xref: /netbsd-src/sys/nfs/nfs_subs.c (revision fad4c9f71477ae11cea2ee75ec82151ac770a534)
1 /*	$NetBSD: nfs_subs.c,v 1.165 2006/06/07 22:34:17 kardel 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 1.165 2006/06/07 22:34:17 kardel 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 #include <sys/once.h>
101 #include <sys/kauth.h>
102 
103 #include <uvm/uvm_extern.h>
104 
105 #include <nfs/rpcv2.h>
106 #include <nfs/nfsproto.h>
107 #include <nfs/nfsnode.h>
108 #include <nfs/nfs.h>
109 #include <nfs/xdr_subs.h>
110 #include <nfs/nfsm_subs.h>
111 #include <nfs/nfsmount.h>
112 #include <nfs/nqnfs.h>
113 #include <nfs/nfsrtt.h>
114 #include <nfs/nfs_var.h>
115 
116 #include <miscfs/specfs/specdev.h>
117 
118 #include <netinet/in.h>
119 #ifdef ISO
120 #include <netiso/iso.h>
121 #endif
122 
123 /*
124  * Data items converted to xdr at startup, since they are constant
125  * This is kinda hokey, but may save a little time doing byte swaps
126  */
127 u_int32_t nfs_xdrneg1;
128 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr,
129 	rpc_mismatch, rpc_auth_unix, rpc_msgaccepted,
130 	rpc_auth_kerb;
131 u_int32_t nfs_prog, nqnfs_prog, nfs_true, nfs_false;
132 
133 /* And other global data */
134 const nfstype nfsv2_type[9] =
135 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON };
136 const nfstype nfsv3_type[9] =
137 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON };
138 const enum vtype nv2tov_type[8] =
139 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON };
140 const enum vtype nv3tov_type[8] =
141 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO };
142 int nfs_ticks;
143 int nfs_commitsize;
144 
145 MALLOC_DEFINE(M_NFSDIROFF, "NFS diroff", "NFS directory cookies");
146 
147 /* NFS client/server stats. */
148 struct nfsstats nfsstats;
149 
150 /*
151  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
152  */
153 const int nfsv3_procid[NFS_NPROCS] = {
154 	NFSPROC_NULL,
155 	NFSPROC_GETATTR,
156 	NFSPROC_SETATTR,
157 	NFSPROC_NOOP,
158 	NFSPROC_LOOKUP,
159 	NFSPROC_READLINK,
160 	NFSPROC_READ,
161 	NFSPROC_NOOP,
162 	NFSPROC_WRITE,
163 	NFSPROC_CREATE,
164 	NFSPROC_REMOVE,
165 	NFSPROC_RENAME,
166 	NFSPROC_LINK,
167 	NFSPROC_SYMLINK,
168 	NFSPROC_MKDIR,
169 	NFSPROC_RMDIR,
170 	NFSPROC_READDIR,
171 	NFSPROC_FSSTAT,
172 	NFSPROC_NOOP,
173 	NFSPROC_NOOP,
174 	NFSPROC_NOOP,
175 	NFSPROC_NOOP,
176 	NFSPROC_NOOP,
177 	NFSPROC_NOOP,
178 	NFSPROC_NOOP,
179 	NFSPROC_NOOP
180 };
181 
182 /*
183  * and the reverse mapping from generic to Version 2 procedure numbers
184  */
185 const int nfsv2_procid[NFS_NPROCS] = {
186 	NFSV2PROC_NULL,
187 	NFSV2PROC_GETATTR,
188 	NFSV2PROC_SETATTR,
189 	NFSV2PROC_LOOKUP,
190 	NFSV2PROC_NOOP,
191 	NFSV2PROC_READLINK,
192 	NFSV2PROC_READ,
193 	NFSV2PROC_WRITE,
194 	NFSV2PROC_CREATE,
195 	NFSV2PROC_MKDIR,
196 	NFSV2PROC_SYMLINK,
197 	NFSV2PROC_CREATE,
198 	NFSV2PROC_REMOVE,
199 	NFSV2PROC_RMDIR,
200 	NFSV2PROC_RENAME,
201 	NFSV2PROC_LINK,
202 	NFSV2PROC_READDIR,
203 	NFSV2PROC_NOOP,
204 	NFSV2PROC_STATFS,
205 	NFSV2PROC_NOOP,
206 	NFSV2PROC_NOOP,
207 	NFSV2PROC_NOOP,
208 	NFSV2PROC_NOOP,
209 	NFSV2PROC_NOOP,
210 	NFSV2PROC_NOOP,
211 	NFSV2PROC_NOOP,
212 };
213 
214 /*
215  * Maps errno values to nfs error numbers.
216  * Use NFSERR_IO as the catch all for ones not specifically defined in
217  * RFC 1094.
218  */
219 static const u_char nfsrv_v2errmap[ELAST] = {
220   NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
221   NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
222   NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
223   NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
224   NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
225   NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
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_IO,	NFSERR_IO,	NFSERR_IO,
231   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
232   NFSERR_IO,	NFSERR_IO,	NFSERR_NAMETOL,	NFSERR_IO,	NFSERR_IO,
233   NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
234   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
235   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
236   NFSERR_IO,	NFSERR_IO,
237 };
238 
239 /*
240  * Maps errno values to nfs error numbers.
241  * Although it is not obvious whether or not NFS clients really care if
242  * a returned error value is in the specified list for the procedure, the
243  * safest thing to do is filter them appropriately. For Version 2, the
244  * X/Open XNFS document is the only specification that defines error values
245  * for each RPC (The RFC simply lists all possible error values for all RPCs),
246  * so I have decided to not do this for Version 2.
247  * The first entry is the default error return and the rest are the valid
248  * errors for that RPC in increasing numeric order.
249  */
250 static const short nfsv3err_null[] = {
251 	0,
252 	0,
253 };
254 
255 static const short nfsv3err_getattr[] = {
256 	NFSERR_IO,
257 	NFSERR_IO,
258 	NFSERR_STALE,
259 	NFSERR_BADHANDLE,
260 	NFSERR_SERVERFAULT,
261 	0,
262 };
263 
264 static const short nfsv3err_setattr[] = {
265 	NFSERR_IO,
266 	NFSERR_PERM,
267 	NFSERR_IO,
268 	NFSERR_ACCES,
269 	NFSERR_INVAL,
270 	NFSERR_NOSPC,
271 	NFSERR_ROFS,
272 	NFSERR_DQUOT,
273 	NFSERR_STALE,
274 	NFSERR_BADHANDLE,
275 	NFSERR_NOT_SYNC,
276 	NFSERR_SERVERFAULT,
277 	0,
278 };
279 
280 static const short nfsv3err_lookup[] = {
281 	NFSERR_IO,
282 	NFSERR_NOENT,
283 	NFSERR_IO,
284 	NFSERR_ACCES,
285 	NFSERR_NOTDIR,
286 	NFSERR_NAMETOL,
287 	NFSERR_STALE,
288 	NFSERR_BADHANDLE,
289 	NFSERR_SERVERFAULT,
290 	0,
291 };
292 
293 static const short nfsv3err_access[] = {
294 	NFSERR_IO,
295 	NFSERR_IO,
296 	NFSERR_STALE,
297 	NFSERR_BADHANDLE,
298 	NFSERR_SERVERFAULT,
299 	0,
300 };
301 
302 static const short nfsv3err_readlink[] = {
303 	NFSERR_IO,
304 	NFSERR_IO,
305 	NFSERR_ACCES,
306 	NFSERR_INVAL,
307 	NFSERR_STALE,
308 	NFSERR_BADHANDLE,
309 	NFSERR_NOTSUPP,
310 	NFSERR_SERVERFAULT,
311 	0,
312 };
313 
314 static const short nfsv3err_read[] = {
315 	NFSERR_IO,
316 	NFSERR_IO,
317 	NFSERR_NXIO,
318 	NFSERR_ACCES,
319 	NFSERR_INVAL,
320 	NFSERR_STALE,
321 	NFSERR_BADHANDLE,
322 	NFSERR_SERVERFAULT,
323 	NFSERR_JUKEBOX,
324 	0,
325 };
326 
327 static const short nfsv3err_write[] = {
328 	NFSERR_IO,
329 	NFSERR_IO,
330 	NFSERR_ACCES,
331 	NFSERR_INVAL,
332 	NFSERR_FBIG,
333 	NFSERR_NOSPC,
334 	NFSERR_ROFS,
335 	NFSERR_DQUOT,
336 	NFSERR_STALE,
337 	NFSERR_BADHANDLE,
338 	NFSERR_SERVERFAULT,
339 	NFSERR_JUKEBOX,
340 	0,
341 };
342 
343 static const short nfsv3err_create[] = {
344 	NFSERR_IO,
345 	NFSERR_IO,
346 	NFSERR_ACCES,
347 	NFSERR_EXIST,
348 	NFSERR_NOTDIR,
349 	NFSERR_NOSPC,
350 	NFSERR_ROFS,
351 	NFSERR_NAMETOL,
352 	NFSERR_DQUOT,
353 	NFSERR_STALE,
354 	NFSERR_BADHANDLE,
355 	NFSERR_NOTSUPP,
356 	NFSERR_SERVERFAULT,
357 	0,
358 };
359 
360 static const short nfsv3err_mkdir[] = {
361 	NFSERR_IO,
362 	NFSERR_IO,
363 	NFSERR_ACCES,
364 	NFSERR_EXIST,
365 	NFSERR_NOTDIR,
366 	NFSERR_NOSPC,
367 	NFSERR_ROFS,
368 	NFSERR_NAMETOL,
369 	NFSERR_DQUOT,
370 	NFSERR_STALE,
371 	NFSERR_BADHANDLE,
372 	NFSERR_NOTSUPP,
373 	NFSERR_SERVERFAULT,
374 	0,
375 };
376 
377 static const short nfsv3err_symlink[] = {
378 	NFSERR_IO,
379 	NFSERR_IO,
380 	NFSERR_ACCES,
381 	NFSERR_EXIST,
382 	NFSERR_NOTDIR,
383 	NFSERR_NOSPC,
384 	NFSERR_ROFS,
385 	NFSERR_NAMETOL,
386 	NFSERR_DQUOT,
387 	NFSERR_STALE,
388 	NFSERR_BADHANDLE,
389 	NFSERR_NOTSUPP,
390 	NFSERR_SERVERFAULT,
391 	0,
392 };
393 
394 static const short nfsv3err_mknod[] = {
395 	NFSERR_IO,
396 	NFSERR_IO,
397 	NFSERR_ACCES,
398 	NFSERR_EXIST,
399 	NFSERR_NOTDIR,
400 	NFSERR_NOSPC,
401 	NFSERR_ROFS,
402 	NFSERR_NAMETOL,
403 	NFSERR_DQUOT,
404 	NFSERR_STALE,
405 	NFSERR_BADHANDLE,
406 	NFSERR_NOTSUPP,
407 	NFSERR_SERVERFAULT,
408 	NFSERR_BADTYPE,
409 	0,
410 };
411 
412 static const short nfsv3err_remove[] = {
413 	NFSERR_IO,
414 	NFSERR_NOENT,
415 	NFSERR_IO,
416 	NFSERR_ACCES,
417 	NFSERR_NOTDIR,
418 	NFSERR_ROFS,
419 	NFSERR_NAMETOL,
420 	NFSERR_STALE,
421 	NFSERR_BADHANDLE,
422 	NFSERR_SERVERFAULT,
423 	0,
424 };
425 
426 static const short nfsv3err_rmdir[] = {
427 	NFSERR_IO,
428 	NFSERR_NOENT,
429 	NFSERR_IO,
430 	NFSERR_ACCES,
431 	NFSERR_EXIST,
432 	NFSERR_NOTDIR,
433 	NFSERR_INVAL,
434 	NFSERR_ROFS,
435 	NFSERR_NAMETOL,
436 	NFSERR_NOTEMPTY,
437 	NFSERR_STALE,
438 	NFSERR_BADHANDLE,
439 	NFSERR_NOTSUPP,
440 	NFSERR_SERVERFAULT,
441 	0,
442 };
443 
444 static const short nfsv3err_rename[] = {
445 	NFSERR_IO,
446 	NFSERR_NOENT,
447 	NFSERR_IO,
448 	NFSERR_ACCES,
449 	NFSERR_EXIST,
450 	NFSERR_XDEV,
451 	NFSERR_NOTDIR,
452 	NFSERR_ISDIR,
453 	NFSERR_INVAL,
454 	NFSERR_NOSPC,
455 	NFSERR_ROFS,
456 	NFSERR_MLINK,
457 	NFSERR_NAMETOL,
458 	NFSERR_NOTEMPTY,
459 	NFSERR_DQUOT,
460 	NFSERR_STALE,
461 	NFSERR_BADHANDLE,
462 	NFSERR_NOTSUPP,
463 	NFSERR_SERVERFAULT,
464 	0,
465 };
466 
467 static const short nfsv3err_link[] = {
468 	NFSERR_IO,
469 	NFSERR_IO,
470 	NFSERR_ACCES,
471 	NFSERR_EXIST,
472 	NFSERR_XDEV,
473 	NFSERR_NOTDIR,
474 	NFSERR_INVAL,
475 	NFSERR_NOSPC,
476 	NFSERR_ROFS,
477 	NFSERR_MLINK,
478 	NFSERR_NAMETOL,
479 	NFSERR_DQUOT,
480 	NFSERR_STALE,
481 	NFSERR_BADHANDLE,
482 	NFSERR_NOTSUPP,
483 	NFSERR_SERVERFAULT,
484 	0,
485 };
486 
487 static const short nfsv3err_readdir[] = {
488 	NFSERR_IO,
489 	NFSERR_IO,
490 	NFSERR_ACCES,
491 	NFSERR_NOTDIR,
492 	NFSERR_STALE,
493 	NFSERR_BADHANDLE,
494 	NFSERR_BAD_COOKIE,
495 	NFSERR_TOOSMALL,
496 	NFSERR_SERVERFAULT,
497 	0,
498 };
499 
500 static const short nfsv3err_readdirplus[] = {
501 	NFSERR_IO,
502 	NFSERR_IO,
503 	NFSERR_ACCES,
504 	NFSERR_NOTDIR,
505 	NFSERR_STALE,
506 	NFSERR_BADHANDLE,
507 	NFSERR_BAD_COOKIE,
508 	NFSERR_NOTSUPP,
509 	NFSERR_TOOSMALL,
510 	NFSERR_SERVERFAULT,
511 	0,
512 };
513 
514 static const short nfsv3err_fsstat[] = {
515 	NFSERR_IO,
516 	NFSERR_IO,
517 	NFSERR_STALE,
518 	NFSERR_BADHANDLE,
519 	NFSERR_SERVERFAULT,
520 	0,
521 };
522 
523 static const short nfsv3err_fsinfo[] = {
524 	NFSERR_STALE,
525 	NFSERR_STALE,
526 	NFSERR_BADHANDLE,
527 	NFSERR_SERVERFAULT,
528 	0,
529 };
530 
531 static const short nfsv3err_pathconf[] = {
532 	NFSERR_STALE,
533 	NFSERR_STALE,
534 	NFSERR_BADHANDLE,
535 	NFSERR_SERVERFAULT,
536 	0,
537 };
538 
539 static const short nfsv3err_commit[] = {
540 	NFSERR_IO,
541 	NFSERR_IO,
542 	NFSERR_STALE,
543 	NFSERR_BADHANDLE,
544 	NFSERR_SERVERFAULT,
545 	0,
546 };
547 
548 static const short * const nfsrv_v3errmap[] = {
549 	nfsv3err_null,
550 	nfsv3err_getattr,
551 	nfsv3err_setattr,
552 	nfsv3err_lookup,
553 	nfsv3err_access,
554 	nfsv3err_readlink,
555 	nfsv3err_read,
556 	nfsv3err_write,
557 	nfsv3err_create,
558 	nfsv3err_mkdir,
559 	nfsv3err_symlink,
560 	nfsv3err_mknod,
561 	nfsv3err_remove,
562 	nfsv3err_rmdir,
563 	nfsv3err_rename,
564 	nfsv3err_link,
565 	nfsv3err_readdir,
566 	nfsv3err_readdirplus,
567 	nfsv3err_fsstat,
568 	nfsv3err_fsinfo,
569 	nfsv3err_pathconf,
570 	nfsv3err_commit,
571 };
572 
573 extern struct nfsrtt nfsrtt;
574 extern time_t nqnfsstarttime;
575 extern int nqsrv_clockskew;
576 extern int nqsrv_writeslack;
577 extern int nqsrv_maxlease;
578 extern const int nqnfs_piggy[NFS_NPROCS];
579 extern struct nfsnodehashhead *nfsnodehashtbl;
580 extern u_long nfsnodehash;
581 
582 u_long nfsdirhashmask;
583 
584 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *));
585 
586 /*
587  * Create the header for an rpc request packet
588  * The hsiz is the size of the rest of the nfs request header.
589  * (just used to decide if a cluster is a good idea)
590  */
591 struct mbuf *
592 nfsm_reqh(np, procid, hsiz, bposp)
593 	struct nfsnode *np;
594 	u_long procid;
595 	int hsiz;
596 	caddr_t *bposp;
597 {
598 	struct mbuf *mb;
599 	caddr_t bpos;
600 #ifndef NFS_V2_ONLY
601 	struct nfsmount *nmp;
602 	u_int32_t *tl;
603 	int nqflag;
604 #endif
605 
606 	mb = m_get(M_WAIT, MT_DATA);
607 	MCLAIM(mb, &nfs_mowner);
608 	if (hsiz >= MINCLSIZE)
609 		m_clget(mb, M_WAIT);
610 	mb->m_len = 0;
611 	bpos = mtod(mb, caddr_t);
612 
613 #ifndef NFS_V2_ONLY
614 	/*
615 	 * For NQNFS, add lease request.
616 	 */
617 	if (np) {
618 		nmp = VFSTONFS(np->n_vnode->v_mount);
619 		if (nmp->nm_flag & NFSMNT_NQNFS) {
620 			nqflag = NQNFS_NEEDLEASE(np, procid);
621 			if (nqflag) {
622 				nfsm_build(tl, u_int32_t *, 2*NFSX_UNSIGNED);
623 				*tl++ = txdr_unsigned(nqflag);
624 				*tl = txdr_unsigned(nmp->nm_leaseterm);
625 			} else {
626 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
627 				*tl = 0;
628 			}
629 		}
630 	}
631 #endif
632 	/* Finally, return values */
633 	*bposp = bpos;
634 	return (mb);
635 }
636 
637 /*
638  * Build the RPC header and fill in the authorization info.
639  * The authorization string argument is only used when the credentials
640  * come from outside of the kernel.
641  * Returns the head of the mbuf list.
642  */
643 struct mbuf *
644 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len,
645 	verf_str, mrest, mrest_len, mbp, xidp)
646 	kauth_cred_t cr;
647 	int nmflag;
648 	int procid;
649 	int auth_type;
650 	int auth_len;
651 	char *auth_str;
652 	int verf_len;
653 	char *verf_str;
654 	struct mbuf *mrest;
655 	int mrest_len;
656 	struct mbuf **mbp;
657 	u_int32_t *xidp;
658 {
659 	struct mbuf *mb;
660 	u_int32_t *tl;
661 	caddr_t bpos;
662 	int i;
663 	struct mbuf *mreq;
664 	int siz, grpsiz, authsiz;
665 
666 	authsiz = nfsm_rndup(auth_len);
667 	mb = m_gethdr(M_WAIT, MT_DATA);
668 	MCLAIM(mb, &nfs_mowner);
669 	if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
670 		m_clget(mb, M_WAIT);
671 	} else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
672 		MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED);
673 	} else {
674 		MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
675 	}
676 	mb->m_len = 0;
677 	mreq = mb;
678 	bpos = mtod(mb, caddr_t);
679 
680 	/*
681 	 * First the RPC header.
682 	 */
683 	nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
684 
685 	*tl++ = *xidp = nfs_getxid();
686 	*tl++ = rpc_call;
687 	*tl++ = rpc_vers;
688 	if (nmflag & NFSMNT_NQNFS) {
689 		*tl++ = txdr_unsigned(NQNFS_PROG);
690 		*tl++ = txdr_unsigned(NQNFS_VER3);
691 	} else {
692 		*tl++ = txdr_unsigned(NFS_PROG);
693 		if (nmflag & NFSMNT_NFSV3)
694 			*tl++ = txdr_unsigned(NFS_VER3);
695 		else
696 			*tl++ = txdr_unsigned(NFS_VER2);
697 	}
698 	if (nmflag & NFSMNT_NFSV3)
699 		*tl++ = txdr_unsigned(procid);
700 	else
701 		*tl++ = txdr_unsigned(nfsv2_procid[procid]);
702 
703 	/*
704 	 * And then the authorization cred.
705 	 */
706 	*tl++ = txdr_unsigned(auth_type);
707 	*tl = txdr_unsigned(authsiz);
708 	switch (auth_type) {
709 	case RPCAUTH_UNIX:
710 		nfsm_build(tl, u_int32_t *, auth_len);
711 		*tl++ = 0;		/* stamp ?? */
712 		*tl++ = 0;		/* NULL hostname */
713 		*tl++ = txdr_unsigned(kauth_cred_geteuid(cr));
714 		*tl++ = txdr_unsigned(kauth_cred_getegid(cr));
715 		grpsiz = (auth_len >> 2) - 5;
716 		*tl++ = txdr_unsigned(grpsiz);
717 		for (i = 0; i < grpsiz; i++)
718 			*tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */
719 		break;
720 	case RPCAUTH_KERB4:
721 		siz = auth_len;
722 		while (siz > 0) {
723 			if (M_TRAILINGSPACE(mb) == 0) {
724 				struct mbuf *mb2;
725 				mb2 = m_get(M_WAIT, MT_DATA);
726 				MCLAIM(mb2, &nfs_mowner);
727 				if (siz >= MINCLSIZE)
728 					m_clget(mb2, M_WAIT);
729 				mb->m_next = mb2;
730 				mb = mb2;
731 				mb->m_len = 0;
732 				bpos = mtod(mb, caddr_t);
733 			}
734 			i = min(siz, M_TRAILINGSPACE(mb));
735 			memcpy(bpos, auth_str, i);
736 			mb->m_len += i;
737 			auth_str += i;
738 			bpos += i;
739 			siz -= i;
740 		}
741 		if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
742 			for (i = 0; i < siz; i++)
743 				*bpos++ = '\0';
744 			mb->m_len += siz;
745 		}
746 		break;
747 	};
748 
749 	/*
750 	 * And the verifier...
751 	 */
752 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
753 	if (verf_str) {
754 		*tl++ = txdr_unsigned(RPCAUTH_KERB4);
755 		*tl = txdr_unsigned(verf_len);
756 		siz = verf_len;
757 		while (siz > 0) {
758 			if (M_TRAILINGSPACE(mb) == 0) {
759 				struct mbuf *mb2;
760 				mb2 = m_get(M_WAIT, MT_DATA);
761 				MCLAIM(mb2, &nfs_mowner);
762 				if (siz >= MINCLSIZE)
763 					m_clget(mb2, M_WAIT);
764 				mb->m_next = mb2;
765 				mb = mb2;
766 				mb->m_len = 0;
767 				bpos = mtod(mb, caddr_t);
768 			}
769 			i = min(siz, M_TRAILINGSPACE(mb));
770 			memcpy(bpos, verf_str, i);
771 			mb->m_len += i;
772 			verf_str += i;
773 			bpos += i;
774 			siz -= i;
775 		}
776 		if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
777 			for (i = 0; i < siz; i++)
778 				*bpos++ = '\0';
779 			mb->m_len += siz;
780 		}
781 	} else {
782 		*tl++ = txdr_unsigned(RPCAUTH_NULL);
783 		*tl = 0;
784 	}
785 	mb->m_next = mrest;
786 	mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
787 	mreq->m_pkthdr.rcvif = (struct ifnet *)0;
788 	*mbp = mb;
789 	return (mreq);
790 }
791 
792 /*
793  * copies mbuf chain to the uio scatter/gather list
794  */
795 int
796 nfsm_mbuftouio(mrep, uiop, siz, dpos)
797 	struct mbuf **mrep;
798 	struct uio *uiop;
799 	int siz;
800 	caddr_t *dpos;
801 {
802 	char *mbufcp, *uiocp;
803 	int xfer, left, len;
804 	struct mbuf *mp;
805 	long uiosiz, rem;
806 	int error = 0;
807 
808 	mp = *mrep;
809 	mbufcp = *dpos;
810 	len = mtod(mp, caddr_t)+mp->m_len-mbufcp;
811 	rem = nfsm_rndup(siz)-siz;
812 	while (siz > 0) {
813 		if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
814 			return (EFBIG);
815 		left = uiop->uio_iov->iov_len;
816 		uiocp = uiop->uio_iov->iov_base;
817 		if (left > siz)
818 			left = siz;
819 		uiosiz = left;
820 		while (left > 0) {
821 			while (len == 0) {
822 				mp = mp->m_next;
823 				if (mp == NULL)
824 					return (EBADRPC);
825 				mbufcp = mtod(mp, caddr_t);
826 				len = mp->m_len;
827 			}
828 			xfer = (left > len) ? len : left;
829 			error = copyout_vmspace(uiop->uio_vmspace, mbufcp,
830 			    uiocp, xfer);
831 			if (error) {
832 				return error;
833 			}
834 			left -= xfer;
835 			len -= xfer;
836 			mbufcp += xfer;
837 			uiocp += xfer;
838 			uiop->uio_offset += xfer;
839 			uiop->uio_resid -= xfer;
840 		}
841 		if (uiop->uio_iov->iov_len <= siz) {
842 			uiop->uio_iovcnt--;
843 			uiop->uio_iov++;
844 		} else {
845 			uiop->uio_iov->iov_base =
846 			    (caddr_t)uiop->uio_iov->iov_base + uiosiz;
847 			uiop->uio_iov->iov_len -= uiosiz;
848 		}
849 		siz -= uiosiz;
850 	}
851 	*dpos = mbufcp;
852 	*mrep = mp;
853 	if (rem > 0) {
854 		if (len < rem)
855 			error = nfs_adv(mrep, dpos, rem, len);
856 		else
857 			*dpos += rem;
858 	}
859 	return (error);
860 }
861 
862 /*
863  * copies a uio scatter/gather list to an mbuf chain.
864  * NOTE: can ony handle iovcnt == 1
865  */
866 int
867 nfsm_uiotombuf(uiop, mq, siz, bpos)
868 	struct uio *uiop;
869 	struct mbuf **mq;
870 	int siz;
871 	caddr_t *bpos;
872 {
873 	char *uiocp;
874 	struct mbuf *mp, *mp2;
875 	int xfer, left, mlen;
876 	int uiosiz, clflg, rem;
877 	char *cp;
878 	int error;
879 
880 #ifdef DIAGNOSTIC
881 	if (uiop->uio_iovcnt != 1)
882 		panic("nfsm_uiotombuf: iovcnt != 1");
883 #endif
884 
885 	if (siz > MLEN)		/* or should it >= MCLBYTES ?? */
886 		clflg = 1;
887 	else
888 		clflg = 0;
889 	rem = nfsm_rndup(siz)-siz;
890 	mp = mp2 = *mq;
891 	while (siz > 0) {
892 		left = uiop->uio_iov->iov_len;
893 		uiocp = uiop->uio_iov->iov_base;
894 		if (left > siz)
895 			left = siz;
896 		uiosiz = left;
897 		while (left > 0) {
898 			mlen = M_TRAILINGSPACE(mp);
899 			if (mlen == 0) {
900 				mp = m_get(M_WAIT, MT_DATA);
901 				MCLAIM(mp, &nfs_mowner);
902 				if (clflg)
903 					m_clget(mp, M_WAIT);
904 				mp->m_len = 0;
905 				mp2->m_next = mp;
906 				mp2 = mp;
907 				mlen = M_TRAILINGSPACE(mp);
908 			}
909 			xfer = (left > mlen) ? mlen : left;
910 			cp = mtod(mp, caddr_t) + mp->m_len;
911 			error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp,
912 			    xfer);
913 			if (error) {
914 				/* XXX */
915 			}
916 			mp->m_len += xfer;
917 			left -= xfer;
918 			uiocp += xfer;
919 			uiop->uio_offset += xfer;
920 			uiop->uio_resid -= xfer;
921 		}
922 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
923 		    uiosiz;
924 		uiop->uio_iov->iov_len -= uiosiz;
925 		siz -= uiosiz;
926 	}
927 	if (rem > 0) {
928 		if (rem > M_TRAILINGSPACE(mp)) {
929 			mp = m_get(M_WAIT, MT_DATA);
930 			MCLAIM(mp, &nfs_mowner);
931 			mp->m_len = 0;
932 			mp2->m_next = mp;
933 		}
934 		cp = mtod(mp, caddr_t) + mp->m_len;
935 		for (left = 0; left < rem; left++)
936 			*cp++ = '\0';
937 		mp->m_len += rem;
938 		*bpos = cp;
939 	} else
940 		*bpos = mtod(mp, caddr_t)+mp->m_len;
941 	*mq = mp;
942 	return (0);
943 }
944 
945 /*
946  * Get at least "siz" bytes of correctly aligned data.
947  * When called the mbuf pointers are not necessarily correct,
948  * dsosp points to what ought to be in m_data and left contains
949  * what ought to be in m_len.
950  * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
951  * cases. (The macros use the vars. dpos and dpos2)
952  */
953 int
954 nfsm_disct(mdp, dposp, siz, left, cp2)
955 	struct mbuf **mdp;
956 	caddr_t *dposp;
957 	int siz;
958 	int left;
959 	caddr_t *cp2;
960 {
961 	struct mbuf *m1, *m2;
962 	struct mbuf *havebuf = NULL;
963 	caddr_t src = *dposp;
964 	caddr_t dst;
965 	int len;
966 
967 #ifdef DEBUG
968 	if (left < 0)
969 		panic("nfsm_disct: left < 0");
970 #endif
971 	m1 = *mdp;
972 	/*
973 	 * Skip through the mbuf chain looking for an mbuf with
974 	 * some data. If the first mbuf found has enough data
975 	 * and it is correctly aligned return it.
976 	 */
977 	while (left == 0) {
978 		havebuf = m1;
979 		*mdp = m1 = m1->m_next;
980 		if (m1 == NULL)
981 			return (EBADRPC);
982 		src = mtod(m1, caddr_t);
983 		left = m1->m_len;
984 		/*
985 		 * If we start a new mbuf and it is big enough
986 		 * and correctly aligned just return it, don't
987 		 * do any pull up.
988 		 */
989 		if (left >= siz && nfsm_aligned(src)) {
990 			*cp2 = src;
991 			*dposp = src + siz;
992 			return (0);
993 		}
994 	}
995 	if (m1->m_flags & M_EXT) {
996 		if (havebuf) {
997 			/* If the first mbuf with data has external data
998 			 * and there is a previous empty mbuf use it
999 			 * to move the data into.
1000 			 */
1001 			m2 = m1;
1002 			*mdp = m1 = havebuf;
1003 			if (m1->m_flags & M_EXT) {
1004 				MEXTREMOVE(m1);
1005 			}
1006 		} else {
1007 			/*
1008 			 * If the first mbuf has a external data
1009 			 * and there is no previous empty mbuf
1010 			 * allocate a new mbuf and move the external
1011 			 * data to the new mbuf. Also make the first
1012 			 * mbuf look empty.
1013 			 */
1014 			m2 = m_get(M_WAIT, MT_DATA);
1015 			m2->m_ext = m1->m_ext;
1016 			m2->m_data = src;
1017 			m2->m_len = left;
1018 			MCLADDREFERENCE(m1, m2);
1019 			MEXTREMOVE(m1);
1020 			m2->m_next = m1->m_next;
1021 			m1->m_next = m2;
1022 		}
1023 		m1->m_len = 0;
1024 		if (m1->m_flags & M_PKTHDR)
1025 			dst = m1->m_pktdat;
1026 		else
1027 			dst = m1->m_dat;
1028 		m1->m_data = dst;
1029 	} else {
1030 		/*
1031 		 * If the first mbuf has no external data
1032 		 * move the data to the front of the mbuf.
1033 		 */
1034 		if (m1->m_flags & M_PKTHDR)
1035 			dst = m1->m_pktdat;
1036 		else
1037 			dst = m1->m_dat;
1038 		m1->m_data = dst;
1039 		if (dst != src)
1040 			memmove(dst, src, left);
1041 		dst += left;
1042 		m1->m_len = left;
1043 		m2 = m1->m_next;
1044 	}
1045 	*cp2 = m1->m_data;
1046 	*dposp = mtod(m1, caddr_t) + siz;
1047 	/*
1048 	 * Loop through mbufs pulling data up into first mbuf until
1049 	 * the first mbuf is full or there is no more data to
1050 	 * pullup.
1051 	 */
1052 	while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) {
1053 		if ((len = min(len, m2->m_len)) != 0)
1054 			memcpy(dst, m2->m_data, len);
1055 		m1->m_len += len;
1056 		dst += len;
1057 		m2->m_data += len;
1058 		m2->m_len -= len;
1059 		m2 = m2->m_next;
1060 	}
1061 	if (m1->m_len < siz)
1062 		return (EBADRPC);
1063 	return (0);
1064 }
1065 
1066 /*
1067  * Advance the position in the mbuf chain.
1068  */
1069 int
1070 nfs_adv(mdp, dposp, offs, left)
1071 	struct mbuf **mdp;
1072 	caddr_t *dposp;
1073 	int offs;
1074 	int left;
1075 {
1076 	struct mbuf *m;
1077 	int s;
1078 
1079 	m = *mdp;
1080 	s = left;
1081 	while (s < offs) {
1082 		offs -= s;
1083 		m = m->m_next;
1084 		if (m == NULL)
1085 			return (EBADRPC);
1086 		s = m->m_len;
1087 	}
1088 	*mdp = m;
1089 	*dposp = mtod(m, caddr_t)+offs;
1090 	return (0);
1091 }
1092 
1093 /*
1094  * Copy a string into mbufs for the hard cases...
1095  */
1096 int
1097 nfsm_strtmbuf(mb, bpos, cp, siz)
1098 	struct mbuf **mb;
1099 	char **bpos;
1100 	const char *cp;
1101 	long siz;
1102 {
1103 	struct mbuf *m1 = NULL, *m2;
1104 	long left, xfer, len, tlen;
1105 	u_int32_t *tl;
1106 	int putsize;
1107 
1108 	putsize = 1;
1109 	m2 = *mb;
1110 	left = M_TRAILINGSPACE(m2);
1111 	if (left > 0) {
1112 		tl = ((u_int32_t *)(*bpos));
1113 		*tl++ = txdr_unsigned(siz);
1114 		putsize = 0;
1115 		left -= NFSX_UNSIGNED;
1116 		m2->m_len += NFSX_UNSIGNED;
1117 		if (left > 0) {
1118 			memcpy((caddr_t) tl, cp, left);
1119 			siz -= left;
1120 			cp += left;
1121 			m2->m_len += left;
1122 			left = 0;
1123 		}
1124 	}
1125 	/* Loop around adding mbufs */
1126 	while (siz > 0) {
1127 		m1 = m_get(M_WAIT, MT_DATA);
1128 		MCLAIM(m1, &nfs_mowner);
1129 		if (siz > MLEN)
1130 			m_clget(m1, M_WAIT);
1131 		m1->m_len = NFSMSIZ(m1);
1132 		m2->m_next = m1;
1133 		m2 = m1;
1134 		tl = mtod(m1, u_int32_t *);
1135 		tlen = 0;
1136 		if (putsize) {
1137 			*tl++ = txdr_unsigned(siz);
1138 			m1->m_len -= NFSX_UNSIGNED;
1139 			tlen = NFSX_UNSIGNED;
1140 			putsize = 0;
1141 		}
1142 		if (siz < m1->m_len) {
1143 			len = nfsm_rndup(siz);
1144 			xfer = siz;
1145 			if (xfer < len)
1146 				*(tl+(xfer>>2)) = 0;
1147 		} else {
1148 			xfer = len = m1->m_len;
1149 		}
1150 		memcpy((caddr_t) tl, cp, xfer);
1151 		m1->m_len = len+tlen;
1152 		siz -= xfer;
1153 		cp += xfer;
1154 	}
1155 	*mb = m1;
1156 	*bpos = mtod(m1, caddr_t)+m1->m_len;
1157 	return (0);
1158 }
1159 
1160 /*
1161  * Directory caching routines. They work as follows:
1162  * - a cache is maintained per VDIR nfsnode.
1163  * - for each offset cookie that is exported to userspace, and can
1164  *   thus be thrown back at us as an offset to VOP_READDIR, store
1165  *   information in the cache.
1166  * - cached are:
1167  *   - cookie itself
1168  *   - blocknumber (essentially just a search key in the buffer cache)
1169  *   - entry number in block.
1170  *   - offset cookie of block in which this entry is stored
1171  *   - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
1172  * - entries are looked up in a hash table
1173  * - also maintained is an LRU list of entries, used to determine
1174  *   which ones to delete if the cache grows too large.
1175  * - if 32 <-> 64 translation mode is requested for a filesystem,
1176  *   the cache also functions as a translation table
1177  * - in the translation case, invalidating the cache does not mean
1178  *   flushing it, but just marking entries as invalid, except for
1179  *   the <64bit cookie, 32bitcookie> pair which is still valid, to
1180  *   still be able to use the cache as a translation table.
1181  * - 32 bit cookies are uniquely created by combining the hash table
1182  *   entry value, and one generation count per hash table entry,
1183  *   incremented each time an entry is appended to the chain.
1184  * - the cache is invalidated each time a direcory is modified
1185  * - sanity checks are also done; if an entry in a block turns
1186  *   out not to have a matching cookie, the cache is invalidated
1187  *   and a new block starting from the wanted offset is fetched from
1188  *   the server.
1189  * - directory entries as read from the server are extended to contain
1190  *   the 64bit and, optionally, the 32bit cookies, for sanity checking
1191  *   the cache and exporting them to userspace through the cookie
1192  *   argument to VOP_READDIR.
1193  */
1194 
1195 u_long
1196 nfs_dirhash(off)
1197 	off_t off;
1198 {
1199 	int i;
1200 	char *cp = (char *)&off;
1201 	u_long sum = 0L;
1202 
1203 	for (i = 0 ; i < sizeof (off); i++)
1204 		sum += *cp++;
1205 
1206 	return sum;
1207 }
1208 
1209 #define	_NFSDC_MTX(np)		(&NFSTOV(np)->v_interlock)
1210 #define	NFSDC_LOCK(np)		simple_lock(_NFSDC_MTX(np))
1211 #define	NFSDC_UNLOCK(np)	simple_unlock(_NFSDC_MTX(np))
1212 #define	NFSDC_ASSERT_LOCKED(np) LOCK_ASSERT(simple_lock_held(_NFSDC_MTX(np)))
1213 
1214 void
1215 nfs_initdircache(vp)
1216 	struct vnode *vp;
1217 {
1218 	struct nfsnode *np = VTONFS(vp);
1219 	struct nfsdirhashhead *dircache;
1220 
1221 	dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF,
1222 	    M_WAITOK, &nfsdirhashmask);
1223 
1224 	NFSDC_LOCK(np);
1225 	if (np->n_dircache == NULL) {
1226 		np->n_dircachesize = 0;
1227 		np->n_dircache = dircache;
1228 		dircache = NULL;
1229 		TAILQ_INIT(&np->n_dirchain);
1230 	}
1231 	NFSDC_UNLOCK(np);
1232 	if (dircache)
1233 		hashdone(dircache, M_NFSDIROFF);
1234 }
1235 
1236 void
1237 nfs_initdirxlatecookie(vp)
1238 	struct vnode *vp;
1239 {
1240 	struct nfsnode *np = VTONFS(vp);
1241 	unsigned *dirgens;
1242 
1243 	KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE);
1244 
1245 	dirgens = malloc(NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF,
1246 	    M_WAITOK|M_ZERO);
1247 	NFSDC_LOCK(np);
1248 	if (np->n_dirgens == NULL) {
1249 		np->n_dirgens = dirgens;
1250 		dirgens = NULL;
1251 	}
1252 	NFSDC_UNLOCK(np);
1253 	if (dirgens)
1254 		free(dirgens, M_NFSDIROFF);
1255 }
1256 
1257 static const struct nfsdircache dzero;
1258 
1259 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *));
1260 static void nfs_putdircache_unlocked __P((struct nfsnode *,
1261     struct nfsdircache *));
1262 
1263 static void
1264 nfs_unlinkdircache(np, ndp)
1265 	struct nfsnode *np;
1266 	struct nfsdircache *ndp;
1267 {
1268 
1269 	NFSDC_ASSERT_LOCKED(np);
1270 	KASSERT(ndp != &dzero);
1271 
1272 	if (LIST_NEXT(ndp, dc_hash) == (void *)-1)
1273 		return;
1274 
1275 	TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1276 	LIST_REMOVE(ndp, dc_hash);
1277 	LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */
1278 
1279 	nfs_putdircache_unlocked(np, ndp);
1280 }
1281 
1282 void
1283 nfs_putdircache(np, ndp)
1284 	struct nfsnode *np;
1285 	struct nfsdircache *ndp;
1286 {
1287 	int ref;
1288 
1289 	if (ndp == &dzero)
1290 		return;
1291 
1292 	KASSERT(ndp->dc_refcnt > 0);
1293 	NFSDC_LOCK(np);
1294 	ref = --ndp->dc_refcnt;
1295 	NFSDC_UNLOCK(np);
1296 
1297 	if (ref == 0)
1298 		free(ndp, M_NFSDIROFF);
1299 }
1300 
1301 static void
1302 nfs_putdircache_unlocked(np, ndp)
1303 	struct nfsnode *np;
1304 	struct nfsdircache *ndp;
1305 {
1306 	int ref;
1307 
1308 	NFSDC_ASSERT_LOCKED(np);
1309 
1310 	if (ndp == &dzero)
1311 		return;
1312 
1313 	KASSERT(ndp->dc_refcnt > 0);
1314 	ref = --ndp->dc_refcnt;
1315 	if (ref == 0)
1316 		free(ndp, M_NFSDIROFF);
1317 }
1318 
1319 struct nfsdircache *
1320 nfs_searchdircache(vp, off, do32, hashent)
1321 	struct vnode *vp;
1322 	off_t off;
1323 	int do32;
1324 	int *hashent;
1325 {
1326 	struct nfsdirhashhead *ndhp;
1327 	struct nfsdircache *ndp = NULL;
1328 	struct nfsnode *np = VTONFS(vp);
1329 	unsigned ent;
1330 
1331 	/*
1332 	 * Zero is always a valid cookie.
1333 	 */
1334 	if (off == 0)
1335 		/* XXXUNCONST */
1336 		return (struct nfsdircache *)__UNCONST(&dzero);
1337 
1338 	if (!np->n_dircache)
1339 		return NULL;
1340 
1341 	/*
1342 	 * We use a 32bit cookie as search key, directly reconstruct
1343 	 * the hashentry. Else use the hashfunction.
1344 	 */
1345 	if (do32) {
1346 		ent = (u_int32_t)off >> 24;
1347 		if (ent >= NFS_DIRHASHSIZ)
1348 			return NULL;
1349 		ndhp = &np->n_dircache[ent];
1350 	} else {
1351 		ndhp = NFSDIRHASH(np, off);
1352 	}
1353 
1354 	if (hashent)
1355 		*hashent = (int)(ndhp - np->n_dircache);
1356 
1357 	NFSDC_LOCK(np);
1358 	if (do32) {
1359 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1360 			if (ndp->dc_cookie32 == (u_int32_t)off) {
1361 				/*
1362 				 * An invalidated entry will become the
1363 				 * start of a new block fetched from
1364 				 * the server.
1365 				 */
1366 				if (ndp->dc_flags & NFSDC_INVALID) {
1367 					ndp->dc_blkcookie = ndp->dc_cookie;
1368 					ndp->dc_entry = 0;
1369 					ndp->dc_flags &= ~NFSDC_INVALID;
1370 				}
1371 				break;
1372 			}
1373 		}
1374 	} else {
1375 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1376 			if (ndp->dc_cookie == off)
1377 				break;
1378 		}
1379 	}
1380 	if (ndp != NULL)
1381 		ndp->dc_refcnt++;
1382 	NFSDC_UNLOCK(np);
1383 	return ndp;
1384 }
1385 
1386 
1387 struct nfsdircache *
1388 nfs_enterdircache(vp, off, blkoff, en, blkno)
1389 	struct vnode *vp;
1390 	off_t off, blkoff;
1391 	int en;
1392 	daddr_t blkno;
1393 {
1394 	struct nfsnode *np = VTONFS(vp);
1395 	struct nfsdirhashhead *ndhp;
1396 	struct nfsdircache *ndp = NULL;
1397 	struct nfsdircache *newndp = NULL;
1398 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1399 	int hashent = 0, gen, overwrite;	/* XXX: GCC */
1400 
1401 	/*
1402 	 * XXX refuse entries for offset 0. amd(8) erroneously sets
1403 	 * cookie 0 for the '.' entry, making this necessary. This
1404 	 * isn't so bad, as 0 is a special case anyway.
1405 	 */
1406 	if (off == 0)
1407 		/* XXXUNCONST */
1408 		return (struct nfsdircache *)__UNCONST(&dzero);
1409 
1410 	if (!np->n_dircache)
1411 		/*
1412 		 * XXX would like to do this in nfs_nget but vtype
1413 		 * isn't known at that time.
1414 		 */
1415 		nfs_initdircache(vp);
1416 
1417 	if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens)
1418 		nfs_initdirxlatecookie(vp);
1419 
1420 retry:
1421 	ndp = nfs_searchdircache(vp, off, 0, &hashent);
1422 
1423 	NFSDC_LOCK(np);
1424 	if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) {
1425 		/*
1426 		 * Overwriting an old entry. Check if it's the same.
1427 		 * If so, just return. If not, remove the old entry.
1428 		 */
1429 		if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
1430 			goto done;
1431 		nfs_unlinkdircache(np, ndp);
1432 		nfs_putdircache_unlocked(np, ndp);
1433 		ndp = NULL;
1434 	}
1435 
1436 	ndhp = &np->n_dircache[hashent];
1437 
1438 	if (!ndp) {
1439 		if (newndp == NULL) {
1440 			NFSDC_UNLOCK(np);
1441 			newndp = malloc(sizeof(*ndp), M_NFSDIROFF, M_WAITOK);
1442 			newndp->dc_refcnt = 1;
1443 			LIST_NEXT(newndp, dc_hash) = (void *)-1;
1444 			goto retry;
1445 		}
1446 		ndp = newndp;
1447 		newndp = NULL;
1448 		overwrite = 0;
1449 		if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1450 			/*
1451 			 * We're allocating a new entry, so bump the
1452 			 * generation number.
1453 			 */
1454 			KASSERT(np->n_dirgens);
1455 			gen = ++np->n_dirgens[hashent];
1456 			if (gen == 0) {
1457 				np->n_dirgens[hashent]++;
1458 				gen++;
1459 			}
1460 			ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
1461 		}
1462 	} else
1463 		overwrite = 1;
1464 
1465 	ndp->dc_cookie = off;
1466 	ndp->dc_blkcookie = blkoff;
1467 	ndp->dc_entry = en;
1468 	ndp->dc_flags = 0;
1469 
1470 	if (overwrite)
1471 		goto done;
1472 
1473 	/*
1474 	 * If the maximum directory cookie cache size has been reached
1475 	 * for this node, take one off the front. The idea is that
1476 	 * directories are typically read front-to-back once, so that
1477 	 * the oldest entries can be thrown away without much performance
1478 	 * loss.
1479 	 */
1480 	if (np->n_dircachesize == NFS_MAXDIRCACHE) {
1481 		nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain));
1482 	} else
1483 		np->n_dircachesize++;
1484 
1485 	KASSERT(ndp->dc_refcnt == 1);
1486 	LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
1487 	TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
1488 	ndp->dc_refcnt++;
1489 done:
1490 	KASSERT(ndp->dc_refcnt > 0);
1491 	NFSDC_UNLOCK(np);
1492 	if (newndp)
1493 		nfs_putdircache(np, newndp);
1494 	return ndp;
1495 }
1496 
1497 void
1498 nfs_invaldircache(vp, flags)
1499 	struct vnode *vp;
1500 	int flags;
1501 {
1502 	struct nfsnode *np = VTONFS(vp);
1503 	struct nfsdircache *ndp = NULL;
1504 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1505 	const boolean_t forcefree = flags & NFS_INVALDIRCACHE_FORCE;
1506 
1507 #ifdef DIAGNOSTIC
1508 	if (vp->v_type != VDIR)
1509 		panic("nfs: invaldircache: not dir");
1510 #endif
1511 
1512 	if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0)
1513 		np->n_flag &= ~NEOFVALID;
1514 
1515 	if (!np->n_dircache)
1516 		return;
1517 
1518 	NFSDC_LOCK(np);
1519 	if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
1520 		while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) {
1521 			KASSERT(!forcefree || ndp->dc_refcnt == 1);
1522 			nfs_unlinkdircache(np, ndp);
1523 		}
1524 		np->n_dircachesize = 0;
1525 		if (forcefree && np->n_dirgens) {
1526 			FREE(np->n_dirgens, M_NFSDIROFF);
1527 			np->n_dirgens = NULL;
1528 		}
1529 	} else {
1530 		TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain)
1531 			ndp->dc_flags |= NFSDC_INVALID;
1532 	}
1533 
1534 	NFSDC_UNLOCK(np);
1535 }
1536 
1537 /*
1538  * Called once before VFS init to initialize shared and
1539  * server-specific data structures.
1540  */
1541 static int
1542 nfs_init0(void)
1543 {
1544 	nfsrtt.pos = 0;
1545 	rpc_vers = txdr_unsigned(RPC_VER2);
1546 	rpc_call = txdr_unsigned(RPC_CALL);
1547 	rpc_reply = txdr_unsigned(RPC_REPLY);
1548 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1549 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1550 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1551 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1552 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1553 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1554 	nfs_prog = txdr_unsigned(NFS_PROG);
1555 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
1556 	nfs_true = txdr_unsigned(TRUE);
1557 	nfs_false = txdr_unsigned(FALSE);
1558 	nfs_xdrneg1 = txdr_unsigned(-1);
1559 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1560 	if (nfs_ticks < 1)
1561 		nfs_ticks = 1;
1562 #ifdef NFSSERVER
1563 	nfsrv_init(0);			/* Init server data structures */
1564 	nfsrv_initcache();		/* Init the server request cache */
1565 #endif /* NFSSERVER */
1566 
1567 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY))
1568 	nfsdreq_init();
1569 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */
1570 
1571 #if defined(NFSSERVER) || !defined(NFS_V2_ONLY)
1572 	/*
1573 	 * Initialize the nqnfs data structures.
1574 	 */
1575 	if (nqnfsstarttime == 0) {
1576 		nqnfsstarttime = boottime.tv_sec + nqsrv_maxlease
1577 			+ nqsrv_clockskew + nqsrv_writeslack;
1578 		NQLOADNOVRAM(nqnfsstarttime);
1579 		CIRCLEQ_INIT(&nqtimerhead);
1580 		nqfhhashtbl = hashinit(NQLCHSZ, HASH_LIST, M_NQLEASE,
1581 		    M_WAITOK, &nqfhhash);
1582 	}
1583 #endif
1584 
1585 	exithook_establish(nfs_exit, NULL);
1586 
1587 	/*
1588 	 * Initialize reply list and start timer
1589 	 */
1590 	TAILQ_INIT(&nfs_reqq);
1591 	nfs_timer(NULL);
1592 	MOWNER_ATTACH(&nfs_mowner);
1593 
1594 #ifdef NFS
1595 	/* Initialize the kqueue structures */
1596 	nfs_kqinit();
1597 	/* Initialize the iod structures */
1598 	nfs_iodinit();
1599 #endif
1600 	return 0;
1601 }
1602 
1603 void
1604 nfs_init(void)
1605 {
1606 	static ONCE_DECL(nfs_init_once);
1607 
1608 	RUN_ONCE(&nfs_init_once, nfs_init0);
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 	/* Initialize NFS server / client shared data. */
1619 	nfs_init();
1620 
1621 	nfs_nhinit();			/* Init the nfsnode table */
1622 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1623 }
1624 
1625 void
1626 nfs_vfs_reinit()
1627 {
1628 	nfs_nhreinit();
1629 }
1630 
1631 void
1632 nfs_vfs_done()
1633 {
1634 	nfs_nhdone();
1635 }
1636 
1637 /*
1638  * Attribute cache routines.
1639  * nfs_loadattrcache() - loads or updates the cache contents from attributes
1640  *	that are on the mbuf list
1641  * nfs_getattrcache() - returns valid attributes if found in cache, returns
1642  *	error otherwise
1643  */
1644 
1645 /*
1646  * Load the attribute cache (that lives in the nfsnode entry) with
1647  * the values on the mbuf list and
1648  * Iff vap not NULL
1649  *    copy the attributes to *vaper
1650  */
1651 int
1652 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags)
1653 	struct vnode **vpp;
1654 	struct mbuf **mdp;
1655 	caddr_t *dposp;
1656 	struct vattr *vaper;
1657 	int flags;
1658 {
1659 	int32_t t1;
1660 	caddr_t cp2;
1661 	int error = 0;
1662 	struct mbuf *md;
1663 	int v3 = NFS_ISV3(*vpp);
1664 
1665 	md = *mdp;
1666 	t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
1667 	error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2);
1668 	if (error)
1669 		return (error);
1670 	return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags);
1671 }
1672 
1673 int
1674 nfs_loadattrcache(vpp, fp, vaper, flags)
1675 	struct vnode **vpp;
1676 	struct nfs_fattr *fp;
1677 	struct vattr *vaper;
1678 	int flags;
1679 {
1680 	struct vnode *vp = *vpp;
1681 	struct vattr *vap;
1682 	int v3 = NFS_ISV3(vp);
1683 	enum vtype vtyp;
1684 	u_short vmode;
1685 	struct timespec mtime;
1686 	struct timespec ctime;
1687 	struct vnode *nvp;
1688 	int32_t rdev;
1689 	struct nfsnode *np;
1690 	extern int (**spec_nfsv2nodeop_p) __P((void *));
1691 	uid_t uid;
1692 	gid_t gid;
1693 
1694 	if (v3) {
1695 		vtyp = nfsv3tov_type(fp->fa_type);
1696 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1697 		rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1),
1698 			fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2));
1699 		fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
1700 		fxdr_nfsv3time(&fp->fa3_ctime, &ctime);
1701 	} else {
1702 		vtyp = nfsv2tov_type(fp->fa_type);
1703 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1704 		if (vtyp == VNON || vtyp == VREG)
1705 			vtyp = IFTOVT(vmode);
1706 		rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
1707 		fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
1708 		ctime.tv_sec = fxdr_unsigned(u_int32_t,
1709 		    fp->fa2_ctime.nfsv2_sec);
1710 		ctime.tv_nsec = 0;
1711 
1712 		/*
1713 		 * Really ugly NFSv2 kludge.
1714 		 */
1715 		if (vtyp == VCHR && rdev == 0xffffffff)
1716 			vtyp = VFIFO;
1717 	}
1718 
1719 	vmode &= ALLPERMS;
1720 
1721 	/*
1722 	 * If v_type == VNON it is a new node, so fill in the v_type,
1723 	 * n_mtime fields. Check to see if it represents a special
1724 	 * device, and if so, check for a possible alias. Once the
1725 	 * correct vnode has been obtained, fill in the rest of the
1726 	 * information.
1727 	 */
1728 	np = VTONFS(vp);
1729 	if (vp->v_type == VNON) {
1730 		vp->v_type = vtyp;
1731 		if (vp->v_type == VFIFO) {
1732 			extern int (**fifo_nfsv2nodeop_p) __P((void *));
1733 			vp->v_op = fifo_nfsv2nodeop_p;
1734 		} else if (vp->v_type == VREG) {
1735 			lockinit(&np->n_commitlock, PINOD, "nfsclock", 0, 0);
1736 		} else if (vp->v_type == VCHR || vp->v_type == VBLK) {
1737 			vp->v_op = spec_nfsv2nodeop_p;
1738 			nvp = checkalias(vp, (dev_t)rdev, vp->v_mount);
1739 			if (nvp) {
1740 				/*
1741 				 * Discard unneeded vnode, but save its nfsnode.
1742 				 * Since the nfsnode does not have a lock, its
1743 				 * vnode lock has to be carried over.
1744 				 */
1745 				/*
1746 				 * XXX is the old node sure to be locked here?
1747 				 */
1748 				KASSERT(lockstatus(&vp->v_lock) ==
1749 				    LK_EXCLUSIVE);
1750 				nvp->v_data = vp->v_data;
1751 				vp->v_data = NULL;
1752 				VOP_UNLOCK(vp, 0);
1753 				vp->v_op = spec_vnodeop_p;
1754 				vrele(vp);
1755 				vgone(vp);
1756 				lockmgr(&nvp->v_lock, LK_EXCLUSIVE,
1757 				    &nvp->v_interlock);
1758 				/*
1759 				 * Reinitialize aliased node.
1760 				 */
1761 				np->n_vnode = nvp;
1762 				*vpp = vp = nvp;
1763 			}
1764 		}
1765 		np->n_mtime = mtime;
1766 	}
1767 	uid = fxdr_unsigned(uid_t, fp->fa_uid);
1768 	gid = fxdr_unsigned(gid_t, fp->fa_gid);
1769 	vap = np->n_vattr;
1770 
1771 	/*
1772 	 * Invalidate access cache if uid, gid, mode or ctime changed.
1773 	 */
1774 	if (np->n_accstamp != -1 &&
1775 	    (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode
1776 	    || timespeccmp(&ctime, &vap->va_ctime, !=)))
1777 		np->n_accstamp = -1;
1778 
1779 	vap->va_type = vtyp;
1780 	vap->va_mode = vmode;
1781 	vap->va_rdev = (dev_t)rdev;
1782 	vap->va_mtime = mtime;
1783 	vap->va_ctime = ctime;
1784 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1785 	switch (vtyp) {
1786 	case VDIR:
1787 		vap->va_blocksize = NFS_DIRFRAGSIZ;
1788 		break;
1789 	case VBLK:
1790 		vap->va_blocksize = BLKDEV_IOSIZE;
1791 		break;
1792 	case VCHR:
1793 		vap->va_blocksize = MAXBSIZE;
1794 		break;
1795 	default:
1796 		vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize :
1797 		    fxdr_unsigned(int32_t, fp->fa2_blocksize);
1798 		break;
1799 	}
1800 	if (v3) {
1801 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1802 		vap->va_uid = uid;
1803 		vap->va_gid = gid;
1804 		vap->va_size = fxdr_hyper(&fp->fa3_size);
1805 		vap->va_bytes = fxdr_hyper(&fp->fa3_used);
1806 		vap->va_fileid = fxdr_hyper(&fp->fa3_fileid);
1807 		fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
1808 		vap->va_flags = 0;
1809 		vap->va_filerev = 0;
1810 	} else {
1811 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1812 		vap->va_uid = uid;
1813 		vap->va_gid = gid;
1814 		vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
1815 		vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks)
1816 		    * NFS_FABLKSIZE;
1817 		vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
1818 		fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
1819 		vap->va_flags = 0;
1820 		vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
1821 		vap->va_filerev = 0;
1822 	}
1823 	if (vap->va_size != np->n_size) {
1824 		if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) {
1825 			vap->va_size = np->n_size;
1826 		} else {
1827 			np->n_size = vap->va_size;
1828 			if (vap->va_type == VREG) {
1829 				/*
1830 				 * we can't free pages if NAC_NOTRUNC because
1831 				 * the pages can be owned by ourselves.
1832 				 */
1833 				if (flags & NAC_NOTRUNC) {
1834 					np->n_flag |= NTRUNCDELAYED;
1835 				} else {
1836 					simple_lock(&vp->v_interlock);
1837 					(void)VOP_PUTPAGES(vp, 0,
1838 					    0, PGO_SYNCIO | PGO_CLEANIT |
1839 					    PGO_FREE | PGO_ALLPAGES);
1840 					uvm_vnp_setsize(vp, np->n_size);
1841 				}
1842 			}
1843 		}
1844 	}
1845 	np->n_attrstamp = time_second;
1846 	if (vaper != NULL) {
1847 		memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(*vap));
1848 		if (np->n_flag & NCHG) {
1849 			if (np->n_flag & NACC)
1850 				vaper->va_atime = np->n_atim;
1851 			if (np->n_flag & NUPD)
1852 				vaper->va_mtime = np->n_mtim;
1853 		}
1854 	}
1855 	return (0);
1856 }
1857 
1858 /*
1859  * Check the time stamp
1860  * If the cache is valid, copy contents to *vap and return 0
1861  * otherwise return an error
1862  */
1863 int
1864 nfs_getattrcache(vp, vaper)
1865 	struct vnode *vp;
1866 	struct vattr *vaper;
1867 {
1868 	struct nfsnode *np = VTONFS(vp);
1869 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1870 	struct vattr *vap;
1871 
1872 	if (np->n_attrstamp == 0 ||
1873 	    (time_second - np->n_attrstamp) >= NFS_ATTRTIMEO(nmp, np)) {
1874 		nfsstats.attrcache_misses++;
1875 		return (ENOENT);
1876 	}
1877 	nfsstats.attrcache_hits++;
1878 	vap = np->n_vattr;
1879 	if (vap->va_size != np->n_size) {
1880 		if (vap->va_type == VREG) {
1881 			if (np->n_flag & NMODIFIED) {
1882 				if (vap->va_size < np->n_size)
1883 					vap->va_size = np->n_size;
1884 				else
1885 					np->n_size = vap->va_size;
1886 			} else
1887 				np->n_size = vap->va_size;
1888 			uvm_vnp_setsize(vp, np->n_size);
1889 		} else
1890 			np->n_size = vap->va_size;
1891 	}
1892 	memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(struct vattr));
1893 	if (np->n_flag & NCHG) {
1894 		if (np->n_flag & NACC)
1895 			vaper->va_atime = np->n_atim;
1896 		if (np->n_flag & NUPD)
1897 			vaper->va_mtime = np->n_mtim;
1898 	}
1899 	return (0);
1900 }
1901 
1902 void
1903 nfs_delayedtruncate(vp)
1904 	struct vnode *vp;
1905 {
1906 	struct nfsnode *np = VTONFS(vp);
1907 
1908 	if (np->n_flag & NTRUNCDELAYED) {
1909 		np->n_flag &= ~NTRUNCDELAYED;
1910 		simple_lock(&vp->v_interlock);
1911 		(void)VOP_PUTPAGES(vp, 0,
1912 		    0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES);
1913 		uvm_vnp_setsize(vp, np->n_size);
1914 	}
1915 }
1916 
1917 #define	NFS_WCCKLUDGE_TIMEOUT	(24 * 60 * 60)	/* 1 day */
1918 #define	NFS_WCCKLUDGE(nmp, now) \
1919 	(((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \
1920 	((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0)
1921 
1922 /*
1923  * nfs_check_wccdata: check inaccurate wcc_data
1924  *
1925  * => return non-zero if we shouldn't trust the wcc_data.
1926  * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed".
1927  */
1928 
1929 int
1930 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime,
1931     struct timespec *mtime, boolean_t docheck)
1932 {
1933 	int error = 0;
1934 
1935 #if !defined(NFS_V2_ONLY)
1936 
1937 	if (docheck) {
1938 		struct vnode *vp = NFSTOV(np);
1939 		struct nfsmount *nmp;
1940 		long now = time_second;
1941 #if defined(DEBUG)
1942 		const char *reason = NULL; /* XXX: gcc */
1943 #endif
1944 
1945 		if (timespeccmp(&np->n_vattr->va_mtime, mtime, <=)) {
1946 #if defined(DEBUG)
1947 			reason = "mtime";
1948 #endif
1949 			error = EINVAL;
1950 		}
1951 
1952 		if (vp->v_type == VDIR &&
1953 		    timespeccmp(&np->n_vattr->va_ctime, ctime, <=)) {
1954 #if defined(DEBUG)
1955 			reason = "ctime";
1956 #endif
1957 			error = EINVAL;
1958 		}
1959 
1960 		nmp = VFSTONFS(vp->v_mount);
1961 		if (error) {
1962 
1963 			/*
1964 			 * despite of the fact that we've updated the file,
1965 			 * timestamps of the file were not updated as we
1966 			 * expected.
1967 			 * it means that the server has incompatible
1968 			 * semantics of timestamps or (more likely)
1969 			 * the server time is not precise enough to
1970 			 * track each modifications.
1971 			 * in that case, we disable wcc processing.
1972 			 *
1973 			 * yes, strictly speaking, we should disable all
1974 			 * caching.  it's a compromise.
1975 			 */
1976 
1977 			simple_lock(&nmp->nm_slock);
1978 #if defined(DEBUG)
1979 			if (!NFS_WCCKLUDGE(nmp, now)) {
1980 				printf("%s: inaccurate wcc data (%s) detected,"
1981 				    " disabling wcc\n",
1982 				    vp->v_mount->mnt_stat.f_mntfromname,
1983 				    reason);
1984 			}
1985 #endif
1986 			nmp->nm_iflag |= NFSMNT_WCCKLUDGE;
1987 			nmp->nm_wcckludgetime = now;
1988 			simple_unlock(&nmp->nm_slock);
1989 		} else if (NFS_WCCKLUDGE(nmp, now)) {
1990 			error = EPERM; /* XXX */
1991 		} else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
1992 			simple_lock(&nmp->nm_slock);
1993 			if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
1994 #if defined(DEBUG)
1995 				printf("%s: re-enabling wcc\n",
1996 				    vp->v_mount->mnt_stat.f_mntfromname);
1997 #endif
1998 				nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE;
1999 			}
2000 			simple_unlock(&nmp->nm_slock);
2001 		}
2002 	}
2003 
2004 #endif /* !defined(NFS_V2_ONLY) */
2005 
2006 	return error;
2007 }
2008 
2009 /*
2010  * Heuristic to see if the server XDR encodes directory cookies or not.
2011  * it is not supposed to, but a lot of servers may do this. Also, since
2012  * most/all servers will implement V2 as well, it is expected that they
2013  * may return just 32 bits worth of cookie information, so we need to
2014  * find out in which 32 bits this information is available. We do this
2015  * to avoid trouble with emulated binaries that can't handle 64 bit
2016  * directory offsets.
2017  */
2018 
2019 void
2020 nfs_cookieheuristic(vp, flagp, l, cred)
2021 	struct vnode *vp;
2022 	int *flagp;
2023 	struct lwp *l;
2024 	kauth_cred_t cred;
2025 {
2026 	struct uio auio;
2027 	struct iovec aiov;
2028 	caddr_t tbuf, cp;
2029 	struct dirent *dp;
2030 	off_t *cookies = NULL, *cop;
2031 	int error, eof, nc, len;
2032 
2033 	MALLOC(tbuf, caddr_t, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK);
2034 
2035 	aiov.iov_base = tbuf;
2036 	aiov.iov_len = NFS_DIRFRAGSIZ;
2037 	auio.uio_iov = &aiov;
2038 	auio.uio_iovcnt = 1;
2039 	auio.uio_rw = UIO_READ;
2040 	auio.uio_resid = NFS_DIRFRAGSIZ;
2041 	auio.uio_offset = 0;
2042 	UIO_SETUP_SYSSPACE(&auio);
2043 
2044 	error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc);
2045 
2046 	len = NFS_DIRFRAGSIZ - auio.uio_resid;
2047 	if (error || len == 0) {
2048 		FREE(tbuf, M_TEMP);
2049 		if (cookies)
2050 			free(cookies, M_TEMP);
2051 		return;
2052 	}
2053 
2054 	/*
2055 	 * Find the first valid entry and look at its offset cookie.
2056 	 */
2057 
2058 	cp = tbuf;
2059 	for (cop = cookies; len > 0; len -= dp->d_reclen) {
2060 		dp = (struct dirent *)cp;
2061 		if (dp->d_fileno != 0 && len >= dp->d_reclen) {
2062 			if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) {
2063 				*flagp |= NFSMNT_SWAPCOOKIE;
2064 				nfs_invaldircache(vp, 0);
2065 				nfs_vinvalbuf(vp, 0, cred, l, 1);
2066 			}
2067 			break;
2068 		}
2069 		cop++;
2070 		cp += dp->d_reclen;
2071 	}
2072 
2073 	FREE(tbuf, M_TEMP);
2074 	free(cookies, M_TEMP);
2075 }
2076 #endif /* NFS */
2077 
2078 #ifdef NFSSERVER
2079 /*
2080  * Set up nameidata for a lookup() call and do it.
2081  *
2082  * If pubflag is set, this call is done for a lookup operation on the
2083  * public filehandle. In that case we allow crossing mountpoints and
2084  * absolute pathnames. However, the caller is expected to check that
2085  * the lookup result is within the public fs, and deny access if
2086  * it is not.
2087  */
2088 int
2089 nfs_namei(ndp, fhp, len, slp, nam, mdp, dposp, retdirp, l, kerbflag, pubflag)
2090 	struct nameidata *ndp;
2091 	fhandle_t *fhp;
2092 	uint32_t len;
2093 	struct nfssvc_sock *slp;
2094 	struct mbuf *nam;
2095 	struct mbuf **mdp;
2096 	caddr_t *dposp;
2097 	struct vnode **retdirp;
2098 	struct lwp *l;
2099 	int kerbflag, pubflag;
2100 {
2101 	int i, rem;
2102 	struct mbuf *md;
2103 	char *fromcp, *tocp, *cp;
2104 	struct iovec aiov;
2105 	struct uio auio;
2106 	struct vnode *dp;
2107 	int error, rdonly, linklen;
2108 	struct componentname *cnp = &ndp->ni_cnd;
2109 
2110 	*retdirp = (struct vnode *)0;
2111 
2112 	if ((len + 1) > MAXPATHLEN)
2113 		return (ENAMETOOLONG);
2114 	if (len == 0)
2115 		return (EACCES);
2116 	cnp->cn_pnbuf = PNBUF_GET();
2117 
2118 	/*
2119 	 * Copy the name from the mbuf list to ndp->ni_pnbuf
2120 	 * and set the various ndp fields appropriately.
2121 	 */
2122 	fromcp = *dposp;
2123 	tocp = cnp->cn_pnbuf;
2124 	md = *mdp;
2125 	rem = mtod(md, caddr_t) + md->m_len - fromcp;
2126 	for (i = 0; i < len; i++) {
2127 		while (rem == 0) {
2128 			md = md->m_next;
2129 			if (md == NULL) {
2130 				error = EBADRPC;
2131 				goto out;
2132 			}
2133 			fromcp = mtod(md, caddr_t);
2134 			rem = md->m_len;
2135 		}
2136 		if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
2137 			error = EACCES;
2138 			goto out;
2139 		}
2140 		*tocp++ = *fromcp++;
2141 		rem--;
2142 	}
2143 	*tocp = '\0';
2144 	*mdp = md;
2145 	*dposp = fromcp;
2146 	len = nfsm_rndup(len)-len;
2147 	if (len > 0) {
2148 		if (rem >= len)
2149 			*dposp += len;
2150 		else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
2151 			goto out;
2152 	}
2153 
2154 	/*
2155 	 * Extract and set starting directory.
2156 	 */
2157 	error = nfsrv_fhtovp(fhp, FALSE, &dp, ndp->ni_cnd.cn_cred, slp,
2158 	    nam, &rdonly, kerbflag, pubflag);
2159 	if (error)
2160 		goto out;
2161 	if (dp->v_type != VDIR) {
2162 		vrele(dp);
2163 		error = ENOTDIR;
2164 		goto out;
2165 	}
2166 
2167 	if (rdonly)
2168 		cnp->cn_flags |= RDONLY;
2169 
2170 	*retdirp = dp;
2171 
2172 	if (pubflag) {
2173 		/*
2174 		 * Oh joy. For WebNFS, handle those pesky '%' escapes,
2175 		 * and the 'native path' indicator.
2176 		 */
2177 		cp = PNBUF_GET();
2178 		fromcp = cnp->cn_pnbuf;
2179 		tocp = cp;
2180 		if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
2181 			switch ((unsigned char)*fromcp) {
2182 			case WEBNFS_NATIVE_CHAR:
2183 				/*
2184 				 * 'Native' path for us is the same
2185 				 * as a path according to the NFS spec,
2186 				 * just skip the escape char.
2187 				 */
2188 				fromcp++;
2189 				break;
2190 			/*
2191 			 * More may be added in the future, range 0x80-0xff
2192 			 */
2193 			default:
2194 				error = EIO;
2195 				PNBUF_PUT(cp);
2196 				goto out;
2197 			}
2198 		}
2199 		/*
2200 		 * Translate the '%' escapes, URL-style.
2201 		 */
2202 		while (*fromcp != '\0') {
2203 			if (*fromcp == WEBNFS_ESC_CHAR) {
2204 				if (fromcp[1] != '\0' && fromcp[2] != '\0') {
2205 					fromcp++;
2206 					*tocp++ = HEXSTRTOI(fromcp);
2207 					fromcp += 2;
2208 					continue;
2209 				} else {
2210 					error = ENOENT;
2211 					PNBUF_PUT(cp);
2212 					goto out;
2213 				}
2214 			} else
2215 				*tocp++ = *fromcp++;
2216 		}
2217 		*tocp = '\0';
2218 		PNBUF_PUT(cnp->cn_pnbuf);
2219 		cnp->cn_pnbuf = cp;
2220 	}
2221 
2222 	ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1;
2223 	ndp->ni_segflg = UIO_SYSSPACE;
2224 	ndp->ni_rootdir = rootvnode;
2225 
2226 	if (pubflag) {
2227 		ndp->ni_loopcnt = 0;
2228 		if (cnp->cn_pnbuf[0] == '/')
2229 			dp = rootvnode;
2230 	} else {
2231 		cnp->cn_flags |= NOCROSSMOUNT;
2232 	}
2233 
2234 	cnp->cn_lwp = l;
2235 	VREF(dp);
2236 
2237     for (;;) {
2238 	cnp->cn_nameptr = cnp->cn_pnbuf;
2239 	ndp->ni_startdir = dp;
2240 	/*
2241 	 * And call lookup() to do the real work
2242 	 */
2243 	error = lookup(ndp);
2244 	if (error) {
2245 		PNBUF_PUT(cnp->cn_pnbuf);
2246 		return (error);
2247 	}
2248 	/*
2249 	 * Check for encountering a symbolic link
2250 	 */
2251 	if ((cnp->cn_flags & ISSYMLINK) == 0) {
2252 		if (cnp->cn_flags & (SAVENAME | SAVESTART))
2253 			cnp->cn_flags |= HASBUF;
2254 		else
2255 			PNBUF_PUT(cnp->cn_pnbuf);
2256 		return (0);
2257 	} else {
2258 		if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
2259 			VOP_UNLOCK(ndp->ni_dvp, 0);
2260 		if (!pubflag) {
2261 			error = EINVAL;
2262 			break;
2263 		}
2264 
2265 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
2266 			error = ELOOP;
2267 			break;
2268 		}
2269 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
2270 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred,
2271 			    cnp->cn_lwp);
2272 			if (error != 0)
2273 				break;
2274 		}
2275 		if (ndp->ni_pathlen > 1)
2276 			cp = PNBUF_GET();
2277 		else
2278 			cp = cnp->cn_pnbuf;
2279 		aiov.iov_base = cp;
2280 		aiov.iov_len = MAXPATHLEN;
2281 		auio.uio_iov = &aiov;
2282 		auio.uio_iovcnt = 1;
2283 		auio.uio_offset = 0;
2284 		auio.uio_rw = UIO_READ;
2285 		auio.uio_resid = MAXPATHLEN;
2286 		UIO_SETUP_SYSSPACE(&auio);
2287 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
2288 		if (error) {
2289 		badlink:
2290 			if (ndp->ni_pathlen > 1)
2291 				PNBUF_PUT(cp);
2292 			break;
2293 		}
2294 		linklen = MAXPATHLEN - auio.uio_resid;
2295 		if (linklen == 0) {
2296 			error = ENOENT;
2297 			goto badlink;
2298 		}
2299 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
2300 			error = ENAMETOOLONG;
2301 			goto badlink;
2302 		}
2303 		if (ndp->ni_pathlen > 1) {
2304 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
2305 			PNBUF_PUT(cnp->cn_pnbuf);
2306 			cnp->cn_pnbuf = cp;
2307 		} else
2308 			cnp->cn_pnbuf[linklen] = '\0';
2309 		ndp->ni_pathlen += linklen;
2310 		vput(ndp->ni_vp);
2311 		dp = ndp->ni_dvp;
2312 		/*
2313 		 * Check if root directory should replace current directory.
2314 		 */
2315 		if (cnp->cn_pnbuf[0] == '/') {
2316 			vrele(dp);
2317 			dp = ndp->ni_rootdir;
2318 			VREF(dp);
2319 		}
2320 	}
2321    }
2322 	vrele(ndp->ni_dvp);
2323 	vput(ndp->ni_vp);
2324 	ndp->ni_vp = NULL;
2325 out:
2326 	PNBUF_PUT(cnp->cn_pnbuf);
2327 	return (error);
2328 }
2329 #endif /* NFSSERVER */
2330 
2331 /*
2332  * A fiddled version of m_adj() that ensures null fill to a 32-bit
2333  * boundary and only trims off the back end
2334  *
2335  * 1. trim off 'len' bytes as m_adj(mp, -len).
2336  * 2. add zero-padding 'nul' bytes at the end of the mbuf chain.
2337  */
2338 void
2339 nfs_zeropad(mp, len, nul)
2340 	struct mbuf *mp;
2341 	int len;
2342 	int nul;
2343 {
2344 	struct mbuf *m;
2345 	int count;
2346 
2347 	/*
2348 	 * Trim from tail.  Scan the mbuf chain,
2349 	 * calculating its length and finding the last mbuf.
2350 	 * If the adjustment only affects this mbuf, then just
2351 	 * adjust and return.  Otherwise, rescan and truncate
2352 	 * after the remaining size.
2353 	 */
2354 	count = 0;
2355 	m = mp;
2356 	for (;;) {
2357 		count += m->m_len;
2358 		if (m->m_next == NULL)
2359 			break;
2360 		m = m->m_next;
2361 	}
2362 
2363 	KDASSERT(count >= len);
2364 
2365 	if (m->m_len >= len) {
2366 		m->m_len -= len;
2367 	} else {
2368 		count -= len;
2369 		/*
2370 		 * Correct length for chain is "count".
2371 		 * Find the mbuf with last data, adjust its length,
2372 		 * and toss data from remaining mbufs on chain.
2373 		 */
2374 		for (m = mp; m; m = m->m_next) {
2375 			if (m->m_len >= count) {
2376 				m->m_len = count;
2377 				break;
2378 			}
2379 			count -= m->m_len;
2380 		}
2381 		KASSERT(m && m->m_next);
2382 		m_freem(m->m_next);
2383 		m->m_next = NULL;
2384 	}
2385 
2386 	KDASSERT(m->m_next == NULL);
2387 
2388 	/*
2389 	 * zero-padding.
2390 	 */
2391 	if (nul > 0) {
2392 		char *cp;
2393 		int i;
2394 
2395 		if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) {
2396 			struct mbuf *n;
2397 
2398 			KDASSERT(MLEN >= nul);
2399 			n = m_get(M_WAIT, MT_DATA);
2400 			MCLAIM(n, &nfs_mowner);
2401 			n->m_len = nul;
2402 			n->m_next = NULL;
2403 			m->m_next = n;
2404 			cp = mtod(n, caddr_t);
2405 		} else {
2406 			cp = mtod(m, caddr_t) + m->m_len;
2407 			m->m_len += nul;
2408 		}
2409 		for (i = 0; i < nul; i++)
2410 			*cp++ = '\0';
2411 	}
2412 	return;
2413 }
2414 
2415 /*
2416  * Make these functions instead of macros, so that the kernel text size
2417  * doesn't get too big...
2418  */
2419 void
2420 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp)
2421 	struct nfsrv_descript *nfsd;
2422 	int before_ret;
2423 	struct vattr *before_vap;
2424 	int after_ret;
2425 	struct vattr *after_vap;
2426 	struct mbuf **mbp;
2427 	char **bposp;
2428 {
2429 	struct mbuf *mb = *mbp;
2430 	char *bpos = *bposp;
2431 	u_int32_t *tl;
2432 
2433 	if (before_ret) {
2434 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2435 		*tl = nfs_false;
2436 	} else {
2437 		nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
2438 		*tl++ = nfs_true;
2439 		txdr_hyper(before_vap->va_size, tl);
2440 		tl += 2;
2441 		txdr_nfsv3time(&(before_vap->va_mtime), tl);
2442 		tl += 2;
2443 		txdr_nfsv3time(&(before_vap->va_ctime), tl);
2444 	}
2445 	*bposp = bpos;
2446 	*mbp = mb;
2447 	nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
2448 }
2449 
2450 void
2451 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp)
2452 	struct nfsrv_descript *nfsd;
2453 	int after_ret;
2454 	struct vattr *after_vap;
2455 	struct mbuf **mbp;
2456 	char **bposp;
2457 {
2458 	struct mbuf *mb = *mbp;
2459 	char *bpos = *bposp;
2460 	u_int32_t *tl;
2461 	struct nfs_fattr *fp;
2462 
2463 	if (after_ret) {
2464 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2465 		*tl = nfs_false;
2466 	} else {
2467 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
2468 		*tl++ = nfs_true;
2469 		fp = (struct nfs_fattr *)tl;
2470 		nfsm_srvfattr(nfsd, after_vap, fp);
2471 	}
2472 	*mbp = mb;
2473 	*bposp = bpos;
2474 }
2475 
2476 void
2477 nfsm_srvfattr(nfsd, vap, fp)
2478 	struct nfsrv_descript *nfsd;
2479 	struct vattr *vap;
2480 	struct nfs_fattr *fp;
2481 {
2482 
2483 	fp->fa_nlink = txdr_unsigned(vap->va_nlink);
2484 	fp->fa_uid = txdr_unsigned(vap->va_uid);
2485 	fp->fa_gid = txdr_unsigned(vap->va_gid);
2486 	if (nfsd->nd_flag & ND_NFSV3) {
2487 		fp->fa_type = vtonfsv3_type(vap->va_type);
2488 		fp->fa_mode = vtonfsv3_mode(vap->va_mode);
2489 		txdr_hyper(vap->va_size, &fp->fa3_size);
2490 		txdr_hyper(vap->va_bytes, &fp->fa3_used);
2491 		fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
2492 		fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
2493 		fp->fa3_fsid.nfsuquad[0] = 0;
2494 		fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
2495 		txdr_hyper(vap->va_fileid, &fp->fa3_fileid);
2496 		txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
2497 		txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
2498 		txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
2499 	} else {
2500 		fp->fa_type = vtonfsv2_type(vap->va_type);
2501 		fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
2502 		fp->fa2_size = txdr_unsigned(vap->va_size);
2503 		fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
2504 		if (vap->va_type == VFIFO)
2505 			fp->fa2_rdev = 0xffffffff;
2506 		else
2507 			fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
2508 		fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
2509 		fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
2510 		fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
2511 		txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
2512 		txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
2513 		txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
2514 	}
2515 }
2516 
2517 #ifdef NFSSERVER
2518 /*
2519  * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
2520  * 	- look up fsid in mount list (if not found ret error)
2521  *	- get vp and export rights by calling VFS_FHTOVP()
2522  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2523  *	- if not lockflag unlock it with VOP_UNLOCK()
2524  */
2525 int
2526 nfsrv_fhtovp(fhp, lockflag, vpp, cred, slp, nam, rdonlyp, kerbflag, pubflag)
2527 	fhandle_t *fhp;
2528 	int lockflag;
2529 	struct vnode **vpp;
2530 	kauth_cred_t cred;
2531 	struct nfssvc_sock *slp;
2532 	struct mbuf *nam;
2533 	int *rdonlyp;
2534 	int kerbflag;
2535 {
2536 	struct mount *mp;
2537 	kauth_cred_t credanon;
2538 	int error, exflags;
2539 	struct sockaddr_in *saddr;
2540 
2541 	*vpp = (struct vnode *)0;
2542 
2543 	if (nfs_ispublicfh(fhp)) {
2544 		if (!pubflag || !nfs_pub.np_valid)
2545 			return (ESTALE);
2546 		fhp = &nfs_pub.np_handle;
2547 	}
2548 
2549 	error = netexport_check(&fhp->fh_fsid, nam, &mp, &exflags, &credanon);
2550 	if (error) {
2551 		return error;
2552 	}
2553 
2554 	error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
2555 	if (error)
2556 		return (error);
2557 
2558 	if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
2559 		saddr = mtod(nam, struct sockaddr_in *);
2560 		if ((saddr->sin_family == AF_INET) &&
2561 		    ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
2562 			vput(*vpp);
2563 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2564 		}
2565 #ifdef INET6
2566 		if ((saddr->sin_family == AF_INET6) &&
2567 		    ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) {
2568 			vput(*vpp);
2569 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2570 		}
2571 #endif
2572 	}
2573 	/*
2574 	 * Check/setup credentials.
2575 	 */
2576 	if (exflags & MNT_EXKERB) {
2577 		if (!kerbflag) {
2578 			vput(*vpp);
2579 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2580 		}
2581 	} else if (kerbflag) {
2582 		vput(*vpp);
2583 		return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2584 	} else if (kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER,
2585 		    NULL) == 0 || (exflags & MNT_EXPORTANON)) {
2586 		kauth_cred_clone(credanon, cred);
2587 	}
2588 	if (exflags & MNT_EXRDONLY)
2589 		*rdonlyp = 1;
2590 	else
2591 		*rdonlyp = 0;
2592 	if (!lockflag)
2593 		VOP_UNLOCK(*vpp, 0);
2594 	return (0);
2595 }
2596 
2597 /*
2598  * WebNFS: check if a filehandle is a public filehandle. For v3, this
2599  * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
2600  * transformed this to all zeroes in both cases, so check for it.
2601  */
2602 int
2603 nfs_ispublicfh(fhp)
2604 	fhandle_t *fhp;
2605 {
2606 	char *cp = (char *)fhp;
2607 	int i;
2608 
2609 	for (i = 0; i < NFSX_V3FH; i++)
2610 		if (*cp++ != 0)
2611 			return (FALSE);
2612 	return (TRUE);
2613 }
2614 #endif /* NFSSERVER */
2615 
2616 /*
2617  * This function compares two net addresses by family and returns TRUE
2618  * if they are the same host.
2619  * If there is any doubt, return FALSE.
2620  * The AF_INET family is handled as a special case so that address mbufs
2621  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
2622  */
2623 int
2624 netaddr_match(family, haddr, nam)
2625 	int family;
2626 	union nethostaddr *haddr;
2627 	struct mbuf *nam;
2628 {
2629 	struct sockaddr_in *inetaddr;
2630 
2631 	switch (family) {
2632 	case AF_INET:
2633 		inetaddr = mtod(nam, struct sockaddr_in *);
2634 		if (inetaddr->sin_family == AF_INET &&
2635 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
2636 			return (1);
2637 		break;
2638 #ifdef INET6
2639 	case AF_INET6:
2640 	    {
2641 		struct sockaddr_in6 *sin6_1, *sin6_2;
2642 
2643 		sin6_1 = mtod(nam, struct sockaddr_in6 *);
2644 		sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
2645 		if (sin6_1->sin6_family == AF_INET6 &&
2646 		    IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
2647 			return 1;
2648 	    }
2649 #endif
2650 #ifdef ISO
2651 	case AF_ISO:
2652 	    {
2653 		struct sockaddr_iso *isoaddr1, *isoaddr2;
2654 
2655 		isoaddr1 = mtod(nam, struct sockaddr_iso *);
2656 		isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *);
2657 		if (isoaddr1->siso_family == AF_ISO &&
2658 		    isoaddr1->siso_nlen > 0 &&
2659 		    isoaddr1->siso_nlen == isoaddr2->siso_nlen &&
2660 		    SAME_ISOADDR(isoaddr1, isoaddr2))
2661 			return (1);
2662 		break;
2663 	    }
2664 #endif	/* ISO */
2665 	default:
2666 		break;
2667 	};
2668 	return (0);
2669 }
2670 
2671 /*
2672  * The write verifier has changed (probably due to a server reboot), so all
2673  * PG_NEEDCOMMIT pages will have to be written again. Since they are marked
2674  * as dirty or are being written out just now, all this takes is clearing
2675  * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for
2676  * the mount point.
2677  */
2678 void
2679 nfs_clearcommit(mp)
2680 	struct mount *mp;
2681 {
2682 	struct vnode *vp;
2683 	struct nfsnode *np;
2684 	struct vm_page *pg;
2685 	struct nfsmount *nmp = VFSTONFS(mp);
2686 
2687 	lockmgr(&nmp->nm_writeverflock, LK_EXCLUSIVE, NULL);
2688 
2689 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2690 		KASSERT(vp->v_mount == mp);
2691 		if (vp->v_type != VREG)
2692 			continue;
2693 		np = VTONFS(vp);
2694 		np->n_pushlo = np->n_pushhi = np->n_pushedlo =
2695 		    np->n_pushedhi = 0;
2696 		np->n_commitflags &=
2697 		    ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
2698 		simple_lock(&vp->v_uobj.vmobjlock);
2699 		TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) {
2700 			pg->flags &= ~PG_NEEDCOMMIT;
2701 		}
2702 		simple_unlock(&vp->v_uobj.vmobjlock);
2703 	}
2704 	simple_lock(&nmp->nm_slock);
2705 	nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF;
2706 	simple_unlock(&nmp->nm_slock);
2707 	lockmgr(&nmp->nm_writeverflock, LK_RELEASE, NULL);
2708 }
2709 
2710 void
2711 nfs_merge_commit_ranges(vp)
2712 	struct vnode *vp;
2713 {
2714 	struct nfsnode *np = VTONFS(vp);
2715 
2716 	KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID);
2717 
2718 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2719 		np->n_pushedlo = np->n_pushlo;
2720 		np->n_pushedhi = np->n_pushhi;
2721 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2722 	} else {
2723 		if (np->n_pushlo < np->n_pushedlo)
2724 			np->n_pushedlo = np->n_pushlo;
2725 		if (np->n_pushhi > np->n_pushedhi)
2726 			np->n_pushedhi = np->n_pushhi;
2727 	}
2728 
2729 	np->n_pushlo = np->n_pushhi = 0;
2730 	np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
2731 
2732 #ifdef NFS_DEBUG_COMMIT
2733 	printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2734 	    (unsigned)np->n_pushedhi);
2735 #endif
2736 }
2737 
2738 int
2739 nfs_in_committed_range(vp, off, len)
2740 	struct vnode *vp;
2741 	off_t off, len;
2742 {
2743 	struct nfsnode *np = VTONFS(vp);
2744 	off_t lo, hi;
2745 
2746 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2747 		return 0;
2748 	lo = off;
2749 	hi = lo + len;
2750 
2751 	return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
2752 }
2753 
2754 int
2755 nfs_in_tobecommitted_range(vp, off, len)
2756 	struct vnode *vp;
2757 	off_t off, len;
2758 {
2759 	struct nfsnode *np = VTONFS(vp);
2760 	off_t lo, hi;
2761 
2762 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2763 		return 0;
2764 	lo = off;
2765 	hi = lo + len;
2766 
2767 	return (lo >= np->n_pushlo && hi <= np->n_pushhi);
2768 }
2769 
2770 void
2771 nfs_add_committed_range(vp, off, len)
2772 	struct vnode *vp;
2773 	off_t off, len;
2774 {
2775 	struct nfsnode *np = VTONFS(vp);
2776 	off_t lo, hi;
2777 
2778 	lo = off;
2779 	hi = lo + len;
2780 
2781 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2782 		np->n_pushedlo = lo;
2783 		np->n_pushedhi = hi;
2784 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2785 	} else {
2786 		if (hi > np->n_pushedhi)
2787 			np->n_pushedhi = hi;
2788 		if (lo < np->n_pushedlo)
2789 			np->n_pushedlo = lo;
2790 	}
2791 #ifdef NFS_DEBUG_COMMIT
2792 	printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2793 	    (unsigned)np->n_pushedhi);
2794 #endif
2795 }
2796 
2797 void
2798 nfs_del_committed_range(vp, off, len)
2799 	struct vnode *vp;
2800 	off_t off, len;
2801 {
2802 	struct nfsnode *np = VTONFS(vp);
2803 	off_t lo, hi;
2804 
2805 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2806 		return;
2807 
2808 	lo = off;
2809 	hi = lo + len;
2810 
2811 	if (lo > np->n_pushedhi || hi < np->n_pushedlo)
2812 		return;
2813 	if (lo <= np->n_pushedlo)
2814 		np->n_pushedlo = hi;
2815 	else if (hi >= np->n_pushedhi)
2816 		np->n_pushedhi = lo;
2817 	else {
2818 		/*
2819 		 * XXX There's only one range. If the deleted range
2820 		 * is in the middle, pick the largest of the
2821 		 * contiguous ranges that it leaves.
2822 		 */
2823 		if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
2824 			np->n_pushedhi = lo;
2825 		else
2826 			np->n_pushedlo = hi;
2827 	}
2828 #ifdef NFS_DEBUG_COMMIT
2829 	printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2830 	    (unsigned)np->n_pushedhi);
2831 #endif
2832 }
2833 
2834 void
2835 nfs_add_tobecommitted_range(vp, off, len)
2836 	struct vnode *vp;
2837 	off_t off, len;
2838 {
2839 	struct nfsnode *np = VTONFS(vp);
2840 	off_t lo, hi;
2841 
2842 	lo = off;
2843 	hi = lo + len;
2844 
2845 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
2846 		np->n_pushlo = lo;
2847 		np->n_pushhi = hi;
2848 		np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
2849 	} else {
2850 		if (lo < np->n_pushlo)
2851 			np->n_pushlo = lo;
2852 		if (hi > np->n_pushhi)
2853 			np->n_pushhi = hi;
2854 	}
2855 #ifdef NFS_DEBUG_COMMIT
2856 	printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2857 	    (unsigned)np->n_pushhi);
2858 #endif
2859 }
2860 
2861 void
2862 nfs_del_tobecommitted_range(vp, off, len)
2863 	struct vnode *vp;
2864 	off_t off, len;
2865 {
2866 	struct nfsnode *np = VTONFS(vp);
2867 	off_t lo, hi;
2868 
2869 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2870 		return;
2871 
2872 	lo = off;
2873 	hi = lo + len;
2874 
2875 	if (lo > np->n_pushhi || hi < np->n_pushlo)
2876 		return;
2877 
2878 	if (lo <= np->n_pushlo)
2879 		np->n_pushlo = hi;
2880 	else if (hi >= np->n_pushhi)
2881 		np->n_pushhi = lo;
2882 	else {
2883 		/*
2884 		 * XXX There's only one range. If the deleted range
2885 		 * is in the middle, pick the largest of the
2886 		 * contiguous ranges that it leaves.
2887 		 */
2888 		if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
2889 			np->n_pushhi = lo;
2890 		else
2891 			np->n_pushlo = hi;
2892 	}
2893 #ifdef NFS_DEBUG_COMMIT
2894 	printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2895 	    (unsigned)np->n_pushhi);
2896 #endif
2897 }
2898 
2899 /*
2900  * Map errnos to NFS error numbers. For Version 3 also filter out error
2901  * numbers not specified for the associated procedure.
2902  */
2903 int
2904 nfsrv_errmap(nd, err)
2905 	struct nfsrv_descript *nd;
2906 	int err;
2907 {
2908 	const short *defaulterrp, *errp;
2909 
2910 	if (nd->nd_flag & ND_NFSV3) {
2911 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
2912 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2913 		while (*++errp) {
2914 			if (*errp == err)
2915 				return (err);
2916 			else if (*errp > err)
2917 				break;
2918 		}
2919 		return ((int)*defaulterrp);
2920 	    } else
2921 		return (err & 0xffff);
2922 	}
2923 	if (err <= ELAST)
2924 		return ((int)nfsrv_v2errmap[err - 1]);
2925 	return (NFSERR_IO);
2926 }
2927 
2928 u_int32_t
2929 nfs_getxid()
2930 {
2931 	static u_int32_t base;
2932 	static u_int32_t nfs_xid = 0;
2933 	static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER;
2934 	u_int32_t newxid;
2935 
2936 	simple_lock(&nfs_xidlock);
2937 	/*
2938 	 * derive initial xid from system time
2939 	 * XXX time is invalid if root not yet mounted
2940 	 */
2941 	if (__predict_false(!base && (rootvp))) {
2942 		struct timeval tv;
2943 
2944 		microtime(&tv);
2945 		base = tv.tv_sec << 12;
2946 		nfs_xid = base;
2947 	}
2948 
2949 	/*
2950 	 * Skip zero xid if it should ever happen.
2951 	 */
2952 	if (__predict_false(++nfs_xid == 0))
2953 		nfs_xid++;
2954 	newxid = nfs_xid;
2955 	simple_unlock(&nfs_xidlock);
2956 
2957 	return txdr_unsigned(newxid);
2958 }
2959 
2960 /*
2961  * assign a new xid for existing request.
2962  * used for NFSERR_JUKEBOX handling.
2963  */
2964 void
2965 nfs_renewxid(struct nfsreq *req)
2966 {
2967 	u_int32_t xid;
2968 	int off;
2969 
2970 	xid = nfs_getxid();
2971 	if (req->r_nmp->nm_sotype == SOCK_STREAM)
2972 		off = sizeof(u_int32_t); /* RPC record mark */
2973 	else
2974 		off = 0;
2975 
2976 	m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid);
2977 	req->r_xid = xid;
2978 }
2979