xref: /netbsd-src/sys/nfs/nfs_subs.c (revision e5548b402ae4c44fb816de42c7bba9581ce23ef5)
1 /*	$NetBSD: nfs_subs.c,v 1.156 2005/12/11 12:25:16 christos Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_subs.c	8.8 (Berkeley) 5/22/95
35  */
36 
37 /*
38  * Copyright 2000 Wasabi Systems, Inc.
39  * All rights reserved.
40  *
41  * Written by Frank van der Linden for Wasabi Systems, Inc.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice, this list of conditions and the following disclaimer.
48  * 2. Redistributions in binary form must reproduce the above copyright
49  *    notice, this list of conditions and the following disclaimer in the
50  *    documentation and/or other materials provided with the distribution.
51  * 3. All advertising materials mentioning features or use of this software
52  *    must display the following acknowledgement:
53  *      This product includes software developed for the NetBSD Project by
54  *      Wasabi Systems, Inc.
55  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
56  *    or promote products derived from this software without specific prior
57  *    written permission.
58  *
59  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
60  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
61  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
62  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
63  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
64  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
65  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
66  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
67  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
68  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
69  * POSSIBILITY OF SUCH DAMAGE.
70  */
71 
72 #include <sys/cdefs.h>
73 __KERNEL_RCSID(0, "$NetBSD: nfs_subs.c,v 1.156 2005/12/11 12:25:16 christos Exp $");
74 
75 #include "fs_nfs.h"
76 #include "opt_nfs.h"
77 #include "opt_nfsserver.h"
78 #include "opt_iso.h"
79 #include "opt_inet.h"
80 
81 /*
82  * These functions support the macros and help fiddle mbuf chains for
83  * the nfs op functions. They do things like create the rpc header and
84  * copy data between mbuf chains and uio lists.
85  */
86 #include <sys/param.h>
87 #include <sys/proc.h>
88 #include <sys/systm.h>
89 #include <sys/kernel.h>
90 #include <sys/mount.h>
91 #include <sys/vnode.h>
92 #include <sys/namei.h>
93 #include <sys/mbuf.h>
94 #include <sys/socket.h>
95 #include <sys/stat.h>
96 #include <sys/malloc.h>
97 #include <sys/filedesc.h>
98 #include <sys/time.h>
99 #include <sys/dirent.h>
100 #include <sys/once.h>
101 
102 #include <uvm/uvm_extern.h>
103 
104 #include <nfs/rpcv2.h>
105 #include <nfs/nfsproto.h>
106 #include <nfs/nfsnode.h>
107 #include <nfs/nfs.h>
108 #include <nfs/xdr_subs.h>
109 #include <nfs/nfsm_subs.h>
110 #include <nfs/nfsmount.h>
111 #include <nfs/nqnfs.h>
112 #include <nfs/nfsrtt.h>
113 #include <nfs/nfs_var.h>
114 
115 #include <miscfs/specfs/specdev.h>
116 
117 #include <netinet/in.h>
118 #ifdef ISO
119 #include <netiso/iso.h>
120 #endif
121 
122 /*
123  * Data items converted to xdr at startup, since they are constant
124  * This is kinda hokey, but may save a little time doing byte swaps
125  */
126 u_int32_t nfs_xdrneg1;
127 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr,
128 	rpc_mismatch, rpc_auth_unix, rpc_msgaccepted,
129 	rpc_auth_kerb;
130 u_int32_t nfs_prog, nqnfs_prog, nfs_true, nfs_false;
131 
132 /* And other global data */
133 const nfstype nfsv2_type[9] =
134 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON };
135 const nfstype nfsv3_type[9] =
136 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON };
137 const enum vtype nv2tov_type[8] =
138 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON };
139 const enum vtype nv3tov_type[8] =
140 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO };
141 int nfs_ticks;
142 int nfs_commitsize;
143 
144 MALLOC_DEFINE(M_NFSDIROFF, "NFS diroff", "NFS directory cookies");
145 
146 /* NFS client/server stats. */
147 struct nfsstats nfsstats;
148 
149 /*
150  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
151  */
152 const int nfsv3_procid[NFS_NPROCS] = {
153 	NFSPROC_NULL,
154 	NFSPROC_GETATTR,
155 	NFSPROC_SETATTR,
156 	NFSPROC_NOOP,
157 	NFSPROC_LOOKUP,
158 	NFSPROC_READLINK,
159 	NFSPROC_READ,
160 	NFSPROC_NOOP,
161 	NFSPROC_WRITE,
162 	NFSPROC_CREATE,
163 	NFSPROC_REMOVE,
164 	NFSPROC_RENAME,
165 	NFSPROC_LINK,
166 	NFSPROC_SYMLINK,
167 	NFSPROC_MKDIR,
168 	NFSPROC_RMDIR,
169 	NFSPROC_READDIR,
170 	NFSPROC_FSSTAT,
171 	NFSPROC_NOOP,
172 	NFSPROC_NOOP,
173 	NFSPROC_NOOP,
174 	NFSPROC_NOOP,
175 	NFSPROC_NOOP,
176 	NFSPROC_NOOP,
177 	NFSPROC_NOOP,
178 	NFSPROC_NOOP
179 };
180 
181 /*
182  * and the reverse mapping from generic to Version 2 procedure numbers
183  */
184 const int nfsv2_procid[NFS_NPROCS] = {
185 	NFSV2PROC_NULL,
186 	NFSV2PROC_GETATTR,
187 	NFSV2PROC_SETATTR,
188 	NFSV2PROC_LOOKUP,
189 	NFSV2PROC_NOOP,
190 	NFSV2PROC_READLINK,
191 	NFSV2PROC_READ,
192 	NFSV2PROC_WRITE,
193 	NFSV2PROC_CREATE,
194 	NFSV2PROC_MKDIR,
195 	NFSV2PROC_SYMLINK,
196 	NFSV2PROC_CREATE,
197 	NFSV2PROC_REMOVE,
198 	NFSV2PROC_RMDIR,
199 	NFSV2PROC_RENAME,
200 	NFSV2PROC_LINK,
201 	NFSV2PROC_READDIR,
202 	NFSV2PROC_NOOP,
203 	NFSV2PROC_STATFS,
204 	NFSV2PROC_NOOP,
205 	NFSV2PROC_NOOP,
206 	NFSV2PROC_NOOP,
207 	NFSV2PROC_NOOP,
208 	NFSV2PROC_NOOP,
209 	NFSV2PROC_NOOP,
210 	NFSV2PROC_NOOP,
211 };
212 
213 /*
214  * Maps errno values to nfs error numbers.
215  * Use NFSERR_IO as the catch all for ones not specifically defined in
216  * RFC 1094.
217  */
218 static const u_char nfsrv_v2errmap[ELAST] = {
219   NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
220   NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
221   NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
222   NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
223   NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
224   NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
225   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
226   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
227   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
228   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
229   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
230   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
231   NFSERR_IO,	NFSERR_IO,	NFSERR_NAMETOL,	NFSERR_IO,	NFSERR_IO,
232   NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
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,
236 };
237 
238 /*
239  * Maps errno values to nfs error numbers.
240  * Although it is not obvious whether or not NFS clients really care if
241  * a returned error value is in the specified list for the procedure, the
242  * safest thing to do is filter them appropriately. For Version 2, the
243  * X/Open XNFS document is the only specification that defines error values
244  * for each RPC (The RFC simply lists all possible error values for all RPCs),
245  * so I have decided to not do this for Version 2.
246  * The first entry is the default error return and the rest are the valid
247  * errors for that RPC in increasing numeric order.
248  */
249 static const short nfsv3err_null[] = {
250 	0,
251 	0,
252 };
253 
254 static const short nfsv3err_getattr[] = {
255 	NFSERR_IO,
256 	NFSERR_IO,
257 	NFSERR_STALE,
258 	NFSERR_BADHANDLE,
259 	NFSERR_SERVERFAULT,
260 	0,
261 };
262 
263 static const short nfsv3err_setattr[] = {
264 	NFSERR_IO,
265 	NFSERR_PERM,
266 	NFSERR_IO,
267 	NFSERR_ACCES,
268 	NFSERR_INVAL,
269 	NFSERR_NOSPC,
270 	NFSERR_ROFS,
271 	NFSERR_DQUOT,
272 	NFSERR_STALE,
273 	NFSERR_BADHANDLE,
274 	NFSERR_NOT_SYNC,
275 	NFSERR_SERVERFAULT,
276 	0,
277 };
278 
279 static const short nfsv3err_lookup[] = {
280 	NFSERR_IO,
281 	NFSERR_NOENT,
282 	NFSERR_IO,
283 	NFSERR_ACCES,
284 	NFSERR_NOTDIR,
285 	NFSERR_NAMETOL,
286 	NFSERR_STALE,
287 	NFSERR_BADHANDLE,
288 	NFSERR_SERVERFAULT,
289 	0,
290 };
291 
292 static const short nfsv3err_access[] = {
293 	NFSERR_IO,
294 	NFSERR_IO,
295 	NFSERR_STALE,
296 	NFSERR_BADHANDLE,
297 	NFSERR_SERVERFAULT,
298 	0,
299 };
300 
301 static const short nfsv3err_readlink[] = {
302 	NFSERR_IO,
303 	NFSERR_IO,
304 	NFSERR_ACCES,
305 	NFSERR_INVAL,
306 	NFSERR_STALE,
307 	NFSERR_BADHANDLE,
308 	NFSERR_NOTSUPP,
309 	NFSERR_SERVERFAULT,
310 	0,
311 };
312 
313 static const short nfsv3err_read[] = {
314 	NFSERR_IO,
315 	NFSERR_IO,
316 	NFSERR_NXIO,
317 	NFSERR_ACCES,
318 	NFSERR_INVAL,
319 	NFSERR_STALE,
320 	NFSERR_BADHANDLE,
321 	NFSERR_SERVERFAULT,
322 	NFSERR_JUKEBOX,
323 	0,
324 };
325 
326 static const short nfsv3err_write[] = {
327 	NFSERR_IO,
328 	NFSERR_IO,
329 	NFSERR_ACCES,
330 	NFSERR_INVAL,
331 	NFSERR_FBIG,
332 	NFSERR_NOSPC,
333 	NFSERR_ROFS,
334 	NFSERR_DQUOT,
335 	NFSERR_STALE,
336 	NFSERR_BADHANDLE,
337 	NFSERR_SERVERFAULT,
338 	NFSERR_JUKEBOX,
339 	0,
340 };
341 
342 static const short nfsv3err_create[] = {
343 	NFSERR_IO,
344 	NFSERR_IO,
345 	NFSERR_ACCES,
346 	NFSERR_EXIST,
347 	NFSERR_NOTDIR,
348 	NFSERR_NOSPC,
349 	NFSERR_ROFS,
350 	NFSERR_NAMETOL,
351 	NFSERR_DQUOT,
352 	NFSERR_STALE,
353 	NFSERR_BADHANDLE,
354 	NFSERR_NOTSUPP,
355 	NFSERR_SERVERFAULT,
356 	0,
357 };
358 
359 static const short nfsv3err_mkdir[] = {
360 	NFSERR_IO,
361 	NFSERR_IO,
362 	NFSERR_ACCES,
363 	NFSERR_EXIST,
364 	NFSERR_NOTDIR,
365 	NFSERR_NOSPC,
366 	NFSERR_ROFS,
367 	NFSERR_NAMETOL,
368 	NFSERR_DQUOT,
369 	NFSERR_STALE,
370 	NFSERR_BADHANDLE,
371 	NFSERR_NOTSUPP,
372 	NFSERR_SERVERFAULT,
373 	0,
374 };
375 
376 static const short nfsv3err_symlink[] = {
377 	NFSERR_IO,
378 	NFSERR_IO,
379 	NFSERR_ACCES,
380 	NFSERR_EXIST,
381 	NFSERR_NOTDIR,
382 	NFSERR_NOSPC,
383 	NFSERR_ROFS,
384 	NFSERR_NAMETOL,
385 	NFSERR_DQUOT,
386 	NFSERR_STALE,
387 	NFSERR_BADHANDLE,
388 	NFSERR_NOTSUPP,
389 	NFSERR_SERVERFAULT,
390 	0,
391 };
392 
393 static const short nfsv3err_mknod[] = {
394 	NFSERR_IO,
395 	NFSERR_IO,
396 	NFSERR_ACCES,
397 	NFSERR_EXIST,
398 	NFSERR_NOTDIR,
399 	NFSERR_NOSPC,
400 	NFSERR_ROFS,
401 	NFSERR_NAMETOL,
402 	NFSERR_DQUOT,
403 	NFSERR_STALE,
404 	NFSERR_BADHANDLE,
405 	NFSERR_NOTSUPP,
406 	NFSERR_SERVERFAULT,
407 	NFSERR_BADTYPE,
408 	0,
409 };
410 
411 static const short nfsv3err_remove[] = {
412 	NFSERR_IO,
413 	NFSERR_NOENT,
414 	NFSERR_IO,
415 	NFSERR_ACCES,
416 	NFSERR_NOTDIR,
417 	NFSERR_ROFS,
418 	NFSERR_NAMETOL,
419 	NFSERR_STALE,
420 	NFSERR_BADHANDLE,
421 	NFSERR_SERVERFAULT,
422 	0,
423 };
424 
425 static const short nfsv3err_rmdir[] = {
426 	NFSERR_IO,
427 	NFSERR_NOENT,
428 	NFSERR_IO,
429 	NFSERR_ACCES,
430 	NFSERR_EXIST,
431 	NFSERR_NOTDIR,
432 	NFSERR_INVAL,
433 	NFSERR_ROFS,
434 	NFSERR_NAMETOL,
435 	NFSERR_NOTEMPTY,
436 	NFSERR_STALE,
437 	NFSERR_BADHANDLE,
438 	NFSERR_NOTSUPP,
439 	NFSERR_SERVERFAULT,
440 	0,
441 };
442 
443 static const short nfsv3err_rename[] = {
444 	NFSERR_IO,
445 	NFSERR_NOENT,
446 	NFSERR_IO,
447 	NFSERR_ACCES,
448 	NFSERR_EXIST,
449 	NFSERR_XDEV,
450 	NFSERR_NOTDIR,
451 	NFSERR_ISDIR,
452 	NFSERR_INVAL,
453 	NFSERR_NOSPC,
454 	NFSERR_ROFS,
455 	NFSERR_MLINK,
456 	NFSERR_NAMETOL,
457 	NFSERR_NOTEMPTY,
458 	NFSERR_DQUOT,
459 	NFSERR_STALE,
460 	NFSERR_BADHANDLE,
461 	NFSERR_NOTSUPP,
462 	NFSERR_SERVERFAULT,
463 	0,
464 };
465 
466 static const short nfsv3err_link[] = {
467 	NFSERR_IO,
468 	NFSERR_IO,
469 	NFSERR_ACCES,
470 	NFSERR_EXIST,
471 	NFSERR_XDEV,
472 	NFSERR_NOTDIR,
473 	NFSERR_INVAL,
474 	NFSERR_NOSPC,
475 	NFSERR_ROFS,
476 	NFSERR_MLINK,
477 	NFSERR_NAMETOL,
478 	NFSERR_DQUOT,
479 	NFSERR_STALE,
480 	NFSERR_BADHANDLE,
481 	NFSERR_NOTSUPP,
482 	NFSERR_SERVERFAULT,
483 	0,
484 };
485 
486 static const short nfsv3err_readdir[] = {
487 	NFSERR_IO,
488 	NFSERR_IO,
489 	NFSERR_ACCES,
490 	NFSERR_NOTDIR,
491 	NFSERR_STALE,
492 	NFSERR_BADHANDLE,
493 	NFSERR_BAD_COOKIE,
494 	NFSERR_TOOSMALL,
495 	NFSERR_SERVERFAULT,
496 	0,
497 };
498 
499 static const short nfsv3err_readdirplus[] = {
500 	NFSERR_IO,
501 	NFSERR_IO,
502 	NFSERR_ACCES,
503 	NFSERR_NOTDIR,
504 	NFSERR_STALE,
505 	NFSERR_BADHANDLE,
506 	NFSERR_BAD_COOKIE,
507 	NFSERR_NOTSUPP,
508 	NFSERR_TOOSMALL,
509 	NFSERR_SERVERFAULT,
510 	0,
511 };
512 
513 static const short nfsv3err_fsstat[] = {
514 	NFSERR_IO,
515 	NFSERR_IO,
516 	NFSERR_STALE,
517 	NFSERR_BADHANDLE,
518 	NFSERR_SERVERFAULT,
519 	0,
520 };
521 
522 static const short nfsv3err_fsinfo[] = {
523 	NFSERR_STALE,
524 	NFSERR_STALE,
525 	NFSERR_BADHANDLE,
526 	NFSERR_SERVERFAULT,
527 	0,
528 };
529 
530 static const short nfsv3err_pathconf[] = {
531 	NFSERR_STALE,
532 	NFSERR_STALE,
533 	NFSERR_BADHANDLE,
534 	NFSERR_SERVERFAULT,
535 	0,
536 };
537 
538 static const short nfsv3err_commit[] = {
539 	NFSERR_IO,
540 	NFSERR_IO,
541 	NFSERR_STALE,
542 	NFSERR_BADHANDLE,
543 	NFSERR_SERVERFAULT,
544 	0,
545 };
546 
547 static const short * const nfsrv_v3errmap[] = {
548 	nfsv3err_null,
549 	nfsv3err_getattr,
550 	nfsv3err_setattr,
551 	nfsv3err_lookup,
552 	nfsv3err_access,
553 	nfsv3err_readlink,
554 	nfsv3err_read,
555 	nfsv3err_write,
556 	nfsv3err_create,
557 	nfsv3err_mkdir,
558 	nfsv3err_symlink,
559 	nfsv3err_mknod,
560 	nfsv3err_remove,
561 	nfsv3err_rmdir,
562 	nfsv3err_rename,
563 	nfsv3err_link,
564 	nfsv3err_readdir,
565 	nfsv3err_readdirplus,
566 	nfsv3err_fsstat,
567 	nfsv3err_fsinfo,
568 	nfsv3err_pathconf,
569 	nfsv3err_commit,
570 };
571 
572 extern struct nfsrtt nfsrtt;
573 extern time_t nqnfsstarttime;
574 extern int nqsrv_clockskew;
575 extern int nqsrv_writeslack;
576 extern int nqsrv_maxlease;
577 extern const int nqnfs_piggy[NFS_NPROCS];
578 extern struct nfsnodehashhead *nfsnodehashtbl;
579 extern u_long nfsnodehash;
580 
581 u_long nfsdirhashmask;
582 
583 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *));
584 
585 /*
586  * Create the header for an rpc request packet
587  * The hsiz is the size of the rest of the nfs request header.
588  * (just used to decide if a cluster is a good idea)
589  */
590 struct mbuf *
591 nfsm_reqh(np, procid, hsiz, bposp)
592 	struct nfsnode *np;
593 	u_long procid;
594 	int hsiz;
595 	caddr_t *bposp;
596 {
597 	struct mbuf *mb;
598 	caddr_t bpos;
599 #ifndef NFS_V2_ONLY
600 	struct nfsmount *nmp;
601 	u_int32_t *tl;
602 	int nqflag;
603 #endif
604 
605 	mb = m_get(M_WAIT, MT_DATA);
606 	MCLAIM(mb, &nfs_mowner);
607 	if (hsiz >= MINCLSIZE)
608 		m_clget(mb, M_WAIT);
609 	mb->m_len = 0;
610 	bpos = mtod(mb, caddr_t);
611 
612 #ifndef NFS_V2_ONLY
613 	/*
614 	 * For NQNFS, add lease request.
615 	 */
616 	if (np) {
617 		nmp = VFSTONFS(np->n_vnode->v_mount);
618 		if (nmp->nm_flag & NFSMNT_NQNFS) {
619 			nqflag = NQNFS_NEEDLEASE(np, procid);
620 			if (nqflag) {
621 				nfsm_build(tl, u_int32_t *, 2*NFSX_UNSIGNED);
622 				*tl++ = txdr_unsigned(nqflag);
623 				*tl = txdr_unsigned(nmp->nm_leaseterm);
624 			} else {
625 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
626 				*tl = 0;
627 			}
628 		}
629 	}
630 #endif
631 	/* Finally, return values */
632 	*bposp = bpos;
633 	return (mb);
634 }
635 
636 /*
637  * Build the RPC header and fill in the authorization info.
638  * The authorization string argument is only used when the credentials
639  * come from outside of the kernel.
640  * Returns the head of the mbuf list.
641  */
642 struct mbuf *
643 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len,
644 	verf_str, mrest, mrest_len, mbp, xidp)
645 	struct ucred *cr;
646 	int nmflag;
647 	int procid;
648 	int auth_type;
649 	int auth_len;
650 	char *auth_str;
651 	int verf_len;
652 	char *verf_str;
653 	struct mbuf *mrest;
654 	int mrest_len;
655 	struct mbuf **mbp;
656 	u_int32_t *xidp;
657 {
658 	struct mbuf *mb;
659 	u_int32_t *tl;
660 	caddr_t bpos;
661 	int i;
662 	struct mbuf *mreq;
663 	int siz, grpsiz, authsiz;
664 
665 	authsiz = nfsm_rndup(auth_len);
666 	mb = m_gethdr(M_WAIT, MT_DATA);
667 	MCLAIM(mb, &nfs_mowner);
668 	if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
669 		m_clget(mb, M_WAIT);
670 	} else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
671 		MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED);
672 	} else {
673 		MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
674 	}
675 	mb->m_len = 0;
676 	mreq = mb;
677 	bpos = mtod(mb, caddr_t);
678 
679 	/*
680 	 * First the RPC header.
681 	 */
682 	nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
683 
684 	*tl++ = *xidp = nfs_getxid();
685 	*tl++ = rpc_call;
686 	*tl++ = rpc_vers;
687 	if (nmflag & NFSMNT_NQNFS) {
688 		*tl++ = txdr_unsigned(NQNFS_PROG);
689 		*tl++ = txdr_unsigned(NQNFS_VER3);
690 	} else {
691 		*tl++ = txdr_unsigned(NFS_PROG);
692 		if (nmflag & NFSMNT_NFSV3)
693 			*tl++ = txdr_unsigned(NFS_VER3);
694 		else
695 			*tl++ = txdr_unsigned(NFS_VER2);
696 	}
697 	if (nmflag & NFSMNT_NFSV3)
698 		*tl++ = txdr_unsigned(procid);
699 	else
700 		*tl++ = txdr_unsigned(nfsv2_procid[procid]);
701 
702 	/*
703 	 * And then the authorization cred.
704 	 */
705 	*tl++ = txdr_unsigned(auth_type);
706 	*tl = txdr_unsigned(authsiz);
707 	switch (auth_type) {
708 	case RPCAUTH_UNIX:
709 		nfsm_build(tl, u_int32_t *, auth_len);
710 		*tl++ = 0;		/* stamp ?? */
711 		*tl++ = 0;		/* NULL hostname */
712 		*tl++ = txdr_unsigned(cr->cr_uid);
713 		*tl++ = txdr_unsigned(cr->cr_gid);
714 		grpsiz = (auth_len >> 2) - 5;
715 		*tl++ = txdr_unsigned(grpsiz);
716 		for (i = 0; i < grpsiz; i++)
717 			*tl++ = txdr_unsigned(cr->cr_groups[i]);
718 		break;
719 	case RPCAUTH_KERB4:
720 		siz = auth_len;
721 		while (siz > 0) {
722 			if (M_TRAILINGSPACE(mb) == 0) {
723 				struct mbuf *mb2;
724 				mb2 = m_get(M_WAIT, MT_DATA);
725 				MCLAIM(mb2, &nfs_mowner);
726 				if (siz >= MINCLSIZE)
727 					m_clget(mb2, M_WAIT);
728 				mb->m_next = mb2;
729 				mb = mb2;
730 				mb->m_len = 0;
731 				bpos = mtod(mb, caddr_t);
732 			}
733 			i = min(siz, M_TRAILINGSPACE(mb));
734 			memcpy(bpos, auth_str, i);
735 			mb->m_len += i;
736 			auth_str += i;
737 			bpos += i;
738 			siz -= i;
739 		}
740 		if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
741 			for (i = 0; i < siz; i++)
742 				*bpos++ = '\0';
743 			mb->m_len += siz;
744 		}
745 		break;
746 	};
747 
748 	/*
749 	 * And the verifier...
750 	 */
751 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
752 	if (verf_str) {
753 		*tl++ = txdr_unsigned(RPCAUTH_KERB4);
754 		*tl = txdr_unsigned(verf_len);
755 		siz = verf_len;
756 		while (siz > 0) {
757 			if (M_TRAILINGSPACE(mb) == 0) {
758 				struct mbuf *mb2;
759 				mb2 = m_get(M_WAIT, MT_DATA);
760 				MCLAIM(mb2, &nfs_mowner);
761 				if (siz >= MINCLSIZE)
762 					m_clget(mb2, M_WAIT);
763 				mb->m_next = mb2;
764 				mb = mb2;
765 				mb->m_len = 0;
766 				bpos = mtod(mb, caddr_t);
767 			}
768 			i = min(siz, M_TRAILINGSPACE(mb));
769 			memcpy(bpos, verf_str, i);
770 			mb->m_len += i;
771 			verf_str += i;
772 			bpos += i;
773 			siz -= i;
774 		}
775 		if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
776 			for (i = 0; i < siz; i++)
777 				*bpos++ = '\0';
778 			mb->m_len += siz;
779 		}
780 	} else {
781 		*tl++ = txdr_unsigned(RPCAUTH_NULL);
782 		*tl = 0;
783 	}
784 	mb->m_next = mrest;
785 	mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
786 	mreq->m_pkthdr.rcvif = (struct ifnet *)0;
787 	*mbp = mb;
788 	return (mreq);
789 }
790 
791 /*
792  * copies mbuf chain to the uio scatter/gather list
793  */
794 int
795 nfsm_mbuftouio(mrep, uiop, siz, dpos)
796 	struct mbuf **mrep;
797 	struct uio *uiop;
798 	int siz;
799 	caddr_t *dpos;
800 {
801 	char *mbufcp, *uiocp;
802 	int xfer, left, len;
803 	struct mbuf *mp;
804 	long uiosiz, rem;
805 	int error = 0;
806 
807 	mp = *mrep;
808 	mbufcp = *dpos;
809 	len = mtod(mp, caddr_t)+mp->m_len-mbufcp;
810 	rem = nfsm_rndup(siz)-siz;
811 	while (siz > 0) {
812 		if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
813 			return (EFBIG);
814 		left = uiop->uio_iov->iov_len;
815 		uiocp = uiop->uio_iov->iov_base;
816 		if (left > siz)
817 			left = siz;
818 		uiosiz = left;
819 		while (left > 0) {
820 			while (len == 0) {
821 				mp = mp->m_next;
822 				if (mp == NULL)
823 					return (EBADRPC);
824 				mbufcp = mtod(mp, caddr_t);
825 				len = mp->m_len;
826 			}
827 			xfer = (left > len) ? len : left;
828 #ifdef notdef
829 			/* Not Yet.. */
830 			if (uiop->uio_iov->iov_op != NULL)
831 				(*(uiop->uio_iov->iov_op))
832 				(mbufcp, uiocp, xfer);
833 			else
834 #endif
835 			if (uiop->uio_segflg == UIO_SYSSPACE)
836 				memcpy(uiocp, mbufcp, xfer);
837 			else
838 				if ((error = copyout_proc(uiop->uio_lwp->l_proc,
839 				    mbufcp, uiocp, xfer)) != 0)
840 					return error;
841 			left -= xfer;
842 			len -= xfer;
843 			mbufcp += xfer;
844 			uiocp += xfer;
845 			uiop->uio_offset += xfer;
846 			uiop->uio_resid -= xfer;
847 		}
848 		if (uiop->uio_iov->iov_len <= siz) {
849 			uiop->uio_iovcnt--;
850 			uiop->uio_iov++;
851 		} else {
852 			uiop->uio_iov->iov_base =
853 			    (caddr_t)uiop->uio_iov->iov_base + uiosiz;
854 			uiop->uio_iov->iov_len -= uiosiz;
855 		}
856 		siz -= uiosiz;
857 	}
858 	*dpos = mbufcp;
859 	*mrep = mp;
860 	if (rem > 0) {
861 		if (len < rem)
862 			error = nfs_adv(mrep, dpos, rem, len);
863 		else
864 			*dpos += rem;
865 	}
866 	return (error);
867 }
868 
869 /*
870  * copies a uio scatter/gather list to an mbuf chain.
871  * NOTE: can ony handle iovcnt == 1
872  */
873 int
874 nfsm_uiotombuf(uiop, mq, siz, bpos)
875 	struct uio *uiop;
876 	struct mbuf **mq;
877 	int siz;
878 	caddr_t *bpos;
879 {
880 	char *uiocp;
881 	struct mbuf *mp, *mp2;
882 	int xfer, left, mlen;
883 	int uiosiz, clflg, rem;
884 	char *cp;
885 
886 #ifdef DIAGNOSTIC
887 	if (uiop->uio_iovcnt != 1)
888 		panic("nfsm_uiotombuf: iovcnt != 1");
889 #endif
890 
891 	if (siz > MLEN)		/* or should it >= MCLBYTES ?? */
892 		clflg = 1;
893 	else
894 		clflg = 0;
895 	rem = nfsm_rndup(siz)-siz;
896 	mp = mp2 = *mq;
897 	while (siz > 0) {
898 		left = uiop->uio_iov->iov_len;
899 		uiocp = uiop->uio_iov->iov_base;
900 		if (left > siz)
901 			left = siz;
902 		uiosiz = left;
903 		while (left > 0) {
904 			mlen = M_TRAILINGSPACE(mp);
905 			if (mlen == 0) {
906 				mp = m_get(M_WAIT, MT_DATA);
907 				MCLAIM(mp, &nfs_mowner);
908 				if (clflg)
909 					m_clget(mp, M_WAIT);
910 				mp->m_len = 0;
911 				mp2->m_next = mp;
912 				mp2 = mp;
913 				mlen = M_TRAILINGSPACE(mp);
914 			}
915 			xfer = (left > mlen) ? mlen : left;
916 			cp = mtod(mp, caddr_t) + mp->m_len;
917 #ifdef notdef
918 			/* Not Yet.. */
919 			if (uiop->uio_iov->iov_op != NULL)
920 				(*(uiop->uio_iov->iov_op))(uiocp, cp, xfer);
921 			else
922 #endif
923 			if (uiop->uio_segflg == UIO_SYSSPACE)
924 				(void)memcpy(cp, uiocp, xfer);
925 			else
926 				/*XXX: Check error */
927 				(void)copyin_proc(uiop->uio_lwp->l_proc, uiocp,
928 				    cp, xfer);
929 			mp->m_len += xfer;
930 			left -= xfer;
931 			uiocp += xfer;
932 			uiop->uio_offset += xfer;
933 			uiop->uio_resid -= xfer;
934 		}
935 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
936 		    uiosiz;
937 		uiop->uio_iov->iov_len -= uiosiz;
938 		siz -= uiosiz;
939 	}
940 	if (rem > 0) {
941 		if (rem > M_TRAILINGSPACE(mp)) {
942 			mp = m_get(M_WAIT, MT_DATA);
943 			MCLAIM(mp, &nfs_mowner);
944 			mp->m_len = 0;
945 			mp2->m_next = mp;
946 		}
947 		cp = mtod(mp, caddr_t) + mp->m_len;
948 		for (left = 0; left < rem; left++)
949 			*cp++ = '\0';
950 		mp->m_len += rem;
951 		*bpos = cp;
952 	} else
953 		*bpos = mtod(mp, caddr_t)+mp->m_len;
954 	*mq = mp;
955 	return (0);
956 }
957 
958 /*
959  * Get at least "siz" bytes of correctly aligned data.
960  * When called the mbuf pointers are not necessarily correct,
961  * dsosp points to what ought to be in m_data and left contains
962  * what ought to be in m_len.
963  * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
964  * cases. (The macros use the vars. dpos and dpos2)
965  */
966 int
967 nfsm_disct(mdp, dposp, siz, left, cp2)
968 	struct mbuf **mdp;
969 	caddr_t *dposp;
970 	int siz;
971 	int left;
972 	caddr_t *cp2;
973 {
974 	struct mbuf *m1, *m2;
975 	struct mbuf *havebuf = NULL;
976 	caddr_t src = *dposp;
977 	caddr_t dst;
978 	int len;
979 
980 #ifdef DEBUG
981 	if (left < 0)
982 		panic("nfsm_disct: left < 0");
983 #endif
984 	m1 = *mdp;
985 	/*
986 	 * Skip through the mbuf chain looking for an mbuf with
987 	 * some data. If the first mbuf found has enough data
988 	 * and it is correctly aligned return it.
989 	 */
990 	while (left == 0) {
991 		havebuf = m1;
992 		*mdp = m1 = m1->m_next;
993 		if (m1 == NULL)
994 			return (EBADRPC);
995 		src = mtod(m1, caddr_t);
996 		left = m1->m_len;
997 		/*
998 		 * If we start a new mbuf and it is big enough
999 		 * and correctly aligned just return it, don't
1000 		 * do any pull up.
1001 		 */
1002 		if (left >= siz && nfsm_aligned(src)) {
1003 			*cp2 = src;
1004 			*dposp = src + siz;
1005 			return (0);
1006 		}
1007 	}
1008 	if (m1->m_flags & M_EXT) {
1009 		if (havebuf) {
1010 			/* If the first mbuf with data has external data
1011 			 * and there is a previous empty mbuf use it
1012 			 * to move the data into.
1013 			 */
1014 			m2 = m1;
1015 			*mdp = m1 = havebuf;
1016 			if (m1->m_flags & M_EXT) {
1017 				MEXTREMOVE(m1);
1018 			}
1019 		} else {
1020 			/*
1021 			 * If the first mbuf has a external data
1022 			 * and there is no previous empty mbuf
1023 			 * allocate a new mbuf and move the external
1024 			 * data to the new mbuf. Also make the first
1025 			 * mbuf look empty.
1026 			 */
1027 			m2 = m_get(M_WAIT, MT_DATA);
1028 			m2->m_ext = m1->m_ext;
1029 			m2->m_data = src;
1030 			m2->m_len = left;
1031 			MCLADDREFERENCE(m1, m2);
1032 			MEXTREMOVE(m1);
1033 			m2->m_next = m1->m_next;
1034 			m1->m_next = m2;
1035 		}
1036 		m1->m_len = 0;
1037 		if (m1->m_flags & M_PKTHDR)
1038 			dst = m1->m_pktdat;
1039 		else
1040 			dst = m1->m_dat;
1041 		m1->m_data = dst;
1042 	} else {
1043 		/*
1044 		 * If the first mbuf has no external data
1045 		 * move the data to the front of the mbuf.
1046 		 */
1047 		if (m1->m_flags & M_PKTHDR)
1048 			dst = m1->m_pktdat;
1049 		else
1050 			dst = m1->m_dat;
1051 		m1->m_data = dst;
1052 		if (dst != src)
1053 			memmove(dst, src, left);
1054 		dst += left;
1055 		m1->m_len = left;
1056 		m2 = m1->m_next;
1057 	}
1058 	*cp2 = m1->m_data;
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 = M_TRAILINGSPACE(m1)) != 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 		m1 = m_get(M_WAIT, MT_DATA);
1141 		MCLAIM(m1, &nfs_mowner);
1142 		if (siz > MLEN)
1143 			m_clget(m1, M_WAIT);
1144 		m1->m_len = NFSMSIZ(m1);
1145 		m2->m_next = m1;
1146 		m2 = m1;
1147 		tl = mtod(m1, u_int32_t *);
1148 		tlen = 0;
1149 		if (putsize) {
1150 			*tl++ = txdr_unsigned(siz);
1151 			m1->m_len -= NFSX_UNSIGNED;
1152 			tlen = NFSX_UNSIGNED;
1153 			putsize = 0;
1154 		}
1155 		if (siz < m1->m_len) {
1156 			len = nfsm_rndup(siz);
1157 			xfer = siz;
1158 			if (xfer < len)
1159 				*(tl+(xfer>>2)) = 0;
1160 		} else {
1161 			xfer = len = m1->m_len;
1162 		}
1163 		memcpy((caddr_t) tl, cp, xfer);
1164 		m1->m_len = len+tlen;
1165 		siz -= xfer;
1166 		cp += xfer;
1167 	}
1168 	*mb = m1;
1169 	*bpos = mtod(m1, caddr_t)+m1->m_len;
1170 	return (0);
1171 }
1172 
1173 /*
1174  * Directory caching routines. They work as follows:
1175  * - a cache is maintained per VDIR nfsnode.
1176  * - for each offset cookie that is exported to userspace, and can
1177  *   thus be thrown back at us as an offset to VOP_READDIR, store
1178  *   information in the cache.
1179  * - cached are:
1180  *   - cookie itself
1181  *   - blocknumber (essentially just a search key in the buffer cache)
1182  *   - entry number in block.
1183  *   - offset cookie of block in which this entry is stored
1184  *   - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
1185  * - entries are looked up in a hash table
1186  * - also maintained is an LRU list of entries, used to determine
1187  *   which ones to delete if the cache grows too large.
1188  * - if 32 <-> 64 translation mode is requested for a filesystem,
1189  *   the cache also functions as a translation table
1190  * - in the translation case, invalidating the cache does not mean
1191  *   flushing it, but just marking entries as invalid, except for
1192  *   the <64bit cookie, 32bitcookie> pair which is still valid, to
1193  *   still be able to use the cache as a translation table.
1194  * - 32 bit cookies are uniquely created by combining the hash table
1195  *   entry value, and one generation count per hash table entry,
1196  *   incremented each time an entry is appended to the chain.
1197  * - the cache is invalidated each time a direcory is modified
1198  * - sanity checks are also done; if an entry in a block turns
1199  *   out not to have a matching cookie, the cache is invalidated
1200  *   and a new block starting from the wanted offset is fetched from
1201  *   the server.
1202  * - directory entries as read from the server are extended to contain
1203  *   the 64bit and, optionally, the 32bit cookies, for sanity checking
1204  *   the cache and exporting them to userspace through the cookie
1205  *   argument to VOP_READDIR.
1206  */
1207 
1208 u_long
1209 nfs_dirhash(off)
1210 	off_t off;
1211 {
1212 	int i;
1213 	char *cp = (char *)&off;
1214 	u_long sum = 0L;
1215 
1216 	for (i = 0 ; i < sizeof (off); i++)
1217 		sum += *cp++;
1218 
1219 	return sum;
1220 }
1221 
1222 #define	_NFSDC_MTX(np)		(&NFSTOV(np)->v_interlock)
1223 #define	NFSDC_LOCK(np)		simple_lock(_NFSDC_MTX(np))
1224 #define	NFSDC_UNLOCK(np)	simple_unlock(_NFSDC_MTX(np))
1225 #define	NFSDC_ASSERT_LOCKED(np) LOCK_ASSERT(simple_lock_held(_NFSDC_MTX(np)))
1226 
1227 void
1228 nfs_initdircache(vp)
1229 	struct vnode *vp;
1230 {
1231 	struct nfsnode *np = VTONFS(vp);
1232 	struct nfsdirhashhead *dircache;
1233 
1234 	dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF,
1235 	    M_WAITOK, &nfsdirhashmask);
1236 
1237 	NFSDC_LOCK(np);
1238 	if (np->n_dircache == NULL) {
1239 		np->n_dircachesize = 0;
1240 		np->n_dircache = dircache;
1241 		dircache = NULL;
1242 		TAILQ_INIT(&np->n_dirchain);
1243 	}
1244 	NFSDC_UNLOCK(np);
1245 	if (dircache)
1246 		hashdone(dircache, M_NFSDIROFF);
1247 }
1248 
1249 void
1250 nfs_initdirxlatecookie(vp)
1251 	struct vnode *vp;
1252 {
1253 	struct nfsnode *np = VTONFS(vp);
1254 	unsigned *dirgens;
1255 
1256 	KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE);
1257 
1258 	dirgens = malloc(NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF,
1259 	    M_WAITOK|M_ZERO);
1260 	NFSDC_LOCK(np);
1261 	if (np->n_dirgens == NULL) {
1262 		np->n_dirgens = dirgens;
1263 		dirgens = NULL;
1264 	}
1265 	NFSDC_UNLOCK(np);
1266 	if (dirgens)
1267 		free(dirgens, M_NFSDIROFF);
1268 }
1269 
1270 static const struct nfsdircache dzero;
1271 
1272 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *));
1273 static void nfs_putdircache_unlocked __P((struct nfsnode *,
1274     struct nfsdircache *));
1275 
1276 static void
1277 nfs_unlinkdircache(np, ndp)
1278 	struct nfsnode *np;
1279 	struct nfsdircache *ndp;
1280 {
1281 
1282 	NFSDC_ASSERT_LOCKED(np);
1283 	KASSERT(ndp != &dzero);
1284 
1285 	if (LIST_NEXT(ndp, dc_hash) == (void *)-1)
1286 		return;
1287 
1288 	TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1289 	LIST_REMOVE(ndp, dc_hash);
1290 	LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */
1291 
1292 	nfs_putdircache_unlocked(np, ndp);
1293 }
1294 
1295 void
1296 nfs_putdircache(np, ndp)
1297 	struct nfsnode *np;
1298 	struct nfsdircache *ndp;
1299 {
1300 	int ref;
1301 
1302 	if (ndp == &dzero)
1303 		return;
1304 
1305 	KASSERT(ndp->dc_refcnt > 0);
1306 	NFSDC_LOCK(np);
1307 	ref = --ndp->dc_refcnt;
1308 	NFSDC_UNLOCK(np);
1309 
1310 	if (ref == 0)
1311 		free(ndp, M_NFSDIROFF);
1312 }
1313 
1314 static void
1315 nfs_putdircache_unlocked(np, ndp)
1316 	struct nfsnode *np;
1317 	struct nfsdircache *ndp;
1318 {
1319 	int ref;
1320 
1321 	NFSDC_ASSERT_LOCKED(np);
1322 
1323 	if (ndp == &dzero)
1324 		return;
1325 
1326 	KASSERT(ndp->dc_refcnt > 0);
1327 	ref = --ndp->dc_refcnt;
1328 	if (ref == 0)
1329 		free(ndp, M_NFSDIROFF);
1330 }
1331 
1332 struct nfsdircache *
1333 nfs_searchdircache(vp, off, do32, hashent)
1334 	struct vnode *vp;
1335 	off_t off;
1336 	int do32;
1337 	int *hashent;
1338 {
1339 	struct nfsdirhashhead *ndhp;
1340 	struct nfsdircache *ndp = NULL;
1341 	struct nfsnode *np = VTONFS(vp);
1342 	unsigned ent;
1343 
1344 	/*
1345 	 * Zero is always a valid cookie.
1346 	 */
1347 	if (off == 0)
1348 		/* XXXUNCONST */
1349 		return (struct nfsdircache *)__UNCONST(&dzero);
1350 
1351 	if (!np->n_dircache)
1352 		return NULL;
1353 
1354 	/*
1355 	 * We use a 32bit cookie as search key, directly reconstruct
1356 	 * the hashentry. Else use the hashfunction.
1357 	 */
1358 	if (do32) {
1359 		ent = (u_int32_t)off >> 24;
1360 		if (ent >= NFS_DIRHASHSIZ)
1361 			return NULL;
1362 		ndhp = &np->n_dircache[ent];
1363 	} else {
1364 		ndhp = NFSDIRHASH(np, off);
1365 	}
1366 
1367 	if (hashent)
1368 		*hashent = (int)(ndhp - np->n_dircache);
1369 
1370 	NFSDC_LOCK(np);
1371 	if (do32) {
1372 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1373 			if (ndp->dc_cookie32 == (u_int32_t)off) {
1374 				/*
1375 				 * An invalidated entry will become the
1376 				 * start of a new block fetched from
1377 				 * the server.
1378 				 */
1379 				if (ndp->dc_flags & NFSDC_INVALID) {
1380 					ndp->dc_blkcookie = ndp->dc_cookie;
1381 					ndp->dc_entry = 0;
1382 					ndp->dc_flags &= ~NFSDC_INVALID;
1383 				}
1384 				break;
1385 			}
1386 		}
1387 	} else {
1388 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1389 			if (ndp->dc_cookie == off)
1390 				break;
1391 		}
1392 	}
1393 	if (ndp != NULL)
1394 		ndp->dc_refcnt++;
1395 	NFSDC_UNLOCK(np);
1396 	return ndp;
1397 }
1398 
1399 
1400 struct nfsdircache *
1401 nfs_enterdircache(vp, off, blkoff, en, blkno)
1402 	struct vnode *vp;
1403 	off_t off, blkoff;
1404 	int en;
1405 	daddr_t blkno;
1406 {
1407 	struct nfsnode *np = VTONFS(vp);
1408 	struct nfsdirhashhead *ndhp;
1409 	struct nfsdircache *ndp = NULL;
1410 	struct nfsdircache *newndp = NULL;
1411 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1412 	int hashent, gen, overwrite;
1413 
1414 	/*
1415 	 * XXX refuse entries for offset 0. amd(8) erroneously sets
1416 	 * cookie 0 for the '.' entry, making this necessary. This
1417 	 * isn't so bad, as 0 is a special case anyway.
1418 	 */
1419 	if (off == 0)
1420 		/* XXXUNCONST */
1421 		return (struct nfsdircache *)__UNCONST(&dzero);
1422 
1423 	if (!np->n_dircache)
1424 		/*
1425 		 * XXX would like to do this in nfs_nget but vtype
1426 		 * isn't known at that time.
1427 		 */
1428 		nfs_initdircache(vp);
1429 
1430 	if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens)
1431 		nfs_initdirxlatecookie(vp);
1432 
1433 retry:
1434 	ndp = nfs_searchdircache(vp, off, 0, &hashent);
1435 
1436 	NFSDC_LOCK(np);
1437 	if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) {
1438 		/*
1439 		 * Overwriting an old entry. Check if it's the same.
1440 		 * If so, just return. If not, remove the old entry.
1441 		 */
1442 		if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
1443 			goto done;
1444 		nfs_unlinkdircache(np, ndp);
1445 		nfs_putdircache_unlocked(np, ndp);
1446 		ndp = NULL;
1447 	}
1448 
1449 	ndhp = &np->n_dircache[hashent];
1450 
1451 	if (!ndp) {
1452 		if (newndp == NULL) {
1453 			NFSDC_UNLOCK(np);
1454 			newndp = malloc(sizeof(*ndp), M_NFSDIROFF, M_WAITOK);
1455 			newndp->dc_refcnt = 1;
1456 			LIST_NEXT(newndp, dc_hash) = (void *)-1;
1457 			goto retry;
1458 		}
1459 		ndp = newndp;
1460 		newndp = NULL;
1461 		overwrite = 0;
1462 		if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1463 			/*
1464 			 * We're allocating a new entry, so bump the
1465 			 * generation number.
1466 			 */
1467 			gen = ++np->n_dirgens[hashent];
1468 			if (gen == 0) {
1469 				np->n_dirgens[hashent]++;
1470 				gen++;
1471 			}
1472 			ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
1473 		}
1474 	} else
1475 		overwrite = 1;
1476 
1477 	ndp->dc_cookie = off;
1478 	ndp->dc_blkcookie = blkoff;
1479 	ndp->dc_entry = en;
1480 	ndp->dc_flags = 0;
1481 
1482 	if (overwrite)
1483 		goto done;
1484 
1485 	/*
1486 	 * If the maximum directory cookie cache size has been reached
1487 	 * for this node, take one off the front. The idea is that
1488 	 * directories are typically read front-to-back once, so that
1489 	 * the oldest entries can be thrown away without much performance
1490 	 * loss.
1491 	 */
1492 	if (np->n_dircachesize == NFS_MAXDIRCACHE) {
1493 		nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain));
1494 	} else
1495 		np->n_dircachesize++;
1496 
1497 	KASSERT(ndp->dc_refcnt == 1);
1498 	LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
1499 	TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
1500 	ndp->dc_refcnt++;
1501 done:
1502 	KASSERT(ndp->dc_refcnt > 0);
1503 	NFSDC_UNLOCK(np);
1504 	if (newndp)
1505 		nfs_putdircache(np, newndp);
1506 	return ndp;
1507 }
1508 
1509 void
1510 nfs_invaldircache(vp, flags)
1511 	struct vnode *vp;
1512 	int flags;
1513 {
1514 	struct nfsnode *np = VTONFS(vp);
1515 	struct nfsdircache *ndp = NULL;
1516 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1517 	const boolean_t forcefree = flags & NFS_INVALDIRCACHE_FORCE;
1518 
1519 #ifdef DIAGNOSTIC
1520 	if (vp->v_type != VDIR)
1521 		panic("nfs: invaldircache: not dir");
1522 #endif
1523 
1524 	if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0)
1525 		np->n_flag &= ~NEOFVALID;
1526 
1527 	if (!np->n_dircache)
1528 		return;
1529 
1530 	NFSDC_LOCK(np);
1531 	if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
1532 		while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) {
1533 			KASSERT(!forcefree || ndp->dc_refcnt == 1);
1534 			nfs_unlinkdircache(np, ndp);
1535 		}
1536 		np->n_dircachesize = 0;
1537 		if (forcefree && np->n_dirgens) {
1538 			FREE(np->n_dirgens, M_NFSDIROFF);
1539 			np->n_dirgens = NULL;
1540 		}
1541 	} else {
1542 		TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain)
1543 			ndp->dc_flags |= NFSDC_INVALID;
1544 	}
1545 
1546 	NFSDC_UNLOCK(np);
1547 }
1548 
1549 /*
1550  * Called once before VFS init to initialize shared and
1551  * server-specific data structures.
1552  */
1553 static void
1554 nfs_init0(void)
1555 {
1556 	nfsrtt.pos = 0;
1557 	rpc_vers = txdr_unsigned(RPC_VER2);
1558 	rpc_call = txdr_unsigned(RPC_CALL);
1559 	rpc_reply = txdr_unsigned(RPC_REPLY);
1560 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1561 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1562 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1563 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1564 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1565 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1566 	nfs_prog = txdr_unsigned(NFS_PROG);
1567 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
1568 	nfs_true = txdr_unsigned(TRUE);
1569 	nfs_false = txdr_unsigned(FALSE);
1570 	nfs_xdrneg1 = txdr_unsigned(-1);
1571 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1572 	if (nfs_ticks < 1)
1573 		nfs_ticks = 1;
1574 #ifdef NFSSERVER
1575 	nfsrv_init(0);			/* Init server data structures */
1576 	nfsrv_initcache();		/* Init the server request cache */
1577 	pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript),
1578 	    0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr);
1579 #endif /* NFSSERVER */
1580 
1581 #if defined(NFSSERVER) || !defined(NFS_V2_ONLY)
1582 	/*
1583 	 * Initialize the nqnfs data structures.
1584 	 */
1585 	if (nqnfsstarttime == 0) {
1586 		nqnfsstarttime = boottime.tv_sec + nqsrv_maxlease
1587 			+ nqsrv_clockskew + nqsrv_writeslack;
1588 		NQLOADNOVRAM(nqnfsstarttime);
1589 		CIRCLEQ_INIT(&nqtimerhead);
1590 		nqfhhashtbl = hashinit(NQLCHSZ, HASH_LIST, M_NQLEASE,
1591 		    M_WAITOK, &nqfhhash);
1592 	}
1593 #endif
1594 
1595 	exithook_establish(nfs_exit, NULL);
1596 
1597 	/*
1598 	 * Initialize reply list and start timer
1599 	 */
1600 	TAILQ_INIT(&nfs_reqq);
1601 	nfs_timer(NULL);
1602 	MOWNER_ATTACH(&nfs_mowner);
1603 
1604 #ifdef NFS
1605 	/* Initialize the kqueue structures */
1606 	nfs_kqinit();
1607 	/* Initialize the iod structures */
1608 	nfs_iodinit();
1609 #endif
1610 }
1611 
1612 void
1613 nfs_init(void)
1614 {
1615 	static ONCE_DECL(nfs_init_once);
1616 
1617 	RUN_ONCE(&nfs_init_once, nfs_init0);
1618 }
1619 
1620 #ifdef NFS
1621 /*
1622  * Called once at VFS init to initialize client-specific data structures.
1623  */
1624 void
1625 nfs_vfs_init()
1626 {
1627 	/* Initialize NFS server / client shared data. */
1628 	nfs_init();
1629 
1630 	nfs_nhinit();			/* Init the nfsnode table */
1631 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1632 }
1633 
1634 void
1635 nfs_vfs_reinit()
1636 {
1637 	nfs_nhreinit();
1638 }
1639 
1640 void
1641 nfs_vfs_done()
1642 {
1643 	nfs_nhdone();
1644 }
1645 
1646 /*
1647  * Attribute cache routines.
1648  * nfs_loadattrcache() - loads or updates the cache contents from attributes
1649  *	that are on the mbuf list
1650  * nfs_getattrcache() - returns valid attributes if found in cache, returns
1651  *	error otherwise
1652  */
1653 
1654 /*
1655  * Load the attribute cache (that lives in the nfsnode entry) with
1656  * the values on the mbuf list and
1657  * Iff vap not NULL
1658  *    copy the attributes to *vaper
1659  */
1660 int
1661 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags)
1662 	struct vnode **vpp;
1663 	struct mbuf **mdp;
1664 	caddr_t *dposp;
1665 	struct vattr *vaper;
1666 	int flags;
1667 {
1668 	int32_t t1;
1669 	caddr_t cp2;
1670 	int error = 0;
1671 	struct mbuf *md;
1672 	int v3 = NFS_ISV3(*vpp);
1673 
1674 	md = *mdp;
1675 	t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
1676 	error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2);
1677 	if (error)
1678 		return (error);
1679 	return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags);
1680 }
1681 
1682 int
1683 nfs_loadattrcache(vpp, fp, vaper, flags)
1684 	struct vnode **vpp;
1685 	struct nfs_fattr *fp;
1686 	struct vattr *vaper;
1687 	int flags;
1688 {
1689 	struct vnode *vp = *vpp;
1690 	struct vattr *vap;
1691 	int v3 = NFS_ISV3(vp);
1692 	enum vtype vtyp;
1693 	u_short vmode;
1694 	struct timespec mtime;
1695 	struct timespec ctime;
1696 	struct vnode *nvp;
1697 	int32_t rdev;
1698 	struct nfsnode *np;
1699 	extern int (**spec_nfsv2nodeop_p) __P((void *));
1700 	uid_t uid;
1701 	gid_t gid;
1702 
1703 	if (v3) {
1704 		vtyp = nfsv3tov_type(fp->fa_type);
1705 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1706 		rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1),
1707 			fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2));
1708 		fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
1709 		fxdr_nfsv3time(&fp->fa3_ctime, &ctime);
1710 	} else {
1711 		vtyp = nfsv2tov_type(fp->fa_type);
1712 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1713 		if (vtyp == VNON || vtyp == VREG)
1714 			vtyp = IFTOVT(vmode);
1715 		rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
1716 		fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
1717 		ctime.tv_sec = fxdr_unsigned(u_int32_t,
1718 		    fp->fa2_ctime.nfsv2_sec);
1719 		ctime.tv_nsec = 0;
1720 
1721 		/*
1722 		 * Really ugly NFSv2 kludge.
1723 		 */
1724 		if (vtyp == VCHR && rdev == 0xffffffff)
1725 			vtyp = VFIFO;
1726 	}
1727 
1728 	vmode &= ALLPERMS;
1729 
1730 	/*
1731 	 * If v_type == VNON it is a new node, so fill in the v_type,
1732 	 * n_mtime fields. Check to see if it represents a special
1733 	 * device, and if so, check for a possible alias. Once the
1734 	 * correct vnode has been obtained, fill in the rest of the
1735 	 * information.
1736 	 */
1737 	np = VTONFS(vp);
1738 	if (vp->v_type == VNON) {
1739 		vp->v_type = vtyp;
1740 		if (vp->v_type == VFIFO) {
1741 			extern int (**fifo_nfsv2nodeop_p) __P((void *));
1742 			vp->v_op = fifo_nfsv2nodeop_p;
1743 		} else if (vp->v_type == VREG) {
1744 			lockinit(&np->n_commitlock, PINOD, "nfsclock", 0, 0);
1745 		} else if (vp->v_type == VCHR || vp->v_type == VBLK) {
1746 			vp->v_op = spec_nfsv2nodeop_p;
1747 			nvp = checkalias(vp, (dev_t)rdev, vp->v_mount);
1748 			if (nvp) {
1749 				/*
1750 				 * Discard unneeded vnode, but save its nfsnode.
1751 				 * Since the nfsnode does not have a lock, its
1752 				 * vnode lock has to be carried over.
1753 				 */
1754 				/*
1755 				 * XXX is the old node sure to be locked here?
1756 				 */
1757 				KASSERT(lockstatus(&vp->v_lock) ==
1758 				    LK_EXCLUSIVE);
1759 				nvp->v_data = vp->v_data;
1760 				vp->v_data = NULL;
1761 				VOP_UNLOCK(vp, 0);
1762 				vp->v_op = spec_vnodeop_p;
1763 				vrele(vp);
1764 				vgone(vp);
1765 				lockmgr(&nvp->v_lock, LK_EXCLUSIVE,
1766 				    &nvp->v_interlock);
1767 				/*
1768 				 * Reinitialize aliased node.
1769 				 */
1770 				np->n_vnode = nvp;
1771 				*vpp = vp = nvp;
1772 			}
1773 		}
1774 		np->n_mtime = mtime;
1775 	}
1776 	uid = fxdr_unsigned(uid_t, fp->fa_uid);
1777 	gid = fxdr_unsigned(gid_t, fp->fa_gid);
1778 	vap = np->n_vattr;
1779 
1780 	/*
1781 	 * Invalidate access cache if uid, gid, mode or ctime changed.
1782 	 */
1783 	if (np->n_accstamp != -1 &&
1784 	    (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode
1785 	    || timespeccmp(&ctime, &vap->va_ctime, !=)))
1786 		np->n_accstamp = -1;
1787 
1788 	vap->va_type = vtyp;
1789 	vap->va_mode = vmode;
1790 	vap->va_rdev = (dev_t)rdev;
1791 	vap->va_mtime = mtime;
1792 	vap->va_ctime = ctime;
1793 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1794 	switch (vtyp) {
1795 	case VDIR:
1796 		vap->va_blocksize = NFS_DIRFRAGSIZ;
1797 		break;
1798 	case VBLK:
1799 		vap->va_blocksize = BLKDEV_IOSIZE;
1800 		break;
1801 	case VCHR:
1802 		vap->va_blocksize = MAXBSIZE;
1803 		break;
1804 	default:
1805 		vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize :
1806 		    fxdr_unsigned(int32_t, fp->fa2_blocksize);
1807 		break;
1808 	}
1809 	if (v3) {
1810 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1811 		vap->va_uid = uid;
1812 		vap->va_gid = gid;
1813 		vap->va_size = fxdr_hyper(&fp->fa3_size);
1814 		vap->va_bytes = fxdr_hyper(&fp->fa3_used);
1815 		vap->va_fileid = fxdr_hyper(&fp->fa3_fileid);
1816 		fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
1817 		vap->va_flags = 0;
1818 		vap->va_filerev = 0;
1819 	} else {
1820 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1821 		vap->va_uid = uid;
1822 		vap->va_gid = gid;
1823 		vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
1824 		vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks)
1825 		    * NFS_FABLKSIZE;
1826 		vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
1827 		fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
1828 		vap->va_flags = 0;
1829 		vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
1830 		vap->va_filerev = 0;
1831 	}
1832 	if (vap->va_size != np->n_size) {
1833 		if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) {
1834 			vap->va_size = np->n_size;
1835 		} else {
1836 			np->n_size = vap->va_size;
1837 			if (vap->va_type == VREG) {
1838 				/*
1839 				 * we can't free pages if NAC_NOTRUNC because
1840 				 * the pages can be owned by ourselves.
1841 				 */
1842 				if (flags & NAC_NOTRUNC) {
1843 					np->n_flag |= NTRUNCDELAYED;
1844 				} else {
1845 					simple_lock(&vp->v_interlock);
1846 					(void)VOP_PUTPAGES(vp, 0,
1847 					    0, PGO_SYNCIO | PGO_CLEANIT |
1848 					    PGO_FREE | PGO_ALLPAGES);
1849 					uvm_vnp_setsize(vp, np->n_size);
1850 				}
1851 			}
1852 		}
1853 	}
1854 	np->n_attrstamp = mono_time.tv_sec;
1855 	if (vaper != NULL) {
1856 		memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(*vap));
1857 		if (np->n_flag & NCHG) {
1858 			if (np->n_flag & NACC)
1859 				vaper->va_atime = np->n_atim;
1860 			if (np->n_flag & NUPD)
1861 				vaper->va_mtime = np->n_mtim;
1862 		}
1863 	}
1864 	return (0);
1865 }
1866 
1867 /*
1868  * Check the time stamp
1869  * If the cache is valid, copy contents to *vap and return 0
1870  * otherwise return an error
1871  */
1872 int
1873 nfs_getattrcache(vp, vaper)
1874 	struct vnode *vp;
1875 	struct vattr *vaper;
1876 {
1877 	struct nfsnode *np = VTONFS(vp);
1878 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1879 	struct vattr *vap;
1880 
1881 	if (np->n_attrstamp == 0 ||
1882 	    (mono_time.tv_sec - np->n_attrstamp) >= NFS_ATTRTIMEO(nmp, np)) {
1883 		nfsstats.attrcache_misses++;
1884 		return (ENOENT);
1885 	}
1886 	nfsstats.attrcache_hits++;
1887 	vap = np->n_vattr;
1888 	if (vap->va_size != np->n_size) {
1889 		if (vap->va_type == VREG) {
1890 			if (np->n_flag & NMODIFIED) {
1891 				if (vap->va_size < np->n_size)
1892 					vap->va_size = np->n_size;
1893 				else
1894 					np->n_size = vap->va_size;
1895 			} else
1896 				np->n_size = vap->va_size;
1897 			uvm_vnp_setsize(vp, np->n_size);
1898 		} else
1899 			np->n_size = vap->va_size;
1900 	}
1901 	memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(struct vattr));
1902 	if (np->n_flag & NCHG) {
1903 		if (np->n_flag & NACC)
1904 			vaper->va_atime = np->n_atim;
1905 		if (np->n_flag & NUPD)
1906 			vaper->va_mtime = np->n_mtim;
1907 	}
1908 	return (0);
1909 }
1910 
1911 void
1912 nfs_delayedtruncate(vp)
1913 	struct vnode *vp;
1914 {
1915 	struct nfsnode *np = VTONFS(vp);
1916 
1917 	if (np->n_flag & NTRUNCDELAYED) {
1918 		np->n_flag &= ~NTRUNCDELAYED;
1919 		simple_lock(&vp->v_interlock);
1920 		(void)VOP_PUTPAGES(vp, 0,
1921 		    0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES);
1922 		uvm_vnp_setsize(vp, np->n_size);
1923 	}
1924 }
1925 
1926 #define	NFS_WCCKLUDGE_TIMEOUT	(24 * 60 * 60)	/* 1 day */
1927 #define	NFS_WCCKLUDGE(nmp, now) \
1928 	(((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \
1929 	((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0)
1930 
1931 /*
1932  * nfs_check_wccdata: check inaccurate wcc_data
1933  *
1934  * => return non-zero if we shouldn't trust the wcc_data.
1935  * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed".
1936  */
1937 
1938 int
1939 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime,
1940     struct timespec *mtime, boolean_t docheck)
1941 {
1942 	int error = 0;
1943 
1944 #if !defined(NFS_V2_ONLY)
1945 
1946 	if (docheck) {
1947 		struct vnode *vp = NFSTOV(np);
1948 		struct nfsmount *nmp;
1949 		long now = mono_time.tv_sec;
1950 #if defined(DEBUG)
1951 		const char *reason = NULL; /* XXX: gcc */
1952 #endif
1953 
1954 		if (timespeccmp(&np->n_vattr->va_mtime, mtime, <=)) {
1955 #if defined(DEBUG)
1956 			reason = "mtime";
1957 #endif
1958 			error = EINVAL;
1959 		}
1960 
1961 		if (vp->v_type == VDIR &&
1962 		    timespeccmp(&np->n_vattr->va_ctime, ctime, <=)) {
1963 #if defined(DEBUG)
1964 			reason = "ctime";
1965 #endif
1966 			error = EINVAL;
1967 		}
1968 
1969 		nmp = VFSTONFS(vp->v_mount);
1970 		if (error) {
1971 
1972 			/*
1973 			 * despite of the fact that we've updated the file,
1974 			 * timestamps of the file were not updated as we
1975 			 * expected.
1976 			 * it means that the server has incompatible
1977 			 * semantics of timestamps or (more likely)
1978 			 * the server time is not precise enough to
1979 			 * track each modifications.
1980 			 * in that case, we disable wcc processing.
1981 			 *
1982 			 * yes, strictly speaking, we should disable all
1983 			 * caching.  it's a compromise.
1984 			 */
1985 
1986 			simple_lock(&nmp->nm_slock);
1987 #if defined(DEBUG)
1988 			if (!NFS_WCCKLUDGE(nmp, now)) {
1989 				printf("%s: inaccurate wcc data (%s) detected,"
1990 				    " disabling wcc\n",
1991 				    vp->v_mount->mnt_stat.f_mntfromname,
1992 				    reason);
1993 			}
1994 #endif
1995 			nmp->nm_iflag |= NFSMNT_WCCKLUDGE;
1996 			nmp->nm_wcckludgetime = now;
1997 			simple_unlock(&nmp->nm_slock);
1998 		} else if (NFS_WCCKLUDGE(nmp, now)) {
1999 			error = EPERM; /* XXX */
2000 		} else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
2001 			simple_lock(&nmp->nm_slock);
2002 			if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
2003 #if defined(DEBUG)
2004 				printf("%s: re-enabling wcc\n",
2005 				    vp->v_mount->mnt_stat.f_mntfromname);
2006 #endif
2007 				nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE;
2008 			}
2009 			simple_unlock(&nmp->nm_slock);
2010 		}
2011 	}
2012 
2013 #endif /* !defined(NFS_V2_ONLY) */
2014 
2015 	return error;
2016 }
2017 
2018 /*
2019  * Heuristic to see if the server XDR encodes directory cookies or not.
2020  * it is not supposed to, but a lot of servers may do this. Also, since
2021  * most/all servers will implement V2 as well, it is expected that they
2022  * may return just 32 bits worth of cookie information, so we need to
2023  * find out in which 32 bits this information is available. We do this
2024  * to avoid trouble with emulated binaries that can't handle 64 bit
2025  * directory offsets.
2026  */
2027 
2028 void
2029 nfs_cookieheuristic(vp, flagp, l, cred)
2030 	struct vnode *vp;
2031 	int *flagp;
2032 	struct lwp *l;
2033 	struct ucred *cred;
2034 {
2035 	struct uio auio;
2036 	struct iovec aiov;
2037 	caddr_t tbuf, cp;
2038 	struct dirent *dp;
2039 	off_t *cookies = NULL, *cop;
2040 	int error, eof, nc, len;
2041 
2042 	MALLOC(tbuf, caddr_t, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK);
2043 
2044 	aiov.iov_base = tbuf;
2045 	aiov.iov_len = NFS_DIRFRAGSIZ;
2046 	auio.uio_iov = &aiov;
2047 	auio.uio_iovcnt = 1;
2048 	auio.uio_rw = UIO_READ;
2049 	auio.uio_segflg = UIO_SYSSPACE;
2050 	auio.uio_lwp = NULL;
2051 	auio.uio_resid = NFS_DIRFRAGSIZ;
2052 	auio.uio_offset = 0;
2053 
2054 	error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc);
2055 
2056 	len = NFS_DIRFRAGSIZ - auio.uio_resid;
2057 	if (error || len == 0) {
2058 		FREE(tbuf, M_TEMP);
2059 		if (cookies)
2060 			free(cookies, M_TEMP);
2061 		return;
2062 	}
2063 
2064 	/*
2065 	 * Find the first valid entry and look at its offset cookie.
2066 	 */
2067 
2068 	cp = tbuf;
2069 	for (cop = cookies; len > 0; len -= dp->d_reclen) {
2070 		dp = (struct dirent *)cp;
2071 		if (dp->d_fileno != 0 && len >= dp->d_reclen) {
2072 			if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) {
2073 				*flagp |= NFSMNT_SWAPCOOKIE;
2074 				nfs_invaldircache(vp, 0);
2075 				nfs_vinvalbuf(vp, 0, cred, l, 1);
2076 			}
2077 			break;
2078 		}
2079 		cop++;
2080 		cp += dp->d_reclen;
2081 	}
2082 
2083 	FREE(tbuf, M_TEMP);
2084 	free(cookies, M_TEMP);
2085 }
2086 #endif /* NFS */
2087 
2088 #ifdef NFSSERVER
2089 /*
2090  * Set up nameidata for a lookup() call and do it.
2091  *
2092  * If pubflag is set, this call is done for a lookup operation on the
2093  * public filehandle. In that case we allow crossing mountpoints and
2094  * absolute pathnames. However, the caller is expected to check that
2095  * the lookup result is within the public fs, and deny access if
2096  * it is not.
2097  */
2098 int
2099 nfs_namei(ndp, fhp, len, slp, nam, mdp, dposp, retdirp, l, kerbflag, pubflag)
2100 	struct nameidata *ndp;
2101 	fhandle_t *fhp;
2102 	uint32_t len;
2103 	struct nfssvc_sock *slp;
2104 	struct mbuf *nam;
2105 	struct mbuf **mdp;
2106 	caddr_t *dposp;
2107 	struct vnode **retdirp;
2108 	struct lwp *l;
2109 	int kerbflag, pubflag;
2110 {
2111 	int i, rem;
2112 	struct mbuf *md;
2113 	char *fromcp, *tocp, *cp;
2114 	struct iovec aiov;
2115 	struct uio auio;
2116 	struct vnode *dp;
2117 	int error, rdonly, linklen;
2118 	struct componentname *cnp = &ndp->ni_cnd;
2119 
2120 	*retdirp = (struct vnode *)0;
2121 
2122 	if ((len + 1) > MAXPATHLEN)
2123 		return (ENAMETOOLONG);
2124 	if (len == 0)
2125 		return (EACCES);
2126 	cnp->cn_pnbuf = PNBUF_GET();
2127 
2128 	/*
2129 	 * Copy the name from the mbuf list to ndp->ni_pnbuf
2130 	 * and set the various ndp fields appropriately.
2131 	 */
2132 	fromcp = *dposp;
2133 	tocp = cnp->cn_pnbuf;
2134 	md = *mdp;
2135 	rem = mtod(md, caddr_t) + md->m_len - fromcp;
2136 	for (i = 0; i < len; i++) {
2137 		while (rem == 0) {
2138 			md = md->m_next;
2139 			if (md == NULL) {
2140 				error = EBADRPC;
2141 				goto out;
2142 			}
2143 			fromcp = mtod(md, caddr_t);
2144 			rem = md->m_len;
2145 		}
2146 		if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
2147 			error = EACCES;
2148 			goto out;
2149 		}
2150 		*tocp++ = *fromcp++;
2151 		rem--;
2152 	}
2153 	*tocp = '\0';
2154 	*mdp = md;
2155 	*dposp = fromcp;
2156 	len = nfsm_rndup(len)-len;
2157 	if (len > 0) {
2158 		if (rem >= len)
2159 			*dposp += len;
2160 		else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
2161 			goto out;
2162 	}
2163 
2164 	/*
2165 	 * Extract and set starting directory.
2166 	 */
2167 	error = nfsrv_fhtovp(fhp, FALSE, &dp, ndp->ni_cnd.cn_cred, slp,
2168 	    nam, &rdonly, kerbflag, pubflag);
2169 	if (error)
2170 		goto out;
2171 	if (dp->v_type != VDIR) {
2172 		vrele(dp);
2173 		error = ENOTDIR;
2174 		goto out;
2175 	}
2176 
2177 	if (rdonly)
2178 		cnp->cn_flags |= RDONLY;
2179 
2180 	*retdirp = dp;
2181 
2182 	if (pubflag) {
2183 		/*
2184 		 * Oh joy. For WebNFS, handle those pesky '%' escapes,
2185 		 * and the 'native path' indicator.
2186 		 */
2187 		cp = PNBUF_GET();
2188 		fromcp = cnp->cn_pnbuf;
2189 		tocp = cp;
2190 		if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
2191 			switch ((unsigned char)*fromcp) {
2192 			case WEBNFS_NATIVE_CHAR:
2193 				/*
2194 				 * 'Native' path for us is the same
2195 				 * as a path according to the NFS spec,
2196 				 * just skip the escape char.
2197 				 */
2198 				fromcp++;
2199 				break;
2200 			/*
2201 			 * More may be added in the future, range 0x80-0xff
2202 			 */
2203 			default:
2204 				error = EIO;
2205 				PNBUF_PUT(cp);
2206 				goto out;
2207 			}
2208 		}
2209 		/*
2210 		 * Translate the '%' escapes, URL-style.
2211 		 */
2212 		while (*fromcp != '\0') {
2213 			if (*fromcp == WEBNFS_ESC_CHAR) {
2214 				if (fromcp[1] != '\0' && fromcp[2] != '\0') {
2215 					fromcp++;
2216 					*tocp++ = HEXSTRTOI(fromcp);
2217 					fromcp += 2;
2218 					continue;
2219 				} else {
2220 					error = ENOENT;
2221 					PNBUF_PUT(cp);
2222 					goto out;
2223 				}
2224 			} else
2225 				*tocp++ = *fromcp++;
2226 		}
2227 		*tocp = '\0';
2228 		PNBUF_PUT(cnp->cn_pnbuf);
2229 		cnp->cn_pnbuf = cp;
2230 	}
2231 
2232 	ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1;
2233 	ndp->ni_segflg = UIO_SYSSPACE;
2234 	ndp->ni_rootdir = rootvnode;
2235 
2236 	if (pubflag) {
2237 		ndp->ni_loopcnt = 0;
2238 		if (cnp->cn_pnbuf[0] == '/')
2239 			dp = rootvnode;
2240 	} else {
2241 		cnp->cn_flags |= NOCROSSMOUNT;
2242 	}
2243 
2244 	cnp->cn_lwp = l;
2245 	VREF(dp);
2246 
2247     for (;;) {
2248 	cnp->cn_nameptr = cnp->cn_pnbuf;
2249 	ndp->ni_startdir = dp;
2250 	/*
2251 	 * And call lookup() to do the real work
2252 	 */
2253 	error = lookup(ndp);
2254 	if (error) {
2255 		PNBUF_PUT(cnp->cn_pnbuf);
2256 		return (error);
2257 	}
2258 	/*
2259 	 * Check for encountering a symbolic link
2260 	 */
2261 	if ((cnp->cn_flags & ISSYMLINK) == 0) {
2262 		if (cnp->cn_flags & (SAVENAME | SAVESTART))
2263 			cnp->cn_flags |= HASBUF;
2264 		else
2265 			PNBUF_PUT(cnp->cn_pnbuf);
2266 		return (0);
2267 	} else {
2268 		if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
2269 			VOP_UNLOCK(ndp->ni_dvp, 0);
2270 		if (!pubflag) {
2271 			error = EINVAL;
2272 			break;
2273 		}
2274 
2275 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
2276 			error = ELOOP;
2277 			break;
2278 		}
2279 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
2280 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred,
2281 			    cnp->cn_lwp);
2282 			if (error != 0)
2283 				break;
2284 		}
2285 		if (ndp->ni_pathlen > 1)
2286 			cp = PNBUF_GET();
2287 		else
2288 			cp = cnp->cn_pnbuf;
2289 		aiov.iov_base = cp;
2290 		aiov.iov_len = MAXPATHLEN;
2291 		auio.uio_iov = &aiov;
2292 		auio.uio_iovcnt = 1;
2293 		auio.uio_offset = 0;
2294 		auio.uio_rw = UIO_READ;
2295 		auio.uio_segflg = UIO_SYSSPACE;
2296 		auio.uio_lwp = NULL;
2297 		auio.uio_resid = MAXPATHLEN;
2298 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
2299 		if (error) {
2300 		badlink:
2301 			if (ndp->ni_pathlen > 1)
2302 				PNBUF_PUT(cp);
2303 			break;
2304 		}
2305 		linklen = MAXPATHLEN - auio.uio_resid;
2306 		if (linklen == 0) {
2307 			error = ENOENT;
2308 			goto badlink;
2309 		}
2310 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
2311 			error = ENAMETOOLONG;
2312 			goto badlink;
2313 		}
2314 		if (ndp->ni_pathlen > 1) {
2315 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
2316 			PNBUF_PUT(cnp->cn_pnbuf);
2317 			cnp->cn_pnbuf = cp;
2318 		} else
2319 			cnp->cn_pnbuf[linklen] = '\0';
2320 		ndp->ni_pathlen += linklen;
2321 		vput(ndp->ni_vp);
2322 		dp = ndp->ni_dvp;
2323 		/*
2324 		 * Check if root directory should replace current directory.
2325 		 */
2326 		if (cnp->cn_pnbuf[0] == '/') {
2327 			vrele(dp);
2328 			dp = ndp->ni_rootdir;
2329 			VREF(dp);
2330 		}
2331 	}
2332    }
2333 	vrele(ndp->ni_dvp);
2334 	vput(ndp->ni_vp);
2335 	ndp->ni_vp = NULL;
2336 out:
2337 	PNBUF_PUT(cnp->cn_pnbuf);
2338 	return (error);
2339 }
2340 #endif /* NFSSERVER */
2341 
2342 /*
2343  * A fiddled version of m_adj() that ensures null fill to a 32-bit
2344  * boundary and only trims off the back end
2345  *
2346  * 1. trim off 'len' bytes as m_adj(mp, -len).
2347  * 2. add zero-padding 'nul' bytes at the end of the mbuf chain.
2348  */
2349 void
2350 nfs_zeropad(mp, len, nul)
2351 	struct mbuf *mp;
2352 	int len;
2353 	int nul;
2354 {
2355 	struct mbuf *m;
2356 	int count;
2357 
2358 	/*
2359 	 * Trim from tail.  Scan the mbuf chain,
2360 	 * calculating its length and finding the last mbuf.
2361 	 * If the adjustment only affects this mbuf, then just
2362 	 * adjust and return.  Otherwise, rescan and truncate
2363 	 * after the remaining size.
2364 	 */
2365 	count = 0;
2366 	m = mp;
2367 	for (;;) {
2368 		count += m->m_len;
2369 		if (m->m_next == NULL)
2370 			break;
2371 		m = m->m_next;
2372 	}
2373 
2374 	KDASSERT(count >= len);
2375 
2376 	if (m->m_len >= len) {
2377 		m->m_len -= len;
2378 	} else {
2379 		count -= len;
2380 		/*
2381 		 * Correct length for chain is "count".
2382 		 * Find the mbuf with last data, adjust its length,
2383 		 * and toss data from remaining mbufs on chain.
2384 		 */
2385 		for (m = mp; m; m = m->m_next) {
2386 			if (m->m_len >= count) {
2387 				m->m_len = count;
2388 				break;
2389 			}
2390 			count -= m->m_len;
2391 		}
2392 		m_freem(m->m_next);
2393 		m->m_next = NULL;
2394 	}
2395 
2396 	KDASSERT(m->m_next == NULL);
2397 
2398 	/*
2399 	 * zero-padding.
2400 	 */
2401 	if (nul > 0) {
2402 		char *cp;
2403 		int i;
2404 
2405 		if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) {
2406 			struct mbuf *n;
2407 
2408 			KDASSERT(MLEN >= nul);
2409 			n = m_get(M_WAIT, MT_DATA);
2410 			MCLAIM(n, &nfs_mowner);
2411 			n->m_len = nul;
2412 			n->m_next = NULL;
2413 			m->m_next = n;
2414 			cp = mtod(n, caddr_t);
2415 		} else {
2416 			cp = mtod(m, caddr_t) + m->m_len;
2417 			m->m_len += nul;
2418 		}
2419 		for (i = 0; i < nul; i++)
2420 			*cp++ = '\0';
2421 	}
2422 	return;
2423 }
2424 
2425 /*
2426  * Make these functions instead of macros, so that the kernel text size
2427  * doesn't get too big...
2428  */
2429 void
2430 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp)
2431 	struct nfsrv_descript *nfsd;
2432 	int before_ret;
2433 	struct vattr *before_vap;
2434 	int after_ret;
2435 	struct vattr *after_vap;
2436 	struct mbuf **mbp;
2437 	char **bposp;
2438 {
2439 	struct mbuf *mb = *mbp;
2440 	char *bpos = *bposp;
2441 	u_int32_t *tl;
2442 
2443 	if (before_ret) {
2444 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2445 		*tl = nfs_false;
2446 	} else {
2447 		nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
2448 		*tl++ = nfs_true;
2449 		txdr_hyper(before_vap->va_size, tl);
2450 		tl += 2;
2451 		txdr_nfsv3time(&(before_vap->va_mtime), tl);
2452 		tl += 2;
2453 		txdr_nfsv3time(&(before_vap->va_ctime), tl);
2454 	}
2455 	*bposp = bpos;
2456 	*mbp = mb;
2457 	nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
2458 }
2459 
2460 void
2461 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp)
2462 	struct nfsrv_descript *nfsd;
2463 	int after_ret;
2464 	struct vattr *after_vap;
2465 	struct mbuf **mbp;
2466 	char **bposp;
2467 {
2468 	struct mbuf *mb = *mbp;
2469 	char *bpos = *bposp;
2470 	u_int32_t *tl;
2471 	struct nfs_fattr *fp;
2472 
2473 	if (after_ret) {
2474 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2475 		*tl = nfs_false;
2476 	} else {
2477 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
2478 		*tl++ = nfs_true;
2479 		fp = (struct nfs_fattr *)tl;
2480 		nfsm_srvfattr(nfsd, after_vap, fp);
2481 	}
2482 	*mbp = mb;
2483 	*bposp = bpos;
2484 }
2485 
2486 void
2487 nfsm_srvfattr(nfsd, vap, fp)
2488 	struct nfsrv_descript *nfsd;
2489 	struct vattr *vap;
2490 	struct nfs_fattr *fp;
2491 {
2492 
2493 	fp->fa_nlink = txdr_unsigned(vap->va_nlink);
2494 	fp->fa_uid = txdr_unsigned(vap->va_uid);
2495 	fp->fa_gid = txdr_unsigned(vap->va_gid);
2496 	if (nfsd->nd_flag & ND_NFSV3) {
2497 		fp->fa_type = vtonfsv3_type(vap->va_type);
2498 		fp->fa_mode = vtonfsv3_mode(vap->va_mode);
2499 		txdr_hyper(vap->va_size, &fp->fa3_size);
2500 		txdr_hyper(vap->va_bytes, &fp->fa3_used);
2501 		fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
2502 		fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
2503 		fp->fa3_fsid.nfsuquad[0] = 0;
2504 		fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
2505 		txdr_hyper(vap->va_fileid, &fp->fa3_fileid);
2506 		txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
2507 		txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
2508 		txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
2509 	} else {
2510 		fp->fa_type = vtonfsv2_type(vap->va_type);
2511 		fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
2512 		fp->fa2_size = txdr_unsigned(vap->va_size);
2513 		fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
2514 		if (vap->va_type == VFIFO)
2515 			fp->fa2_rdev = 0xffffffff;
2516 		else
2517 			fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
2518 		fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
2519 		fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
2520 		fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
2521 		txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
2522 		txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
2523 		txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
2524 	}
2525 }
2526 
2527 #ifdef NFSSERVER
2528 /*
2529  * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
2530  * 	- look up fsid in mount list (if not found ret error)
2531  *	- get vp and export rights by calling VFS_FHTOVP()
2532  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2533  *	- if not lockflag unlock it with VOP_UNLOCK()
2534  */
2535 int
2536 nfsrv_fhtovp(fhp, lockflag, vpp, cred, slp, nam, rdonlyp, kerbflag, pubflag)
2537 	fhandle_t *fhp;
2538 	int lockflag;
2539 	struct vnode **vpp;
2540 	struct ucred *cred;
2541 	struct nfssvc_sock *slp;
2542 	struct mbuf *nam;
2543 	int *rdonlyp;
2544 	int kerbflag;
2545 {
2546 	struct mount *mp;
2547 	int i;
2548 	struct ucred *credanon;
2549 	int error, exflags;
2550 	struct sockaddr_in *saddr;
2551 
2552 	*vpp = (struct vnode *)0;
2553 
2554 	if (nfs_ispublicfh(fhp)) {
2555 		if (!pubflag || !nfs_pub.np_valid)
2556 			return (ESTALE);
2557 		fhp = &nfs_pub.np_handle;
2558 	}
2559 
2560 	error = netexport_check(&fhp->fh_fsid, nam, &mp, &exflags, &credanon);
2561 	if (error) {
2562 		return error;
2563 	}
2564 
2565 	error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
2566 	if (error)
2567 		return (error);
2568 
2569 	if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
2570 		saddr = mtod(nam, struct sockaddr_in *);
2571 		if ((saddr->sin_family == AF_INET) &&
2572 		    ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
2573 			vput(*vpp);
2574 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2575 		}
2576 #ifdef INET6
2577 		if ((saddr->sin_family == AF_INET6) &&
2578 		    ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) {
2579 			vput(*vpp);
2580 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2581 		}
2582 #endif
2583 	}
2584 	/*
2585 	 * Check/setup credentials.
2586 	 */
2587 	if (exflags & MNT_EXKERB) {
2588 		if (!kerbflag) {
2589 			vput(*vpp);
2590 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2591 		}
2592 	} else if (kerbflag) {
2593 		vput(*vpp);
2594 		return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2595 	} else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
2596 		cred->cr_uid = credanon->cr_uid;
2597 		cred->cr_gid = credanon->cr_gid;
2598 		for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
2599 			cred->cr_groups[i] = credanon->cr_groups[i];
2600 		cred->cr_ngroups = i;
2601 	}
2602 	if (exflags & MNT_EXRDONLY)
2603 		*rdonlyp = 1;
2604 	else
2605 		*rdonlyp = 0;
2606 	if (!lockflag)
2607 		VOP_UNLOCK(*vpp, 0);
2608 	return (0);
2609 }
2610 
2611 /*
2612  * WebNFS: check if a filehandle is a public filehandle. For v3, this
2613  * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
2614  * transformed this to all zeroes in both cases, so check for it.
2615  */
2616 int
2617 nfs_ispublicfh(fhp)
2618 	fhandle_t *fhp;
2619 {
2620 	char *cp = (char *)fhp;
2621 	int i;
2622 
2623 	for (i = 0; i < NFSX_V3FH; i++)
2624 		if (*cp++ != 0)
2625 			return (FALSE);
2626 	return (TRUE);
2627 }
2628 #endif /* NFSSERVER */
2629 
2630 /*
2631  * This function compares two net addresses by family and returns TRUE
2632  * if they are the same host.
2633  * If there is any doubt, return FALSE.
2634  * The AF_INET family is handled as a special case so that address mbufs
2635  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
2636  */
2637 int
2638 netaddr_match(family, haddr, nam)
2639 	int family;
2640 	union nethostaddr *haddr;
2641 	struct mbuf *nam;
2642 {
2643 	struct sockaddr_in *inetaddr;
2644 
2645 	switch (family) {
2646 	case AF_INET:
2647 		inetaddr = mtod(nam, struct sockaddr_in *);
2648 		if (inetaddr->sin_family == AF_INET &&
2649 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
2650 			return (1);
2651 		break;
2652 #ifdef INET6
2653 	case AF_INET6:
2654 	    {
2655 		struct sockaddr_in6 *sin6_1, *sin6_2;
2656 
2657 		sin6_1 = mtod(nam, struct sockaddr_in6 *);
2658 		sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
2659 		if (sin6_1->sin6_family == AF_INET6 &&
2660 		    IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
2661 			return 1;
2662 	    }
2663 #endif
2664 #ifdef ISO
2665 	case AF_ISO:
2666 	    {
2667 		struct sockaddr_iso *isoaddr1, *isoaddr2;
2668 
2669 		isoaddr1 = mtod(nam, struct sockaddr_iso *);
2670 		isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *);
2671 		if (isoaddr1->siso_family == AF_ISO &&
2672 		    isoaddr1->siso_nlen > 0 &&
2673 		    isoaddr1->siso_nlen == isoaddr2->siso_nlen &&
2674 		    SAME_ISOADDR(isoaddr1, isoaddr2))
2675 			return (1);
2676 		break;
2677 	    }
2678 #endif	/* ISO */
2679 	default:
2680 		break;
2681 	};
2682 	return (0);
2683 }
2684 
2685 /*
2686  * The write verifier has changed (probably due to a server reboot), so all
2687  * PG_NEEDCOMMIT pages will have to be written again. Since they are marked
2688  * as dirty or are being written out just now, all this takes is clearing
2689  * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for
2690  * the mount point.
2691  */
2692 void
2693 nfs_clearcommit(mp)
2694 	struct mount *mp;
2695 {
2696 	struct vnode *vp;
2697 	struct nfsnode *np;
2698 	struct vm_page *pg;
2699 	struct nfsmount *nmp = VFSTONFS(mp);
2700 
2701 	lockmgr(&nmp->nm_writeverflock, LK_EXCLUSIVE, NULL);
2702 
2703 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2704 		KASSERT(vp->v_mount == mp);
2705 		if (vp->v_type != VREG)
2706 			continue;
2707 		np = VTONFS(vp);
2708 		np->n_pushlo = np->n_pushhi = np->n_pushedlo =
2709 		    np->n_pushedhi = 0;
2710 		np->n_commitflags &=
2711 		    ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
2712 		simple_lock(&vp->v_uobj.vmobjlock);
2713 		TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) {
2714 			pg->flags &= ~PG_NEEDCOMMIT;
2715 		}
2716 		simple_unlock(&vp->v_uobj.vmobjlock);
2717 	}
2718 	simple_lock(&nmp->nm_slock);
2719 	nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF;
2720 	simple_unlock(&nmp->nm_slock);
2721 	lockmgr(&nmp->nm_writeverflock, LK_RELEASE, NULL);
2722 }
2723 
2724 void
2725 nfs_merge_commit_ranges(vp)
2726 	struct vnode *vp;
2727 {
2728 	struct nfsnode *np = VTONFS(vp);
2729 
2730 	KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID);
2731 
2732 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2733 		np->n_pushedlo = np->n_pushlo;
2734 		np->n_pushedhi = np->n_pushhi;
2735 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2736 	} else {
2737 		if (np->n_pushlo < np->n_pushedlo)
2738 			np->n_pushedlo = np->n_pushlo;
2739 		if (np->n_pushhi > np->n_pushedhi)
2740 			np->n_pushedhi = np->n_pushhi;
2741 	}
2742 
2743 	np->n_pushlo = np->n_pushhi = 0;
2744 	np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
2745 
2746 #ifdef NFS_DEBUG_COMMIT
2747 	printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2748 	    (unsigned)np->n_pushedhi);
2749 #endif
2750 }
2751 
2752 int
2753 nfs_in_committed_range(vp, off, len)
2754 	struct vnode *vp;
2755 	off_t off, len;
2756 {
2757 	struct nfsnode *np = VTONFS(vp);
2758 	off_t lo, hi;
2759 
2760 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2761 		return 0;
2762 	lo = off;
2763 	hi = lo + len;
2764 
2765 	return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
2766 }
2767 
2768 int
2769 nfs_in_tobecommitted_range(vp, off, len)
2770 	struct vnode *vp;
2771 	off_t off, len;
2772 {
2773 	struct nfsnode *np = VTONFS(vp);
2774 	off_t lo, hi;
2775 
2776 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2777 		return 0;
2778 	lo = off;
2779 	hi = lo + len;
2780 
2781 	return (lo >= np->n_pushlo && hi <= np->n_pushhi);
2782 }
2783 
2784 void
2785 nfs_add_committed_range(vp, off, len)
2786 	struct vnode *vp;
2787 	off_t off, len;
2788 {
2789 	struct nfsnode *np = VTONFS(vp);
2790 	off_t lo, hi;
2791 
2792 	lo = off;
2793 	hi = lo + len;
2794 
2795 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2796 		np->n_pushedlo = lo;
2797 		np->n_pushedhi = hi;
2798 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2799 	} else {
2800 		if (hi > np->n_pushedhi)
2801 			np->n_pushedhi = hi;
2802 		if (lo < np->n_pushedlo)
2803 			np->n_pushedlo = lo;
2804 	}
2805 #ifdef NFS_DEBUG_COMMIT
2806 	printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2807 	    (unsigned)np->n_pushedhi);
2808 #endif
2809 }
2810 
2811 void
2812 nfs_del_committed_range(vp, off, len)
2813 	struct vnode *vp;
2814 	off_t off, len;
2815 {
2816 	struct nfsnode *np = VTONFS(vp);
2817 	off_t lo, hi;
2818 
2819 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2820 		return;
2821 
2822 	lo = off;
2823 	hi = lo + len;
2824 
2825 	if (lo > np->n_pushedhi || hi < np->n_pushedlo)
2826 		return;
2827 	if (lo <= np->n_pushedlo)
2828 		np->n_pushedlo = hi;
2829 	else if (hi >= np->n_pushedhi)
2830 		np->n_pushedhi = lo;
2831 	else {
2832 		/*
2833 		 * XXX There's only one range. If the deleted range
2834 		 * is in the middle, pick the largest of the
2835 		 * contiguous ranges that it leaves.
2836 		 */
2837 		if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
2838 			np->n_pushedhi = lo;
2839 		else
2840 			np->n_pushedlo = hi;
2841 	}
2842 #ifdef NFS_DEBUG_COMMIT
2843 	printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2844 	    (unsigned)np->n_pushedhi);
2845 #endif
2846 }
2847 
2848 void
2849 nfs_add_tobecommitted_range(vp, off, len)
2850 	struct vnode *vp;
2851 	off_t off, len;
2852 {
2853 	struct nfsnode *np = VTONFS(vp);
2854 	off_t lo, hi;
2855 
2856 	lo = off;
2857 	hi = lo + len;
2858 
2859 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
2860 		np->n_pushlo = lo;
2861 		np->n_pushhi = hi;
2862 		np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
2863 	} else {
2864 		if (lo < np->n_pushlo)
2865 			np->n_pushlo = lo;
2866 		if (hi > np->n_pushhi)
2867 			np->n_pushhi = hi;
2868 	}
2869 #ifdef NFS_DEBUG_COMMIT
2870 	printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2871 	    (unsigned)np->n_pushhi);
2872 #endif
2873 }
2874 
2875 void
2876 nfs_del_tobecommitted_range(vp, off, len)
2877 	struct vnode *vp;
2878 	off_t off, len;
2879 {
2880 	struct nfsnode *np = VTONFS(vp);
2881 	off_t lo, hi;
2882 
2883 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2884 		return;
2885 
2886 	lo = off;
2887 	hi = lo + len;
2888 
2889 	if (lo > np->n_pushhi || hi < np->n_pushlo)
2890 		return;
2891 
2892 	if (lo <= np->n_pushlo)
2893 		np->n_pushlo = hi;
2894 	else if (hi >= np->n_pushhi)
2895 		np->n_pushhi = lo;
2896 	else {
2897 		/*
2898 		 * XXX There's only one range. If the deleted range
2899 		 * is in the middle, pick the largest of the
2900 		 * contiguous ranges that it leaves.
2901 		 */
2902 		if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
2903 			np->n_pushhi = lo;
2904 		else
2905 			np->n_pushlo = hi;
2906 	}
2907 #ifdef NFS_DEBUG_COMMIT
2908 	printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2909 	    (unsigned)np->n_pushhi);
2910 #endif
2911 }
2912 
2913 /*
2914  * Map errnos to NFS error numbers. For Version 3 also filter out error
2915  * numbers not specified for the associated procedure.
2916  */
2917 int
2918 nfsrv_errmap(nd, err)
2919 	struct nfsrv_descript *nd;
2920 	int err;
2921 {
2922 	const short *defaulterrp, *errp;
2923 
2924 	if (nd->nd_flag & ND_NFSV3) {
2925 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
2926 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2927 		while (*++errp) {
2928 			if (*errp == err)
2929 				return (err);
2930 			else if (*errp > err)
2931 				break;
2932 		}
2933 		return ((int)*defaulterrp);
2934 	    } else
2935 		return (err & 0xffff);
2936 	}
2937 	if (err <= ELAST)
2938 		return ((int)nfsrv_v2errmap[err - 1]);
2939 	return (NFSERR_IO);
2940 }
2941 
2942 /*
2943  * Sort the group list in increasing numerical order.
2944  * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
2945  *  that used to be here.)
2946  */
2947 void
2948 nfsrvw_sort(list, num)
2949         gid_t *list;
2950         int num;
2951 {
2952 	int i, j;
2953 	gid_t v;
2954 
2955 	/* Insertion sort. */
2956 	for (i = 1; i < num; i++) {
2957 		v = list[i];
2958 		/* find correct slot for value v, moving others up */
2959 		for (j = i; --j >= 0 && v < list[j];)
2960 			list[j + 1] = list[j];
2961 		list[j + 1] = v;
2962 	}
2963 }
2964 
2965 /*
2966  * copy credentials making sure that the result can be compared with memcmp().
2967  */
2968 void
2969 nfsrv_setcred(incred, outcred)
2970 	struct ucred *incred, *outcred;
2971 {
2972 	int i;
2973 
2974 	memset((caddr_t)outcred, 0, sizeof (struct ucred));
2975 	outcred->cr_ref = 1;
2976 	outcred->cr_uid = incred->cr_uid;
2977 	outcred->cr_gid = incred->cr_gid;
2978 	outcred->cr_ngroups = incred->cr_ngroups;
2979 	for (i = 0; i < incred->cr_ngroups; i++)
2980 		outcred->cr_groups[i] = incred->cr_groups[i];
2981 	nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
2982 }
2983 
2984 u_int32_t
2985 nfs_getxid()
2986 {
2987 	static u_int32_t base;
2988 	static u_int32_t nfs_xid = 0;
2989 	static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER;
2990 	u_int32_t newxid;
2991 
2992 	simple_lock(&nfs_xidlock);
2993 	/*
2994 	 * derive initial xid from system time
2995 	 * XXX time is invalid if root not yet mounted
2996 	 */
2997 	if (__predict_false(!base && (rootvp))) {
2998 		struct timeval tv;
2999 
3000 		microtime(&tv);
3001 		base = tv.tv_sec << 12;
3002 		nfs_xid = base;
3003 	}
3004 
3005 	/*
3006 	 * Skip zero xid if it should ever happen.
3007 	 */
3008 	if (__predict_false(++nfs_xid == 0))
3009 		nfs_xid++;
3010 	newxid = nfs_xid;
3011 	simple_unlock(&nfs_xidlock);
3012 
3013 	return txdr_unsigned(newxid);
3014 }
3015 
3016 /*
3017  * assign a new xid for existing request.
3018  * used for NFSERR_JUKEBOX handling.
3019  */
3020 void
3021 nfs_renewxid(struct nfsreq *req)
3022 {
3023 	u_int32_t xid;
3024 	int off;
3025 
3026 	xid = nfs_getxid();
3027 	if (req->r_nmp->nm_sotype == SOCK_STREAM)
3028 		off = sizeof(u_int32_t); /* RPC record mark */
3029 	else
3030 		off = 0;
3031 
3032 	m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid);
3033 	req->r_xid = xid;
3034 }
3035