xref: /netbsd-src/sys/nfs/nfs_subs.c (revision 1ffa7b76c40339c17a0fb2a09fac93f287cfc046)
1 /*	$NetBSD: nfs_subs.c,v 1.118 2003/05/03 18:07:42 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. 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 1.118 2003/05/03 18:07:42 yamt 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 MALLOC_DEFINE(M_NFSDIROFF, "NFS diroff", "NFS directory cookies");
149 
150 /* NFS client/server stats. */
151 struct nfsstats nfsstats;
152 
153 /*
154  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
155  */
156 const int nfsv3_procid[NFS_NPROCS] = {
157 	NFSPROC_NULL,
158 	NFSPROC_GETATTR,
159 	NFSPROC_SETATTR,
160 	NFSPROC_NOOP,
161 	NFSPROC_LOOKUP,
162 	NFSPROC_READLINK,
163 	NFSPROC_READ,
164 	NFSPROC_NOOP,
165 	NFSPROC_WRITE,
166 	NFSPROC_CREATE,
167 	NFSPROC_REMOVE,
168 	NFSPROC_RENAME,
169 	NFSPROC_LINK,
170 	NFSPROC_SYMLINK,
171 	NFSPROC_MKDIR,
172 	NFSPROC_RMDIR,
173 	NFSPROC_READDIR,
174 	NFSPROC_FSSTAT,
175 	NFSPROC_NOOP,
176 	NFSPROC_NOOP,
177 	NFSPROC_NOOP,
178 	NFSPROC_NOOP,
179 	NFSPROC_NOOP,
180 	NFSPROC_NOOP,
181 	NFSPROC_NOOP,
182 	NFSPROC_NOOP
183 };
184 
185 /*
186  * and the reverse mapping from generic to Version 2 procedure numbers
187  */
188 const int nfsv2_procid[NFS_NPROCS] = {
189 	NFSV2PROC_NULL,
190 	NFSV2PROC_GETATTR,
191 	NFSV2PROC_SETATTR,
192 	NFSV2PROC_LOOKUP,
193 	NFSV2PROC_NOOP,
194 	NFSV2PROC_READLINK,
195 	NFSV2PROC_READ,
196 	NFSV2PROC_WRITE,
197 	NFSV2PROC_CREATE,
198 	NFSV2PROC_MKDIR,
199 	NFSV2PROC_SYMLINK,
200 	NFSV2PROC_CREATE,
201 	NFSV2PROC_REMOVE,
202 	NFSV2PROC_RMDIR,
203 	NFSV2PROC_RENAME,
204 	NFSV2PROC_LINK,
205 	NFSV2PROC_READDIR,
206 	NFSV2PROC_NOOP,
207 	NFSV2PROC_STATFS,
208 	NFSV2PROC_NOOP,
209 	NFSV2PROC_NOOP,
210 	NFSV2PROC_NOOP,
211 	NFSV2PROC_NOOP,
212 	NFSV2PROC_NOOP,
213 	NFSV2PROC_NOOP,
214 	NFSV2PROC_NOOP,
215 };
216 
217 /*
218  * Maps errno values to nfs error numbers.
219  * Use NFSERR_IO as the catch all for ones not specifically defined in
220  * RFC 1094.
221  */
222 static const u_char nfsrv_v2errmap[ELAST] = {
223   NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
224   NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
225   NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
226   NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
227   NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
228   NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
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_IO,	NFSERR_IO,	NFSERR_IO,
234   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
235   NFSERR_IO,	NFSERR_IO,	NFSERR_NAMETOL,	NFSERR_IO,	NFSERR_IO,
236   NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
237   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
238   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
239   NFSERR_IO,	NFSERR_IO,
240 };
241 
242 /*
243  * Maps errno values to nfs error numbers.
244  * Although it is not obvious whether or not NFS clients really care if
245  * a returned error value is in the specified list for the procedure, the
246  * safest thing to do is filter them appropriately. For Version 2, the
247  * X/Open XNFS document is the only specification that defines error values
248  * for each RPC (The RFC simply lists all possible error values for all RPCs),
249  * so I have decided to not do this for Version 2.
250  * The first entry is the default error return and the rest are the valid
251  * errors for that RPC in increasing numeric order.
252  */
253 static const short nfsv3err_null[] = {
254 	0,
255 	0,
256 };
257 
258 static const short nfsv3err_getattr[] = {
259 	NFSERR_IO,
260 	NFSERR_IO,
261 	NFSERR_STALE,
262 	NFSERR_BADHANDLE,
263 	NFSERR_SERVERFAULT,
264 	0,
265 };
266 
267 static const short nfsv3err_setattr[] = {
268 	NFSERR_IO,
269 	NFSERR_PERM,
270 	NFSERR_IO,
271 	NFSERR_ACCES,
272 	NFSERR_INVAL,
273 	NFSERR_NOSPC,
274 	NFSERR_ROFS,
275 	NFSERR_DQUOT,
276 	NFSERR_STALE,
277 	NFSERR_BADHANDLE,
278 	NFSERR_NOT_SYNC,
279 	NFSERR_SERVERFAULT,
280 	0,
281 };
282 
283 static const short nfsv3err_lookup[] = {
284 	NFSERR_IO,
285 	NFSERR_NOENT,
286 	NFSERR_IO,
287 	NFSERR_ACCES,
288 	NFSERR_NOTDIR,
289 	NFSERR_NAMETOL,
290 	NFSERR_STALE,
291 	NFSERR_BADHANDLE,
292 	NFSERR_SERVERFAULT,
293 	0,
294 };
295 
296 static const short nfsv3err_access[] = {
297 	NFSERR_IO,
298 	NFSERR_IO,
299 	NFSERR_STALE,
300 	NFSERR_BADHANDLE,
301 	NFSERR_SERVERFAULT,
302 	0,
303 };
304 
305 static const short nfsv3err_readlink[] = {
306 	NFSERR_IO,
307 	NFSERR_IO,
308 	NFSERR_ACCES,
309 	NFSERR_INVAL,
310 	NFSERR_STALE,
311 	NFSERR_BADHANDLE,
312 	NFSERR_NOTSUPP,
313 	NFSERR_SERVERFAULT,
314 	0,
315 };
316 
317 static const short nfsv3err_read[] = {
318 	NFSERR_IO,
319 	NFSERR_IO,
320 	NFSERR_NXIO,
321 	NFSERR_ACCES,
322 	NFSERR_INVAL,
323 	NFSERR_STALE,
324 	NFSERR_BADHANDLE,
325 	NFSERR_SERVERFAULT,
326 	NFSERR_JUKEBOX,
327 	0,
328 };
329 
330 static const short nfsv3err_write[] = {
331 	NFSERR_IO,
332 	NFSERR_IO,
333 	NFSERR_ACCES,
334 	NFSERR_INVAL,
335 	NFSERR_FBIG,
336 	NFSERR_NOSPC,
337 	NFSERR_ROFS,
338 	NFSERR_DQUOT,
339 	NFSERR_STALE,
340 	NFSERR_BADHANDLE,
341 	NFSERR_SERVERFAULT,
342 	NFSERR_JUKEBOX,
343 	0,
344 };
345 
346 static const short nfsv3err_create[] = {
347 	NFSERR_IO,
348 	NFSERR_IO,
349 	NFSERR_ACCES,
350 	NFSERR_EXIST,
351 	NFSERR_NOTDIR,
352 	NFSERR_NOSPC,
353 	NFSERR_ROFS,
354 	NFSERR_NAMETOL,
355 	NFSERR_DQUOT,
356 	NFSERR_STALE,
357 	NFSERR_BADHANDLE,
358 	NFSERR_NOTSUPP,
359 	NFSERR_SERVERFAULT,
360 	0,
361 };
362 
363 static const short nfsv3err_mkdir[] = {
364 	NFSERR_IO,
365 	NFSERR_IO,
366 	NFSERR_ACCES,
367 	NFSERR_EXIST,
368 	NFSERR_NOTDIR,
369 	NFSERR_NOSPC,
370 	NFSERR_ROFS,
371 	NFSERR_NAMETOL,
372 	NFSERR_DQUOT,
373 	NFSERR_STALE,
374 	NFSERR_BADHANDLE,
375 	NFSERR_NOTSUPP,
376 	NFSERR_SERVERFAULT,
377 	0,
378 };
379 
380 static const short nfsv3err_symlink[] = {
381 	NFSERR_IO,
382 	NFSERR_IO,
383 	NFSERR_ACCES,
384 	NFSERR_EXIST,
385 	NFSERR_NOTDIR,
386 	NFSERR_NOSPC,
387 	NFSERR_ROFS,
388 	NFSERR_NAMETOL,
389 	NFSERR_DQUOT,
390 	NFSERR_STALE,
391 	NFSERR_BADHANDLE,
392 	NFSERR_NOTSUPP,
393 	NFSERR_SERVERFAULT,
394 	0,
395 };
396 
397 static const short nfsv3err_mknod[] = {
398 	NFSERR_IO,
399 	NFSERR_IO,
400 	NFSERR_ACCES,
401 	NFSERR_EXIST,
402 	NFSERR_NOTDIR,
403 	NFSERR_NOSPC,
404 	NFSERR_ROFS,
405 	NFSERR_NAMETOL,
406 	NFSERR_DQUOT,
407 	NFSERR_STALE,
408 	NFSERR_BADHANDLE,
409 	NFSERR_NOTSUPP,
410 	NFSERR_SERVERFAULT,
411 	NFSERR_BADTYPE,
412 	0,
413 };
414 
415 static const short nfsv3err_remove[] = {
416 	NFSERR_IO,
417 	NFSERR_NOENT,
418 	NFSERR_IO,
419 	NFSERR_ACCES,
420 	NFSERR_NOTDIR,
421 	NFSERR_ROFS,
422 	NFSERR_NAMETOL,
423 	NFSERR_STALE,
424 	NFSERR_BADHANDLE,
425 	NFSERR_SERVERFAULT,
426 	0,
427 };
428 
429 static const short nfsv3err_rmdir[] = {
430 	NFSERR_IO,
431 	NFSERR_NOENT,
432 	NFSERR_IO,
433 	NFSERR_ACCES,
434 	NFSERR_EXIST,
435 	NFSERR_NOTDIR,
436 	NFSERR_INVAL,
437 	NFSERR_ROFS,
438 	NFSERR_NAMETOL,
439 	NFSERR_NOTEMPTY,
440 	NFSERR_STALE,
441 	NFSERR_BADHANDLE,
442 	NFSERR_NOTSUPP,
443 	NFSERR_SERVERFAULT,
444 	0,
445 };
446 
447 static const short nfsv3err_rename[] = {
448 	NFSERR_IO,
449 	NFSERR_NOENT,
450 	NFSERR_IO,
451 	NFSERR_ACCES,
452 	NFSERR_EXIST,
453 	NFSERR_XDEV,
454 	NFSERR_NOTDIR,
455 	NFSERR_ISDIR,
456 	NFSERR_INVAL,
457 	NFSERR_NOSPC,
458 	NFSERR_ROFS,
459 	NFSERR_MLINK,
460 	NFSERR_NAMETOL,
461 	NFSERR_NOTEMPTY,
462 	NFSERR_DQUOT,
463 	NFSERR_STALE,
464 	NFSERR_BADHANDLE,
465 	NFSERR_NOTSUPP,
466 	NFSERR_SERVERFAULT,
467 	0,
468 };
469 
470 static const short nfsv3err_link[] = {
471 	NFSERR_IO,
472 	NFSERR_IO,
473 	NFSERR_ACCES,
474 	NFSERR_EXIST,
475 	NFSERR_XDEV,
476 	NFSERR_NOTDIR,
477 	NFSERR_INVAL,
478 	NFSERR_NOSPC,
479 	NFSERR_ROFS,
480 	NFSERR_MLINK,
481 	NFSERR_NAMETOL,
482 	NFSERR_DQUOT,
483 	NFSERR_STALE,
484 	NFSERR_BADHANDLE,
485 	NFSERR_NOTSUPP,
486 	NFSERR_SERVERFAULT,
487 	0,
488 };
489 
490 static const short nfsv3err_readdir[] = {
491 	NFSERR_IO,
492 	NFSERR_IO,
493 	NFSERR_ACCES,
494 	NFSERR_NOTDIR,
495 	NFSERR_STALE,
496 	NFSERR_BADHANDLE,
497 	NFSERR_BAD_COOKIE,
498 	NFSERR_TOOSMALL,
499 	NFSERR_SERVERFAULT,
500 	0,
501 };
502 
503 static const short nfsv3err_readdirplus[] = {
504 	NFSERR_IO,
505 	NFSERR_IO,
506 	NFSERR_ACCES,
507 	NFSERR_NOTDIR,
508 	NFSERR_STALE,
509 	NFSERR_BADHANDLE,
510 	NFSERR_BAD_COOKIE,
511 	NFSERR_NOTSUPP,
512 	NFSERR_TOOSMALL,
513 	NFSERR_SERVERFAULT,
514 	0,
515 };
516 
517 static const short nfsv3err_fsstat[] = {
518 	NFSERR_IO,
519 	NFSERR_IO,
520 	NFSERR_STALE,
521 	NFSERR_BADHANDLE,
522 	NFSERR_SERVERFAULT,
523 	0,
524 };
525 
526 static const short nfsv3err_fsinfo[] = {
527 	NFSERR_STALE,
528 	NFSERR_STALE,
529 	NFSERR_BADHANDLE,
530 	NFSERR_SERVERFAULT,
531 	0,
532 };
533 
534 static const short nfsv3err_pathconf[] = {
535 	NFSERR_STALE,
536 	NFSERR_STALE,
537 	NFSERR_BADHANDLE,
538 	NFSERR_SERVERFAULT,
539 	0,
540 };
541 
542 static const short nfsv3err_commit[] = {
543 	NFSERR_IO,
544 	NFSERR_IO,
545 	NFSERR_STALE,
546 	NFSERR_BADHANDLE,
547 	NFSERR_SERVERFAULT,
548 	0,
549 };
550 
551 static const short * const nfsrv_v3errmap[] = {
552 	nfsv3err_null,
553 	nfsv3err_getattr,
554 	nfsv3err_setattr,
555 	nfsv3err_lookup,
556 	nfsv3err_access,
557 	nfsv3err_readlink,
558 	nfsv3err_read,
559 	nfsv3err_write,
560 	nfsv3err_create,
561 	nfsv3err_mkdir,
562 	nfsv3err_symlink,
563 	nfsv3err_mknod,
564 	nfsv3err_remove,
565 	nfsv3err_rmdir,
566 	nfsv3err_rename,
567 	nfsv3err_link,
568 	nfsv3err_readdir,
569 	nfsv3err_readdirplus,
570 	nfsv3err_fsstat,
571 	nfsv3err_fsinfo,
572 	nfsv3err_pathconf,
573 	nfsv3err_commit,
574 };
575 
576 extern struct nfsrtt nfsrtt;
577 extern time_t nqnfsstarttime;
578 extern int nqsrv_clockskew;
579 extern int nqsrv_writeslack;
580 extern int nqsrv_maxlease;
581 extern const int nqnfs_piggy[NFS_NPROCS];
582 extern struct nfsnodehashhead *nfsnodehashtbl;
583 extern u_long nfsnodehash;
584 
585 u_long nfsdirhashmask;
586 
587 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *));
588 
589 /*
590  * Create the header for an rpc request packet
591  * The hsiz is the size of the rest of the nfs request header.
592  * (just used to decide if a cluster is a good idea)
593  */
594 struct mbuf *
595 nfsm_reqh(np, procid, hsiz, bposp)
596 	struct nfsnode *np;
597 	u_long procid;
598 	int hsiz;
599 	caddr_t *bposp;
600 {
601 	struct mbuf *mb;
602 	caddr_t bpos;
603 #ifndef NFS_V2_ONLY
604 	struct nfsmount *nmp;
605 	u_int32_t *tl;
606 	int nqflag;
607 #endif
608 
609 	mb = m_get(M_WAIT, MT_DATA);
610 	MCLAIM(mb, &nfs_mowner);
611 	if (hsiz >= MINCLSIZE)
612 		m_clget(mb, M_WAIT);
613 	mb->m_len = 0;
614 	bpos = mtod(mb, caddr_t);
615 
616 #ifndef NFS_V2_ONLY
617 	/*
618 	 * For NQNFS, add lease request.
619 	 */
620 	if (np) {
621 		nmp = VFSTONFS(np->n_vnode->v_mount);
622 		if (nmp->nm_flag & NFSMNT_NQNFS) {
623 			nqflag = NQNFS_NEEDLEASE(np, procid);
624 			if (nqflag) {
625 				nfsm_build(tl, u_int32_t *, 2*NFSX_UNSIGNED);
626 				*tl++ = txdr_unsigned(nqflag);
627 				*tl = txdr_unsigned(nmp->nm_leaseterm);
628 			} else {
629 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
630 				*tl = 0;
631 			}
632 		}
633 	}
634 #endif
635 	/* Finally, return values */
636 	*bposp = bpos;
637 	return (mb);
638 }
639 
640 /*
641  * Build the RPC header and fill in the authorization info.
642  * The authorization string argument is only used when the credentials
643  * come from outside of the kernel.
644  * Returns the head of the mbuf list.
645  */
646 struct mbuf *
647 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len,
648 	verf_str, mrest, mrest_len, mbp, xidp)
649 	struct ucred *cr;
650 	int nmflag;
651 	int procid;
652 	int auth_type;
653 	int auth_len;
654 	char *auth_str;
655 	int verf_len;
656 	char *verf_str;
657 	struct mbuf *mrest;
658 	int mrest_len;
659 	struct mbuf **mbp;
660 	u_int32_t *xidp;
661 {
662 	struct mbuf *mb;
663 	u_int32_t *tl;
664 	caddr_t bpos;
665 	int i;
666 	struct mbuf *mreq;
667 	int siz, grpsiz, authsiz;
668 	struct timeval tv;
669 	static u_int32_t base;
670 
671 	authsiz = nfsm_rndup(auth_len);
672 	mb = m_gethdr(M_WAIT, MT_DATA);
673 	MCLAIM(mb, &nfs_mowner);
674 	if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
675 		m_clget(mb, M_WAIT);
676 	} else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
677 		MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED);
678 	} else {
679 		MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
680 	}
681 	mb->m_len = 0;
682 	mreq = mb;
683 	bpos = mtod(mb, caddr_t);
684 
685 	/*
686 	 * First the RPC header.
687 	 */
688 	nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
689 
690 	/*
691 	 * derive initial xid from system time
692 	 * XXX time is invalid if root not yet mounted
693 	 */
694 	if (!base && (rootvp)) {
695 		microtime(&tv);
696 		base = tv.tv_sec << 12;
697 		nfs_xid = base;
698 	}
699 	/*
700 	 * Skip zero xid if it should ever happen.
701 	 */
702 	if (++nfs_xid == 0)
703 		nfs_xid++;
704 
705 	*tl++ = *xidp = txdr_unsigned(nfs_xid);
706 	*tl++ = rpc_call;
707 	*tl++ = rpc_vers;
708 	if (nmflag & NFSMNT_NQNFS) {
709 		*tl++ = txdr_unsigned(NQNFS_PROG);
710 		*tl++ = txdr_unsigned(NQNFS_VER3);
711 	} else {
712 		*tl++ = txdr_unsigned(NFS_PROG);
713 		if (nmflag & NFSMNT_NFSV3)
714 			*tl++ = txdr_unsigned(NFS_VER3);
715 		else
716 			*tl++ = txdr_unsigned(NFS_VER2);
717 	}
718 	if (nmflag & NFSMNT_NFSV3)
719 		*tl++ = txdr_unsigned(procid);
720 	else
721 		*tl++ = txdr_unsigned(nfsv2_procid[procid]);
722 
723 	/*
724 	 * And then the authorization cred.
725 	 */
726 	*tl++ = txdr_unsigned(auth_type);
727 	*tl = txdr_unsigned(authsiz);
728 	switch (auth_type) {
729 	case RPCAUTH_UNIX:
730 		nfsm_build(tl, u_int32_t *, auth_len);
731 		*tl++ = 0;		/* stamp ?? */
732 		*tl++ = 0;		/* NULL hostname */
733 		*tl++ = txdr_unsigned(cr->cr_uid);
734 		*tl++ = txdr_unsigned(cr->cr_gid);
735 		grpsiz = (auth_len >> 2) - 5;
736 		*tl++ = txdr_unsigned(grpsiz);
737 		for (i = 0; i < grpsiz; i++)
738 			*tl++ = txdr_unsigned(cr->cr_groups[i]);
739 		break;
740 	case RPCAUTH_KERB4:
741 		siz = auth_len;
742 		while (siz > 0) {
743 			if (M_TRAILINGSPACE(mb) == 0) {
744 				struct mbuf *mb2;
745 				mb2 = m_get(M_WAIT, MT_DATA);
746 				MCLAIM(mb2, &nfs_mowner);
747 				if (siz >= MINCLSIZE)
748 					m_clget(mb2, M_WAIT);
749 				mb->m_next = mb2;
750 				mb = mb2;
751 				mb->m_len = 0;
752 				bpos = mtod(mb, caddr_t);
753 			}
754 			i = min(siz, M_TRAILINGSPACE(mb));
755 			memcpy(bpos, auth_str, i);
756 			mb->m_len += i;
757 			auth_str += i;
758 			bpos += i;
759 			siz -= i;
760 		}
761 		if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
762 			for (i = 0; i < siz; i++)
763 				*bpos++ = '\0';
764 			mb->m_len += siz;
765 		}
766 		break;
767 	};
768 
769 	/*
770 	 * And the verifier...
771 	 */
772 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
773 	if (verf_str) {
774 		*tl++ = txdr_unsigned(RPCAUTH_KERB4);
775 		*tl = txdr_unsigned(verf_len);
776 		siz = verf_len;
777 		while (siz > 0) {
778 			if (M_TRAILINGSPACE(mb) == 0) {
779 				struct mbuf *mb2;
780 				mb2 = m_get(M_WAIT, MT_DATA);
781 				MCLAIM(mb2, &nfs_mowner);
782 				if (siz >= MINCLSIZE)
783 					m_clget(mb2, M_WAIT);
784 				mb->m_next = mb2;
785 				mb = mb2;
786 				mb->m_len = 0;
787 				bpos = mtod(mb, caddr_t);
788 			}
789 			i = min(siz, M_TRAILINGSPACE(mb));
790 			memcpy(bpos, verf_str, i);
791 			mb->m_len += i;
792 			verf_str += i;
793 			bpos += i;
794 			siz -= i;
795 		}
796 		if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
797 			for (i = 0; i < siz; i++)
798 				*bpos++ = '\0';
799 			mb->m_len += siz;
800 		}
801 	} else {
802 		*tl++ = txdr_unsigned(RPCAUTH_NULL);
803 		*tl = 0;
804 	}
805 	mb->m_next = mrest;
806 	mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
807 	mreq->m_pkthdr.rcvif = (struct ifnet *)0;
808 	*mbp = mb;
809 	return (mreq);
810 }
811 
812 /*
813  * copies mbuf chain to the uio scatter/gather list
814  */
815 int
816 nfsm_mbuftouio(mrep, uiop, siz, dpos)
817 	struct mbuf **mrep;
818 	struct uio *uiop;
819 	int siz;
820 	caddr_t *dpos;
821 {
822 	char *mbufcp, *uiocp;
823 	int xfer, left, len;
824 	struct mbuf *mp;
825 	long uiosiz, rem;
826 	int error = 0;
827 
828 	mp = *mrep;
829 	mbufcp = *dpos;
830 	len = mtod(mp, caddr_t)+mp->m_len-mbufcp;
831 	rem = nfsm_rndup(siz)-siz;
832 	while (siz > 0) {
833 		if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
834 			return (EFBIG);
835 		left = uiop->uio_iov->iov_len;
836 		uiocp = uiop->uio_iov->iov_base;
837 		if (left > siz)
838 			left = siz;
839 		uiosiz = left;
840 		while (left > 0) {
841 			while (len == 0) {
842 				mp = mp->m_next;
843 				if (mp == NULL)
844 					return (EBADRPC);
845 				mbufcp = mtod(mp, caddr_t);
846 				len = mp->m_len;
847 			}
848 			xfer = (left > len) ? len : left;
849 #ifdef notdef
850 			/* Not Yet.. */
851 			if (uiop->uio_iov->iov_op != NULL)
852 				(*(uiop->uio_iov->iov_op))
853 				(mbufcp, uiocp, xfer);
854 			else
855 #endif
856 			if (uiop->uio_segflg == UIO_SYSSPACE)
857 				memcpy(uiocp, mbufcp, xfer);
858 			else
859 				copyout(mbufcp, uiocp, xfer);
860 			left -= xfer;
861 			len -= xfer;
862 			mbufcp += xfer;
863 			uiocp += xfer;
864 			uiop->uio_offset += xfer;
865 			uiop->uio_resid -= xfer;
866 		}
867 		if (uiop->uio_iov->iov_len <= siz) {
868 			uiop->uio_iovcnt--;
869 			uiop->uio_iov++;
870 		} else {
871 			uiop->uio_iov->iov_base =
872 			    (caddr_t)uiop->uio_iov->iov_base + uiosiz;
873 			uiop->uio_iov->iov_len -= uiosiz;
874 		}
875 		siz -= uiosiz;
876 	}
877 	*dpos = mbufcp;
878 	*mrep = mp;
879 	if (rem > 0) {
880 		if (len < rem)
881 			error = nfs_adv(mrep, dpos, rem, len);
882 		else
883 			*dpos += rem;
884 	}
885 	return (error);
886 }
887 
888 /*
889  * copies a uio scatter/gather list to an mbuf chain.
890  * NOTE: can ony handle iovcnt == 1
891  */
892 int
893 nfsm_uiotombuf(uiop, mq, siz, bpos)
894 	struct uio *uiop;
895 	struct mbuf **mq;
896 	int siz;
897 	caddr_t *bpos;
898 {
899 	char *uiocp;
900 	struct mbuf *mp, *mp2;
901 	int xfer, left, mlen;
902 	int uiosiz, clflg, rem;
903 	char *cp;
904 
905 #ifdef DIAGNOSTIC
906 	if (uiop->uio_iovcnt != 1)
907 		panic("nfsm_uiotombuf: iovcnt != 1");
908 #endif
909 
910 	if (siz > MLEN)		/* or should it >= MCLBYTES ?? */
911 		clflg = 1;
912 	else
913 		clflg = 0;
914 	rem = nfsm_rndup(siz)-siz;
915 	mp = mp2 = *mq;
916 	while (siz > 0) {
917 		left = uiop->uio_iov->iov_len;
918 		uiocp = uiop->uio_iov->iov_base;
919 		if (left > siz)
920 			left = siz;
921 		uiosiz = left;
922 		while (left > 0) {
923 			mlen = M_TRAILINGSPACE(mp);
924 			if (mlen == 0) {
925 				mp = m_get(M_WAIT, MT_DATA);
926 				MCLAIM(mp, &nfs_mowner);
927 				if (clflg)
928 					m_clget(mp, M_WAIT);
929 				mp->m_len = 0;
930 				mp2->m_next = mp;
931 				mp2 = mp;
932 				mlen = M_TRAILINGSPACE(mp);
933 			}
934 			xfer = (left > mlen) ? mlen : left;
935 #ifdef notdef
936 			/* Not Yet.. */
937 			if (uiop->uio_iov->iov_op != NULL)
938 				(*(uiop->uio_iov->iov_op))
939 				(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
940 			else
941 #endif
942 			if (uiop->uio_segflg == UIO_SYSSPACE)
943 				memcpy(mtod(mp, caddr_t)+mp->m_len, uiocp, xfer);
944 			else
945 				copyin(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
946 			mp->m_len += xfer;
947 			left -= xfer;
948 			uiocp += xfer;
949 			uiop->uio_offset += xfer;
950 			uiop->uio_resid -= xfer;
951 		}
952 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
953 		    uiosiz;
954 		uiop->uio_iov->iov_len -= uiosiz;
955 		siz -= uiosiz;
956 	}
957 	if (rem > 0) {
958 		if (rem > M_TRAILINGSPACE(mp)) {
959 			mp = m_get(M_WAIT, MT_DATA);
960 			MCLAIM(mp, &nfs_mowner);
961 			mp->m_len = 0;
962 			mp2->m_next = mp;
963 		}
964 		cp = mtod(mp, caddr_t)+mp->m_len;
965 		for (left = 0; left < rem; left++)
966 			*cp++ = '\0';
967 		mp->m_len += rem;
968 		*bpos = cp;
969 	} else
970 		*bpos = mtod(mp, caddr_t)+mp->m_len;
971 	*mq = mp;
972 	return (0);
973 }
974 
975 /*
976  * Get at least "siz" bytes of correctly aligned data.
977  * When called the mbuf pointers are not necessarily correct,
978  * dsosp points to what ought to be in m_data and left contains
979  * what ought to be in m_len.
980  * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
981  * cases. (The macros use the vars. dpos and dpos2)
982  */
983 int
984 nfsm_disct(mdp, dposp, siz, left, cp2)
985 	struct mbuf **mdp;
986 	caddr_t *dposp;
987 	int siz;
988 	int left;
989 	caddr_t *cp2;
990 {
991 	struct mbuf *m1, *m2;
992 	struct mbuf *havebuf = NULL;
993 	caddr_t src = *dposp;
994 	caddr_t dst;
995 	int len;
996 
997 #ifdef DEBUG
998 	if (left < 0)
999 		panic("nfsm_disct: left < 0");
1000 #endif
1001 	m1 = *mdp;
1002 	/*
1003 	 * Skip through the mbuf chain looking for an mbuf with
1004 	 * some data. If the first mbuf found has enough data
1005 	 * and it is correctly aligned return it.
1006 	 */
1007 	while (left == 0) {
1008 		havebuf = m1;
1009 		*mdp = m1 = m1->m_next;
1010 		if (m1 == NULL)
1011 			return (EBADRPC);
1012 		src = mtod(m1, caddr_t);
1013 		left = m1->m_len;
1014 		/*
1015 		 * If we start a new mbuf and it is big enough
1016 		 * and correctly aligned just return it, don't
1017 		 * do any pull up.
1018 		 */
1019 		if (left >= siz && nfsm_aligned(src)) {
1020 			*cp2 = src;
1021 			*dposp = src + siz;
1022 			return (0);
1023 		}
1024 	}
1025 	if (m1->m_flags & M_EXT) {
1026 		if (havebuf) {
1027 			/* If the first mbuf with data has external data
1028 			 * and there is a previous empty mbuf use it
1029 			 * to move the data into.
1030 			 */
1031 			m2 = m1;
1032 			*mdp = m1 = havebuf;
1033 			if (m1->m_flags & M_EXT) {
1034 				MEXTREMOVE(m1);
1035 			}
1036 		} else {
1037 			/*
1038 			 * If the first mbuf has a external data
1039 			 * and there is no previous empty mbuf
1040 			 * allocate a new mbuf and move the external
1041 			 * data to the new mbuf. Also make the first
1042 			 * mbuf look empty.
1043 			 */
1044 			m2 = m_get(M_WAIT, MT_DATA);
1045 			m2->m_ext = m1->m_ext;
1046 			m2->m_data = src;
1047 			m2->m_len = left;
1048 			MCLADDREFERENCE(m1, m2);
1049 			MEXTREMOVE(m1);
1050 			m2->m_next = m1->m_next;
1051 			m1->m_next = m2;
1052 		}
1053 		m1->m_len = 0;
1054 		dst = m1->m_dat;
1055 	} else {
1056 		/*
1057 		 * If the first mbuf has no external data
1058 		 * move the data to the front of the mbuf.
1059 		 */
1060 		if ((dst = m1->m_dat) != src)
1061 			memmove(dst, src, left);
1062 		dst += left;
1063 		m1->m_len = left;
1064 		m2 = m1->m_next;
1065 	}
1066 	m1->m_flags &= ~M_PKTHDR;
1067 	*cp2 = m1->m_data = m1->m_dat;   /* data is at beginning of buffer */
1068 	*dposp = mtod(m1, caddr_t) + siz;
1069 	/*
1070 	 * Loop through mbufs pulling data up into first mbuf until
1071 	 * the first mbuf is full or there is no more data to
1072 	 * pullup.
1073 	 */
1074 	while ((len = (MLEN - m1->m_len)) != 0 && m2) {
1075 		if ((len = min(len, m2->m_len)) != 0)
1076 			memcpy(dst, m2->m_data, len);
1077 		m1->m_len += len;
1078 		dst += len;
1079 		m2->m_data += len;
1080 		m2->m_len -= len;
1081 		m2 = m2->m_next;
1082 	}
1083 	if (m1->m_len < siz)
1084 		return (EBADRPC);
1085 	return (0);
1086 }
1087 
1088 /*
1089  * Advance the position in the mbuf chain.
1090  */
1091 int
1092 nfs_adv(mdp, dposp, offs, left)
1093 	struct mbuf **mdp;
1094 	caddr_t *dposp;
1095 	int offs;
1096 	int left;
1097 {
1098 	struct mbuf *m;
1099 	int s;
1100 
1101 	m = *mdp;
1102 	s = left;
1103 	while (s < offs) {
1104 		offs -= s;
1105 		m = m->m_next;
1106 		if (m == NULL)
1107 			return (EBADRPC);
1108 		s = m->m_len;
1109 	}
1110 	*mdp = m;
1111 	*dposp = mtod(m, caddr_t)+offs;
1112 	return (0);
1113 }
1114 
1115 /*
1116  * Copy a string into mbufs for the hard cases...
1117  */
1118 int
1119 nfsm_strtmbuf(mb, bpos, cp, siz)
1120 	struct mbuf **mb;
1121 	char **bpos;
1122 	const char *cp;
1123 	long siz;
1124 {
1125 	struct mbuf *m1 = NULL, *m2;
1126 	long left, xfer, len, tlen;
1127 	u_int32_t *tl;
1128 	int putsize;
1129 
1130 	putsize = 1;
1131 	m2 = *mb;
1132 	left = M_TRAILINGSPACE(m2);
1133 	if (left > 0) {
1134 		tl = ((u_int32_t *)(*bpos));
1135 		*tl++ = txdr_unsigned(siz);
1136 		putsize = 0;
1137 		left -= NFSX_UNSIGNED;
1138 		m2->m_len += NFSX_UNSIGNED;
1139 		if (left > 0) {
1140 			memcpy((caddr_t) tl, cp, left);
1141 			siz -= left;
1142 			cp += left;
1143 			m2->m_len += left;
1144 			left = 0;
1145 		}
1146 	}
1147 	/* Loop around adding mbufs */
1148 	while (siz > 0) {
1149 		m1 = m_get(M_WAIT, MT_DATA);
1150 		MCLAIM(m1, &nfs_mowner);
1151 		if (siz > MLEN)
1152 			m_clget(m1, M_WAIT);
1153 		m1->m_len = NFSMSIZ(m1);
1154 		m2->m_next = m1;
1155 		m2 = m1;
1156 		tl = mtod(m1, u_int32_t *);
1157 		tlen = 0;
1158 		if (putsize) {
1159 			*tl++ = txdr_unsigned(siz);
1160 			m1->m_len -= NFSX_UNSIGNED;
1161 			tlen = NFSX_UNSIGNED;
1162 			putsize = 0;
1163 		}
1164 		if (siz < m1->m_len) {
1165 			len = nfsm_rndup(siz);
1166 			xfer = siz;
1167 			if (xfer < len)
1168 				*(tl+(xfer>>2)) = 0;
1169 		} else {
1170 			xfer = len = m1->m_len;
1171 		}
1172 		memcpy((caddr_t) tl, cp, xfer);
1173 		m1->m_len = len+tlen;
1174 		siz -= xfer;
1175 		cp += xfer;
1176 	}
1177 	*mb = m1;
1178 	*bpos = mtod(m1, caddr_t)+m1->m_len;
1179 	return (0);
1180 }
1181 
1182 /*
1183  * Directory caching routines. They work as follows:
1184  * - a cache is maintained per VDIR nfsnode.
1185  * - for each offset cookie that is exported to userspace, and can
1186  *   thus be thrown back at us as an offset to VOP_READDIR, store
1187  *   information in the cache.
1188  * - cached are:
1189  *   - cookie itself
1190  *   - blocknumber (essentially just a search key in the buffer cache)
1191  *   - entry number in block.
1192  *   - offset cookie of block in which this entry is stored
1193  *   - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
1194  * - entries are looked up in a hash table
1195  * - also maintained is an LRU list of entries, used to determine
1196  *   which ones to delete if the cache grows too large.
1197  * - if 32 <-> 64 translation mode is requested for a filesystem,
1198  *   the cache also functions as a translation table
1199  * - in the translation case, invalidating the cache does not mean
1200  *   flushing it, but just marking entries as invalid, except for
1201  *   the <64bit cookie, 32bitcookie> pair which is still valid, to
1202  *   still be able to use the cache as a translation table.
1203  * - 32 bit cookies are uniquely created by combining the hash table
1204  *   entry value, and one generation count per hash table entry,
1205  *   incremented each time an entry is appended to the chain.
1206  * - the cache is invalidated each time a direcory is modified
1207  * - sanity checks are also done; if an entry in a block turns
1208  *   out not to have a matching cookie, the cache is invalidated
1209  *   and a new block starting from the wanted offset is fetched from
1210  *   the server.
1211  * - directory entries as read from the server are extended to contain
1212  *   the 64bit and, optionally, the 32bit cookies, for sanity checking
1213  *   the cache and exporting them to userspace through the cookie
1214  *   argument to VOP_READDIR.
1215  */
1216 
1217 u_long
1218 nfs_dirhash(off)
1219 	off_t off;
1220 {
1221 	int i;
1222 	char *cp = (char *)&off;
1223 	u_long sum = 0L;
1224 
1225 	for (i = 0 ; i < sizeof (off); i++)
1226 		sum += *cp++;
1227 
1228 	return sum;
1229 }
1230 
1231 void
1232 nfs_initdircache(vp)
1233 	struct vnode *vp;
1234 {
1235 	struct nfsnode *np = VTONFS(vp);
1236 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1237 
1238 	np->n_dircachesize = 0;
1239 	np->n_dblkno = 1;
1240 	np->n_dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF,
1241 	    M_WAITOK, &nfsdirhashmask);
1242 	TAILQ_INIT(&np->n_dirchain);
1243 	if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1244 		MALLOC(np->n_dirgens, unsigned *,
1245 		    NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF,
1246 		    M_WAITOK);
1247 		memset((caddr_t)np->n_dirgens, 0,
1248 		    NFS_DIRHASHSIZ * sizeof (unsigned));
1249 	}
1250 }
1251 
1252 static struct nfsdircache dzero = {0, 0, {0, 0}, {0, 0}, 0, 0, 0};
1253 
1254 struct nfsdircache *
1255 nfs_searchdircache(vp, off, do32, hashent)
1256 	struct vnode *vp;
1257 	off_t off;
1258 	int do32;
1259 	int *hashent;
1260 {
1261 	struct nfsdirhashhead *ndhp;
1262 	struct nfsdircache *ndp = NULL;
1263 	struct nfsnode *np = VTONFS(vp);
1264 	unsigned ent;
1265 
1266 	/*
1267 	 * Zero is always a valid cookie.
1268 	 */
1269 	if (off == 0)
1270 		return &dzero;
1271 
1272 	/*
1273 	 * We use a 32bit cookie as search key, directly reconstruct
1274 	 * the hashentry. Else use the hashfunction.
1275 	 */
1276 	if (do32) {
1277 		ent = (u_int32_t)off >> 24;
1278 		if (ent >= NFS_DIRHASHSIZ)
1279 			return NULL;
1280 		ndhp = &np->n_dircache[ent];
1281 	} else {
1282 		ndhp = NFSDIRHASH(np, off);
1283 	}
1284 
1285 	if (hashent)
1286 		*hashent = (int)(ndhp - np->n_dircache);
1287 	if (do32) {
1288 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1289 			if (ndp->dc_cookie32 == (u_int32_t)off) {
1290 				/*
1291 				 * An invalidated entry will become the
1292 				 * start of a new block fetched from
1293 				 * the server.
1294 				 */
1295 				if (ndp->dc_blkno == -1) {
1296 					ndp->dc_blkcookie = ndp->dc_cookie;
1297 					ndp->dc_blkno = np->n_dblkno++;
1298 					ndp->dc_entry = 0;
1299 				}
1300 				break;
1301 			}
1302 		}
1303 	} else {
1304 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1305 			if (ndp->dc_cookie == off)
1306 				break;
1307 		}
1308 	}
1309 	return ndp;
1310 }
1311 
1312 
1313 struct nfsdircache *
1314 nfs_enterdircache(vp, off, blkoff, en, blkno)
1315 	struct vnode *vp;
1316 	off_t off, blkoff;
1317 	int en;
1318 	daddr_t blkno;
1319 {
1320 	struct nfsnode *np = VTONFS(vp);
1321 	struct nfsdirhashhead *ndhp;
1322 	struct nfsdircache *ndp = NULL, *first;
1323 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1324 	int hashent, gen, overwrite;
1325 
1326 	if (!np->n_dircache)
1327 		/*
1328 		 * XXX would like to do this in nfs_nget but vtype
1329 		 * isn't known at that time.
1330 		 */
1331 		nfs_initdircache(vp);
1332 
1333 	/*
1334 	 * XXX refuse entries for offset 0. amd(8) erroneously sets
1335 	 * cookie 0 for the '.' entry, making this necessary. This
1336 	 * isn't so bad, as 0 is a special case anyway.
1337 	 */
1338 	if (off == 0)
1339 		return &dzero;
1340 
1341 	ndp = nfs_searchdircache(vp, off, 0, &hashent);
1342 
1343 	if (ndp && ndp->dc_blkno != -1) {
1344 		/*
1345 		 * Overwriting an old entry. Check if it's the same.
1346 		 * If so, just return. If not, remove the old entry.
1347 		 */
1348 		if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
1349 			return ndp;
1350 		TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1351 		LIST_REMOVE(ndp, dc_hash);
1352 		FREE(ndp, M_NFSDIROFF);
1353 		ndp = 0;
1354 	}
1355 
1356 	ndhp = &np->n_dircache[hashent];
1357 
1358 	if (!ndp) {
1359 		MALLOC(ndp, struct nfsdircache *, sizeof (*ndp), M_NFSDIROFF,
1360 		    M_WAITOK);
1361 		overwrite = 0;
1362 		if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1363 			/*
1364 			 * We're allocating a new entry, so bump the
1365 			 * generation number.
1366 			 */
1367 			gen = ++np->n_dirgens[hashent];
1368 			if (gen == 0) {
1369 				np->n_dirgens[hashent]++;
1370 				gen++;
1371 			}
1372 			ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
1373 		}
1374 	} else
1375 		overwrite = 1;
1376 
1377 	/*
1378 	 * If the entry number is 0, we are at the start of a new block, so
1379 	 * allocate a new blocknumber.
1380 	 */
1381 	if (en == 0)
1382 		ndp->dc_blkno = np->n_dblkno++;
1383 	else
1384 		ndp->dc_blkno = blkno;
1385 
1386 	ndp->dc_cookie = off;
1387 	ndp->dc_blkcookie = blkoff;
1388 	ndp->dc_entry = en;
1389 
1390 	if (overwrite)
1391 		return ndp;
1392 
1393 	/*
1394 	 * If the maximum directory cookie cache size has been reached
1395 	 * for this node, take one off the front. The idea is that
1396 	 * directories are typically read front-to-back once, so that
1397 	 * the oldest entries can be thrown away without much performance
1398 	 * loss.
1399 	 */
1400 	if (np->n_dircachesize == NFS_MAXDIRCACHE) {
1401 		first = TAILQ_FIRST(&np->n_dirchain);
1402 		TAILQ_REMOVE(&np->n_dirchain, first, dc_chain);
1403 		LIST_REMOVE(first, dc_hash);
1404 		FREE(first, M_NFSDIROFF);
1405 	} else
1406 		np->n_dircachesize++;
1407 
1408 	LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
1409 	TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
1410 	return ndp;
1411 }
1412 
1413 void
1414 nfs_invaldircache(vp, forcefree)
1415 	struct vnode *vp;
1416 	int forcefree;
1417 {
1418 	struct nfsnode *np = VTONFS(vp);
1419 	struct nfsdircache *ndp = NULL;
1420 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1421 
1422 #ifdef DIAGNOSTIC
1423 	if (vp->v_type != VDIR)
1424 		panic("nfs: invaldircache: not dir");
1425 #endif
1426 
1427 	if (!np->n_dircache)
1428 		return;
1429 
1430 	if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
1431 		while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != 0) {
1432 			TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1433 			LIST_REMOVE(ndp, dc_hash);
1434 			FREE(ndp, M_NFSDIROFF);
1435 		}
1436 		np->n_dircachesize = 0;
1437 		if (forcefree && np->n_dirgens) {
1438 			FREE(np->n_dirgens, M_NFSDIROFF);
1439 		}
1440 	} else {
1441 		TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain) {
1442 			ndp->dc_blkno = -1;
1443 		}
1444 	}
1445 
1446 	np->n_dblkno = 1;
1447 }
1448 
1449 /*
1450  * Called once before VFS init to initialize shared and
1451  * server-specific data structures.
1452  */
1453 void
1454 nfs_init()
1455 {
1456 	nfsrtt.pos = 0;
1457 	rpc_vers = txdr_unsigned(RPC_VER2);
1458 	rpc_call = txdr_unsigned(RPC_CALL);
1459 	rpc_reply = txdr_unsigned(RPC_REPLY);
1460 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1461 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1462 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1463 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1464 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1465 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1466 	nfs_prog = txdr_unsigned(NFS_PROG);
1467 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
1468 	nfs_true = txdr_unsigned(TRUE);
1469 	nfs_false = txdr_unsigned(FALSE);
1470 	nfs_xdrneg1 = txdr_unsigned(-1);
1471 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1472 	if (nfs_ticks < 1)
1473 		nfs_ticks = 1;
1474 #ifdef NFSSERVER
1475 	nfsrv_init(0);			/* Init server data structures */
1476 	nfsrv_initcache();		/* Init the server request cache */
1477 #endif /* NFSSERVER */
1478 
1479 #if defined(NFSSERVER) || !defined(NFS_V2_ONLY)
1480 	/*
1481 	 * Initialize the nqnfs data structures.
1482 	 */
1483 	if (nqnfsstarttime == 0) {
1484 		nqnfsstarttime = boottime.tv_sec + nqsrv_maxlease
1485 			+ nqsrv_clockskew + nqsrv_writeslack;
1486 		NQLOADNOVRAM(nqnfsstarttime);
1487 		CIRCLEQ_INIT(&nqtimerhead);
1488 		nqfhhashtbl = hashinit(NQLCHSZ, HASH_LIST, M_NQLEASE,
1489 		    M_WAITOK, &nqfhhash);
1490 	}
1491 #endif
1492 
1493 	exithook_establish(nfs_exit, NULL);
1494 
1495 	/*
1496 	 * Initialize reply list and start timer
1497 	 */
1498 	TAILQ_INIT(&nfs_reqq);
1499 	nfs_timer(NULL);
1500 	MOWNER_ATTACH(&nfs_mowner);
1501 
1502 #ifdef NFS
1503 	/* Initialize the kqueue structures */
1504 	nfs_kqinit();
1505 #endif
1506 }
1507 
1508 #ifdef NFS
1509 /*
1510  * Called once at VFS init to initialize client-specific data structures.
1511  */
1512 void
1513 nfs_vfs_init()
1514 {
1515 	nfs_nhinit();			/* Init the nfsnode table */
1516 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1517 }
1518 
1519 void
1520 nfs_vfs_reinit()
1521 {
1522 	nfs_nhreinit();
1523 }
1524 
1525 void
1526 nfs_vfs_done()
1527 {
1528 	nfs_nhdone();
1529 }
1530 
1531 /*
1532  * Attribute cache routines.
1533  * nfs_loadattrcache() - loads or updates the cache contents from attributes
1534  *	that are on the mbuf list
1535  * nfs_getattrcache() - returns valid attributes if found in cache, returns
1536  *	error otherwise
1537  */
1538 
1539 /*
1540  * Load the attribute cache (that lives in the nfsnode entry) with
1541  * the values on the mbuf list and
1542  * Iff vap not NULL
1543  *    copy the attributes to *vaper
1544  */
1545 int
1546 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags)
1547 	struct vnode **vpp;
1548 	struct mbuf **mdp;
1549 	caddr_t *dposp;
1550 	struct vattr *vaper;
1551 	int flags;
1552 {
1553 	int32_t t1;
1554 	caddr_t cp2;
1555 	int error = 0;
1556 	struct mbuf *md;
1557 	int v3 = NFS_ISV3(*vpp);
1558 
1559 	md = *mdp;
1560 	t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
1561 	error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2);
1562 	if (error)
1563 		return (error);
1564 	return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags);
1565 }
1566 
1567 int
1568 nfs_loadattrcache(vpp, fp, vaper, flags)
1569 	struct vnode **vpp;
1570 	struct nfs_fattr *fp;
1571 	struct vattr *vaper;
1572 	int flags;
1573 {
1574 	struct vnode *vp = *vpp;
1575 	struct vattr *vap;
1576 	int v3 = NFS_ISV3(vp);
1577 	enum vtype vtyp;
1578 	u_short vmode;
1579 	struct timespec mtime;
1580 	struct vnode *nvp;
1581 	int32_t rdev;
1582 	struct nfsnode *np;
1583 	extern int (**spec_nfsv2nodeop_p) __P((void *));
1584 	uid_t uid;
1585 	gid_t gid;
1586 
1587 	if (v3) {
1588 		vtyp = nfsv3tov_type(fp->fa_type);
1589 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1590 		rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1),
1591 			fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2));
1592 		fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
1593 	} else {
1594 		vtyp = nfsv2tov_type(fp->fa_type);
1595 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1596 		if (vtyp == VNON || vtyp == VREG)
1597 			vtyp = IFTOVT(vmode);
1598 		rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
1599 		fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
1600 
1601 		/*
1602 		 * Really ugly NFSv2 kludge.
1603 		 */
1604 		if (vtyp == VCHR && rdev == 0xffffffff)
1605 			vtyp = VFIFO;
1606 	}
1607 
1608 	vmode &= ALLPERMS;
1609 
1610 	/*
1611 	 * If v_type == VNON it is a new node, so fill in the v_type,
1612 	 * n_mtime fields. Check to see if it represents a special
1613 	 * device, and if so, check for a possible alias. Once the
1614 	 * correct vnode has been obtained, fill in the rest of the
1615 	 * information.
1616 	 */
1617 	np = VTONFS(vp);
1618 	if (vp->v_type == VNON) {
1619 		vp->v_type = vtyp;
1620 		if (vp->v_type == VFIFO) {
1621 			extern int (**fifo_nfsv2nodeop_p) __P((void *));
1622 			vp->v_op = fifo_nfsv2nodeop_p;
1623 		}
1624 		if (vp->v_type == VCHR || vp->v_type == VBLK) {
1625 			vp->v_op = spec_nfsv2nodeop_p;
1626 			nvp = checkalias(vp, (dev_t)rdev, vp->v_mount);
1627 			if (nvp) {
1628 				/*
1629 				 * Discard unneeded vnode, but save its nfsnode.
1630 				 * Since the nfsnode does not have a lock, its
1631 				 * vnode lock has to be carried over.
1632 				 */
1633 				/*
1634 				 * XXX is the old node sure to be locked here?
1635 				 */
1636 				KASSERT(lockstatus(&vp->v_lock) ==
1637 				    LK_EXCLUSIVE);
1638 				nvp->v_data = vp->v_data;
1639 				vp->v_data = NULL;
1640 				VOP_UNLOCK(vp, 0);
1641 				vp->v_op = spec_vnodeop_p;
1642 				vrele(vp);
1643 				vgone(vp);
1644 				lockmgr(&nvp->v_lock, LK_EXCLUSIVE,
1645 				    &nvp->v_interlock);
1646 				/*
1647 				 * Reinitialize aliased node.
1648 				 */
1649 				np->n_vnode = nvp;
1650 				*vpp = vp = nvp;
1651 			}
1652 		}
1653 		np->n_mtime = mtime.tv_sec;
1654 	}
1655 	uid = fxdr_unsigned(uid_t, fp->fa_uid);
1656 	gid = fxdr_unsigned(gid_t, fp->fa_gid);
1657 	vap = np->n_vattr;
1658 
1659 	/*
1660 	 * Invalidate access cache if uid, gid or mode changed.
1661 	 */
1662 	if (np->n_accstamp != -1 &&
1663 	    (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode))
1664 		np->n_accstamp = -1;
1665 
1666 	vap->va_type = vtyp;
1667 	vap->va_mode = vmode;
1668 	vap->va_rdev = (dev_t)rdev;
1669 	vap->va_mtime = mtime;
1670 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1671 	switch (vtyp) {
1672 	case VDIR:
1673 		vap->va_blocksize = NFS_DIRFRAGSIZ;
1674 		break;
1675 	case VBLK:
1676 		vap->va_blocksize = BLKDEV_IOSIZE;
1677 		break;
1678 	case VCHR:
1679 		vap->va_blocksize = MAXBSIZE;
1680 		break;
1681 	default:
1682 		vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize :
1683 		    fxdr_unsigned(int32_t, fp->fa2_blocksize);
1684 		break;
1685 	}
1686 	if (v3) {
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_hyper(&fp->fa3_size);
1691 		vap->va_bytes = fxdr_hyper(&fp->fa3_used);
1692 		vap->va_fileid = fxdr_unsigned(int32_t,
1693 		    fp->fa3_fileid.nfsuquad[1]);
1694 		fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
1695 		fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime);
1696 		vap->va_flags = 0;
1697 		vap->va_filerev = 0;
1698 	} else {
1699 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1700 		vap->va_uid = uid;
1701 		vap->va_gid = gid;
1702 		vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
1703 		vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks)
1704 		    * NFS_FABLKSIZE;
1705 		vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
1706 		fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
1707 		vap->va_flags = 0;
1708 		vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t,
1709 		    fp->fa2_ctime.nfsv2_sec);
1710 		vap->va_ctime.tv_nsec = 0;
1711 		vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
1712 		vap->va_filerev = 0;
1713 	}
1714 	if (vap->va_size != np->n_size) {
1715 		if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) {
1716 			vap->va_size = np->n_size;
1717 		} else {
1718 			np->n_size = vap->va_size;
1719 			if (vap->va_type == VREG) {
1720 				if ((flags & NAC_NOTRUNC)
1721 				    && np->n_size < vp->v_size) {
1722 					/*
1723 					 * we can't free pages now because
1724 					 * the pages can be owned by ourselves.
1725 					 */
1726 					np->n_flag |= NTRUNCDELAYED;
1727 				}
1728 				else {
1729 					uvm_vnp_setsize(vp, np->n_size);
1730 				}
1731 			}
1732 		}
1733 	}
1734 	np->n_attrstamp = time.tv_sec;
1735 	if (vaper != NULL) {
1736 		memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(*vap));
1737 		if (np->n_flag & NCHG) {
1738 			if (np->n_flag & NACC)
1739 				vaper->va_atime = np->n_atim;
1740 			if (np->n_flag & NUPD)
1741 				vaper->va_mtime = np->n_mtim;
1742 		}
1743 	}
1744 	return (0);
1745 }
1746 
1747 /*
1748  * Check the time stamp
1749  * If the cache is valid, copy contents to *vap and return 0
1750  * otherwise return an error
1751  */
1752 int
1753 nfs_getattrcache(vp, vaper)
1754 	struct vnode *vp;
1755 	struct vattr *vaper;
1756 {
1757 	struct nfsnode *np = VTONFS(vp);
1758 	struct vattr *vap;
1759 
1760 	if ((time.tv_sec - np->n_attrstamp) >= NFS_ATTRTIMEO(np)) {
1761 		nfsstats.attrcache_misses++;
1762 		return (ENOENT);
1763 	}
1764 	nfsstats.attrcache_hits++;
1765 	vap = np->n_vattr;
1766 	if (vap->va_size != np->n_size) {
1767 		if (vap->va_type == VREG) {
1768 			if (np->n_flag & NMODIFIED) {
1769 				if (vap->va_size < np->n_size)
1770 					vap->va_size = np->n_size;
1771 				else
1772 					np->n_size = vap->va_size;
1773 			} else
1774 				np->n_size = vap->va_size;
1775 			uvm_vnp_setsize(vp, np->n_size);
1776 		} else
1777 			np->n_size = vap->va_size;
1778 	}
1779 	memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(struct vattr));
1780 	if (np->n_flag & NCHG) {
1781 		if (np->n_flag & NACC)
1782 			vaper->va_atime = np->n_atim;
1783 		if (np->n_flag & NUPD)
1784 			vaper->va_mtime = np->n_mtim;
1785 	}
1786 	return (0);
1787 }
1788 
1789 void
1790 nfs_delayedtruncate(vp)
1791 	struct vnode *vp;
1792 {
1793 	struct nfsnode *np = VTONFS(vp);
1794 
1795 	if (np->n_flag & NTRUNCDELAYED) {
1796 		np->n_flag &= ~NTRUNCDELAYED;
1797 		uvm_vnp_setsize(vp, np->n_size);
1798 	}
1799 }
1800 
1801 /*
1802  * Heuristic to see if the server XDR encodes directory cookies or not.
1803  * it is not supposed to, but a lot of servers may do this. Also, since
1804  * most/all servers will implement V2 as well, it is expected that they
1805  * may return just 32 bits worth of cookie information, so we need to
1806  * find out in which 32 bits this information is available. We do this
1807  * to avoid trouble with emulated binaries that can't handle 64 bit
1808  * directory offsets.
1809  */
1810 
1811 void
1812 nfs_cookieheuristic(vp, flagp, p, cred)
1813 	struct vnode *vp;
1814 	int *flagp;
1815 	struct proc *p;
1816 	struct ucred *cred;
1817 {
1818 	struct uio auio;
1819 	struct iovec aiov;
1820 	caddr_t buf, cp;
1821 	struct dirent *dp;
1822 	off_t *cookies = NULL, *cop;
1823 	int error, eof, nc, len;
1824 
1825 	MALLOC(buf, caddr_t, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK);
1826 
1827 	aiov.iov_base = buf;
1828 	aiov.iov_len = NFS_DIRFRAGSIZ;
1829 	auio.uio_iov = &aiov;
1830 	auio.uio_iovcnt = 1;
1831 	auio.uio_rw = UIO_READ;
1832 	auio.uio_segflg = UIO_SYSSPACE;
1833 	auio.uio_procp = p;
1834 	auio.uio_resid = NFS_DIRFRAGSIZ;
1835 	auio.uio_offset = 0;
1836 
1837 	error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc);
1838 
1839 	len = NFS_DIRFRAGSIZ - auio.uio_resid;
1840 	if (error || len == 0) {
1841 		FREE(buf, M_TEMP);
1842 		if (cookies)
1843 			free(cookies, M_TEMP);
1844 		return;
1845 	}
1846 
1847 	/*
1848 	 * Find the first valid entry and look at its offset cookie.
1849 	 */
1850 
1851 	cp = buf;
1852 	for (cop = cookies; len > 0; len -= dp->d_reclen) {
1853 		dp = (struct dirent *)cp;
1854 		if (dp->d_fileno != 0 && len >= dp->d_reclen) {
1855 			if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) {
1856 				*flagp |= NFSMNT_SWAPCOOKIE;
1857 				nfs_invaldircache(vp, 0);
1858 				nfs_vinvalbuf(vp, 0, cred, p, 1);
1859 			}
1860 			break;
1861 		}
1862 		cop++;
1863 		cp += dp->d_reclen;
1864 	}
1865 
1866 	FREE(buf, M_TEMP);
1867 	free(cookies, M_TEMP);
1868 }
1869 #endif /* NFS */
1870 
1871 /*
1872  * Set up nameidata for a lookup() call and do it.
1873  *
1874  * If pubflag is set, this call is done for a lookup operation on the
1875  * public filehandle. In that case we allow crossing mountpoints and
1876  * absolute pathnames. However, the caller is expected to check that
1877  * the lookup result is within the public fs, and deny access if
1878  * it is not.
1879  */
1880 int
1881 nfs_namei(ndp, fhp, len, slp, nam, mdp, dposp, retdirp, p, kerbflag, pubflag)
1882 	struct nameidata *ndp;
1883 	fhandle_t *fhp;
1884 	uint32_t len;
1885 	struct nfssvc_sock *slp;
1886 	struct mbuf *nam;
1887 	struct mbuf **mdp;
1888 	caddr_t *dposp;
1889 	struct vnode **retdirp;
1890 	struct proc *p;
1891 	int kerbflag, pubflag;
1892 {
1893 	int i, rem;
1894 	struct mbuf *md;
1895 	char *fromcp, *tocp, *cp;
1896 	struct iovec aiov;
1897 	struct uio auio;
1898 	struct vnode *dp;
1899 	int error, rdonly, linklen;
1900 	struct componentname *cnp = &ndp->ni_cnd;
1901 
1902 	*retdirp = (struct vnode *)0;
1903 
1904 	if ((len + 1) > MAXPATHLEN)
1905 		return (ENAMETOOLONG);
1906 	cnp->cn_pnbuf = PNBUF_GET();
1907 
1908 	/*
1909 	 * Copy the name from the mbuf list to ndp->ni_pnbuf
1910 	 * and set the various ndp fields appropriately.
1911 	 */
1912 	fromcp = *dposp;
1913 	tocp = cnp->cn_pnbuf;
1914 	md = *mdp;
1915 	rem = mtod(md, caddr_t) + md->m_len - fromcp;
1916 	for (i = 0; i < len; i++) {
1917 		while (rem == 0) {
1918 			md = md->m_next;
1919 			if (md == NULL) {
1920 				error = EBADRPC;
1921 				goto out;
1922 			}
1923 			fromcp = mtod(md, caddr_t);
1924 			rem = md->m_len;
1925 		}
1926 		if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
1927 			error = EACCES;
1928 			goto out;
1929 		}
1930 		*tocp++ = *fromcp++;
1931 		rem--;
1932 	}
1933 	*tocp = '\0';
1934 	*mdp = md;
1935 	*dposp = fromcp;
1936 	len = nfsm_rndup(len)-len;
1937 	if (len > 0) {
1938 		if (rem >= len)
1939 			*dposp += len;
1940 		else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
1941 			goto out;
1942 	}
1943 
1944 	/*
1945 	 * Extract and set starting directory.
1946 	 */
1947 	error = nfsrv_fhtovp(fhp, FALSE, &dp, ndp->ni_cnd.cn_cred, slp,
1948 	    nam, &rdonly, kerbflag, pubflag);
1949 	if (error)
1950 		goto out;
1951 	if (dp->v_type != VDIR) {
1952 		vrele(dp);
1953 		error = ENOTDIR;
1954 		goto out;
1955 	}
1956 
1957 	if (rdonly)
1958 		cnp->cn_flags |= RDONLY;
1959 
1960 	*retdirp = dp;
1961 
1962 	if (pubflag) {
1963 		/*
1964 		 * Oh joy. For WebNFS, handle those pesky '%' escapes,
1965 		 * and the 'native path' indicator.
1966 		 */
1967 		cp = PNBUF_GET();
1968 		fromcp = cnp->cn_pnbuf;
1969 		tocp = cp;
1970 		if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
1971 			switch ((unsigned char)*fromcp) {
1972 			case WEBNFS_NATIVE_CHAR:
1973 				/*
1974 				 * 'Native' path for us is the same
1975 				 * as a path according to the NFS spec,
1976 				 * just skip the escape char.
1977 				 */
1978 				fromcp++;
1979 				break;
1980 			/*
1981 			 * More may be added in the future, range 0x80-0xff
1982 			 */
1983 			default:
1984 				error = EIO;
1985 				PNBUF_PUT(cp);
1986 				goto out;
1987 			}
1988 		}
1989 		/*
1990 		 * Translate the '%' escapes, URL-style.
1991 		 */
1992 		while (*fromcp != '\0') {
1993 			if (*fromcp == WEBNFS_ESC_CHAR) {
1994 				if (fromcp[1] != '\0' && fromcp[2] != '\0') {
1995 					fromcp++;
1996 					*tocp++ = HEXSTRTOI(fromcp);
1997 					fromcp += 2;
1998 					continue;
1999 				} else {
2000 					error = ENOENT;
2001 					PNBUF_PUT(cp);
2002 					goto out;
2003 				}
2004 			} else
2005 				*tocp++ = *fromcp++;
2006 		}
2007 		*tocp = '\0';
2008 		PNBUF_PUT(cnp->cn_pnbuf);
2009 		cnp->cn_pnbuf = cp;
2010 	}
2011 
2012 	ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1;
2013 	ndp->ni_segflg = UIO_SYSSPACE;
2014 	ndp->ni_rootdir = rootvnode;
2015 
2016 	if (pubflag) {
2017 		ndp->ni_loopcnt = 0;
2018 		if (cnp->cn_pnbuf[0] == '/')
2019 			dp = rootvnode;
2020 	} else {
2021 		cnp->cn_flags |= NOCROSSMOUNT;
2022 	}
2023 
2024 	cnp->cn_proc = p;
2025 	VREF(dp);
2026 
2027     for (;;) {
2028 	cnp->cn_nameptr = cnp->cn_pnbuf;
2029 	ndp->ni_startdir = dp;
2030 	/*
2031 	 * And call lookup() to do the real work
2032 	 */
2033 	error = lookup(ndp);
2034 	if (error) {
2035 		PNBUF_PUT(cnp->cn_pnbuf);
2036 		return (error);
2037 	}
2038 	/*
2039 	 * Check for encountering a symbolic link
2040 	 */
2041 	if ((cnp->cn_flags & ISSYMLINK) == 0) {
2042 		if (cnp->cn_flags & (SAVENAME | SAVESTART))
2043 			cnp->cn_flags |= HASBUF;
2044 		else
2045 			PNBUF_PUT(cnp->cn_pnbuf);
2046 		return (0);
2047 	} else {
2048 		if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
2049 			VOP_UNLOCK(ndp->ni_dvp, 0);
2050 		if (!pubflag) {
2051 			error = EINVAL;
2052 			break;
2053 		}
2054 
2055 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
2056 			error = ELOOP;
2057 			break;
2058 		}
2059 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
2060 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred,
2061 			    cnp->cn_proc);
2062 			if (error != 0)
2063 				break;
2064 		}
2065 		if (ndp->ni_pathlen > 1)
2066 			cp = PNBUF_GET();
2067 		else
2068 			cp = cnp->cn_pnbuf;
2069 		aiov.iov_base = cp;
2070 		aiov.iov_len = MAXPATHLEN;
2071 		auio.uio_iov = &aiov;
2072 		auio.uio_iovcnt = 1;
2073 		auio.uio_offset = 0;
2074 		auio.uio_rw = UIO_READ;
2075 		auio.uio_segflg = UIO_SYSSPACE;
2076 		auio.uio_procp = (struct proc *)0;
2077 		auio.uio_resid = MAXPATHLEN;
2078 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
2079 		if (error) {
2080 		badlink:
2081 			if (ndp->ni_pathlen > 1)
2082 				PNBUF_PUT(cp);
2083 			break;
2084 		}
2085 		linklen = MAXPATHLEN - auio.uio_resid;
2086 		if (linklen == 0) {
2087 			error = ENOENT;
2088 			goto badlink;
2089 		}
2090 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
2091 			error = ENAMETOOLONG;
2092 			goto badlink;
2093 		}
2094 		if (ndp->ni_pathlen > 1) {
2095 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
2096 			PNBUF_PUT(cnp->cn_pnbuf);
2097 			cnp->cn_pnbuf = cp;
2098 		} else
2099 			cnp->cn_pnbuf[linklen] = '\0';
2100 		ndp->ni_pathlen += linklen;
2101 		vput(ndp->ni_vp);
2102 		dp = ndp->ni_dvp;
2103 		/*
2104 		 * Check if root directory should replace current directory.
2105 		 */
2106 		if (cnp->cn_pnbuf[0] == '/') {
2107 			vrele(dp);
2108 			dp = ndp->ni_rootdir;
2109 			VREF(dp);
2110 		}
2111 	}
2112    }
2113 	vrele(ndp->ni_dvp);
2114 	vput(ndp->ni_vp);
2115 	ndp->ni_vp = NULL;
2116 out:
2117 	PNBUF_PUT(cnp->cn_pnbuf);
2118 	return (error);
2119 }
2120 
2121 /*
2122  * A fiddled version of m_adj() that ensures null fill to a 32-bit
2123  * boundary and only trims off the back end
2124  */
2125 void
2126 nfsm_adj(mp, len, nul)
2127 	struct mbuf *mp;
2128 	int len;
2129 	int nul;
2130 {
2131 	struct mbuf *m;
2132 	int count, i;
2133 	char *cp;
2134 
2135 	/*
2136 	 * Trim from tail.  Scan the mbuf chain,
2137 	 * calculating its length and finding the last mbuf.
2138 	 * If the adjustment only affects this mbuf, then just
2139 	 * adjust and return.  Otherwise, rescan and truncate
2140 	 * after the remaining size.
2141 	 */
2142 	count = 0;
2143 	m = mp;
2144 	for (;;) {
2145 		count += m->m_len;
2146 		if (m->m_next == (struct mbuf *)0)
2147 			break;
2148 		m = m->m_next;
2149 	}
2150 	if (m->m_len > len) {
2151 		m->m_len -= len;
2152 		if (nul > 0) {
2153 			if (M_ROMAP(m)) {
2154 				struct mbuf *n;
2155 
2156 				KDASSERT(MLEN >= nul);
2157 				n = m_get(M_WAIT, MT_DATA);
2158 				MCLAIM(n, &nfs_mowner);
2159 				n->m_len = nul;
2160 				n->m_next = m->m_next;
2161 				m->m_len -= nul;
2162 				m->m_next = n;
2163 				m = n;
2164 			}
2165 			cp = mtod(m, caddr_t)+m->m_len-nul;
2166 			for (i = 0; i < nul; i++)
2167 				*cp++ = '\0';
2168 		}
2169 		return;
2170 	}
2171 	count -= len;
2172 	if (count < 0)
2173 		count = 0;
2174 	/*
2175 	 * Correct length for chain is "count".
2176 	 * Find the mbuf with last data, adjust its length,
2177 	 * and toss data from remaining mbufs on chain.
2178 	 */
2179 	for (m = mp; m; m = m->m_next) {
2180 		if (m->m_len >= count) {
2181 			m->m_len = count;
2182 			if (nul > 0) {
2183 				if (M_ROMAP(m)) {
2184 					struct mbuf *n;
2185 
2186 					KDASSERT(MLEN >= nul);
2187 					n = m_get(M_WAIT, MT_DATA);
2188 					MCLAIM(n, &nfs_mowner);
2189 					n->m_len = nul;
2190 					n->m_next = m->m_next;
2191 					m->m_len -= nul;
2192 					m->m_next = n;
2193 					m = n;
2194 				}
2195 				cp = mtod(m, caddr_t)+m->m_len-nul;
2196 				for (i = 0; i < nul; i++)
2197 					*cp++ = '\0';
2198 			}
2199 			break;
2200 		}
2201 		count -= m->m_len;
2202 	}
2203 	for (m = m->m_next;m;m = m->m_next)
2204 		m->m_len = 0;
2205 }
2206 
2207 /*
2208  * Make these functions instead of macros, so that the kernel text size
2209  * doesn't get too big...
2210  */
2211 void
2212 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp)
2213 	struct nfsrv_descript *nfsd;
2214 	int before_ret;
2215 	struct vattr *before_vap;
2216 	int after_ret;
2217 	struct vattr *after_vap;
2218 	struct mbuf **mbp;
2219 	char **bposp;
2220 {
2221 	struct mbuf *mb = *mbp;
2222 	char *bpos = *bposp;
2223 	u_int32_t *tl;
2224 
2225 	if (before_ret) {
2226 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2227 		*tl = nfs_false;
2228 	} else {
2229 		nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
2230 		*tl++ = nfs_true;
2231 		txdr_hyper(before_vap->va_size, tl);
2232 		tl += 2;
2233 		txdr_nfsv3time(&(before_vap->va_mtime), tl);
2234 		tl += 2;
2235 		txdr_nfsv3time(&(before_vap->va_ctime), tl);
2236 	}
2237 	*bposp = bpos;
2238 	*mbp = mb;
2239 	nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
2240 }
2241 
2242 void
2243 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp)
2244 	struct nfsrv_descript *nfsd;
2245 	int after_ret;
2246 	struct vattr *after_vap;
2247 	struct mbuf **mbp;
2248 	char **bposp;
2249 {
2250 	struct mbuf *mb = *mbp;
2251 	char *bpos = *bposp;
2252 	u_int32_t *tl;
2253 	struct nfs_fattr *fp;
2254 
2255 	if (after_ret) {
2256 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2257 		*tl = nfs_false;
2258 	} else {
2259 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
2260 		*tl++ = nfs_true;
2261 		fp = (struct nfs_fattr *)tl;
2262 		nfsm_srvfattr(nfsd, after_vap, fp);
2263 	}
2264 	*mbp = mb;
2265 	*bposp = bpos;
2266 }
2267 
2268 void
2269 nfsm_srvfattr(nfsd, vap, fp)
2270 	struct nfsrv_descript *nfsd;
2271 	struct vattr *vap;
2272 	struct nfs_fattr *fp;
2273 {
2274 
2275 	fp->fa_nlink = txdr_unsigned(vap->va_nlink);
2276 	fp->fa_uid = txdr_unsigned(vap->va_uid);
2277 	fp->fa_gid = txdr_unsigned(vap->va_gid);
2278 	if (nfsd->nd_flag & ND_NFSV3) {
2279 		fp->fa_type = vtonfsv3_type(vap->va_type);
2280 		fp->fa_mode = vtonfsv3_mode(vap->va_mode);
2281 		txdr_hyper(vap->va_size, &fp->fa3_size);
2282 		txdr_hyper(vap->va_bytes, &fp->fa3_used);
2283 		fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
2284 		fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
2285 		fp->fa3_fsid.nfsuquad[0] = 0;
2286 		fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
2287 		fp->fa3_fileid.nfsuquad[0] = 0;
2288 		fp->fa3_fileid.nfsuquad[1] = txdr_unsigned(vap->va_fileid);
2289 		txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
2290 		txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
2291 		txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
2292 	} else {
2293 		fp->fa_type = vtonfsv2_type(vap->va_type);
2294 		fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
2295 		fp->fa2_size = txdr_unsigned(vap->va_size);
2296 		fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
2297 		if (vap->va_type == VFIFO)
2298 			fp->fa2_rdev = 0xffffffff;
2299 		else
2300 			fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
2301 		fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
2302 		fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
2303 		fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
2304 		txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
2305 		txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
2306 		txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
2307 	}
2308 }
2309 
2310 /*
2311  * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
2312  * 	- look up fsid in mount list (if not found ret error)
2313  *	- get vp and export rights by calling VFS_FHTOVP()
2314  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2315  *	- if not lockflag unlock it with VOP_UNLOCK()
2316  */
2317 int
2318 nfsrv_fhtovp(fhp, lockflag, vpp, cred, slp, nam, rdonlyp, kerbflag, pubflag)
2319 	fhandle_t *fhp;
2320 	int lockflag;
2321 	struct vnode **vpp;
2322 	struct ucred *cred;
2323 	struct nfssvc_sock *slp;
2324 	struct mbuf *nam;
2325 	int *rdonlyp;
2326 	int kerbflag;
2327 {
2328 	struct mount *mp;
2329 	int i;
2330 	struct ucred *credanon;
2331 	int error, exflags;
2332 	struct sockaddr_in *saddr;
2333 
2334 	*vpp = (struct vnode *)0;
2335 
2336 	if (nfs_ispublicfh(fhp)) {
2337 		if (!pubflag || !nfs_pub.np_valid)
2338 			return (ESTALE);
2339 		fhp = &nfs_pub.np_handle;
2340 	}
2341 
2342 	mp = vfs_getvfs(&fhp->fh_fsid);
2343 	if (!mp)
2344 		return (ESTALE);
2345 	error = VFS_CHECKEXP(mp, nam, &exflags, &credanon);
2346 	if (error)
2347 		return (error);
2348 	error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
2349 	if (error)
2350 		return (error);
2351 
2352 	if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
2353 		saddr = mtod(nam, struct sockaddr_in *);
2354 		if ((saddr->sin_family == AF_INET) &&
2355 		    ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
2356 			vput(*vpp);
2357 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2358 		}
2359 #ifdef INET6
2360 		if ((saddr->sin_family == AF_INET6) &&
2361 		    ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) {
2362 			vput(*vpp);
2363 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2364 		}
2365 #endif
2366 	}
2367 	/*
2368 	 * Check/setup credentials.
2369 	 */
2370 	if (exflags & MNT_EXKERB) {
2371 		if (!kerbflag) {
2372 			vput(*vpp);
2373 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2374 		}
2375 	} else if (kerbflag) {
2376 		vput(*vpp);
2377 		return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2378 	} else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
2379 		cred->cr_uid = credanon->cr_uid;
2380 		cred->cr_gid = credanon->cr_gid;
2381 		for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
2382 			cred->cr_groups[i] = credanon->cr_groups[i];
2383 		cred->cr_ngroups = i;
2384 	}
2385 	if (exflags & MNT_EXRDONLY)
2386 		*rdonlyp = 1;
2387 	else
2388 		*rdonlyp = 0;
2389 	if (!lockflag)
2390 		VOP_UNLOCK(*vpp, 0);
2391 	return (0);
2392 }
2393 
2394 /*
2395  * WebNFS: check if a filehandle is a public filehandle. For v3, this
2396  * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
2397  * transformed this to all zeroes in both cases, so check for it.
2398  */
2399 int
2400 nfs_ispublicfh(fhp)
2401 	fhandle_t *fhp;
2402 {
2403 	char *cp = (char *)fhp;
2404 	int i;
2405 
2406 	for (i = 0; i < NFSX_V3FH; i++)
2407 		if (*cp++ != 0)
2408 			return (FALSE);
2409 	return (TRUE);
2410 }
2411 
2412 /*
2413  * This function compares two net addresses by family and returns TRUE
2414  * if they are the same host.
2415  * If there is any doubt, return FALSE.
2416  * The AF_INET family is handled as a special case so that address mbufs
2417  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
2418  */
2419 int
2420 netaddr_match(family, haddr, nam)
2421 	int family;
2422 	union nethostaddr *haddr;
2423 	struct mbuf *nam;
2424 {
2425 	struct sockaddr_in *inetaddr;
2426 
2427 	switch (family) {
2428 	case AF_INET:
2429 		inetaddr = mtod(nam, struct sockaddr_in *);
2430 		if (inetaddr->sin_family == AF_INET &&
2431 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
2432 			return (1);
2433 		break;
2434 #ifdef INET6
2435 	case AF_INET6:
2436 	    {
2437 		struct sockaddr_in6 *sin6_1, *sin6_2;
2438 
2439 		sin6_1 = mtod(nam, struct sockaddr_in6 *);
2440 		sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
2441 		if (sin6_1->sin6_family == AF_INET6 &&
2442 		    IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
2443 			return 1;
2444 	    }
2445 #endif
2446 #ifdef ISO
2447 	case AF_ISO:
2448 	    {
2449 		struct sockaddr_iso *isoaddr1, *isoaddr2;
2450 
2451 		isoaddr1 = mtod(nam, struct sockaddr_iso *);
2452 		isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *);
2453 		if (isoaddr1->siso_family == AF_ISO &&
2454 		    isoaddr1->siso_nlen > 0 &&
2455 		    isoaddr1->siso_nlen == isoaddr2->siso_nlen &&
2456 		    SAME_ISOADDR(isoaddr1, isoaddr2))
2457 			return (1);
2458 		break;
2459 	    }
2460 #endif	/* ISO */
2461 	default:
2462 		break;
2463 	};
2464 	return (0);
2465 }
2466 
2467 /*
2468  * The write verifier has changed (probably due to a server reboot), so all
2469  * PG_NEEDCOMMIT pages will have to be written again. Since they are marked
2470  * as dirty or are being written out just now, all this takes is clearing
2471  * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for
2472  * the mount point.
2473  */
2474 void
2475 nfs_clearcommit(mp)
2476 	struct mount *mp;
2477 {
2478 	struct vnode *vp;
2479 	struct nfsnode *np;
2480 	struct vm_page *pg;
2481 	struct nfsmount *nmp = VFSTONFS(mp);
2482 
2483 	lockmgr(&nmp->nm_writeverflock, LK_EXCLUSIVE, NULL);
2484 
2485 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2486 		KASSERT(vp->v_mount == mp);
2487 		if (vp->v_type == VNON)
2488 			continue;
2489 		np = VTONFS(vp);
2490 		np->n_pushlo = np->n_pushhi = np->n_pushedlo =
2491 		    np->n_pushedhi = 0;
2492 		np->n_commitflags &=
2493 		    ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
2494 		simple_lock(&vp->v_uobj.vmobjlock);
2495 		TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) {
2496 			pg->flags &= ~PG_NEEDCOMMIT;
2497 		}
2498 		simple_unlock(&vp->v_uobj.vmobjlock);
2499 	}
2500 	simple_lock(&nmp->nm_slock);
2501 	nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF;
2502 	simple_unlock(&nmp->nm_slock);
2503 	lockmgr(&nmp->nm_writeverflock, LK_RELEASE, NULL);
2504 }
2505 
2506 void
2507 nfs_merge_commit_ranges(vp)
2508 	struct vnode *vp;
2509 {
2510 	struct nfsnode *np = VTONFS(vp);
2511 
2512 	KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID);
2513 
2514 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2515 		np->n_pushedlo = np->n_pushlo;
2516 		np->n_pushedhi = np->n_pushhi;
2517 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2518 	} else {
2519 		if (np->n_pushlo < np->n_pushedlo)
2520 			np->n_pushedlo = np->n_pushlo;
2521 		if (np->n_pushhi > np->n_pushedhi)
2522 			np->n_pushedhi = np->n_pushhi;
2523 	}
2524 
2525 	np->n_pushlo = np->n_pushhi = 0;
2526 	np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
2527 
2528 #ifdef NFS_DEBUG_COMMIT
2529 	printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2530 	    (unsigned)np->n_pushedhi);
2531 #endif
2532 }
2533 
2534 int
2535 nfs_in_committed_range(vp, off, len)
2536 	struct vnode *vp;
2537 	off_t off, len;
2538 {
2539 	struct nfsnode *np = VTONFS(vp);
2540 	off_t lo, hi;
2541 
2542 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2543 		return 0;
2544 	lo = off;
2545 	hi = lo + len;
2546 
2547 	return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
2548 }
2549 
2550 int
2551 nfs_in_tobecommitted_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_PUSH_VALID))
2559 		return 0;
2560 	lo = off;
2561 	hi = lo + len;
2562 
2563 	return (lo >= np->n_pushlo && hi <= np->n_pushhi);
2564 }
2565 
2566 void
2567 nfs_add_committed_range(vp, off, len)
2568 	struct vnode *vp;
2569 	off_t off, len;
2570 {
2571 	struct nfsnode *np = VTONFS(vp);
2572 	off_t lo, hi;
2573 
2574 	lo = off;
2575 	hi = lo + len;
2576 
2577 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2578 		np->n_pushedlo = lo;
2579 		np->n_pushedhi = hi;
2580 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2581 	} else {
2582 		if (hi > np->n_pushedhi)
2583 			np->n_pushedhi = hi;
2584 		if (lo < np->n_pushedlo)
2585 			np->n_pushedlo = lo;
2586 	}
2587 #ifdef NFS_DEBUG_COMMIT
2588 	printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2589 	    (unsigned)np->n_pushedhi);
2590 #endif
2591 }
2592 
2593 void
2594 nfs_del_committed_range(vp, off, len)
2595 	struct vnode *vp;
2596 	off_t off, len;
2597 {
2598 	struct nfsnode *np = VTONFS(vp);
2599 	off_t lo, hi;
2600 
2601 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2602 		return;
2603 
2604 	lo = off;
2605 	hi = lo + len;
2606 
2607 	if (lo > np->n_pushedhi || hi < np->n_pushedlo)
2608 		return;
2609 	if (lo <= np->n_pushedlo)
2610 		np->n_pushedlo = hi;
2611 	else if (hi >= np->n_pushedhi)
2612 		np->n_pushedhi = lo;
2613 	else {
2614 		/*
2615 		 * XXX There's only one range. If the deleted range
2616 		 * is in the middle, pick the largest of the
2617 		 * contiguous ranges that it leaves.
2618 		 */
2619 		if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
2620 			np->n_pushedhi = lo;
2621 		else
2622 			np->n_pushedlo = hi;
2623 	}
2624 #ifdef NFS_DEBUG_COMMIT
2625 	printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2626 	    (unsigned)np->n_pushedhi);
2627 #endif
2628 }
2629 
2630 void
2631 nfs_add_tobecommitted_range(vp, off, len)
2632 	struct vnode *vp;
2633 	off_t off, len;
2634 {
2635 	struct nfsnode *np = VTONFS(vp);
2636 	off_t lo, hi;
2637 
2638 	lo = off;
2639 	hi = lo + len;
2640 
2641 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
2642 		np->n_pushlo = lo;
2643 		np->n_pushhi = hi;
2644 		np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
2645 	} else {
2646 		if (lo < np->n_pushlo)
2647 			np->n_pushlo = lo;
2648 		if (hi > np->n_pushhi)
2649 			np->n_pushhi = hi;
2650 	}
2651 #ifdef NFS_DEBUG_COMMIT
2652 	printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2653 	    (unsigned)np->n_pushhi);
2654 #endif
2655 }
2656 
2657 void
2658 nfs_del_tobecommitted_range(vp, off, len)
2659 	struct vnode *vp;
2660 	off_t off, len;
2661 {
2662 	struct nfsnode *np = VTONFS(vp);
2663 	off_t lo, hi;
2664 
2665 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2666 		return;
2667 
2668 	lo = off;
2669 	hi = lo + len;
2670 
2671 	if (lo > np->n_pushhi || hi < np->n_pushlo)
2672 		return;
2673 
2674 	if (lo <= np->n_pushlo)
2675 		np->n_pushlo = hi;
2676 	else if (hi >= np->n_pushhi)
2677 		np->n_pushhi = lo;
2678 	else {
2679 		/*
2680 		 * XXX There's only one range. If the deleted range
2681 		 * is in the middle, pick the largest of the
2682 		 * contiguous ranges that it leaves.
2683 		 */
2684 		if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
2685 			np->n_pushhi = lo;
2686 		else
2687 			np->n_pushlo = hi;
2688 	}
2689 #ifdef NFS_DEBUG_COMMIT
2690 	printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2691 	    (unsigned)np->n_pushhi);
2692 #endif
2693 }
2694 
2695 /*
2696  * Map errnos to NFS error numbers. For Version 3 also filter out error
2697  * numbers not specified for the associated procedure.
2698  */
2699 int
2700 nfsrv_errmap(nd, err)
2701 	struct nfsrv_descript *nd;
2702 	int err;
2703 {
2704 	const short *defaulterrp, *errp;
2705 
2706 	if (nd->nd_flag & ND_NFSV3) {
2707 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
2708 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2709 		while (*++errp) {
2710 			if (*errp == err)
2711 				return (err);
2712 			else if (*errp > err)
2713 				break;
2714 		}
2715 		return ((int)*defaulterrp);
2716 	    } else
2717 		return (err & 0xffff);
2718 	}
2719 	if (err <= ELAST)
2720 		return ((int)nfsrv_v2errmap[err - 1]);
2721 	return (NFSERR_IO);
2722 }
2723 
2724 /*
2725  * Sort the group list in increasing numerical order.
2726  * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
2727  *  that used to be here.)
2728  */
2729 void
2730 nfsrvw_sort(list, num)
2731         gid_t *list;
2732         int num;
2733 {
2734 	int i, j;
2735 	gid_t v;
2736 
2737 	/* Insertion sort. */
2738 	for (i = 1; i < num; i++) {
2739 		v = list[i];
2740 		/* find correct slot for value v, moving others up */
2741 		for (j = i; --j >= 0 && v < list[j];)
2742 			list[j + 1] = list[j];
2743 		list[j + 1] = v;
2744 	}
2745 }
2746 
2747 /*
2748  * copy credentials making sure that the result can be compared with memcmp().
2749  */
2750 void
2751 nfsrv_setcred(incred, outcred)
2752 	struct ucred *incred, *outcred;
2753 {
2754 	int i;
2755 
2756 	memset((caddr_t)outcred, 0, sizeof (struct ucred));
2757 	outcred->cr_ref = 1;
2758 	outcred->cr_uid = incred->cr_uid;
2759 	outcred->cr_gid = incred->cr_gid;
2760 	outcred->cr_ngroups = incred->cr_ngroups;
2761 	for (i = 0; i < incred->cr_ngroups; i++)
2762 		outcred->cr_groups[i] = incred->cr_groups[i];
2763 	nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
2764 }
2765