xref: /netbsd-src/sys/nfs/nfs_subs.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: nfs_subs.c,v 1.199 2008/02/13 09:51:37 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. 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 1.199 2008/02/13 09:51:37 yamt 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/malloc.h>
98 #include <sys/filedesc.h>
99 #include <sys/time.h>
100 #include <sys/dirent.h>
101 #include <sys/once.h>
102 #include <sys/kauth.h>
103 
104 #include <uvm/uvm_extern.h>
105 
106 #include <nfs/rpcv2.h>
107 #include <nfs/nfsproto.h>
108 #include <nfs/nfsnode.h>
109 #include <nfs/nfs.h>
110 #include <nfs/xdr_subs.h>
111 #include <nfs/nfsm_subs.h>
112 #include <nfs/nfsmount.h>
113 #include <nfs/nfsrtt.h>
114 #include <nfs/nfs_var.h>
115 
116 #include <miscfs/specfs/specdev.h>
117 
118 #include <netinet/in.h>
119 #ifdef ISO
120 #include <netiso/iso.h>
121 #endif
122 
123 /*
124  * Data items converted to xdr at startup, since they are constant
125  * This is kinda hokey, but may save a little time doing byte swaps
126  */
127 u_int32_t nfs_xdrneg1;
128 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr,
129 	rpc_mismatch, rpc_auth_unix, rpc_msgaccepted,
130 	rpc_auth_kerb;
131 u_int32_t nfs_prog, nfs_true, nfs_false;
132 
133 /* And other global data */
134 const nfstype nfsv2_type[9] =
135 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON };
136 const nfstype nfsv3_type[9] =
137 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON };
138 const enum vtype nv2tov_type[8] =
139 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON };
140 const enum vtype nv3tov_type[8] =
141 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO };
142 int nfs_ticks;
143 int nfs_commitsize;
144 
145 MALLOC_DEFINE(M_NFSDIROFF, "NFS diroff", "NFS directory cookies");
146 
147 /* NFS client/server stats. */
148 struct nfsstats nfsstats;
149 
150 /*
151  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
152  */
153 const int nfsv3_procid[NFS_NPROCS] = {
154 	NFSPROC_NULL,
155 	NFSPROC_GETATTR,
156 	NFSPROC_SETATTR,
157 	NFSPROC_NOOP,
158 	NFSPROC_LOOKUP,
159 	NFSPROC_READLINK,
160 	NFSPROC_READ,
161 	NFSPROC_NOOP,
162 	NFSPROC_WRITE,
163 	NFSPROC_CREATE,
164 	NFSPROC_REMOVE,
165 	NFSPROC_RENAME,
166 	NFSPROC_LINK,
167 	NFSPROC_SYMLINK,
168 	NFSPROC_MKDIR,
169 	NFSPROC_RMDIR,
170 	NFSPROC_READDIR,
171 	NFSPROC_FSSTAT,
172 	NFSPROC_NOOP,
173 	NFSPROC_NOOP,
174 	NFSPROC_NOOP,
175 	NFSPROC_NOOP,
176 	NFSPROC_NOOP
177 };
178 
179 /*
180  * and the reverse mapping from generic to Version 2 procedure numbers
181  */
182 const int nfsv2_procid[NFS_NPROCS] = {
183 	NFSV2PROC_NULL,
184 	NFSV2PROC_GETATTR,
185 	NFSV2PROC_SETATTR,
186 	NFSV2PROC_LOOKUP,
187 	NFSV2PROC_NOOP,
188 	NFSV2PROC_READLINK,
189 	NFSV2PROC_READ,
190 	NFSV2PROC_WRITE,
191 	NFSV2PROC_CREATE,
192 	NFSV2PROC_MKDIR,
193 	NFSV2PROC_SYMLINK,
194 	NFSV2PROC_CREATE,
195 	NFSV2PROC_REMOVE,
196 	NFSV2PROC_RMDIR,
197 	NFSV2PROC_RENAME,
198 	NFSV2PROC_LINK,
199 	NFSV2PROC_READDIR,
200 	NFSV2PROC_NOOP,
201 	NFSV2PROC_STATFS,
202 	NFSV2PROC_NOOP,
203 	NFSV2PROC_NOOP,
204 	NFSV2PROC_NOOP,
205 	NFSV2PROC_NOOP,
206 };
207 
208 /*
209  * Maps errno values to nfs error numbers.
210  * Use NFSERR_IO as the catch all for ones not specifically defined in
211  * RFC 1094.
212  */
213 static const u_char nfsrv_v2errmap[ELAST] = {
214   NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
215   NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
216   NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
217   NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
218   NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
219   NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
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_IO,	NFSERR_IO,	NFSERR_IO,
226   NFSERR_IO,	NFSERR_IO,	NFSERR_NAMETOL,	NFSERR_IO,	NFSERR_IO,
227   NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
228   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
229   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
230   NFSERR_IO,	NFSERR_IO,
231 };
232 
233 /*
234  * Maps errno values to nfs error numbers.
235  * Although it is not obvious whether or not NFS clients really care if
236  * a returned error value is in the specified list for the procedure, the
237  * safest thing to do is filter them appropriately. For Version 2, the
238  * X/Open XNFS document is the only specification that defines error values
239  * for each RPC (The RFC simply lists all possible error values for all RPCs),
240  * so I have decided to not do this for Version 2.
241  * The first entry is the default error return and the rest are the valid
242  * errors for that RPC in increasing numeric order.
243  */
244 static const short nfsv3err_null[] = {
245 	0,
246 	0,
247 };
248 
249 static const short nfsv3err_getattr[] = {
250 	NFSERR_IO,
251 	NFSERR_IO,
252 	NFSERR_STALE,
253 	NFSERR_BADHANDLE,
254 	NFSERR_SERVERFAULT,
255 	0,
256 };
257 
258 static const short nfsv3err_setattr[] = {
259 	NFSERR_IO,
260 	NFSERR_PERM,
261 	NFSERR_IO,
262 	NFSERR_ACCES,
263 	NFSERR_INVAL,
264 	NFSERR_NOSPC,
265 	NFSERR_ROFS,
266 	NFSERR_DQUOT,
267 	NFSERR_STALE,
268 	NFSERR_BADHANDLE,
269 	NFSERR_NOT_SYNC,
270 	NFSERR_SERVERFAULT,
271 	0,
272 };
273 
274 static const short nfsv3err_lookup[] = {
275 	NFSERR_IO,
276 	NFSERR_NOENT,
277 	NFSERR_IO,
278 	NFSERR_ACCES,
279 	NFSERR_NOTDIR,
280 	NFSERR_NAMETOL,
281 	NFSERR_STALE,
282 	NFSERR_BADHANDLE,
283 	NFSERR_SERVERFAULT,
284 	0,
285 };
286 
287 static const short nfsv3err_access[] = {
288 	NFSERR_IO,
289 	NFSERR_IO,
290 	NFSERR_STALE,
291 	NFSERR_BADHANDLE,
292 	NFSERR_SERVERFAULT,
293 	0,
294 };
295 
296 static const short nfsv3err_readlink[] = {
297 	NFSERR_IO,
298 	NFSERR_IO,
299 	NFSERR_ACCES,
300 	NFSERR_INVAL,
301 	NFSERR_STALE,
302 	NFSERR_BADHANDLE,
303 	NFSERR_NOTSUPP,
304 	NFSERR_SERVERFAULT,
305 	0,
306 };
307 
308 static const short nfsv3err_read[] = {
309 	NFSERR_IO,
310 	NFSERR_IO,
311 	NFSERR_NXIO,
312 	NFSERR_ACCES,
313 	NFSERR_INVAL,
314 	NFSERR_STALE,
315 	NFSERR_BADHANDLE,
316 	NFSERR_SERVERFAULT,
317 	NFSERR_JUKEBOX,
318 	0,
319 };
320 
321 static const short nfsv3err_write[] = {
322 	NFSERR_IO,
323 	NFSERR_IO,
324 	NFSERR_ACCES,
325 	NFSERR_INVAL,
326 	NFSERR_FBIG,
327 	NFSERR_NOSPC,
328 	NFSERR_ROFS,
329 	NFSERR_DQUOT,
330 	NFSERR_STALE,
331 	NFSERR_BADHANDLE,
332 	NFSERR_SERVERFAULT,
333 	NFSERR_JUKEBOX,
334 	0,
335 };
336 
337 static const short nfsv3err_create[] = {
338 	NFSERR_IO,
339 	NFSERR_IO,
340 	NFSERR_ACCES,
341 	NFSERR_EXIST,
342 	NFSERR_NOTDIR,
343 	NFSERR_NOSPC,
344 	NFSERR_ROFS,
345 	NFSERR_NAMETOL,
346 	NFSERR_DQUOT,
347 	NFSERR_STALE,
348 	NFSERR_BADHANDLE,
349 	NFSERR_NOTSUPP,
350 	NFSERR_SERVERFAULT,
351 	0,
352 };
353 
354 static const short nfsv3err_mkdir[] = {
355 	NFSERR_IO,
356 	NFSERR_IO,
357 	NFSERR_ACCES,
358 	NFSERR_EXIST,
359 	NFSERR_NOTDIR,
360 	NFSERR_NOSPC,
361 	NFSERR_ROFS,
362 	NFSERR_NAMETOL,
363 	NFSERR_DQUOT,
364 	NFSERR_STALE,
365 	NFSERR_BADHANDLE,
366 	NFSERR_NOTSUPP,
367 	NFSERR_SERVERFAULT,
368 	0,
369 };
370 
371 static const short nfsv3err_symlink[] = {
372 	NFSERR_IO,
373 	NFSERR_IO,
374 	NFSERR_ACCES,
375 	NFSERR_EXIST,
376 	NFSERR_NOTDIR,
377 	NFSERR_NOSPC,
378 	NFSERR_ROFS,
379 	NFSERR_NAMETOL,
380 	NFSERR_DQUOT,
381 	NFSERR_STALE,
382 	NFSERR_BADHANDLE,
383 	NFSERR_NOTSUPP,
384 	NFSERR_SERVERFAULT,
385 	0,
386 };
387 
388 static const short nfsv3err_mknod[] = {
389 	NFSERR_IO,
390 	NFSERR_IO,
391 	NFSERR_ACCES,
392 	NFSERR_EXIST,
393 	NFSERR_NOTDIR,
394 	NFSERR_NOSPC,
395 	NFSERR_ROFS,
396 	NFSERR_NAMETOL,
397 	NFSERR_DQUOT,
398 	NFSERR_STALE,
399 	NFSERR_BADHANDLE,
400 	NFSERR_NOTSUPP,
401 	NFSERR_SERVERFAULT,
402 	NFSERR_BADTYPE,
403 	0,
404 };
405 
406 static const short nfsv3err_remove[] = {
407 	NFSERR_IO,
408 	NFSERR_NOENT,
409 	NFSERR_IO,
410 	NFSERR_ACCES,
411 	NFSERR_NOTDIR,
412 	NFSERR_ROFS,
413 	NFSERR_NAMETOL,
414 	NFSERR_STALE,
415 	NFSERR_BADHANDLE,
416 	NFSERR_SERVERFAULT,
417 	0,
418 };
419 
420 static const short nfsv3err_rmdir[] = {
421 	NFSERR_IO,
422 	NFSERR_NOENT,
423 	NFSERR_IO,
424 	NFSERR_ACCES,
425 	NFSERR_EXIST,
426 	NFSERR_NOTDIR,
427 	NFSERR_INVAL,
428 	NFSERR_ROFS,
429 	NFSERR_NAMETOL,
430 	NFSERR_NOTEMPTY,
431 	NFSERR_STALE,
432 	NFSERR_BADHANDLE,
433 	NFSERR_NOTSUPP,
434 	NFSERR_SERVERFAULT,
435 	0,
436 };
437 
438 static const short nfsv3err_rename[] = {
439 	NFSERR_IO,
440 	NFSERR_NOENT,
441 	NFSERR_IO,
442 	NFSERR_ACCES,
443 	NFSERR_EXIST,
444 	NFSERR_XDEV,
445 	NFSERR_NOTDIR,
446 	NFSERR_ISDIR,
447 	NFSERR_INVAL,
448 	NFSERR_NOSPC,
449 	NFSERR_ROFS,
450 	NFSERR_MLINK,
451 	NFSERR_NAMETOL,
452 	NFSERR_NOTEMPTY,
453 	NFSERR_DQUOT,
454 	NFSERR_STALE,
455 	NFSERR_BADHANDLE,
456 	NFSERR_NOTSUPP,
457 	NFSERR_SERVERFAULT,
458 	0,
459 };
460 
461 static const short nfsv3err_link[] = {
462 	NFSERR_IO,
463 	NFSERR_IO,
464 	NFSERR_ACCES,
465 	NFSERR_EXIST,
466 	NFSERR_XDEV,
467 	NFSERR_NOTDIR,
468 	NFSERR_INVAL,
469 	NFSERR_NOSPC,
470 	NFSERR_ROFS,
471 	NFSERR_MLINK,
472 	NFSERR_NAMETOL,
473 	NFSERR_DQUOT,
474 	NFSERR_STALE,
475 	NFSERR_BADHANDLE,
476 	NFSERR_NOTSUPP,
477 	NFSERR_SERVERFAULT,
478 	0,
479 };
480 
481 static const short nfsv3err_readdir[] = {
482 	NFSERR_IO,
483 	NFSERR_IO,
484 	NFSERR_ACCES,
485 	NFSERR_NOTDIR,
486 	NFSERR_STALE,
487 	NFSERR_BADHANDLE,
488 	NFSERR_BAD_COOKIE,
489 	NFSERR_TOOSMALL,
490 	NFSERR_SERVERFAULT,
491 	0,
492 };
493 
494 static const short nfsv3err_readdirplus[] = {
495 	NFSERR_IO,
496 	NFSERR_IO,
497 	NFSERR_ACCES,
498 	NFSERR_NOTDIR,
499 	NFSERR_STALE,
500 	NFSERR_BADHANDLE,
501 	NFSERR_BAD_COOKIE,
502 	NFSERR_NOTSUPP,
503 	NFSERR_TOOSMALL,
504 	NFSERR_SERVERFAULT,
505 	0,
506 };
507 
508 static const short nfsv3err_fsstat[] = {
509 	NFSERR_IO,
510 	NFSERR_IO,
511 	NFSERR_STALE,
512 	NFSERR_BADHANDLE,
513 	NFSERR_SERVERFAULT,
514 	0,
515 };
516 
517 static const short nfsv3err_fsinfo[] = {
518 	NFSERR_STALE,
519 	NFSERR_STALE,
520 	NFSERR_BADHANDLE,
521 	NFSERR_SERVERFAULT,
522 	0,
523 };
524 
525 static const short nfsv3err_pathconf[] = {
526 	NFSERR_STALE,
527 	NFSERR_STALE,
528 	NFSERR_BADHANDLE,
529 	NFSERR_SERVERFAULT,
530 	0,
531 };
532 
533 static const short nfsv3err_commit[] = {
534 	NFSERR_IO,
535 	NFSERR_IO,
536 	NFSERR_STALE,
537 	NFSERR_BADHANDLE,
538 	NFSERR_SERVERFAULT,
539 	0,
540 };
541 
542 static const short * const nfsrv_v3errmap[] = {
543 	nfsv3err_null,
544 	nfsv3err_getattr,
545 	nfsv3err_setattr,
546 	nfsv3err_lookup,
547 	nfsv3err_access,
548 	nfsv3err_readlink,
549 	nfsv3err_read,
550 	nfsv3err_write,
551 	nfsv3err_create,
552 	nfsv3err_mkdir,
553 	nfsv3err_symlink,
554 	nfsv3err_mknod,
555 	nfsv3err_remove,
556 	nfsv3err_rmdir,
557 	nfsv3err_rename,
558 	nfsv3err_link,
559 	nfsv3err_readdir,
560 	nfsv3err_readdirplus,
561 	nfsv3err_fsstat,
562 	nfsv3err_fsinfo,
563 	nfsv3err_pathconf,
564 	nfsv3err_commit,
565 };
566 
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) {
952 		if (havebuf) {
953 			/* If the first mbuf with data has external data
954 			 * and there is a previous empty mbuf use it
955 			 * to move the data into.
956 			 */
957 			m2 = m1;
958 			*mdp = m1 = havebuf;
959 			if (m1->m_flags & M_EXT) {
960 				MEXTREMOVE(m1);
961 			}
962 		} else {
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 = m_get(M_WAIT, MT_DATA);
971 			m2->m_ext = m1->m_ext;
972 			m2->m_data = src;
973 			m2->m_len = left;
974 			MCLADDREFERENCE(m1, m2);
975 			MEXTREMOVE(m1);
976 			m2->m_next = m1->m_next;
977 			m1->m_next = m2;
978 		}
979 		m1->m_len = 0;
980 		if (m1->m_flags & M_PKTHDR)
981 			dst = m1->m_pktdat;
982 		else
983 			dst = m1->m_dat;
984 		m1->m_data = dst;
985 	} else {
986 		/*
987 		 * If the first mbuf has no external data
988 		 * move the data to the front of the mbuf.
989 		 */
990 		if (m1->m_flags & M_PKTHDR)
991 			dst = m1->m_pktdat;
992 		else
993 			dst = m1->m_dat;
994 		m1->m_data = dst;
995 		if (dst != src)
996 			memmove(dst, src, left);
997 		dst += left;
998 		m1->m_len = left;
999 		m2 = m1->m_next;
1000 	}
1001 	*cp2 = m1->m_data;
1002 	*dposp = mtod(m1, char *) + siz;
1003 	/*
1004 	 * Loop through mbufs pulling data up into first mbuf until
1005 	 * the first mbuf is full or there is no more data to
1006 	 * pullup.
1007 	 */
1008 	while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) {
1009 		if ((len = min(len, m2->m_len)) != 0)
1010 			memcpy(dst, m2->m_data, len);
1011 		m1->m_len += len;
1012 		dst += len;
1013 		m2->m_data += len;
1014 		m2->m_len -= len;
1015 		m2 = m2->m_next;
1016 	}
1017 	if (m1->m_len < siz)
1018 		return (EBADRPC);
1019 	return (0);
1020 }
1021 
1022 /*
1023  * Advance the position in the mbuf chain.
1024  */
1025 int
1026 nfs_adv(mdp, dposp, offs, left)
1027 	struct mbuf **mdp;
1028 	char **dposp;
1029 	int offs;
1030 	int left;
1031 {
1032 	struct mbuf *m;
1033 	int s;
1034 
1035 	m = *mdp;
1036 	s = left;
1037 	while (s < offs) {
1038 		offs -= s;
1039 		m = m->m_next;
1040 		if (m == NULL)
1041 			return (EBADRPC);
1042 		s = m->m_len;
1043 	}
1044 	*mdp = m;
1045 	*dposp = mtod(m, char *) + offs;
1046 	return (0);
1047 }
1048 
1049 /*
1050  * Copy a string into mbufs for the hard cases...
1051  */
1052 int
1053 nfsm_strtmbuf(mb, bpos, cp, siz)
1054 	struct mbuf **mb;
1055 	char **bpos;
1056 	const char *cp;
1057 	long siz;
1058 {
1059 	struct mbuf *m1 = NULL, *m2;
1060 	long left, xfer, len, tlen;
1061 	u_int32_t *tl;
1062 	int putsize;
1063 
1064 	putsize = 1;
1065 	m2 = *mb;
1066 	left = M_TRAILINGSPACE(m2);
1067 	if (left > 0) {
1068 		tl = ((u_int32_t *)(*bpos));
1069 		*tl++ = txdr_unsigned(siz);
1070 		putsize = 0;
1071 		left -= NFSX_UNSIGNED;
1072 		m2->m_len += NFSX_UNSIGNED;
1073 		if (left > 0) {
1074 			memcpy((void *) tl, cp, left);
1075 			siz -= left;
1076 			cp += left;
1077 			m2->m_len += left;
1078 			left = 0;
1079 		}
1080 	}
1081 	/* Loop around adding mbufs */
1082 	while (siz > 0) {
1083 		m1 = m_get(M_WAIT, MT_DATA);
1084 		MCLAIM(m1, &nfs_mowner);
1085 		if (siz > MLEN)
1086 			m_clget(m1, M_WAIT);
1087 		m1->m_len = NFSMSIZ(m1);
1088 		m2->m_next = m1;
1089 		m2 = m1;
1090 		tl = mtod(m1, u_int32_t *);
1091 		tlen = 0;
1092 		if (putsize) {
1093 			*tl++ = txdr_unsigned(siz);
1094 			m1->m_len -= NFSX_UNSIGNED;
1095 			tlen = NFSX_UNSIGNED;
1096 			putsize = 0;
1097 		}
1098 		if (siz < m1->m_len) {
1099 			len = nfsm_rndup(siz);
1100 			xfer = siz;
1101 			if (xfer < len)
1102 				*(tl+(xfer>>2)) = 0;
1103 		} else {
1104 			xfer = len = m1->m_len;
1105 		}
1106 		memcpy((void *) tl, cp, xfer);
1107 		m1->m_len = len+tlen;
1108 		siz -= xfer;
1109 		cp += xfer;
1110 	}
1111 	*mb = m1;
1112 	*bpos = mtod(m1, char *) + m1->m_len;
1113 	return (0);
1114 }
1115 
1116 /*
1117  * Directory caching routines. They work as follows:
1118  * - a cache is maintained per VDIR nfsnode.
1119  * - for each offset cookie that is exported to userspace, and can
1120  *   thus be thrown back at us as an offset to VOP_READDIR, store
1121  *   information in the cache.
1122  * - cached are:
1123  *   - cookie itself
1124  *   - blocknumber (essentially just a search key in the buffer cache)
1125  *   - entry number in block.
1126  *   - offset cookie of block in which this entry is stored
1127  *   - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
1128  * - entries are looked up in a hash table
1129  * - also maintained is an LRU list of entries, used to determine
1130  *   which ones to delete if the cache grows too large.
1131  * - if 32 <-> 64 translation mode is requested for a filesystem,
1132  *   the cache also functions as a translation table
1133  * - in the translation case, invalidating the cache does not mean
1134  *   flushing it, but just marking entries as invalid, except for
1135  *   the <64bit cookie, 32bitcookie> pair which is still valid, to
1136  *   still be able to use the cache as a translation table.
1137  * - 32 bit cookies are uniquely created by combining the hash table
1138  *   entry value, and one generation count per hash table entry,
1139  *   incremented each time an entry is appended to the chain.
1140  * - the cache is invalidated each time a direcory is modified
1141  * - sanity checks are also done; if an entry in a block turns
1142  *   out not to have a matching cookie, the cache is invalidated
1143  *   and a new block starting from the wanted offset is fetched from
1144  *   the server.
1145  * - directory entries as read from the server are extended to contain
1146  *   the 64bit and, optionally, the 32bit cookies, for sanity checking
1147  *   the cache and exporting them to userspace through the cookie
1148  *   argument to VOP_READDIR.
1149  */
1150 
1151 u_long
1152 nfs_dirhash(off)
1153 	off_t off;
1154 {
1155 	int i;
1156 	char *cp = (char *)&off;
1157 	u_long sum = 0L;
1158 
1159 	for (i = 0 ; i < sizeof (off); i++)
1160 		sum += *cp++;
1161 
1162 	return sum;
1163 }
1164 
1165 #define	_NFSDC_MTX(np)		(&NFSTOV(np)->v_interlock)
1166 #define	NFSDC_LOCK(np)		mutex_enter(_NFSDC_MTX(np))
1167 #define	NFSDC_UNLOCK(np)	mutex_exit(_NFSDC_MTX(np))
1168 #define	NFSDC_ASSERT_LOCKED(np) KASSERT(mutex_owned(_NFSDC_MTX(np)))
1169 
1170 void
1171 nfs_initdircache(vp)
1172 	struct vnode *vp;
1173 {
1174 	struct nfsnode *np = VTONFS(vp);
1175 	struct nfsdirhashhead *dircache;
1176 
1177 	dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF,
1178 	    M_WAITOK, &nfsdirhashmask);
1179 
1180 	NFSDC_LOCK(np);
1181 	if (np->n_dircache == NULL) {
1182 		np->n_dircachesize = 0;
1183 		np->n_dircache = dircache;
1184 		dircache = NULL;
1185 		TAILQ_INIT(&np->n_dirchain);
1186 	}
1187 	NFSDC_UNLOCK(np);
1188 	if (dircache)
1189 		hashdone(dircache, M_NFSDIROFF);
1190 }
1191 
1192 void
1193 nfs_initdirxlatecookie(vp)
1194 	struct vnode *vp;
1195 {
1196 	struct nfsnode *np = VTONFS(vp);
1197 	unsigned *dirgens;
1198 
1199 	KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE);
1200 
1201 	dirgens = kmem_zalloc(NFS_DIRHASHSIZ * sizeof(unsigned), KM_SLEEP);
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 		kmem_free(dirgens, NFS_DIRHASHSIZ * sizeof(unsigned));
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 		kmem_free(ndp, sizeof(*ndp));
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 		kmem_free(ndp, sizeof(*ndp));
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 = kmem_alloc(sizeof(*newndp), KM_SLEEP);
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 			kmem_free(np->n_dirgens,
1477 			    NFS_DIRHASHSIZ * sizeof(unsigned));
1478 			np->n_dirgens = NULL;
1479 		}
1480 	} else {
1481 		TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain)
1482 			ndp->dc_flags |= NFSDC_INVALID;
1483 	}
1484 
1485 	NFSDC_UNLOCK(np);
1486 }
1487 
1488 /*
1489  * Called once before VFS init to initialize shared and
1490  * server-specific data structures.
1491  */
1492 static int
1493 nfs_init0(void)
1494 {
1495 
1496 	nfsrtt.pos = 0;
1497 	rpc_vers = txdr_unsigned(RPC_VER2);
1498 	rpc_call = txdr_unsigned(RPC_CALL);
1499 	rpc_reply = txdr_unsigned(RPC_REPLY);
1500 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1501 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1502 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1503 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1504 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1505 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1506 	nfs_prog = txdr_unsigned(NFS_PROG);
1507 	nfs_true = txdr_unsigned(true);
1508 	nfs_false = txdr_unsigned(false);
1509 	nfs_xdrneg1 = txdr_unsigned(-1);
1510 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1511 	if (nfs_ticks < 1)
1512 		nfs_ticks = 1;
1513 #ifdef NFSSERVER
1514 	nfsrv_init(0);			/* Init server data structures */
1515 	nfsrv_initcache();		/* Init the server request cache */
1516 	{
1517 		extern krwlock_t netexport_lock;	/* XXX */
1518 		rw_init(&netexport_lock);
1519 	}
1520 #endif /* NFSSERVER */
1521 
1522 #if defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY))
1523 	nfsdreq_init();
1524 #endif /* defined(NFSSERVER) || (defined(NFS) && !defined(NFS_V2_ONLY)) */
1525 
1526 	/*
1527 	 * Initialize reply list and start timer
1528 	 */
1529 	TAILQ_INIT(&nfs_reqq);
1530 	nfs_timer_init();
1531 	MOWNER_ATTACH(&nfs_mowner);
1532 
1533 #ifdef NFS
1534 	/* Initialize the kqueue structures */
1535 	nfs_kqinit();
1536 	/* Initialize the iod structures */
1537 	nfs_iodinit();
1538 #endif
1539 	return 0;
1540 }
1541 
1542 void
1543 nfs_init(void)
1544 {
1545 	static ONCE_DECL(nfs_init_once);
1546 
1547 	RUN_ONCE(&nfs_init_once, nfs_init0);
1548 }
1549 
1550 #ifdef NFS
1551 /*
1552  * Called once at VFS init to initialize client-specific data structures.
1553  */
1554 void
1555 nfs_vfs_init()
1556 {
1557 	/* Initialize NFS server / client shared data. */
1558 	nfs_init();
1559 
1560 	nfs_nhinit();			/* Init the nfsnode table */
1561 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1562 }
1563 
1564 void
1565 nfs_vfs_reinit()
1566 {
1567 	nfs_nhreinit();
1568 }
1569 
1570 void
1571 nfs_vfs_done()
1572 {
1573 	nfs_nhdone();
1574 }
1575 
1576 /*
1577  * Attribute cache routines.
1578  * nfs_loadattrcache() - loads or updates the cache contents from attributes
1579  *	that are on the mbuf list
1580  * nfs_getattrcache() - returns valid attributes if found in cache, returns
1581  *	error otherwise
1582  */
1583 
1584 /*
1585  * Load the attribute cache (that lives in the nfsnode entry) with
1586  * the values on the mbuf list and
1587  * Iff vap not NULL
1588  *    copy the attributes to *vaper
1589  */
1590 int
1591 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags)
1592 	struct vnode **vpp;
1593 	struct mbuf **mdp;
1594 	char **dposp;
1595 	struct vattr *vaper;
1596 	int flags;
1597 {
1598 	int32_t t1;
1599 	char *cp2;
1600 	int error = 0;
1601 	struct mbuf *md;
1602 	int v3 = NFS_ISV3(*vpp);
1603 
1604 	md = *mdp;
1605 	t1 = (mtod(md, char *) + md->m_len) - *dposp;
1606 	error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2);
1607 	if (error)
1608 		return (error);
1609 	return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags);
1610 }
1611 
1612 int
1613 nfs_loadattrcache(vpp, fp, vaper, flags)
1614 	struct vnode **vpp;
1615 	struct nfs_fattr *fp;
1616 	struct vattr *vaper;
1617 	int flags;
1618 {
1619 	struct vnode *vp = *vpp;
1620 	struct vattr *vap;
1621 	int v3 = NFS_ISV3(vp);
1622 	enum vtype vtyp;
1623 	u_short vmode;
1624 	struct timespec mtime;
1625 	struct timespec ctime;
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 			spec_node_init(vp, (dev_t)rdev);
1677 		}
1678 		np->n_mtime = mtime;
1679 	}
1680 	uid = fxdr_unsigned(uid_t, fp->fa_uid);
1681 	gid = fxdr_unsigned(gid_t, fp->fa_gid);
1682 	vap = np->n_vattr;
1683 
1684 	/*
1685 	 * Invalidate access cache if uid, gid, mode or ctime changed.
1686 	 */
1687 	if (np->n_accstamp != -1 &&
1688 	    (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode
1689 	    || timespeccmp(&ctime, &vap->va_ctime, !=)))
1690 		np->n_accstamp = -1;
1691 
1692 	vap->va_type = vtyp;
1693 	vap->va_mode = vmode;
1694 	vap->va_rdev = (dev_t)rdev;
1695 	vap->va_mtime = mtime;
1696 	vap->va_ctime = ctime;
1697 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
1698 	switch (vtyp) {
1699 	case VDIR:
1700 		vap->va_blocksize = NFS_DIRFRAGSIZ;
1701 		break;
1702 	case VBLK:
1703 		vap->va_blocksize = BLKDEV_IOSIZE;
1704 		break;
1705 	case VCHR:
1706 		vap->va_blocksize = MAXBSIZE;
1707 		break;
1708 	default:
1709 		vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize :
1710 		    fxdr_unsigned(int32_t, fp->fa2_blocksize);
1711 		break;
1712 	}
1713 	if (v3) {
1714 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1715 		vap->va_uid = uid;
1716 		vap->va_gid = gid;
1717 		vap->va_size = fxdr_hyper(&fp->fa3_size);
1718 		vap->va_bytes = fxdr_hyper(&fp->fa3_used);
1719 		vap->va_fileid = fxdr_hyper(&fp->fa3_fileid);
1720 		fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
1721 		vap->va_flags = 0;
1722 		vap->va_filerev = 0;
1723 	} else {
1724 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1725 		vap->va_uid = uid;
1726 		vap->va_gid = gid;
1727 		vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
1728 		vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks)
1729 		    * NFS_FABLKSIZE;
1730 		vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
1731 		fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
1732 		vap->va_flags = 0;
1733 		vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
1734 		vap->va_filerev = 0;
1735 	}
1736 	if (vap->va_size > VFSTONFS(vp->v_mount)->nm_maxfilesize) {
1737 		return EFBIG;
1738 	}
1739 	if (vap->va_size != np->n_size) {
1740 		if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) {
1741 			vap->va_size = np->n_size;
1742 		} else {
1743 			np->n_size = vap->va_size;
1744 			if (vap->va_type == VREG) {
1745 				/*
1746 				 * we can't free pages if NAC_NOTRUNC because
1747 				 * the pages can be owned by ourselves.
1748 				 */
1749 				if (flags & NAC_NOTRUNC) {
1750 					np->n_flag |= NTRUNCDELAYED;
1751 				} else {
1752 					genfs_node_wrlock(vp);
1753 					mutex_enter(&vp->v_interlock);
1754 					(void)VOP_PUTPAGES(vp, 0,
1755 					    0, PGO_SYNCIO | PGO_CLEANIT |
1756 					    PGO_FREE | PGO_ALLPAGES);
1757 					uvm_vnp_setsize(vp, np->n_size);
1758 					genfs_node_unlock(vp);
1759 				}
1760 			}
1761 		}
1762 	}
1763 	np->n_attrstamp = time_second;
1764 	if (vaper != NULL) {
1765 		memcpy((void *)vaper, (void *)vap, sizeof(*vap));
1766 		if (np->n_flag & NCHG) {
1767 			if (np->n_flag & NACC)
1768 				vaper->va_atime = np->n_atim;
1769 			if (np->n_flag & NUPD)
1770 				vaper->va_mtime = np->n_mtim;
1771 		}
1772 	}
1773 	return (0);
1774 }
1775 
1776 /*
1777  * Check the time stamp
1778  * If the cache is valid, copy contents to *vap and return 0
1779  * otherwise return an error
1780  */
1781 int
1782 nfs_getattrcache(vp, vaper)
1783 	struct vnode *vp;
1784 	struct vattr *vaper;
1785 {
1786 	struct nfsnode *np = VTONFS(vp);
1787 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1788 	struct vattr *vap;
1789 
1790 	if (np->n_attrstamp == 0 ||
1791 	    (time_second - np->n_attrstamp) >= nfs_attrtimeo(nmp, np)) {
1792 		nfsstats.attrcache_misses++;
1793 		return (ENOENT);
1794 	}
1795 	nfsstats.attrcache_hits++;
1796 	vap = np->n_vattr;
1797 	if (vap->va_size != np->n_size) {
1798 		if (vap->va_type == VREG) {
1799 			if ((np->n_flag & NMODIFIED) != 0 &&
1800 			    vap->va_size < np->n_size) {
1801 				vap->va_size = np->n_size;
1802 			} else {
1803 				np->n_size = vap->va_size;
1804 			}
1805 			genfs_node_wrlock(vp);
1806 			uvm_vnp_setsize(vp, np->n_size);
1807 			genfs_node_unlock(vp);
1808 		} else
1809 			np->n_size = vap->va_size;
1810 	}
1811 	memcpy((void *)vaper, (void *)vap, sizeof(struct vattr));
1812 	if (np->n_flag & NCHG) {
1813 		if (np->n_flag & NACC)
1814 			vaper->va_atime = np->n_atim;
1815 		if (np->n_flag & NUPD)
1816 			vaper->va_mtime = np->n_mtim;
1817 	}
1818 	return (0);
1819 }
1820 
1821 void
1822 nfs_delayedtruncate(vp)
1823 	struct vnode *vp;
1824 {
1825 	struct nfsnode *np = VTONFS(vp);
1826 
1827 	if (np->n_flag & NTRUNCDELAYED) {
1828 		np->n_flag &= ~NTRUNCDELAYED;
1829 		genfs_node_wrlock(vp);
1830 		mutex_enter(&vp->v_interlock);
1831 		(void)VOP_PUTPAGES(vp, 0,
1832 		    0, PGO_SYNCIO | PGO_CLEANIT | PGO_FREE | PGO_ALLPAGES);
1833 		uvm_vnp_setsize(vp, np->n_size);
1834 		genfs_node_unlock(vp);
1835 	}
1836 }
1837 
1838 #define	NFS_WCCKLUDGE_TIMEOUT	(24 * 60 * 60)	/* 1 day */
1839 #define	NFS_WCCKLUDGE(nmp, now) \
1840 	(((nmp)->nm_iflag & NFSMNT_WCCKLUDGE) && \
1841 	((now) - (nmp)->nm_wcckludgetime - NFS_WCCKLUDGE_TIMEOUT) < 0)
1842 
1843 /*
1844  * nfs_check_wccdata: check inaccurate wcc_data
1845  *
1846  * => return non-zero if we shouldn't trust the wcc_data.
1847  * => NFS_WCCKLUDGE_TIMEOUT is for the case that the server is "fixed".
1848  */
1849 
1850 int
1851 nfs_check_wccdata(struct nfsnode *np, const struct timespec *ctime,
1852     struct timespec *mtime, bool docheck)
1853 {
1854 	int error = 0;
1855 
1856 #if !defined(NFS_V2_ONLY)
1857 
1858 	if (docheck) {
1859 		struct vnode *vp = NFSTOV(np);
1860 		struct nfsmount *nmp;
1861 		long now = time_second;
1862 		const struct timespec *omtime = &np->n_vattr->va_mtime;
1863 		const struct timespec *octime = &np->n_vattr->va_ctime;
1864 		const char *reason = NULL; /* XXX: gcc */
1865 
1866 		if (timespeccmp(omtime, mtime, <=)) {
1867 			reason = "mtime";
1868 			error = EINVAL;
1869 		}
1870 
1871 		if (vp->v_type == VDIR && timespeccmp(octime, ctime, <=)) {
1872 			reason = "ctime";
1873 			error = EINVAL;
1874 		}
1875 
1876 		nmp = VFSTONFS(vp->v_mount);
1877 		if (error) {
1878 
1879 			/*
1880 			 * despite of the fact that we've updated the file,
1881 			 * timestamps of the file were not updated as we
1882 			 * expected.
1883 			 * it means that the server has incompatible
1884 			 * semantics of timestamps or (more likely)
1885 			 * the server time is not precise enough to
1886 			 * track each modifications.
1887 			 * in that case, we disable wcc processing.
1888 			 *
1889 			 * yes, strictly speaking, we should disable all
1890 			 * caching.  it's a compromise.
1891 			 */
1892 
1893 			mutex_enter(&nmp->nm_lock);
1894 			if (!NFS_WCCKLUDGE(nmp, now)) {
1895 				printf("%s: inaccurate wcc data (%s) detected,"
1896 				    " disabling wcc"
1897 				    " (ctime %u.%09u %u.%09u,"
1898 				    " mtime %u.%09u %u.%09u)\n",
1899 				    vp->v_mount->mnt_stat.f_mntfromname,
1900 				    reason,
1901 				    (unsigned int)octime->tv_sec,
1902 				    (unsigned int)octime->tv_nsec,
1903 				    (unsigned int)ctime->tv_sec,
1904 				    (unsigned int)ctime->tv_nsec,
1905 				    (unsigned int)omtime->tv_sec,
1906 				    (unsigned int)omtime->tv_nsec,
1907 				    (unsigned int)mtime->tv_sec,
1908 				    (unsigned int)mtime->tv_nsec);
1909 			}
1910 			nmp->nm_iflag |= NFSMNT_WCCKLUDGE;
1911 			nmp->nm_wcckludgetime = now;
1912 			mutex_exit(&nmp->nm_lock);
1913 		} else if (NFS_WCCKLUDGE(nmp, now)) {
1914 			error = EPERM; /* XXX */
1915 		} else if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
1916 			mutex_enter(&nmp->nm_lock);
1917 			if (nmp->nm_iflag & NFSMNT_WCCKLUDGE) {
1918 				printf("%s: re-enabling wcc\n",
1919 				    vp->v_mount->mnt_stat.f_mntfromname);
1920 				nmp->nm_iflag &= ~NFSMNT_WCCKLUDGE;
1921 			}
1922 			mutex_exit(&nmp->nm_lock);
1923 		}
1924 	}
1925 
1926 #endif /* !defined(NFS_V2_ONLY) */
1927 
1928 	return error;
1929 }
1930 
1931 /*
1932  * Heuristic to see if the server XDR encodes directory cookies or not.
1933  * it is not supposed to, but a lot of servers may do this. Also, since
1934  * most/all servers will implement V2 as well, it is expected that they
1935  * may return just 32 bits worth of cookie information, so we need to
1936  * find out in which 32 bits this information is available. We do this
1937  * to avoid trouble with emulated binaries that can't handle 64 bit
1938  * directory offsets.
1939  */
1940 
1941 void
1942 nfs_cookieheuristic(vp, flagp, l, cred)
1943 	struct vnode *vp;
1944 	int *flagp;
1945 	struct lwp *l;
1946 	kauth_cred_t cred;
1947 {
1948 	struct uio auio;
1949 	struct iovec aiov;
1950 	char *tbuf, *cp;
1951 	struct dirent *dp;
1952 	off_t *cookies = NULL, *cop;
1953 	int error, eof, nc, len;
1954 
1955 	MALLOC(tbuf, void *, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK);
1956 
1957 	aiov.iov_base = tbuf;
1958 	aiov.iov_len = NFS_DIRFRAGSIZ;
1959 	auio.uio_iov = &aiov;
1960 	auio.uio_iovcnt = 1;
1961 	auio.uio_rw = UIO_READ;
1962 	auio.uio_resid = NFS_DIRFRAGSIZ;
1963 	auio.uio_offset = 0;
1964 	UIO_SETUP_SYSSPACE(&auio);
1965 
1966 	error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc);
1967 
1968 	len = NFS_DIRFRAGSIZ - auio.uio_resid;
1969 	if (error || len == 0) {
1970 		FREE(tbuf, M_TEMP);
1971 		if (cookies)
1972 			free(cookies, M_TEMP);
1973 		return;
1974 	}
1975 
1976 	/*
1977 	 * Find the first valid entry and look at its offset cookie.
1978 	 */
1979 
1980 	cp = tbuf;
1981 	for (cop = cookies; len > 0; len -= dp->d_reclen) {
1982 		dp = (struct dirent *)cp;
1983 		if (dp->d_fileno != 0 && len >= dp->d_reclen) {
1984 			if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) {
1985 				*flagp |= NFSMNT_SWAPCOOKIE;
1986 				nfs_invaldircache(vp, 0);
1987 				nfs_vinvalbuf(vp, 0, cred, l, 1);
1988 			}
1989 			break;
1990 		}
1991 		cop++;
1992 		cp += dp->d_reclen;
1993 	}
1994 
1995 	FREE(tbuf, M_TEMP);
1996 	free(cookies, M_TEMP);
1997 }
1998 #endif /* NFS */
1999 
2000 #ifdef NFSSERVER
2001 /*
2002  * Set up nameidata for a lookup() call and do it.
2003  *
2004  * If pubflag is set, this call is done for a lookup operation on the
2005  * public filehandle. In that case we allow crossing mountpoints and
2006  * absolute pathnames. However, the caller is expected to check that
2007  * the lookup result is within the public fs, and deny access if
2008  * it is not.
2009  */
2010 int
2011 nfs_namei(ndp, nsfh, len, slp, nam, mdp, dposp, retdirp, l, kerbflag, pubflag)
2012 	struct nameidata *ndp;
2013 	nfsrvfh_t *nsfh;
2014 	uint32_t len;
2015 	struct nfssvc_sock *slp;
2016 	struct mbuf *nam;
2017 	struct mbuf **mdp;
2018 	char **dposp;
2019 	struct vnode **retdirp;
2020 	struct lwp *l;
2021 	int kerbflag, pubflag;
2022 {
2023 	int i, rem;
2024 	struct mbuf *md;
2025 	char *fromcp, *tocp, *cp;
2026 	struct iovec aiov;
2027 	struct uio auio;
2028 	struct vnode *dp;
2029 	int error, rdonly, linklen;
2030 	struct componentname *cnp = &ndp->ni_cnd;
2031 
2032 	*retdirp = NULL;
2033 
2034 	if ((len + 1) > MAXPATHLEN)
2035 		return (ENAMETOOLONG);
2036 	if (len == 0)
2037 		return (EACCES);
2038 	cnp->cn_pnbuf = PNBUF_GET();
2039 
2040 	/*
2041 	 * Copy the name from the mbuf list to ndp->ni_pnbuf
2042 	 * and set the various ndp fields appropriately.
2043 	 */
2044 	fromcp = *dposp;
2045 	tocp = cnp->cn_pnbuf;
2046 	md = *mdp;
2047 	rem = mtod(md, char *) + md->m_len - fromcp;
2048 	for (i = 0; i < len; i++) {
2049 		while (rem == 0) {
2050 			md = md->m_next;
2051 			if (md == NULL) {
2052 				error = EBADRPC;
2053 				goto out;
2054 			}
2055 			fromcp = mtod(md, void *);
2056 			rem = md->m_len;
2057 		}
2058 		if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
2059 			error = EACCES;
2060 			goto out;
2061 		}
2062 		*tocp++ = *fromcp++;
2063 		rem--;
2064 	}
2065 	*tocp = '\0';
2066 	*mdp = md;
2067 	*dposp = fromcp;
2068 	len = nfsm_rndup(len)-len;
2069 	if (len > 0) {
2070 		if (rem >= len)
2071 			*dposp += len;
2072 		else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
2073 			goto out;
2074 	}
2075 
2076 	/*
2077 	 * Extract and set starting directory.
2078 	 */
2079 	error = nfsrv_fhtovp(nsfh, false, &dp, ndp->ni_cnd.cn_cred, slp,
2080 	    nam, &rdonly, kerbflag, pubflag);
2081 	if (error)
2082 		goto out;
2083 	if (dp->v_type != VDIR) {
2084 		vrele(dp);
2085 		error = ENOTDIR;
2086 		goto out;
2087 	}
2088 
2089 	if (rdonly)
2090 		cnp->cn_flags |= RDONLY;
2091 
2092 	*retdirp = dp;
2093 
2094 	if (pubflag) {
2095 		/*
2096 		 * Oh joy. For WebNFS, handle those pesky '%' escapes,
2097 		 * and the 'native path' indicator.
2098 		 */
2099 		cp = PNBUF_GET();
2100 		fromcp = cnp->cn_pnbuf;
2101 		tocp = cp;
2102 		if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
2103 			switch ((unsigned char)*fromcp) {
2104 			case WEBNFS_NATIVE_CHAR:
2105 				/*
2106 				 * 'Native' path for us is the same
2107 				 * as a path according to the NFS spec,
2108 				 * just skip the escape char.
2109 				 */
2110 				fromcp++;
2111 				break;
2112 			/*
2113 			 * More may be added in the future, range 0x80-0xff
2114 			 */
2115 			default:
2116 				error = EIO;
2117 				vrele(dp);
2118 				PNBUF_PUT(cp);
2119 				goto out;
2120 			}
2121 		}
2122 		/*
2123 		 * Translate the '%' escapes, URL-style.
2124 		 */
2125 		while (*fromcp != '\0') {
2126 			if (*fromcp == WEBNFS_ESC_CHAR) {
2127 				if (fromcp[1] != '\0' && fromcp[2] != '\0') {
2128 					fromcp++;
2129 					*tocp++ = HEXSTRTOI(fromcp);
2130 					fromcp += 2;
2131 					continue;
2132 				} else {
2133 					error = ENOENT;
2134 					vrele(dp);
2135 					PNBUF_PUT(cp);
2136 					goto out;
2137 				}
2138 			} else
2139 				*tocp++ = *fromcp++;
2140 		}
2141 		*tocp = '\0';
2142 		PNBUF_PUT(cnp->cn_pnbuf);
2143 		cnp->cn_pnbuf = cp;
2144 	}
2145 
2146 	ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1;
2147 	ndp->ni_segflg = UIO_SYSSPACE;
2148 	ndp->ni_rootdir = rootvnode;
2149 	ndp->ni_erootdir = NULL;
2150 
2151 	if (pubflag) {
2152 		ndp->ni_loopcnt = 0;
2153 		if (cnp->cn_pnbuf[0] == '/')
2154 			dp = rootvnode;
2155 	} else {
2156 		cnp->cn_flags |= NOCROSSMOUNT;
2157 	}
2158 
2159 	VREF(dp);
2160 	vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
2161 
2162     for (;;) {
2163 	cnp->cn_nameptr = cnp->cn_pnbuf;
2164 	ndp->ni_startdir = dp;
2165 
2166 	/*
2167 	 * And call lookup() to do the real work
2168 	 */
2169 	error = lookup(ndp);
2170 	if (error) {
2171 		if (ndp->ni_dvp) {
2172 			vput(ndp->ni_dvp);
2173 		}
2174 		PNBUF_PUT(cnp->cn_pnbuf);
2175 		return (error);
2176 	}
2177 
2178 	/*
2179 	 * Check for encountering a symbolic link
2180 	 */
2181 	if ((cnp->cn_flags & ISSYMLINK) == 0) {
2182 		if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp) {
2183 			if (ndp->ni_dvp == ndp->ni_vp) {
2184 				vrele(ndp->ni_dvp);
2185 			} else {
2186 				vput(ndp->ni_dvp);
2187 			}
2188 		}
2189 		if (cnp->cn_flags & (SAVENAME | SAVESTART))
2190 			cnp->cn_flags |= HASBUF;
2191 		else
2192 			PNBUF_PUT(cnp->cn_pnbuf);
2193 		return (0);
2194 	} else {
2195 		if (!pubflag) {
2196 			error = EINVAL;
2197 			break;
2198 		}
2199 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
2200 			error = ELOOP;
2201 			break;
2202 		}
2203 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
2204 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred);
2205 			if (error != 0)
2206 				break;
2207 		}
2208 		if (ndp->ni_pathlen > 1)
2209 			cp = PNBUF_GET();
2210 		else
2211 			cp = cnp->cn_pnbuf;
2212 		aiov.iov_base = cp;
2213 		aiov.iov_len = MAXPATHLEN;
2214 		auio.uio_iov = &aiov;
2215 		auio.uio_iovcnt = 1;
2216 		auio.uio_offset = 0;
2217 		auio.uio_rw = UIO_READ;
2218 		auio.uio_resid = MAXPATHLEN;
2219 		UIO_SETUP_SYSSPACE(&auio);
2220 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
2221 		if (error) {
2222 badlink:
2223 			if (ndp->ni_pathlen > 1)
2224 				PNBUF_PUT(cp);
2225 			break;
2226 		}
2227 		linklen = MAXPATHLEN - auio.uio_resid;
2228 		if (linklen == 0) {
2229 			error = ENOENT;
2230 			goto badlink;
2231 		}
2232 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
2233 			error = ENAMETOOLONG;
2234 			goto badlink;
2235 		}
2236 		if (ndp->ni_pathlen > 1) {
2237 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
2238 			PNBUF_PUT(cnp->cn_pnbuf);
2239 			cnp->cn_pnbuf = cp;
2240 		} else
2241 			cnp->cn_pnbuf[linklen] = '\0';
2242 		ndp->ni_pathlen += linklen;
2243 		vput(ndp->ni_vp);
2244 		dp = ndp->ni_dvp;
2245 
2246 		/*
2247 		 * Check if root directory should replace current directory.
2248 		 */
2249 		if (cnp->cn_pnbuf[0] == '/') {
2250 			vput(dp);
2251 			dp = ndp->ni_rootdir;
2252 			VREF(dp);
2253 			vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
2254 		}
2255 	}
2256    }
2257 	vput(ndp->ni_dvp);
2258 	vput(ndp->ni_vp);
2259 	ndp->ni_vp = NULL;
2260 out:
2261 	PNBUF_PUT(cnp->cn_pnbuf);
2262 	return (error);
2263 }
2264 #endif /* NFSSERVER */
2265 
2266 /*
2267  * A fiddled version of m_adj() that ensures null fill to a 32-bit
2268  * boundary and only trims off the back end
2269  *
2270  * 1. trim off 'len' bytes as m_adj(mp, -len).
2271  * 2. add zero-padding 'nul' bytes at the end of the mbuf chain.
2272  */
2273 void
2274 nfs_zeropad(mp, len, nul)
2275 	struct mbuf *mp;
2276 	int len;
2277 	int nul;
2278 {
2279 	struct mbuf *m;
2280 	int count;
2281 
2282 	/*
2283 	 * Trim from tail.  Scan the mbuf chain,
2284 	 * calculating its length and finding the last mbuf.
2285 	 * If the adjustment only affects this mbuf, then just
2286 	 * adjust and return.  Otherwise, rescan and truncate
2287 	 * after the remaining size.
2288 	 */
2289 	count = 0;
2290 	m = mp;
2291 	for (;;) {
2292 		count += m->m_len;
2293 		if (m->m_next == NULL)
2294 			break;
2295 		m = m->m_next;
2296 	}
2297 
2298 	KDASSERT(count >= len);
2299 
2300 	if (m->m_len >= len) {
2301 		m->m_len -= len;
2302 	} else {
2303 		count -= len;
2304 		/*
2305 		 * Correct length for chain is "count".
2306 		 * Find the mbuf with last data, adjust its length,
2307 		 * and toss data from remaining mbufs on chain.
2308 		 */
2309 		for (m = mp; m; m = m->m_next) {
2310 			if (m->m_len >= count) {
2311 				m->m_len = count;
2312 				break;
2313 			}
2314 			count -= m->m_len;
2315 		}
2316 		KASSERT(m && m->m_next);
2317 		m_freem(m->m_next);
2318 		m->m_next = NULL;
2319 	}
2320 
2321 	KDASSERT(m->m_next == NULL);
2322 
2323 	/*
2324 	 * zero-padding.
2325 	 */
2326 	if (nul > 0) {
2327 		char *cp;
2328 		int i;
2329 
2330 		if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) {
2331 			struct mbuf *n;
2332 
2333 			KDASSERT(MLEN >= nul);
2334 			n = m_get(M_WAIT, MT_DATA);
2335 			MCLAIM(n, &nfs_mowner);
2336 			n->m_len = nul;
2337 			n->m_next = NULL;
2338 			m->m_next = n;
2339 			cp = mtod(n, void *);
2340 		} else {
2341 			cp = mtod(m, char *) + m->m_len;
2342 			m->m_len += nul;
2343 		}
2344 		for (i = 0; i < nul; i++)
2345 			*cp++ = '\0';
2346 	}
2347 	return;
2348 }
2349 
2350 /*
2351  * Make these functions instead of macros, so that the kernel text size
2352  * doesn't get too big...
2353  */
2354 void
2355 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp)
2356 	struct nfsrv_descript *nfsd;
2357 	int before_ret;
2358 	struct vattr *before_vap;
2359 	int after_ret;
2360 	struct vattr *after_vap;
2361 	struct mbuf **mbp;
2362 	char **bposp;
2363 {
2364 	struct mbuf *mb = *mbp;
2365 	char *bpos = *bposp;
2366 	u_int32_t *tl;
2367 
2368 	if (before_ret) {
2369 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2370 		*tl = nfs_false;
2371 	} else {
2372 		nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
2373 		*tl++ = nfs_true;
2374 		txdr_hyper(before_vap->va_size, tl);
2375 		tl += 2;
2376 		txdr_nfsv3time(&(before_vap->va_mtime), tl);
2377 		tl += 2;
2378 		txdr_nfsv3time(&(before_vap->va_ctime), tl);
2379 	}
2380 	*bposp = bpos;
2381 	*mbp = mb;
2382 	nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
2383 }
2384 
2385 void
2386 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp)
2387 	struct nfsrv_descript *nfsd;
2388 	int after_ret;
2389 	struct vattr *after_vap;
2390 	struct mbuf **mbp;
2391 	char **bposp;
2392 {
2393 	struct mbuf *mb = *mbp;
2394 	char *bpos = *bposp;
2395 	u_int32_t *tl;
2396 	struct nfs_fattr *fp;
2397 
2398 	if (after_ret) {
2399 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2400 		*tl = nfs_false;
2401 	} else {
2402 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
2403 		*tl++ = nfs_true;
2404 		fp = (struct nfs_fattr *)tl;
2405 		nfsm_srvfattr(nfsd, after_vap, fp);
2406 	}
2407 	*mbp = mb;
2408 	*bposp = bpos;
2409 }
2410 
2411 void
2412 nfsm_srvfattr(nfsd, vap, fp)
2413 	struct nfsrv_descript *nfsd;
2414 	struct vattr *vap;
2415 	struct nfs_fattr *fp;
2416 {
2417 
2418 	fp->fa_nlink = txdr_unsigned(vap->va_nlink);
2419 	fp->fa_uid = txdr_unsigned(vap->va_uid);
2420 	fp->fa_gid = txdr_unsigned(vap->va_gid);
2421 	if (nfsd->nd_flag & ND_NFSV3) {
2422 		fp->fa_type = vtonfsv3_type(vap->va_type);
2423 		fp->fa_mode = vtonfsv3_mode(vap->va_mode);
2424 		txdr_hyper(vap->va_size, &fp->fa3_size);
2425 		txdr_hyper(vap->va_bytes, &fp->fa3_used);
2426 		fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
2427 		fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
2428 		fp->fa3_fsid.nfsuquad[0] = 0;
2429 		fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
2430 		txdr_hyper(vap->va_fileid, &fp->fa3_fileid);
2431 		txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
2432 		txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
2433 		txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
2434 	} else {
2435 		fp->fa_type = vtonfsv2_type(vap->va_type);
2436 		fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
2437 		fp->fa2_size = txdr_unsigned(vap->va_size);
2438 		fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
2439 		if (vap->va_type == VFIFO)
2440 			fp->fa2_rdev = 0xffffffff;
2441 		else
2442 			fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
2443 		fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
2444 		fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
2445 		fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
2446 		txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
2447 		txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
2448 		txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
2449 	}
2450 }
2451 
2452 #ifdef NFSSERVER
2453 /*
2454  * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
2455  * 	- look up fsid in mount list (if not found ret error)
2456  *	- get vp and export rights by calling VFS_FHTOVP()
2457  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2458  *	- if not lockflag unlock it with VOP_UNLOCK()
2459  */
2460 int
2461 nfsrv_fhtovp(nfsrvfh_t *nsfh, int lockflag, struct vnode **vpp,
2462     kauth_cred_t cred, struct nfssvc_sock *slp, struct mbuf *nam, int *rdonlyp,
2463     int kerbflag, int pubflag)
2464 {
2465 	struct mount *mp;
2466 	kauth_cred_t credanon;
2467 	int error, exflags;
2468 	struct sockaddr_in *saddr;
2469 	fhandle_t *fhp;
2470 
2471 	fhp = NFSRVFH_FHANDLE(nsfh);
2472 	*vpp = (struct vnode *)0;
2473 
2474 	if (nfs_ispublicfh(nsfh)) {
2475 		if (!pubflag || !nfs_pub.np_valid)
2476 			return (ESTALE);
2477 		fhp = nfs_pub.np_handle;
2478 	}
2479 
2480 	error = netexport_check(&fhp->fh_fsid, nam, &mp, &exflags, &credanon);
2481 	if (error) {
2482 		return error;
2483 	}
2484 
2485 	error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
2486 	if (error)
2487 		return (error);
2488 
2489 	if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
2490 		saddr = mtod(nam, struct sockaddr_in *);
2491 		if ((saddr->sin_family == AF_INET) &&
2492 		    ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
2493 			vput(*vpp);
2494 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2495 		}
2496 #ifdef INET6
2497 		if ((saddr->sin_family == AF_INET6) &&
2498 		    ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) {
2499 			vput(*vpp);
2500 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2501 		}
2502 #endif
2503 	}
2504 	/*
2505 	 * Check/setup credentials.
2506 	 */
2507 	if (exflags & MNT_EXKERB) {
2508 		if (!kerbflag) {
2509 			vput(*vpp);
2510 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2511 		}
2512 	} else if (kerbflag) {
2513 		vput(*vpp);
2514 		return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2515 	} else if (kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER,
2516 		    NULL) == 0 || (exflags & MNT_EXPORTANON)) {
2517 		kauth_cred_clone(credanon, cred);
2518 	}
2519 	if (exflags & MNT_EXRDONLY)
2520 		*rdonlyp = 1;
2521 	else
2522 		*rdonlyp = 0;
2523 	if (!lockflag)
2524 		VOP_UNLOCK(*vpp, 0);
2525 	return (0);
2526 }
2527 
2528 /*
2529  * WebNFS: check if a filehandle is a public filehandle. For v3, this
2530  * means a length of 0, for v2 it means all zeroes.
2531  */
2532 int
2533 nfs_ispublicfh(const nfsrvfh_t *nsfh)
2534 {
2535 	const char *cp = (const void *)(NFSRVFH_DATA(nsfh));
2536 	int i;
2537 
2538 	if (NFSRVFH_SIZE(nsfh) == 0) {
2539 		return true;
2540 	}
2541 	if (NFSRVFH_SIZE(nsfh) != NFSX_V2FH) {
2542 		return false;
2543 	}
2544 	for (i = 0; i < NFSX_V2FH; i++)
2545 		if (*cp++ != 0)
2546 			return false;
2547 	return true;
2548 }
2549 #endif /* NFSSERVER */
2550 
2551 /*
2552  * This function compares two net addresses by family and returns true
2553  * if they are the same host.
2554  * If there is any doubt, return false.
2555  * The AF_INET family is handled as a special case so that address mbufs
2556  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
2557  */
2558 int
2559 netaddr_match(family, haddr, nam)
2560 	int family;
2561 	union nethostaddr *haddr;
2562 	struct mbuf *nam;
2563 {
2564 	struct sockaddr_in *inetaddr;
2565 
2566 	switch (family) {
2567 	case AF_INET:
2568 		inetaddr = mtod(nam, struct sockaddr_in *);
2569 		if (inetaddr->sin_family == AF_INET &&
2570 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
2571 			return (1);
2572 		break;
2573 #ifdef INET6
2574 	case AF_INET6:
2575 	    {
2576 		struct sockaddr_in6 *sin6_1, *sin6_2;
2577 
2578 		sin6_1 = mtod(nam, struct sockaddr_in6 *);
2579 		sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
2580 		if (sin6_1->sin6_family == AF_INET6 &&
2581 		    IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
2582 			return 1;
2583 	    }
2584 #endif
2585 #ifdef ISO
2586 	case AF_ISO:
2587 	    {
2588 		struct sockaddr_iso *isoaddr1, *isoaddr2;
2589 
2590 		isoaddr1 = mtod(nam, struct sockaddr_iso *);
2591 		isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *);
2592 		if (isoaddr1->siso_family == AF_ISO &&
2593 		    isoaddr1->siso_nlen > 0 &&
2594 		    isoaddr1->siso_nlen == isoaddr2->siso_nlen &&
2595 		    SAME_ISOADDR(isoaddr1, isoaddr2))
2596 			return (1);
2597 		break;
2598 	    }
2599 #endif	/* ISO */
2600 	default:
2601 		break;
2602 	};
2603 	return (0);
2604 }
2605 
2606 /*
2607  * The write verifier has changed (probably due to a server reboot), so all
2608  * PG_NEEDCOMMIT pages will have to be written again. Since they are marked
2609  * as dirty or are being written out just now, all this takes is clearing
2610  * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for
2611  * the mount point.
2612  */
2613 void
2614 nfs_clearcommit(mp)
2615 	struct mount *mp;
2616 {
2617 	struct vnode *vp;
2618 	struct nfsnode *np;
2619 	struct vm_page *pg;
2620 	struct nfsmount *nmp = VFSTONFS(mp);
2621 
2622 	rw_enter(&nmp->nm_writeverflock, RW_WRITER);
2623 	mutex_enter(&mntvnode_lock);
2624 	TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2625 		KASSERT(vp->v_mount == mp);
2626 		if (vp->v_type != VREG)
2627 			continue;
2628 		np = VTONFS(vp);
2629 		np->n_pushlo = np->n_pushhi = np->n_pushedlo =
2630 		    np->n_pushedhi = 0;
2631 		np->n_commitflags &=
2632 		    ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
2633 		mutex_enter(&vp->v_uobj.vmobjlock);
2634 		TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) {
2635 			pg->flags &= ~PG_NEEDCOMMIT;
2636 		}
2637 		mutex_exit(&vp->v_uobj.vmobjlock);
2638 	}
2639 	mutex_exit(&mntvnode_lock);
2640 	mutex_enter(&nmp->nm_lock);
2641 	nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF;
2642 	mutex_exit(&nmp->nm_lock);
2643 	rw_exit(&nmp->nm_writeverflock);
2644 }
2645 
2646 void
2647 nfs_merge_commit_ranges(vp)
2648 	struct vnode *vp;
2649 {
2650 	struct nfsnode *np = VTONFS(vp);
2651 
2652 	KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID);
2653 
2654 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2655 		np->n_pushedlo = np->n_pushlo;
2656 		np->n_pushedhi = np->n_pushhi;
2657 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2658 	} else {
2659 		if (np->n_pushlo < np->n_pushedlo)
2660 			np->n_pushedlo = np->n_pushlo;
2661 		if (np->n_pushhi > np->n_pushedhi)
2662 			np->n_pushedhi = np->n_pushhi;
2663 	}
2664 
2665 	np->n_pushlo = np->n_pushhi = 0;
2666 	np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
2667 
2668 #ifdef NFS_DEBUG_COMMIT
2669 	printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2670 	    (unsigned)np->n_pushedhi);
2671 #endif
2672 }
2673 
2674 int
2675 nfs_in_committed_range(vp, off, len)
2676 	struct vnode *vp;
2677 	off_t off, len;
2678 {
2679 	struct nfsnode *np = VTONFS(vp);
2680 	off_t lo, hi;
2681 
2682 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2683 		return 0;
2684 	lo = off;
2685 	hi = lo + len;
2686 
2687 	return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
2688 }
2689 
2690 int
2691 nfs_in_tobecommitted_range(vp, off, len)
2692 	struct vnode *vp;
2693 	off_t off, len;
2694 {
2695 	struct nfsnode *np = VTONFS(vp);
2696 	off_t lo, hi;
2697 
2698 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2699 		return 0;
2700 	lo = off;
2701 	hi = lo + len;
2702 
2703 	return (lo >= np->n_pushlo && hi <= np->n_pushhi);
2704 }
2705 
2706 void
2707 nfs_add_committed_range(vp, off, len)
2708 	struct vnode *vp;
2709 	off_t off, len;
2710 {
2711 	struct nfsnode *np = VTONFS(vp);
2712 	off_t lo, hi;
2713 
2714 	lo = off;
2715 	hi = lo + len;
2716 
2717 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2718 		np->n_pushedlo = lo;
2719 		np->n_pushedhi = hi;
2720 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2721 	} else {
2722 		if (hi > np->n_pushedhi)
2723 			np->n_pushedhi = hi;
2724 		if (lo < np->n_pushedlo)
2725 			np->n_pushedlo = lo;
2726 	}
2727 #ifdef NFS_DEBUG_COMMIT
2728 	printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2729 	    (unsigned)np->n_pushedhi);
2730 #endif
2731 }
2732 
2733 void
2734 nfs_del_committed_range(vp, off, len)
2735 	struct vnode *vp;
2736 	off_t off, len;
2737 {
2738 	struct nfsnode *np = VTONFS(vp);
2739 	off_t lo, hi;
2740 
2741 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2742 		return;
2743 
2744 	lo = off;
2745 	hi = lo + len;
2746 
2747 	if (lo > np->n_pushedhi || hi < np->n_pushedlo)
2748 		return;
2749 	if (lo <= np->n_pushedlo)
2750 		np->n_pushedlo = hi;
2751 	else if (hi >= np->n_pushedhi)
2752 		np->n_pushedhi = lo;
2753 	else {
2754 		/*
2755 		 * XXX There's only one range. If the deleted range
2756 		 * is in the middle, pick the largest of the
2757 		 * contiguous ranges that it leaves.
2758 		 */
2759 		if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
2760 			np->n_pushedhi = lo;
2761 		else
2762 			np->n_pushedlo = hi;
2763 	}
2764 #ifdef NFS_DEBUG_COMMIT
2765 	printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2766 	    (unsigned)np->n_pushedhi);
2767 #endif
2768 }
2769 
2770 void
2771 nfs_add_tobecommitted_range(vp, off, len)
2772 	struct vnode *vp;
2773 	off_t off, len;
2774 {
2775 	struct nfsnode *np = VTONFS(vp);
2776 	off_t lo, hi;
2777 
2778 	lo = off;
2779 	hi = lo + len;
2780 
2781 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
2782 		np->n_pushlo = lo;
2783 		np->n_pushhi = hi;
2784 		np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
2785 	} else {
2786 		if (lo < np->n_pushlo)
2787 			np->n_pushlo = lo;
2788 		if (hi > np->n_pushhi)
2789 			np->n_pushhi = hi;
2790 	}
2791 #ifdef NFS_DEBUG_COMMIT
2792 	printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2793 	    (unsigned)np->n_pushhi);
2794 #endif
2795 }
2796 
2797 void
2798 nfs_del_tobecommitted_range(vp, off, len)
2799 	struct vnode *vp;
2800 	off_t off, len;
2801 {
2802 	struct nfsnode *np = VTONFS(vp);
2803 	off_t lo, hi;
2804 
2805 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2806 		return;
2807 
2808 	lo = off;
2809 	hi = lo + len;
2810 
2811 	if (lo > np->n_pushhi || hi < np->n_pushlo)
2812 		return;
2813 
2814 	if (lo <= np->n_pushlo)
2815 		np->n_pushlo = hi;
2816 	else if (hi >= np->n_pushhi)
2817 		np->n_pushhi = lo;
2818 	else {
2819 		/*
2820 		 * XXX There's only one range. If the deleted range
2821 		 * is in the middle, pick the largest of the
2822 		 * contiguous ranges that it leaves.
2823 		 */
2824 		if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
2825 			np->n_pushhi = lo;
2826 		else
2827 			np->n_pushlo = hi;
2828 	}
2829 #ifdef NFS_DEBUG_COMMIT
2830 	printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2831 	    (unsigned)np->n_pushhi);
2832 #endif
2833 }
2834 
2835 /*
2836  * Map errnos to NFS error numbers. For Version 3 also filter out error
2837  * numbers not specified for the associated procedure.
2838  */
2839 int
2840 nfsrv_errmap(nd, err)
2841 	struct nfsrv_descript *nd;
2842 	int err;
2843 {
2844 	const short *defaulterrp, *errp;
2845 
2846 	if (nd->nd_flag & ND_NFSV3) {
2847 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
2848 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2849 		while (*++errp) {
2850 			if (*errp == err)
2851 				return (err);
2852 			else if (*errp > err)
2853 				break;
2854 		}
2855 		return ((int)*defaulterrp);
2856 	    } else
2857 		return (err & 0xffff);
2858 	}
2859 	if (err <= ELAST)
2860 		return ((int)nfsrv_v2errmap[err - 1]);
2861 	return (NFSERR_IO);
2862 }
2863 
2864 u_int32_t
2865 nfs_getxid()
2866 {
2867 	static u_int32_t base;
2868 	static u_int32_t nfs_xid = 0;
2869 	static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER;
2870 	u_int32_t newxid;
2871 
2872 	simple_lock(&nfs_xidlock);
2873 	/*
2874 	 * derive initial xid from system time
2875 	 * XXX time is invalid if root not yet mounted
2876 	 */
2877 	if (__predict_false(!base && (rootvp))) {
2878 		struct timeval tv;
2879 
2880 		microtime(&tv);
2881 		base = tv.tv_sec << 12;
2882 		nfs_xid = base;
2883 	}
2884 
2885 	/*
2886 	 * Skip zero xid if it should ever happen.
2887 	 */
2888 	if (__predict_false(++nfs_xid == 0))
2889 		nfs_xid++;
2890 	newxid = nfs_xid;
2891 	simple_unlock(&nfs_xidlock);
2892 
2893 	return txdr_unsigned(newxid);
2894 }
2895 
2896 /*
2897  * assign a new xid for existing request.
2898  * used for NFSERR_JUKEBOX handling.
2899  */
2900 void
2901 nfs_renewxid(struct nfsreq *req)
2902 {
2903 	u_int32_t xid;
2904 	int off;
2905 
2906 	xid = nfs_getxid();
2907 	if (req->r_nmp->nm_sotype == SOCK_STREAM)
2908 		off = sizeof(u_int32_t); /* RPC record mark */
2909 	else
2910 		off = 0;
2911 
2912 	m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid);
2913 	req->r_xid = xid;
2914 }
2915 
2916 #if defined(NFSSERVER)
2917 int
2918 nfsrv_composefh(struct vnode *vp, nfsrvfh_t *nsfh, bool v3)
2919 {
2920 	int error;
2921 	size_t fhsize;
2922 
2923 	fhsize = NFSD_MAXFHSIZE;
2924 	error = vfs_composefh(vp, (void *)NFSRVFH_DATA(nsfh), &fhsize);
2925 	if (NFSX_FHTOOBIG_P(fhsize, v3)) {
2926 		error = EOPNOTSUPP;
2927 	}
2928 	if (error != 0) {
2929 		return error;
2930 	}
2931 	if (!v3 && fhsize < NFSX_V2FH) {
2932 		memset((char *)NFSRVFH_DATA(nsfh) + fhsize, 0,
2933 		    NFSX_V2FH - fhsize);
2934 		fhsize = NFSX_V2FH;
2935 	}
2936 	if ((fhsize % NFSX_UNSIGNED) != 0) {
2937 		return EOPNOTSUPP;
2938 	}
2939 	nsfh->nsfh_size = fhsize;
2940 	return 0;
2941 }
2942 
2943 int
2944 nfsrv_comparefh(const nfsrvfh_t *fh1, const nfsrvfh_t *fh2)
2945 {
2946 
2947 	if (NFSRVFH_SIZE(fh1) != NFSRVFH_SIZE(fh2)) {
2948 		return NFSRVFH_SIZE(fh2) - NFSRVFH_SIZE(fh1);
2949 	}
2950 	return memcmp(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), NFSRVFH_SIZE(fh1));
2951 }
2952 
2953 void
2954 nfsrv_copyfh(nfsrvfh_t *fh1, const nfsrvfh_t *fh2)
2955 {
2956 	size_t size;
2957 
2958 	fh1->nsfh_size = size = NFSRVFH_SIZE(fh2);
2959 	memcpy(NFSRVFH_DATA(fh1), NFSRVFH_DATA(fh2), size);
2960 }
2961 #endif /* defined(NFSSERVER) */
2962 
2963 #if defined(NFS)
2964 /*
2965  * Set the attribute timeout based on how recently the file has been modified.
2966  */
2967 
2968 time_t
2969 nfs_attrtimeo(struct nfsmount *nmp, struct nfsnode *np)
2970 {
2971 	time_t timeo;
2972 
2973 	if ((nmp->nm_flag & NFSMNT_NOAC) != 0)
2974 		return 0;
2975 
2976 	if (((np)->n_flag & NMODIFIED) != 0)
2977 		return NFS_MINATTRTIMO;
2978 
2979 	timeo = (time_second - np->n_mtime.tv_sec) / 10;
2980 	timeo = max(timeo, NFS_MINATTRTIMO);
2981 	timeo = min(timeo, NFS_MAXATTRTIMO);
2982 	return timeo;
2983 }
2984 #endif /* defined(NFS) */
2985