xref: /netbsd-src/sys/nfs/nfs_subs.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: nfs_subs.c,v 1.194 2007/12/08 19:29:51 pooka 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 1.194 2007/12/08 19:29:51 pooka Exp $");
74 
75 #include "fs_nfs.h"
76 #include "opt_nfs.h"
77 #include "opt_nfsserver.h"
78 #include "opt_iso.h"
79 #include "opt_inet.h"
80 
81 /*
82  * These functions support the macros and help fiddle mbuf chains for
83  * the nfs op functions. They do things like create the rpc header and
84  * copy data between mbuf chains and uio lists.
85  */
86 #include <sys/param.h>
87 #include <sys/proc.h>
88 #include <sys/systm.h>
89 #include <sys/kernel.h>
90 #include <sys/mount.h>
91 #include <sys/vnode.h>
92 #include <sys/namei.h>
93 #include <sys/mbuf.h>
94 #include <sys/socket.h>
95 #include <sys/stat.h>
96 #include <sys/malloc.h>
97 #include <sys/filedesc.h>
98 #include <sys/time.h>
99 #include <sys/dirent.h>
100 #include <sys/once.h>
101 #include <sys/kauth.h>
102 
103 #include <uvm/uvm_extern.h>
104 
105 #include <nfs/rpcv2.h>
106 #include <nfs/nfsproto.h>
107 #include <nfs/nfsnode.h>
108 #include <nfs/nfs.h>
109 #include <nfs/xdr_subs.h>
110 #include <nfs/nfsm_subs.h>
111 #include <nfs/nfsmount.h>
112 #include <nfs/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 nfsrtt nfsrtt;
567 extern struct nfsnodehashhead *nfsnodehashtbl;
568 extern u_long nfsnodehash;
569 
570 u_long nfsdirhashmask;
571 
572 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *));
573 
574 /*
575  * Create the header for an rpc request packet
576  * The hsiz is the size of the rest of the nfs request header.
577  * (just used to decide if a cluster is a good idea)
578  */
579 struct mbuf *
580 nfsm_reqh(struct nfsnode *np, u_long procid, int hsiz, char **bposp)
581 {
582 	struct mbuf *mb;
583 	char *bpos;
584 
585 	mb = m_get(M_WAIT, MT_DATA);
586 	MCLAIM(mb, &nfs_mowner);
587 	if (hsiz >= MINCLSIZE)
588 		m_clget(mb, M_WAIT);
589 	mb->m_len = 0;
590 	bpos = mtod(mb, void *);
591 
592 	/* Finally, return values */
593 	*bposp = bpos;
594 	return (mb);
595 }
596 
597 /*
598  * Build the RPC header and fill in the authorization info.
599  * The authorization string argument is only used when the credentials
600  * come from outside of the kernel.
601  * Returns the head of the mbuf list.
602  */
603 struct mbuf *
604 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len,
605 	verf_str, mrest, mrest_len, mbp, xidp)
606 	kauth_cred_t cr;
607 	int nmflag;
608 	int procid;
609 	int auth_type;
610 	int auth_len;
611 	char *auth_str;
612 	int verf_len;
613 	char *verf_str;
614 	struct mbuf *mrest;
615 	int mrest_len;
616 	struct mbuf **mbp;
617 	u_int32_t *xidp;
618 {
619 	struct mbuf *mb;
620 	u_int32_t *tl;
621 	char *bpos;
622 	int i;
623 	struct mbuf *mreq;
624 	int siz, grpsiz, authsiz;
625 
626 	authsiz = nfsm_rndup(auth_len);
627 	mb = m_gethdr(M_WAIT, MT_DATA);
628 	MCLAIM(mb, &nfs_mowner);
629 	if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
630 		m_clget(mb, M_WAIT);
631 	} else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
632 		MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED);
633 	} else {
634 		MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
635 	}
636 	mb->m_len = 0;
637 	mreq = mb;
638 	bpos = mtod(mb, void *);
639 
640 	/*
641 	 * First the RPC header.
642 	 */
643 	nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
644 
645 	*tl++ = *xidp = nfs_getxid();
646 	*tl++ = rpc_call;
647 	*tl++ = rpc_vers;
648 	*tl++ = txdr_unsigned(NFS_PROG);
649 	if (nmflag & NFSMNT_NFSV3)
650 		*tl++ = txdr_unsigned(NFS_VER3);
651 	else
652 		*tl++ = txdr_unsigned(NFS_VER2);
653 	if (nmflag & NFSMNT_NFSV3)
654 		*tl++ = txdr_unsigned(procid);
655 	else
656 		*tl++ = txdr_unsigned(nfsv2_procid[procid]);
657 
658 	/*
659 	 * And then the authorization cred.
660 	 */
661 	*tl++ = txdr_unsigned(auth_type);
662 	*tl = txdr_unsigned(authsiz);
663 	switch (auth_type) {
664 	case RPCAUTH_UNIX:
665 		nfsm_build(tl, u_int32_t *, auth_len);
666 		*tl++ = 0;		/* stamp ?? */
667 		*tl++ = 0;		/* NULL hostname */
668 		*tl++ = txdr_unsigned(kauth_cred_geteuid(cr));
669 		*tl++ = txdr_unsigned(kauth_cred_getegid(cr));
670 		grpsiz = (auth_len >> 2) - 5;
671 		*tl++ = txdr_unsigned(grpsiz);
672 		for (i = 0; i < grpsiz; i++)
673 			*tl++ = txdr_unsigned(kauth_cred_group(cr, i)); /* XXX elad review */
674 		break;
675 	case RPCAUTH_KERB4:
676 		siz = auth_len;
677 		while (siz > 0) {
678 			if (M_TRAILINGSPACE(mb) == 0) {
679 				struct mbuf *mb2;
680 				mb2 = m_get(M_WAIT, MT_DATA);
681 				MCLAIM(mb2, &nfs_mowner);
682 				if (siz >= MINCLSIZE)
683 					m_clget(mb2, M_WAIT);
684 				mb->m_next = mb2;
685 				mb = mb2;
686 				mb->m_len = 0;
687 				bpos = mtod(mb, void *);
688 			}
689 			i = min(siz, M_TRAILINGSPACE(mb));
690 			memcpy(bpos, auth_str, i);
691 			mb->m_len += i;
692 			auth_str += i;
693 			bpos += i;
694 			siz -= i;
695 		}
696 		if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
697 			for (i = 0; i < siz; i++)
698 				*bpos++ = '\0';
699 			mb->m_len += siz;
700 		}
701 		break;
702 	};
703 
704 	/*
705 	 * And the verifier...
706 	 */
707 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
708 	if (verf_str) {
709 		*tl++ = txdr_unsigned(RPCAUTH_KERB4);
710 		*tl = txdr_unsigned(verf_len);
711 		siz = verf_len;
712 		while (siz > 0) {
713 			if (M_TRAILINGSPACE(mb) == 0) {
714 				struct mbuf *mb2;
715 				mb2 = m_get(M_WAIT, MT_DATA);
716 				MCLAIM(mb2, &nfs_mowner);
717 				if (siz >= MINCLSIZE)
718 					m_clget(mb2, M_WAIT);
719 				mb->m_next = mb2;
720 				mb = mb2;
721 				mb->m_len = 0;
722 				bpos = mtod(mb, void *);
723 			}
724 			i = min(siz, M_TRAILINGSPACE(mb));
725 			memcpy(bpos, verf_str, i);
726 			mb->m_len += i;
727 			verf_str += i;
728 			bpos += i;
729 			siz -= i;
730 		}
731 		if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
732 			for (i = 0; i < siz; i++)
733 				*bpos++ = '\0';
734 			mb->m_len += siz;
735 		}
736 	} else {
737 		*tl++ = txdr_unsigned(RPCAUTH_NULL);
738 		*tl = 0;
739 	}
740 	mb->m_next = mrest;
741 	mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
742 	mreq->m_pkthdr.rcvif = (struct ifnet *)0;
743 	*mbp = mb;
744 	return (mreq);
745 }
746 
747 /*
748  * copies mbuf chain to the uio scatter/gather list
749  */
750 int
751 nfsm_mbuftouio(mrep, uiop, siz, dpos)
752 	struct mbuf **mrep;
753 	struct uio *uiop;
754 	int siz;
755 	char **dpos;
756 {
757 	char *mbufcp, *uiocp;
758 	int xfer, left, len;
759 	struct mbuf *mp;
760 	long uiosiz, rem;
761 	int error = 0;
762 
763 	mp = *mrep;
764 	mbufcp = *dpos;
765 	len = mtod(mp, char *) + mp->m_len - mbufcp;
766 	rem = nfsm_rndup(siz)-siz;
767 	while (siz > 0) {
768 		if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
769 			return (EFBIG);
770 		left = uiop->uio_iov->iov_len;
771 		uiocp = uiop->uio_iov->iov_base;
772 		if (left > siz)
773 			left = siz;
774 		uiosiz = left;
775 		while (left > 0) {
776 			while (len == 0) {
777 				mp = mp->m_next;
778 				if (mp == NULL)
779 					return (EBADRPC);
780 				mbufcp = mtod(mp, void *);
781 				len = mp->m_len;
782 			}
783 			xfer = (left > len) ? len : left;
784 			error = copyout_vmspace(uiop->uio_vmspace, mbufcp,
785 			    uiocp, xfer);
786 			if (error) {
787 				return error;
788 			}
789 			left -= xfer;
790 			len -= xfer;
791 			mbufcp += xfer;
792 			uiocp += xfer;
793 			uiop->uio_offset += xfer;
794 			uiop->uio_resid -= xfer;
795 		}
796 		if (uiop->uio_iov->iov_len <= siz) {
797 			uiop->uio_iovcnt--;
798 			uiop->uio_iov++;
799 		} else {
800 			uiop->uio_iov->iov_base =
801 			    (char *)uiop->uio_iov->iov_base + uiosiz;
802 			uiop->uio_iov->iov_len -= uiosiz;
803 		}
804 		siz -= uiosiz;
805 	}
806 	*dpos = mbufcp;
807 	*mrep = mp;
808 	if (rem > 0) {
809 		if (len < rem)
810 			error = nfs_adv(mrep, dpos, rem, len);
811 		else
812 			*dpos += rem;
813 	}
814 	return (error);
815 }
816 
817 /*
818  * copies a uio scatter/gather list to an mbuf chain.
819  * NOTE: can ony handle iovcnt == 1
820  */
821 int
822 nfsm_uiotombuf(uiop, mq, siz, bpos)
823 	struct uio *uiop;
824 	struct mbuf **mq;
825 	int siz;
826 	char **bpos;
827 {
828 	char *uiocp;
829 	struct mbuf *mp, *mp2;
830 	int xfer, left, mlen;
831 	int uiosiz, clflg, rem;
832 	char *cp;
833 	int error;
834 
835 #ifdef DIAGNOSTIC
836 	if (uiop->uio_iovcnt != 1)
837 		panic("nfsm_uiotombuf: iovcnt != 1");
838 #endif
839 
840 	if (siz > MLEN)		/* or should it >= MCLBYTES ?? */
841 		clflg = 1;
842 	else
843 		clflg = 0;
844 	rem = nfsm_rndup(siz)-siz;
845 	mp = mp2 = *mq;
846 	while (siz > 0) {
847 		left = uiop->uio_iov->iov_len;
848 		uiocp = uiop->uio_iov->iov_base;
849 		if (left > siz)
850 			left = siz;
851 		uiosiz = left;
852 		while (left > 0) {
853 			mlen = M_TRAILINGSPACE(mp);
854 			if (mlen == 0) {
855 				mp = m_get(M_WAIT, MT_DATA);
856 				MCLAIM(mp, &nfs_mowner);
857 				if (clflg)
858 					m_clget(mp, M_WAIT);
859 				mp->m_len = 0;
860 				mp2->m_next = mp;
861 				mp2 = mp;
862 				mlen = M_TRAILINGSPACE(mp);
863 			}
864 			xfer = (left > mlen) ? mlen : left;
865 			cp = mtod(mp, char *) + mp->m_len;
866 			error = copyin_vmspace(uiop->uio_vmspace, uiocp, cp,
867 			    xfer);
868 			if (error) {
869 				/* XXX */
870 			}
871 			mp->m_len += xfer;
872 			left -= xfer;
873 			uiocp += xfer;
874 			uiop->uio_offset += xfer;
875 			uiop->uio_resid -= xfer;
876 		}
877 		uiop->uio_iov->iov_base = (char *)uiop->uio_iov->iov_base +
878 		    uiosiz;
879 		uiop->uio_iov->iov_len -= uiosiz;
880 		siz -= uiosiz;
881 	}
882 	if (rem > 0) {
883 		if (rem > M_TRAILINGSPACE(mp)) {
884 			mp = m_get(M_WAIT, MT_DATA);
885 			MCLAIM(mp, &nfs_mowner);
886 			mp->m_len = 0;
887 			mp2->m_next = mp;
888 		}
889 		cp = mtod(mp, char *) + mp->m_len;
890 		for (left = 0; left < rem; left++)
891 			*cp++ = '\0';
892 		mp->m_len += rem;
893 		*bpos = cp;
894 	} else
895 		*bpos = mtod(mp, char *) + mp->m_len;
896 	*mq = mp;
897 	return (0);
898 }
899 
900 /*
901  * Get at least "siz" bytes of correctly aligned data.
902  * When called the mbuf pointers are not necessarily correct,
903  * dsosp points to what ought to be in m_data and left contains
904  * what ought to be in m_len.
905  * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
906  * cases. (The macros use the vars. dpos and dpos2)
907  */
908 int
909 nfsm_disct(mdp, dposp, siz, left, cp2)
910 	struct mbuf **mdp;
911 	char **dposp;
912 	int siz;
913 	int left;
914 	char **cp2;
915 {
916 	struct mbuf *m1, *m2;
917 	struct mbuf *havebuf = NULL;
918 	char *src = *dposp;
919 	char *dst;
920 	int len;
921 
922 #ifdef DEBUG
923 	if (left < 0)
924 		panic("nfsm_disct: left < 0");
925 #endif
926 	m1 = *mdp;
927 	/*
928 	 * Skip through the mbuf chain looking for an mbuf with
929 	 * some data. If the first mbuf found has enough data
930 	 * and it is correctly aligned return it.
931 	 */
932 	while (left == 0) {
933 		havebuf = m1;
934 		*mdp = m1 = m1->m_next;
935 		if (m1 == NULL)
936 			return (EBADRPC);
937 		src = mtod(m1, void *);
938 		left = m1->m_len;
939 		/*
940 		 * If we start a new mbuf and it is big enough
941 		 * and correctly aligned just return it, don't
942 		 * do any pull up.
943 		 */
944 		if (left >= siz && nfsm_aligned(src)) {
945 			*cp2 = src;
946 			*dposp = src + siz;
947 			return (0);
948 		}
949 	}
950 	if (m1->m_flags & M_EXT) {
951 		if (havebuf) {
952 			/* If the first mbuf with data has external data
953 			 * and there is a previous empty mbuf use it
954 			 * to move the data into.
955 			 */
956 			m2 = m1;
957 			*mdp = m1 = havebuf;
958 			if (m1->m_flags & M_EXT) {
959 				MEXTREMOVE(m1);
960 			}
961 		} else {
962 			/*
963 			 * If the first mbuf has a external data
964 			 * and there is no previous empty mbuf
965 			 * allocate a new mbuf and move the external
966 			 * data to the new mbuf. Also make the first
967 			 * mbuf look empty.
968 			 */
969 			m2 = m_get(M_WAIT, MT_DATA);
970 			m2->m_ext = m1->m_ext;
971 			m2->m_data = src;
972 			m2->m_len = left;
973 			MCLADDREFERENCE(m1, m2);
974 			MEXTREMOVE(m1);
975 			m2->m_next = m1->m_next;
976 			m1->m_next = m2;
977 		}
978 		m1->m_len = 0;
979 		if (m1->m_flags & M_PKTHDR)
980 			dst = m1->m_pktdat;
981 		else
982 			dst = m1->m_dat;
983 		m1->m_data = dst;
984 	} else {
985 		/*
986 		 * If the first mbuf has no external data
987 		 * move the data to the front of the mbuf.
988 		 */
989 		if (m1->m_flags & M_PKTHDR)
990 			dst = m1->m_pktdat;
991 		else
992 			dst = m1->m_dat;
993 		m1->m_data = dst;
994 		if (dst != src)
995 			memmove(dst, src, left);
996 		dst += left;
997 		m1->m_len = left;
998 		m2 = m1->m_next;
999 	}
1000 	*cp2 = m1->m_data;
1001 	*dposp = mtod(m1, char *) + siz;
1002 	/*
1003 	 * Loop through mbufs pulling data up into first mbuf until
1004 	 * the first mbuf is full or there is no more data to
1005 	 * pullup.
1006 	 */
1007 	while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) {
1008 		if ((len = min(len, m2->m_len)) != 0)
1009 			memcpy(dst, m2->m_data, len);
1010 		m1->m_len += len;
1011 		dst += len;
1012 		m2->m_data += len;
1013 		m2->m_len -= len;
1014 		m2 = m2->m_next;
1015 	}
1016 	if (m1->m_len < siz)
1017 		return (EBADRPC);
1018 	return (0);
1019 }
1020 
1021 /*
1022  * Advance the position in the mbuf chain.
1023  */
1024 int
1025 nfs_adv(mdp, dposp, offs, left)
1026 	struct mbuf **mdp;
1027 	char **dposp;
1028 	int offs;
1029 	int left;
1030 {
1031 	struct mbuf *m;
1032 	int s;
1033 
1034 	m = *mdp;
1035 	s = left;
1036 	while (s < offs) {
1037 		offs -= s;
1038 		m = m->m_next;
1039 		if (m == NULL)
1040 			return (EBADRPC);
1041 		s = m->m_len;
1042 	}
1043 	*mdp = m;
1044 	*dposp = mtod(m, char *) + offs;
1045 	return (0);
1046 }
1047 
1048 /*
1049  * Copy a string into mbufs for the hard cases...
1050  */
1051 int
1052 nfsm_strtmbuf(mb, bpos, cp, siz)
1053 	struct mbuf **mb;
1054 	char **bpos;
1055 	const char *cp;
1056 	long siz;
1057 {
1058 	struct mbuf *m1 = NULL, *m2;
1059 	long left, xfer, len, tlen;
1060 	u_int32_t *tl;
1061 	int putsize;
1062 
1063 	putsize = 1;
1064 	m2 = *mb;
1065 	left = M_TRAILINGSPACE(m2);
1066 	if (left > 0) {
1067 		tl = ((u_int32_t *)(*bpos));
1068 		*tl++ = txdr_unsigned(siz);
1069 		putsize = 0;
1070 		left -= NFSX_UNSIGNED;
1071 		m2->m_len += NFSX_UNSIGNED;
1072 		if (left > 0) {
1073 			memcpy((void *) tl, cp, left);
1074 			siz -= left;
1075 			cp += left;
1076 			m2->m_len += left;
1077 			left = 0;
1078 		}
1079 	}
1080 	/* Loop around adding mbufs */
1081 	while (siz > 0) {
1082 		m1 = m_get(M_WAIT, MT_DATA);
1083 		MCLAIM(m1, &nfs_mowner);
1084 		if (siz > MLEN)
1085 			m_clget(m1, M_WAIT);
1086 		m1->m_len = NFSMSIZ(m1);
1087 		m2->m_next = m1;
1088 		m2 = m1;
1089 		tl = mtod(m1, u_int32_t *);
1090 		tlen = 0;
1091 		if (putsize) {
1092 			*tl++ = txdr_unsigned(siz);
1093 			m1->m_len -= NFSX_UNSIGNED;
1094 			tlen = NFSX_UNSIGNED;
1095 			putsize = 0;
1096 		}
1097 		if (siz < m1->m_len) {
1098 			len = nfsm_rndup(siz);
1099 			xfer = siz;
1100 			if (xfer < len)
1101 				*(tl+(xfer>>2)) = 0;
1102 		} else {
1103 			xfer = len = m1->m_len;
1104 		}
1105 		memcpy((void *) tl, cp, xfer);
1106 		m1->m_len = len+tlen;
1107 		siz -= xfer;
1108 		cp += xfer;
1109 	}
1110 	*mb = m1;
1111 	*bpos = mtod(m1, char *) + m1->m_len;
1112 	return (0);
1113 }
1114 
1115 /*
1116  * Directory caching routines. They work as follows:
1117  * - a cache is maintained per VDIR nfsnode.
1118  * - for each offset cookie that is exported to userspace, and can
1119  *   thus be thrown back at us as an offset to VOP_READDIR, store
1120  *   information in the cache.
1121  * - cached are:
1122  *   - cookie itself
1123  *   - blocknumber (essentially just a search key in the buffer cache)
1124  *   - entry number in block.
1125  *   - offset cookie of block in which this entry is stored
1126  *   - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
1127  * - entries are looked up in a hash table
1128  * - also maintained is an LRU list of entries, used to determine
1129  *   which ones to delete if the cache grows too large.
1130  * - if 32 <-> 64 translation mode is requested for a filesystem,
1131  *   the cache also functions as a translation table
1132  * - in the translation case, invalidating the cache does not mean
1133  *   flushing it, but just marking entries as invalid, except for
1134  *   the <64bit cookie, 32bitcookie> pair which is still valid, to
1135  *   still be able to use the cache as a translation table.
1136  * - 32 bit cookies are uniquely created by combining the hash table
1137  *   entry value, and one generation count per hash table entry,
1138  *   incremented each time an entry is appended to the chain.
1139  * - the cache is invalidated each time a direcory is modified
1140  * - sanity checks are also done; if an entry in a block turns
1141  *   out not to have a matching cookie, the cache is invalidated
1142  *   and a new block starting from the wanted offset is fetched from
1143  *   the server.
1144  * - directory entries as read from the server are extended to contain
1145  *   the 64bit and, optionally, the 32bit cookies, for sanity checking
1146  *   the cache and exporting them to userspace through the cookie
1147  *   argument to VOP_READDIR.
1148  */
1149 
1150 u_long
1151 nfs_dirhash(off)
1152 	off_t off;
1153 {
1154 	int i;
1155 	char *cp = (char *)&off;
1156 	u_long sum = 0L;
1157 
1158 	for (i = 0 ; i < sizeof (off); i++)
1159 		sum += *cp++;
1160 
1161 	return sum;
1162 }
1163 
1164 #define	_NFSDC_MTX(np)		(&NFSTOV(np)->v_interlock)
1165 #define	NFSDC_LOCK(np)		simple_lock(_NFSDC_MTX(np))
1166 #define	NFSDC_UNLOCK(np)	simple_unlock(_NFSDC_MTX(np))
1167 #define	NFSDC_ASSERT_LOCKED(np) LOCK_ASSERT(simple_lock_held(_NFSDC_MTX(np)))
1168 
1169 void
1170 nfs_initdircache(vp)
1171 	struct vnode *vp;
1172 {
1173 	struct nfsnode *np = VTONFS(vp);
1174 	struct nfsdirhashhead *dircache;
1175 
1176 	dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF,
1177 	    M_WAITOK, &nfsdirhashmask);
1178 
1179 	NFSDC_LOCK(np);
1180 	if (np->n_dircache == NULL) {
1181 		np->n_dircachesize = 0;
1182 		np->n_dircache = dircache;
1183 		dircache = NULL;
1184 		TAILQ_INIT(&np->n_dirchain);
1185 	}
1186 	NFSDC_UNLOCK(np);
1187 	if (dircache)
1188 		hashdone(dircache, M_NFSDIROFF);
1189 }
1190 
1191 void
1192 nfs_initdirxlatecookie(vp)
1193 	struct vnode *vp;
1194 {
1195 	struct nfsnode *np = VTONFS(vp);
1196 	unsigned *dirgens;
1197 
1198 	KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE);
1199 
1200 	dirgens = malloc(NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF,
1201 	    M_WAITOK|M_ZERO);
1202 	NFSDC_LOCK(np);
1203 	if (np->n_dirgens == NULL) {
1204 		np->n_dirgens = dirgens;
1205 		dirgens = NULL;
1206 	}
1207 	NFSDC_UNLOCK(np);
1208 	if (dirgens)
1209 		free(dirgens, M_NFSDIROFF);
1210 }
1211 
1212 static const struct nfsdircache dzero;
1213 
1214 static void nfs_unlinkdircache __P((struct nfsnode *np, struct nfsdircache *));
1215 static void nfs_putdircache_unlocked __P((struct nfsnode *,
1216     struct nfsdircache *));
1217 
1218 static void
1219 nfs_unlinkdircache(np, ndp)
1220 	struct nfsnode *np;
1221 	struct nfsdircache *ndp;
1222 {
1223 
1224 	NFSDC_ASSERT_LOCKED(np);
1225 	KASSERT(ndp != &dzero);
1226 
1227 	if (LIST_NEXT(ndp, dc_hash) == (void *)-1)
1228 		return;
1229 
1230 	TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1231 	LIST_REMOVE(ndp, dc_hash);
1232 	LIST_NEXT(ndp, dc_hash) = (void *)-1; /* mark as unlinked */
1233 
1234 	nfs_putdircache_unlocked(np, ndp);
1235 }
1236 
1237 void
1238 nfs_putdircache(np, ndp)
1239 	struct nfsnode *np;
1240 	struct nfsdircache *ndp;
1241 {
1242 	int ref;
1243 
1244 	if (ndp == &dzero)
1245 		return;
1246 
1247 	KASSERT(ndp->dc_refcnt > 0);
1248 	NFSDC_LOCK(np);
1249 	ref = --ndp->dc_refcnt;
1250 	NFSDC_UNLOCK(np);
1251 
1252 	if (ref == 0)
1253 		free(ndp, M_NFSDIROFF);
1254 }
1255 
1256 static void
1257 nfs_putdircache_unlocked(struct nfsnode *np, struct nfsdircache *ndp)
1258 {
1259 	int ref;
1260 
1261 	NFSDC_ASSERT_LOCKED(np);
1262 
1263 	if (ndp == &dzero)
1264 		return;
1265 
1266 	KASSERT(ndp->dc_refcnt > 0);
1267 	ref = --ndp->dc_refcnt;
1268 	if (ref == 0)
1269 		free(ndp, M_NFSDIROFF);
1270 }
1271 
1272 struct nfsdircache *
1273 nfs_searchdircache(vp, off, do32, hashent)
1274 	struct vnode *vp;
1275 	off_t off;
1276 	int do32;
1277 	int *hashent;
1278 {
1279 	struct nfsdirhashhead *ndhp;
1280 	struct nfsdircache *ndp = NULL;
1281 	struct nfsnode *np = VTONFS(vp);
1282 	unsigned ent;
1283 
1284 	/*
1285 	 * Zero is always a valid cookie.
1286 	 */
1287 	if (off == 0)
1288 		/* XXXUNCONST */
1289 		return (struct nfsdircache *)__UNCONST(&dzero);
1290 
1291 	if (!np->n_dircache)
1292 		return NULL;
1293 
1294 	/*
1295 	 * We use a 32bit cookie as search key, directly reconstruct
1296 	 * the hashentry. Else use the hashfunction.
1297 	 */
1298 	if (do32) {
1299 		ent = (u_int32_t)off >> 24;
1300 		if (ent >= NFS_DIRHASHSIZ)
1301 			return NULL;
1302 		ndhp = &np->n_dircache[ent];
1303 	} else {
1304 		ndhp = NFSDIRHASH(np, off);
1305 	}
1306 
1307 	if (hashent)
1308 		*hashent = (int)(ndhp - np->n_dircache);
1309 
1310 	NFSDC_LOCK(np);
1311 	if (do32) {
1312 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1313 			if (ndp->dc_cookie32 == (u_int32_t)off) {
1314 				/*
1315 				 * An invalidated entry will become the
1316 				 * start of a new block fetched from
1317 				 * the server.
1318 				 */
1319 				if (ndp->dc_flags & NFSDC_INVALID) {
1320 					ndp->dc_blkcookie = ndp->dc_cookie;
1321 					ndp->dc_entry = 0;
1322 					ndp->dc_flags &= ~NFSDC_INVALID;
1323 				}
1324 				break;
1325 			}
1326 		}
1327 	} else {
1328 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1329 			if (ndp->dc_cookie == off)
1330 				break;
1331 		}
1332 	}
1333 	if (ndp != NULL)
1334 		ndp->dc_refcnt++;
1335 	NFSDC_UNLOCK(np);
1336 	return ndp;
1337 }
1338 
1339 
1340 struct nfsdircache *
1341 nfs_enterdircache(struct vnode *vp, off_t off, off_t blkoff, int en,
1342     daddr_t blkno)
1343 {
1344 	struct nfsnode *np = VTONFS(vp);
1345 	struct nfsdirhashhead *ndhp;
1346 	struct nfsdircache *ndp = NULL;
1347 	struct nfsdircache *newndp = NULL;
1348 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1349 	int hashent = 0, gen, overwrite;	/* XXX: GCC */
1350 
1351 	/*
1352 	 * XXX refuse entries for offset 0. amd(8) erroneously sets
1353 	 * cookie 0 for the '.' entry, making this necessary. This
1354 	 * isn't so bad, as 0 is a special case anyway.
1355 	 */
1356 	if (off == 0)
1357 		/* XXXUNCONST */
1358 		return (struct nfsdircache *)__UNCONST(&dzero);
1359 
1360 	if (!np->n_dircache)
1361 		/*
1362 		 * XXX would like to do this in nfs_nget but vtype
1363 		 * isn't known at that time.
1364 		 */
1365 		nfs_initdircache(vp);
1366 
1367 	if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens)
1368 		nfs_initdirxlatecookie(vp);
1369 
1370 retry:
1371 	ndp = nfs_searchdircache(vp, off, 0, &hashent);
1372 
1373 	NFSDC_LOCK(np);
1374 	if (ndp && (ndp->dc_flags & NFSDC_INVALID) == 0) {
1375 		/*
1376 		 * Overwriting an old entry. Check if it's the same.
1377 		 * If so, just return. If not, remove the old entry.
1378 		 */
1379 		if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
1380 			goto done;
1381 		nfs_unlinkdircache(np, ndp);
1382 		nfs_putdircache_unlocked(np, ndp);
1383 		ndp = NULL;
1384 	}
1385 
1386 	ndhp = &np->n_dircache[hashent];
1387 
1388 	if (!ndp) {
1389 		if (newndp == NULL) {
1390 			NFSDC_UNLOCK(np);
1391 			newndp = malloc(sizeof(*ndp), M_NFSDIROFF, M_WAITOK);
1392 			newndp->dc_refcnt = 1;
1393 			LIST_NEXT(newndp, dc_hash) = (void *)-1;
1394 			goto retry;
1395 		}
1396 		ndp = newndp;
1397 		newndp = NULL;
1398 		overwrite = 0;
1399 		if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1400 			/*
1401 			 * We're allocating a new entry, so bump the
1402 			 * generation number.
1403 			 */
1404 			KASSERT(np->n_dirgens);
1405 			gen = ++np->n_dirgens[hashent];
1406 			if (gen == 0) {
1407 				np->n_dirgens[hashent]++;
1408 				gen++;
1409 			}
1410 			ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
1411 		}
1412 	} else
1413 		overwrite = 1;
1414 
1415 	ndp->dc_cookie = off;
1416 	ndp->dc_blkcookie = blkoff;
1417 	ndp->dc_entry = en;
1418 	ndp->dc_flags = 0;
1419 
1420 	if (overwrite)
1421 		goto done;
1422 
1423 	/*
1424 	 * If the maximum directory cookie cache size has been reached
1425 	 * for this node, take one off the front. The idea is that
1426 	 * directories are typically read front-to-back once, so that
1427 	 * the oldest entries can be thrown away without much performance
1428 	 * loss.
1429 	 */
1430 	if (np->n_dircachesize == NFS_MAXDIRCACHE) {
1431 		nfs_unlinkdircache(np, TAILQ_FIRST(&np->n_dirchain));
1432 	} else
1433 		np->n_dircachesize++;
1434 
1435 	KASSERT(ndp->dc_refcnt == 1);
1436 	LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
1437 	TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
1438 	ndp->dc_refcnt++;
1439 done:
1440 	KASSERT(ndp->dc_refcnt > 0);
1441 	NFSDC_UNLOCK(np);
1442 	if (newndp)
1443 		nfs_putdircache(np, newndp);
1444 	return ndp;
1445 }
1446 
1447 void
1448 nfs_invaldircache(vp, flags)
1449 	struct vnode *vp;
1450 	int flags;
1451 {
1452 	struct nfsnode *np = VTONFS(vp);
1453 	struct nfsdircache *ndp = NULL;
1454 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1455 	const bool forcefree = flags & NFS_INVALDIRCACHE_FORCE;
1456 
1457 #ifdef DIAGNOSTIC
1458 	if (vp->v_type != VDIR)
1459 		panic("nfs: invaldircache: not dir");
1460 #endif
1461 
1462 	if ((flags & NFS_INVALDIRCACHE_KEEPEOF) == 0)
1463 		np->n_flag &= ~NEOFVALID;
1464 
1465 	if (!np->n_dircache)
1466 		return;
1467 
1468 	NFSDC_LOCK(np);
1469 	if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
1470 		while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != NULL) {
1471 			KASSERT(!forcefree || ndp->dc_refcnt == 1);
1472 			nfs_unlinkdircache(np, ndp);
1473 		}
1474 		np->n_dircachesize = 0;
1475 		if (forcefree && np->n_dirgens) {
1476 			FREE(np->n_dirgens, M_NFSDIROFF);
1477 			np->n_dirgens = NULL;
1478 		}
1479 	} else {
1480 		TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain)
1481 			ndp->dc_flags |= NFSDC_INVALID;
1482 	}
1483 
1484 	NFSDC_UNLOCK(np);
1485 }
1486 
1487 /*
1488  * Called once before VFS init to initialize shared and
1489  * server-specific data structures.
1490  */
1491 static int
1492 nfs_init0(void)
1493 {
1494 
1495 	nfsrtt.pos = 0;
1496 	rpc_vers = txdr_unsigned(RPC_VER2);
1497 	rpc_call = txdr_unsigned(RPC_CALL);
1498 	rpc_reply = txdr_unsigned(RPC_REPLY);
1499 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1500 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1501 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1502 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1503 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1504 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1505 	nfs_prog = txdr_unsigned(NFS_PROG);
1506 	nfs_true = txdr_unsigned(true);
1507 	nfs_false = txdr_unsigned(false);
1508 	nfs_xdrneg1 = txdr_unsigned(-1);
1509 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1510 	if (nfs_ticks < 1)
1511 		nfs_ticks = 1;
1512 #ifdef NFSSERVER
1513 	nfsrv_init(0);			/* Init server data structures */
1514 	nfsrv_initcache();		/* Init the server request cache */
1515 	{
1516 		extern krwlock_t netexport_lock;	/* XXX */
1517 		rw_init(&netexport_lock);
1518 	}
1519 #endif /* NFSSERVER */
1520 
1521 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY))
1522 	nfsdreq_init();
1523 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */
1524 
1525 	/*
1526 	 * Initialize reply list and start timer
1527 	 */
1528 	TAILQ_INIT(&nfs_reqq);
1529 	nfs_timer_init();
1530 	MOWNER_ATTACH(&nfs_mowner);
1531 
1532 #ifdef NFS
1533 	/* Initialize the kqueue structures */
1534 	nfs_kqinit();
1535 	/* Initialize the iod structures */
1536 	nfs_iodinit();
1537 #endif
1538 	return 0;
1539 }
1540 
1541 void
1542 nfs_init(void)
1543 {
1544 	static ONCE_DECL(nfs_init_once);
1545 
1546 	RUN_ONCE(&nfs_init_once, nfs_init0);
1547 }
1548 
1549 #ifdef NFS
1550 /*
1551  * Called once at VFS init to initialize client-specific data structures.
1552  */
1553 void
1554 nfs_vfs_init()
1555 {
1556 	/* Initialize NFS server / client shared data. */
1557 	nfs_init();
1558 
1559 	nfs_nhinit();			/* Init the nfsnode table */
1560 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1561 }
1562 
1563 void
1564 nfs_vfs_reinit()
1565 {
1566 	nfs_nhreinit();
1567 }
1568 
1569 void
1570 nfs_vfs_done()
1571 {
1572 	nfs_nhdone();
1573 }
1574 
1575 /*
1576  * Attribute cache routines.
1577  * nfs_loadattrcache() - loads or updates the cache contents from attributes
1578  *	that are on the mbuf list
1579  * nfs_getattrcache() - returns valid attributes if found in cache, returns
1580  *	error otherwise
1581  */
1582 
1583 /*
1584  * Load the attribute cache (that lives in the nfsnode entry) with
1585  * the values on the mbuf list and
1586  * Iff vap not NULL
1587  *    copy the attributes to *vaper
1588  */
1589 int
1590 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags)
1591 	struct vnode **vpp;
1592 	struct mbuf **mdp;
1593 	char **dposp;
1594 	struct vattr *vaper;
1595 	int flags;
1596 {
1597 	int32_t t1;
1598 	char *cp2;
1599 	int error = 0;
1600 	struct mbuf *md;
1601 	int v3 = NFS_ISV3(*vpp);
1602 
1603 	md = *mdp;
1604 	t1 = (mtod(md, char *) + md->m_len) - *dposp;
1605 	error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2);
1606 	if (error)
1607 		return (error);
1608 	return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags);
1609 }
1610 
1611 int
1612 nfs_loadattrcache(vpp, fp, vaper, flags)
1613 	struct vnode **vpp;
1614 	struct nfs_fattr *fp;
1615 	struct vattr *vaper;
1616 	int flags;
1617 {
1618 	struct vnode *vp = *vpp;
1619 	struct vattr *vap;
1620 	int v3 = NFS_ISV3(vp);
1621 	enum vtype vtyp;
1622 	u_short vmode;
1623 	struct timespec mtime;
1624 	struct timespec ctime;
1625 	struct vnode *nvp;
1626 	int32_t rdev;
1627 	struct nfsnode *np;
1628 	extern int (**spec_nfsv2nodeop_p) __P((void *));
1629 	uid_t uid;
1630 	gid_t gid;
1631 
1632 	if (v3) {
1633 		vtyp = nfsv3tov_type(fp->fa_type);
1634 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1635 		rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1),
1636 			fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2));
1637 		fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
1638 		fxdr_nfsv3time(&fp->fa3_ctime, &ctime);
1639 	} else {
1640 		vtyp = nfsv2tov_type(fp->fa_type);
1641 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1642 		if (vtyp == VNON || vtyp == VREG)
1643 			vtyp = IFTOVT(vmode);
1644 		rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
1645 		fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
1646 		ctime.tv_sec = fxdr_unsigned(u_int32_t,
1647 		    fp->fa2_ctime.nfsv2_sec);
1648 		ctime.tv_nsec = 0;
1649 
1650 		/*
1651 		 * Really ugly NFSv2 kludge.
1652 		 */
1653 		if (vtyp == VCHR && rdev == 0xffffffff)
1654 			vtyp = VFIFO;
1655 	}
1656 
1657 	vmode &= ALLPERMS;
1658 
1659 	/*
1660 	 * If v_type == VNON it is a new node, so fill in the v_type,
1661 	 * n_mtime fields. Check to see if it represents a special
1662 	 * device, and if so, check for a possible alias. Once the
1663 	 * correct vnode has been obtained, fill in the rest of the
1664 	 * information.
1665 	 */
1666 	np = VTONFS(vp);
1667 	if (vp->v_type == VNON) {
1668 		vp->v_type = vtyp;
1669 		if (vp->v_type == VFIFO) {
1670 			extern int (**fifo_nfsv2nodeop_p) __P((void *));
1671 			vp->v_op = fifo_nfsv2nodeop_p;
1672 		} else if (vp->v_type == VREG) {
1673 			mutex_init(&np->n_commitlock, MUTEX_DEFAULT, IPL_NONE);
1674 		} else if (vp->v_type == VCHR || vp->v_type == VBLK) {
1675 			vp->v_op = spec_nfsv2nodeop_p;
1676 			nvp = checkalias(vp, (dev_t)rdev, vp->v_mount);
1677 			if (nvp) {
1678 				/*
1679 				 * Discard unneeded vnode, but save its nfsnode.
1680 				 * Since the nfsnode does not have a lock, its
1681 				 * vnode lock has to be carried over.
1682 				 */
1683 				/*
1684 				 * XXX is the old node sure to be locked here?
1685 				 */
1686 				KASSERT(lockstatus(&vp->v_lock) ==
1687 				    LK_EXCLUSIVE);
1688 				nvp->v_data = vp->v_data;
1689 				vp->v_data = NULL;
1690 				VOP_UNLOCK(vp, 0);
1691 				vp->v_op = spec_vnodeop_p;
1692 				vrele(vp);
1693 				vgone(vp);
1694 				lockmgr(&nvp->v_lock, LK_EXCLUSIVE,
1695 				    &nvp->v_interlock);
1696 				/*
1697 				 * Reinitialize aliased node.
1698 				 */
1699 				np->n_vnode = nvp;
1700 				*vpp = vp = nvp;
1701 			}
1702 		}
1703 		np->n_mtime = mtime;
1704 	}
1705 	uid = fxdr_unsigned(uid_t, fp->fa_uid);
1706 	gid = fxdr_unsigned(gid_t, fp->fa_gid);
1707 	vap = np->n_vattr;
1708 
1709 	/*
1710 	 * Invalidate access cache if uid, gid, mode or ctime changed.
1711 	 */
1712 	if (np->n_accstamp != -1 &&
1713 	    (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode
1714 	    || timespeccmp(&ctime, &vap->va_ctime, !=)))
1715 		np->n_accstamp = -1;
1716 
1717 	vap->va_type = vtyp;
1718 	vap->va_mode = vmode;
1719 	vap->va_rdev = (dev_t)rdev;
1720 	vap->va_mtime = mtime;
1721 	vap->va_ctime = ctime;
1722 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1723 	switch (vtyp) {
1724 	case VDIR:
1725 		vap->va_blocksize = NFS_DIRFRAGSIZ;
1726 		break;
1727 	case VBLK:
1728 		vap->va_blocksize = BLKDEV_IOSIZE;
1729 		break;
1730 	case VCHR:
1731 		vap->va_blocksize = MAXBSIZE;
1732 		break;
1733 	default:
1734 		vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize :
1735 		    fxdr_unsigned(int32_t, fp->fa2_blocksize);
1736 		break;
1737 	}
1738 	if (v3) {
1739 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1740 		vap->va_uid = uid;
1741 		vap->va_gid = gid;
1742 		vap->va_size = fxdr_hyper(&fp->fa3_size);
1743 		vap->va_bytes = fxdr_hyper(&fp->fa3_used);
1744 		vap->va_fileid = fxdr_hyper(&fp->fa3_fileid);
1745 		fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
1746 		vap->va_flags = 0;
1747 		vap->va_filerev = 0;
1748 	} else {
1749 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1750 		vap->va_uid = uid;
1751 		vap->va_gid = gid;
1752 		vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
1753 		vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks)
1754 		    * NFS_FABLKSIZE;
1755 		vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
1756 		fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
1757 		vap->va_flags = 0;
1758 		vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
1759 		vap->va_filerev = 0;
1760 	}
1761 	if (vap->va_size != np->n_size) {
1762 		if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) {
1763 			vap->va_size = np->n_size;
1764 		} else {
1765 			np->n_size = vap->va_size;
1766 			if (vap->va_type == VREG) {
1767 				/*
1768 				 * we can't free pages if NAC_NOTRUNC because
1769 				 * the pages can be owned by ourselves.
1770 				 */
1771 				if (flags & NAC_NOTRUNC) {
1772 					np->n_flag |= NTRUNCDELAYED;
1773 				} else {
1774 					genfs_node_wrlock(vp);
1775 					simple_lock(&vp->v_interlock);
1776 					(void)VOP_PUTPAGES(vp, 0,
1777 					    0, PGO_SYNCIO | PGO_CLEANIT |
1778 					    PGO_FREE | PGO_ALLPAGES);
1779 					uvm_vnp_setsize(vp, np->n_size);
1780 					genfs_node_unlock(vp);
1781 				}
1782 			}
1783 		}
1784 	}
1785 	np->n_attrstamp = time_second;
1786 	if (vaper != NULL) {
1787 		memcpy((void *)vaper, (void *)vap, sizeof(*vap));
1788 		if (np->n_flag & NCHG) {
1789 			if (np->n_flag & NACC)
1790 				vaper->va_atime = np->n_atim;
1791 			if (np->n_flag & NUPD)
1792 				vaper->va_mtime = np->n_mtim;
1793 		}
1794 	}
1795 	return (0);
1796 }
1797 
1798 /*
1799  * Check the time stamp
1800  * If the cache is valid, copy contents to *vap and return 0
1801  * otherwise return an error
1802  */
1803 int
1804 nfs_getattrcache(vp, vaper)
1805 	struct vnode *vp;
1806 	struct vattr *vaper;
1807 {
1808 	struct nfsnode *np = VTONFS(vp);
1809 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1810 	struct vattr *vap;
1811 
1812 	if (np->n_attrstamp == 0 ||
1813 	    (time_second - np->n_attrstamp) >= nfs_attrtimeo(nmp, np)) {
1814 		nfsstats.attrcache_misses++;
1815 		return (ENOENT);
1816 	}
1817 	nfsstats.attrcache_hits++;
1818 	vap = np->n_vattr;
1819 	if (vap->va_size != np->n_size) {
1820 		if (vap->va_type == VREG) {
1821 			if ((np->n_flag & NMODIFIED) != 0 &&
1822 			    vap->va_size < np->n_size) {
1823 				vap->va_size = np->n_size;
1824 			} else {
1825 				np->n_size = vap->va_size;
1826 			}
1827 			genfs_node_wrlock(vp);
1828 			uvm_vnp_setsize(vp, np->n_size);
1829 			genfs_node_unlock(vp);
1830 		} else
1831 			np->n_size = vap->va_size;
1832 	}
1833 	memcpy((void *)vaper, (void *)vap, sizeof(struct vattr));
1834 	if (np->n_flag & NCHG) {
1835 		if (np->n_flag & NACC)
1836 			vaper->va_atime = np->n_atim;
1837 		if (np->n_flag & NUPD)
1838 			vaper->va_mtime = np->n_mtim;
1839 	}
1840 	return (0);
1841 }
1842 
1843 void
1844 nfs_delayedtruncate(vp)
1845 	struct vnode *vp;
1846 {
1847 	struct nfsnode *np = VTONFS(vp);
1848 
1849 	if (np->n_flag & NTRUNCDELAYED) {
1850 		np->n_flag &= ~NTRUNCDELAYED;
1851 		genfs_node_wrlock(vp);
1852 		simple_lock(&vp->v_interlock);
1853 		(void)VOP_PUTPAGES(vp, 0,
1854 		    0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES);
1855 		uvm_vnp_setsize(vp, np->n_size);
1856 		genfs_node_unlock(vp);
1857 	}
1858 }
1859 
1860 #define	NFS_WCCKLUDGE_TIMEOUT	(24 * 60 * 60)	/* 1 day */
1861 #define	NFS_WCCKLUDGE(nmp, now) \
1862 	(((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \
1863 	((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0)
1864 
1865 /*
1866  * nfs_check_wccdata: check inaccurate wcc_data
1867  *
1868  * => return non-zero if we shouldn't trust the wcc_data.
1869  * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed".
1870  */
1871 
1872 int
1873 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime,
1874     struct timespec *mtime, bool docheck)
1875 {
1876 	int error = 0;
1877 
1878 #if !defined(NFS_V2_ONLY)
1879 
1880 	if (docheck) {
1881 		struct vnode *vp = NFSTOV(np);
1882 		struct nfsmount *nmp;
1883 		long now = time_second;
1884 		const struct timespec *omtime = &np->n_vattr->va_mtime;
1885 		const struct timespec *octime = &np->n_vattr->va_ctime;
1886 #if defined(DEBUG)
1887 		const char *reason = NULL; /* XXX: gcc */
1888 #endif
1889 
1890 		if (timespeccmp(omtime, mtime, <=)) {
1891 #if defined(DEBUG)
1892 			reason = "mtime";
1893 #endif
1894 			error = EINVAL;
1895 		}
1896 
1897 		if (vp->v_type == VDIR && timespeccmp(octime, ctime, <=)) {
1898 #if defined(DEBUG)
1899 			reason = "ctime";
1900 #endif
1901 			error = EINVAL;
1902 		}
1903 
1904 		nmp = VFSTONFS(vp->v_mount);
1905 		if (error) {
1906 
1907 			/*
1908 			 * despite of the fact that we've updated the file,
1909 			 * timestamps of the file were not updated as we
1910 			 * expected.
1911 			 * it means that the server has incompatible
1912 			 * semantics of timestamps or (more likely)
1913 			 * the server time is not precise enough to
1914 			 * track each modifications.
1915 			 * in that case, we disable wcc processing.
1916 			 *
1917 			 * yes, strictly speaking, we should disable all
1918 			 * caching.  it's a compromise.
1919 			 */
1920 
1921 			mutex_enter(&nmp->nm_lock);
1922 #if defined(DEBUG)
1923 			if (!NFS_WCCKLUDGE(nmp, now)) {
1924 				printf("%s: inaccurate wcc data (%s) detected,"
1925 				    " disabling wcc"
1926 				    " (ctime %u.%09u %u.%09u,"
1927 				    " mtime %u.%09u %u.%09u)\n",
1928 				    vp->v_mount->mnt_stat.f_mntfromname,
1929 				    reason,
1930 				    (unsigned int)octime->tv_sec,
1931 				    (unsigned int)octime->tv_nsec,
1932 				    (unsigned int)ctime->tv_sec,
1933 				    (unsigned int)ctime->tv_nsec,
1934 				    (unsigned int)omtime->tv_sec,
1935 				    (unsigned int)omtime->tv_nsec,
1936 				    (unsigned int)mtime->tv_sec,
1937 				    (unsigned int)mtime->tv_nsec);
1938 			}
1939 #endif
1940 			nmp->nm_iflag |= NFSMNT_WCCKLUDGE;
1941 			nmp->nm_wcckludgetime = now;
1942 			mutex_exit(&nmp->nm_lock);
1943 		} else if (NFS_WCCKLUDGE(nmp, now)) {
1944 			error = EPERM; /* XXX */
1945 		} else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
1946 			mutex_enter(&nmp->nm_lock);
1947 			if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
1948 #if defined(DEBUG)
1949 				printf("%s: re-enabling wcc\n",
1950 				    vp->v_mount->mnt_stat.f_mntfromname);
1951 #endif
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_authorize_generic(cred, KAUTH_GENERIC_ISSUSER,
2548 		    NULL) == 0 || (exflags & MNT_EXPORTANON)) {
2549 		kauth_cred_clone(credanon, cred);
2550 	}
2551 	if (exflags & MNT_EXRDONLY)
2552 		*rdonlyp = 1;
2553 	else
2554 		*rdonlyp = 0;
2555 	if (!lockflag)
2556 		VOP_UNLOCK(*vpp, 0);
2557 	return (0);
2558 }
2559 
2560 /*
2561  * WebNFS: check if a filehandle is a public filehandle. For v3, this
2562  * means a length of 0, for v2 it means all zeroes.
2563  */
2564 int
2565 nfs_ispublicfh(const nfsrvfh_t *nsfh)
2566 {
2567 	const char *cp = (const void *)(NFSRVFH_DATA(nsfh));
2568 	int i;
2569 
2570 	if (NFSRVFH_SIZE(nsfh) == 0) {
2571 		return true;
2572 	}
2573 	if (NFSRVFH_SIZE(nsfh) != NFSX_V2FH) {
2574 		return false;
2575 	}
2576 	for (i = 0; i < NFSX_V2FH; i++)
2577 		if (*cp++ != 0)
2578 			return false;
2579 	return true;
2580 }
2581 #endif /* NFSSERVER */
2582 
2583 /*
2584  * This function compares two net addresses by family and returns true
2585  * if they are the same host.
2586  * If there is any doubt, return false.
2587  * The AF_INET family is handled as a special case so that address mbufs
2588  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
2589  */
2590 int
2591 netaddr_match(family, haddr, nam)
2592 	int family;
2593 	union nethostaddr *haddr;
2594 	struct mbuf *nam;
2595 {
2596 	struct sockaddr_in *inetaddr;
2597 
2598 	switch (family) {
2599 	case AF_INET:
2600 		inetaddr = mtod(nam, struct sockaddr_in *);
2601 		if (inetaddr->sin_family == AF_INET &&
2602 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
2603 			return (1);
2604 		break;
2605 #ifdef INET6
2606 	case AF_INET6:
2607 	    {
2608 		struct sockaddr_in6 *sin6_1, *sin6_2;
2609 
2610 		sin6_1 = mtod(nam, struct sockaddr_in6 *);
2611 		sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
2612 		if (sin6_1->sin6_family == AF_INET6 &&
2613 		    IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
2614 			return 1;
2615 	    }
2616 #endif
2617 #ifdef ISO
2618 	case AF_ISO:
2619 	    {
2620 		struct sockaddr_iso *isoaddr1, *isoaddr2;
2621 
2622 		isoaddr1 = mtod(nam, struct sockaddr_iso *);
2623 		isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *);
2624 		if (isoaddr1->siso_family == AF_ISO &&
2625 		    isoaddr1->siso_nlen > 0 &&
2626 		    isoaddr1->siso_nlen == isoaddr2->siso_nlen &&
2627 		    SAME_ISOADDR(isoaddr1, isoaddr2))
2628 			return (1);
2629 		break;
2630 	    }
2631 #endif	/* ISO */
2632 	default:
2633 		break;
2634 	};
2635 	return (0);
2636 }
2637 
2638 /*
2639  * The write verifier has changed (probably due to a server reboot), so all
2640  * PG_NEEDCOMMIT pages will have to be written again. Since they are marked
2641  * as dirty or are being written out just now, all this takes is clearing
2642  * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for
2643  * the mount point.
2644  */
2645 void
2646 nfs_clearcommit(mp)
2647 	struct mount *mp;
2648 {
2649 	struct vnode *vp;
2650 	struct nfsnode *np;
2651 	struct vm_page *pg;
2652 	struct nfsmount *nmp = VFSTONFS(mp);
2653 
2654 	rw_enter(&nmp->nm_writeverflock, RW_WRITER);
2655 
2656 	TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2657 		KASSERT(vp->v_mount == mp);
2658 		if (vp->v_type != VREG)
2659 			continue;
2660 		np = VTONFS(vp);
2661 		np->n_pushlo = np->n_pushhi = np->n_pushedlo =
2662 		    np->n_pushedhi = 0;
2663 		np->n_commitflags &=
2664 		    ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
2665 		simple_lock(&vp->v_uobj.vmobjlock);
2666 		TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) {
2667 			pg->flags &= ~PG_NEEDCOMMIT;
2668 		}
2669 		simple_unlock(&vp->v_uobj.vmobjlock);
2670 	}
2671 	mutex_enter(&nmp->nm_lock);
2672 	nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF;
2673 	mutex_exit(&nmp->nm_lock);
2674 	rw_exit(&nmp->nm_writeverflock);
2675 }
2676 
2677 void
2678 nfs_merge_commit_ranges(vp)
2679 	struct vnode *vp;
2680 {
2681 	struct nfsnode *np = VTONFS(vp);
2682 
2683 	KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID);
2684 
2685 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2686 		np->n_pushedlo = np->n_pushlo;
2687 		np->n_pushedhi = np->n_pushhi;
2688 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2689 	} else {
2690 		if (np->n_pushlo < np->n_pushedlo)
2691 			np->n_pushedlo = np->n_pushlo;
2692 		if (np->n_pushhi > np->n_pushedhi)
2693 			np->n_pushedhi = np->n_pushhi;
2694 	}
2695 
2696 	np->n_pushlo = np->n_pushhi = 0;
2697 	np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
2698 
2699 #ifdef NFS_DEBUG_COMMIT
2700 	printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2701 	    (unsigned)np->n_pushedhi);
2702 #endif
2703 }
2704 
2705 int
2706 nfs_in_committed_range(vp, off, len)
2707 	struct vnode *vp;
2708 	off_t off, len;
2709 {
2710 	struct nfsnode *np = VTONFS(vp);
2711 	off_t lo, hi;
2712 
2713 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2714 		return 0;
2715 	lo = off;
2716 	hi = lo + len;
2717 
2718 	return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
2719 }
2720 
2721 int
2722 nfs_in_tobecommitted_range(vp, off, len)
2723 	struct vnode *vp;
2724 	off_t off, len;
2725 {
2726 	struct nfsnode *np = VTONFS(vp);
2727 	off_t lo, hi;
2728 
2729 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2730 		return 0;
2731 	lo = off;
2732 	hi = lo + len;
2733 
2734 	return (lo >= np->n_pushlo && hi <= np->n_pushhi);
2735 }
2736 
2737 void
2738 nfs_add_committed_range(vp, off, len)
2739 	struct vnode *vp;
2740 	off_t off, len;
2741 {
2742 	struct nfsnode *np = VTONFS(vp);
2743 	off_t lo, hi;
2744 
2745 	lo = off;
2746 	hi = lo + len;
2747 
2748 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2749 		np->n_pushedlo = lo;
2750 		np->n_pushedhi = hi;
2751 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2752 	} else {
2753 		if (hi > np->n_pushedhi)
2754 			np->n_pushedhi = hi;
2755 		if (lo < np->n_pushedlo)
2756 			np->n_pushedlo = lo;
2757 	}
2758 #ifdef NFS_DEBUG_COMMIT
2759 	printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2760 	    (unsigned)np->n_pushedhi);
2761 #endif
2762 }
2763 
2764 void
2765 nfs_del_committed_range(vp, off, len)
2766 	struct vnode *vp;
2767 	off_t off, len;
2768 {
2769 	struct nfsnode *np = VTONFS(vp);
2770 	off_t lo, hi;
2771 
2772 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2773 		return;
2774 
2775 	lo = off;
2776 	hi = lo + len;
2777 
2778 	if (lo > np->n_pushedhi || hi < np->n_pushedlo)
2779 		return;
2780 	if (lo <= np->n_pushedlo)
2781 		np->n_pushedlo = hi;
2782 	else if (hi >= np->n_pushedhi)
2783 		np->n_pushedhi = lo;
2784 	else {
2785 		/*
2786 		 * XXX There's only one range. If the deleted range
2787 		 * is in the middle, pick the largest of the
2788 		 * contiguous ranges that it leaves.
2789 		 */
2790 		if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
2791 			np->n_pushedhi = lo;
2792 		else
2793 			np->n_pushedlo = hi;
2794 	}
2795 #ifdef NFS_DEBUG_COMMIT
2796 	printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2797 	    (unsigned)np->n_pushedhi);
2798 #endif
2799 }
2800 
2801 void
2802 nfs_add_tobecommitted_range(vp, off, len)
2803 	struct vnode *vp;
2804 	off_t off, len;
2805 {
2806 	struct nfsnode *np = VTONFS(vp);
2807 	off_t lo, hi;
2808 
2809 	lo = off;
2810 	hi = lo + len;
2811 
2812 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
2813 		np->n_pushlo = lo;
2814 		np->n_pushhi = hi;
2815 		np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
2816 	} else {
2817 		if (lo < np->n_pushlo)
2818 			np->n_pushlo = lo;
2819 		if (hi > np->n_pushhi)
2820 			np->n_pushhi = hi;
2821 	}
2822 #ifdef NFS_DEBUG_COMMIT
2823 	printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2824 	    (unsigned)np->n_pushhi);
2825 #endif
2826 }
2827 
2828 void
2829 nfs_del_tobecommitted_range(vp, off, len)
2830 	struct vnode *vp;
2831 	off_t off, len;
2832 {
2833 	struct nfsnode *np = VTONFS(vp);
2834 	off_t lo, hi;
2835 
2836 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2837 		return;
2838 
2839 	lo = off;
2840 	hi = lo + len;
2841 
2842 	if (lo > np->n_pushhi || hi < np->n_pushlo)
2843 		return;
2844 
2845 	if (lo <= np->n_pushlo)
2846 		np->n_pushlo = hi;
2847 	else if (hi >= np->n_pushhi)
2848 		np->n_pushhi = lo;
2849 	else {
2850 		/*
2851 		 * XXX There's only one range. If the deleted range
2852 		 * is in the middle, pick the largest of the
2853 		 * contiguous ranges that it leaves.
2854 		 */
2855 		if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
2856 			np->n_pushhi = lo;
2857 		else
2858 			np->n_pushlo = hi;
2859 	}
2860 #ifdef NFS_DEBUG_COMMIT
2861 	printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2862 	    (unsigned)np->n_pushhi);
2863 #endif
2864 }
2865 
2866 /*
2867  * Map errnos to NFS error numbers. For Version 3 also filter out error
2868  * numbers not specified for the associated procedure.
2869  */
2870 int
2871 nfsrv_errmap(nd, err)
2872 	struct nfsrv_descript *nd;
2873 	int err;
2874 {
2875 	const short *defaulterrp, *errp;
2876 
2877 	if (nd->nd_flag & ND_NFSV3) {
2878 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
2879 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2880 		while (*++errp) {
2881 			if (*errp == err)
2882 				return (err);
2883 			else if (*errp > err)
2884 				break;
2885 		}
2886 		return ((int)*defaulterrp);
2887 	    } else
2888 		return (err & 0xffff);
2889 	}
2890 	if (err <= ELAST)
2891 		return ((int)nfsrv_v2errmap[err - 1]);
2892 	return (NFSERR_IO);
2893 }
2894 
2895 u_int32_t
2896 nfs_getxid()
2897 {
2898 	static u_int32_t base;
2899 	static u_int32_t nfs_xid = 0;
2900 	static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER;
2901 	u_int32_t newxid;
2902 
2903 	simple_lock(&nfs_xidlock);
2904 	/*
2905 	 * derive initial xid from system time
2906 	 * XXX time is invalid if root not yet mounted
2907 	 */
2908 	if (__predict_false(!base && (rootvp))) {
2909 		struct timeval tv;
2910 
2911 		microtime(&tv);
2912 		base = tv.tv_sec << 12;
2913 		nfs_xid = base;
2914 	}
2915 
2916 	/*
2917 	 * Skip zero xid if it should ever happen.
2918 	 */
2919 	if (__predict_false(++nfs_xid == 0))
2920 		nfs_xid++;
2921 	newxid = nfs_xid;
2922 	simple_unlock(&nfs_xidlock);
2923 
2924 	return txdr_unsigned(newxid);
2925 }
2926 
2927 /*
2928  * assign a new xid for existing request.
2929  * used for NFSERR_JUKEBOX handling.
2930  */
2931 void
2932 nfs_renewxid(struct nfsreq *req)
2933 {
2934 	u_int32_t xid;
2935 	int off;
2936 
2937 	xid = nfs_getxid();
2938 	if (req->r_nmp->nm_sotype == SOCK_STREAM)
2939 		off = sizeof(u_int32_t); /* RPC record mark */
2940 	else
2941 		off = 0;
2942 
2943 	m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid);
2944 	req->r_xid = xid;
2945 }
2946 
2947 #if defined(NFSSERVER)
2948 int
2949 nfsrv_composefh(struct vnode *vp, nfsrvfh_t *nsfh, bool v3)
2950 {
2951 	int error;
2952 	size_t fhsize;
2953 
2954 	fhsize = NFSD_MAXFHSIZE;
2955 	error = vfs_composefh(vp, (void *)NFSRVFH_DATA(nsfh), &fhsize);
2956 	if (NFSX_FHTOOBIG_P(fhsize, v3)) {
2957 		error = EOPNOTSUPP;
2958 	}
2959 	if (error != 0) {
2960 		return error;
2961 	}
2962 	if (!v3 && fhsize < NFSX_V2FH) {
2963 		memset((char *)NFSRVFH_DATA(nsfh) + fhsize, 0,
2964 		    NFSX_V2FH - fhsize);
2965 		fhsize = NFSX_V2FH;
2966 	}
2967 	if ((fhsize % NFSX_UNSIGNED) != 0) {
2968 		return EOPNOTSUPP;
2969 	}
2970 	nsfh->nsfh_size = fhsize;
2971 	return 0;
2972 }
2973 
2974 int
2975 nfsrv_comparefh(const nfsrvfh_t *fh1, const nfsrvfh_t *fh2)
2976 {
2977 
2978 	if (NFSRVFH_SIZE(fh1) != NFSRVFH_SIZE(fh2)) {
2979 		return NFSRVFH_SIZE(fh2) - NFSRVFH_SIZE(fh1);
2980 	}
2981 	return memcmp(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), NFSRVFH_SIZE(fh1));
2982 }
2983 
2984 void
2985 nfsrv_copyfh(nfsrvfh_t *fh1, const nfsrvfh_t *fh2)
2986 {
2987 	size_t size;
2988 
2989 	fh1->nsfh_size = size = NFSRVFH_SIZE(fh2);
2990 	memcpy(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), size);
2991 }
2992 #endif /* defined(NFSSERVER) */
2993 
2994 #if defined(NFS)
2995 /*
2996  * Set the attribute timeout based on how recently the file has been modified.
2997  */
2998 
2999 time_t
3000 nfs_attrtimeo(struct nfsmount *nmp, struct nfsnode *np)
3001 {
3002 	time_t timeo;
3003 
3004 	if ((nmp->nm_flag & NFSMNT_NOAC) != 0)
3005 		return 0;
3006 
3007 	if (((np)->n_flag & NMODIFIED) != 0)
3008 		return NFS_MINATTRTIMO;
3009 
3010 	timeo = (time_second - np->n_mtime.tv_sec) / 10;
3011 	timeo = max(timeo, NFS_MINATTRTIMO);
3012 	timeo = min(timeo, NFS_MAXATTRTIMO);
3013 	return timeo;
3014 }
3015 #endif /* defined(NFS) */
3016