xref: /netbsd/sys/nfs/nfs_subs.c (revision c4a72b64)
1 /*	$NetBSD: nfs_subs.c,v 1.107 2002/12/01 23:02:11 matt 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. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)nfs_subs.c	8.8 (Berkeley) 5/22/95
39  */
40 
41 /*
42  * Copyright 2000 Wasabi Systems, Inc.
43  * All rights reserved.
44  *
45  * Written by Frank van der Linden for Wasabi Systems, Inc.
46  *
47  * Redistribution and use in source and binary forms, with or without
48  * modification, are permitted provided that the following conditions
49  * are met:
50  * 1. Redistributions of source code must retain the above copyright
51  *    notice, this list of conditions and the following disclaimer.
52  * 2. Redistributions in binary form must reproduce the above copyright
53  *    notice, this list of conditions and the following disclaimer in the
54  *    documentation and/or other materials provided with the distribution.
55  * 3. All advertising materials mentioning features or use of this software
56  *    must display the following acknowledgement:
57  *      This product includes software developed for the NetBSD Project by
58  *      Wasabi Systems, Inc.
59  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
60  *    or promote products derived from this software without specific prior
61  *    written permission.
62  *
63  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
64  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
65  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
66  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
67  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
68  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
69  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
70  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
71  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
72  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
73  * POSSIBILITY OF SUCH DAMAGE.
74  */
75 
76 #include <sys/cdefs.h>
77 __KERNEL_RCSID(0, "$NetBSD: nfs_subs.c,v 1.107 2002/12/01 23:02:11 matt Exp $");
78 
79 #include "fs_nfs.h"
80 #include "opt_nfs.h"
81 #include "opt_nfsserver.h"
82 #include "opt_iso.h"
83 #include "opt_inet.h"
84 
85 /*
86  * These functions support the macros and help fiddle mbuf chains for
87  * the nfs op functions. They do things like create the rpc header and
88  * copy data between mbuf chains and uio lists.
89  */
90 #include <sys/param.h>
91 #include <sys/proc.h>
92 #include <sys/systm.h>
93 #include <sys/kernel.h>
94 #include <sys/mount.h>
95 #include <sys/vnode.h>
96 #include <sys/namei.h>
97 #include <sys/mbuf.h>
98 #include <sys/socket.h>
99 #include <sys/stat.h>
100 #include <sys/malloc.h>
101 #include <sys/filedesc.h>
102 #include <sys/time.h>
103 #include <sys/dirent.h>
104 
105 #include <uvm/uvm_extern.h>
106 
107 #include <nfs/rpcv2.h>
108 #include <nfs/nfsproto.h>
109 #include <nfs/nfsnode.h>
110 #include <nfs/nfs.h>
111 #include <nfs/xdr_subs.h>
112 #include <nfs/nfsm_subs.h>
113 #include <nfs/nfsmount.h>
114 #include <nfs/nqnfs.h>
115 #include <nfs/nfsrtt.h>
116 #include <nfs/nfs_var.h>
117 
118 #include <miscfs/specfs/specdev.h>
119 
120 #include <netinet/in.h>
121 #ifdef ISO
122 #include <netiso/iso.h>
123 #endif
124 
125 /*
126  * Data items converted to xdr at startup, since they are constant
127  * This is kinda hokey, but may save a little time doing byte swaps
128  */
129 u_int32_t nfs_xdrneg1;
130 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr,
131 	rpc_mismatch, rpc_auth_unix, rpc_msgaccepted,
132 	rpc_auth_kerb;
133 u_int32_t nfs_prog, nqnfs_prog, nfs_true, nfs_false;
134 
135 /* And other global data */
136 static u_int32_t nfs_xid = 0;
137 const nfstype nfsv2_type[9] =
138 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON };
139 const nfstype nfsv3_type[9] =
140 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON };
141 const enum vtype nv2tov_type[8] =
142 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON };
143 const enum vtype nv3tov_type[8] =
144 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO };
145 int nfs_ticks;
146 int nfs_commitsize;
147 
148 /* NFS client/server stats. */
149 struct nfsstats nfsstats;
150 
151 /*
152  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
153  */
154 const int nfsv3_procid[NFS_NPROCS] = {
155 	NFSPROC_NULL,
156 	NFSPROC_GETATTR,
157 	NFSPROC_SETATTR,
158 	NFSPROC_NOOP,
159 	NFSPROC_LOOKUP,
160 	NFSPROC_READLINK,
161 	NFSPROC_READ,
162 	NFSPROC_NOOP,
163 	NFSPROC_WRITE,
164 	NFSPROC_CREATE,
165 	NFSPROC_REMOVE,
166 	NFSPROC_RENAME,
167 	NFSPROC_LINK,
168 	NFSPROC_SYMLINK,
169 	NFSPROC_MKDIR,
170 	NFSPROC_RMDIR,
171 	NFSPROC_READDIR,
172 	NFSPROC_FSSTAT,
173 	NFSPROC_NOOP,
174 	NFSPROC_NOOP,
175 	NFSPROC_NOOP,
176 	NFSPROC_NOOP,
177 	NFSPROC_NOOP,
178 	NFSPROC_NOOP,
179 	NFSPROC_NOOP,
180 	NFSPROC_NOOP
181 };
182 
183 /*
184  * and the reverse mapping from generic to Version 2 procedure numbers
185  */
186 const int nfsv2_procid[NFS_NPROCS] = {
187 	NFSV2PROC_NULL,
188 	NFSV2PROC_GETATTR,
189 	NFSV2PROC_SETATTR,
190 	NFSV2PROC_LOOKUP,
191 	NFSV2PROC_NOOP,
192 	NFSV2PROC_READLINK,
193 	NFSV2PROC_READ,
194 	NFSV2PROC_WRITE,
195 	NFSV2PROC_CREATE,
196 	NFSV2PROC_MKDIR,
197 	NFSV2PROC_SYMLINK,
198 	NFSV2PROC_CREATE,
199 	NFSV2PROC_REMOVE,
200 	NFSV2PROC_RMDIR,
201 	NFSV2PROC_RENAME,
202 	NFSV2PROC_LINK,
203 	NFSV2PROC_READDIR,
204 	NFSV2PROC_NOOP,
205 	NFSV2PROC_STATFS,
206 	NFSV2PROC_NOOP,
207 	NFSV2PROC_NOOP,
208 	NFSV2PROC_NOOP,
209 	NFSV2PROC_NOOP,
210 	NFSV2PROC_NOOP,
211 	NFSV2PROC_NOOP,
212 	NFSV2PROC_NOOP,
213 };
214 
215 /*
216  * Maps errno values to nfs error numbers.
217  * Use NFSERR_IO as the catch all for ones not specifically defined in
218  * RFC 1094.
219  */
220 static const u_char nfsrv_v2errmap[ELAST] = {
221   NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
222   NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
223   NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
224   NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
225   NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
226   NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
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_IO,	NFSERR_IO,	NFSERR_IO,
233   NFSERR_IO,	NFSERR_IO,	NFSERR_NAMETOL,	NFSERR_IO,	NFSERR_IO,
234   NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
235   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
236   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
237   NFSERR_IO,	NFSERR_IO,
238 };
239 
240 /*
241  * Maps errno values to nfs error numbers.
242  * Although it is not obvious whether or not NFS clients really care if
243  * a returned error value is in the specified list for the procedure, the
244  * safest thing to do is filter them appropriately. For Version 2, the
245  * X/Open XNFS document is the only specification that defines error values
246  * for each RPC (The RFC simply lists all possible error values for all RPCs),
247  * so I have decided to not do this for Version 2.
248  * The first entry is the default error return and the rest are the valid
249  * errors for that RPC in increasing numeric order.
250  */
251 static const short nfsv3err_null[] = {
252 	0,
253 	0,
254 };
255 
256 static const short nfsv3err_getattr[] = {
257 	NFSERR_IO,
258 	NFSERR_IO,
259 	NFSERR_STALE,
260 	NFSERR_BADHANDLE,
261 	NFSERR_SERVERFAULT,
262 	0,
263 };
264 
265 static const short nfsv3err_setattr[] = {
266 	NFSERR_IO,
267 	NFSERR_PERM,
268 	NFSERR_IO,
269 	NFSERR_ACCES,
270 	NFSERR_INVAL,
271 	NFSERR_NOSPC,
272 	NFSERR_ROFS,
273 	NFSERR_DQUOT,
274 	NFSERR_STALE,
275 	NFSERR_BADHANDLE,
276 	NFSERR_NOT_SYNC,
277 	NFSERR_SERVERFAULT,
278 	0,
279 };
280 
281 static const short nfsv3err_lookup[] = {
282 	NFSERR_IO,
283 	NFSERR_NOENT,
284 	NFSERR_IO,
285 	NFSERR_ACCES,
286 	NFSERR_NOTDIR,
287 	NFSERR_NAMETOL,
288 	NFSERR_STALE,
289 	NFSERR_BADHANDLE,
290 	NFSERR_SERVERFAULT,
291 	0,
292 };
293 
294 static const short nfsv3err_access[] = {
295 	NFSERR_IO,
296 	NFSERR_IO,
297 	NFSERR_STALE,
298 	NFSERR_BADHANDLE,
299 	NFSERR_SERVERFAULT,
300 	0,
301 };
302 
303 static const short nfsv3err_readlink[] = {
304 	NFSERR_IO,
305 	NFSERR_IO,
306 	NFSERR_ACCES,
307 	NFSERR_INVAL,
308 	NFSERR_STALE,
309 	NFSERR_BADHANDLE,
310 	NFSERR_NOTSUPP,
311 	NFSERR_SERVERFAULT,
312 	0,
313 };
314 
315 static const short nfsv3err_read[] = {
316 	NFSERR_IO,
317 	NFSERR_IO,
318 	NFSERR_NXIO,
319 	NFSERR_ACCES,
320 	NFSERR_INVAL,
321 	NFSERR_STALE,
322 	NFSERR_BADHANDLE,
323 	NFSERR_SERVERFAULT,
324 	NFSERR_JUKEBOX,
325 	0,
326 };
327 
328 static const short nfsv3err_write[] = {
329 	NFSERR_IO,
330 	NFSERR_IO,
331 	NFSERR_ACCES,
332 	NFSERR_INVAL,
333 	NFSERR_FBIG,
334 	NFSERR_NOSPC,
335 	NFSERR_ROFS,
336 	NFSERR_DQUOT,
337 	NFSERR_STALE,
338 	NFSERR_BADHANDLE,
339 	NFSERR_SERVERFAULT,
340 	NFSERR_JUKEBOX,
341 	0,
342 };
343 
344 static const short nfsv3err_create[] = {
345 	NFSERR_IO,
346 	NFSERR_IO,
347 	NFSERR_ACCES,
348 	NFSERR_EXIST,
349 	NFSERR_NOTDIR,
350 	NFSERR_NOSPC,
351 	NFSERR_ROFS,
352 	NFSERR_NAMETOL,
353 	NFSERR_DQUOT,
354 	NFSERR_STALE,
355 	NFSERR_BADHANDLE,
356 	NFSERR_NOTSUPP,
357 	NFSERR_SERVERFAULT,
358 	0,
359 };
360 
361 static const short nfsv3err_mkdir[] = {
362 	NFSERR_IO,
363 	NFSERR_IO,
364 	NFSERR_ACCES,
365 	NFSERR_EXIST,
366 	NFSERR_NOTDIR,
367 	NFSERR_NOSPC,
368 	NFSERR_ROFS,
369 	NFSERR_NAMETOL,
370 	NFSERR_DQUOT,
371 	NFSERR_STALE,
372 	NFSERR_BADHANDLE,
373 	NFSERR_NOTSUPP,
374 	NFSERR_SERVERFAULT,
375 	0,
376 };
377 
378 static const short nfsv3err_symlink[] = {
379 	NFSERR_IO,
380 	NFSERR_IO,
381 	NFSERR_ACCES,
382 	NFSERR_EXIST,
383 	NFSERR_NOTDIR,
384 	NFSERR_NOSPC,
385 	NFSERR_ROFS,
386 	NFSERR_NAMETOL,
387 	NFSERR_DQUOT,
388 	NFSERR_STALE,
389 	NFSERR_BADHANDLE,
390 	NFSERR_NOTSUPP,
391 	NFSERR_SERVERFAULT,
392 	0,
393 };
394 
395 static const short nfsv3err_mknod[] = {
396 	NFSERR_IO,
397 	NFSERR_IO,
398 	NFSERR_ACCES,
399 	NFSERR_EXIST,
400 	NFSERR_NOTDIR,
401 	NFSERR_NOSPC,
402 	NFSERR_ROFS,
403 	NFSERR_NAMETOL,
404 	NFSERR_DQUOT,
405 	NFSERR_STALE,
406 	NFSERR_BADHANDLE,
407 	NFSERR_NOTSUPP,
408 	NFSERR_SERVERFAULT,
409 	NFSERR_BADTYPE,
410 	0,
411 };
412 
413 static const short nfsv3err_remove[] = {
414 	NFSERR_IO,
415 	NFSERR_NOENT,
416 	NFSERR_IO,
417 	NFSERR_ACCES,
418 	NFSERR_NOTDIR,
419 	NFSERR_ROFS,
420 	NFSERR_NAMETOL,
421 	NFSERR_STALE,
422 	NFSERR_BADHANDLE,
423 	NFSERR_SERVERFAULT,
424 	0,
425 };
426 
427 static const short nfsv3err_rmdir[] = {
428 	NFSERR_IO,
429 	NFSERR_NOENT,
430 	NFSERR_IO,
431 	NFSERR_ACCES,
432 	NFSERR_EXIST,
433 	NFSERR_NOTDIR,
434 	NFSERR_INVAL,
435 	NFSERR_ROFS,
436 	NFSERR_NAMETOL,
437 	NFSERR_NOTEMPTY,
438 	NFSERR_STALE,
439 	NFSERR_BADHANDLE,
440 	NFSERR_NOTSUPP,
441 	NFSERR_SERVERFAULT,
442 	0,
443 };
444 
445 static const short nfsv3err_rename[] = {
446 	NFSERR_IO,
447 	NFSERR_NOENT,
448 	NFSERR_IO,
449 	NFSERR_ACCES,
450 	NFSERR_EXIST,
451 	NFSERR_XDEV,
452 	NFSERR_NOTDIR,
453 	NFSERR_ISDIR,
454 	NFSERR_INVAL,
455 	NFSERR_NOSPC,
456 	NFSERR_ROFS,
457 	NFSERR_MLINK,
458 	NFSERR_NAMETOL,
459 	NFSERR_NOTEMPTY,
460 	NFSERR_DQUOT,
461 	NFSERR_STALE,
462 	NFSERR_BADHANDLE,
463 	NFSERR_NOTSUPP,
464 	NFSERR_SERVERFAULT,
465 	0,
466 };
467 
468 static const short nfsv3err_link[] = {
469 	NFSERR_IO,
470 	NFSERR_IO,
471 	NFSERR_ACCES,
472 	NFSERR_EXIST,
473 	NFSERR_XDEV,
474 	NFSERR_NOTDIR,
475 	NFSERR_INVAL,
476 	NFSERR_NOSPC,
477 	NFSERR_ROFS,
478 	NFSERR_MLINK,
479 	NFSERR_NAMETOL,
480 	NFSERR_DQUOT,
481 	NFSERR_STALE,
482 	NFSERR_BADHANDLE,
483 	NFSERR_NOTSUPP,
484 	NFSERR_SERVERFAULT,
485 	0,
486 };
487 
488 static const short nfsv3err_readdir[] = {
489 	NFSERR_IO,
490 	NFSERR_IO,
491 	NFSERR_ACCES,
492 	NFSERR_NOTDIR,
493 	NFSERR_STALE,
494 	NFSERR_BADHANDLE,
495 	NFSERR_BAD_COOKIE,
496 	NFSERR_TOOSMALL,
497 	NFSERR_SERVERFAULT,
498 	0,
499 };
500 
501 static const short nfsv3err_readdirplus[] = {
502 	NFSERR_IO,
503 	NFSERR_IO,
504 	NFSERR_ACCES,
505 	NFSERR_NOTDIR,
506 	NFSERR_STALE,
507 	NFSERR_BADHANDLE,
508 	NFSERR_BAD_COOKIE,
509 	NFSERR_NOTSUPP,
510 	NFSERR_TOOSMALL,
511 	NFSERR_SERVERFAULT,
512 	0,
513 };
514 
515 static const short nfsv3err_fsstat[] = {
516 	NFSERR_IO,
517 	NFSERR_IO,
518 	NFSERR_STALE,
519 	NFSERR_BADHANDLE,
520 	NFSERR_SERVERFAULT,
521 	0,
522 };
523 
524 static const short nfsv3err_fsinfo[] = {
525 	NFSERR_STALE,
526 	NFSERR_STALE,
527 	NFSERR_BADHANDLE,
528 	NFSERR_SERVERFAULT,
529 	0,
530 };
531 
532 static const short nfsv3err_pathconf[] = {
533 	NFSERR_STALE,
534 	NFSERR_STALE,
535 	NFSERR_BADHANDLE,
536 	NFSERR_SERVERFAULT,
537 	0,
538 };
539 
540 static const short nfsv3err_commit[] = {
541 	NFSERR_IO,
542 	NFSERR_IO,
543 	NFSERR_STALE,
544 	NFSERR_BADHANDLE,
545 	NFSERR_SERVERFAULT,
546 	0,
547 };
548 
549 static const short * const nfsrv_v3errmap[] = {
550 	nfsv3err_null,
551 	nfsv3err_getattr,
552 	nfsv3err_setattr,
553 	nfsv3err_lookup,
554 	nfsv3err_access,
555 	nfsv3err_readlink,
556 	nfsv3err_read,
557 	nfsv3err_write,
558 	nfsv3err_create,
559 	nfsv3err_mkdir,
560 	nfsv3err_symlink,
561 	nfsv3err_mknod,
562 	nfsv3err_remove,
563 	nfsv3err_rmdir,
564 	nfsv3err_rename,
565 	nfsv3err_link,
566 	nfsv3err_readdir,
567 	nfsv3err_readdirplus,
568 	nfsv3err_fsstat,
569 	nfsv3err_fsinfo,
570 	nfsv3err_pathconf,
571 	nfsv3err_commit,
572 };
573 
574 extern struct nfsrtt nfsrtt;
575 extern time_t nqnfsstarttime;
576 extern int nqsrv_clockskew;
577 extern int nqsrv_writeslack;
578 extern int nqsrv_maxlease;
579 extern const int nqnfs_piggy[NFS_NPROCS];
580 extern struct nfsnodehashhead *nfsnodehashtbl;
581 extern u_long nfsnodehash;
582 
583 u_long nfsdirhashmask;
584 
585 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *));
586 
587 /*
588  * Create the header for an rpc request packet
589  * The hsiz is the size of the rest of the nfs request header.
590  * (just used to decide if a cluster is a good idea)
591  */
592 struct mbuf *
593 nfsm_reqh(vp, procid, hsiz, bposp)
594 	struct vnode *vp;
595 	u_long procid;
596 	int hsiz;
597 	caddr_t *bposp;
598 {
599 	struct mbuf *mb;
600 	caddr_t bpos;
601 	struct nfsmount *nmp;
602 #ifndef NFS_V2_ONLY
603 	u_int32_t *tl;
604 	struct mbuf *mb2;
605 	int nqflag;
606 #endif
607 
608 	MGET(mb, M_WAIT, MT_DATA);
609 	if (hsiz >= MINCLSIZE)
610 		MCLGET(mb, M_WAIT);
611 	mb->m_len = 0;
612 	bpos = mtod(mb, caddr_t);
613 
614 	/*
615 	 * For NQNFS, add lease request.
616 	 */
617 	if (vp) {
618 		nmp = VFSTONFS(vp->v_mount);
619 #ifndef NFS_V2_ONLY
620 		if (nmp->nm_flag & NFSMNT_NQNFS) {
621 			nqflag = NQNFS_NEEDLEASE(vp, procid);
622 			if (nqflag) {
623 				nfsm_build(tl, u_int32_t *, 2*NFSX_UNSIGNED);
624 				*tl++ = txdr_unsigned(nqflag);
625 				*tl = txdr_unsigned(nmp->nm_leaseterm);
626 			} else {
627 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
628 				*tl = 0;
629 			}
630 		}
631 #endif
632 	}
633 	/* Finally, return values */
634 	*bposp = bpos;
635 	return (mb);
636 }
637 
638 /*
639  * Build the RPC header and fill in the authorization info.
640  * The authorization string argument is only used when the credentials
641  * come from outside of the kernel.
642  * Returns the head of the mbuf list.
643  */
644 struct mbuf *
645 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len,
646 	verf_str, mrest, mrest_len, mbp, xidp)
647 	struct ucred *cr;
648 	int nmflag;
649 	int procid;
650 	int auth_type;
651 	int auth_len;
652 	char *auth_str;
653 	int verf_len;
654 	char *verf_str;
655 	struct mbuf *mrest;
656 	int mrest_len;
657 	struct mbuf **mbp;
658 	u_int32_t *xidp;
659 {
660 	struct mbuf *mb;
661 	u_int32_t *tl;
662 	caddr_t bpos;
663 	int i;
664 	struct mbuf *mreq, *mb2;
665 	int siz, grpsiz, authsiz;
666 	struct timeval tv;
667 	static u_int32_t base;
668 
669 	authsiz = nfsm_rndup(auth_len);
670 	MGETHDR(mb, M_WAIT, MT_DATA);
671 	if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
672 		MCLGET(mb, M_WAIT);
673 	} else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
674 		MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED);
675 	} else {
676 		MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
677 	}
678 	mb->m_len = 0;
679 	mreq = mb;
680 	bpos = mtod(mb, caddr_t);
681 
682 	/*
683 	 * First the RPC header.
684 	 */
685 	nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
686 
687 	/*
688 	 * derive initial xid from system time
689 	 * XXX time is invalid if root not yet mounted
690 	 */
691 	if (!base && (rootvp)) {
692 		microtime(&tv);
693 		base = tv.tv_sec << 12;
694 		nfs_xid = base;
695 	}
696 	/*
697 	 * Skip zero xid if it should ever happen.
698 	 */
699 	if (++nfs_xid == 0)
700 		nfs_xid++;
701 
702 	*tl++ = *xidp = txdr_unsigned(nfs_xid);
703 	*tl++ = rpc_call;
704 	*tl++ = rpc_vers;
705 	if (nmflag & NFSMNT_NQNFS) {
706 		*tl++ = txdr_unsigned(NQNFS_PROG);
707 		*tl++ = txdr_unsigned(NQNFS_VER3);
708 	} else {
709 		*tl++ = txdr_unsigned(NFS_PROG);
710 		if (nmflag & NFSMNT_NFSV3)
711 			*tl++ = txdr_unsigned(NFS_VER3);
712 		else
713 			*tl++ = txdr_unsigned(NFS_VER2);
714 	}
715 	if (nmflag & NFSMNT_NFSV3)
716 		*tl++ = txdr_unsigned(procid);
717 	else
718 		*tl++ = txdr_unsigned(nfsv2_procid[procid]);
719 
720 	/*
721 	 * And then the authorization cred.
722 	 */
723 	*tl++ = txdr_unsigned(auth_type);
724 	*tl = txdr_unsigned(authsiz);
725 	switch (auth_type) {
726 	case RPCAUTH_UNIX:
727 		nfsm_build(tl, u_int32_t *, auth_len);
728 		*tl++ = 0;		/* stamp ?? */
729 		*tl++ = 0;		/* NULL hostname */
730 		*tl++ = txdr_unsigned(cr->cr_uid);
731 		*tl++ = txdr_unsigned(cr->cr_gid);
732 		grpsiz = (auth_len >> 2) - 5;
733 		*tl++ = txdr_unsigned(grpsiz);
734 		for (i = 0; i < grpsiz; i++)
735 			*tl++ = txdr_unsigned(cr->cr_groups[i]);
736 		break;
737 	case RPCAUTH_KERB4:
738 		siz = auth_len;
739 		while (siz > 0) {
740 			if (M_TRAILINGSPACE(mb) == 0) {
741 				MGET(mb2, M_WAIT, MT_DATA);
742 				if (siz >= MINCLSIZE)
743 					MCLGET(mb2, M_WAIT);
744 				mb->m_next = mb2;
745 				mb = mb2;
746 				mb->m_len = 0;
747 				bpos = mtod(mb, caddr_t);
748 			}
749 			i = min(siz, M_TRAILINGSPACE(mb));
750 			memcpy(bpos, auth_str, i);
751 			mb->m_len += i;
752 			auth_str += i;
753 			bpos += i;
754 			siz -= i;
755 		}
756 		if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
757 			for (i = 0; i < siz; i++)
758 				*bpos++ = '\0';
759 			mb->m_len += siz;
760 		}
761 		break;
762 	};
763 
764 	/*
765 	 * And the verifier...
766 	 */
767 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
768 	if (verf_str) {
769 		*tl++ = txdr_unsigned(RPCAUTH_KERB4);
770 		*tl = txdr_unsigned(verf_len);
771 		siz = verf_len;
772 		while (siz > 0) {
773 			if (M_TRAILINGSPACE(mb) == 0) {
774 				MGET(mb2, M_WAIT, MT_DATA);
775 				if (siz >= MINCLSIZE)
776 					MCLGET(mb2, M_WAIT);
777 				mb->m_next = mb2;
778 				mb = mb2;
779 				mb->m_len = 0;
780 				bpos = mtod(mb, caddr_t);
781 			}
782 			i = min(siz, M_TRAILINGSPACE(mb));
783 			memcpy(bpos, verf_str, i);
784 			mb->m_len += i;
785 			verf_str += i;
786 			bpos += i;
787 			siz -= i;
788 		}
789 		if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
790 			for (i = 0; i < siz; i++)
791 				*bpos++ = '\0';
792 			mb->m_len += siz;
793 		}
794 	} else {
795 		*tl++ = txdr_unsigned(RPCAUTH_NULL);
796 		*tl = 0;
797 	}
798 	mb->m_next = mrest;
799 	mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
800 	mreq->m_pkthdr.rcvif = (struct ifnet *)0;
801 	*mbp = mb;
802 	return (mreq);
803 }
804 
805 /*
806  * copies mbuf chain to the uio scatter/gather list
807  */
808 int
809 nfsm_mbuftouio(mrep, uiop, siz, dpos)
810 	struct mbuf **mrep;
811 	struct uio *uiop;
812 	int siz;
813 	caddr_t *dpos;
814 {
815 	char *mbufcp, *uiocp;
816 	int xfer, left, len;
817 	struct mbuf *mp;
818 	long uiosiz, rem;
819 	int error = 0;
820 
821 	mp = *mrep;
822 	mbufcp = *dpos;
823 	len = mtod(mp, caddr_t)+mp->m_len-mbufcp;
824 	rem = nfsm_rndup(siz)-siz;
825 	while (siz > 0) {
826 		if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
827 			return (EFBIG);
828 		left = uiop->uio_iov->iov_len;
829 		uiocp = uiop->uio_iov->iov_base;
830 		if (left > siz)
831 			left = siz;
832 		uiosiz = left;
833 		while (left > 0) {
834 			while (len == 0) {
835 				mp = mp->m_next;
836 				if (mp == NULL)
837 					return (EBADRPC);
838 				mbufcp = mtod(mp, caddr_t);
839 				len = mp->m_len;
840 			}
841 			xfer = (left > len) ? len : left;
842 #ifdef notdef
843 			/* Not Yet.. */
844 			if (uiop->uio_iov->iov_op != NULL)
845 				(*(uiop->uio_iov->iov_op))
846 				(mbufcp, uiocp, xfer);
847 			else
848 #endif
849 			if (uiop->uio_segflg == UIO_SYSSPACE)
850 				memcpy(uiocp, mbufcp, xfer);
851 			else
852 				copyout(mbufcp, uiocp, xfer);
853 			left -= xfer;
854 			len -= xfer;
855 			mbufcp += xfer;
856 			uiocp += xfer;
857 			uiop->uio_offset += xfer;
858 			uiop->uio_resid -= xfer;
859 		}
860 		if (uiop->uio_iov->iov_len <= siz) {
861 			uiop->uio_iovcnt--;
862 			uiop->uio_iov++;
863 		} else {
864 			uiop->uio_iov->iov_base =
865 			    (caddr_t)uiop->uio_iov->iov_base + uiosiz;
866 			uiop->uio_iov->iov_len -= uiosiz;
867 		}
868 		siz -= uiosiz;
869 	}
870 	*dpos = mbufcp;
871 	*mrep = mp;
872 	if (rem > 0) {
873 		if (len < rem)
874 			error = nfs_adv(mrep, dpos, rem, len);
875 		else
876 			*dpos += rem;
877 	}
878 	return (error);
879 }
880 
881 /*
882  * copies a uio scatter/gather list to an mbuf chain.
883  * NOTE: can ony handle iovcnt == 1
884  */
885 int
886 nfsm_uiotombuf(uiop, mq, siz, bpos)
887 	struct uio *uiop;
888 	struct mbuf **mq;
889 	int siz;
890 	caddr_t *bpos;
891 {
892 	char *uiocp;
893 	struct mbuf *mp, *mp2;
894 	int xfer, left, mlen;
895 	int uiosiz, clflg, rem;
896 	char *cp;
897 
898 #ifdef DIAGNOSTIC
899 	if (uiop->uio_iovcnt != 1)
900 		panic("nfsm_uiotombuf: iovcnt != 1");
901 #endif
902 
903 	if (siz > MLEN)		/* or should it >= MCLBYTES ?? */
904 		clflg = 1;
905 	else
906 		clflg = 0;
907 	rem = nfsm_rndup(siz)-siz;
908 	mp = mp2 = *mq;
909 	while (siz > 0) {
910 		left = uiop->uio_iov->iov_len;
911 		uiocp = uiop->uio_iov->iov_base;
912 		if (left > siz)
913 			left = siz;
914 		uiosiz = left;
915 		while (left > 0) {
916 			mlen = M_TRAILINGSPACE(mp);
917 			if (mlen == 0) {
918 				MGET(mp, M_WAIT, MT_DATA);
919 				if (clflg)
920 					MCLGET(mp, M_WAIT);
921 				mp->m_len = 0;
922 				mp2->m_next = mp;
923 				mp2 = mp;
924 				mlen = M_TRAILINGSPACE(mp);
925 			}
926 			xfer = (left > mlen) ? mlen : left;
927 #ifdef notdef
928 			/* Not Yet.. */
929 			if (uiop->uio_iov->iov_op != NULL)
930 				(*(uiop->uio_iov->iov_op))
931 				(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
932 			else
933 #endif
934 			if (uiop->uio_segflg == UIO_SYSSPACE)
935 				memcpy(mtod(mp, caddr_t)+mp->m_len, uiocp, xfer);
936 			else
937 				copyin(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
938 			mp->m_len += xfer;
939 			left -= xfer;
940 			uiocp += xfer;
941 			uiop->uio_offset += xfer;
942 			uiop->uio_resid -= xfer;
943 		}
944 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
945 		    uiosiz;
946 		uiop->uio_iov->iov_len -= uiosiz;
947 		siz -= uiosiz;
948 	}
949 	if (rem > 0) {
950 		if (rem > M_TRAILINGSPACE(mp)) {
951 			MGET(mp, M_WAIT, MT_DATA);
952 			mp->m_len = 0;
953 			mp2->m_next = mp;
954 		}
955 		cp = mtod(mp, caddr_t)+mp->m_len;
956 		for (left = 0; left < rem; left++)
957 			*cp++ = '\0';
958 		mp->m_len += rem;
959 		*bpos = cp;
960 	} else
961 		*bpos = mtod(mp, caddr_t)+mp->m_len;
962 	*mq = mp;
963 	return (0);
964 }
965 
966 /*
967  * Get at least "siz" bytes of correctly aligned data.
968  * When called the mbuf pointers are not necessarily correct,
969  * dsosp points to what ought to be in m_data and left contains
970  * what ought to be in m_len.
971  * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
972  * cases. (The macros use the vars. dpos and dpos2)
973  */
974 int
975 nfsm_disct(mdp, dposp, siz, left, cp2)
976 	struct mbuf **mdp;
977 	caddr_t *dposp;
978 	int siz;
979 	int left;
980 	caddr_t *cp2;
981 {
982 	struct mbuf *m1, *m2;
983 	struct mbuf *havebuf = NULL;
984 	caddr_t src = *dposp;
985 	caddr_t dst;
986 	int len;
987 
988 #ifdef DEBUG
989 	if (left < 0)
990 		panic("nfsm_disct: left < 0");
991 #endif
992 	m1 = *mdp;
993 	/*
994 	 * Skip through the mbuf chain looking for an mbuf with
995 	 * some data. If the first mbuf found has enough data
996 	 * and it is correctly aligned return it.
997 	 */
998 	while (left == 0) {
999 		havebuf = m1;
1000 		*mdp = m1 = m1->m_next;
1001 		if (m1 == NULL)
1002 			return (EBADRPC);
1003 		src = mtod(m1, caddr_t);
1004 		left = m1->m_len;
1005 		/*
1006 		 * If we start a new mbuf and it is big enough
1007 		 * and correctly aligned just return it, don't
1008 		 * do any pull up.
1009 		 */
1010 		if (left >= siz && nfsm_aligned(src)) {
1011 			*cp2 = src;
1012 			*dposp = src + siz;
1013 			return (0);
1014 		}
1015 	}
1016 	if (m1->m_flags & M_EXT) {
1017 		if (havebuf) {
1018 			/* If the first mbuf with data has external data
1019 			 * and there is a previous empty mbuf use it
1020 			 * to move the data into.
1021 			 */
1022 			m2 = m1;
1023 			*mdp = m1 = havebuf;
1024 			if (m1->m_flags & M_EXT) {
1025 				MEXTREMOVE(m1);
1026 			}
1027 		} else {
1028 			/*
1029 			 * If the first mbuf has a external data
1030 			 * and there is no previous empty mbuf
1031 			 * allocate a new mbuf and move the external
1032 			 * data to the new mbuf. Also make the first
1033 			 * mbuf look empty.
1034 			 */
1035 			m2 = m_get(M_WAIT, MT_DATA);
1036 			m2->m_ext = m1->m_ext;
1037 			m2->m_data = src;
1038 			m2->m_len = left;
1039 			MCLADDREFERENCE(m1, m2);
1040 			MEXTREMOVE(m1);
1041 			m2->m_next = m1->m_next;
1042 			m1->m_next = m2;
1043 		}
1044 		m1->m_len = 0;
1045 		dst = m1->m_dat;
1046 	} else {
1047 		/*
1048 		 * If the first mbuf has no external data
1049 		 * move the data to the front of the mbuf.
1050 		 */
1051 		if ((dst = m1->m_dat) != src)
1052 			memmove(dst, src, left);
1053 		dst += left;
1054 		m1->m_len = left;
1055 		m2 = m1->m_next;
1056 	}
1057 	m1->m_flags &= ~M_PKTHDR;
1058 	*cp2 = m1->m_data = m1->m_dat;   /* data is at beginning of buffer */
1059 	*dposp = mtod(m1, caddr_t) + siz;
1060 	/*
1061 	 * Loop through mbufs pulling data up into first mbuf until
1062 	 * the first mbuf is full or there is no more data to
1063 	 * pullup.
1064 	 */
1065 	while ((len = (MLEN - m1->m_len)) != 0 && m2) {
1066 		if ((len = min(len, m2->m_len)) != 0)
1067 			memcpy(dst, m2->m_data, len);
1068 		m1->m_len += len;
1069 		dst += len;
1070 		m2->m_data += len;
1071 		m2->m_len -= len;
1072 		m2 = m2->m_next;
1073 	}
1074 	if (m1->m_len < siz)
1075 		return (EBADRPC);
1076 	return (0);
1077 }
1078 
1079 /*
1080  * Advance the position in the mbuf chain.
1081  */
1082 int
1083 nfs_adv(mdp, dposp, offs, left)
1084 	struct mbuf **mdp;
1085 	caddr_t *dposp;
1086 	int offs;
1087 	int left;
1088 {
1089 	struct mbuf *m;
1090 	int s;
1091 
1092 	m = *mdp;
1093 	s = left;
1094 	while (s < offs) {
1095 		offs -= s;
1096 		m = m->m_next;
1097 		if (m == NULL)
1098 			return (EBADRPC);
1099 		s = m->m_len;
1100 	}
1101 	*mdp = m;
1102 	*dposp = mtod(m, caddr_t)+offs;
1103 	return (0);
1104 }
1105 
1106 /*
1107  * Copy a string into mbufs for the hard cases...
1108  */
1109 int
1110 nfsm_strtmbuf(mb, bpos, cp, siz)
1111 	struct mbuf **mb;
1112 	char **bpos;
1113 	const char *cp;
1114 	long siz;
1115 {
1116 	struct mbuf *m1 = NULL, *m2;
1117 	long left, xfer, len, tlen;
1118 	u_int32_t *tl;
1119 	int putsize;
1120 
1121 	putsize = 1;
1122 	m2 = *mb;
1123 	left = M_TRAILINGSPACE(m2);
1124 	if (left > 0) {
1125 		tl = ((u_int32_t *)(*bpos));
1126 		*tl++ = txdr_unsigned(siz);
1127 		putsize = 0;
1128 		left -= NFSX_UNSIGNED;
1129 		m2->m_len += NFSX_UNSIGNED;
1130 		if (left > 0) {
1131 			memcpy((caddr_t) tl, cp, left);
1132 			siz -= left;
1133 			cp += left;
1134 			m2->m_len += left;
1135 			left = 0;
1136 		}
1137 	}
1138 	/* Loop around adding mbufs */
1139 	while (siz > 0) {
1140 		MGET(m1, M_WAIT, MT_DATA);
1141 		if (siz > MLEN)
1142 			MCLGET(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 void
1222 nfs_initdircache(vp)
1223 	struct vnode *vp;
1224 {
1225 	struct nfsnode *np = VTONFS(vp);
1226 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1227 
1228 	np->n_dircachesize = 0;
1229 	np->n_dblkno = 1;
1230 	np->n_dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF,
1231 	    M_WAITOK, &nfsdirhashmask);
1232 	TAILQ_INIT(&np->n_dirchain);
1233 	if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1234 		MALLOC(np->n_dirgens, unsigned *,
1235 		    NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF,
1236 		    M_WAITOK);
1237 		memset((caddr_t)np->n_dirgens, 0,
1238 		    NFS_DIRHASHSIZ * sizeof (unsigned));
1239 	}
1240 }
1241 
1242 static struct nfsdircache dzero = {0, 0, {0, 0}, {0, 0}, 0, 0, 0};
1243 
1244 struct nfsdircache *
1245 nfs_searchdircache(vp, off, do32, hashent)
1246 	struct vnode *vp;
1247 	off_t off;
1248 	int do32;
1249 	int *hashent;
1250 {
1251 	struct nfsdirhashhead *ndhp;
1252 	struct nfsdircache *ndp = NULL;
1253 	struct nfsnode *np = VTONFS(vp);
1254 	unsigned ent;
1255 
1256 	/*
1257 	 * Zero is always a valid cookie.
1258 	 */
1259 	if (off == 0)
1260 		return &dzero;
1261 
1262 	/*
1263 	 * We use a 32bit cookie as search key, directly reconstruct
1264 	 * the hashentry. Else use the hashfunction.
1265 	 */
1266 	if (do32) {
1267 		ent = (u_int32_t)off >> 24;
1268 		if (ent >= NFS_DIRHASHSIZ)
1269 			return NULL;
1270 		ndhp = &np->n_dircache[ent];
1271 	} else {
1272 		ndhp = NFSDIRHASH(np, off);
1273 	}
1274 
1275 	if (hashent)
1276 		*hashent = (int)(ndhp - np->n_dircache);
1277 	if (do32) {
1278 		for (ndp = ndhp->lh_first; ndp; ndp = ndp->dc_hash.le_next) {
1279 			if (ndp->dc_cookie32 == (u_int32_t)off) {
1280 				/*
1281 				 * An invalidated entry will become the
1282 				 * start of a new block fetched from
1283 				 * the server.
1284 				 */
1285 				if (ndp->dc_blkno == -1) {
1286 					ndp->dc_blkcookie = ndp->dc_cookie;
1287 					ndp->dc_blkno = np->n_dblkno++;
1288 					ndp->dc_entry = 0;
1289 				}
1290 				break;
1291 			}
1292 		}
1293 	} else {
1294 		for (ndp = ndhp->lh_first; ndp; ndp = ndp->dc_hash.le_next)
1295 			if (ndp->dc_cookie == off)
1296 				break;
1297 	}
1298 	return ndp;
1299 }
1300 
1301 
1302 struct nfsdircache *
1303 nfs_enterdircache(vp, off, blkoff, en, blkno)
1304 	struct vnode *vp;
1305 	off_t off, blkoff;
1306 	int en;
1307 	daddr_t blkno;
1308 {
1309 	struct nfsnode *np = VTONFS(vp);
1310 	struct nfsdirhashhead *ndhp;
1311 	struct nfsdircache *ndp = NULL, *first;
1312 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1313 	int hashent, gen, overwrite;
1314 
1315 	if (!np->n_dircache)
1316 		/*
1317 		 * XXX would like to do this in nfs_nget but vtype
1318 		 * isn't known at that time.
1319 		 */
1320 		nfs_initdircache(vp);
1321 
1322 	/*
1323 	 * XXX refuse entries for offset 0. amd(8) erroneously sets
1324 	 * cookie 0 for the '.' entry, making this necessary. This
1325 	 * isn't so bad, as 0 is a special case anyway.
1326 	 */
1327 	if (off == 0)
1328 		return &dzero;
1329 
1330 	ndp = nfs_searchdircache(vp, off, 0, &hashent);
1331 
1332 	if (ndp && ndp->dc_blkno != -1) {
1333 		/*
1334 		 * Overwriting an old entry. Check if it's the same.
1335 		 * If so, just return. If not, remove the old entry.
1336 		 */
1337 		if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
1338 			return ndp;
1339 		TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1340 		LIST_REMOVE(ndp, dc_hash);
1341 		FREE(ndp, M_NFSDIROFF);
1342 		ndp = 0;
1343 	}
1344 
1345 	ndhp = &np->n_dircache[hashent];
1346 
1347 	if (!ndp) {
1348 		MALLOC(ndp, struct nfsdircache *, sizeof (*ndp), M_NFSDIROFF,
1349 		    M_WAITOK);
1350 		overwrite = 0;
1351 		if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1352 			/*
1353 			 * We're allocating a new entry, so bump the
1354 			 * generation number.
1355 			 */
1356 			gen = ++np->n_dirgens[hashent];
1357 			if (gen == 0) {
1358 				np->n_dirgens[hashent]++;
1359 				gen++;
1360 			}
1361 			ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
1362 		}
1363 	} else
1364 		overwrite = 1;
1365 
1366 	/*
1367 	 * If the entry number is 0, we are at the start of a new block, so
1368 	 * allocate a new blocknumber.
1369 	 */
1370 	if (en == 0)
1371 		ndp->dc_blkno = np->n_dblkno++;
1372 	else
1373 		ndp->dc_blkno = blkno;
1374 
1375 	ndp->dc_cookie = off;
1376 	ndp->dc_blkcookie = blkoff;
1377 	ndp->dc_entry = en;
1378 
1379 	if (overwrite)
1380 		return ndp;
1381 
1382 	/*
1383 	 * If the maximum directory cookie cache size has been reached
1384 	 * for this node, take one off the front. The idea is that
1385 	 * directories are typically read front-to-back once, so that
1386 	 * the oldest entries can be thrown away without much performance
1387 	 * loss.
1388 	 */
1389 	if (np->n_dircachesize == NFS_MAXDIRCACHE) {
1390 		first = np->n_dirchain.tqh_first;
1391 		TAILQ_REMOVE(&np->n_dirchain, first, dc_chain);
1392 		LIST_REMOVE(first, dc_hash);
1393 		FREE(first, M_NFSDIROFF);
1394 	} else
1395 		np->n_dircachesize++;
1396 
1397 	LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
1398 	TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
1399 	return ndp;
1400 }
1401 
1402 void
1403 nfs_invaldircache(vp, forcefree)
1404 	struct vnode *vp;
1405 	int forcefree;
1406 {
1407 	struct nfsnode *np = VTONFS(vp);
1408 	struct nfsdircache *ndp = NULL;
1409 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1410 
1411 #ifdef DIAGNOSTIC
1412 	if (vp->v_type != VDIR)
1413 		panic("nfs: invaldircache: not dir");
1414 #endif
1415 
1416 	if (!np->n_dircache)
1417 		return;
1418 
1419 	if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
1420 		while ((ndp = np->n_dirchain.tqh_first)) {
1421 			TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1422 			LIST_REMOVE(ndp, dc_hash);
1423 			FREE(ndp, M_NFSDIROFF);
1424 		}
1425 		np->n_dircachesize = 0;
1426 		if (forcefree && np->n_dirgens) {
1427 			FREE(np->n_dirgens, M_NFSDIROFF);
1428 		}
1429 	} else {
1430 		for (ndp = np->n_dirchain.tqh_first; ndp;
1431 		    ndp = ndp->dc_chain.tqe_next)
1432 			ndp->dc_blkno = -1;
1433 	}
1434 
1435 	np->n_dblkno = 1;
1436 }
1437 
1438 /*
1439  * Called once before VFS init to initialize shared and
1440  * server-specific data structures.
1441  */
1442 void
1443 nfs_init()
1444 {
1445 	nfsrtt.pos = 0;
1446 	rpc_vers = txdr_unsigned(RPC_VER2);
1447 	rpc_call = txdr_unsigned(RPC_CALL);
1448 	rpc_reply = txdr_unsigned(RPC_REPLY);
1449 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1450 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1451 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1452 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1453 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1454 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1455 	nfs_prog = txdr_unsigned(NFS_PROG);
1456 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
1457 	nfs_true = txdr_unsigned(TRUE);
1458 	nfs_false = txdr_unsigned(FALSE);
1459 	nfs_xdrneg1 = txdr_unsigned(-1);
1460 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1461 	if (nfs_ticks < 1)
1462 		nfs_ticks = 1;
1463 #ifdef NFSSERVER
1464 	nfsrv_init(0);			/* Init server data structures */
1465 	nfsrv_initcache();		/* Init the server request cache */
1466 #endif /* NFSSERVER */
1467 
1468 #if defined(NFSSERVER) || !defined(NFS_V2_ONLY)
1469 	/*
1470 	 * Initialize the nqnfs data structures.
1471 	 */
1472 	if (nqnfsstarttime == 0) {
1473 		nqnfsstarttime = boottime.tv_sec + nqsrv_maxlease
1474 			+ nqsrv_clockskew + nqsrv_writeslack;
1475 		NQLOADNOVRAM(nqnfsstarttime);
1476 		CIRCLEQ_INIT(&nqtimerhead);
1477 		nqfhhashtbl = hashinit(NQLCHSZ, HASH_LIST, M_NQLEASE,
1478 		    M_WAITOK, &nqfhhash);
1479 	}
1480 #endif
1481 
1482 	exithook_establish(nfs_exit, NULL);
1483 
1484 	/*
1485 	 * Initialize reply list and start timer
1486 	 */
1487 	TAILQ_INIT(&nfs_reqq);
1488 	nfs_timer(NULL);
1489 
1490 #ifdef NFS
1491 	/* Initialize the kqueue structures */
1492 	nfs_kqinit();
1493 #endif
1494 }
1495 
1496 #ifdef NFS
1497 /*
1498  * Called once at VFS init to initialize client-specific data structures.
1499  */
1500 void
1501 nfs_vfs_init()
1502 {
1503 	nfs_nhinit();			/* Init the nfsnode table */
1504 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1505 }
1506 
1507 void
1508 nfs_vfs_reinit()
1509 {
1510 	nfs_nhreinit();
1511 }
1512 
1513 void
1514 nfs_vfs_done()
1515 {
1516 	nfs_nhdone();
1517 }
1518 
1519 /*
1520  * Attribute cache routines.
1521  * nfs_loadattrcache() - loads or updates the cache contents from attributes
1522  *	that are on the mbuf list
1523  * nfs_getattrcache() - returns valid attributes if found in cache, returns
1524  *	error otherwise
1525  */
1526 
1527 /*
1528  * Load the attribute cache (that lives in the nfsnode entry) with
1529  * the values on the mbuf list and
1530  * Iff vap not NULL
1531  *    copy the attributes to *vaper
1532  */
1533 int
1534 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags)
1535 	struct vnode **vpp;
1536 	struct mbuf **mdp;
1537 	caddr_t *dposp;
1538 	struct vattr *vaper;
1539 	int flags;
1540 {
1541 	int32_t t1;
1542 	caddr_t cp2;
1543 	int error = 0;
1544 	struct mbuf *md;
1545 	int v3 = NFS_ISV3(*vpp);
1546 
1547 	md = *mdp;
1548 	t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
1549 	error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2);
1550 	if (error)
1551 		return (error);
1552 	return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags);
1553 }
1554 
1555 int
1556 nfs_loadattrcache(vpp, fp, vaper, flags)
1557 	struct vnode **vpp;
1558 	struct nfs_fattr *fp;
1559 	struct vattr *vaper;
1560 	int flags;
1561 {
1562 	struct vnode *vp = *vpp;
1563 	struct vattr *vap;
1564 	int v3 = NFS_ISV3(vp);
1565 	enum vtype vtyp;
1566 	u_short vmode;
1567 	struct timespec mtime;
1568 	struct vnode *nvp;
1569 	int32_t rdev;
1570 	struct nfsnode *np;
1571 	extern int (**spec_nfsv2nodeop_p) __P((void *));
1572 	uid_t uid;
1573 	gid_t gid;
1574 
1575 	if (v3) {
1576 		vtyp = nfsv3tov_type(fp->fa_type);
1577 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1578 		rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1),
1579 			fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2));
1580 		fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
1581 	} else {
1582 		vtyp = nfsv2tov_type(fp->fa_type);
1583 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1584 		if (vtyp == VNON || vtyp == VREG)
1585 			vtyp = IFTOVT(vmode);
1586 		rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
1587 		fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
1588 
1589 		/*
1590 		 * Really ugly NFSv2 kludge.
1591 		 */
1592 		if (vtyp == VCHR && rdev == 0xffffffff)
1593 			vtyp = VFIFO;
1594 	}
1595 
1596 	vmode &= ALLPERMS;
1597 
1598 	/*
1599 	 * If v_type == VNON it is a new node, so fill in the v_type,
1600 	 * n_mtime fields. Check to see if it represents a special
1601 	 * device, and if so, check for a possible alias. Once the
1602 	 * correct vnode has been obtained, fill in the rest of the
1603 	 * information.
1604 	 */
1605 	np = VTONFS(vp);
1606 	if (vp->v_type == VNON) {
1607 		vp->v_type = vtyp;
1608 		if (vp->v_type == VFIFO) {
1609 			extern int (**fifo_nfsv2nodeop_p) __P((void *));
1610 			vp->v_op = fifo_nfsv2nodeop_p;
1611 		}
1612 		if (vp->v_type == VCHR || vp->v_type == VBLK) {
1613 			vp->v_op = spec_nfsv2nodeop_p;
1614 			nvp = checkalias(vp, (dev_t)rdev, vp->v_mount);
1615 			if (nvp) {
1616 				/*
1617 				 * Discard unneeded vnode, but save its nfsnode.
1618 				 * Since the nfsnode does not have a lock, its
1619 				 * vnode lock has to be carried over.
1620 				 */
1621 				/*
1622 				 * XXX is the old node sure to be locked here?
1623 				 */
1624 				KASSERT(lockstatus(&vp->v_lock) ==
1625 				    LK_EXCLUSIVE);
1626 				nvp->v_data = vp->v_data;
1627 				vp->v_data = NULL;
1628 				VOP_UNLOCK(vp, 0);
1629 				vp->v_op = spec_vnodeop_p;
1630 				vrele(vp);
1631 				vgone(vp);
1632 				lockmgr(&nvp->v_lock, LK_EXCLUSIVE,
1633 				    &nvp->v_interlock);
1634 				/*
1635 				 * Reinitialize aliased node.
1636 				 */
1637 				np->n_vnode = nvp;
1638 				*vpp = vp = nvp;
1639 			}
1640 		}
1641 		np->n_mtime = mtime.tv_sec;
1642 	}
1643 	uid = fxdr_unsigned(uid_t, fp->fa_uid);
1644 	gid = fxdr_unsigned(gid_t, fp->fa_gid);
1645 	vap = np->n_vattr;
1646 
1647 	/*
1648 	 * Invalidate access cache if uid, gid or mode changed.
1649 	 */
1650 	if (np->n_accstamp != -1 &&
1651 	    (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode))
1652 		np->n_accstamp = -1;
1653 
1654 	vap->va_type = vtyp;
1655 	vap->va_mode = vmode;
1656 	vap->va_rdev = (dev_t)rdev;
1657 	vap->va_mtime = mtime;
1658 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1659 	switch (vtyp) {
1660 	case VDIR:
1661 		vap->va_blocksize = NFS_DIRFRAGSIZ;
1662 		break;
1663 	case VBLK:
1664 		vap->va_blocksize = BLKDEV_IOSIZE;
1665 		break;
1666 	case VCHR:
1667 		vap->va_blocksize = MAXBSIZE;
1668 		break;
1669 	default:
1670 		vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize :
1671 		    fxdr_unsigned(int32_t, fp->fa2_blocksize);
1672 		break;
1673 	}
1674 	if (v3) {
1675 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1676 		vap->va_uid = uid;
1677 		vap->va_gid = gid;
1678 		vap->va_size = fxdr_hyper(&fp->fa3_size);
1679 		vap->va_bytes = fxdr_hyper(&fp->fa3_used);
1680 		vap->va_fileid = fxdr_unsigned(int32_t,
1681 		    fp->fa3_fileid.nfsuquad[1]);
1682 		fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
1683 		fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime);
1684 		vap->va_flags = 0;
1685 		vap->va_filerev = 0;
1686 	} else {
1687 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1688 		vap->va_uid = uid;
1689 		vap->va_gid = gid;
1690 		vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
1691 		vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks)
1692 		    * NFS_FABLKSIZE;
1693 		vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
1694 		fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
1695 		vap->va_flags = 0;
1696 		vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t,
1697 		    fp->fa2_ctime.nfsv2_sec);
1698 		vap->va_ctime.tv_nsec = 0;
1699 		vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
1700 		vap->va_filerev = 0;
1701 	}
1702 	if (vap->va_size != np->n_size) {
1703 		if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) {
1704 			vap->va_size = np->n_size;
1705 		} else {
1706 			np->n_size = vap->va_size;
1707 			if (vap->va_type == VREG) {
1708 				if ((flags & NAC_NOTRUNC)
1709 				    && np->n_size < vp->v_size) {
1710 					/*
1711 					 * we can't free pages now because
1712 					 * the pages can be owned by ourselves.
1713 					 */
1714 					np->n_flag |= NTRUNCDELAYED;
1715 				}
1716 				else {
1717 					uvm_vnp_setsize(vp, np->n_size);
1718 				}
1719 			}
1720 		}
1721 	}
1722 	np->n_attrstamp = time.tv_sec;
1723 	if (vaper != NULL) {
1724 		memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(*vap));
1725 		if (np->n_flag & NCHG) {
1726 			if (np->n_flag & NACC)
1727 				vaper->va_atime = np->n_atim;
1728 			if (np->n_flag & NUPD)
1729 				vaper->va_mtime = np->n_mtim;
1730 		}
1731 	}
1732 	return (0);
1733 }
1734 
1735 /*
1736  * Check the time stamp
1737  * If the cache is valid, copy contents to *vap and return 0
1738  * otherwise return an error
1739  */
1740 int
1741 nfs_getattrcache(vp, vaper)
1742 	struct vnode *vp;
1743 	struct vattr *vaper;
1744 {
1745 	struct nfsnode *np = VTONFS(vp);
1746 	struct vattr *vap;
1747 
1748 	if ((time.tv_sec - np->n_attrstamp) >= NFS_ATTRTIMEO(np)) {
1749 		nfsstats.attrcache_misses++;
1750 		return (ENOENT);
1751 	}
1752 	nfsstats.attrcache_hits++;
1753 	vap = np->n_vattr;
1754 	if (vap->va_size != np->n_size) {
1755 		if (vap->va_type == VREG) {
1756 			if (np->n_flag & NMODIFIED) {
1757 				if (vap->va_size < np->n_size)
1758 					vap->va_size = np->n_size;
1759 				else
1760 					np->n_size = vap->va_size;
1761 			} else
1762 				np->n_size = vap->va_size;
1763 			uvm_vnp_setsize(vp, np->n_size);
1764 		} else
1765 			np->n_size = vap->va_size;
1766 	}
1767 	memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(struct vattr));
1768 	if (np->n_flag & NCHG) {
1769 		if (np->n_flag & NACC)
1770 			vaper->va_atime = np->n_atim;
1771 		if (np->n_flag & NUPD)
1772 			vaper->va_mtime = np->n_mtim;
1773 	}
1774 	return (0);
1775 }
1776 
1777 void
1778 nfs_delayedtruncate(vp)
1779 	struct vnode *vp;
1780 {
1781 	struct nfsnode *np = VTONFS(vp);
1782 
1783 	if (np->n_flag & NTRUNCDELAYED) {
1784 		np->n_flag &= ~NTRUNCDELAYED;
1785 		uvm_vnp_setsize(vp, np->n_size);
1786 	}
1787 }
1788 
1789 /*
1790  * Heuristic to see if the server XDR encodes directory cookies or not.
1791  * it is not supposed to, but a lot of servers may do this. Also, since
1792  * most/all servers will implement V2 as well, it is expected that they
1793  * may return just 32 bits worth of cookie information, so we need to
1794  * find out in which 32 bits this information is available. We do this
1795  * to avoid trouble with emulated binaries that can't handle 64 bit
1796  * directory offsets.
1797  */
1798 
1799 void
1800 nfs_cookieheuristic(vp, flagp, p, cred)
1801 	struct vnode *vp;
1802 	int *flagp;
1803 	struct proc *p;
1804 	struct ucred *cred;
1805 {
1806 	struct uio auio;
1807 	struct iovec aiov;
1808 	caddr_t buf, cp;
1809 	struct dirent *dp;
1810 	off_t *cookies = NULL, *cop;
1811 	int error, eof, nc, len;
1812 
1813 	MALLOC(buf, caddr_t, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK);
1814 
1815 	aiov.iov_base = buf;
1816 	aiov.iov_len = NFS_DIRFRAGSIZ;
1817 	auio.uio_iov = &aiov;
1818 	auio.uio_iovcnt = 1;
1819 	auio.uio_rw = UIO_READ;
1820 	auio.uio_segflg = UIO_SYSSPACE;
1821 	auio.uio_procp = p;
1822 	auio.uio_resid = NFS_DIRFRAGSIZ;
1823 	auio.uio_offset = 0;
1824 
1825 	error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc);
1826 
1827 	len = NFS_DIRFRAGSIZ - auio.uio_resid;
1828 	if (error || len == 0) {
1829 		FREE(buf, M_TEMP);
1830 		if (cookies)
1831 			free(cookies, M_TEMP);
1832 		return;
1833 	}
1834 
1835 	/*
1836 	 * Find the first valid entry and look at its offset cookie.
1837 	 */
1838 
1839 	cp = buf;
1840 	for (cop = cookies; len > 0; len -= dp->d_reclen) {
1841 		dp = (struct dirent *)cp;
1842 		if (dp->d_fileno != 0 && len >= dp->d_reclen) {
1843 			if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) {
1844 				*flagp |= NFSMNT_SWAPCOOKIE;
1845 				nfs_invaldircache(vp, 0);
1846 				nfs_vinvalbuf(vp, 0, cred, p, 1);
1847 			}
1848 			break;
1849 		}
1850 		cop++;
1851 		cp += dp->d_reclen;
1852 	}
1853 
1854 	FREE(buf, M_TEMP);
1855 	free(cookies, M_TEMP);
1856 }
1857 #endif /* NFS */
1858 
1859 /*
1860  * Set up nameidata for a lookup() call and do it.
1861  *
1862  * If pubflag is set, this call is done for a lookup operation on the
1863  * public filehandle. In that case we allow crossing mountpoints and
1864  * absolute pathnames. However, the caller is expected to check that
1865  * the lookup result is within the public fs, and deny access if
1866  * it is not.
1867  */
1868 int
1869 nfs_namei(ndp, fhp, len, slp, nam, mdp, dposp, retdirp, p, kerbflag, pubflag)
1870 	struct nameidata *ndp;
1871 	fhandle_t *fhp;
1872 	int len;
1873 	struct nfssvc_sock *slp;
1874 	struct mbuf *nam;
1875 	struct mbuf **mdp;
1876 	caddr_t *dposp;
1877 	struct vnode **retdirp;
1878 	struct proc *p;
1879 	int kerbflag, pubflag;
1880 {
1881 	int i, rem;
1882 	struct mbuf *md;
1883 	char *fromcp, *tocp, *cp;
1884 	struct iovec aiov;
1885 	struct uio auio;
1886 	struct vnode *dp;
1887 	int error, rdonly, linklen;
1888 	struct componentname *cnp = &ndp->ni_cnd;
1889 
1890 	*retdirp = (struct vnode *)0;
1891 
1892 	if ((len + 1) > MAXPATHLEN)
1893 		return (ENAMETOOLONG);
1894 	cnp->cn_pnbuf = PNBUF_GET();
1895 
1896 	/*
1897 	 * Copy the name from the mbuf list to ndp->ni_pnbuf
1898 	 * and set the various ndp fields appropriately.
1899 	 */
1900 	fromcp = *dposp;
1901 	tocp = cnp->cn_pnbuf;
1902 	md = *mdp;
1903 	rem = mtod(md, caddr_t) + md->m_len - fromcp;
1904 	for (i = 0; i < len; i++) {
1905 		while (rem == 0) {
1906 			md = md->m_next;
1907 			if (md == NULL) {
1908 				error = EBADRPC;
1909 				goto out;
1910 			}
1911 			fromcp = mtod(md, caddr_t);
1912 			rem = md->m_len;
1913 		}
1914 		if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
1915 			error = EACCES;
1916 			goto out;
1917 		}
1918 		*tocp++ = *fromcp++;
1919 		rem--;
1920 	}
1921 	*tocp = '\0';
1922 	*mdp = md;
1923 	*dposp = fromcp;
1924 	len = nfsm_rndup(len)-len;
1925 	if (len > 0) {
1926 		if (rem >= len)
1927 			*dposp += len;
1928 		else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
1929 			goto out;
1930 	}
1931 
1932 	/*
1933 	 * Extract and set starting directory.
1934 	 */
1935 	error = nfsrv_fhtovp(fhp, FALSE, &dp, ndp->ni_cnd.cn_cred, slp,
1936 	    nam, &rdonly, kerbflag, pubflag);
1937 	if (error)
1938 		goto out;
1939 	if (dp->v_type != VDIR) {
1940 		vrele(dp);
1941 		error = ENOTDIR;
1942 		goto out;
1943 	}
1944 
1945 	if (rdonly)
1946 		cnp->cn_flags |= RDONLY;
1947 
1948 	*retdirp = dp;
1949 
1950 	if (pubflag) {
1951 		/*
1952 		 * Oh joy. For WebNFS, handle those pesky '%' escapes,
1953 		 * and the 'native path' indicator.
1954 		 */
1955 		cp = PNBUF_GET();
1956 		fromcp = cnp->cn_pnbuf;
1957 		tocp = cp;
1958 		if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
1959 			switch ((unsigned char)*fromcp) {
1960 			case WEBNFS_NATIVE_CHAR:
1961 				/*
1962 				 * 'Native' path for us is the same
1963 				 * as a path according to the NFS spec,
1964 				 * just skip the escape char.
1965 				 */
1966 				fromcp++;
1967 				break;
1968 			/*
1969 			 * More may be added in the future, range 0x80-0xff
1970 			 */
1971 			default:
1972 				error = EIO;
1973 				PNBUF_PUT(cp);
1974 				goto out;
1975 			}
1976 		}
1977 		/*
1978 		 * Translate the '%' escapes, URL-style.
1979 		 */
1980 		while (*fromcp != '\0') {
1981 			if (*fromcp == WEBNFS_ESC_CHAR) {
1982 				if (fromcp[1] != '\0' && fromcp[2] != '\0') {
1983 					fromcp++;
1984 					*tocp++ = HEXSTRTOI(fromcp);
1985 					fromcp += 2;
1986 					continue;
1987 				} else {
1988 					error = ENOENT;
1989 					PNBUF_PUT(cp);
1990 					goto out;
1991 				}
1992 			} else
1993 				*tocp++ = *fromcp++;
1994 		}
1995 		*tocp = '\0';
1996 		PNBUF_PUT(cnp->cn_pnbuf);
1997 		cnp->cn_pnbuf = cp;
1998 	}
1999 
2000 	ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1;
2001 	ndp->ni_segflg = UIO_SYSSPACE;
2002 	ndp->ni_rootdir = rootvnode;
2003 
2004 	if (pubflag) {
2005 		ndp->ni_loopcnt = 0;
2006 		if (cnp->cn_pnbuf[0] == '/')
2007 			dp = rootvnode;
2008 	} else {
2009 		cnp->cn_flags |= NOCROSSMOUNT;
2010 	}
2011 
2012 	cnp->cn_proc = p;
2013 	VREF(dp);
2014 
2015     for (;;) {
2016 	cnp->cn_nameptr = cnp->cn_pnbuf;
2017 	ndp->ni_startdir = dp;
2018 	/*
2019 	 * And call lookup() to do the real work
2020 	 */
2021 	error = lookup(ndp);
2022 	if (error) {
2023 		PNBUF_PUT(cnp->cn_pnbuf);
2024 		return (error);
2025 	}
2026 	/*
2027 	 * Check for encountering a symbolic link
2028 	 */
2029 	if ((cnp->cn_flags & ISSYMLINK) == 0) {
2030 		if (cnp->cn_flags & (SAVENAME | SAVESTART))
2031 			cnp->cn_flags |= HASBUF;
2032 		else
2033 			PNBUF_PUT(cnp->cn_pnbuf);
2034 		return (0);
2035 	} else {
2036 		if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
2037 			VOP_UNLOCK(ndp->ni_dvp, 0);
2038 		if (!pubflag) {
2039 			error = EINVAL;
2040 			break;
2041 		}
2042 
2043 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
2044 			error = ELOOP;
2045 			break;
2046 		}
2047 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
2048 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred,
2049 			    cnp->cn_proc);
2050 			if (error != 0)
2051 				break;
2052 		}
2053 		if (ndp->ni_pathlen > 1)
2054 			cp = PNBUF_GET();
2055 		else
2056 			cp = cnp->cn_pnbuf;
2057 		aiov.iov_base = cp;
2058 		aiov.iov_len = MAXPATHLEN;
2059 		auio.uio_iov = &aiov;
2060 		auio.uio_iovcnt = 1;
2061 		auio.uio_offset = 0;
2062 		auio.uio_rw = UIO_READ;
2063 		auio.uio_segflg = UIO_SYSSPACE;
2064 		auio.uio_procp = (struct proc *)0;
2065 		auio.uio_resid = MAXPATHLEN;
2066 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
2067 		if (error) {
2068 		badlink:
2069 			if (ndp->ni_pathlen > 1)
2070 				PNBUF_PUT(cp);
2071 			break;
2072 		}
2073 		linklen = MAXPATHLEN - auio.uio_resid;
2074 		if (linklen == 0) {
2075 			error = ENOENT;
2076 			goto badlink;
2077 		}
2078 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
2079 			error = ENAMETOOLONG;
2080 			goto badlink;
2081 		}
2082 		if (ndp->ni_pathlen > 1) {
2083 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
2084 			PNBUF_PUT(cnp->cn_pnbuf);
2085 			cnp->cn_pnbuf = cp;
2086 		} else
2087 			cnp->cn_pnbuf[linklen] = '\0';
2088 		ndp->ni_pathlen += linklen;
2089 		vput(ndp->ni_vp);
2090 		dp = ndp->ni_dvp;
2091 		/*
2092 		 * Check if root directory should replace current directory.
2093 		 */
2094 		if (cnp->cn_pnbuf[0] == '/') {
2095 			vrele(dp);
2096 			dp = ndp->ni_rootdir;
2097 			VREF(dp);
2098 		}
2099 	}
2100    }
2101 	vrele(ndp->ni_dvp);
2102 	vput(ndp->ni_vp);
2103 	ndp->ni_vp = NULL;
2104 out:
2105 	PNBUF_PUT(cnp->cn_pnbuf);
2106 	return (error);
2107 }
2108 
2109 /*
2110  * A fiddled version of m_adj() that ensures null fill to a long
2111  * boundary and only trims off the back end
2112  */
2113 void
2114 nfsm_adj(mp, len, nul)
2115 	struct mbuf *mp;
2116 	int len;
2117 	int nul;
2118 {
2119 	struct mbuf *m;
2120 	int count, i;
2121 	char *cp;
2122 
2123 	/*
2124 	 * Trim from tail.  Scan the mbuf chain,
2125 	 * calculating its length and finding the last mbuf.
2126 	 * If the adjustment only affects this mbuf, then just
2127 	 * adjust and return.  Otherwise, rescan and truncate
2128 	 * after the remaining size.
2129 	 */
2130 	count = 0;
2131 	m = mp;
2132 	for (;;) {
2133 		count += m->m_len;
2134 		if (m->m_next == (struct mbuf *)0)
2135 			break;
2136 		m = m->m_next;
2137 	}
2138 	if (m->m_len > len) {
2139 		m->m_len -= len;
2140 		if (nul > 0) {
2141 			cp = mtod(m, caddr_t)+m->m_len-nul;
2142 			for (i = 0; i < nul; i++)
2143 				*cp++ = '\0';
2144 		}
2145 		return;
2146 	}
2147 	count -= len;
2148 	if (count < 0)
2149 		count = 0;
2150 	/*
2151 	 * Correct length for chain is "count".
2152 	 * Find the mbuf with last data, adjust its length,
2153 	 * and toss data from remaining mbufs on chain.
2154 	 */
2155 	for (m = mp; m; m = m->m_next) {
2156 		if (m->m_len >= count) {
2157 			m->m_len = count;
2158 			if (nul > 0) {
2159 				cp = mtod(m, caddr_t)+m->m_len-nul;
2160 				for (i = 0; i < nul; i++)
2161 					*cp++ = '\0';
2162 			}
2163 			break;
2164 		}
2165 		count -= m->m_len;
2166 	}
2167 	for (m = m->m_next;m;m = m->m_next)
2168 		m->m_len = 0;
2169 }
2170 
2171 /*
2172  * Make these functions instead of macros, so that the kernel text size
2173  * doesn't get too big...
2174  */
2175 void
2176 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp)
2177 	struct nfsrv_descript *nfsd;
2178 	int before_ret;
2179 	struct vattr *before_vap;
2180 	int after_ret;
2181 	struct vattr *after_vap;
2182 	struct mbuf **mbp;
2183 	char **bposp;
2184 {
2185 	struct mbuf *mb = *mbp, *mb2;
2186 	char *bpos = *bposp;
2187 	u_int32_t *tl;
2188 
2189 	if (before_ret) {
2190 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2191 		*tl = nfs_false;
2192 	} else {
2193 		nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
2194 		*tl++ = nfs_true;
2195 		txdr_hyper(before_vap->va_size, tl);
2196 		tl += 2;
2197 		txdr_nfsv3time(&(before_vap->va_mtime), tl);
2198 		tl += 2;
2199 		txdr_nfsv3time(&(before_vap->va_ctime), tl);
2200 	}
2201 	*bposp = bpos;
2202 	*mbp = mb;
2203 	nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
2204 }
2205 
2206 void
2207 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp)
2208 	struct nfsrv_descript *nfsd;
2209 	int after_ret;
2210 	struct vattr *after_vap;
2211 	struct mbuf **mbp;
2212 	char **bposp;
2213 {
2214 	struct mbuf *mb = *mbp, *mb2;
2215 	char *bpos = *bposp;
2216 	u_int32_t *tl;
2217 	struct nfs_fattr *fp;
2218 
2219 	if (after_ret) {
2220 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2221 		*tl = nfs_false;
2222 	} else {
2223 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
2224 		*tl++ = nfs_true;
2225 		fp = (struct nfs_fattr *)tl;
2226 		nfsm_srvfattr(nfsd, after_vap, fp);
2227 	}
2228 	*mbp = mb;
2229 	*bposp = bpos;
2230 }
2231 
2232 void
2233 nfsm_srvfattr(nfsd, vap, fp)
2234 	struct nfsrv_descript *nfsd;
2235 	struct vattr *vap;
2236 	struct nfs_fattr *fp;
2237 {
2238 
2239 	fp->fa_nlink = txdr_unsigned(vap->va_nlink);
2240 	fp->fa_uid = txdr_unsigned(vap->va_uid);
2241 	fp->fa_gid = txdr_unsigned(vap->va_gid);
2242 	if (nfsd->nd_flag & ND_NFSV3) {
2243 		fp->fa_type = vtonfsv3_type(vap->va_type);
2244 		fp->fa_mode = vtonfsv3_mode(vap->va_mode);
2245 		txdr_hyper(vap->va_size, &fp->fa3_size);
2246 		txdr_hyper(vap->va_bytes, &fp->fa3_used);
2247 		fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
2248 		fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
2249 		fp->fa3_fsid.nfsuquad[0] = 0;
2250 		fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
2251 		fp->fa3_fileid.nfsuquad[0] = 0;
2252 		fp->fa3_fileid.nfsuquad[1] = txdr_unsigned(vap->va_fileid);
2253 		txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
2254 		txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
2255 		txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
2256 	} else {
2257 		fp->fa_type = vtonfsv2_type(vap->va_type);
2258 		fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
2259 		fp->fa2_size = txdr_unsigned(vap->va_size);
2260 		fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
2261 		if (vap->va_type == VFIFO)
2262 			fp->fa2_rdev = 0xffffffff;
2263 		else
2264 			fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
2265 		fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
2266 		fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
2267 		fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
2268 		txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
2269 		txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
2270 		txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
2271 	}
2272 }
2273 
2274 /*
2275  * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
2276  * 	- look up fsid in mount list (if not found ret error)
2277  *	- get vp and export rights by calling VFS_FHTOVP()
2278  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2279  *	- if not lockflag unlock it with VOP_UNLOCK()
2280  */
2281 int
2282 nfsrv_fhtovp(fhp, lockflag, vpp, cred, slp, nam, rdonlyp, kerbflag, pubflag)
2283 	fhandle_t *fhp;
2284 	int lockflag;
2285 	struct vnode **vpp;
2286 	struct ucred *cred;
2287 	struct nfssvc_sock *slp;
2288 	struct mbuf *nam;
2289 	int *rdonlyp;
2290 	int kerbflag;
2291 {
2292 	struct mount *mp;
2293 	int i;
2294 	struct ucred *credanon;
2295 	int error, exflags;
2296 	struct sockaddr_in *saddr;
2297 
2298 	*vpp = (struct vnode *)0;
2299 
2300 	if (nfs_ispublicfh(fhp)) {
2301 		if (!pubflag || !nfs_pub.np_valid)
2302 			return (ESTALE);
2303 		fhp = &nfs_pub.np_handle;
2304 	}
2305 
2306 	mp = vfs_getvfs(&fhp->fh_fsid);
2307 	if (!mp)
2308 		return (ESTALE);
2309 	error = VFS_CHECKEXP(mp, nam, &exflags, &credanon);
2310 	if (error)
2311 		return (error);
2312 	error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
2313 	if (error)
2314 		return (error);
2315 
2316 	if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
2317 		saddr = mtod(nam, struct sockaddr_in *);
2318 		if ((saddr->sin_family == AF_INET) &&
2319 		    ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
2320 			vput(*vpp);
2321 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2322 		}
2323 #ifdef INET6
2324 		if ((saddr->sin_family == AF_INET6) &&
2325 		    ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) {
2326 			vput(*vpp);
2327 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2328 		}
2329 #endif
2330 	}
2331 	/*
2332 	 * Check/setup credentials.
2333 	 */
2334 	if (exflags & MNT_EXKERB) {
2335 		if (!kerbflag) {
2336 			vput(*vpp);
2337 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2338 		}
2339 	} else if (kerbflag) {
2340 		vput(*vpp);
2341 		return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2342 	} else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
2343 		cred->cr_uid = credanon->cr_uid;
2344 		cred->cr_gid = credanon->cr_gid;
2345 		for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
2346 			cred->cr_groups[i] = credanon->cr_groups[i];
2347 		cred->cr_ngroups = i;
2348 	}
2349 	if (exflags & MNT_EXRDONLY)
2350 		*rdonlyp = 1;
2351 	else
2352 		*rdonlyp = 0;
2353 	if (!lockflag)
2354 		VOP_UNLOCK(*vpp, 0);
2355 	return (0);
2356 }
2357 
2358 /*
2359  * WebNFS: check if a filehandle is a public filehandle. For v3, this
2360  * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
2361  * transformed this to all zeroes in both cases, so check for it.
2362  */
2363 int
2364 nfs_ispublicfh(fhp)
2365 	fhandle_t *fhp;
2366 {
2367 	char *cp = (char *)fhp;
2368 	int i;
2369 
2370 	for (i = 0; i < NFSX_V3FH; i++)
2371 		if (*cp++ != 0)
2372 			return (FALSE);
2373 	return (TRUE);
2374 }
2375 
2376 /*
2377  * This function compares two net addresses by family and returns TRUE
2378  * if they are the same host.
2379  * If there is any doubt, return FALSE.
2380  * The AF_INET family is handled as a special case so that address mbufs
2381  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
2382  */
2383 int
2384 netaddr_match(family, haddr, nam)
2385 	int family;
2386 	union nethostaddr *haddr;
2387 	struct mbuf *nam;
2388 {
2389 	struct sockaddr_in *inetaddr;
2390 
2391 	switch (family) {
2392 	case AF_INET:
2393 		inetaddr = mtod(nam, struct sockaddr_in *);
2394 		if (inetaddr->sin_family == AF_INET &&
2395 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
2396 			return (1);
2397 		break;
2398 #ifdef INET6
2399 	case AF_INET6:
2400 	    {
2401 		struct sockaddr_in6 *sin6_1, *sin6_2;
2402 
2403 		sin6_1 = mtod(nam, struct sockaddr_in6 *);
2404 		sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
2405 		if (sin6_1->sin6_family == AF_INET6 &&
2406 		    IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
2407 			return 1;
2408 	    }
2409 #endif
2410 #ifdef ISO
2411 	case AF_ISO:
2412 	    {
2413 		struct sockaddr_iso *isoaddr1, *isoaddr2;
2414 
2415 		isoaddr1 = mtod(nam, struct sockaddr_iso *);
2416 		isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *);
2417 		if (isoaddr1->siso_family == AF_ISO &&
2418 		    isoaddr1->siso_nlen > 0 &&
2419 		    isoaddr1->siso_nlen == isoaddr2->siso_nlen &&
2420 		    SAME_ISOADDR(isoaddr1, isoaddr2))
2421 			return (1);
2422 		break;
2423 	    }
2424 #endif	/* ISO */
2425 	default:
2426 		break;
2427 	};
2428 	return (0);
2429 }
2430 
2431 /*
2432  * The write verifier has changed (probably due to a server reboot), so all
2433  * B_NEEDCOMMIT blocks will have to be written again. Since they are on the
2434  * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT
2435  * flag. Once done the new write verifier can be set for the mount point.
2436  */
2437 void
2438 nfs_clearcommit(mp)
2439 	struct mount *mp;
2440 {
2441 	struct vnode *vp;
2442 	struct nfsnode *np;
2443 	struct vm_page *pg;
2444 	int s;
2445 
2446 	s = splbio();
2447 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2448 		KASSERT(vp->v_mount == mp);
2449 		if (vp->v_type == VNON)
2450 			continue;
2451 		np = VTONFS(vp);
2452 		np->n_pushlo = np->n_pushhi = np->n_pushedlo =
2453 		    np->n_pushedhi = 0;
2454 		np->n_commitflags &=
2455 		    ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
2456 		simple_lock(&vp->v_uobj.vmobjlock);
2457 		TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) {
2458 			pg->flags &= ~PG_NEEDCOMMIT;
2459 		}
2460 		simple_unlock(&vp->v_uobj.vmobjlock);
2461 	}
2462 	splx(s);
2463 }
2464 
2465 void
2466 nfs_merge_commit_ranges(vp)
2467 	struct vnode *vp;
2468 {
2469 	struct nfsnode *np = VTONFS(vp);
2470 
2471 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2472 		np->n_pushedlo = np->n_pushlo;
2473 		np->n_pushedhi = np->n_pushhi;
2474 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2475 	} else {
2476 		if (np->n_pushlo < np->n_pushedlo)
2477 			np->n_pushedlo = np->n_pushlo;
2478 		if (np->n_pushhi > np->n_pushedhi)
2479 			np->n_pushedhi = np->n_pushhi;
2480 	}
2481 
2482 	np->n_pushlo = np->n_pushhi = 0;
2483 	np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
2484 
2485 #ifdef fvdl_debug
2486 	printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2487 	    (unsigned)np->n_pushedhi);
2488 #endif
2489 }
2490 
2491 int
2492 nfs_in_committed_range(vp, off, len)
2493 	struct vnode *vp;
2494 	off_t off, len;
2495 {
2496 	struct nfsnode *np = VTONFS(vp);
2497 	off_t lo, hi;
2498 
2499 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2500 		return 0;
2501 	lo = off;
2502 	hi = lo + len;
2503 
2504 	return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
2505 }
2506 
2507 int
2508 nfs_in_tobecommitted_range(vp, off, len)
2509 	struct vnode *vp;
2510 	off_t off, len;
2511 {
2512 	struct nfsnode *np = VTONFS(vp);
2513 	off_t lo, hi;
2514 
2515 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2516 		return 0;
2517 	lo = off;
2518 	hi = lo + len;
2519 
2520 	return (lo >= np->n_pushlo && hi <= np->n_pushhi);
2521 }
2522 
2523 void
2524 nfs_add_committed_range(vp, off, len)
2525 	struct vnode *vp;
2526 	off_t off, len;
2527 {
2528 	struct nfsnode *np = VTONFS(vp);
2529 	off_t lo, hi;
2530 
2531 	lo = off;
2532 	hi = lo + len;
2533 
2534 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2535 		np->n_pushedlo = lo;
2536 		np->n_pushedhi = hi;
2537 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2538 	} else {
2539 		if (hi > np->n_pushedhi)
2540 			np->n_pushedhi = hi;
2541 		if (lo < np->n_pushedlo)
2542 			np->n_pushedlo = lo;
2543 	}
2544 #ifdef fvdl_debug
2545 	printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2546 	    (unsigned)np->n_pushedhi);
2547 #endif
2548 }
2549 
2550 void
2551 nfs_del_committed_range(vp, off, len)
2552 	struct vnode *vp;
2553 	off_t off, len;
2554 {
2555 	struct nfsnode *np = VTONFS(vp);
2556 	off_t lo, hi;
2557 
2558 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2559 		return;
2560 
2561 	lo = off;
2562 	hi = lo + len;
2563 
2564 	if (lo > np->n_pushedhi || hi < np->n_pushedlo)
2565 		return;
2566 	if (lo <= np->n_pushedlo)
2567 		np->n_pushedlo = hi;
2568 	else if (hi >= np->n_pushedhi)
2569 		np->n_pushedhi = lo;
2570 	else {
2571 		/*
2572 		 * XXX There's only one range. If the deleted range
2573 		 * is in the middle, pick the largest of the
2574 		 * contiguous ranges that it leaves.
2575 		 */
2576 		if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
2577 			np->n_pushedhi = lo;
2578 		else
2579 			np->n_pushedlo = hi;
2580 	}
2581 #ifdef fvdl_debug
2582 	printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2583 	    (unsigned)np->n_pushedhi);
2584 #endif
2585 }
2586 
2587 void
2588 nfs_add_tobecommitted_range(vp, off, len)
2589 	struct vnode *vp;
2590 	off_t off, len;
2591 {
2592 	struct nfsnode *np = VTONFS(vp);
2593 	off_t lo, hi;
2594 
2595 	lo = off;
2596 	hi = lo + len;
2597 
2598 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
2599 		np->n_pushlo = lo;
2600 		np->n_pushhi = hi;
2601 		np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
2602 	} else {
2603 		if (lo < np->n_pushlo)
2604 			np->n_pushlo = lo;
2605 		if (hi > np->n_pushhi)
2606 			np->n_pushhi = hi;
2607 	}
2608 #ifdef fvdl_debug
2609 	printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2610 	    (unsigned)np->n_pushhi);
2611 #endif
2612 }
2613 
2614 void
2615 nfs_del_tobecommitted_range(vp, off, len)
2616 	struct vnode *vp;
2617 	off_t off, len;
2618 {
2619 	struct nfsnode *np = VTONFS(vp);
2620 	off_t lo, hi;
2621 
2622 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2623 		return;
2624 
2625 	lo = off;
2626 	hi = lo + len;
2627 
2628 	if (lo > np->n_pushhi || hi < np->n_pushlo)
2629 		return;
2630 
2631 	if (lo <= np->n_pushlo)
2632 		np->n_pushlo = hi;
2633 	else if (hi >= np->n_pushhi)
2634 		np->n_pushhi = lo;
2635 	else {
2636 		/*
2637 		 * XXX There's only one range. If the deleted range
2638 		 * is in the middle, pick the largest of the
2639 		 * contiguous ranges that it leaves.
2640 		 */
2641 		if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
2642 			np->n_pushhi = lo;
2643 		else
2644 			np->n_pushlo = hi;
2645 	}
2646 #ifdef fvdl_debug
2647 	printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2648 	    (unsigned)np->n_pushhi);
2649 #endif
2650 }
2651 
2652 /*
2653  * Map errnos to NFS error numbers. For Version 3 also filter out error
2654  * numbers not specified for the associated procedure.
2655  */
2656 int
2657 nfsrv_errmap(nd, err)
2658 	struct nfsrv_descript *nd;
2659 	int err;
2660 {
2661 	const short *defaulterrp, *errp;
2662 
2663 	if (nd->nd_flag & ND_NFSV3) {
2664 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
2665 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2666 		while (*++errp) {
2667 			if (*errp == err)
2668 				return (err);
2669 			else if (*errp > err)
2670 				break;
2671 		}
2672 		return ((int)*defaulterrp);
2673 	    } else
2674 		return (err & 0xffff);
2675 	}
2676 	if (err <= ELAST)
2677 		return ((int)nfsrv_v2errmap[err - 1]);
2678 	return (NFSERR_IO);
2679 }
2680 
2681 /*
2682  * Sort the group list in increasing numerical order.
2683  * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
2684  *  that used to be here.)
2685  */
2686 void
2687 nfsrvw_sort(list, num)
2688         gid_t *list;
2689         int num;
2690 {
2691 	int i, j;
2692 	gid_t v;
2693 
2694 	/* Insertion sort. */
2695 	for (i = 1; i < num; i++) {
2696 		v = list[i];
2697 		/* find correct slot for value v, moving others up */
2698 		for (j = i; --j >= 0 && v < list[j];)
2699 			list[j + 1] = list[j];
2700 		list[j + 1] = v;
2701 	}
2702 }
2703 
2704 /*
2705  * copy credentials making sure that the result can be compared with memcmp().
2706  */
2707 void
2708 nfsrv_setcred(incred, outcred)
2709 	struct ucred *incred, *outcred;
2710 {
2711 	int i;
2712 
2713 	memset((caddr_t)outcred, 0, sizeof (struct ucred));
2714 	outcred->cr_ref = 1;
2715 	outcred->cr_uid = incred->cr_uid;
2716 	outcred->cr_gid = incred->cr_gid;
2717 	outcred->cr_ngroups = incred->cr_ngroups;
2718 	for (i = 0; i < incred->cr_ngroups; i++)
2719 		outcred->cr_groups[i] = incred->cr_groups[i];
2720 	nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
2721 }
2722