
980 	if (left < 0)
981 		panic("nfsm_disct: left < 0");
982 #endif
983 	m1 = *mdp;
984 	/*
985 	 * Skip through the mbuf chain looking for an mbuf with
986 	 * some data. If the first mbuf found has enough data
987 	 * and it is correctly aligned return it.
988 	 */
989 	while (left == 0) {
990 		havebuf = m1;
991 		*mdp = m1 = m1->m_next;
992 		if (m1 == NULL)
993 			return (EBADRPC);
994 		src = mtod(m1, caddr_t);
995 		left = m1->m_len;
996 		/*
997 		 * If we start a new mbuf and it is big enough
998 		 * and correctly aligned just return it, don't
999 		 * do any pull up.
1000 		 */
1001 		if (left >= siz && nfsm_aligned(src)) {
1002 			*cp2 = src;
1003 			*dposp = src + siz;
1004 			return (0);
1005 		}
1006 	}
1007 	if (m1->m_flags & M_EXT) {
1008 		if (havebuf) {
1009 			/* If the first mbuf with data has external data
1010 			 * and there is a previous empty mbuf use it
1011 			 * to move the data into.
1012 			 */
1013 			m2 = m1;
1014 			*mdp = m1 = havebuf;
1015 			if (m1->m_flags & M_EXT) {
1016 				MEXTREMOVE(m1);
1017 			}
1018 		} else {
1019 			/*
1020 			 * If the first mbuf has a external data
1021 			 * and there is no previous empty mbuf
1022 			 * allocate a new mbuf and move the external
1023 			 * data to the new mbuf. Also make the first
1024 			 * mbuf look empty.
1025 			 */
1026 			m2 = m_get(M_WAIT, MT_DATA);
1027 			m2->m_ext = m1->m_ext;
1028 			m2->m_data = src;
1029 			m2->m_len = left;
1030 			MCLADDREFERENCE(m1, m2);
1031 			MEXTREMOVE(m1);
1032 			m2->m_next = m1->m_next;
1033 			m1->m_next = m2;
1034 		}
1035 		m1->m_len = 0;
1036 		if (m1->m_flags & M_PKTHDR)
1037 			dst = m1->m_pktdat;
1038 		else
1039 			dst = m1->m_dat;
1040 		m1->m_data = dst;
1041 	} else {
1042 		/*
1043 		 * If the first mbuf has no external data
1044 		 * move the data to the front of the mbuf.
1045 		 */
1046 		if (m1->m_flags & M_PKTHDR)
1047 			dst = m1->m_pktdat;
1048 		else
1049 			dst = m1->m_dat;
1050 		m1->m_data = dst;
1051 		if (dst != src)
1052 			memmove(dst, src, left);
1053 		dst += left;
1054 		m1->m_len = left;
1055 		m2 = m1->m_next;
1056 	}
1057 	*cp2 = m1->m_data;
1058 	*dposp = mtod(m1, caddr_t) + siz;
1059 	/*
1060 	 * Loop through mbufs pulling data up into first mbuf until
1061 	 * the first mbuf is full or there is no more data to
1062 	 * pullup.
1063 	 */
1064 	while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) {
1065 		if ((len = min(len, m2->m_len)) != 0)
1066 			memcpy(dst, m2->m_data, len);
1067 		m1->m_len += len;
1068 		dst += len;
1069 		m2->m_data += len;
1070 		m2->m_len -= len;
1071 		m2 = m2->m_next;
1072 	}
1073 	if (m1->m_len < siz)
1074 		return (EBADRPC);
1075 	return (0);
1076 }
1077 
1078 /*
1079  * Advance the position in the mbuf chain.
1080  */
1081 int
1082 nfs_adv(mdp, dposp, offs, left)
1083 	struct mbuf **mdp;
1084 	caddr_t *dposp;
1085 	int offs;
1086 	int left;
1087 {
1088 	struct mbuf *m;
1089 	int s;
1090 
1091 	m = *mdp;
1092 	s = left;
1093 	while (s < offs) {
1094 		offs -= s;
1095 		m = m->m_next;
1096 		if (m == NULL)
1097 			return (EBADRPC);
1098 		s = m->m_len;
1099 	}
1100 	*mdp = m;
1101 	*dposp = mtod(m, caddr_t)+offs;
1102 	return (0);
1103 }
1104 
1105 /*
1106  * Copy a string into mbufs for the hard cases...
1107  */
1108 int
1109 nfsm_strtmbuf(mb, bpos, cp, siz)
1110 	struct mbuf **mb;
1111 	char **bpos;
1112 	const char *cp;
1113 	long siz;
1114 {
1115 	struct mbuf *m1 = NULL, *m2;
1116 	long left, xfer, len, tlen;
1117 	u_int32_t *tl;
1118 	int putsize;
1119 
1120 	putsize = 1;
1121 	m2 = *mb;
1122 	left = M_TRAILINGSPACE(m2);
1123 	if (left > 0) {
1124 		tl = ((u_int32_t *)(*bpos));
1125 		*tl++ = txdr_unsigned(siz);
1126 		putsize = 0;
1127 		left -= NFSX_UNSIGNED;
1128 		m2->m_len += NFSX_UNSIGNED;
1129 		if (left > 0) {
1130 			memcpy((caddr_t) tl, cp, left);
1131 			siz -= left;
1132 			cp += left;
1133 			m2->m_len += left;
1134 			left = 0;
1135 		}
1136 	}
1137 	/* Loop around adding mbufs */
1138 	while (siz > 0) {
1139 		m1 = m_get(M_WAIT, MT_DATA);
1140 		MCLAIM(m1, &nfs_mowner);
1141 		if (siz > MLEN)
1142 			m_clget(m1, M_WAIT);
1143 		m1->m_len = NFSMSIZ(m1);
1144 		m2->m_next = m1;
1145 		m2 = m1;
1146 		tl = mtod(m1, u_int32_t *);
1147 		tlen = 0;
1148 		if (putsize) {
1149 			*tl++ = txdr_unsigned(siz);
1150 			m1->m_len -= NFSX_UNSIGNED;
1151 			tlen = NFSX_UNSIGNED;
1152 			putsize = 0;
1153 		}
1154 		if (siz < m1->m_len) {
1155 			len = nfsm_rndup(siz);
1156 			xfer = siz;
1157 			if (xfer < len)
1158 				*(tl+(xfer>>2)) = 0;
1159 		} else {
1160 			xfer = len = m1->m_len;
1161 		}
1162 		memcpy((caddr_t) tl, cp, xfer);
1163 		m1->m_len = len+tlen;
1164 		siz -= xfer;
1165 		cp += xfer;
1166 	}
1167 	*mb = m1;
1168 	*bpos = mtod(m1, caddr_t)+m1->m_len;
1169 	return (0);
1170 }
1171 
1172 /*
1173  * Directory caching routines. They work as follows:
1174  * - a cache is maintained per VDIR nfsnode.
1175  * - for each offset cookie that is exported to userspace, and can
1176  *   thus be thrown back at us as an offset to VOP_READDIR, store
1177  *   information in the cache.
1178  * - cached are:
1179  *   - cookie itself
1180  *   - blocknumber (essentially just a search key in the buffer cache)
1181  *   - entry number in block.
1182  *   - offset cookie of block in which this entry is stored
1183  *   - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
1184  * - entries are looked up in a hash table
1185  * - also maintained is an LRU list of entries, used to determine
1186  *   which ones to delete if the cache grows too large.
1187  * - if 32 <-> 64 translation mode is requested for a filesystem,
1188  *   the cache also functions as a translation table
1189  * - in the translation case, invalidating the cache does not mean
1190  *   flushing it, but just marking entries as invalid, except for
1191  *   the <64bit cookie, 32bitcookie> pair which is still valid, to
1192  *   still be able to use the cache as a translation table.
1193  * - 32 bit cookies are uniquely created by combining the hash table
1194  *   entry value, and one generation count per hash table entry,
1195  *   incremented each time an entry is appended to the chain.
1196  * - the cache is invalidated each time a direcory is modified
1197  * - sanity checks are also done; if an entry in a block turns
1198  *   out not to have a matching cookie, the cache is invalidated
1199  *   and a new block starting from the wanted offset is fetched from
1200  *   the server.
1201  * - directory entries as read from the server are extended to contain
1202  *   the 64bit and, optionally, the 32bit cookies, for sanity checking
1203  *   the cache and exporting them to userspace through the cookie
1204  *   argument to VOP_READDIR.
1205  */
1206 
1207 u_long
1208 nfs_dirhash(off)
1209 	off_t off;
1210 {
1211 	int i;
1212 	char *cp = (char *)&off;
1213 	u_long sum = 0L;
1214 
1215 	for (i = 0 ; i < sizeof (off); i++)
1216 		sum += *cp++;
1217 
1218 	return sum;
1219 }
1220 
1221 #define	_NFSDC_MTX(np)		(&NFSTOV(np)->v_interlock)
1222 #define	NFSDC_LOCK(np)		simple_lock(_NFSDC_MTX(np))
1223 #define	NFSDC_UNLOCK(np)	simple_unlock(_NFSDC_MTX(np))
1224 #define	NFSDC_ASSERT_LOCKED(np) LOCK_ASSERT(simple_lock_held(_NFSDC_MTX(np)))
1225 
1226 void
1227 nfs_initdircache(vp)
1228 	struct vnode *vp;
1229 {
1230 	struct nfsnode *np = VTONFS(vp);
1231 	struct nfsdirhashhead *dircache;
1232 
1233 	dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF,
1234 	    M_WAITOK, &nfsdirhashmask);
1235 
1236 	NFSDC_LOCK(np);
1237 	if (np->n_dircache == NULL) {
1238 		np->n_dircachesize = 0;
1239 		np->n_dircache = dircache;
1240 		dircache = NULL;
1241 		TAILQ_INIT(&np->n_dirchain);
1242 	}
1243 	NFSDC_UNLOCK(np);
1244 	if (dircache)
1245 		hashdone(dircache, M_NFSDIROFF);
1246 }
1247 
1248 void
1249 nfs_initdirxlatecookie(vp)
1250 	struct vnode *vp;
1251 {
1252 	struct nfsnode *np = VTONFS(vp);
1253 	unsigned *dirgens;
1254 
1255 	KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE);
1256 
1257 	dirgens = malloc(NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF,
1258 	    M_WAITOK|M_ZERO);
1259 	NFSDC_LOCK(np);
1260 	if (np->n_dirgens == NULL) {
1261 		np->n_dirgens = dirgens;
1262 		dirgens = NULL;
1263 	}
1264 	NFSDC_UNLOCK(np);
1265 	if (dirgens)
1266 		free(dirgens, M_NFSDIROFF);
1267 }
1268 
1269 static const struct nfsdircache dzero;
1270 
1271 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *));
1272 static void nfs_putdircache_unlocked __P((struct nfsnode *,
1273     struct nfsdircache *));
1274 
1275 static void
1276 nfs_unlinkdircache(np, ndp)
1277 	struct nfsnode *np;
1278 	struct nfsdircache *ndp;
1279 {
1280 
1281 	NFSDC_ASSERT_LOCKED(np);
1282 	KASSERT(ndp != &dzero);
1283 
1284 	if (LIST_NEXT(ndp, dc_hash) == (void *)-1)
1285 		return;
1286 
1287 	TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1288 	LIST_REMOVE(ndp, dc_hash);
1289 	LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */
1290 
1291 	nfs_putdircache_unlocked(np, ndp);
1292 }
1293 
1294 void
1295 nfs_putdircache(np, ndp)
1296 	struct nfsnode *np;
1297 	struct nfsdircache *ndp;
1298 {
1299 	int ref;
1300 
1301 	if (ndp == &dzero)
1302 		return;
1303 
1304 	KASSERT(ndp->dc_refcnt > 0);
1305 	NFSDC_LOCK(np);
1306 	ref = --ndp->dc_refcnt;
1307 	NFSDC_UNLOCK(np);
1308 
1309 	if (ref == 0)
1310 		free(ndp, M_NFSDIROFF);
1311 }
1312 
1313 static void
1314 nfs_putdircache_unlocked(np, ndp)
1315 	struct nfsnode *np;
1316 	struct nfsdircache *ndp;
1317 {
1318 	int ref;
1319 
1320 	NFSDC_ASSERT_LOCKED(np);
1321 
1322 	if (ndp == &dzero)
1323 		return;
1324 
1325 	KASSERT(ndp->dc_refcnt > 0);
1326 	ref = --ndp->dc_refcnt;
1327 	if (ref == 0)
1328 		free(ndp, M_NFSDIROFF);
1329 }
1330 
1331 struct nfsdircache *
1332 nfs_searchdircache(vp, off, do32, hashent)
1333 	struct vnode *vp;
1334 	off_t off;
1335 	int do32;
1336 	int *hashent;
1337 {
1338 	struct nfsdirhashhead *ndhp;
1339 	struct nfsdircache *ndp = NULL;
1340 	struct nfsnode *np = VTONFS(vp);
1341 	unsigned ent;
1342 
1343 	/*
1344 	 * Zero is always a valid cookie.
1345 	 */
1346 	if (off == 0)
1347 		/* XXXUNCONST */
1348 		return (struct nfsdircache *)__UNCONST(&dzero);
1349 
1350 	if (!np->n_dircache)
1351 		return NULL;
1352 
1353 	/*
1354 	 * We use a 32bit cookie as search key, directly reconstruct
1355 	 * the hashentry. Else use the hashfunction.
1356 	 */
1357 	if (do32) {
1358 		ent = (u_int32_t)off >> 24;
1359 		if (ent >= NFS_DIRHASHSIZ)
1360 			return NULL;
1361 		ndhp = &np->n_dircache[ent];
1362 	} else {
1363 		ndhp = NFSDIRHASH(np, off);
1364 	}
1365 
1366 	if (hashent)
1367 		*hashent = (int)(ndhp - np->n_dircache);
1368 
1369 	NFSDC_LOCK(np);
1370 	if (do32) {
1371 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1372 			if (ndp->dc_cookie32 == (u_int32_t)off) {
1373 				/*
1374 				 * An invalidated entry will become the
1375 				 * start of a new block fetched from
1376 				 * the server.
1377 				 */
1378 				if (ndp->dc_flags & NFSDC_INVALID) {
1379 					ndp->dc_blkcookie = ndp->dc_cookie;
1380 					ndp->dc_entry = 0;
1381 					ndp->dc_flags &= ~NFSDC_INVALID;
1382 				}
1383 				break;
1384 			}
1385 		}
1386 	} else {
1387 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1388 			if (ndp->dc_cookie == off)
1389 				break;
1390 		}
1391 	}
1392 	if (ndp != NULL)
1393 		ndp->dc_refcnt++;
1394 	NFSDC_UNLOCK(np);
1395 	return ndp;
1396 }
1397 
1398 
1399 struct nfsdircache *
1400 nfs_enterdircache(vp, off, blkoff, en, blkno)
1401 	struct vnode *vp;
1402 	off_t off, blkoff;
1403 	int en;
1404 	daddr_t blkno;
1405 {
1406 	struct nfsnode *np = VTONFS(vp);
1407 	struct nfsdirhashhead *ndhp;
1408 	struct nfsdircache *ndp = NULL;
1409 	struct nfsdircache *newndp = NULL;
1410 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1411 	int hashent, gen, overwrite;
1412 
1413 	/*
1414 	 * XXX refuse entries for offset 0. amd(8) erroneously sets
1415 	 * cookie 0 for the '.' entry, making this necessary. This
1416 	 * isn't so bad, as 0 is a special case anyway.
1417 	 */
1418 	if (off == 0)
1419 		/* XXXUNCONST */
1420 		return (struct nfsdircache *)__UNCONST(&dzero);
1421 
1422 	if (!np->n_dircache)
1423 		/*
1424 		 * XXX would like to do this in nfs_nget but vtype
1425 		 * isn't known at that time.
1426 		 */
1427 		nfs_initdircache(vp);
1428 
1429 	if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens)
1430 		nfs_initdirxlatecookie(vp);
1431 
1432 retry:
1433 	ndp = nfs_searchdircache(vp, off, 0, &hashent);
1434 
1435 	NFSDC_LOCK(np);
1436 	if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) {
1437 		/*
1438 		 * Overwriting an old entry. Check if it's the same.
1439 		 * If so, just return. If not, remove the old entry.
1440 		 */
1441 		if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
1442 			goto done;
1443 		nfs_unlinkdircache(np, ndp);
1444 		nfs_putdircache_unlocked(np, ndp);
1445 		ndp = NULL;
1446 	}
1447 
1448 	ndhp = &np->n_dircache[hashent];
1449 
1450 	if (!ndp) {
1451 		if (newndp == NULL) {
1452 			NFSDC_UNLOCK(np);
1453 			newndp = malloc(sizeof(*ndp), M_NFSDIROFF, M_WAITOK);
1454 			newndp->dc_refcnt = 1;
1455 			LIST_NEXT(newndp, dc_hash) = (void *)-1;
1456 			goto retry;
1457 		}
1458 		ndp = newndp;
1459 		newndp = NULL;
1460 		overwrite = 0;
1461 		if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1462 			/*
1463 			 * We're allocating a new entry, so bump the
1464 			 * generation number.
1465 			 */
1466 			gen = ++np->n_dirgens[hashent];
1467 			if (gen == 0) {
1468 				np->n_dirgens[hashent]++;
1469 				gen++;
1470 			}
1471 			ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
1472 		}
1473 	} else
1474 		overwrite = 1;
1475 
1476 	ndp->dc_cookie = off;
1477 	ndp->dc_blkcookie = blkoff;
1478 	ndp->dc_entry = en;
1479 	ndp->dc_flags = 0;
1480 
1481 	if (overwrite)
1482 		goto done;
1483 
1484 	/*
1485 	 * If the maximum directory cookie cache size has been reached
1486 	 * for this node, take one off the front. The idea is that
1487 	 * directories are typically read front-to-back once, so that
1488 	 * the oldest entries can be thrown away without much performance
1489 	 * loss.
1490 	 */
1491 	if (np->n_dircachesize == NFS_MAXDIRCACHE) {
1492 		nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain));
1493 	} else
1494 		np->n_dircachesize++;
1495 
1496 	KASSERT(ndp->dc_refcnt == 1);
1497 	LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
1498 	TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
1499 	ndp->dc_refcnt++;
1500 done:
1501 	KASSERT(ndp->dc_refcnt > 0);
1502 	NFSDC_UNLOCK(np);
1503 	if (newndp)
1504 		nfs_putdircache(np, newndp);
1505 	return ndp;
1506 }
1507 
1508 void
1509 nfs_invaldircache(vp, flags)
1510 	struct vnode *vp;
1511 	int flags;
1512 {
1513 	struct nfsnode *np = VTONFS(vp);
1514 	struct nfsdircache *ndp = NULL;
1515 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1516 	const boolean_t forcefree = flags & NFS_INVALDIRCACHE_FORCE;
1517 
1518 #ifdef DIAGNOSTIC
1519 	if (vp->v_type != VDIR)
1520 		panic("nfs: invaldircache: not dir");
1521 #endif
1522 
1523 	if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0)
1524 		np->n_flag &= ~NEOFVALID;
1525 
1526 	if (!np->n_dircache)
1527 		return;
1528 
1529 	NFSDC_LOCK(np);
1530 	if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
1531 		while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) {
1532 			KASSERT(!forcefree || ndp->dc_refcnt == 1);
1533 			nfs_unlinkdircache(np, ndp);
1534 		}
1535 		np->n_dircachesize = 0;
1536 		if (forcefree && np->n_dirgens) {
1537 			FREE(np->n_dirgens, M_NFSDIROFF);
1538 			np->n_dirgens = NULL;
1539 		}
1540 	} else {
1541 		TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain)
1542 			ndp->dc_flags |= NFSDC_INVALID;
1543 	}
1544 
1545 	NFSDC_UNLOCK(np);
1546 }
1547 
1548 /*
1549  * Called once before VFS init to initialize shared and
1550  * server-specific data structures.
1551  */
1552 void
1553 nfs_init()
1554 {
1555 	nfsrtt.pos = 0;
1556 	rpc_vers = txdr_unsigned(RPC_VER2);
1557 	rpc_call = txdr_unsigned(RPC_CALL);
1558 	rpc_reply = txdr_unsigned(RPC_REPLY);
1559 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1560 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1561 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1562 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1563 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1564 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1565 	nfs_prog = txdr_unsigned(NFS_PROG);
1566 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
1567 	nfs_true = txdr_unsigned(TRUE);
1568 	nfs_false = txdr_unsigned(FALSE);
1569 	nfs_xdrneg1 = txdr_unsigned(-1);
1570 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1571 	if (nfs_ticks < 1)
1572 		nfs_ticks = 1;
1573 #ifdef NFSSERVER
1574 	nfsrv_init(0);			/* Init server data structures */
1575 	nfsrv_initcache();		/* Init the server request cache */
1576 	pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript),
1577 	    0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr);
1578 #endif /* NFSSERVER */
1579 
1580 #if defined(NFSSERVER) || !defined(NFS_V2_ONLY)
1581 	/*
1582 	 * Initialize the nqnfs data structures.
1583 	 */
1584 	if (nqnfsstarttime == 0) {
1585 		nqnfsstarttime = boottime.tv_sec + nqsrv_maxlease
1586 			+ nqsrv_clockskew + nqsrv_writeslack;
1587 		NQLOADNOVRAM(nqnfsstarttime);
1588 		CIRCLEQ_INIT(&nqtimerhead);
1589 		nqfhhashtbl = hashinit(NQLCHSZ, HASH_LIST, M_NQLEASE,
1590 		    M_WAITOK, &nqfhhash);
1591 	}
1592 #endif
1593 
1594 	exithook_establish(nfs_exit, NULL);
1595 
1596 	/*
1597 	 * Initialize reply list and start timer
1598 	 */
1599 	TAILQ_INIT(&nfs_reqq);
1600 	nfs_timer(NULL);
1601 	MOWNER_ATTACH(&nfs_mowner);
1602 
1603 #ifdef NFS
1604 	/* Initialize the kqueue structures */
1605 	nfs_kqinit();
1606 	/* Initialize the iod structures */
1607 	nfs_iodinit();
1608 #endif
1609 }
1610 
1611 #ifdef NFS
1612 /*
1613  * Called once at VFS init to initialize client-specific data structures.
1614  */
1615 void
1616 nfs_vfs_init()
1617 {
1618 	nfs_nhinit();			/* Init the nfsnode table */
1619 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1620 }
1621 
1622 void
1623 nfs_vfs_reinit()
1624 {
1625 	nfs_nhreinit();
1626 }
1627 
1628 void
1629 nfs_vfs_done()
1630 {
1631 	nfs_nhdone();
1632 }
1633 
1634 /*
1635  * Attribute cache routines.
1636  * nfs_loadattrcache() - loads or updates the cache contents from attributes
1637  *	that are on the mbuf list
1638  * nfs_getattrcache() - returns valid attributes if found in cache, returns
1639  *	error otherwise
1640  */
1641 
1642 /*
1643  * Load the attribute cache (that lives in the nfsnode entry) with
1644  * the values on the mbuf list and
1645  * Iff vap not NULL
1646  *    copy the attributes to *vaper
1647  */
1648 int
1649 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags)
1650 	struct vnode **vpp;
1651 	struct mbuf **mdp;
1652 	caddr_t *dposp;
1653 	struct vattr *vaper;
1654 	int flags;
1655 {
1656 	int32_t t1;
1657 	caddr_t cp2;
1658 	int error = 0;
1659 	struct mbuf *md;
1660 	int v3 = NFS_ISV3(*vpp);
1661 
1662 	md = *mdp;
1663 	t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
1664 	error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2);
1665 	if (error)
1666 		return (error);
1667 	return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags);
1668 }
1669 
1670 int
1671 nfs_loadattrcache(vpp, fp, vaper, flags)
1672 	struct vnode **vpp;
1673 	struct nfs_fattr *fp;
1674 	struct vattr *vaper;
1675 	int flags;
1676 {
1677 	struct vnode *vp = *vpp;
1678 	struct vattr *vap;
1679 	int v3 = NFS_ISV3(vp);
1680 	enum vtype vtyp;
1681 	u_short vmode;
1682 	struct timespec mtime;
1683 	struct timespec ctime;
1684 	struct vnode *nvp;
1685 	int32_t rdev;
1686 	struct nfsnode *np;
1687 	extern int (**spec_nfsv2nodeop_p) __P((void *));
1688 	uid_t uid;
1689 	gid_t gid;
1690 
1691 	if (v3) {
1692 		vtyp = nfsv3tov_type(fp->fa_type);
1693 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1694 		rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1),
1695 			fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2));
1696 		fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
1697 		fxdr_nfsv3time(&fp->fa3_ctime, &ctime);
1698 	} else {
1699 		vtyp = nfsv2tov_type(fp->fa_type);
1700 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1701 		if (vtyp == VNON || vtyp == VREG)
1702 			vtyp = IFTOVT(vmode);
1703 		rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
1704 		fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
1705 		ctime.tv_sec = fxdr_unsigned(u_int32_t,
1706 		    fp->fa2_ctime.nfsv2_sec);
1707 		ctime.tv_nsec = 0;
1708 
1709 		/*
1710 		 * Really ugly NFSv2 kludge.
1711 		 */
1712 		if (vtyp == VCHR && rdev == 0xffffffff)
1713 			vtyp = VFIFO;
1714 	}
1715 
1716 	vmode &= ALLPERMS;
1717 
1718 	/*
1719 	 * If v_type == VNON it is a new node, so fill in the v_type,
1720 	 * n_mtime fields. Check to see if it represents a special
1721 	 * device, and if so, check for a possible alias. Once the
1722 	 * correct vnode has been obtained, fill in the rest of the
1723 	 * information.
1724 	 */
1725 	np = VTONFS(vp);
1726 	if (vp->v_type == VNON) {
1727 		vp->v_type = vtyp;
1728 		if (vp->v_type == VFIFO) {
1729 			extern int (**fifo_nfsv2nodeop_p) __P((void *));
1730 			vp->v_op = fifo_nfsv2nodeop_p;
1731 		} else if (vp->v_type == VREG) {
1732 			lockinit(&np->n_commitlock, PINOD, "nfsclock", 0, 0);
1733 		} else if (vp->v_type == VCHR || vp->v_type == VBLK) {
1734 			vp->v_op = spec_nfsv2nodeop_p;
1735 			nvp = checkalias(vp, (dev_t)rdev, vp->v_mount);
1736 			if (nvp) {
1737 				/*
1738 				 * Discard unneeded vnode, but save its nfsnode.
1739 				 * Since the nfsnode does not have a lock, its
1740 				 * vnode lock has to be carried over.
1741 				 */
1742 				/*
1743 				 * XXX is the old node sure to be locked here?
1744 				 */
1745 				KASSERT(lockstatus(&vp->v_lock) ==
1746 				    LK_EXCLUSIVE);
1747 				nvp->v_data = vp->v_data;
1748 				vp->v_data = NULL;
1749 				VOP_UNLOCK(vp, 0);
1750 				vp->v_op = spec_vnodeop_p;
1751 				vrele(vp);
1752 				vgone(vp);
1753 				lockmgr(&nvp->v_lock, LK_EXCLUSIVE,
1754 				    &nvp->v_interlock);
1755 				/*
1756 				 * Reinitialize aliased node.
1757 				 */
1758 				np->n_vnode = nvp;
1759 				*vpp = vp = nvp;
1760 			}
1761 		}
1762 		np->n_mtime = mtime;
1763 	}
1764 	uid = fxdr_unsigned(uid_t, fp->fa_uid);
1765 	gid = fxdr_unsigned(gid_t, fp->fa_gid);
1766 	vap = np->n_vattr;
1767 
1768 	/*
1769 	 * Invalidate access cache if uid, gid, mode or ctime changed.
1770 	 */
1771 	if (np->n_accstamp != -1 &&
1772 	    (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode
1773 	    || timespeccmp(&ctime, &vap->va_ctime, !=)))
1774 		np->n_accstamp = -1;
1775 
1776 	vap->va_type = vtyp;
1777 	vap->va_mode = vmode;
1778 	vap->va_rdev = (dev_t)rdev;
1779 	vap->va_mtime = mtime;
1780 	vap->va_ctime = ctime;
1781 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1782 	switch (vtyp) {
1783 	case VDIR:
1784 		vap->va_blocksize = NFS_DIRFRAGSIZ;
1785 		break;
1786 	case VBLK:
1787 		vap->va_blocksize = BLKDEV_IOSIZE;
1788 		break;
1789 	case VCHR:
1790 		vap->va_blocksize = MAXBSIZE;
1791 		break;
1792 	default:
1793 		vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize :
1794 		    fxdr_unsigned(int32_t, fp->fa2_blocksize);
1795 		break;
1796 	}
1797 	if (v3) {
1798 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1799 		vap->va_uid = uid;
1800 		vap->va_gid = gid;
1801 		vap->va_size = fxdr_hyper(&fp->fa3_size);
1802 		vap->va_bytes = fxdr_hyper(&fp->fa3_used);
1803 		vap->va_fileid = fxdr_hyper(&fp->fa3_fileid);
1804 		fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
1805 		vap->va_flags = 0;
1806 		vap->va_filerev = 0;
1807 	} else {
1808 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1809 		vap->va_uid = uid;
1810 		vap->va_gid = gid;
1811 		vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
1812 		vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks)
1813 		    * NFS_FABLKSIZE;
1814 		vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
1815 		fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
1816 		vap->va_flags = 0;
1817 		vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
1818 		vap->va_filerev = 0;
1819 	}
1820 	if (vap->va_size != np->n_size) {
1821 		if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) {
1822 			vap->va_size = np->n_size;
1823 		} else {
1824 			np->n_size = vap->va_size;
1825 			if (vap->va_type == VREG) {
1826 				/*
1827 				 * we can't free pages if NAC_NOTRUNC because
1828 				 * the pages can be owned by ourselves.
1829 				 */
1830 				if (flags & NAC_NOTRUNC) {
1831 					np->n_flag |= NTRUNCDELAYED;
1832 				} else {
1833 					simple_lock(&vp->v_interlock);
1834 					(void)VOP_PUTPAGES(vp, 0,
1835 					    0, PGO_SYNCIO | PGO_CLEANIT |
1836 					    PGO_FREE | PGO_ALLPAGES);
1837 					uvm_vnp_setsize(vp, np->n_size);
1838 				}
1839 			}
1840 		}
1841 	}
1842 	np->n_attrstamp = mono_time.tv_sec;
1843 	if (vaper != NULL) {
1844 		memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(*vap));
1845 		if (np->n_flag & NCHG) {
1846 			if (np->n_flag & NACC)
1847 				vaper->va_atime = np->n_atim;
1848 			if (np->n_flag & NUPD)
1849 				vaper->va_mtime = np->n_mtim;
1850 		}
1851 	}
1852 	return (0);
1853 }
1854 
1855 /*
1856  * Check the time stamp
1857  * If the cache is valid, copy contents to *vap and return 0
1858  * otherwise return an error
1859  */
1860 int
1861 nfs_getattrcache(vp, vaper)
1862 	struct vnode *vp;
1863 	struct vattr *vaper;
1864 {
1865 	struct nfsnode *np = VTONFS(vp);
1866 	struct vattr *vap;
1867 
1868 	if (np->n_attrstamp == 0 ||
1869 	    (mono_time.tv_sec - np->n_attrstamp) >= NFS_ATTRTIMEO(np)) {
1870 		nfsstats.attrcache_misses++;
1871 		return (ENOENT);
1872 	}
1873 	nfsstats.attrcache_hits++;
1874 	vap = np->n_vattr;
1875 	if (vap->va_size != np->n_size) {
1876 		if (vap->va_type == VREG) {
1877 			if (np->n_flag & NMODIFIED) {
1878 				if (vap->va_size < np->n_size)
1879 					vap->va_size = np->n_size;
1880 				else
1881 					np->n_size = vap->va_size;
1882 			} else
1883 				np->n_size = vap->va_size;
1884 			uvm_vnp_setsize(vp, np->n_size);
1885 		} else
1886 			np->n_size = vap->va_size;
1887 	}
1888 	memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(struct vattr));
1889 	if (np->n_flag & NCHG) {
1890 		if (np->n_flag & NACC)
1891 			vaper->va_atime = np->n_atim;
1892 		if (np->n_flag & NUPD)
1893 			vaper->va_mtime = np->n_mtim;
1894 	}
1895 	return (0);
1896 }
1897 
1898 void
1899 nfs_delayedtruncate(vp)
1900 	struct vnode *vp;
1901 {
1902 	struct nfsnode *np = VTONFS(vp);
1903 
1904 	if (np->n_flag & NTRUNCDELAYED) {
1905 		np->n_flag &= ~NTRUNCDELAYED;
1906 		simple_lock(&vp->v_interlock);
1907 		(void)VOP_PUTPAGES(vp, 0,
1908 		    0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES);
1909 		uvm_vnp_setsize(vp, np->n_size);
1910 	}
1911 }
1912 
1913 #define	NFS_WCCKLUDGE_TIMEOUT	(24 * 60 * 60)	/* 1 day */
1914 #define	NFS_WCCKLUDGE(nmp, now) \
1915 	(((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \
1916 	((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0)
1917 
1918 /*
1919  * nfs_check_wccdata: check inaccurate wcc_data
1920  *
1921  * => return non-zero if we shouldn't trust the wcc_data.
1922  * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed".
1923  */
1924 
1925 int
1926 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime,
1927     struct timespec *mtime, boolean_t docheck)
1928 {
1929 	int error = 0;
1930 
1931 #if !defined(NFS_V2_ONLY)
1932 
1933 	if (docheck) {
1934 		struct vnode *vp = NFSTOV(np);
1935 		struct nfsmount *nmp;
1936 		long now = mono_time.tv_sec;
1937 #if defined(DEBUG)
1938 		const char *reason = NULL; /* XXX: gcc */
1939 #endif
1940 
1941 		if (timespeccmp(&np->n_vattr->va_mtime, mtime, <=)) {
1942 #if defined(DEBUG)
1943 			reason = "mtime";
1944 #endif
1945 			error = EINVAL;
1946 		}
1947 
1948 		if (vp->v_type == VDIR &&
1949 		    timespeccmp(&np->n_vattr->va_ctime, ctime, <=)) {
1950 #if defined(DEBUG)
1951 			reason = "ctime";
1952 #endif
1953 			error = EINVAL;
1954 		}
1955 
1956 		nmp = VFSTONFS(vp->v_mount);
1957 		if (error) {
1958 
1959 			/*
1960 			 * despite of the fact that we've updated the file,
1961 			 * timestamps of the file were not updated as we
1962 			 * expected.
1963 			 * it means that the server has incompatible
1964 			 * semantics of timestamps or (more likely)
1965 			 * the server time is not precise enough to
1966 			 * track each modifications.
1967 			 * in that case, we disable wcc processing.
1968 			 *
1969 			 * yes, strictly speaking, we should disable all
1970 			 * caching.  it's a compromise.
1971 			 */
1972 
1973 			simple_lock(&nmp->nm_slock);
1974 #if defined(DEBUG)
1975 			if (!NFS_WCCKLUDGE(nmp, now)) {
1976 				printf("%s: inaccurate wcc data (%s) detected,"
1977 				    " disabling wcc\n",
1978 				    vp->v_mount->mnt_stat.f_mntfromname,
1979 				    reason);
1980 			}
1981 #endif
1982 			nmp->nm_iflag |= NFSMNT_WCCKLUDGE;
1983 			nmp->nm_wcckludgetime = now;
1984 			simple_unlock(&nmp->nm_slock);
1985 		} else if (NFS_WCCKLUDGE(nmp, now)) {
1986 			error = EPERM; /* XXX */
1987 		} else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
1988 			simple_lock(&nmp->nm_slock);
1989 			if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
1990 #if defined(DEBUG)
1991 				printf("%s: re-enabling wcc\n",
1992 				    vp->v_mount->mnt_stat.f_mntfromname);
1993 #endif
1994 				nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE;
1995 			}
1996 			simple_unlock(&nmp->nm_slock);
1997 		}
1998 	}
1999 
2000 #endif /* !defined(NFS_V2_ONLY) */
2001 
2002 	return error;
2003 }
2004 
2005 /*
2006  * Heuristic to see if the server XDR encodes directory cookies or not.
2007  * it is not supposed to, but a lot of servers may do this. Also, since
2008  * most/all servers will implement V2 as well, it is expected that they
2009  * may return just 32 bits worth of cookie information, so we need to
2010  * find out in which 32 bits this information is available. We do this
2011  * to avoid trouble with emulated binaries that can't handle 64 bit
2012  * directory offsets.
2013  */
2014 
2015 void
2016 nfs_cookieheuristic(vp, flagp, p, cred)
2017 	struct vnode *vp;
2018 	int *flagp;
2019 	struct proc *p;
2020 	struct ucred *cred;
2021 {
2022 	struct uio auio;
2023 	struct iovec aiov;
2024 	caddr_t tbuf, cp;
2025 	struct dirent *dp;
2026 	off_t *cookies = NULL, *cop;
2027 	int error, eof, nc, len;
2028 
2029 	MALLOC(tbuf, caddr_t, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK);
2030 
2031 	aiov.iov_base = tbuf;
2032 	aiov.iov_len = NFS_DIRFRAGSIZ;
2033 	auio.uio_iov = &aiov;
2034 	auio.uio_iovcnt = 1;
2035 	auio.uio_rw = UIO_READ;
2036 	auio.uio_segflg = UIO_SYSSPACE;
2037 	auio.uio_procp = NULL;
2038 	auio.uio_resid = NFS_DIRFRAGSIZ;
2039 	auio.uio_offset = 0;
2040 
2041 	error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc);
2042 
2043 	len = NFS_DIRFRAGSIZ - auio.uio_resid;
2044 	if (error || len == 0) {
2045 		FREE(tbuf, M_TEMP);
2046 		if (cookies)
2047 			free(cookies, M_TEMP);
2048 		return;
2049 	}
2050 
2051 	/*
2052 	 * Find the first valid entry and look at its offset cookie.
2053 	 */
2054 
2055 	cp = tbuf;
2056 	for (cop = cookies; len > 0; len -= dp->d_reclen) {
2057 		dp = (struct dirent *)cp;
2058 		if (dp->d_fileno != 0 && len >= dp->d_reclen) {
2059 			if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) {
2060 				*flagp |= NFSMNT_SWAPCOOKIE;
2061 				nfs_invaldircache(vp, 0);
2062 				nfs_vinvalbuf(vp, 0, cred, p, 1);
2063 			}
2064 			break;
2065 		}
2066 		cop++;
2067 		cp += dp->d_reclen;
2068 	}
2069 
2070 	FREE(tbuf, M_TEMP);
2071 	free(cookies, M_TEMP);
2072 }
2073 #endif /* NFS */
2074 
2075 /*
2076  * Set up nameidata for a lookup() call and do it.
2077  *
2078  * If pubflag is set, this call is done for a lookup operation on the
2079  * public filehandle. In that case we allow crossing mountpoints and
2080  * absolute pathnames. However, the caller is expected to check that
2081  * the lookup result is within the public fs, and deny access if
2082  * it is not.
2083  */
2084 int
2085 nfs_namei(ndp, fhp, len, slp, nam, mdp, dposp, retdirp, p, kerbflag, pubflag)
2086 	struct nameidata *ndp;
2087 	fhandle_t *fhp;
2088 	uint32_t len;
2089 	struct nfssvc_sock *slp;
2090 	struct mbuf *nam;
2091 	struct mbuf **mdp;
2092 	caddr_t *dposp;
2093 	struct vnode **retdirp;
2094 	struct proc *p;
2095 	int kerbflag, pubflag;
2096 {
2097 	int i, rem;
2098 	struct mbuf *md;
2099 	char *fromcp, *tocp, *cp;
2100 	struct iovec aiov;
2101 	struct uio auio;
2102 	struct vnode *dp;
2103 	int error, rdonly, linklen;
2104 	struct componentname *cnp = &ndp->ni_cnd;
2105 
2106 	*retdirp = (struct vnode *)0;
2107 
2108 	if ((len + 1) > MAXPATHLEN)
2109 		return (ENAMETOOLONG);
2110 	if (len == 0)
2111 		return (EACCES);
2112 	cnp->cn_pnbuf = PNBUF_GET();
2113 
2114 	/*
2115 	 * Copy the name from the mbuf list to ndp->ni_pnbuf
2116 	 * and set the various ndp fields appropriately.
2117 	 */
2118 	fromcp = *dposp;
2119 	tocp = cnp->cn_pnbuf;
2120 	md = *mdp;
2121 	rem = mtod(md, caddr_t) + md->m_len - fromcp;
2122 	for (i = 0; i < len; i++) {
2123 		while (rem == 0) {
2124 			md = md->m_next;
2125 			if (md == NULL) {
2126 				error = EBADRPC;
2127 				goto out;
2128 			}
2129 			fromcp = mtod(md, caddr_t);
2130 			rem = md->m_len;
2131 		}
2132 		if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
2133 			error = EACCES;
2134 			goto out;
2135 		}
2136 		*tocp++ = *fromcp++;
2137 		rem--;
2138 	}
2139 	*tocp = '\0';
2140 	*mdp = md;
2141 	*dposp = fromcp;
2142 	len = nfsm_rndup(len)-len;
2143 	if (len > 0) {
2144 		if (rem >= len)
2145 			*dposp += len;
2146 		else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
2147 			goto out;
2148 	}
2149 
2150 	/*
2151 	 * Extract and set starting directory.
2152 	 */
2153 	error = nfsrv_fhtovp(fhp, FALSE, &dp, ndp->ni_cnd.cn_cred, slp,
2154 	    nam, &rdonly, kerbflag, pubflag);
2155 	if (error)
2156 		goto out;
2157 	if (dp->v_type != VDIR) {
2158 		vrele(dp);
2159 		error = ENOTDIR;
2160 		goto out;
2161 	}
2162 
2163 	if (rdonly)
2164 		cnp->cn_flags |= RDONLY;
2165 
2166 	*retdirp = dp;
2167 
2168 	if (pubflag) {
2169 		/*
2170 		 * Oh joy. For WebNFS, handle those pesky '%' escapes,
2171 		 * and the 'native path' indicator.
2172 		 */
2173 		cp = PNBUF_GET();
2174 		fromcp = cnp->cn_pnbuf;
2175 		tocp = cp;
2176 		if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
2177 			switch ((unsigned char)*fromcp) {
2178 			case WEBNFS_NATIVE_CHAR:
2179 				/*
2180 				 * 'Native' path for us is the same
2181 				 * as a path according to the NFS spec,
2182 				 * just skip the escape char.
2183 				 */
2184 				fromcp++;
2185 				break;
2186 			/*
2187 			 * More may be added in the future, range 0x80-0xff
2188 			 */
2189 			default:
2190 				error = EIO;
2191 				PNBUF_PUT(cp);
2192 				goto out;
2193 			}
2194 		}
2195 		/*
2196 		 * Translate the '%' escapes, URL-style.
2197 		 */
2198 		while (*fromcp != '\0') {
2199 			if (*fromcp == WEBNFS_ESC_CHAR) {
2200 				if (fromcp[1] != '\0' && fromcp[2] != '\0') {
2201 					fromcp++;
2202 					*tocp++ = HEXSTRTOI(fromcp);
2203 					fromcp += 2;
2204 					continue;
2205 				} else {
2206 					error = ENOENT;
2207 					PNBUF_PUT(cp);
2208 					goto out;
2209 				}
2210 			} else
2211 				*tocp++ = *fromcp++;
2212 		}
2213 		*tocp = '\0';
2214 		PNBUF_PUT(cnp->cn_pnbuf);
2215 		cnp->cn_pnbuf = cp;
2216 	}
2217 
2218 	ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1;
2219 	ndp->ni_segflg = UIO_SYSSPACE;
2220 	ndp->ni_rootdir = rootvnode;
2221 
2222 	if (pubflag) {
2223 		ndp->ni_loopcnt = 0;
2224 		if (cnp->cn_pnbuf[0] == '/')
2225 			dp = rootvnode;
2226 	} else {
2227 		cnp->cn_flags |= NOCROSSMOUNT;
2228 	}
2229 
2230 	cnp->cn_proc = p;
2231 	VREF(dp);
2232 
2233     for (;;) {
2234 	cnp->cn_nameptr = cnp->cn_pnbuf;
2235 	ndp->ni_startdir = dp;
2236 	/*
2237 	 * And call lookup() to do the real work
2238 	 */
2239 	error = lookup(ndp);
2240 	if (error) {
2241 		PNBUF_PUT(cnp->cn_pnbuf);
2242 		return (error);
2243 	}
2244 	/*
2245 	 * Check for encountering a symbolic link
2246 	 */
2247 	if ((cnp->cn_flags & ISSYMLINK) == 0) {
2248 		if (cnp->cn_flags & (SAVENAME | SAVESTART))
2249 			cnp->cn_flags |= HASBUF;
2250 		else
2251 			PNBUF_PUT(cnp->cn_pnbuf);
2252 		return (0);
2253 	} else {
2254 		if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
2255 			VOP_UNLOCK(ndp->ni_dvp, 0);
2256 		if (!pubflag) {
2257 			error = EINVAL;
2258 			break;
2259 		}
2260 
2261 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
2262 			error = ELOOP;
2263 			break;
2264 		}
2265 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
2266 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred,
2267 			    cnp->cn_proc);
2268 			if (error != 0)
2269 				break;
2270 		}
2271 		if (ndp->ni_pathlen > 1)
2272 			cp = PNBUF_GET();
2273 		else
2274 			cp = cnp->cn_pnbuf;
2275 		aiov.iov_base = cp;
2276 		aiov.iov_len = MAXPATHLEN;
2277 		auio.uio_iov = &aiov;
2278 		auio.uio_iovcnt = 1;
2279 		auio.uio_offset = 0;
2280 		auio.uio_rw = UIO_READ;
2281 		auio.uio_segflg = UIO_SYSSPACE;
2282 		auio.uio_procp = NULL;
2283 		auio.uio_resid = MAXPATHLEN;
2284 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
2285 		if (error) {
2286 		badlink:
2287 			if (ndp->ni_pathlen > 1)
2288 				PNBUF_PUT(cp);
2289 			break;
2290 		}
2291 		linklen = MAXPATHLEN - auio.uio_resid;
2292 		if (linklen == 0) {
2293 			error = ENOENT;
2294 			goto badlink;
2295 		}
2296 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
2297 			error = ENAMETOOLONG;
2298 			goto badlink;
2299 		}
2300 		if (ndp->ni_pathlen > 1) {
2301 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
2302 			PNBUF_PUT(cnp->cn_pnbuf);
2303 			cnp->cn_pnbuf = cp;
2304 		} else
2305 			cnp->cn_pnbuf[linklen] = '\0';
2306 		ndp->ni_pathlen += linklen;
2307 		vput(ndp->ni_vp);
2308 		dp = ndp->ni_dvp;
2309 		/*
2310 		 * Check if root directory should replace current directory.
2311 		 */
2312 		if (cnp->cn_pnbuf[0] == '/') {
2313 			vrele(dp);
2314 			dp = ndp->ni_rootdir;
2315 			VREF(dp);
2316 		}
2317 	}
2318    }
2319 	vrele(ndp->ni_dvp);
2320 	vput(ndp->ni_vp);
2321 	ndp->ni_vp = NULL;
2322 out:
2323 	PNBUF_PUT(cnp->cn_pnbuf);
2324 	return (error);
2325 }
2326 
2327 /*
2328  * A fiddled version of m_adj() that ensures null fill to a 32-bit
2329  * boundary and only trims off the back end
2330  *
2331  * 1. trim off 'len' bytes as m_adj(mp, -len).
2332  * 2. add zero-padding 'nul' bytes at the end of the mbuf chain.
2333  */
2334 void
2335 nfs_zeropad(mp, len, nul)
2336 	struct mbuf *mp;
2337 	int len;
2338 	int nul;
2339 {
2340 	struct mbuf *m;
2341 	int count;
2342 
2343 	/*
2344 	 * Trim from tail.  Scan the mbuf chain,
2345 	 * calculating its length and finding the last mbuf.
2346 	 * If the adjustment only affects this mbuf, then just
2347 	 * adjust and return.  Otherwise, rescan and truncate
2348 	 * after the remaining size.
2349 	 */
2350 	count = 0;
2351 	m = mp;
2352 	for (;;) {
2353 		count += m->m_len;
2354 		if (m->m_next == NULL)
2355 			break;
2356 		m = m->m_next;
2357 	}
2358 
2359 	KDASSERT(count >= len);
2360 
2361 	if (m->m_len >= len) {
2362 		m->m_len -= len;
2363 	} else {
2364 		count -= len;
2365 		/*
2366 		 * Correct length for chain is "count".
2367 		 * Find the mbuf with last data, adjust its length,
2368 		 * and toss data from remaining mbufs on chain.
2369 		 */
2370 		for (m = mp; m; m = m->m_next) {
2371 			if (m->m_len >= count) {
2372 				m->m_len = count;
2373 				break;
2374 			}
2375 			count -= m->m_len;
2376 		}
2377 		m_freem(m->m_next);
2378 		m->m_next = NULL;
2379 	}
2380 
2381 	KDASSERT(m->m_next == NULL);
2382 
2383 	/*
2384 	 * zero-padding.
2385 	 */
2386 	if (nul > 0) {
2387 		char *cp;
2388 		int i;
2389 
2390 		if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) {
2391 			struct mbuf *n;
2392 
2393 			KDASSERT(MLEN >= nul);
2394 			n = m_get(M_WAIT, MT_DATA);
2395 			MCLAIM(n, &nfs_mowner);
2396 			n->m_len = nul;
2397 			n->m_next = NULL;
2398 			m->m_next = n;
2399 			cp = mtod(n, caddr_t);
2400 		} else {
2401 			cp = mtod(m, caddr_t) + m->m_len;
2402 			m->m_len += nul;
2403 		}
2404 		for (i = 0; i < nul; i++)
2405 			*cp++ = '\0';
2406 	}
2407 	return;
2408 }
2409 
2410 /*
2411  * Make these functions instead of macros, so that the kernel text size
2412  * doesn't get too big...
2413  */
2414 void
2415 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp)
2416 	struct nfsrv_descript *nfsd;
2417 	int before_ret;
2418 	struct vattr *before_vap;
2419 	int after_ret;
2420 	struct vattr *after_vap;
2421 	struct mbuf **mbp;
2422 	char **bposp;
2423 {
2424 	struct mbuf *mb = *mbp;
2425 	char *bpos = *bposp;
2426 	u_int32_t *tl;
2427 
2428 	if (before_ret) {
2429 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2430 		*tl = nfs_false;
2431 	} else {
2432 		nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
2433 		*tl++ = nfs_true;
2434 		txdr_hyper(before_vap->va_size, tl);
2435 		tl += 2;
2436 		txdr_nfsv3time(&(before_vap->va_mtime), tl);
2437 		tl += 2;
2438 		txdr_nfsv3time(&(before_vap->va_ctime), tl);
2439 	}
2440 	*bposp = bpos;
2441 	*mbp = mb;
2442 	nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
2443 }
2444 
2445 void
2446 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp)
2447 	struct nfsrv_descript *nfsd;
2448 	int after_ret;
2449 	struct vattr *after_vap;
2450 	struct mbuf **mbp;
2451 	char **bposp;
2452 {
2453 	struct mbuf *mb = *mbp;
2454 	char *bpos = *bposp;
2455 	u_int32_t *tl;
2456 	struct nfs_fattr *fp;
2457 
2458 	if (after_ret) {
2459 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2460 		*tl = nfs_false;
2461 	} else {
2462 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
2463 		*tl++ = nfs_true;
2464 		fp = (struct nfs_fattr *)tl;
2465 		nfsm_srvfattr(nfsd, after_vap, fp);
2466 	}
2467 	*mbp = mb;
2468 	*bposp = bpos;
2469 }
2470 
2471 void
2472 nfsm_srvfattr(nfsd, vap, fp)
2473 	struct nfsrv_descript *nfsd;
2474 	struct vattr *vap;
2475 	struct nfs_fattr *fp;
2476 {
2477 
2478 	fp->fa_nlink = txdr_unsigned(vap->va_nlink);
2479 	fp->fa_uid = txdr_unsigned(vap->va_uid);
2480 	fp->fa_gid = txdr_unsigned(vap->va_gid);
2481 	if (nfsd->nd_flag & ND_NFSV3) {
2482 		fp->fa_type = vtonfsv3_type(vap->va_type);
2483 		fp->fa_mode = vtonfsv3_mode(vap->va_mode);
2484 		txdr_hyper(vap->va_size, &fp->fa3_size);
2485 		txdr_hyper(vap->va_bytes, &fp->fa3_used);
2486 		fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
2487 		fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
2488 		fp->fa3_fsid.nfsuquad[0] = 0;
2489 		fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
2490 		txdr_hyper(vap->va_fileid, &fp->fa3_fileid);
2491 		txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
2492 		txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
2493 		txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
2494 	} else {
2495 		fp->fa_type = vtonfsv2_type(vap->va_type);
2496 		fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
2497 		fp->fa2_size = txdr_unsigned(vap->va_size);
2498 		fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
2499 		if (vap->va_type == VFIFO)
2500 			fp->fa2_rdev = 0xffffffff;
2501 		else
2502 			fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
2503 		fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
2504 		fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
2505 		fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
2506 		txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
2507 		txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
2508 		txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
2509 	}
2510 }
2511 
2512 /*
2513  * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
2514  * 	- look up fsid in mount list (if not found ret error)
2515  *	- get vp and export rights by calling VFS_FHTOVP()
2516  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2517  *	- if not lockflag unlock it with VOP_UNLOCK()
2518  */
2519 int
2520 nfsrv_fhtovp(fhp, lockflag, vpp, cred, slp, nam, rdonlyp, kerbflag, pubflag)
2521 	fhandle_t *fhp;
2522 	int lockflag;
2523 	struct vnode **vpp;
2524 	struct ucred *cred;
2525 	struct nfssvc_sock *slp;
2526 	struct mbuf *nam;
2527 	int *rdonlyp;
2528 	int kerbflag;
2529 {
2530 	struct mount *mp;
2531 	int i;
2532 	struct ucred *credanon;
2533 	int error, exflags;
2534 	struct sockaddr_in *saddr;
2535 
2536 	*vpp = (struct vnode *)0;
2537 
2538 	if (nfs_ispublicfh(fhp)) {
2539 		if (!pubflag || !nfs_pub.np_valid)
2540 			return (ESTALE);
2541 		fhp = &nfs_pub.np_handle;
2542 	}
2543 
2544 	mp = vfs_getvfs(&fhp->fh_fsid);
2545 	if (!mp)
2546 		return (ESTALE);
2547 	error = VFS_CHECKEXP(mp, nam, &exflags, &credanon);
2548 	if (error)
2549 		return (error);
2550 	error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
2551 	if (error)
2552 		return (error);
2553 
2554 	if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
2555 		saddr = mtod(nam, struct sockaddr_in *);
2556 		if ((saddr->sin_family == AF_INET) &&
2557 		    ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
2558 			vput(*vpp);
2559 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2560 		}
2561 #ifdef INET6
2562 		if ((saddr->sin_family == AF_INET6) &&
2563 		    ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) {
2564 			vput(*vpp);
2565 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2566 		}
2567 #endif
2568 	}
2569 	/*
2570 	 * Check/setup credentials.
2571 	 */
2572 	if (exflags & MNT_EXKERB) {
2573 		if (!kerbflag) {
2574 			vput(*vpp);
2575 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2576 		}
2577 	} else if (kerbflag) {
2578 		vput(*vpp);
2579 		return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2580 	} else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
2581 		cred->cr_uid = credanon->cr_uid;
2582 		cred->cr_gid = credanon->cr_gid;
2583 		for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
2584 			cred->cr_groups[i] = credanon->cr_groups[i];
2585 		cred->cr_ngroups = i;
2586 	}
2587 	if (exflags & MNT_EXRDONLY)
2588 		*rdonlyp = 1;
2589 	else
2590 		*rdonlyp = 0;
2591 	if (!lockflag)
2592 		VOP_UNLOCK(*vpp, 0);
2593 	return (0);
2594 }
2595 
2596 /*
2597  * WebNFS: check if a filehandle is a public filehandle. For v3, this
2598  * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
2599  * transformed this to all zeroes in both cases, so check for it.
2600  */
2601 int
2602 nfs_ispublicfh(fhp)
2603 	fhandle_t *fhp;
2604 {
2605 	char *cp = (char *)fhp;
2606 	int i;
2607 
2608 	for (i = 0; i < NFSX_V3FH; i++)
2609 		if (*cp++ != 0)
2610 			return (FALSE);
2611 	return (TRUE);
2612 }
2613 
2614 /*
2615  * This function compares two net addresses by family and returns TRUE
2616  * if they are the same host.
2617  * If there is any doubt, return FALSE.
2618  * The AF_INET family is handled as a special case so that address mbufs
2619  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
2620  */
2621 int
2622 netaddr_match(family, haddr, nam)
2623 	int family;
2624 	union nethostaddr *haddr;
2625 	struct mbuf *nam;
2626 {
2627 	struct sockaddr_in *inetaddr;
2628 
2629 	switch (family) {
2630 	case AF_INET:
2631 		inetaddr = mtod(nam, struct sockaddr_in *);
2632 		if (inetaddr->sin_family == AF_INET &&
2633 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
2634 			return (1);
2635 		break;
2636 #ifdef INET6
2637 	case AF_INET6:
2638 	    {
2639 		struct sockaddr_in6 *sin6_1, *sin6_2;
2640 
2641 		sin6_1 = mtod(nam, struct sockaddr_in6 *);
2642 		sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
2643 		if (sin6_1->sin6_family == AF_INET6 &&
2644 		    IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
2645 			return 1;
2646 	    }
2647 #endif
2648 #ifdef ISO
2649 	case AF_ISO:
2650 	    {
2651 		struct sockaddr_iso *isoaddr1, *isoaddr2;
2652 
2653 		isoaddr1 = mtod(nam, struct sockaddr_iso *);
2654 		isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *);
2655 		if (isoaddr1->siso_family == AF_ISO &&
2656 		    isoaddr1->siso_nlen > 0 &&
2657 		    isoaddr1->siso_nlen == isoaddr2->siso_nlen &&
2658 		    SAME_ISOADDR(isoaddr1, isoaddr2))
2659 			return (1);
2660 		break;
2661 	    }
2662 #endif	/* ISO */
2663 	default:
2664 		break;
2665 	};
2666 	return (0);
2667 }
2668 
2669 /*
2670  * The write verifier has changed (probably due to a server reboot), so all
2671  * PG_NEEDCOMMIT pages will have to be written again. Since they are marked
2672  * as dirty or are being written out just now, all this takes is clearing
2673  * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for
2674  * the mount point.
2675  */
2676 void
2677 nfs_clearcommit(mp)
2678 	struct mount *mp;
2679 {
2680 	struct vnode *vp;
2681 	struct nfsnode *np;
2682 	struct vm_page *pg;
2683 	struct nfsmount *nmp = VFSTONFS(mp);
2684 
2685 	lockmgr(&nmp->nm_writeverflock, LK_EXCLUSIVE, NULL);
2686 
2687 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2688 		KASSERT(vp->v_mount == mp);
2689 		if (vp->v_type != VREG)
2690 			continue;
2691 		np = VTONFS(vp);
2692 		np->n_pushlo = np->n_pushhi = np->n_pushedlo =
2693 		    np->n_pushedhi = 0;
2694 		np->n_commitflags &=
2695 		    ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
2696 		simple_lock(&vp->v_uobj.vmobjlock);
2697 		TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) {
2698 			pg->flags &= ~PG_NEEDCOMMIT;
2699 		}
2700 		simple_unlock(&vp->v_uobj.vmobjlock);
2701 	}
2702 	simple_lock(&nmp->nm_slock);
2703 	nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF;
2704 	simple_unlock(&nmp->nm_slock);
2705 	lockmgr(&nmp->nm_writeverflock, LK_RELEASE, NULL);
2706 }
2707 
2708 void
2709 nfs_merge_commit_ranges(vp)
2710 	struct vnode *vp;
2711 {
2712 	struct nfsnode *np = VTONFS(vp);
2713 
2714 	KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID);
2715 
2716 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2717 		np->n_pushedlo = np->n_pushlo;
2718 		np->n_pushedhi = np->n_pushhi;
2719 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2720 	} else {
2721 		if (np->n_pushlo < np->n_pushedlo)
2722 			np->n_pushedlo = np->n_pushlo;
2723 		if (np->n_pushhi > np->n_pushedhi)
2724 			np->n_pushedhi = np->n_pushhi;
2725 	}
2726 
2727 	np->n_pushlo = np->n_pushhi = 0;
2728 	np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
2729 
2730 #ifdef NFS_DEBUG_COMMIT
2731 	printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2732 	    (unsigned)np->n_pushedhi);
2733 #endif
2734 }
2735 
2736 int
2737 nfs_in_committed_range(vp, off, len)
2738 	struct vnode *vp;
2739 	off_t off, len;
2740 {
2741 	struct nfsnode *np = VTONFS(vp);
2742 	off_t lo, hi;
2743 
2744 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2745 		return 0;
2746 	lo = off;
2747 	hi = lo + len;
2748 
2749 	return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
2750 }
2751 
2752 int
2753 nfs_in_tobecommitted_range(vp, off, len)
2754 	struct vnode *vp;
2755 	off_t off, len;
2756 {
2757 	struct nfsnode *np = VTONFS(vp);
2758 	off_t lo, hi;
2759 
2760 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2761 		return 0;
2762 	lo = off;
2763 	hi = lo + len;
2764 
2765 	return (lo >= np->n_pushlo && hi <= np->n_pushhi);
2766 }
2767 
2768 void
2769 nfs_add_committed_range(vp, off, len)
2770 	struct vnode *vp;
2771 	off_t off, len;
2772 {
2773 	struct nfsnode *np = VTONFS(vp);
2774 	off_t lo, hi;
2775 
2776 	lo = off;
2777 	hi = lo + len;
2778 
2779 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2780 		np->n_pushedlo = lo;
2781 		np->n_pushedhi = hi;
2782 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2783 	} else {
2784 		if (hi > np->n_pushedhi)
2785 			np->n_pushedhi = hi;
2786 		if (lo < np->n_pushedlo)
2787 			np->n_pushedlo = lo;
2788 	}
2789 #ifdef NFS_DEBUG_COMMIT
2790 	printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2791 	    (unsigned)np->n_pushedhi);
2792 #endif
2793 }
2794 
2795 void
2796 nfs_del_committed_range(vp, off, len)
2797 	struct vnode *vp;
2798 	off_t off, len;
2799 {
2800 	struct nfsnode *np = VTONFS(vp);
2801 	off_t lo, hi;
2802 
2803 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2804 		return;
2805 
2806 	lo = off;
2807 	hi = lo + len;
2808 
2809 	if (lo > np->n_pushedhi || hi < np->n_pushedlo)
2810 		return;
2811 	if (lo <= np->n_pushedlo)
2812 		np->n_pushedlo = hi;
2813 	else if (hi >= np->n_pushedhi)
2814 		np->n_pushedhi = lo;
2815 	else {
2816 		/*
2817 		 * XXX There's only one range. If the deleted range
2818 		 * is in the middle, pick the largest of the
2819 		 * contiguous ranges that it leaves.
2820 		 */
2821 		if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
2822 			np->n_pushedhi = lo;
2823 		else
2824 			np->n_pushedlo = hi;
2825 	}
2826 #ifdef NFS_DEBUG_COMMIT
2827 	printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2828 	    (unsigned)np->n_pushedhi);
2829 #endif
2830 }
2831 
2832 void
2833 nfs_add_tobecommitted_range(vp, off, len)
2834 	struct vnode *vp;
2835 	off_t off, len;
2836 {
2837 	struct nfsnode *np = VTONFS(vp);
2838 	off_t lo, hi;
2839 
2840 	lo = off;
2841 	hi = lo + len;
2842 
2843 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
2844 		np->n_pushlo = lo;
2845 		np->n_pushhi = hi;
2846 		np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
2847 	} else {
2848 		if (lo < np->n_pushlo)
2849 			np->n_pushlo = lo;
2850 		if (hi > np->n_pushhi)
2851 			np->n_pushhi = hi;
2852 	}
2853 #ifdef NFS_DEBUG_COMMIT
2854 	printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2855 	    (unsigned)np->n_pushhi);
2856 #endif
2857 }
2858 
2859 void
2860 nfs_del_tobecommitted_range(vp, off, len)
2861 	struct vnode *vp;
2862 	off_t off, len;
2863 {
2864 	struct nfsnode *np = VTONFS(vp);
2865 	off_t lo, hi;
2866 
2867 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2868 		return;
2869 
2870 	lo = off;
2871 	hi = lo + len;
2872 
2873 	if (lo > np->n_pushhi || hi < np->n_pushlo)
2874 		return;
2875 
2876 	if (lo <= np->n_pushlo)
2877 		np->n_pushlo = hi;
2878 	else if (hi >= np->n_pushhi)
2879 		np->n_pushhi = lo;
2880 	else {
2881 		/*
2882 		 * XXX There's only one range. If the deleted range
2883 		 * is in the middle, pick the largest of the
2884 		 * contiguous ranges that it leaves.
2885 		 */
2886 		if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
2887 			np->n_pushhi = lo;
2888 		else
2889 			np->n_pushlo = hi;
2890 	}
2891 #ifdef NFS_DEBUG_COMMIT
2892 	printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2893 	    (unsigned)np->n_pushhi);
2894 #endif
2895 }
2896 
2897 /*
2898  * Map errnos to NFS error numbers. For Version 3 also filter out error
2899  * numbers not specified for the associated procedure.
2900  */
2901 int
2902 nfsrv_errmap(nd, err)
2903 	struct nfsrv_descript *nd;
2904 	int err;
2905 {
2906 	const short *defaulterrp, *errp;
2907 
2908 	if (nd->nd_flag & ND_NFSV3) {
2909 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
2910 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2911 		while (*++errp) {
2912 			if (*errp == err)
2913 				return (err);
2914 			else if (*errp > err)
2915 				break;
2916 		}
2917 		return ((int)*defaulterrp);
2918 	    } else
2919 		return (err & 0xffff);
2920 	}
2921 	if (err <= ELAST)
2922 		return ((int)nfsrv_v2errmap[err - 1]);
2923 	return (NFSERR_IO);
2924 }
2925 
2926 /*
2927  * Sort the group list in increasing numerical order.
2928  * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
2929  *  that used to be here.)
2930  */
2931 void
2932 nfsrvw_sort(list, num)
2933         gid_t *list;
2934         int num;
2935 {
2936 	int i, j;
2937 	gid_t v;
2938 
2939 	/* Insertion sort. */
2940 	for (i = 1; i < num; i++) {
2941 		v = list[i];
2942 		/* find correct slot for value v, moving others up */
2943 		for (j = i; --j >= 0 && v < list[j];)
2944 			list[j + 1] = list[j];
2945 		list[j + 1] = v;
2946 	}
2947 }
2948 
2949 /*
2950  * copy credentials making sure that the result can be compared with memcmp().
2951  */
2952 void
2953 nfsrv_setcred(incred, outcred)
2954 	struct ucred *incred, *outcred;
2955 {
2956 	int i;
2957 
2958 	memset((caddr_t)outcred, 0, sizeof (struct ucred));
2959 	outcred->cr_ref = 1;
2960 	outcred->cr_uid = incred->cr_uid;
2961 	outcred->cr_gid = incred->cr_gid;
2962 	outcred->cr_ngroups = incred->cr_ngroups;
2963 	for (i = 0; i < incred->cr_ngroups; i++)
2964 		outcred->cr_groups[i] = incred->cr_groups[i];
2965 	nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
2966 }
2967 
2968 u_int32_t
2969 nfs_getxid()
2970 {
2971 	static u_int32_t base;
2972 	static u_int32_t nfs_xid = 0;
2973 	static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER;
2974 	u_int32_t newxid;
2975 
2976 	simple_lock(&nfs_xidlock);
2977 	/*
2978 	 * derive initial xid from system time
2979 	 * XXX time is invalid if root not yet mounted
2980 	 */
2981 	if (__predict_false(!base && (rootvp))) {
2982 		struct timeval tv;
2983 
2984 		microtime(&tv);
2985 		base = tv.tv_sec << 12;
2986 		nfs_xid = base;
2987 	}
2988 
2989 	/*
2990 	 * Skip zero xid if it should ever happen.
2991 	 */
2992 	if (__predict_false(++nfs_xid == 0))
2993 		nfs_xid++;
2994 	newxid = nfs_xid;
2995 	simple_unlock(&nfs_xidlock);
2996 
2997 	return txdr_unsigned(newxid);
2998 }
2999 
3000 /*
3001  * assign a new xid for existing request.
3002  * used for NFSERR_JUKEBOX handling.
3003  */
3004 void
3005 nfs_renewxid(struct nfsreq *req)
3006 {
3007 	u_int32_t xid;
3008 	int off;
3009 
3010 	xid = nfs_getxid();
3011 	if (req->r_nmp->nm_sotype == SOCK_STREAM)
3012 		off = sizeof(u_int32_t); /* RPC record mark */
3013 	else
3014 		off = 0;
3015 
3016 	m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid);
3017 	req->r_xid = xid;
3018 }
3019