xref: /netbsd-src/sys/nfs/nfs_subs.c (revision cac8e449158efc7261bebc8657cbb0125a2cfdde)
1 /*	$NetBSD: nfs_subs.c,v 1.205 2008/07/15 16:06:58 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 1.205 2008/07/15 16:06:58 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/kmem.h>
91 #include <sys/mount.h>
92 #include <sys/vnode.h>
93 #include <sys/namei.h>
94 #include <sys/mbuf.h>
95 #include <sys/socket.h>
96 #include <sys/stat.h>
97 #include <sys/filedesc.h>
98 #include <sys/time.h>
99 #include <sys/dirent.h>
100 #include <sys/once.h>
101 #include <sys/kauth.h>
102 
103 #include <uvm/uvm_extern.h>
104 
105 #include <nfs/rpcv2.h>
106 #include <nfs/nfsproto.h>
107 #include <nfs/nfsnode.h>
108 #include <nfs/nfs.h>
109 #include <nfs/xdr_subs.h>
110 #include <nfs/nfsm_subs.h>
111 #include <nfs/nfsmount.h>
112 #include <nfs/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, 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 };
177 
178 /*
179  * and the reverse mapping from generic to Version 2 procedure numbers
180  */
181 const int nfsv2_procid[NFS_NPROCS] = {
182 	NFSV2PROC_NULL,
183 	NFSV2PROC_GETATTR,
184 	NFSV2PROC_SETATTR,
185 	NFSV2PROC_LOOKUP,
186 	NFSV2PROC_NOOP,
187 	NFSV2PROC_READLINK,
188 	NFSV2PROC_READ,
189 	NFSV2PROC_WRITE,
190 	NFSV2PROC_CREATE,
191 	NFSV2PROC_MKDIR,
192 	NFSV2PROC_SYMLINK,
193 	NFSV2PROC_CREATE,
194 	NFSV2PROC_REMOVE,
195 	NFSV2PROC_RMDIR,
196 	NFSV2PROC_RENAME,
197 	NFSV2PROC_LINK,
198 	NFSV2PROC_READDIR,
199 	NFSV2PROC_NOOP,
200 	NFSV2PROC_STATFS,
201 	NFSV2PROC_NOOP,
202 	NFSV2PROC_NOOP,
203 	NFSV2PROC_NOOP,
204 	NFSV2PROC_NOOP,
205 };
206 
207 /*
208  * Maps errno values to nfs error numbers.
209  * Use NFSERR_IO as the catch all for ones not specifically defined in
210  * RFC 1094.
211  */
212 static const u_char nfsrv_v2errmap[ELAST] = {
213   NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
214   NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
215   NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
216   NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
217   NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
218   NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
219   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
220   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
221   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
222   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
223   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
224   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
225   NFSERR_IO,	NFSERR_IO,	NFSERR_NAMETOL,	NFSERR_IO,	NFSERR_IO,
226   NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
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,
230 };
231 
232 /*
233  * Maps errno values to nfs error numbers.
234  * Although it is not obvious whether or not NFS clients really care if
235  * a returned error value is in the specified list for the procedure, the
236  * safest thing to do is filter them appropriately. For Version 2, the
237  * X/Open XNFS document is the only specification that defines error values
238  * for each RPC (The RFC simply lists all possible error values for all RPCs),
239  * so I have decided to not do this for Version 2.
240  * The first entry is the default error return and the rest are the valid
241  * errors for that RPC in increasing numeric order.
242  */
243 static const short nfsv3err_null[] = {
244 	0,
245 	0,
246 };
247 
248 static const short nfsv3err_getattr[] = {
249 	NFSERR_IO,
250 	NFSERR_IO,
251 	NFSERR_STALE,
252 	NFSERR_BADHANDLE,
253 	NFSERR_SERVERFAULT,
254 	0,
255 };
256 
257 static const short nfsv3err_setattr[] = {
258 	NFSERR_IO,
259 	NFSERR_PERM,
260 	NFSERR_IO,
261 	NFSERR_ACCES,
262 	NFSERR_INVAL,
263 	NFSERR_NOSPC,
264 	NFSERR_ROFS,
265 	NFSERR_DQUOT,
266 	NFSERR_STALE,
267 	NFSERR_BADHANDLE,
268 	NFSERR_NOT_SYNC,
269 	NFSERR_SERVERFAULT,
270 	0,
271 };
272 
273 static const short nfsv3err_lookup[] = {
274 	NFSERR_IO,
275 	NFSERR_NOENT,
276 	NFSERR_IO,
277 	NFSERR_ACCES,
278 	NFSERR_NOTDIR,
279 	NFSERR_NAMETOL,
280 	NFSERR_STALE,
281 	NFSERR_BADHANDLE,
282 	NFSERR_SERVERFAULT,
283 	0,
284 };
285 
286 static const short nfsv3err_access[] = {
287 	NFSERR_IO,
288 	NFSERR_IO,
289 	NFSERR_STALE,
290 	NFSERR_BADHANDLE,
291 	NFSERR_SERVERFAULT,
292 	0,
293 };
294 
295 static const short nfsv3err_readlink[] = {
296 	NFSERR_IO,
297 	NFSERR_IO,
298 	NFSERR_ACCES,
299 	NFSERR_INVAL,
300 	NFSERR_STALE,
301 	NFSERR_BADHANDLE,
302 	NFSERR_NOTSUPP,
303 	NFSERR_SERVERFAULT,
304 	0,
305 };
306 
307 static const short nfsv3err_read[] = {
308 	NFSERR_IO,
309 	NFSERR_IO,
310 	NFSERR_NXIO,
311 	NFSERR_ACCES,
312 	NFSERR_INVAL,
313 	NFSERR_STALE,
314 	NFSERR_BADHANDLE,
315 	NFSERR_SERVERFAULT,
316 	NFSERR_JUKEBOX,
317 	0,
318 };
319 
320 static const short nfsv3err_write[] = {
321 	NFSERR_IO,
322 	NFSERR_IO,
323 	NFSERR_ACCES,
324 	NFSERR_INVAL,
325 	NFSERR_FBIG,
326 	NFSERR_NOSPC,
327 	NFSERR_ROFS,
328 	NFSERR_DQUOT,
329 	NFSERR_STALE,
330 	NFSERR_BADHANDLE,
331 	NFSERR_SERVERFAULT,
332 	NFSERR_JUKEBOX,
333 	0,
334 };
335 
336 static const short nfsv3err_create[] = {
337 	NFSERR_IO,
338 	NFSERR_IO,
339 	NFSERR_ACCES,
340 	NFSERR_EXIST,
341 	NFSERR_NOTDIR,
342 	NFSERR_NOSPC,
343 	NFSERR_ROFS,
344 	NFSERR_NAMETOL,
345 	NFSERR_DQUOT,
346 	NFSERR_STALE,
347 	NFSERR_BADHANDLE,
348 	NFSERR_NOTSUPP,
349 	NFSERR_SERVERFAULT,
350 	0,
351 };
352 
353 static const short nfsv3err_mkdir[] = {
354 	NFSERR_IO,
355 	NFSERR_IO,
356 	NFSERR_ACCES,
357 	NFSERR_EXIST,
358 	NFSERR_NOTDIR,
359 	NFSERR_NOSPC,
360 	NFSERR_ROFS,
361 	NFSERR_NAMETOL,
362 	NFSERR_DQUOT,
363 	NFSERR_STALE,
364 	NFSERR_BADHANDLE,
365 	NFSERR_NOTSUPP,
366 	NFSERR_SERVERFAULT,
367 	0,
368 };
369 
370 static const short nfsv3err_symlink[] = {
371 	NFSERR_IO,
372 	NFSERR_IO,
373 	NFSERR_ACCES,
374 	NFSERR_EXIST,
375 	NFSERR_NOTDIR,
376 	NFSERR_NOSPC,
377 	NFSERR_ROFS,
378 	NFSERR_NAMETOL,
379 	NFSERR_DQUOT,
380 	NFSERR_STALE,
381 	NFSERR_BADHANDLE,
382 	NFSERR_NOTSUPP,
383 	NFSERR_SERVERFAULT,
384 	0,
385 };
386 
387 static const short nfsv3err_mknod[] = {
388 	NFSERR_IO,
389 	NFSERR_IO,
390 	NFSERR_ACCES,
391 	NFSERR_EXIST,
392 	NFSERR_NOTDIR,
393 	NFSERR_NOSPC,
394 	NFSERR_ROFS,
395 	NFSERR_NAMETOL,
396 	NFSERR_DQUOT,
397 	NFSERR_STALE,
398 	NFSERR_BADHANDLE,
399 	NFSERR_NOTSUPP,
400 	NFSERR_SERVERFAULT,
401 	NFSERR_BADTYPE,
402 	0,
403 };
404 
405 static const short nfsv3err_remove[] = {
406 	NFSERR_IO,
407 	NFSERR_NOENT,
408 	NFSERR_IO,
409 	NFSERR_ACCES,
410 	NFSERR_NOTDIR,
411 	NFSERR_ROFS,
412 	NFSERR_NAMETOL,
413 	NFSERR_STALE,
414 	NFSERR_BADHANDLE,
415 	NFSERR_SERVERFAULT,
416 	0,
417 };
418 
419 static const short nfsv3err_rmdir[] = {
420 	NFSERR_IO,
421 	NFSERR_NOENT,
422 	NFSERR_IO,
423 	NFSERR_ACCES,
424 	NFSERR_EXIST,
425 	NFSERR_NOTDIR,
426 	NFSERR_INVAL,
427 	NFSERR_ROFS,
428 	NFSERR_NAMETOL,
429 	NFSERR_NOTEMPTY,
430 	NFSERR_STALE,
431 	NFSERR_BADHANDLE,
432 	NFSERR_NOTSUPP,
433 	NFSERR_SERVERFAULT,
434 	0,
435 };
436 
437 static const short nfsv3err_rename[] = {
438 	NFSERR_IO,
439 	NFSERR_NOENT,
440 	NFSERR_IO,
441 	NFSERR_ACCES,
442 	NFSERR_EXIST,
443 	NFSERR_XDEV,
444 	NFSERR_NOTDIR,
445 	NFSERR_ISDIR,
446 	NFSERR_INVAL,
447 	NFSERR_NOSPC,
448 	NFSERR_ROFS,
449 	NFSERR_MLINK,
450 	NFSERR_NAMETOL,
451 	NFSERR_NOTEMPTY,
452 	NFSERR_DQUOT,
453 	NFSERR_STALE,
454 	NFSERR_BADHANDLE,
455 	NFSERR_NOTSUPP,
456 	NFSERR_SERVERFAULT,
457 	0,
458 };
459 
460 static const short nfsv3err_link[] = {
461 	NFSERR_IO,
462 	NFSERR_IO,
463 	NFSERR_ACCES,
464 	NFSERR_EXIST,
465 	NFSERR_XDEV,
466 	NFSERR_NOTDIR,
467 	NFSERR_INVAL,
468 	NFSERR_NOSPC,
469 	NFSERR_ROFS,
470 	NFSERR_MLINK,
471 	NFSERR_NAMETOL,
472 	NFSERR_DQUOT,
473 	NFSERR_STALE,
474 	NFSERR_BADHANDLE,
475 	NFSERR_NOTSUPP,
476 	NFSERR_SERVERFAULT,
477 	0,
478 };
479 
480 static const short nfsv3err_readdir[] = {
481 	NFSERR_IO,
482 	NFSERR_IO,
483 	NFSERR_ACCES,
484 	NFSERR_NOTDIR,
485 	NFSERR_STALE,
486 	NFSERR_BADHANDLE,
487 	NFSERR_BAD_COOKIE,
488 	NFSERR_TOOSMALL,
489 	NFSERR_SERVERFAULT,
490 	0,
491 };
492 
493 static const short nfsv3err_readdirplus[] = {
494 	NFSERR_IO,
495 	NFSERR_IO,
496 	NFSERR_ACCES,
497 	NFSERR_NOTDIR,
498 	NFSERR_STALE,
499 	NFSERR_BADHANDLE,
500 	NFSERR_BAD_COOKIE,
501 	NFSERR_NOTSUPP,
502 	NFSERR_TOOSMALL,
503 	NFSERR_SERVERFAULT,
504 	0,
505 };
506 
507 static const short nfsv3err_fsstat[] = {
508 	NFSERR_IO,
509 	NFSERR_IO,
510 	NFSERR_STALE,
511 	NFSERR_BADHANDLE,
512 	NFSERR_SERVERFAULT,
513 	0,
514 };
515 
516 static const short nfsv3err_fsinfo[] = {
517 	NFSERR_STALE,
518 	NFSERR_STALE,
519 	NFSERR_BADHANDLE,
520 	NFSERR_SERVERFAULT,
521 	0,
522 };
523 
524 static const short nfsv3err_pathconf[] = {
525 	NFSERR_STALE,
526 	NFSERR_STALE,
527 	NFSERR_BADHANDLE,
528 	NFSERR_SERVERFAULT,
529 	0,
530 };
531 
532 static const short nfsv3err_commit[] = {
533 	NFSERR_IO,
534 	NFSERR_IO,
535 	NFSERR_STALE,
536 	NFSERR_BADHANDLE,
537 	NFSERR_SERVERFAULT,
538 	0,
539 };
540 
541 static const short * const nfsrv_v3errmap[] = {
542 	nfsv3err_null,
543 	nfsv3err_getattr,
544 	nfsv3err_setattr,
545 	nfsv3err_lookup,
546 	nfsv3err_access,
547 	nfsv3err_readlink,
548 	nfsv3err_read,
549 	nfsv3err_write,
550 	nfsv3err_create,
551 	nfsv3err_mkdir,
552 	nfsv3err_symlink,
553 	nfsv3err_mknod,
554 	nfsv3err_remove,
555 	nfsv3err_rmdir,
556 	nfsv3err_rename,
557 	nfsv3err_link,
558 	nfsv3err_readdir,
559 	nfsv3err_readdirplus,
560 	nfsv3err_fsstat,
561 	nfsv3err_fsinfo,
562 	nfsv3err_pathconf,
563 	nfsv3err_commit,
564 };
565 
566 extern struct vfs_hooks nfs_export_hooks;
567 extern struct nfsrtt nfsrtt;
568 extern struct nfsnodehashhead *nfsnodehashtbl;
569 extern u_long nfsnodehash;
570 
571 u_long nfsdirhashmask;
572 
573 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *));
574 
575 /*
576  * Create the header for an rpc request packet
577  * The hsiz is the size of the rest of the nfs request header.
578  * (just used to decide if a cluster is a good idea)
579  */
580 struct mbuf *
581 nfsm_reqh(struct nfsnode *np, u_long procid, int hsiz, char **bposp)
582 {
583 	struct mbuf *mb;
584 	char *bpos;
585 
586 	mb = m_get(M_WAIT, MT_DATA);
587 	MCLAIM(mb, &nfs_mowner);
588 	if (hsiz >= MINCLSIZE)
589 		m_clget(mb, M_WAIT);
590 	mb->m_len = 0;
591 	bpos = mtod(mb, void *);
592 
593 	/* Finally, return values */
594 	*bposp = bpos;
595 	return (mb);
596 }
597 
598 /*
599  * Build the RPC header and fill in the authorization info.
600  * The authorization string argument is only used when the credentials
601  * come from outside of the kernel.
602  * Returns the head of the mbuf list.
603  */
604 struct mbuf *
605 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len,
606 	verf_str, mrest, mrest_len, mbp, xidp)
607 	kauth_cred_t cr;
608 	int nmflag;
609 	int procid;
610 	int auth_type;
611 	int auth_len;
612 	char *auth_str;
613 	int verf_len;
614 	char *verf_str;
615 	struct mbuf *mrest;
616 	int mrest_len;
617 	struct mbuf **mbp;
618 	u_int32_t *xidp;
619 {
620 	struct mbuf *mb;
621 	u_int32_t *tl;
622 	char *bpos;
623 	int i;
624 	struct mbuf *mreq;
625 	int siz, grpsiz, authsiz;
626 
627 	authsiz = nfsm_rndup(auth_len);
628 	mb = m_gethdr(M_WAIT, MT_DATA);
629 	MCLAIM(mb, &nfs_mowner);
630 	if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
631 		m_clget(mb, M_WAIT);
632 	} else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
633 		MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED);
634 	} else {
635 		MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
636 	}
637 	mb->m_len = 0;
638 	mreq = mb;
639 	bpos = mtod(mb, void *);
640 
641 	/*
642 	 * First the RPC header.
643 	 */
644 	nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
645 
646 	*tl++ = *xidp = nfs_getxid();
647 	*tl++ = rpc_call;
648 	*tl++ = rpc_vers;
649 	*tl++ = txdr_unsigned(NFS_PROG);
650 	if (nmflag & NFSMNT_NFSV3)
651 		*tl++ = txdr_unsigned(NFS_VER3);
652 	else
653 		*tl++ = txdr_unsigned(NFS_VER2);
654 	if (nmflag & NFSMNT_NFSV3)
655 		*tl++ = txdr_unsigned(procid);
656 	else
657 		*tl++ = txdr_unsigned(nfsv2_procid[procid]);
658 
659 	/*
660 	 * And then the authorization cred.
661 	 */
662 	*tl++ = txdr_unsigned(auth_type);
663 	*tl = txdr_unsigned(authsiz);
664 	switch (auth_type) {
665 	case RPCAUTH_UNIX:
666 		nfsm_build(tl, u_int32_t *, auth_len);
667 		*tl++ = 0;		/* stamp ?? */
668 		*tl++ = 0;		/* NULL hostname */
669 		*tl++ = txdr_unsigned(kauth_cred_geteuid(cr));
670 		*tl++ = txdr_unsigned(kauth_cred_getegid(cr));
671 		grpsiz = (auth_len >> 2) - 5;
672 		*tl++ = txdr_unsigned(grpsiz);
673 		for (i = 0; i < grpsiz; i++)
674 			*tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */
675 		break;
676 	case RPCAUTH_KERB4:
677 		siz = auth_len;
678 		while (siz > 0) {
679 			if (M_TRAILINGSPACE(mb) == 0) {
680 				struct mbuf *mb2;
681 				mb2 = m_get(M_WAIT, MT_DATA);
682 				MCLAIM(mb2, &nfs_mowner);
683 				if (siz >= MINCLSIZE)
684 					m_clget(mb2, M_WAIT);
685 				mb->m_next = mb2;
686 				mb = mb2;
687 				mb->m_len = 0;
688 				bpos = mtod(mb, void *);
689 			}
690 			i = min(siz, M_TRAILINGSPACE(mb));
691 			memcpy(bpos, auth_str, i);
692 			mb->m_len += i;
693 			auth_str += i;
694 			bpos += i;
695 			siz -= i;
696 		}
697 		if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
698 			for (i = 0; i < siz; i++)
699 				*bpos++ = '\0';
700 			mb->m_len += siz;
701 		}
702 		break;
703 	};
704 
705 	/*
706 	 * And the verifier...
707 	 */
708 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
709 	if (verf_str) {
710 		*tl++ = txdr_unsigned(RPCAUTH_KERB4);
711 		*tl = txdr_unsigned(verf_len);
712 		siz = verf_len;
713 		while (siz > 0) {
714 			if (M_TRAILINGSPACE(mb) == 0) {
715 				struct mbuf *mb2;
716 				mb2 = m_get(M_WAIT, MT_DATA);
717 				MCLAIM(mb2, &nfs_mowner);
718 				if (siz >= MINCLSIZE)
719 					m_clget(mb2, M_WAIT);
720 				mb->m_next = mb2;
721 				mb = mb2;
722 				mb->m_len = 0;
723 				bpos = mtod(mb, void *);
724 			}
725 			i = min(siz, M_TRAILINGSPACE(mb));
726 			memcpy(bpos, verf_str, i);
727 			mb->m_len += i;
728 			verf_str += i;
729 			bpos += i;
730 			siz -= i;
731 		}
732 		if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
733 			for (i = 0; i < siz; i++)
734 				*bpos++ = '\0';
735 			mb->m_len += siz;
736 		}
737 	} else {
738 		*tl++ = txdr_unsigned(RPCAUTH_NULL);
739 		*tl = 0;
740 	}
741 	mb->m_next = mrest;
742 	mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
743 	mreq->m_pkthdr.rcvif = (struct ifnet *)0;
744 	*mbp = mb;
745 	return (mreq);
746 }
747 
748 /*
749  * copies mbuf chain to the uio scatter/gather list
750  */
751 int
752 nfsm_mbuftouio(mrep, uiop, siz, dpos)
753 	struct mbuf **mrep;
754 	struct uio *uiop;
755 	int siz;
756 	char **dpos;
757 {
758 	char *mbufcp, *uiocp;
759 	int xfer, left, len;
760 	struct mbuf *mp;
761 	long uiosiz, rem;
762 	int error = 0;
763 
764 	mp = *mrep;
765 	mbufcp = *dpos;
766 	len = mtod(mp, char *) + mp->m_len - mbufcp;
767 	rem = nfsm_rndup(siz)-siz;
768 	while (siz > 0) {
769 		if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
770 			return (EFBIG);
771 		left = uiop->uio_iov->iov_len;
772 		uiocp = uiop->uio_iov->iov_base;
773 		if (left > siz)
774 			left = siz;
775 		uiosiz = left;
776 		while (left > 0) {
777 			while (len == 0) {
778 				mp = mp->m_next;
779 				if (mp == NULL)
780 					return (EBADRPC);
781 				mbufcp = mtod(mp, void *);
782 				len = mp->m_len;
783 			}
784 			xfer = (left > len) ? len : left;
785 			error = copyout_vmspace(uiop->uio_vmspace, mbufcp,
786 			    uiocp, xfer);
787 			if (error) {
788 				return error;
789 			}
790 			left -= xfer;
791 			len -= xfer;
792 			mbufcp += xfer;
793 			uiocp += xfer;
794 			uiop->uio_offset += xfer;
795 			uiop->uio_resid -= xfer;
796 		}
797 		if (uiop->uio_iov->iov_len <= siz) {
798 			uiop->uio_iovcnt--;
799 			uiop->uio_iov++;
800 		} else {
801 			uiop->uio_iov->iov_base =
802 			    (char *)uiop->uio_iov->iov_base + uiosiz;
803 			uiop->uio_iov->iov_len -= uiosiz;
804 		}
805 		siz -= uiosiz;
806 	}
807 	*dpos = mbufcp;
808 	*mrep = mp;
809 	if (rem > 0) {
810 		if (len < rem)
811 			error = nfs_adv(mrep, dpos, rem, len);
812 		else
813 			*dpos += rem;
814 	}
815 	return (error);
816 }
817 
818 /*
819  * copies a uio scatter/gather list to an mbuf chain.
820  * NOTE: can ony handle iovcnt == 1
821  */
822 int
823 nfsm_uiotombuf(uiop, mq, siz, bpos)
824 	struct uio *uiop;
825 	struct mbuf **mq;
826 	int siz;
827 	char **bpos;
828 {
829 	char *uiocp;
830 	struct mbuf *mp, *mp2;
831 	int xfer, left, mlen;
832 	int uiosiz, clflg, rem;
833 	char *cp;
834 	int error;
835 
836 #ifdef DIAGNOSTIC
837 	if (uiop->uio_iovcnt != 1)
838 		panic("nfsm_uiotombuf: iovcnt != 1");
839 #endif
840 
841 	if (siz > MLEN)		/* or should it >= MCLBYTES ?? */
842 		clflg = 1;
843 	else
844 		clflg = 0;
845 	rem = nfsm_rndup(siz)-siz;
846 	mp = mp2 = *mq;
847 	while (siz > 0) {
848 		left = uiop->uio_iov->iov_len;
849 		uiocp = uiop->uio_iov->iov_base;
850 		if (left > siz)
851 			left = siz;
852 		uiosiz = left;
853 		while (left > 0) {
854 			mlen = M_TRAILINGSPACE(mp);
855 			if (mlen == 0) {
856 				mp = m_get(M_WAIT, MT_DATA);
857 				MCLAIM(mp, &nfs_mowner);
858 				if (clflg)
859 					m_clget(mp, M_WAIT);
860 				mp->m_len = 0;
861 				mp2->m_next = mp;
862 				mp2 = mp;
863 				mlen = M_TRAILINGSPACE(mp);
864 			}
865 			xfer = (left > mlen) ? mlen : left;
866 			cp = mtod(mp, char *) + mp->m_len;
867 			error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp,
868 			    xfer);
869 			if (error) {
870 				/* XXX */
871 			}
872 			mp->m_len += xfer;
873 			left -= xfer;
874 			uiocp += xfer;
875 			uiop->uio_offset += xfer;
876 			uiop->uio_resid -= xfer;
877 		}
878 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base +
879 		    uiosiz;
880 		uiop->uio_iov->iov_len -= uiosiz;
881 		siz -= uiosiz;
882 	}
883 	if (rem > 0) {
884 		if (rem > M_TRAILINGSPACE(mp)) {
885 			mp = m_get(M_WAIT, MT_DATA);
886 			MCLAIM(mp, &nfs_mowner);
887 			mp->m_len = 0;
888 			mp2->m_next = mp;
889 		}
890 		cp = mtod(mp, char *) + mp->m_len;
891 		for (left = 0; left < rem; left++)
892 			*cp++ = '\0';
893 		mp->m_len += rem;
894 		*bpos = cp;
895 	} else
896 		*bpos = mtod(mp, char *) + mp->m_len;
897 	*mq = mp;
898 	return (0);
899 }
900 
901 /*
902  * Get at least "siz" bytes of correctly aligned data.
903  * When called the mbuf pointers are not necessarily correct,
904  * dsosp points to what ought to be in m_data and left contains
905  * what ought to be in m_len.
906  * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
907  * cases. (The macros use the vars. dpos and dpos2)
908  */
909 int
910 nfsm_disct(mdp, dposp, siz, left, cp2)
911 	struct mbuf **mdp;
912 	char **dposp;
913 	int siz;
914 	int left;
915 	char **cp2;
916 {
917 	struct mbuf *m1, *m2;
918 	struct mbuf *havebuf = NULL;
919 	char *src = *dposp;
920 	char *dst;
921 	int len;
922 
923 #ifdef DEBUG
924 	if (left < 0)
925 		panic("nfsm_disct: left < 0");
926 #endif
927 	m1 = *mdp;
928 	/*
929 	 * Skip through the mbuf chain looking for an mbuf with
930 	 * some data. If the first mbuf found has enough data
931 	 * and it is correctly aligned return it.
932 	 */
933 	while (left == 0) {
934 		havebuf = m1;
935 		*mdp = m1 = m1->m_next;
936 		if (m1 == NULL)
937 			return (EBADRPC);
938 		src = mtod(m1, void *);
939 		left = m1->m_len;
940 		/*
941 		 * If we start a new mbuf and it is big enough
942 		 * and correctly aligned just return it, don't
943 		 * do any pull up.
944 		 */
945 		if (left >= siz && nfsm_aligned(src)) {
946 			*cp2 = src;
947 			*dposp = src + siz;
948 			return (0);
949 		}
950 	}
951 	if ((m1->m_flags & M_EXT) != 0) {
952 		if (havebuf && M_TRAILINGSPACE(havebuf) >= siz &&
953 		    nfsm_aligned(mtod(havebuf, char *) + havebuf->m_len)) {
954 			/*
955 			 * If the first mbuf with data has external data
956 			 * and there is a previous mbuf with some trailing
957 			 * space, use it to move the data into.
958 			 */
959 			m2 = m1;
960 			*mdp = m1 = havebuf;
961 			*cp2 = mtod(m1, char *) + m1->m_len;
962 		} else if (havebuf) {
963 			/*
964 			 * If the first mbuf has a external data
965 			 * and there is no previous empty mbuf
966 			 * allocate a new mbuf and move the external
967 			 * data to the new mbuf. Also make the first
968 			 * mbuf look empty.
969 			 */
970 			m2 = m1;
971 			*mdp = m1 = m_get(M_WAIT, MT_DATA);
972 			MCLAIM(m1, m2->m_owner);
973 			if ((m2->m_flags & M_PKTHDR) != 0) {
974 				/* XXX MOVE */
975 				M_COPY_PKTHDR(m1, m2);
976 				m_tag_delete_chain(m2, NULL);
977 				m2->m_flags &= ~M_PKTHDR;
978 			}
979 			if (havebuf) {
980 				havebuf->m_next = m1;
981 			}
982 			m1->m_next = m2;
983 			MRESETDATA(m1);
984 			m1->m_len = 0;
985 			m2->m_data = src;
986 			m2->m_len = left;
987 			*cp2 = mtod(m1, char *);
988 		} else {
989 			struct mbuf **nextp = &m1->m_next;
990 
991 			m1->m_len -= left;
992 			do {
993 				m2 = m_get(M_WAIT, MT_DATA);
994 				MCLAIM(m2, m1->m_owner);
995 				if (left >= MINCLSIZE) {
996 					MCLGET(m2, M_WAIT);
997 				}
998 				m2->m_next = *nextp;
999 				*nextp = m2;
1000 				nextp = &m2->m_next;
1001 				len = (m2->m_flags & M_EXT) != 0 ?
1002 				    MCLBYTES : MLEN;
1003 				if (len > left) {
1004 					len = left;
1005 				}
1006 				memcpy(mtod(m2, char *), src, len);
1007 				m2->m_len = len;
1008 				src += len;
1009 				left -= len;
1010 			} while (left > 0);
1011 			*mdp = m1 = m1->m_next;
1012 			m2 = m1->m_next;
1013 			*cp2 = mtod(m1, char *);
1014 		}
1015 	} else {
1016 		/*
1017 		 * If the first mbuf has no external data
1018 		 * move the data to the front of the mbuf.
1019 		 */
1020 		MRESETDATA(m1);
1021 		dst = mtod(m1, char *);
1022 		if (dst != src) {
1023 			memmove(dst, src, left);
1024 		}
1025 		m1->m_len = left;
1026 		m2 = m1->m_next;
1027 		*cp2 = m1->m_data;
1028 	}
1029 	*dposp = *cp2 + siz;
1030 	/*
1031 	 * Loop through mbufs pulling data up into first mbuf until
1032 	 * the first mbuf is full or there is no more data to
1033 	 * pullup.
1034 	 */
1035 	dst = mtod(m1, char *) + m1->m_len;
1036 	while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) {
1037 		if ((len = min(len, m2->m_len)) != 0) {
1038 			memcpy(dst, mtod(m2, char *), len);
1039 		}
1040 		m1->m_len += len;
1041 		dst += len;
1042 		m2->m_data += len;
1043 		m2->m_len -= len;
1044 		m2 = m2->m_next;
1045 	}
1046 	if (m1->m_len < siz)
1047 		return (EBADRPC);
1048 	return (0);
1049 }
1050 
1051 /*
1052  * Advance the position in the mbuf chain.
1053  */
1054 int
1055 nfs_adv(mdp, dposp, offs, left)
1056 	struct mbuf **mdp;
1057 	char **dposp;
1058 	int offs;
1059 	int left;
1060 {
1061 	struct mbuf *m;
1062 	int s;
1063 
1064 	m = *mdp;
1065 	s = left;
1066 	while (s < offs) {
1067 		offs -= s;
1068 		m = m->m_next;
1069 		if (m == NULL)
1070 			return (EBADRPC);
1071 		s = m->m_len;
1072 	}
1073 	*mdp = m;
1074 	*dposp = mtod(m, char *) + offs;
1075 	return (0);
1076 }
1077 
1078 /*
1079  * Copy a string into mbufs for the hard cases...
1080  */
1081 int
1082 nfsm_strtmbuf(mb, bpos, cp, siz)
1083 	struct mbuf **mb;
1084 	char **bpos;
1085 	const char *cp;
1086 	long siz;
1087 {
1088 	struct mbuf *m1 = NULL, *m2;
1089 	long left, xfer, len, tlen;
1090 	u_int32_t *tl;
1091 	int putsize;
1092 
1093 	putsize = 1;
1094 	m2 = *mb;
1095 	left = M_TRAILINGSPACE(m2);
1096 	if (left > 0) {
1097 		tl = ((u_int32_t *)(*bpos));
1098 		*tl++ = txdr_unsigned(siz);
1099 		putsize = 0;
1100 		left -= NFSX_UNSIGNED;
1101 		m2->m_len += NFSX_UNSIGNED;
1102 		if (left > 0) {
1103 			memcpy((void *) tl, cp, left);
1104 			siz -= left;
1105 			cp += left;
1106 			m2->m_len += left;
1107 			left = 0;
1108 		}
1109 	}
1110 	/* Loop around adding mbufs */
1111 	while (siz > 0) {
1112 		m1 = m_get(M_WAIT, MT_DATA);
1113 		MCLAIM(m1, &nfs_mowner);
1114 		if (siz > MLEN)
1115 			m_clget(m1, M_WAIT);
1116 		m1->m_len = NFSMSIZ(m1);
1117 		m2->m_next = m1;
1118 		m2 = m1;
1119 		tl = mtod(m1, u_int32_t *);
1120 		tlen = 0;
1121 		if (putsize) {
1122 			*tl++ = txdr_unsigned(siz);
1123 			m1->m_len -= NFSX_UNSIGNED;
1124 			tlen = NFSX_UNSIGNED;
1125 			putsize = 0;
1126 		}
1127 		if (siz < m1->m_len) {
1128 			len = nfsm_rndup(siz);
1129 			xfer = siz;
1130 			if (xfer < len)
1131 				*(tl+(xfer>>2)) = 0;
1132 		} else {
1133 			xfer = len = m1->m_len;
1134 		}
1135 		memcpy((void *) tl, cp, xfer);
1136 		m1->m_len = len+tlen;
1137 		siz -= xfer;
1138 		cp += xfer;
1139 	}
1140 	*mb = m1;
1141 	*bpos = mtod(m1, char *) + m1->m_len;
1142 	return (0);
1143 }
1144 
1145 /*
1146  * Directory caching routines. They work as follows:
1147  * - a cache is maintained per VDIR nfsnode.
1148  * - for each offset cookie that is exported to userspace, and can
1149  *   thus be thrown back at us as an offset to VOP_READDIR, store
1150  *   information in the cache.
1151  * - cached are:
1152  *   - cookie itself
1153  *   - blocknumber (essentially just a search key in the buffer cache)
1154  *   - entry number in block.
1155  *   - offset cookie of block in which this entry is stored
1156  *   - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
1157  * - entries are looked up in a hash table
1158  * - also maintained is an LRU list of entries, used to determine
1159  *   which ones to delete if the cache grows too large.
1160  * - if 32 <-> 64 translation mode is requested for a filesystem,
1161  *   the cache also functions as a translation table
1162  * - in the translation case, invalidating the cache does not mean
1163  *   flushing it, but just marking entries as invalid, except for
1164  *   the <64bit cookie, 32bitcookie> pair which is still valid, to
1165  *   still be able to use the cache as a translation table.
1166  * - 32 bit cookies are uniquely created by combining the hash table
1167  *   entry value, and one generation count per hash table entry,
1168  *   incremented each time an entry is appended to the chain.
1169  * - the cache is invalidated each time a direcory is modified
1170  * - sanity checks are also done; if an entry in a block turns
1171  *   out not to have a matching cookie, the cache is invalidated
1172  *   and a new block starting from the wanted offset is fetched from
1173  *   the server.
1174  * - directory entries as read from the server are extended to contain
1175  *   the 64bit and, optionally, the 32bit cookies, for sanity checking
1176  *   the cache and exporting them to userspace through the cookie
1177  *   argument to VOP_READDIR.
1178  */
1179 
1180 u_long
1181 nfs_dirhash(off)
1182 	off_t off;
1183 {
1184 	int i;
1185 	char *cp = (char *)&off;
1186 	u_long sum = 0L;
1187 
1188 	for (i = 0 ; i < sizeof (off); i++)
1189 		sum += *cp++;
1190 
1191 	return sum;
1192 }
1193 
1194 #define	_NFSDC_MTX(np)		(&NFSTOV(np)->v_interlock)
1195 #define	NFSDC_LOCK(np)		mutex_enter(_NFSDC_MTX(np))
1196 #define	NFSDC_UNLOCK(np)	mutex_exit(_NFSDC_MTX(np))
1197 #define	NFSDC_ASSERT_LOCKED(np) KASSERT(mutex_owned(_NFSDC_MTX(np)))
1198 
1199 void
1200 nfs_initdircache(vp)
1201 	struct vnode *vp;
1202 {
1203 	struct nfsnode *np = VTONFS(vp);
1204 	struct nfsdirhashhead *dircache;
1205 
1206 	dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, true,
1207 	    &nfsdirhashmask);
1208 
1209 	NFSDC_LOCK(np);
1210 	if (np->n_dircache == NULL) {
1211 		np->n_dircachesize = 0;
1212 		np->n_dircache = dircache;
1213 		dircache = NULL;
1214 		TAILQ_INIT(&np->n_dirchain);
1215 	}
1216 	NFSDC_UNLOCK(np);
1217 	if (dircache)
1218 		hashdone(dircache, HASH_LIST, nfsdirhashmask);
1219 }
1220 
1221 void
1222 nfs_initdirxlatecookie(vp)
1223 	struct vnode *vp;
1224 {
1225 	struct nfsnode *np = VTONFS(vp);
1226 	unsigned *dirgens;
1227 
1228 	KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE);
1229 
1230 	dirgens = kmem_zalloc(NFS_DIRHASHSIZ * sizeof(unsigned), KM_SLEEP);
1231 	NFSDC_LOCK(np);
1232 	if (np->n_dirgens == NULL) {
1233 		np->n_dirgens = dirgens;
1234 		dirgens = NULL;
1235 	}
1236 	NFSDC_UNLOCK(np);
1237 	if (dirgens)
1238 		kmem_free(dirgens, NFS_DIRHASHSIZ * sizeof(unsigned));
1239 }
1240 
1241 static const struct nfsdircache dzero;
1242 
1243 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *));
1244 static void nfs_putdircache_unlocked __P((struct nfsnode *,
1245     struct nfsdircache *));
1246 
1247 static void
1248 nfs_unlinkdircache(np, ndp)
1249 	struct nfsnode *np;
1250 	struct nfsdircache *ndp;
1251 {
1252 
1253 	NFSDC_ASSERT_LOCKED(np);
1254 	KASSERT(ndp != &dzero);
1255 
1256 	if (LIST_NEXT(ndp, dc_hash) == (void *)-1)
1257 		return;
1258 
1259 	TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1260 	LIST_REMOVE(ndp, dc_hash);
1261 	LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */
1262 
1263 	nfs_putdircache_unlocked(np, ndp);
1264 }
1265 
1266 void
1267 nfs_putdircache(np, ndp)
1268 	struct nfsnode *np;
1269 	struct nfsdircache *ndp;
1270 {
1271 	int ref;
1272 
1273 	if (ndp == &dzero)
1274 		return;
1275 
1276 	KASSERT(ndp->dc_refcnt > 0);
1277 	NFSDC_LOCK(np);
1278 	ref = --ndp->dc_refcnt;
1279 	NFSDC_UNLOCK(np);
1280 
1281 	if (ref == 0)
1282 		kmem_free(ndp, sizeof(*ndp));
1283 }
1284 
1285 static void
1286 nfs_putdircache_unlocked(struct nfsnode *np, struct nfsdircache *ndp)
1287 {
1288 	int ref;
1289 
1290 	NFSDC_ASSERT_LOCKED(np);
1291 
1292 	if (ndp == &dzero)
1293 		return;
1294 
1295 	KASSERT(ndp->dc_refcnt > 0);
1296 	ref = --ndp->dc_refcnt;
1297 	if (ref == 0)
1298 		kmem_free(ndp, sizeof(*ndp));
1299 }
1300 
1301 struct nfsdircache *
1302 nfs_searchdircache(vp, off, do32, hashent)
1303 	struct vnode *vp;
1304 	off_t off;
1305 	int do32;
1306 	int *hashent;
1307 {
1308 	struct nfsdirhashhead *ndhp;
1309 	struct nfsdircache *ndp = NULL;
1310 	struct nfsnode *np = VTONFS(vp);
1311 	unsigned ent;
1312 
1313 	/*
1314 	 * Zero is always a valid cookie.
1315 	 */
1316 	if (off == 0)
1317 		/* XXXUNCONST */
1318 		return (struct nfsdircache *)__UNCONST(&dzero);
1319 
1320 	if (!np->n_dircache)
1321 		return NULL;
1322 
1323 	/*
1324 	 * We use a 32bit cookie as search key, directly reconstruct
1325 	 * the hashentry. Else use the hashfunction.
1326 	 */
1327 	if (do32) {
1328 		ent = (u_int32_t)off >> 24;
1329 		if (ent >= NFS_DIRHASHSIZ)
1330 			return NULL;
1331 		ndhp = &np->n_dircache[ent];
1332 	} else {
1333 		ndhp = NFSDIRHASH(np, off);
1334 	}
1335 
1336 	if (hashent)
1337 		*hashent = (int)(ndhp - np->n_dircache);
1338 
1339 	NFSDC_LOCK(np);
1340 	if (do32) {
1341 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1342 			if (ndp->dc_cookie32 == (u_int32_t)off) {
1343 				/*
1344 				 * An invalidated entry will become the
1345 				 * start of a new block fetched from
1346 				 * the server.
1347 				 */
1348 				if (ndp->dc_flags & NFSDC_INVALID) {
1349 					ndp->dc_blkcookie = ndp->dc_cookie;
1350 					ndp->dc_entry = 0;
1351 					ndp->dc_flags &= ~NFSDC_INVALID;
1352 				}
1353 				break;
1354 			}
1355 		}
1356 	} else {
1357 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1358 			if (ndp->dc_cookie == off)
1359 				break;
1360 		}
1361 	}
1362 	if (ndp != NULL)
1363 		ndp->dc_refcnt++;
1364 	NFSDC_UNLOCK(np);
1365 	return ndp;
1366 }
1367 
1368 
1369 struct nfsdircache *
1370 nfs_enterdircache(struct vnode *vp, off_t off, off_t blkoff, int en,
1371     daddr_t blkno)
1372 {
1373 	struct nfsnode *np = VTONFS(vp);
1374 	struct nfsdirhashhead *ndhp;
1375 	struct nfsdircache *ndp = NULL;
1376 	struct nfsdircache *newndp = NULL;
1377 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1378 	int hashent = 0, gen, overwrite;	/* XXX: GCC */
1379 
1380 	/*
1381 	 * XXX refuse entries for offset 0. amd(8) erroneously sets
1382 	 * cookie 0 for the '.' entry, making this necessary. This
1383 	 * isn't so bad, as 0 is a special case anyway.
1384 	 */
1385 	if (off == 0)
1386 		/* XXXUNCONST */
1387 		return (struct nfsdircache *)__UNCONST(&dzero);
1388 
1389 	if (!np->n_dircache)
1390 		/*
1391 		 * XXX would like to do this in nfs_nget but vtype
1392 		 * isn't known at that time.
1393 		 */
1394 		nfs_initdircache(vp);
1395 
1396 	if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens)
1397 		nfs_initdirxlatecookie(vp);
1398 
1399 retry:
1400 	ndp = nfs_searchdircache(vp, off, 0, &hashent);
1401 
1402 	NFSDC_LOCK(np);
1403 	if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) {
1404 		/*
1405 		 * Overwriting an old entry. Check if it's the same.
1406 		 * If so, just return. If not, remove the old entry.
1407 		 */
1408 		if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
1409 			goto done;
1410 		nfs_unlinkdircache(np, ndp);
1411 		nfs_putdircache_unlocked(np, ndp);
1412 		ndp = NULL;
1413 	}
1414 
1415 	ndhp = &np->n_dircache[hashent];
1416 
1417 	if (!ndp) {
1418 		if (newndp == NULL) {
1419 			NFSDC_UNLOCK(np);
1420 			newndp = kmem_alloc(sizeof(*newndp), KM_SLEEP);
1421 			newndp->dc_refcnt = 1;
1422 			LIST_NEXT(newndp, dc_hash) = (void *)-1;
1423 			goto retry;
1424 		}
1425 		ndp = newndp;
1426 		newndp = NULL;
1427 		overwrite = 0;
1428 		if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1429 			/*
1430 			 * We're allocating a new entry, so bump the
1431 			 * generation number.
1432 			 */
1433 			KASSERT(np->n_dirgens);
1434 			gen = ++np->n_dirgens[hashent];
1435 			if (gen == 0) {
1436 				np->n_dirgens[hashent]++;
1437 				gen++;
1438 			}
1439 			ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
1440 		}
1441 	} else
1442 		overwrite = 1;
1443 
1444 	ndp->dc_cookie = off;
1445 	ndp->dc_blkcookie = blkoff;
1446 	ndp->dc_entry = en;
1447 	ndp->dc_flags = 0;
1448 
1449 	if (overwrite)
1450 		goto done;
1451 
1452 	/*
1453 	 * If the maximum directory cookie cache size has been reached
1454 	 * for this node, take one off the front. The idea is that
1455 	 * directories are typically read front-to-back once, so that
1456 	 * the oldest entries can be thrown away without much performance
1457 	 * loss.
1458 	 */
1459 	if (np->n_dircachesize == NFS_MAXDIRCACHE) {
1460 		nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain));
1461 	} else
1462 		np->n_dircachesize++;
1463 
1464 	KASSERT(ndp->dc_refcnt == 1);
1465 	LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
1466 	TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
1467 	ndp->dc_refcnt++;
1468 done:
1469 	KASSERT(ndp->dc_refcnt > 0);
1470 	NFSDC_UNLOCK(np);
1471 	if (newndp)
1472 		nfs_putdircache(np, newndp);
1473 	return ndp;
1474 }
1475 
1476 void
1477 nfs_invaldircache(vp, flags)
1478 	struct vnode *vp;
1479 	int flags;
1480 {
1481 	struct nfsnode *np = VTONFS(vp);
1482 	struct nfsdircache *ndp = NULL;
1483 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1484 	const bool forcefree = flags & NFS_INVALDIRCACHE_FORCE;
1485 
1486 #ifdef DIAGNOSTIC
1487 	if (vp->v_type != VDIR)
1488 		panic("nfs: invaldircache: not dir");
1489 #endif
1490 
1491 	if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0)
1492 		np->n_flag &= ~NEOFVALID;
1493 
1494 	if (!np->n_dircache)
1495 		return;
1496 
1497 	NFSDC_LOCK(np);
1498 	if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
1499 		while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) {
1500 			KASSERT(!forcefree || ndp->dc_refcnt == 1);
1501 			nfs_unlinkdircache(np, ndp);
1502 		}
1503 		np->n_dircachesize = 0;
1504 		if (forcefree && np->n_dirgens) {
1505 			kmem_free(np->n_dirgens,
1506 			    NFS_DIRHASHSIZ * sizeof(unsigned));
1507 			np->n_dirgens = NULL;
1508 		}
1509 	} else {
1510 		TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain)
1511 			ndp->dc_flags |= NFSDC_INVALID;
1512 	}
1513 
1514 	NFSDC_UNLOCK(np);
1515 }
1516 
1517 /*
1518  * Called once before VFS init to initialize shared and
1519  * server-specific data structures.
1520  */
1521 static int
1522 nfs_init0(void)
1523 {
1524 
1525 	nfsrtt.pos = 0;
1526 	rpc_vers = txdr_unsigned(RPC_VER2);
1527 	rpc_call = txdr_unsigned(RPC_CALL);
1528 	rpc_reply = txdr_unsigned(RPC_REPLY);
1529 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1530 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1531 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1532 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1533 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1534 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1535 	nfs_prog = txdr_unsigned(NFS_PROG);
1536 	nfs_true = txdr_unsigned(true);
1537 	nfs_false = txdr_unsigned(false);
1538 	nfs_xdrneg1 = txdr_unsigned(-1);
1539 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1540 	if (nfs_ticks < 1)
1541 		nfs_ticks = 1;
1542 #ifdef NFSSERVER
1543 	vfs_hooks_attach(&nfs_export_hooks);
1544 	nfsrv_init(0);			/* Init server data structures */
1545 	nfsrv_initcache();		/* Init the server request cache */
1546 	{
1547 		extern krwlock_t netexport_lock;	/* XXX */
1548 		rw_init(&netexport_lock);
1549 	}
1550 #endif /* NFSSERVER */
1551 
1552 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY))
1553 	nfsdreq_init();
1554 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */
1555 
1556 	/*
1557 	 * Initialize reply list and start timer
1558 	 */
1559 	TAILQ_INIT(&nfs_reqq);
1560 	nfs_timer_init();
1561 	MOWNER_ATTACH(&nfs_mowner);
1562 
1563 #ifdef NFS
1564 	/* Initialize the kqueue structures */
1565 	nfs_kqinit();
1566 	/* Initialize the iod structures */
1567 	nfs_iodinit();
1568 #endif
1569 	return 0;
1570 }
1571 
1572 void
1573 nfs_init(void)
1574 {
1575 	static ONCE_DECL(nfs_init_once);
1576 
1577 	RUN_ONCE(&nfs_init_once, nfs_init0);
1578 }
1579 
1580 #ifdef NFS
1581 /*
1582  * Called once at VFS init to initialize client-specific data structures.
1583  */
1584 void
1585 nfs_vfs_init()
1586 {
1587 	/* Initialize NFS server / client shared data. */
1588 	nfs_init();
1589 
1590 	nfs_nhinit();			/* Init the nfsnode table */
1591 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1592 }
1593 
1594 void
1595 nfs_vfs_reinit()
1596 {
1597 	nfs_nhreinit();
1598 }
1599 
1600 void
1601 nfs_vfs_done()
1602 {
1603 	nfs_nhdone();
1604 }
1605 
1606 /*
1607  * Attribute cache routines.
1608  * nfs_loadattrcache() - loads or updates the cache contents from attributes
1609  *	that are on the mbuf list
1610  * nfs_getattrcache() - returns valid attributes if found in cache, returns
1611  *	error otherwise
1612  */
1613 
1614 /*
1615  * Load the attribute cache (that lives in the nfsnode entry) with
1616  * the values on the mbuf list and
1617  * Iff vap not NULL
1618  *    copy the attributes to *vaper
1619  */
1620 int
1621 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags)
1622 	struct vnode **vpp;
1623 	struct mbuf **mdp;
1624 	char **dposp;
1625 	struct vattr *vaper;
1626 	int flags;
1627 {
1628 	int32_t t1;
1629 	char *cp2;
1630 	int error = 0;
1631 	struct mbuf *md;
1632 	int v3 = NFS_ISV3(*vpp);
1633 
1634 	md = *mdp;
1635 	t1 = (mtod(md, char *) + md->m_len) - *dposp;
1636 	error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2);
1637 	if (error)
1638 		return (error);
1639 	return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags);
1640 }
1641 
1642 int
1643 nfs_loadattrcache(vpp, fp, vaper, flags)
1644 	struct vnode **vpp;
1645 	struct nfs_fattr *fp;
1646 	struct vattr *vaper;
1647 	int flags;
1648 {
1649 	struct vnode *vp = *vpp;
1650 	struct vattr *vap;
1651 	int v3 = NFS_ISV3(vp);
1652 	enum vtype vtyp;
1653 	u_short vmode;
1654 	struct timespec mtime;
1655 	struct timespec ctime;
1656 	int32_t rdev;
1657 	struct nfsnode *np;
1658 	extern int (**spec_nfsv2nodeop_p) __P((void *));
1659 	uid_t uid;
1660 	gid_t gid;
1661 
1662 	if (v3) {
1663 		vtyp = nfsv3tov_type(fp->fa_type);
1664 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1665 		rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1),
1666 			fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2));
1667 		fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
1668 		fxdr_nfsv3time(&fp->fa3_ctime, &ctime);
1669 	} else {
1670 		vtyp = nfsv2tov_type(fp->fa_type);
1671 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1672 		if (vtyp == VNON || vtyp == VREG)
1673 			vtyp = IFTOVT(vmode);
1674 		rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
1675 		fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
1676 		ctime.tv_sec = fxdr_unsigned(u_int32_t,
1677 		    fp->fa2_ctime.nfsv2_sec);
1678 		ctime.tv_nsec = 0;
1679 
1680 		/*
1681 		 * Really ugly NFSv2 kludge.
1682 		 */
1683 		if (vtyp == VCHR && rdev == 0xffffffff)
1684 			vtyp = VFIFO;
1685 	}
1686 
1687 	vmode &= ALLPERMS;
1688 
1689 	/*
1690 	 * If v_type == VNON it is a new node, so fill in the v_type,
1691 	 * n_mtime fields. Check to see if it represents a special
1692 	 * device, and if so, check for a possible alias. Once the
1693 	 * correct vnode has been obtained, fill in the rest of the
1694 	 * information.
1695 	 */
1696 	np = VTONFS(vp);
1697 	if (vp->v_type == VNON) {
1698 		vp->v_type = vtyp;
1699 		if (vp->v_type == VFIFO) {
1700 			extern int (**fifo_nfsv2nodeop_p) __P((void *));
1701 			vp->v_op = fifo_nfsv2nodeop_p;
1702 		} else if (vp->v_type == VREG) {
1703 			mutex_init(&np->n_commitlock, MUTEX_DEFAULT, IPL_NONE);
1704 		} else if (vp->v_type == VCHR || vp->v_type == VBLK) {
1705 			vp->v_op = spec_nfsv2nodeop_p;
1706 			spec_node_init(vp, (dev_t)rdev);
1707 		}
1708 		np->n_mtime = mtime;
1709 	}
1710 	uid = fxdr_unsigned(uid_t, fp->fa_uid);
1711 	gid = fxdr_unsigned(gid_t, fp->fa_gid);
1712 	vap = np->n_vattr;
1713 
1714 	/*
1715 	 * Invalidate access cache if uid, gid, mode or ctime changed.
1716 	 */
1717 	if (np->n_accstamp != -1 &&
1718 	    (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode
1719 	    || timespeccmp(&ctime, &vap->va_ctime, !=)))
1720 		np->n_accstamp = -1;
1721 
1722 	vap->va_type = vtyp;
1723 	vap->va_mode = vmode;
1724 	vap->va_rdev = (dev_t)rdev;
1725 	vap->va_mtime = mtime;
1726 	vap->va_ctime = ctime;
1727 	vap->va_birthtime.tv_sec = VNOVAL;
1728 	vap->va_birthtime.tv_nsec = VNOVAL;
1729 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1730 	switch (vtyp) {
1731 	case VDIR:
1732 		vap->va_blocksize = NFS_DIRFRAGSIZ;
1733 		break;
1734 	case VBLK:
1735 		vap->va_blocksize = BLKDEV_IOSIZE;
1736 		break;
1737 	case VCHR:
1738 		vap->va_blocksize = MAXBSIZE;
1739 		break;
1740 	default:
1741 		vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize :
1742 		    fxdr_unsigned(int32_t, fp->fa2_blocksize);
1743 		break;
1744 	}
1745 	if (v3) {
1746 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1747 		vap->va_uid = uid;
1748 		vap->va_gid = gid;
1749 		vap->va_size = fxdr_hyper(&fp->fa3_size);
1750 		vap->va_bytes = fxdr_hyper(&fp->fa3_used);
1751 		vap->va_fileid = fxdr_hyper(&fp->fa3_fileid);
1752 		fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
1753 		vap->va_flags = 0;
1754 		vap->va_filerev = 0;
1755 	} else {
1756 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1757 		vap->va_uid = uid;
1758 		vap->va_gid = gid;
1759 		vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
1760 		vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks)
1761 		    * NFS_FABLKSIZE;
1762 		vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
1763 		fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
1764 		vap->va_flags = 0;
1765 		vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
1766 		vap->va_filerev = 0;
1767 	}
1768 	if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) {
1769 		return EFBIG;
1770 	}
1771 	if (vap->va_size != np->n_size) {
1772 		if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) {
1773 			vap->va_size = np->n_size;
1774 		} else {
1775 			np->n_size = vap->va_size;
1776 			if (vap->va_type == VREG) {
1777 				/*
1778 				 * we can't free pages if NAC_NOTRUNC because
1779 				 * the pages can be owned by ourselves.
1780 				 */
1781 				if (flags & NAC_NOTRUNC) {
1782 					np->n_flag |= NTRUNCDELAYED;
1783 				} else {
1784 					genfs_node_wrlock(vp);
1785 					mutex_enter(&vp->v_interlock);
1786 					(void)VOP_PUTPAGES(vp, 0,
1787 					    0, PGO_SYNCIO | PGO_CLEANIT |
1788 					    PGO_FREE | PGO_ALLPAGES);
1789 					uvm_vnp_setsize(vp, np->n_size);
1790 					genfs_node_unlock(vp);
1791 				}
1792 			}
1793 		}
1794 	}
1795 	np->n_attrstamp = time_second;
1796 	if (vaper != NULL) {
1797 		memcpy((void *)vaper, (void *)vap, sizeof(*vap));
1798 		if (np->n_flag & NCHG) {
1799 			if (np->n_flag & NACC)
1800 				vaper->va_atime = np->n_atim;
1801 			if (np->n_flag & NUPD)
1802 				vaper->va_mtime = np->n_mtim;
1803 		}
1804 	}
1805 	return (0);
1806 }
1807 
1808 /*
1809  * Check the time stamp
1810  * If the cache is valid, copy contents to *vap and return 0
1811  * otherwise return an error
1812  */
1813 int
1814 nfs_getattrcache(vp, vaper)
1815 	struct vnode *vp;
1816 	struct vattr *vaper;
1817 {
1818 	struct nfsnode *np = VTONFS(vp);
1819 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1820 	struct vattr *vap;
1821 
1822 	if (np->n_attrstamp == 0 ||
1823 	    (time_second - np->n_attrstamp) >= nfs_attrtimeo(nmp, np)) {
1824 		nfsstats.attrcache_misses++;
1825 		return (ENOENT);
1826 	}
1827 	nfsstats.attrcache_hits++;
1828 	vap = np->n_vattr;
1829 	if (vap->va_size != np->n_size) {
1830 		if (vap->va_type == VREG) {
1831 			if ((np->n_flag & NMODIFIED) != 0 &&
1832 			    vap->va_size < np->n_size) {
1833 				vap->va_size = np->n_size;
1834 			} else {
1835 				np->n_size = vap->va_size;
1836 			}
1837 			genfs_node_wrlock(vp);
1838 			uvm_vnp_setsize(vp, np->n_size);
1839 			genfs_node_unlock(vp);
1840 		} else
1841 			np->n_size = vap->va_size;
1842 	}
1843 	memcpy((void *)vaper, (void *)vap, sizeof(struct vattr));
1844 	if (np->n_flag & NCHG) {
1845 		if (np->n_flag & NACC)
1846 			vaper->va_atime = np->n_atim;
1847 		if (np->n_flag & NUPD)
1848 			vaper->va_mtime = np->n_mtim;
1849 	}
1850 	return (0);
1851 }
1852 
1853 void
1854 nfs_delayedtruncate(vp)
1855 	struct vnode *vp;
1856 {
1857 	struct nfsnode *np = VTONFS(vp);
1858 
1859 	if (np->n_flag & NTRUNCDELAYED) {
1860 		np->n_flag &= ~NTRUNCDELAYED;
1861 		genfs_node_wrlock(vp);
1862 		mutex_enter(&vp->v_interlock);
1863 		(void)VOP_PUTPAGES(vp, 0,
1864 		    0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES);
1865 		uvm_vnp_setsize(vp, np->n_size);
1866 		genfs_node_unlock(vp);
1867 	}
1868 }
1869 
1870 #define	NFS_WCCKLUDGE_TIMEOUT	(24 * 60 * 60)	/* 1 day */
1871 #define	NFS_WCCKLUDGE(nmp, now) \
1872 	(((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \
1873 	((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0)
1874 
1875 /*
1876  * nfs_check_wccdata: check inaccurate wcc_data
1877  *
1878  * => return non-zero if we shouldn't trust the wcc_data.
1879  * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed".
1880  */
1881 
1882 int
1883 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime,
1884     struct timespec *mtime, bool docheck)
1885 {
1886 	int error = 0;
1887 
1888 #if !defined(NFS_V2_ONLY)
1889 
1890 	if (docheck) {
1891 		struct vnode *vp = NFSTOV(np);
1892 		struct nfsmount *nmp;
1893 		long now = time_second;
1894 		const struct timespec *omtime = &np->n_vattr->va_mtime;
1895 		const struct timespec *octime = &np->n_vattr->va_ctime;
1896 		const char *reason = NULL; /* XXX: gcc */
1897 
1898 		if (timespeccmp(omtime, mtime, <=)) {
1899 			reason = "mtime";
1900 			error = EINVAL;
1901 		}
1902 
1903 		if (vp->v_type == VDIR && timespeccmp(octime, ctime, <=)) {
1904 			reason = "ctime";
1905 			error = EINVAL;
1906 		}
1907 
1908 		nmp = VFSTONFS(vp->v_mount);
1909 		if (error) {
1910 
1911 			/*
1912 			 * despite of the fact that we've updated the file,
1913 			 * timestamps of the file were not updated as we
1914 			 * expected.
1915 			 * it means that the server has incompatible
1916 			 * semantics of timestamps or (more likely)
1917 			 * the server time is not precise enough to
1918 			 * track each modifications.
1919 			 * in that case, we disable wcc processing.
1920 			 *
1921 			 * yes, strictly speaking, we should disable all
1922 			 * caching.  it's a compromise.
1923 			 */
1924 
1925 			mutex_enter(&nmp->nm_lock);
1926 			if (!NFS_WCCKLUDGE(nmp, now)) {
1927 				printf("%s: inaccurate wcc data (%s) detected,"
1928 				    " disabling wcc"
1929 				    " (ctime %u.%09u %u.%09u,"
1930 				    " mtime %u.%09u %u.%09u)\n",
1931 				    vp->v_mount->mnt_stat.f_mntfromname,
1932 				    reason,
1933 				    (unsigned int)octime->tv_sec,
1934 				    (unsigned int)octime->tv_nsec,
1935 				    (unsigned int)ctime->tv_sec,
1936 				    (unsigned int)ctime->tv_nsec,
1937 				    (unsigned int)omtime->tv_sec,
1938 				    (unsigned int)omtime->tv_nsec,
1939 				    (unsigned int)mtime->tv_sec,
1940 				    (unsigned int)mtime->tv_nsec);
1941 			}
1942 			nmp->nm_iflag |= NFSMNT_WCCKLUDGE;
1943 			nmp->nm_wcckludgetime = now;
1944 			mutex_exit(&nmp->nm_lock);
1945 		} else if (NFS_WCCKLUDGE(nmp, now)) {
1946 			error = EPERM; /* XXX */
1947 		} else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
1948 			mutex_enter(&nmp->nm_lock);
1949 			if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
1950 				printf("%s: re-enabling wcc\n",
1951 				    vp->v_mount->mnt_stat.f_mntfromname);
1952 				nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE;
1953 			}
1954 			mutex_exit(&nmp->nm_lock);
1955 		}
1956 	}
1957 
1958 #endif /* !defined(NFS_V2_ONLY) */
1959 
1960 	return error;
1961 }
1962 
1963 /*
1964  * Heuristic to see if the server XDR encodes directory cookies or not.
1965  * it is not supposed to, but a lot of servers may do this. Also, since
1966  * most/all servers will implement V2 as well, it is expected that they
1967  * may return just 32 bits worth of cookie information, so we need to
1968  * find out in which 32 bits this information is available. We do this
1969  * to avoid trouble with emulated binaries that can't handle 64 bit
1970  * directory offsets.
1971  */
1972 
1973 void
1974 nfs_cookieheuristic(vp, flagp, l, cred)
1975 	struct vnode *vp;
1976 	int *flagp;
1977 	struct lwp *l;
1978 	kauth_cred_t cred;
1979 {
1980 	struct uio auio;
1981 	struct iovec aiov;
1982 	char *tbuf, *cp;
1983 	struct dirent *dp;
1984 	off_t *cookies = NULL, *cop;
1985 	int error, eof, nc, len;
1986 
1987 	MALLOC(tbuf, void *, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK);
1988 
1989 	aiov.iov_base = tbuf;
1990 	aiov.iov_len = NFS_DIRFRAGSIZ;
1991 	auio.uio_iov = &aiov;
1992 	auio.uio_iovcnt = 1;
1993 	auio.uio_rw = UIO_READ;
1994 	auio.uio_resid = NFS_DIRFRAGSIZ;
1995 	auio.uio_offset = 0;
1996 	UIO_SETUP_SYSSPACE(&auio);
1997 
1998 	error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc);
1999 
2000 	len = NFS_DIRFRAGSIZ - auio.uio_resid;
2001 	if (error || len == 0) {
2002 		FREE(tbuf, M_TEMP);
2003 		if (cookies)
2004 			free(cookies, M_TEMP);
2005 		return;
2006 	}
2007 
2008 	/*
2009 	 * Find the first valid entry and look at its offset cookie.
2010 	 */
2011 
2012 	cp = tbuf;
2013 	for (cop = cookies; len > 0; len -= dp->d_reclen) {
2014 		dp = (struct dirent *)cp;
2015 		if (dp->d_fileno != 0 && len >= dp->d_reclen) {
2016 			if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) {
2017 				*flagp |= NFSMNT_SWAPCOOKIE;
2018 				nfs_invaldircache(vp, 0);
2019 				nfs_vinvalbuf(vp, 0, cred, l, 1);
2020 			}
2021 			break;
2022 		}
2023 		cop++;
2024 		cp += dp->d_reclen;
2025 	}
2026 
2027 	FREE(tbuf, M_TEMP);
2028 	free(cookies, M_TEMP);
2029 }
2030 #endif /* NFS */
2031 
2032 #ifdef NFSSERVER
2033 /*
2034  * Set up nameidata for a lookup() call and do it.
2035  *
2036  * If pubflag is set, this call is done for a lookup operation on the
2037  * public filehandle. In that case we allow crossing mountpoints and
2038  * absolute pathnames. However, the caller is expected to check that
2039  * the lookup result is within the public fs, and deny access if
2040  * it is not.
2041  */
2042 int
2043 nfs_namei(ndp, nsfh, len, slp, nam, mdp, dposp, retdirp, l, kerbflag, pubflag)
2044 	struct nameidata *ndp;
2045 	nfsrvfh_t *nsfh;
2046 	uint32_t len;
2047 	struct nfssvc_sock *slp;
2048 	struct mbuf *nam;
2049 	struct mbuf **mdp;
2050 	char **dposp;
2051 	struct vnode **retdirp;
2052 	struct lwp *l;
2053 	int kerbflag, pubflag;
2054 {
2055 	int i, rem;
2056 	struct mbuf *md;
2057 	char *fromcp, *tocp, *cp;
2058 	struct iovec aiov;
2059 	struct uio auio;
2060 	struct vnode *dp;
2061 	int error, rdonly, linklen;
2062 	struct componentname *cnp = &ndp->ni_cnd;
2063 
2064 	*retdirp = NULL;
2065 
2066 	if ((len + 1) > MAXPATHLEN)
2067 		return (ENAMETOOLONG);
2068 	if (len == 0)
2069 		return (EACCES);
2070 	cnp->cn_pnbuf = PNBUF_GET();
2071 
2072 	/*
2073 	 * Copy the name from the mbuf list to ndp->ni_pnbuf
2074 	 * and set the various ndp fields appropriately.
2075 	 */
2076 	fromcp = *dposp;
2077 	tocp = cnp->cn_pnbuf;
2078 	md = *mdp;
2079 	rem = mtod(md, char *) + md->m_len - fromcp;
2080 	for (i = 0; i < len; i++) {
2081 		while (rem == 0) {
2082 			md = md->m_next;
2083 			if (md == NULL) {
2084 				error = EBADRPC;
2085 				goto out;
2086 			}
2087 			fromcp = mtod(md, void *);
2088 			rem = md->m_len;
2089 		}
2090 		if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
2091 			error = EACCES;
2092 			goto out;
2093 		}
2094 		*tocp++ = *fromcp++;
2095 		rem--;
2096 	}
2097 	*tocp = '\0';
2098 	*mdp = md;
2099 	*dposp = fromcp;
2100 	len = nfsm_rndup(len)-len;
2101 	if (len > 0) {
2102 		if (rem >= len)
2103 			*dposp += len;
2104 		else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
2105 			goto out;
2106 	}
2107 
2108 	/*
2109 	 * Extract and set starting directory.
2110 	 */
2111 	error = nfsrv_fhtovp(nsfh, false, &dp, ndp->ni_cnd.cn_cred, slp,
2112 	    nam, &rdonly, kerbflag, pubflag);
2113 	if (error)
2114 		goto out;
2115 	if (dp->v_type != VDIR) {
2116 		vrele(dp);
2117 		error = ENOTDIR;
2118 		goto out;
2119 	}
2120 
2121 	if (rdonly)
2122 		cnp->cn_flags |= RDONLY;
2123 
2124 	*retdirp = dp;
2125 
2126 	if (pubflag) {
2127 		/*
2128 		 * Oh joy. For WebNFS, handle those pesky '%' escapes,
2129 		 * and the 'native path' indicator.
2130 		 */
2131 		cp = PNBUF_GET();
2132 		fromcp = cnp->cn_pnbuf;
2133 		tocp = cp;
2134 		if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
2135 			switch ((unsigned char)*fromcp) {
2136 			case WEBNFS_NATIVE_CHAR:
2137 				/*
2138 				 * 'Native' path for us is the same
2139 				 * as a path according to the NFS spec,
2140 				 * just skip the escape char.
2141 				 */
2142 				fromcp++;
2143 				break;
2144 			/*
2145 			 * More may be added in the future, range 0x80-0xff
2146 			 */
2147 			default:
2148 				error = EIO;
2149 				vrele(dp);
2150 				PNBUF_PUT(cp);
2151 				goto out;
2152 			}
2153 		}
2154 		/*
2155 		 * Translate the '%' escapes, URL-style.
2156 		 */
2157 		while (*fromcp != '\0') {
2158 			if (*fromcp == WEBNFS_ESC_CHAR) {
2159 				if (fromcp[1] != '\0' && fromcp[2] != '\0') {
2160 					fromcp++;
2161 					*tocp++ = HEXSTRTOI(fromcp);
2162 					fromcp += 2;
2163 					continue;
2164 				} else {
2165 					error = ENOENT;
2166 					vrele(dp);
2167 					PNBUF_PUT(cp);
2168 					goto out;
2169 				}
2170 			} else
2171 				*tocp++ = *fromcp++;
2172 		}
2173 		*tocp = '\0';
2174 		PNBUF_PUT(cnp->cn_pnbuf);
2175 		cnp->cn_pnbuf = cp;
2176 	}
2177 
2178 	ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1;
2179 	ndp->ni_segflg = UIO_SYSSPACE;
2180 	ndp->ni_rootdir = rootvnode;
2181 	ndp->ni_erootdir = NULL;
2182 
2183 	if (pubflag) {
2184 		ndp->ni_loopcnt = 0;
2185 		if (cnp->cn_pnbuf[0] == '/')
2186 			dp = rootvnode;
2187 	} else {
2188 		cnp->cn_flags |= NOCROSSMOUNT;
2189 	}
2190 
2191 	VREF(dp);
2192 	vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
2193 
2194     for (;;) {
2195 	cnp->cn_nameptr = cnp->cn_pnbuf;
2196 	ndp->ni_startdir = dp;
2197 
2198 	/*
2199 	 * And call lookup() to do the real work
2200 	 */
2201 	error = lookup(ndp);
2202 	if (error) {
2203 		if (ndp->ni_dvp) {
2204 			vput(ndp->ni_dvp);
2205 		}
2206 		PNBUF_PUT(cnp->cn_pnbuf);
2207 		return (error);
2208 	}
2209 
2210 	/*
2211 	 * Check for encountering a symbolic link
2212 	 */
2213 	if ((cnp->cn_flags & ISSYMLINK) == 0) {
2214 		if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp) {
2215 			if (ndp->ni_dvp == ndp->ni_vp) {
2216 				vrele(ndp->ni_dvp);
2217 			} else {
2218 				vput(ndp->ni_dvp);
2219 			}
2220 		}
2221 		if (cnp->cn_flags & (SAVENAME | SAVESTART))
2222 			cnp->cn_flags |= HASBUF;
2223 		else
2224 			PNBUF_PUT(cnp->cn_pnbuf);
2225 		return (0);
2226 	} else {
2227 		if (!pubflag) {
2228 			error = EINVAL;
2229 			break;
2230 		}
2231 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
2232 			error = ELOOP;
2233 			break;
2234 		}
2235 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
2236 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred);
2237 			if (error != 0)
2238 				break;
2239 		}
2240 		if (ndp->ni_pathlen > 1)
2241 			cp = PNBUF_GET();
2242 		else
2243 			cp = cnp->cn_pnbuf;
2244 		aiov.iov_base = cp;
2245 		aiov.iov_len = MAXPATHLEN;
2246 		auio.uio_iov = &aiov;
2247 		auio.uio_iovcnt = 1;
2248 		auio.uio_offset = 0;
2249 		auio.uio_rw = UIO_READ;
2250 		auio.uio_resid = MAXPATHLEN;
2251 		UIO_SETUP_SYSSPACE(&auio);
2252 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
2253 		if (error) {
2254 badlink:
2255 			if (ndp->ni_pathlen > 1)
2256 				PNBUF_PUT(cp);
2257 			break;
2258 		}
2259 		linklen = MAXPATHLEN - auio.uio_resid;
2260 		if (linklen == 0) {
2261 			error = ENOENT;
2262 			goto badlink;
2263 		}
2264 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
2265 			error = ENAMETOOLONG;
2266 			goto badlink;
2267 		}
2268 		if (ndp->ni_pathlen > 1) {
2269 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
2270 			PNBUF_PUT(cnp->cn_pnbuf);
2271 			cnp->cn_pnbuf = cp;
2272 		} else
2273 			cnp->cn_pnbuf[linklen] = '\0';
2274 		ndp->ni_pathlen += linklen;
2275 		vput(ndp->ni_vp);
2276 		dp = ndp->ni_dvp;
2277 
2278 		/*
2279 		 * Check if root directory should replace current directory.
2280 		 */
2281 		if (cnp->cn_pnbuf[0] == '/') {
2282 			vput(dp);
2283 			dp = ndp->ni_rootdir;
2284 			VREF(dp);
2285 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
2286 		}
2287 	}
2288    }
2289 	vput(ndp->ni_dvp);
2290 	vput(ndp->ni_vp);
2291 	ndp->ni_vp = NULL;
2292 out:
2293 	PNBUF_PUT(cnp->cn_pnbuf);
2294 	return (error);
2295 }
2296 #endif /* NFSSERVER */
2297 
2298 /*
2299  * A fiddled version of m_adj() that ensures null fill to a 32-bit
2300  * boundary and only trims off the back end
2301  *
2302  * 1. trim off 'len' bytes as m_adj(mp, -len).
2303  * 2. add zero-padding 'nul' bytes at the end of the mbuf chain.
2304  */
2305 void
2306 nfs_zeropad(mp, len, nul)
2307 	struct mbuf *mp;
2308 	int len;
2309 	int nul;
2310 {
2311 	struct mbuf *m;
2312 	int count;
2313 
2314 	/*
2315 	 * Trim from tail.  Scan the mbuf chain,
2316 	 * calculating its length and finding the last mbuf.
2317 	 * If the adjustment only affects this mbuf, then just
2318 	 * adjust and return.  Otherwise, rescan and truncate
2319 	 * after the remaining size.
2320 	 */
2321 	count = 0;
2322 	m = mp;
2323 	for (;;) {
2324 		count += m->m_len;
2325 		if (m->m_next == NULL)
2326 			break;
2327 		m = m->m_next;
2328 	}
2329 
2330 	KDASSERT(count >= len);
2331 
2332 	if (m->m_len >= len) {
2333 		m->m_len -= len;
2334 	} else {
2335 		count -= len;
2336 		/*
2337 		 * Correct length for chain is "count".
2338 		 * Find the mbuf with last data, adjust its length,
2339 		 * and toss data from remaining mbufs on chain.
2340 		 */
2341 		for (m = mp; m; m = m->m_next) {
2342 			if (m->m_len >= count) {
2343 				m->m_len = count;
2344 				break;
2345 			}
2346 			count -= m->m_len;
2347 		}
2348 		KASSERT(m && m->m_next);
2349 		m_freem(m->m_next);
2350 		m->m_next = NULL;
2351 	}
2352 
2353 	KDASSERT(m->m_next == NULL);
2354 
2355 	/*
2356 	 * zero-padding.
2357 	 */
2358 	if (nul > 0) {
2359 		char *cp;
2360 		int i;
2361 
2362 		if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) {
2363 			struct mbuf *n;
2364 
2365 			KDASSERT(MLEN >= nul);
2366 			n = m_get(M_WAIT, MT_DATA);
2367 			MCLAIM(n, &nfs_mowner);
2368 			n->m_len = nul;
2369 			n->m_next = NULL;
2370 			m->m_next = n;
2371 			cp = mtod(n, void *);
2372 		} else {
2373 			cp = mtod(m, char *) + m->m_len;
2374 			m->m_len += nul;
2375 		}
2376 		for (i = 0; i < nul; i++)
2377 			*cp++ = '\0';
2378 	}
2379 	return;
2380 }
2381 
2382 /*
2383  * Make these functions instead of macros, so that the kernel text size
2384  * doesn't get too big...
2385  */
2386 void
2387 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp)
2388 	struct nfsrv_descript *nfsd;
2389 	int before_ret;
2390 	struct vattr *before_vap;
2391 	int after_ret;
2392 	struct vattr *after_vap;
2393 	struct mbuf **mbp;
2394 	char **bposp;
2395 {
2396 	struct mbuf *mb = *mbp;
2397 	char *bpos = *bposp;
2398 	u_int32_t *tl;
2399 
2400 	if (before_ret) {
2401 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2402 		*tl = nfs_false;
2403 	} else {
2404 		nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
2405 		*tl++ = nfs_true;
2406 		txdr_hyper(before_vap->va_size, tl);
2407 		tl += 2;
2408 		txdr_nfsv3time(&(before_vap->va_mtime), tl);
2409 		tl += 2;
2410 		txdr_nfsv3time(&(before_vap->va_ctime), tl);
2411 	}
2412 	*bposp = bpos;
2413 	*mbp = mb;
2414 	nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
2415 }
2416 
2417 void
2418 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp)
2419 	struct nfsrv_descript *nfsd;
2420 	int after_ret;
2421 	struct vattr *after_vap;
2422 	struct mbuf **mbp;
2423 	char **bposp;
2424 {
2425 	struct mbuf *mb = *mbp;
2426 	char *bpos = *bposp;
2427 	u_int32_t *tl;
2428 	struct nfs_fattr *fp;
2429 
2430 	if (after_ret) {
2431 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2432 		*tl = nfs_false;
2433 	} else {
2434 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
2435 		*tl++ = nfs_true;
2436 		fp = (struct nfs_fattr *)tl;
2437 		nfsm_srvfattr(nfsd, after_vap, fp);
2438 	}
2439 	*mbp = mb;
2440 	*bposp = bpos;
2441 }
2442 
2443 void
2444 nfsm_srvfattr(nfsd, vap, fp)
2445 	struct nfsrv_descript *nfsd;
2446 	struct vattr *vap;
2447 	struct nfs_fattr *fp;
2448 {
2449 
2450 	fp->fa_nlink = txdr_unsigned(vap->va_nlink);
2451 	fp->fa_uid = txdr_unsigned(vap->va_uid);
2452 	fp->fa_gid = txdr_unsigned(vap->va_gid);
2453 	if (nfsd->nd_flag & ND_NFSV3) {
2454 		fp->fa_type = vtonfsv3_type(vap->va_type);
2455 		fp->fa_mode = vtonfsv3_mode(vap->va_mode);
2456 		txdr_hyper(vap->va_size, &fp->fa3_size);
2457 		txdr_hyper(vap->va_bytes, &fp->fa3_used);
2458 		fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
2459 		fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
2460 		fp->fa3_fsid.nfsuquad[0] = 0;
2461 		fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
2462 		txdr_hyper(vap->va_fileid, &fp->fa3_fileid);
2463 		txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
2464 		txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
2465 		txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
2466 	} else {
2467 		fp->fa_type = vtonfsv2_type(vap->va_type);
2468 		fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
2469 		fp->fa2_size = txdr_unsigned(vap->va_size);
2470 		fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
2471 		if (vap->va_type == VFIFO)
2472 			fp->fa2_rdev = 0xffffffff;
2473 		else
2474 			fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
2475 		fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
2476 		fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
2477 		fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
2478 		txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
2479 		txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
2480 		txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
2481 	}
2482 }
2483 
2484 #ifdef NFSSERVER
2485 /*
2486  * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
2487  * 	- look up fsid in mount list (if not found ret error)
2488  *	- get vp and export rights by calling VFS_FHTOVP()
2489  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2490  *	- if not lockflag unlock it with VOP_UNLOCK()
2491  */
2492 int
2493 nfsrv_fhtovp(nfsrvfh_t *nsfh, int lockflag, struct vnode **vpp,
2494     kauth_cred_t cred, struct nfssvc_sock *slp, struct mbuf *nam, int *rdonlyp,
2495     int kerbflag, int pubflag)
2496 {
2497 	struct mount *mp;
2498 	kauth_cred_t credanon;
2499 	int error, exflags;
2500 	struct sockaddr_in *saddr;
2501 	fhandle_t *fhp;
2502 
2503 	fhp = NFSRVFH_FHANDLE(nsfh);
2504 	*vpp = (struct vnode *)0;
2505 
2506 	if (nfs_ispublicfh(nsfh)) {
2507 		if (!pubflag || !nfs_pub.np_valid)
2508 			return (ESTALE);
2509 		fhp = nfs_pub.np_handle;
2510 	}
2511 
2512 	error = netexport_check(&fhp->fh_fsid, nam, &mp, &exflags, &credanon);
2513 	if (error) {
2514 		return error;
2515 	}
2516 
2517 	error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
2518 	if (error)
2519 		return (error);
2520 
2521 	if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
2522 		saddr = mtod(nam, struct sockaddr_in *);
2523 		if ((saddr->sin_family == AF_INET) &&
2524 		    ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
2525 			vput(*vpp);
2526 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2527 		}
2528 #ifdef INET6
2529 		if ((saddr->sin_family == AF_INET6) &&
2530 		    ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) {
2531 			vput(*vpp);
2532 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2533 		}
2534 #endif
2535 	}
2536 	/*
2537 	 * Check/setup credentials.
2538 	 */
2539 	if (exflags & MNT_EXKERB) {
2540 		if (!kerbflag) {
2541 			vput(*vpp);
2542 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2543 		}
2544 	} else if (kerbflag) {
2545 		vput(*vpp);
2546 		return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2547 	} else if (kauth_cred_geteuid(cred) == 0 || /* NFS maproot, see below */
2548 	    (exflags & MNT_EXPORTANON)) {
2549 		/*
2550 		 * This is used by the NFS maproot option. While we can change
2551 		 * the secmodel on our own host, we can't change it on the
2552 		 * clients. As means of least surprise, we're doing the
2553 		 * traditional thing here.
2554 		 * Should look into adding a "mapprivileged" or similar where
2555 		 * the users can be explicitly specified...
2556 		 * [elad, yamt 2008-03-05]
2557 		 */
2558 		kauth_cred_clone(credanon, cred);
2559 	}
2560 	if (exflags & MNT_EXRDONLY)
2561 		*rdonlyp = 1;
2562 	else
2563 		*rdonlyp = 0;
2564 	if (!lockflag)
2565 		VOP_UNLOCK(*vpp, 0);
2566 	return (0);
2567 }
2568 
2569 /*
2570  * WebNFS: check if a filehandle is a public filehandle. For v3, this
2571  * means a length of 0, for v2 it means all zeroes.
2572  */
2573 int
2574 nfs_ispublicfh(const nfsrvfh_t *nsfh)
2575 {
2576 	const char *cp = (const void *)(NFSRVFH_DATA(nsfh));
2577 	int i;
2578 
2579 	if (NFSRVFH_SIZE(nsfh) == 0) {
2580 		return true;
2581 	}
2582 	if (NFSRVFH_SIZE(nsfh) != NFSX_V2FH) {
2583 		return false;
2584 	}
2585 	for (i = 0; i < NFSX_V2FH; i++)
2586 		if (*cp++ != 0)
2587 			return false;
2588 	return true;
2589 }
2590 #endif /* NFSSERVER */
2591 
2592 /*
2593  * This function compares two net addresses by family and returns true
2594  * if they are the same host.
2595  * If there is any doubt, return false.
2596  * The AF_INET family is handled as a special case so that address mbufs
2597  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
2598  */
2599 int
2600 netaddr_match(family, haddr, nam)
2601 	int family;
2602 	union nethostaddr *haddr;
2603 	struct mbuf *nam;
2604 {
2605 	struct sockaddr_in *inetaddr;
2606 
2607 	switch (family) {
2608 	case AF_INET:
2609 		inetaddr = mtod(nam, struct sockaddr_in *);
2610 		if (inetaddr->sin_family == AF_INET &&
2611 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
2612 			return (1);
2613 		break;
2614 #ifdef INET6
2615 	case AF_INET6:
2616 	    {
2617 		struct sockaddr_in6 *sin6_1, *sin6_2;
2618 
2619 		sin6_1 = mtod(nam, struct sockaddr_in6 *);
2620 		sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
2621 		if (sin6_1->sin6_family == AF_INET6 &&
2622 		    IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
2623 			return 1;
2624 	    }
2625 #endif
2626 #ifdef ISO
2627 	case AF_ISO:
2628 	    {
2629 		struct sockaddr_iso *isoaddr1, *isoaddr2;
2630 
2631 		isoaddr1 = mtod(nam, struct sockaddr_iso *);
2632 		isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *);
2633 		if (isoaddr1->siso_family == AF_ISO &&
2634 		    isoaddr1->siso_nlen > 0 &&
2635 		    isoaddr1->siso_nlen == isoaddr2->siso_nlen &&
2636 		    SAME_ISOADDR(isoaddr1, isoaddr2))
2637 			return (1);
2638 		break;
2639 	    }
2640 #endif	/* ISO */
2641 	default:
2642 		break;
2643 	};
2644 	return (0);
2645 }
2646 
2647 /*
2648  * The write verifier has changed (probably due to a server reboot), so all
2649  * PG_NEEDCOMMIT pages will have to be written again. Since they are marked
2650  * as dirty or are being written out just now, all this takes is clearing
2651  * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for
2652  * the mount point.
2653  */
2654 void
2655 nfs_clearcommit(mp)
2656 	struct mount *mp;
2657 {
2658 	struct vnode *vp;
2659 	struct nfsnode *np;
2660 	struct vm_page *pg;
2661 	struct nfsmount *nmp = VFSTONFS(mp);
2662 
2663 	rw_enter(&nmp->nm_writeverflock, RW_WRITER);
2664 	mutex_enter(&mntvnode_lock);
2665 	TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2666 		KASSERT(vp->v_mount == mp);
2667 		if (vp->v_type != VREG)
2668 			continue;
2669 		np = VTONFS(vp);
2670 		np->n_pushlo = np->n_pushhi = np->n_pushedlo =
2671 		    np->n_pushedhi = 0;
2672 		np->n_commitflags &=
2673 		    ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
2674 		mutex_enter(&vp->v_uobj.vmobjlock);
2675 		TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq.queue) {
2676 			pg->flags &= ~PG_NEEDCOMMIT;
2677 		}
2678 		mutex_exit(&vp->v_uobj.vmobjlock);
2679 	}
2680 	mutex_exit(&mntvnode_lock);
2681 	mutex_enter(&nmp->nm_lock);
2682 	nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF;
2683 	mutex_exit(&nmp->nm_lock);
2684 	rw_exit(&nmp->nm_writeverflock);
2685 }
2686 
2687 void
2688 nfs_merge_commit_ranges(vp)
2689 	struct vnode *vp;
2690 {
2691 	struct nfsnode *np = VTONFS(vp);
2692 
2693 	KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID);
2694 
2695 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2696 		np->n_pushedlo = np->n_pushlo;
2697 		np->n_pushedhi = np->n_pushhi;
2698 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2699 	} else {
2700 		if (np->n_pushlo < np->n_pushedlo)
2701 			np->n_pushedlo = np->n_pushlo;
2702 		if (np->n_pushhi > np->n_pushedhi)
2703 			np->n_pushedhi = np->n_pushhi;
2704 	}
2705 
2706 	np->n_pushlo = np->n_pushhi = 0;
2707 	np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
2708 
2709 #ifdef NFS_DEBUG_COMMIT
2710 	printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2711 	    (unsigned)np->n_pushedhi);
2712 #endif
2713 }
2714 
2715 int
2716 nfs_in_committed_range(vp, off, len)
2717 	struct vnode *vp;
2718 	off_t off, len;
2719 {
2720 	struct nfsnode *np = VTONFS(vp);
2721 	off_t lo, hi;
2722 
2723 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2724 		return 0;
2725 	lo = off;
2726 	hi = lo + len;
2727 
2728 	return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
2729 }
2730 
2731 int
2732 nfs_in_tobecommitted_range(vp, off, len)
2733 	struct vnode *vp;
2734 	off_t off, len;
2735 {
2736 	struct nfsnode *np = VTONFS(vp);
2737 	off_t lo, hi;
2738 
2739 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2740 		return 0;
2741 	lo = off;
2742 	hi = lo + len;
2743 
2744 	return (lo >= np->n_pushlo && hi <= np->n_pushhi);
2745 }
2746 
2747 void
2748 nfs_add_committed_range(vp, off, len)
2749 	struct vnode *vp;
2750 	off_t off, len;
2751 {
2752 	struct nfsnode *np = VTONFS(vp);
2753 	off_t lo, hi;
2754 
2755 	lo = off;
2756 	hi = lo + len;
2757 
2758 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2759 		np->n_pushedlo = lo;
2760 		np->n_pushedhi = hi;
2761 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2762 	} else {
2763 		if (hi > np->n_pushedhi)
2764 			np->n_pushedhi = hi;
2765 		if (lo < np->n_pushedlo)
2766 			np->n_pushedlo = lo;
2767 	}
2768 #ifdef NFS_DEBUG_COMMIT
2769 	printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2770 	    (unsigned)np->n_pushedhi);
2771 #endif
2772 }
2773 
2774 void
2775 nfs_del_committed_range(vp, off, len)
2776 	struct vnode *vp;
2777 	off_t off, len;
2778 {
2779 	struct nfsnode *np = VTONFS(vp);
2780 	off_t lo, hi;
2781 
2782 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2783 		return;
2784 
2785 	lo = off;
2786 	hi = lo + len;
2787 
2788 	if (lo > np->n_pushedhi || hi < np->n_pushedlo)
2789 		return;
2790 	if (lo <= np->n_pushedlo)
2791 		np->n_pushedlo = hi;
2792 	else if (hi >= np->n_pushedhi)
2793 		np->n_pushedhi = lo;
2794 	else {
2795 		/*
2796 		 * XXX There's only one range. If the deleted range
2797 		 * is in the middle, pick the largest of the
2798 		 * contiguous ranges that it leaves.
2799 		 */
2800 		if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
2801 			np->n_pushedhi = lo;
2802 		else
2803 			np->n_pushedlo = hi;
2804 	}
2805 #ifdef NFS_DEBUG_COMMIT
2806 	printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2807 	    (unsigned)np->n_pushedhi);
2808 #endif
2809 }
2810 
2811 void
2812 nfs_add_tobecommitted_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 	lo = off;
2820 	hi = lo + len;
2821 
2822 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
2823 		np->n_pushlo = lo;
2824 		np->n_pushhi = hi;
2825 		np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
2826 	} else {
2827 		if (lo < np->n_pushlo)
2828 			np->n_pushlo = lo;
2829 		if (hi > np->n_pushhi)
2830 			np->n_pushhi = hi;
2831 	}
2832 #ifdef NFS_DEBUG_COMMIT
2833 	printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2834 	    (unsigned)np->n_pushhi);
2835 #endif
2836 }
2837 
2838 void
2839 nfs_del_tobecommitted_range(vp, off, len)
2840 	struct vnode *vp;
2841 	off_t off, len;
2842 {
2843 	struct nfsnode *np = VTONFS(vp);
2844 	off_t lo, hi;
2845 
2846 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2847 		return;
2848 
2849 	lo = off;
2850 	hi = lo + len;
2851 
2852 	if (lo > np->n_pushhi || hi < np->n_pushlo)
2853 		return;
2854 
2855 	if (lo <= np->n_pushlo)
2856 		np->n_pushlo = hi;
2857 	else if (hi >= np->n_pushhi)
2858 		np->n_pushhi = lo;
2859 	else {
2860 		/*
2861 		 * XXX There's only one range. If the deleted range
2862 		 * is in the middle, pick the largest of the
2863 		 * contiguous ranges that it leaves.
2864 		 */
2865 		if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
2866 			np->n_pushhi = lo;
2867 		else
2868 			np->n_pushlo = hi;
2869 	}
2870 #ifdef NFS_DEBUG_COMMIT
2871 	printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2872 	    (unsigned)np->n_pushhi);
2873 #endif
2874 }
2875 
2876 /*
2877  * Map errnos to NFS error numbers. For Version 3 also filter out error
2878  * numbers not specified for the associated procedure.
2879  */
2880 int
2881 nfsrv_errmap(nd, err)
2882 	struct nfsrv_descript *nd;
2883 	int err;
2884 {
2885 	const short *defaulterrp, *errp;
2886 
2887 	if (nd->nd_flag & ND_NFSV3) {
2888 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
2889 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2890 		while (*++errp) {
2891 			if (*errp == err)
2892 				return (err);
2893 			else if (*errp > err)
2894 				break;
2895 		}
2896 		return ((int)*defaulterrp);
2897 	    } else
2898 		return (err & 0xffff);
2899 	}
2900 	if (err <= ELAST)
2901 		return ((int)nfsrv_v2errmap[err - 1]);
2902 	return (NFSERR_IO);
2903 }
2904 
2905 u_int32_t
2906 nfs_getxid()
2907 {
2908 	static u_int32_t base;
2909 	static u_int32_t nfs_xid = 0;
2910 	static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER;
2911 	u_int32_t newxid;
2912 
2913 	simple_lock(&nfs_xidlock);
2914 	/*
2915 	 * derive initial xid from system time
2916 	 * XXX time is invalid if root not yet mounted
2917 	 */
2918 	if (__predict_false(!base && (rootvp))) {
2919 		struct timeval tv;
2920 
2921 		microtime(&tv);
2922 		base = tv.tv_sec << 12;
2923 		nfs_xid = base;
2924 	}
2925 
2926 	/*
2927 	 * Skip zero xid if it should ever happen.
2928 	 */
2929 	if (__predict_false(++nfs_xid == 0))
2930 		nfs_xid++;
2931 	newxid = nfs_xid;
2932 	simple_unlock(&nfs_xidlock);
2933 
2934 	return txdr_unsigned(newxid);
2935 }
2936 
2937 /*
2938  * assign a new xid for existing request.
2939  * used for NFSERR_JUKEBOX handling.
2940  */
2941 void
2942 nfs_renewxid(struct nfsreq *req)
2943 {
2944 	u_int32_t xid;
2945 	int off;
2946 
2947 	xid = nfs_getxid();
2948 	if (req->r_nmp->nm_sotype == SOCK_STREAM)
2949 		off = sizeof(u_int32_t); /* RPC record mark */
2950 	else
2951 		off = 0;
2952 
2953 	m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid);
2954 	req->r_xid = xid;
2955 }
2956 
2957 #if defined(NFSSERVER)
2958 int
2959 nfsrv_composefh(struct vnode *vp, nfsrvfh_t *nsfh, bool v3)
2960 {
2961 	int error;
2962 	size_t fhsize;
2963 
2964 	fhsize = NFSD_MAXFHSIZE;
2965 	error = vfs_composefh(vp, (void *)NFSRVFH_DATA(nsfh), &fhsize);
2966 	if (NFSX_FHTOOBIG_P(fhsize, v3)) {
2967 		error = EOPNOTSUPP;
2968 	}
2969 	if (error != 0) {
2970 		return error;
2971 	}
2972 	if (!v3 && fhsize < NFSX_V2FH) {
2973 		memset((char *)NFSRVFH_DATA(nsfh) + fhsize, 0,
2974 		    NFSX_V2FH - fhsize);
2975 		fhsize = NFSX_V2FH;
2976 	}
2977 	if ((fhsize % NFSX_UNSIGNED) != 0) {
2978 		return EOPNOTSUPP;
2979 	}
2980 	nsfh->nsfh_size = fhsize;
2981 	return 0;
2982 }
2983 
2984 int
2985 nfsrv_comparefh(const nfsrvfh_t *fh1, const nfsrvfh_t *fh2)
2986 {
2987 
2988 	if (NFSRVFH_SIZE(fh1) != NFSRVFH_SIZE(fh2)) {
2989 		return NFSRVFH_SIZE(fh2) - NFSRVFH_SIZE(fh1);
2990 	}
2991 	return memcmp(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), NFSRVFH_SIZE(fh1));
2992 }
2993 
2994 void
2995 nfsrv_copyfh(nfsrvfh_t *fh1, const nfsrvfh_t *fh2)
2996 {
2997 	size_t size;
2998 
2999 	fh1->nsfh_size = size = NFSRVFH_SIZE(fh2);
3000 	memcpy(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), size);
3001 }
3002 #endif /* defined(NFSSERVER) */
3003 
3004 #if defined(NFS)
3005 /*
3006  * Set the attribute timeout based on how recently the file has been modified.
3007  */
3008 
3009 time_t
3010 nfs_attrtimeo(struct nfsmount *nmp, struct nfsnode *np)
3011 {
3012 	time_t timeo;
3013 
3014 	if ((nmp->nm_flag & NFSMNT_NOAC) != 0)
3015 		return 0;
3016 
3017 	if (((np)->n_flag & NMODIFIED) != 0)
3018 		return NFS_MINATTRTIMO;
3019 
3020 	timeo = (time_second - np->n_mtime.tv_sec) / 10;
3021 	timeo = max(timeo, NFS_MINATTRTIMO);
3022 	timeo = min(timeo, NFS_MAXATTRTIMO);
3023 	return timeo;
3024 }
3025 #endif /* defined(NFS) */
3026