xref: /netbsd-src/sys/nfs/nfs_subs.c (revision d20841bb642898112fe68f0ad3f7b26dddf56f07)
1 /*	$NetBSD: nfs_subs.c,v 1.130 2003/11/29 19:31:13 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 1.130 2003/11/29 19:31:13 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/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 
101 #include <uvm/uvm_extern.h>
102 
103 #include <nfs/rpcv2.h>
104 #include <nfs/nfsproto.h>
105 #include <nfs/nfsnode.h>
106 #include <nfs/nfs.h>
107 #include <nfs/xdr_subs.h>
108 #include <nfs/nfsm_subs.h>
109 #include <nfs/nfsmount.h>
110 #include <nfs/nqnfs.h>
111 #include <nfs/nfsrtt.h>
112 #include <nfs/nfs_var.h>
113 
114 #include <miscfs/specfs/specdev.h>
115 
116 #include <netinet/in.h>
117 #ifdef ISO
118 #include <netiso/iso.h>
119 #endif
120 
121 /*
122  * Data items converted to xdr at startup, since they are constant
123  * This is kinda hokey, but may save a little time doing byte swaps
124  */
125 u_int32_t nfs_xdrneg1;
126 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr,
127 	rpc_mismatch, rpc_auth_unix, rpc_msgaccepted,
128 	rpc_auth_kerb;
129 u_int32_t nfs_prog, nqnfs_prog, nfs_true, nfs_false;
130 
131 /* And other global data */
132 const nfstype nfsv2_type[9] =
133 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON, NFCHR, NFNON };
134 const nfstype nfsv3_type[9] =
135 	{ NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK, NFFIFO, NFNON };
136 const enum vtype nv2tov_type[8] =
137 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON };
138 const enum vtype nv3tov_type[8] =
139 	{ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO };
140 int nfs_ticks;
141 int nfs_commitsize;
142 
143 MALLOC_DEFINE(M_NFSDIROFF, "NFS diroff", "NFS directory cookies");
144 
145 /* NFS client/server stats. */
146 struct nfsstats nfsstats;
147 
148 /*
149  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
150  */
151 const int nfsv3_procid[NFS_NPROCS] = {
152 	NFSPROC_NULL,
153 	NFSPROC_GETATTR,
154 	NFSPROC_SETATTR,
155 	NFSPROC_NOOP,
156 	NFSPROC_LOOKUP,
157 	NFSPROC_READLINK,
158 	NFSPROC_READ,
159 	NFSPROC_NOOP,
160 	NFSPROC_WRITE,
161 	NFSPROC_CREATE,
162 	NFSPROC_REMOVE,
163 	NFSPROC_RENAME,
164 	NFSPROC_LINK,
165 	NFSPROC_SYMLINK,
166 	NFSPROC_MKDIR,
167 	NFSPROC_RMDIR,
168 	NFSPROC_READDIR,
169 	NFSPROC_FSSTAT,
170 	NFSPROC_NOOP,
171 	NFSPROC_NOOP,
172 	NFSPROC_NOOP,
173 	NFSPROC_NOOP,
174 	NFSPROC_NOOP,
175 	NFSPROC_NOOP,
176 	NFSPROC_NOOP,
177 	NFSPROC_NOOP
178 };
179 
180 /*
181  * and the reverse mapping from generic to Version 2 procedure numbers
182  */
183 const int nfsv2_procid[NFS_NPROCS] = {
184 	NFSV2PROC_NULL,
185 	NFSV2PROC_GETATTR,
186 	NFSV2PROC_SETATTR,
187 	NFSV2PROC_LOOKUP,
188 	NFSV2PROC_NOOP,
189 	NFSV2PROC_READLINK,
190 	NFSV2PROC_READ,
191 	NFSV2PROC_WRITE,
192 	NFSV2PROC_CREATE,
193 	NFSV2PROC_MKDIR,
194 	NFSV2PROC_SYMLINK,
195 	NFSV2PROC_CREATE,
196 	NFSV2PROC_REMOVE,
197 	NFSV2PROC_RMDIR,
198 	NFSV2PROC_RENAME,
199 	NFSV2PROC_LINK,
200 	NFSV2PROC_READDIR,
201 	NFSV2PROC_NOOP,
202 	NFSV2PROC_STATFS,
203 	NFSV2PROC_NOOP,
204 	NFSV2PROC_NOOP,
205 	NFSV2PROC_NOOP,
206 	NFSV2PROC_NOOP,
207 	NFSV2PROC_NOOP,
208 	NFSV2PROC_NOOP,
209 	NFSV2PROC_NOOP,
210 };
211 
212 /*
213  * Maps errno values to nfs error numbers.
214  * Use NFSERR_IO as the catch all for ones not specifically defined in
215  * RFC 1094.
216  */
217 static const u_char nfsrv_v2errmap[ELAST] = {
218   NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
219   NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
220   NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
221   NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
222   NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
223   NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
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_IO,	NFSERR_IO,	NFSERR_IO,
227   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
228   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
229   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
230   NFSERR_IO,	NFSERR_IO,	NFSERR_NAMETOL,	NFSERR_IO,	NFSERR_IO,
231   NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
232   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
233   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
234   NFSERR_IO,	NFSERR_IO,
235 };
236 
237 /*
238  * Maps errno values to nfs error numbers.
239  * Although it is not obvious whether or not NFS clients really care if
240  * a returned error value is in the specified list for the procedure, the
241  * safest thing to do is filter them appropriately. For Version 2, the
242  * X/Open XNFS document is the only specification that defines error values
243  * for each RPC (The RFC simply lists all possible error values for all RPCs),
244  * so I have decided to not do this for Version 2.
245  * The first entry is the default error return and the rest are the valid
246  * errors for that RPC in increasing numeric order.
247  */
248 static const short nfsv3err_null[] = {
249 	0,
250 	0,
251 };
252 
253 static const short nfsv3err_getattr[] = {
254 	NFSERR_IO,
255 	NFSERR_IO,
256 	NFSERR_STALE,
257 	NFSERR_BADHANDLE,
258 	NFSERR_SERVERFAULT,
259 	0,
260 };
261 
262 static const short nfsv3err_setattr[] = {
263 	NFSERR_IO,
264 	NFSERR_PERM,
265 	NFSERR_IO,
266 	NFSERR_ACCES,
267 	NFSERR_INVAL,
268 	NFSERR_NOSPC,
269 	NFSERR_ROFS,
270 	NFSERR_DQUOT,
271 	NFSERR_STALE,
272 	NFSERR_BADHANDLE,
273 	NFSERR_NOT_SYNC,
274 	NFSERR_SERVERFAULT,
275 	0,
276 };
277 
278 static const short nfsv3err_lookup[] = {
279 	NFSERR_IO,
280 	NFSERR_NOENT,
281 	NFSERR_IO,
282 	NFSERR_ACCES,
283 	NFSERR_NOTDIR,
284 	NFSERR_NAMETOL,
285 	NFSERR_STALE,
286 	NFSERR_BADHANDLE,
287 	NFSERR_SERVERFAULT,
288 	0,
289 };
290 
291 static const short nfsv3err_access[] = {
292 	NFSERR_IO,
293 	NFSERR_IO,
294 	NFSERR_STALE,
295 	NFSERR_BADHANDLE,
296 	NFSERR_SERVERFAULT,
297 	0,
298 };
299 
300 static const short nfsv3err_readlink[] = {
301 	NFSERR_IO,
302 	NFSERR_IO,
303 	NFSERR_ACCES,
304 	NFSERR_INVAL,
305 	NFSERR_STALE,
306 	NFSERR_BADHANDLE,
307 	NFSERR_NOTSUPP,
308 	NFSERR_SERVERFAULT,
309 	0,
310 };
311 
312 static const short nfsv3err_read[] = {
313 	NFSERR_IO,
314 	NFSERR_IO,
315 	NFSERR_NXIO,
316 	NFSERR_ACCES,
317 	NFSERR_INVAL,
318 	NFSERR_STALE,
319 	NFSERR_BADHANDLE,
320 	NFSERR_SERVERFAULT,
321 	NFSERR_JUKEBOX,
322 	0,
323 };
324 
325 static const short nfsv3err_write[] = {
326 	NFSERR_IO,
327 	NFSERR_IO,
328 	NFSERR_ACCES,
329 	NFSERR_INVAL,
330 	NFSERR_FBIG,
331 	NFSERR_NOSPC,
332 	NFSERR_ROFS,
333 	NFSERR_DQUOT,
334 	NFSERR_STALE,
335 	NFSERR_BADHANDLE,
336 	NFSERR_SERVERFAULT,
337 	NFSERR_JUKEBOX,
338 	0,
339 };
340 
341 static const short nfsv3err_create[] = {
342 	NFSERR_IO,
343 	NFSERR_IO,
344 	NFSERR_ACCES,
345 	NFSERR_EXIST,
346 	NFSERR_NOTDIR,
347 	NFSERR_NOSPC,
348 	NFSERR_ROFS,
349 	NFSERR_NAMETOL,
350 	NFSERR_DQUOT,
351 	NFSERR_STALE,
352 	NFSERR_BADHANDLE,
353 	NFSERR_NOTSUPP,
354 	NFSERR_SERVERFAULT,
355 	0,
356 };
357 
358 static const short nfsv3err_mkdir[] = {
359 	NFSERR_IO,
360 	NFSERR_IO,
361 	NFSERR_ACCES,
362 	NFSERR_EXIST,
363 	NFSERR_NOTDIR,
364 	NFSERR_NOSPC,
365 	NFSERR_ROFS,
366 	NFSERR_NAMETOL,
367 	NFSERR_DQUOT,
368 	NFSERR_STALE,
369 	NFSERR_BADHANDLE,
370 	NFSERR_NOTSUPP,
371 	NFSERR_SERVERFAULT,
372 	0,
373 };
374 
375 static const short nfsv3err_symlink[] = {
376 	NFSERR_IO,
377 	NFSERR_IO,
378 	NFSERR_ACCES,
379 	NFSERR_EXIST,
380 	NFSERR_NOTDIR,
381 	NFSERR_NOSPC,
382 	NFSERR_ROFS,
383 	NFSERR_NAMETOL,
384 	NFSERR_DQUOT,
385 	NFSERR_STALE,
386 	NFSERR_BADHANDLE,
387 	NFSERR_NOTSUPP,
388 	NFSERR_SERVERFAULT,
389 	0,
390 };
391 
392 static const short nfsv3err_mknod[] = {
393 	NFSERR_IO,
394 	NFSERR_IO,
395 	NFSERR_ACCES,
396 	NFSERR_EXIST,
397 	NFSERR_NOTDIR,
398 	NFSERR_NOSPC,
399 	NFSERR_ROFS,
400 	NFSERR_NAMETOL,
401 	NFSERR_DQUOT,
402 	NFSERR_STALE,
403 	NFSERR_BADHANDLE,
404 	NFSERR_NOTSUPP,
405 	NFSERR_SERVERFAULT,
406 	NFSERR_BADTYPE,
407 	0,
408 };
409 
410 static const short nfsv3err_remove[] = {
411 	NFSERR_IO,
412 	NFSERR_NOENT,
413 	NFSERR_IO,
414 	NFSERR_ACCES,
415 	NFSERR_NOTDIR,
416 	NFSERR_ROFS,
417 	NFSERR_NAMETOL,
418 	NFSERR_STALE,
419 	NFSERR_BADHANDLE,
420 	NFSERR_SERVERFAULT,
421 	0,
422 };
423 
424 static const short nfsv3err_rmdir[] = {
425 	NFSERR_IO,
426 	NFSERR_NOENT,
427 	NFSERR_IO,
428 	NFSERR_ACCES,
429 	NFSERR_EXIST,
430 	NFSERR_NOTDIR,
431 	NFSERR_INVAL,
432 	NFSERR_ROFS,
433 	NFSERR_NAMETOL,
434 	NFSERR_NOTEMPTY,
435 	NFSERR_STALE,
436 	NFSERR_BADHANDLE,
437 	NFSERR_NOTSUPP,
438 	NFSERR_SERVERFAULT,
439 	0,
440 };
441 
442 static const short nfsv3err_rename[] = {
443 	NFSERR_IO,
444 	NFSERR_NOENT,
445 	NFSERR_IO,
446 	NFSERR_ACCES,
447 	NFSERR_EXIST,
448 	NFSERR_XDEV,
449 	NFSERR_NOTDIR,
450 	NFSERR_ISDIR,
451 	NFSERR_INVAL,
452 	NFSERR_NOSPC,
453 	NFSERR_ROFS,
454 	NFSERR_MLINK,
455 	NFSERR_NAMETOL,
456 	NFSERR_NOTEMPTY,
457 	NFSERR_DQUOT,
458 	NFSERR_STALE,
459 	NFSERR_BADHANDLE,
460 	NFSERR_NOTSUPP,
461 	NFSERR_SERVERFAULT,
462 	0,
463 };
464 
465 static const short nfsv3err_link[] = {
466 	NFSERR_IO,
467 	NFSERR_IO,
468 	NFSERR_ACCES,
469 	NFSERR_EXIST,
470 	NFSERR_XDEV,
471 	NFSERR_NOTDIR,
472 	NFSERR_INVAL,
473 	NFSERR_NOSPC,
474 	NFSERR_ROFS,
475 	NFSERR_MLINK,
476 	NFSERR_NAMETOL,
477 	NFSERR_DQUOT,
478 	NFSERR_STALE,
479 	NFSERR_BADHANDLE,
480 	NFSERR_NOTSUPP,
481 	NFSERR_SERVERFAULT,
482 	0,
483 };
484 
485 static const short nfsv3err_readdir[] = {
486 	NFSERR_IO,
487 	NFSERR_IO,
488 	NFSERR_ACCES,
489 	NFSERR_NOTDIR,
490 	NFSERR_STALE,
491 	NFSERR_BADHANDLE,
492 	NFSERR_BAD_COOKIE,
493 	NFSERR_TOOSMALL,
494 	NFSERR_SERVERFAULT,
495 	0,
496 };
497 
498 static const short nfsv3err_readdirplus[] = {
499 	NFSERR_IO,
500 	NFSERR_IO,
501 	NFSERR_ACCES,
502 	NFSERR_NOTDIR,
503 	NFSERR_STALE,
504 	NFSERR_BADHANDLE,
505 	NFSERR_BAD_COOKIE,
506 	NFSERR_NOTSUPP,
507 	NFSERR_TOOSMALL,
508 	NFSERR_SERVERFAULT,
509 	0,
510 };
511 
512 static const short nfsv3err_fsstat[] = {
513 	NFSERR_IO,
514 	NFSERR_IO,
515 	NFSERR_STALE,
516 	NFSERR_BADHANDLE,
517 	NFSERR_SERVERFAULT,
518 	0,
519 };
520 
521 static const short nfsv3err_fsinfo[] = {
522 	NFSERR_STALE,
523 	NFSERR_STALE,
524 	NFSERR_BADHANDLE,
525 	NFSERR_SERVERFAULT,
526 	0,
527 };
528 
529 static const short nfsv3err_pathconf[] = {
530 	NFSERR_STALE,
531 	NFSERR_STALE,
532 	NFSERR_BADHANDLE,
533 	NFSERR_SERVERFAULT,
534 	0,
535 };
536 
537 static const short nfsv3err_commit[] = {
538 	NFSERR_IO,
539 	NFSERR_IO,
540 	NFSERR_STALE,
541 	NFSERR_BADHANDLE,
542 	NFSERR_SERVERFAULT,
543 	0,
544 };
545 
546 static const short * const nfsrv_v3errmap[] = {
547 	nfsv3err_null,
548 	nfsv3err_getattr,
549 	nfsv3err_setattr,
550 	nfsv3err_lookup,
551 	nfsv3err_access,
552 	nfsv3err_readlink,
553 	nfsv3err_read,
554 	nfsv3err_write,
555 	nfsv3err_create,
556 	nfsv3err_mkdir,
557 	nfsv3err_symlink,
558 	nfsv3err_mknod,
559 	nfsv3err_remove,
560 	nfsv3err_rmdir,
561 	nfsv3err_rename,
562 	nfsv3err_link,
563 	nfsv3err_readdir,
564 	nfsv3err_readdirplus,
565 	nfsv3err_fsstat,
566 	nfsv3err_fsinfo,
567 	nfsv3err_pathconf,
568 	nfsv3err_commit,
569 };
570 
571 extern struct nfsrtt nfsrtt;
572 extern time_t nqnfsstarttime;
573 extern int nqsrv_clockskew;
574 extern int nqsrv_writeslack;
575 extern int nqsrv_maxlease;
576 extern const int nqnfs_piggy[NFS_NPROCS];
577 extern struct nfsnodehashhead *nfsnodehashtbl;
578 extern u_long nfsnodehash;
579 
580 u_long nfsdirhashmask;
581 
582 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *));
583 
584 /*
585  * Create the header for an rpc request packet
586  * The hsiz is the size of the rest of the nfs request header.
587  * (just used to decide if a cluster is a good idea)
588  */
589 struct mbuf *
590 nfsm_reqh(np, procid, hsiz, bposp)
591 	struct nfsnode *np;
592 	u_long procid;
593 	int hsiz;
594 	caddr_t *bposp;
595 {
596 	struct mbuf *mb;
597 	caddr_t bpos;
598 #ifndef NFS_V2_ONLY
599 	struct nfsmount *nmp;
600 	u_int32_t *tl;
601 	int nqflag;
602 #endif
603 
604 	mb = m_get(M_WAIT, MT_DATA);
605 	MCLAIM(mb, &nfs_mowner);
606 	if (hsiz >= MINCLSIZE)
607 		m_clget(mb, M_WAIT);
608 	mb->m_len = 0;
609 	bpos = mtod(mb, caddr_t);
610 
611 #ifndef NFS_V2_ONLY
612 	/*
613 	 * For NQNFS, add lease request.
614 	 */
615 	if (np) {
616 		nmp = VFSTONFS(np->n_vnode->v_mount);
617 		if (nmp->nm_flag & NFSMNT_NQNFS) {
618 			nqflag = NQNFS_NEEDLEASE(np, procid);
619 			if (nqflag) {
620 				nfsm_build(tl, u_int32_t *, 2*NFSX_UNSIGNED);
621 				*tl++ = txdr_unsigned(nqflag);
622 				*tl = txdr_unsigned(nmp->nm_leaseterm);
623 			} else {
624 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
625 				*tl = 0;
626 			}
627 		}
628 	}
629 #endif
630 	/* Finally, return values */
631 	*bposp = bpos;
632 	return (mb);
633 }
634 
635 /*
636  * Build the RPC header and fill in the authorization info.
637  * The authorization string argument is only used when the credentials
638  * come from outside of the kernel.
639  * Returns the head of the mbuf list.
640  */
641 struct mbuf *
642 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len,
643 	verf_str, mrest, mrest_len, mbp, xidp)
644 	struct ucred *cr;
645 	int nmflag;
646 	int procid;
647 	int auth_type;
648 	int auth_len;
649 	char *auth_str;
650 	int verf_len;
651 	char *verf_str;
652 	struct mbuf *mrest;
653 	int mrest_len;
654 	struct mbuf **mbp;
655 	u_int32_t *xidp;
656 {
657 	struct mbuf *mb;
658 	u_int32_t *tl;
659 	caddr_t bpos;
660 	int i;
661 	struct mbuf *mreq;
662 	int siz, grpsiz, authsiz;
663 
664 	authsiz = nfsm_rndup(auth_len);
665 	mb = m_gethdr(M_WAIT, MT_DATA);
666 	MCLAIM(mb, &nfs_mowner);
667 	if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
668 		m_clget(mb, M_WAIT);
669 	} else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
670 		MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED);
671 	} else {
672 		MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
673 	}
674 	mb->m_len = 0;
675 	mreq = mb;
676 	bpos = mtod(mb, caddr_t);
677 
678 	/*
679 	 * First the RPC header.
680 	 */
681 	nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
682 
683 	*tl++ = *xidp = nfs_getxid();
684 	*tl++ = rpc_call;
685 	*tl++ = rpc_vers;
686 	if (nmflag & NFSMNT_NQNFS) {
687 		*tl++ = txdr_unsigned(NQNFS_PROG);
688 		*tl++ = txdr_unsigned(NQNFS_VER3);
689 	} else {
690 		*tl++ = txdr_unsigned(NFS_PROG);
691 		if (nmflag & NFSMNT_NFSV3)
692 			*tl++ = txdr_unsigned(NFS_VER3);
693 		else
694 			*tl++ = txdr_unsigned(NFS_VER2);
695 	}
696 	if (nmflag & NFSMNT_NFSV3)
697 		*tl++ = txdr_unsigned(procid);
698 	else
699 		*tl++ = txdr_unsigned(nfsv2_procid[procid]);
700 
701 	/*
702 	 * And then the authorization cred.
703 	 */
704 	*tl++ = txdr_unsigned(auth_type);
705 	*tl = txdr_unsigned(authsiz);
706 	switch (auth_type) {
707 	case RPCAUTH_UNIX:
708 		nfsm_build(tl, u_int32_t *, auth_len);
709 		*tl++ = 0;		/* stamp ?? */
710 		*tl++ = 0;		/* NULL hostname */
711 		*tl++ = txdr_unsigned(cr->cr_uid);
712 		*tl++ = txdr_unsigned(cr->cr_gid);
713 		grpsiz = (auth_len >> 2) - 5;
714 		*tl++ = txdr_unsigned(grpsiz);
715 		for (i = 0; i < grpsiz; i++)
716 			*tl++ = txdr_unsigned(cr->cr_groups[i]);
717 		break;
718 	case RPCAUTH_KERB4:
719 		siz = auth_len;
720 		while (siz > 0) {
721 			if (M_TRAILINGSPACE(mb) == 0) {
722 				struct mbuf *mb2;
723 				mb2 = m_get(M_WAIT, MT_DATA);
724 				MCLAIM(mb2, &nfs_mowner);
725 				if (siz >= MINCLSIZE)
726 					m_clget(mb2, M_WAIT);
727 				mb->m_next = mb2;
728 				mb = mb2;
729 				mb->m_len = 0;
730 				bpos = mtod(mb, caddr_t);
731 			}
732 			i = min(siz, M_TRAILINGSPACE(mb));
733 			memcpy(bpos, auth_str, i);
734 			mb->m_len += i;
735 			auth_str += i;
736 			bpos += i;
737 			siz -= i;
738 		}
739 		if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
740 			for (i = 0; i < siz; i++)
741 				*bpos++ = '\0';
742 			mb->m_len += siz;
743 		}
744 		break;
745 	};
746 
747 	/*
748 	 * And the verifier...
749 	 */
750 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
751 	if (verf_str) {
752 		*tl++ = txdr_unsigned(RPCAUTH_KERB4);
753 		*tl = txdr_unsigned(verf_len);
754 		siz = verf_len;
755 		while (siz > 0) {
756 			if (M_TRAILINGSPACE(mb) == 0) {
757 				struct mbuf *mb2;
758 				mb2 = m_get(M_WAIT, MT_DATA);
759 				MCLAIM(mb2, &nfs_mowner);
760 				if (siz >= MINCLSIZE)
761 					m_clget(mb2, M_WAIT);
762 				mb->m_next = mb2;
763 				mb = mb2;
764 				mb->m_len = 0;
765 				bpos = mtod(mb, caddr_t);
766 			}
767 			i = min(siz, M_TRAILINGSPACE(mb));
768 			memcpy(bpos, verf_str, i);
769 			mb->m_len += i;
770 			verf_str += i;
771 			bpos += i;
772 			siz -= i;
773 		}
774 		if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
775 			for (i = 0; i < siz; i++)
776 				*bpos++ = '\0';
777 			mb->m_len += siz;
778 		}
779 	} else {
780 		*tl++ = txdr_unsigned(RPCAUTH_NULL);
781 		*tl = 0;
782 	}
783 	mb->m_next = mrest;
784 	mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
785 	mreq->m_pkthdr.rcvif = (struct ifnet *)0;
786 	*mbp = mb;
787 	return (mreq);
788 }
789 
790 /*
791  * copies mbuf chain to the uio scatter/gather list
792  */
793 int
794 nfsm_mbuftouio(mrep, uiop, siz, dpos)
795 	struct mbuf **mrep;
796 	struct uio *uiop;
797 	int siz;
798 	caddr_t *dpos;
799 {
800 	char *mbufcp, *uiocp;
801 	int xfer, left, len;
802 	struct mbuf *mp;
803 	long uiosiz, rem;
804 	int error = 0;
805 
806 	mp = *mrep;
807 	mbufcp = *dpos;
808 	len = mtod(mp, caddr_t)+mp->m_len-mbufcp;
809 	rem = nfsm_rndup(siz)-siz;
810 	while (siz > 0) {
811 		if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
812 			return (EFBIG);
813 		left = uiop->uio_iov->iov_len;
814 		uiocp = uiop->uio_iov->iov_base;
815 		if (left > siz)
816 			left = siz;
817 		uiosiz = left;
818 		while (left > 0) {
819 			while (len == 0) {
820 				mp = mp->m_next;
821 				if (mp == NULL)
822 					return (EBADRPC);
823 				mbufcp = mtod(mp, caddr_t);
824 				len = mp->m_len;
825 			}
826 			xfer = (left > len) ? len : left;
827 #ifdef notdef
828 			/* Not Yet.. */
829 			if (uiop->uio_iov->iov_op != NULL)
830 				(*(uiop->uio_iov->iov_op))
831 				(mbufcp, uiocp, xfer);
832 			else
833 #endif
834 			if (uiop->uio_segflg == UIO_SYSSPACE)
835 				memcpy(uiocp, mbufcp, xfer);
836 			else
837 				copyout(mbufcp, uiocp, xfer);
838 			left -= xfer;
839 			len -= xfer;
840 			mbufcp += xfer;
841 			uiocp += xfer;
842 			uiop->uio_offset += xfer;
843 			uiop->uio_resid -= xfer;
844 		}
845 		if (uiop->uio_iov->iov_len <= siz) {
846 			uiop->uio_iovcnt--;
847 			uiop->uio_iov++;
848 		} else {
849 			uiop->uio_iov->iov_base =
850 			    (caddr_t)uiop->uio_iov->iov_base + uiosiz;
851 			uiop->uio_iov->iov_len -= uiosiz;
852 		}
853 		siz -= uiosiz;
854 	}
855 	*dpos = mbufcp;
856 	*mrep = mp;
857 	if (rem > 0) {
858 		if (len < rem)
859 			error = nfs_adv(mrep, dpos, rem, len);
860 		else
861 			*dpos += rem;
862 	}
863 	return (error);
864 }
865 
866 /*
867  * copies a uio scatter/gather list to an mbuf chain.
868  * NOTE: can ony handle iovcnt == 1
869  */
870 int
871 nfsm_uiotombuf(uiop, mq, siz, bpos)
872 	struct uio *uiop;
873 	struct mbuf **mq;
874 	int siz;
875 	caddr_t *bpos;
876 {
877 	char *uiocp;
878 	struct mbuf *mp, *mp2;
879 	int xfer, left, mlen;
880 	int uiosiz, clflg, rem;
881 	char *cp;
882 
883 #ifdef DIAGNOSTIC
884 	if (uiop->uio_iovcnt != 1)
885 		panic("nfsm_uiotombuf: iovcnt != 1");
886 #endif
887 
888 	if (siz > MLEN)		/* or should it >= MCLBYTES ?? */
889 		clflg = 1;
890 	else
891 		clflg = 0;
892 	rem = nfsm_rndup(siz)-siz;
893 	mp = mp2 = *mq;
894 	while (siz > 0) {
895 		left = uiop->uio_iov->iov_len;
896 		uiocp = uiop->uio_iov->iov_base;
897 		if (left > siz)
898 			left = siz;
899 		uiosiz = left;
900 		while (left > 0) {
901 			mlen = M_TRAILINGSPACE(mp);
902 			if (mlen == 0) {
903 				mp = m_get(M_WAIT, MT_DATA);
904 				MCLAIM(mp, &nfs_mowner);
905 				if (clflg)
906 					m_clget(mp, M_WAIT);
907 				mp->m_len = 0;
908 				mp2->m_next = mp;
909 				mp2 = mp;
910 				mlen = M_TRAILINGSPACE(mp);
911 			}
912 			xfer = (left > mlen) ? mlen : left;
913 #ifdef notdef
914 			/* Not Yet.. */
915 			if (uiop->uio_iov->iov_op != NULL)
916 				(*(uiop->uio_iov->iov_op))
917 				(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
918 			else
919 #endif
920 			if (uiop->uio_segflg == UIO_SYSSPACE)
921 				memcpy(mtod(mp, caddr_t)+mp->m_len, uiocp, xfer);
922 			else
923 				copyin(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
924 			mp->m_len += xfer;
925 			left -= xfer;
926 			uiocp += xfer;
927 			uiop->uio_offset += xfer;
928 			uiop->uio_resid -= xfer;
929 		}
930 		uiop->uio_iov->iov_base = (caddr_t)uiop->uio_iov->iov_base +
931 		    uiosiz;
932 		uiop->uio_iov->iov_len -= uiosiz;
933 		siz -= uiosiz;
934 	}
935 	if (rem > 0) {
936 		if (rem > M_TRAILINGSPACE(mp)) {
937 			mp = m_get(M_WAIT, MT_DATA);
938 			MCLAIM(mp, &nfs_mowner);
939 			mp->m_len = 0;
940 			mp2->m_next = mp;
941 		}
942 		cp = mtod(mp, caddr_t)+mp->m_len;
943 		for (left = 0; left < rem; left++)
944 			*cp++ = '\0';
945 		mp->m_len += rem;
946 		*bpos = cp;
947 	} else
948 		*bpos = mtod(mp, caddr_t)+mp->m_len;
949 	*mq = mp;
950 	return (0);
951 }
952 
953 /*
954  * Get at least "siz" bytes of correctly aligned data.
955  * When called the mbuf pointers are not necessarily correct,
956  * dsosp points to what ought to be in m_data and left contains
957  * what ought to be in m_len.
958  * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
959  * cases. (The macros use the vars. dpos and dpos2)
960  */
961 int
962 nfsm_disct(mdp, dposp, siz, left, cp2)
963 	struct mbuf **mdp;
964 	caddr_t *dposp;
965 	int siz;
966 	int left;
967 	caddr_t *cp2;
968 {
969 	struct mbuf *m1, *m2;
970 	struct mbuf *havebuf = NULL;
971 	caddr_t src = *dposp;
972 	caddr_t dst;
973 	int len;
974 
975 #ifdef DEBUG
976 	if (left < 0)
977 		panic("nfsm_disct: left < 0");
978 #endif
979 	m1 = *mdp;
980 	/*
981 	 * Skip through the mbuf chain looking for an mbuf with
982 	 * some data. If the first mbuf found has enough data
983 	 * and it is correctly aligned return it.
984 	 */
985 	while (left == 0) {
986 		havebuf = m1;
987 		*mdp = m1 = m1->m_next;
988 		if (m1 == NULL)
989 			return (EBADRPC);
990 		src = mtod(m1, caddr_t);
991 		left = m1->m_len;
992 		/*
993 		 * If we start a new mbuf and it is big enough
994 		 * and correctly aligned just return it, don't
995 		 * do any pull up.
996 		 */
997 		if (left >= siz && nfsm_aligned(src)) {
998 			*cp2 = src;
999 			*dposp = src + siz;
1000 			return (0);
1001 		}
1002 	}
1003 	if (m1->m_flags & M_EXT) {
1004 		if (havebuf) {
1005 			/* If the first mbuf with data has external data
1006 			 * and there is a previous empty mbuf use it
1007 			 * to move the data into.
1008 			 */
1009 			m2 = m1;
1010 			*mdp = m1 = havebuf;
1011 			if (m1->m_flags & M_EXT) {
1012 				MEXTREMOVE(m1);
1013 			}
1014 		} else {
1015 			/*
1016 			 * If the first mbuf has a external data
1017 			 * and there is no previous empty mbuf
1018 			 * allocate a new mbuf and move the external
1019 			 * data to the new mbuf. Also make the first
1020 			 * mbuf look empty.
1021 			 */
1022 			m2 = m_get(M_WAIT, MT_DATA);
1023 			m2->m_ext = m1->m_ext;
1024 			m2->m_data = src;
1025 			m2->m_len = left;
1026 			MCLADDREFERENCE(m1, m2);
1027 			MEXTREMOVE(m1);
1028 			m2->m_next = m1->m_next;
1029 			m1->m_next = m2;
1030 		}
1031 		m1->m_len = 0;
1032 		if (m1->m_flags & M_PKTHDR)
1033 			dst = m1->m_pktdat;
1034 		else
1035 			dst = m1->m_dat;
1036 		m1->m_data = dst;
1037 	} else {
1038 		/*
1039 		 * If the first mbuf has no external data
1040 		 * move the data to the front of the mbuf.
1041 		 */
1042 		if (m1->m_flags & M_PKTHDR)
1043 			dst = m1->m_pktdat;
1044 		else
1045 			dst = m1->m_dat;
1046 		m1->m_data = dst;
1047 		if (dst != src)
1048 			memmove(dst, src, left);
1049 		dst += left;
1050 		m1->m_len = left;
1051 		m2 = m1->m_next;
1052 	}
1053 	*cp2 = m1->m_data;
1054 	*dposp = mtod(m1, caddr_t) + siz;
1055 	/*
1056 	 * Loop through mbufs pulling data up into first mbuf until
1057 	 * the first mbuf is full or there is no more data to
1058 	 * pullup.
1059 	 */
1060 	while ((len = M_TRAILINGSPACE(m1)) != 0 && m2) {
1061 		if ((len = min(len, m2->m_len)) != 0)
1062 			memcpy(dst, m2->m_data, len);
1063 		m1->m_len += len;
1064 		dst += len;
1065 		m2->m_data += len;
1066 		m2->m_len -= len;
1067 		m2 = m2->m_next;
1068 	}
1069 	if (m1->m_len < siz)
1070 		return (EBADRPC);
1071 	return (0);
1072 }
1073 
1074 /*
1075  * Advance the position in the mbuf chain.
1076  */
1077 int
1078 nfs_adv(mdp, dposp, offs, left)
1079 	struct mbuf **mdp;
1080 	caddr_t *dposp;
1081 	int offs;
1082 	int left;
1083 {
1084 	struct mbuf *m;
1085 	int s;
1086 
1087 	m = *mdp;
1088 	s = left;
1089 	while (s < offs) {
1090 		offs -= s;
1091 		m = m->m_next;
1092 		if (m == NULL)
1093 			return (EBADRPC);
1094 		s = m->m_len;
1095 	}
1096 	*mdp = m;
1097 	*dposp = mtod(m, caddr_t)+offs;
1098 	return (0);
1099 }
1100 
1101 /*
1102  * Copy a string into mbufs for the hard cases...
1103  */
1104 int
1105 nfsm_strtmbuf(mb, bpos, cp, siz)
1106 	struct mbuf **mb;
1107 	char **bpos;
1108 	const char *cp;
1109 	long siz;
1110 {
1111 	struct mbuf *m1 = NULL, *m2;
1112 	long left, xfer, len, tlen;
1113 	u_int32_t *tl;
1114 	int putsize;
1115 
1116 	putsize = 1;
1117 	m2 = *mb;
1118 	left = M_TRAILINGSPACE(m2);
1119 	if (left > 0) {
1120 		tl = ((u_int32_t *)(*bpos));
1121 		*tl++ = txdr_unsigned(siz);
1122 		putsize = 0;
1123 		left -= NFSX_UNSIGNED;
1124 		m2->m_len += NFSX_UNSIGNED;
1125 		if (left > 0) {
1126 			memcpy((caddr_t) tl, cp, left);
1127 			siz -= left;
1128 			cp += left;
1129 			m2->m_len += left;
1130 			left = 0;
1131 		}
1132 	}
1133 	/* Loop around adding mbufs */
1134 	while (siz > 0) {
1135 		m1 = m_get(M_WAIT, MT_DATA);
1136 		MCLAIM(m1, &nfs_mowner);
1137 		if (siz > MLEN)
1138 			m_clget(m1, M_WAIT);
1139 		m1->m_len = NFSMSIZ(m1);
1140 		m2->m_next = m1;
1141 		m2 = m1;
1142 		tl = mtod(m1, u_int32_t *);
1143 		tlen = 0;
1144 		if (putsize) {
1145 			*tl++ = txdr_unsigned(siz);
1146 			m1->m_len -= NFSX_UNSIGNED;
1147 			tlen = NFSX_UNSIGNED;
1148 			putsize = 0;
1149 		}
1150 		if (siz < m1->m_len) {
1151 			len = nfsm_rndup(siz);
1152 			xfer = siz;
1153 			if (xfer < len)
1154 				*(tl+(xfer>>2)) = 0;
1155 		} else {
1156 			xfer = len = m1->m_len;
1157 		}
1158 		memcpy((caddr_t) tl, cp, xfer);
1159 		m1->m_len = len+tlen;
1160 		siz -= xfer;
1161 		cp += xfer;
1162 	}
1163 	*mb = m1;
1164 	*bpos = mtod(m1, caddr_t)+m1->m_len;
1165 	return (0);
1166 }
1167 
1168 /*
1169  * Directory caching routines. They work as follows:
1170  * - a cache is maintained per VDIR nfsnode.
1171  * - for each offset cookie that is exported to userspace, and can
1172  *   thus be thrown back at us as an offset to VOP_READDIR, store
1173  *   information in the cache.
1174  * - cached are:
1175  *   - cookie itself
1176  *   - blocknumber (essentially just a search key in the buffer cache)
1177  *   - entry number in block.
1178  *   - offset cookie of block in which this entry is stored
1179  *   - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
1180  * - entries are looked up in a hash table
1181  * - also maintained is an LRU list of entries, used to determine
1182  *   which ones to delete if the cache grows too large.
1183  * - if 32 <-> 64 translation mode is requested for a filesystem,
1184  *   the cache also functions as a translation table
1185  * - in the translation case, invalidating the cache does not mean
1186  *   flushing it, but just marking entries as invalid, except for
1187  *   the <64bit cookie, 32bitcookie> pair which is still valid, to
1188  *   still be able to use the cache as a translation table.
1189  * - 32 bit cookies are uniquely created by combining the hash table
1190  *   entry value, and one generation count per hash table entry,
1191  *   incremented each time an entry is appended to the chain.
1192  * - the cache is invalidated each time a direcory is modified
1193  * - sanity checks are also done; if an entry in a block turns
1194  *   out not to have a matching cookie, the cache is invalidated
1195  *   and a new block starting from the wanted offset is fetched from
1196  *   the server.
1197  * - directory entries as read from the server are extended to contain
1198  *   the 64bit and, optionally, the 32bit cookies, for sanity checking
1199  *   the cache and exporting them to userspace through the cookie
1200  *   argument to VOP_READDIR.
1201  */
1202 
1203 u_long
1204 nfs_dirhash(off)
1205 	off_t off;
1206 {
1207 	int i;
1208 	char *cp = (char *)&off;
1209 	u_long sum = 0L;
1210 
1211 	for (i = 0 ; i < sizeof (off); i++)
1212 		sum += *cp++;
1213 
1214 	return sum;
1215 }
1216 
1217 void
1218 nfs_initdircache(vp)
1219 	struct vnode *vp;
1220 {
1221 	struct nfsnode *np = VTONFS(vp);
1222 
1223 	KASSERT(np->n_dircache == NULL);
1224 
1225 	np->n_dircachesize = 0;
1226 	np->n_dblkno = 1;
1227 	np->n_dircache = hashinit(NFS_DIRHASHSIZ, HASH_LIST, M_NFSDIROFF,
1228 	    M_WAITOK, &nfsdirhashmask);
1229 	TAILQ_INIT(&np->n_dirchain);
1230 }
1231 
1232 void
1233 nfs_initdirxlatecookie(vp)
1234 	struct vnode *vp;
1235 {
1236 	struct nfsnode *np = VTONFS(vp);
1237 
1238 	KASSERT(VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_XLATECOOKIE);
1239 	KASSERT(np->n_dirgens == NULL);
1240 
1241 	MALLOC(np->n_dirgens, unsigned *,
1242 	    NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF, M_WAITOK);
1243 	memset((caddr_t)np->n_dirgens, 0, NFS_DIRHASHSIZ * sizeof (unsigned));
1244 }
1245 
1246 static struct nfsdircache dzero = {0, 0, {0, 0}, {0, 0}, 0, 0, 0};
1247 
1248 struct nfsdircache *
1249 nfs_searchdircache(vp, off, do32, hashent)
1250 	struct vnode *vp;
1251 	off_t off;
1252 	int do32;
1253 	int *hashent;
1254 {
1255 	struct nfsdirhashhead *ndhp;
1256 	struct nfsdircache *ndp = NULL;
1257 	struct nfsnode *np = VTONFS(vp);
1258 	unsigned ent;
1259 
1260 	/*
1261 	 * Zero is always a valid cookie.
1262 	 */
1263 	if (off == 0)
1264 		return &dzero;
1265 
1266 	/*
1267 	 * We use a 32bit cookie as search key, directly reconstruct
1268 	 * the hashentry. Else use the hashfunction.
1269 	 */
1270 	if (do32) {
1271 		ent = (u_int32_t)off >> 24;
1272 		if (ent >= NFS_DIRHASHSIZ)
1273 			return NULL;
1274 		ndhp = &np->n_dircache[ent];
1275 	} else {
1276 		ndhp = NFSDIRHASH(np, off);
1277 	}
1278 
1279 	if (hashent)
1280 		*hashent = (int)(ndhp - np->n_dircache);
1281 	if (do32) {
1282 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1283 			if (ndp->dc_cookie32 == (u_int32_t)off) {
1284 				/*
1285 				 * An invalidated entry will become the
1286 				 * start of a new block fetched from
1287 				 * the server.
1288 				 */
1289 				if (ndp->dc_blkno == -1) {
1290 					ndp->dc_blkcookie = ndp->dc_cookie;
1291 					ndp->dc_blkno = np->n_dblkno++;
1292 					ndp->dc_entry = 0;
1293 				}
1294 				break;
1295 			}
1296 		}
1297 	} else {
1298 		LIST_FOREACH(ndp, ndhp, dc_hash) {
1299 			if (ndp->dc_cookie == off)
1300 				break;
1301 		}
1302 	}
1303 	return ndp;
1304 }
1305 
1306 
1307 struct nfsdircache *
1308 nfs_enterdircache(vp, off, blkoff, en, blkno)
1309 	struct vnode *vp;
1310 	off_t off, blkoff;
1311 	int en;
1312 	daddr_t blkno;
1313 {
1314 	struct nfsnode *np = VTONFS(vp);
1315 	struct nfsdirhashhead *ndhp;
1316 	struct nfsdircache *ndp = NULL, *first;
1317 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1318 	int hashent, gen, overwrite;
1319 
1320 	if (!np->n_dircache)
1321 		/*
1322 		 * XXX would like to do this in nfs_nget but vtype
1323 		 * isn't known at that time.
1324 		 */
1325 		nfs_initdircache(vp);
1326 
1327 	if ((nmp->nm_flag & NFSMNT_XLATECOOKIE) && !np->n_dirgens)
1328 		nfs_initdirxlatecookie(vp);
1329 
1330 	/*
1331 	 * XXX refuse entries for offset 0. amd(8) erroneously sets
1332 	 * cookie 0 for the '.' entry, making this necessary. This
1333 	 * isn't so bad, as 0 is a special case anyway.
1334 	 */
1335 	if (off == 0)
1336 		return &dzero;
1337 
1338 	ndp = nfs_searchdircache(vp, off, 0, &hashent);
1339 
1340 	if (ndp && ndp->dc_blkno != -1) {
1341 		/*
1342 		 * Overwriting an old entry. Check if it's the same.
1343 		 * If so, just return. If not, remove the old entry.
1344 		 */
1345 		if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
1346 			return ndp;
1347 		TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1348 		LIST_REMOVE(ndp, dc_hash);
1349 		FREE(ndp, M_NFSDIROFF);
1350 		ndp = 0;
1351 	}
1352 
1353 	ndhp = &np->n_dircache[hashent];
1354 
1355 	if (!ndp) {
1356 		MALLOC(ndp, struct nfsdircache *, sizeof (*ndp), M_NFSDIROFF,
1357 		    M_WAITOK);
1358 		overwrite = 0;
1359 		if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1360 			/*
1361 			 * We're allocating a new entry, so bump the
1362 			 * generation number.
1363 			 */
1364 			gen = ++np->n_dirgens[hashent];
1365 			if (gen == 0) {
1366 				np->n_dirgens[hashent]++;
1367 				gen++;
1368 			}
1369 			ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
1370 		}
1371 	} else
1372 		overwrite = 1;
1373 
1374 	/*
1375 	 * If the entry number is 0, we are at the start of a new block, so
1376 	 * allocate a new blocknumber.
1377 	 */
1378 	if (en == 0)
1379 		ndp->dc_blkno = np->n_dblkno++;
1380 	else
1381 		ndp->dc_blkno = blkno;
1382 
1383 	ndp->dc_cookie = off;
1384 	ndp->dc_blkcookie = blkoff;
1385 	ndp->dc_entry = en;
1386 
1387 	if (overwrite)
1388 		return ndp;
1389 
1390 	/*
1391 	 * If the maximum directory cookie cache size has been reached
1392 	 * for this node, take one off the front. The idea is that
1393 	 * directories are typically read front-to-back once, so that
1394 	 * the oldest entries can be thrown away without much performance
1395 	 * loss.
1396 	 */
1397 	if (np->n_dircachesize == NFS_MAXDIRCACHE) {
1398 		first = TAILQ_FIRST(&np->n_dirchain);
1399 		TAILQ_REMOVE(&np->n_dirchain, first, dc_chain);
1400 		LIST_REMOVE(first, dc_hash);
1401 		FREE(first, M_NFSDIROFF);
1402 	} else
1403 		np->n_dircachesize++;
1404 
1405 	LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
1406 	TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
1407 	return ndp;
1408 }
1409 
1410 void
1411 nfs_invaldircache(vp, forcefree)
1412 	struct vnode *vp;
1413 	int forcefree;
1414 {
1415 	struct nfsnode *np = VTONFS(vp);
1416 	struct nfsdircache *ndp = NULL;
1417 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1418 
1419 #ifdef DIAGNOSTIC
1420 	if (vp->v_type != VDIR)
1421 		panic("nfs: invaldircache: not dir");
1422 #endif
1423 
1424 	if (!np->n_dircache)
1425 		return;
1426 
1427 	if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
1428 		while ((ndp = TAILQ_FIRST(&np->n_dirchain)) != 0) {
1429 			TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1430 			LIST_REMOVE(ndp, dc_hash);
1431 			FREE(ndp, M_NFSDIROFF);
1432 		}
1433 		np->n_dircachesize = 0;
1434 		if (forcefree && np->n_dirgens) {
1435 			FREE(np->n_dirgens, M_NFSDIROFF);
1436 			np->n_dirgens = NULL;
1437 		}
1438 	} else {
1439 		TAILQ_FOREACH(ndp, &np->n_dirchain, dc_chain) {
1440 			ndp->dc_blkno = -1;
1441 		}
1442 	}
1443 
1444 	np->n_dblkno = 1;
1445 }
1446 
1447 /*
1448  * Called once before VFS init to initialize shared and
1449  * server-specific data structures.
1450  */
1451 void
1452 nfs_init()
1453 {
1454 	nfsrtt.pos = 0;
1455 	rpc_vers = txdr_unsigned(RPC_VER2);
1456 	rpc_call = txdr_unsigned(RPC_CALL);
1457 	rpc_reply = txdr_unsigned(RPC_REPLY);
1458 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1459 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1460 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1461 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1462 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1463 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1464 	nfs_prog = txdr_unsigned(NFS_PROG);
1465 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
1466 	nfs_true = txdr_unsigned(TRUE);
1467 	nfs_false = txdr_unsigned(FALSE);
1468 	nfs_xdrneg1 = txdr_unsigned(-1);
1469 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1470 	if (nfs_ticks < 1)
1471 		nfs_ticks = 1;
1472 #ifdef NFSSERVER
1473 	nfsrv_init(0);			/* Init server data structures */
1474 	nfsrv_initcache();		/* Init the server request cache */
1475 	pool_init(&nfs_srvdesc_pool, sizeof(struct nfsrv_descript),
1476 	    0, 0, 0, "nfsrvdescpl", &pool_allocator_nointr);
1477 #endif /* NFSSERVER */
1478 
1479 #if defined(NFSSERVER) || !defined(NFS_V2_ONLY)
1480 	/*
1481 	 * Initialize the nqnfs data structures.
1482 	 */
1483 	if (nqnfsstarttime == 0) {
1484 		nqnfsstarttime = boottime.tv_sec + nqsrv_maxlease
1485 			+ nqsrv_clockskew + nqsrv_writeslack;
1486 		NQLOADNOVRAM(nqnfsstarttime);
1487 		CIRCLEQ_INIT(&nqtimerhead);
1488 		nqfhhashtbl = hashinit(NQLCHSZ, HASH_LIST, M_NQLEASE,
1489 		    M_WAITOK, &nqfhhash);
1490 	}
1491 #endif
1492 
1493 	exithook_establish(nfs_exit, NULL);
1494 
1495 	/*
1496 	 * Initialize reply list and start timer
1497 	 */
1498 	TAILQ_INIT(&nfs_reqq);
1499 	nfs_timer(NULL);
1500 	MOWNER_ATTACH(&nfs_mowner);
1501 
1502 #ifdef NFS
1503 	/* Initialize the kqueue structures */
1504 	nfs_kqinit();
1505 	/* Initialize the iod structures */
1506 	nfs_iodinit();
1507 #endif
1508 }
1509 
1510 #ifdef NFS
1511 /*
1512  * Called once at VFS init to initialize client-specific data structures.
1513  */
1514 void
1515 nfs_vfs_init()
1516 {
1517 	nfs_nhinit();			/* Init the nfsnode table */
1518 	nfs_commitsize = uvmexp.npages << (PAGE_SHIFT - 4);
1519 }
1520 
1521 void
1522 nfs_vfs_reinit()
1523 {
1524 	nfs_nhreinit();
1525 }
1526 
1527 void
1528 nfs_vfs_done()
1529 {
1530 	nfs_nhdone();
1531 }
1532 
1533 /*
1534  * Attribute cache routines.
1535  * nfs_loadattrcache() - loads or updates the cache contents from attributes
1536  *	that are on the mbuf list
1537  * nfs_getattrcache() - returns valid attributes if found in cache, returns
1538  *	error otherwise
1539  */
1540 
1541 /*
1542  * Load the attribute cache (that lives in the nfsnode entry) with
1543  * the values on the mbuf list and
1544  * Iff vap not NULL
1545  *    copy the attributes to *vaper
1546  */
1547 int
1548 nfsm_loadattrcache(vpp, mdp, dposp, vaper, flags)
1549 	struct vnode **vpp;
1550 	struct mbuf **mdp;
1551 	caddr_t *dposp;
1552 	struct vattr *vaper;
1553 	int flags;
1554 {
1555 	int32_t t1;
1556 	caddr_t cp2;
1557 	int error = 0;
1558 	struct mbuf *md;
1559 	int v3 = NFS_ISV3(*vpp);
1560 
1561 	md = *mdp;
1562 	t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
1563 	error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2);
1564 	if (error)
1565 		return (error);
1566 	return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper, flags);
1567 }
1568 
1569 int
1570 nfs_loadattrcache(vpp, fp, vaper, flags)
1571 	struct vnode **vpp;
1572 	struct nfs_fattr *fp;
1573 	struct vattr *vaper;
1574 	int flags;
1575 {
1576 	struct vnode *vp = *vpp;
1577 	struct vattr *vap;
1578 	int v3 = NFS_ISV3(vp);
1579 	enum vtype vtyp;
1580 	u_short vmode;
1581 	struct timespec mtime;
1582 	struct vnode *nvp;
1583 	int32_t rdev;
1584 	struct nfsnode *np;
1585 	extern int (**spec_nfsv2nodeop_p) __P((void *));
1586 	uid_t uid;
1587 	gid_t gid;
1588 
1589 	if (v3) {
1590 		vtyp = nfsv3tov_type(fp->fa_type);
1591 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1592 		rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1),
1593 			fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2));
1594 		fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
1595 	} else {
1596 		vtyp = nfsv2tov_type(fp->fa_type);
1597 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1598 		if (vtyp == VNON || vtyp == VREG)
1599 			vtyp = IFTOVT(vmode);
1600 		rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
1601 		fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
1602 
1603 		/*
1604 		 * Really ugly NFSv2 kludge.
1605 		 */
1606 		if (vtyp == VCHR && rdev == 0xffffffff)
1607 			vtyp = VFIFO;
1608 	}
1609 
1610 	vmode &= ALLPERMS;
1611 
1612 	/*
1613 	 * If v_type == VNON it is a new node, so fill in the v_type,
1614 	 * n_mtime fields. Check to see if it represents a special
1615 	 * device, and if so, check for a possible alias. Once the
1616 	 * correct vnode has been obtained, fill in the rest of the
1617 	 * information.
1618 	 */
1619 	np = VTONFS(vp);
1620 	if (vp->v_type == VNON) {
1621 		vp->v_type = vtyp;
1622 		if (vp->v_type == VFIFO) {
1623 			extern int (**fifo_nfsv2nodeop_p) __P((void *));
1624 			vp->v_op = fifo_nfsv2nodeop_p;
1625 		}
1626 		if (vp->v_type == VCHR || vp->v_type == VBLK) {
1627 			vp->v_op = spec_nfsv2nodeop_p;
1628 			nvp = checkalias(vp, (dev_t)rdev, vp->v_mount);
1629 			if (nvp) {
1630 				/*
1631 				 * Discard unneeded vnode, but save its nfsnode.
1632 				 * Since the nfsnode does not have a lock, its
1633 				 * vnode lock has to be carried over.
1634 				 */
1635 				/*
1636 				 * XXX is the old node sure to be locked here?
1637 				 */
1638 				KASSERT(lockstatus(&vp->v_lock) ==
1639 				    LK_EXCLUSIVE);
1640 				nvp->v_data = vp->v_data;
1641 				vp->v_data = NULL;
1642 				VOP_UNLOCK(vp, 0);
1643 				vp->v_op = spec_vnodeop_p;
1644 				vrele(vp);
1645 				vgone(vp);
1646 				lockmgr(&nvp->v_lock, LK_EXCLUSIVE,
1647 				    &nvp->v_interlock);
1648 				/*
1649 				 * Reinitialize aliased node.
1650 				 */
1651 				np->n_vnode = nvp;
1652 				*vpp = vp = nvp;
1653 			}
1654 		}
1655 		np->n_mtime = mtime;
1656 	}
1657 	uid = fxdr_unsigned(uid_t, fp->fa_uid);
1658 	gid = fxdr_unsigned(gid_t, fp->fa_gid);
1659 	vap = np->n_vattr;
1660 
1661 	/*
1662 	 * Invalidate access cache if uid, gid or mode changed.
1663 	 */
1664 	if (np->n_accstamp != -1 &&
1665 	    (gid != vap->va_gid || uid != vap->va_uid || vmode != vap->va_mode))
1666 		np->n_accstamp = -1;
1667 
1668 	vap->va_type = vtyp;
1669 	vap->va_mode = vmode;
1670 	vap->va_rdev = (dev_t)rdev;
1671 	vap->va_mtime = mtime;
1672 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1673 	switch (vtyp) {
1674 	case VDIR:
1675 		vap->va_blocksize = NFS_DIRFRAGSIZ;
1676 		break;
1677 	case VBLK:
1678 		vap->va_blocksize = BLKDEV_IOSIZE;
1679 		break;
1680 	case VCHR:
1681 		vap->va_blocksize = MAXBSIZE;
1682 		break;
1683 	default:
1684 		vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize :
1685 		    fxdr_unsigned(int32_t, fp->fa2_blocksize);
1686 		break;
1687 	}
1688 	if (v3) {
1689 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1690 		vap->va_uid = uid;
1691 		vap->va_gid = gid;
1692 		vap->va_size = fxdr_hyper(&fp->fa3_size);
1693 		vap->va_bytes = fxdr_hyper(&fp->fa3_used);
1694 		vap->va_fileid = fxdr_unsigned(int32_t,
1695 		    fp->fa3_fileid.nfsuquad[1]);
1696 		fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
1697 		fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime);
1698 		vap->va_flags = 0;
1699 		vap->va_filerev = 0;
1700 	} else {
1701 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1702 		vap->va_uid = uid;
1703 		vap->va_gid = gid;
1704 		vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
1705 		vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks)
1706 		    * NFS_FABLKSIZE;
1707 		vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
1708 		fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
1709 		vap->va_flags = 0;
1710 		vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t,
1711 		    fp->fa2_ctime.nfsv2_sec);
1712 		vap->va_ctime.tv_nsec = 0;
1713 		vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
1714 		vap->va_filerev = 0;
1715 	}
1716 	if (vap->va_size != np->n_size) {
1717 		if ((np->n_flag & NMODIFIED) && vap->va_size < np->n_size) {
1718 			vap->va_size = np->n_size;
1719 		} else {
1720 			np->n_size = vap->va_size;
1721 			if (vap->va_type == VREG) {
1722 				if ((flags & NAC_NOTRUNC)
1723 				    && np->n_size < vp->v_size) {
1724 					/*
1725 					 * we can't free pages now because
1726 					 * the pages can be owned by ourselves.
1727 					 */
1728 					np->n_flag |= NTRUNCDELAYED;
1729 				}
1730 				else {
1731 					uvm_vnp_setsize(vp, np->n_size);
1732 				}
1733 			}
1734 		}
1735 	}
1736 	np->n_attrstamp = time.tv_sec;
1737 	if (vaper != NULL) {
1738 		memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(*vap));
1739 		if (np->n_flag & NCHG) {
1740 			if (np->n_flag & NACC)
1741 				vaper->va_atime = np->n_atim;
1742 			if (np->n_flag & NUPD)
1743 				vaper->va_mtime = np->n_mtim;
1744 		}
1745 	}
1746 	return (0);
1747 }
1748 
1749 /*
1750  * Check the time stamp
1751  * If the cache is valid, copy contents to *vap and return 0
1752  * otherwise return an error
1753  */
1754 int
1755 nfs_getattrcache(vp, vaper)
1756 	struct vnode *vp;
1757 	struct vattr *vaper;
1758 {
1759 	struct nfsnode *np = VTONFS(vp);
1760 	struct vattr *vap;
1761 
1762 	if ((time.tv_sec - np->n_attrstamp) >= NFS_ATTRTIMEO(np)) {
1763 		nfsstats.attrcache_misses++;
1764 		return (ENOENT);
1765 	}
1766 	nfsstats.attrcache_hits++;
1767 	vap = np->n_vattr;
1768 	if (vap->va_size != np->n_size) {
1769 		if (vap->va_type == VREG) {
1770 			if (np->n_flag & NMODIFIED) {
1771 				if (vap->va_size < np->n_size)
1772 					vap->va_size = np->n_size;
1773 				else
1774 					np->n_size = vap->va_size;
1775 			} else
1776 				np->n_size = vap->va_size;
1777 			uvm_vnp_setsize(vp, np->n_size);
1778 		} else
1779 			np->n_size = vap->va_size;
1780 	}
1781 	memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(struct vattr));
1782 	if (np->n_flag & NCHG) {
1783 		if (np->n_flag & NACC)
1784 			vaper->va_atime = np->n_atim;
1785 		if (np->n_flag & NUPD)
1786 			vaper->va_mtime = np->n_mtim;
1787 	}
1788 	return (0);
1789 }
1790 
1791 void
1792 nfs_delayedtruncate(vp)
1793 	struct vnode *vp;
1794 {
1795 	struct nfsnode *np = VTONFS(vp);
1796 
1797 	if (np->n_flag & NTRUNCDELAYED) {
1798 		np->n_flag &= ~NTRUNCDELAYED;
1799 		uvm_vnp_setsize(vp, np->n_size);
1800 	}
1801 }
1802 
1803 /*
1804  * Heuristic to see if the server XDR encodes directory cookies or not.
1805  * it is not supposed to, but a lot of servers may do this. Also, since
1806  * most/all servers will implement V2 as well, it is expected that they
1807  * may return just 32 bits worth of cookie information, so we need to
1808  * find out in which 32 bits this information is available. We do this
1809  * to avoid trouble with emulated binaries that can't handle 64 bit
1810  * directory offsets.
1811  */
1812 
1813 void
1814 nfs_cookieheuristic(vp, flagp, p, cred)
1815 	struct vnode *vp;
1816 	int *flagp;
1817 	struct proc *p;
1818 	struct ucred *cred;
1819 {
1820 	struct uio auio;
1821 	struct iovec aiov;
1822 	caddr_t buf, cp;
1823 	struct dirent *dp;
1824 	off_t *cookies = NULL, *cop;
1825 	int error, eof, nc, len;
1826 
1827 	MALLOC(buf, caddr_t, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK);
1828 
1829 	aiov.iov_base = buf;
1830 	aiov.iov_len = NFS_DIRFRAGSIZ;
1831 	auio.uio_iov = &aiov;
1832 	auio.uio_iovcnt = 1;
1833 	auio.uio_rw = UIO_READ;
1834 	auio.uio_segflg = UIO_SYSSPACE;
1835 	auio.uio_procp = p;
1836 	auio.uio_resid = NFS_DIRFRAGSIZ;
1837 	auio.uio_offset = 0;
1838 
1839 	error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc);
1840 
1841 	len = NFS_DIRFRAGSIZ - auio.uio_resid;
1842 	if (error || len == 0) {
1843 		FREE(buf, M_TEMP);
1844 		if (cookies)
1845 			free(cookies, M_TEMP);
1846 		return;
1847 	}
1848 
1849 	/*
1850 	 * Find the first valid entry and look at its offset cookie.
1851 	 */
1852 
1853 	cp = buf;
1854 	for (cop = cookies; len > 0; len -= dp->d_reclen) {
1855 		dp = (struct dirent *)cp;
1856 		if (dp->d_fileno != 0 && len >= dp->d_reclen) {
1857 			if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) {
1858 				*flagp |= NFSMNT_SWAPCOOKIE;
1859 				nfs_invaldircache(vp, 0);
1860 				nfs_vinvalbuf(vp, 0, cred, p, 1);
1861 			}
1862 			break;
1863 		}
1864 		cop++;
1865 		cp += dp->d_reclen;
1866 	}
1867 
1868 	FREE(buf, M_TEMP);
1869 	free(cookies, M_TEMP);
1870 }
1871 #endif /* NFS */
1872 
1873 /*
1874  * Set up nameidata for a lookup() call and do it.
1875  *
1876  * If pubflag is set, this call is done for a lookup operation on the
1877  * public filehandle. In that case we allow crossing mountpoints and
1878  * absolute pathnames. However, the caller is expected to check that
1879  * the lookup result is within the public fs, and deny access if
1880  * it is not.
1881  */
1882 int
1883 nfs_namei(ndp, fhp, len, slp, nam, mdp, dposp, retdirp, p, kerbflag, pubflag)
1884 	struct nameidata *ndp;
1885 	fhandle_t *fhp;
1886 	uint32_t len;
1887 	struct nfssvc_sock *slp;
1888 	struct mbuf *nam;
1889 	struct mbuf **mdp;
1890 	caddr_t *dposp;
1891 	struct vnode **retdirp;
1892 	struct proc *p;
1893 	int kerbflag, pubflag;
1894 {
1895 	int i, rem;
1896 	struct mbuf *md;
1897 	char *fromcp, *tocp, *cp;
1898 	struct iovec aiov;
1899 	struct uio auio;
1900 	struct vnode *dp;
1901 	int error, rdonly, linklen;
1902 	struct componentname *cnp = &ndp->ni_cnd;
1903 
1904 	*retdirp = (struct vnode *)0;
1905 
1906 	if ((len + 1) > MAXPATHLEN)
1907 		return (ENAMETOOLONG);
1908 	cnp->cn_pnbuf = PNBUF_GET();
1909 
1910 	/*
1911 	 * Copy the name from the mbuf list to ndp->ni_pnbuf
1912 	 * and set the various ndp fields appropriately.
1913 	 */
1914 	fromcp = *dposp;
1915 	tocp = cnp->cn_pnbuf;
1916 	md = *mdp;
1917 	rem = mtod(md, caddr_t) + md->m_len - fromcp;
1918 	for (i = 0; i < len; i++) {
1919 		while (rem == 0) {
1920 			md = md->m_next;
1921 			if (md == NULL) {
1922 				error = EBADRPC;
1923 				goto out;
1924 			}
1925 			fromcp = mtod(md, caddr_t);
1926 			rem = md->m_len;
1927 		}
1928 		if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
1929 			error = EACCES;
1930 			goto out;
1931 		}
1932 		*tocp++ = *fromcp++;
1933 		rem--;
1934 	}
1935 	*tocp = '\0';
1936 	*mdp = md;
1937 	*dposp = fromcp;
1938 	len = nfsm_rndup(len)-len;
1939 	if (len > 0) {
1940 		if (rem >= len)
1941 			*dposp += len;
1942 		else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
1943 			goto out;
1944 	}
1945 
1946 	/*
1947 	 * Extract and set starting directory.
1948 	 */
1949 	error = nfsrv_fhtovp(fhp, FALSE, &dp, ndp->ni_cnd.cn_cred, slp,
1950 	    nam, &rdonly, kerbflag, pubflag);
1951 	if (error)
1952 		goto out;
1953 	if (dp->v_type != VDIR) {
1954 		vrele(dp);
1955 		error = ENOTDIR;
1956 		goto out;
1957 	}
1958 
1959 	if (rdonly)
1960 		cnp->cn_flags |= RDONLY;
1961 
1962 	*retdirp = dp;
1963 
1964 	if (pubflag) {
1965 		/*
1966 		 * Oh joy. For WebNFS, handle those pesky '%' escapes,
1967 		 * and the 'native path' indicator.
1968 		 */
1969 		cp = PNBUF_GET();
1970 		fromcp = cnp->cn_pnbuf;
1971 		tocp = cp;
1972 		if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
1973 			switch ((unsigned char)*fromcp) {
1974 			case WEBNFS_NATIVE_CHAR:
1975 				/*
1976 				 * 'Native' path for us is the same
1977 				 * as a path according to the NFS spec,
1978 				 * just skip the escape char.
1979 				 */
1980 				fromcp++;
1981 				break;
1982 			/*
1983 			 * More may be added in the future, range 0x80-0xff
1984 			 */
1985 			default:
1986 				error = EIO;
1987 				PNBUF_PUT(cp);
1988 				goto out;
1989 			}
1990 		}
1991 		/*
1992 		 * Translate the '%' escapes, URL-style.
1993 		 */
1994 		while (*fromcp != '\0') {
1995 			if (*fromcp == WEBNFS_ESC_CHAR) {
1996 				if (fromcp[1] != '\0' && fromcp[2] != '\0') {
1997 					fromcp++;
1998 					*tocp++ = HEXSTRTOI(fromcp);
1999 					fromcp += 2;
2000 					continue;
2001 				} else {
2002 					error = ENOENT;
2003 					PNBUF_PUT(cp);
2004 					goto out;
2005 				}
2006 			} else
2007 				*tocp++ = *fromcp++;
2008 		}
2009 		*tocp = '\0';
2010 		PNBUF_PUT(cnp->cn_pnbuf);
2011 		cnp->cn_pnbuf = cp;
2012 	}
2013 
2014 	ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1;
2015 	ndp->ni_segflg = UIO_SYSSPACE;
2016 	ndp->ni_rootdir = rootvnode;
2017 
2018 	if (pubflag) {
2019 		ndp->ni_loopcnt = 0;
2020 		if (cnp->cn_pnbuf[0] == '/')
2021 			dp = rootvnode;
2022 	} else {
2023 		cnp->cn_flags |= NOCROSSMOUNT;
2024 	}
2025 
2026 	cnp->cn_proc = p;
2027 	VREF(dp);
2028 
2029     for (;;) {
2030 	cnp->cn_nameptr = cnp->cn_pnbuf;
2031 	ndp->ni_startdir = dp;
2032 	/*
2033 	 * And call lookup() to do the real work
2034 	 */
2035 	error = lookup(ndp);
2036 	if (error) {
2037 		PNBUF_PUT(cnp->cn_pnbuf);
2038 		return (error);
2039 	}
2040 	/*
2041 	 * Check for encountering a symbolic link
2042 	 */
2043 	if ((cnp->cn_flags & ISSYMLINK) == 0) {
2044 		if (cnp->cn_flags & (SAVENAME | SAVESTART))
2045 			cnp->cn_flags |= HASBUF;
2046 		else
2047 			PNBUF_PUT(cnp->cn_pnbuf);
2048 		return (0);
2049 	} else {
2050 		if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN))
2051 			VOP_UNLOCK(ndp->ni_dvp, 0);
2052 		if (!pubflag) {
2053 			error = EINVAL;
2054 			break;
2055 		}
2056 
2057 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
2058 			error = ELOOP;
2059 			break;
2060 		}
2061 		if (ndp->ni_vp->v_mount->mnt_flag & MNT_SYMPERM) {
2062 			error = VOP_ACCESS(ndp->ni_vp, VEXEC, cnp->cn_cred,
2063 			    cnp->cn_proc);
2064 			if (error != 0)
2065 				break;
2066 		}
2067 		if (ndp->ni_pathlen > 1)
2068 			cp = PNBUF_GET();
2069 		else
2070 			cp = cnp->cn_pnbuf;
2071 		aiov.iov_base = cp;
2072 		aiov.iov_len = MAXPATHLEN;
2073 		auio.uio_iov = &aiov;
2074 		auio.uio_iovcnt = 1;
2075 		auio.uio_offset = 0;
2076 		auio.uio_rw = UIO_READ;
2077 		auio.uio_segflg = UIO_SYSSPACE;
2078 		auio.uio_procp = (struct proc *)0;
2079 		auio.uio_resid = MAXPATHLEN;
2080 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
2081 		if (error) {
2082 		badlink:
2083 			if (ndp->ni_pathlen > 1)
2084 				PNBUF_PUT(cp);
2085 			break;
2086 		}
2087 		linklen = MAXPATHLEN - auio.uio_resid;
2088 		if (linklen == 0) {
2089 			error = ENOENT;
2090 			goto badlink;
2091 		}
2092 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
2093 			error = ENAMETOOLONG;
2094 			goto badlink;
2095 		}
2096 		if (ndp->ni_pathlen > 1) {
2097 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
2098 			PNBUF_PUT(cnp->cn_pnbuf);
2099 			cnp->cn_pnbuf = cp;
2100 		} else
2101 			cnp->cn_pnbuf[linklen] = '\0';
2102 		ndp->ni_pathlen += linklen;
2103 		vput(ndp->ni_vp);
2104 		dp = ndp->ni_dvp;
2105 		/*
2106 		 * Check if root directory should replace current directory.
2107 		 */
2108 		if (cnp->cn_pnbuf[0] == '/') {
2109 			vrele(dp);
2110 			dp = ndp->ni_rootdir;
2111 			VREF(dp);
2112 		}
2113 	}
2114    }
2115 	vrele(ndp->ni_dvp);
2116 	vput(ndp->ni_vp);
2117 	ndp->ni_vp = NULL;
2118 out:
2119 	PNBUF_PUT(cnp->cn_pnbuf);
2120 	return (error);
2121 }
2122 
2123 /*
2124  * A fiddled version of m_adj() that ensures null fill to a 32-bit
2125  * boundary and only trims off the back end
2126  *
2127  * 1. trim off 'len' bytes as m_adj(mp, -len).
2128  * 2. add zero-padding 'nul' bytes at the end of the mbuf chain.
2129  */
2130 void
2131 nfs_zeropad(mp, len, nul)
2132 	struct mbuf *mp;
2133 	int len;
2134 	int nul;
2135 {
2136 	struct mbuf *m;
2137 	int count;
2138 
2139 	/*
2140 	 * Trim from tail.  Scan the mbuf chain,
2141 	 * calculating its length and finding the last mbuf.
2142 	 * If the adjustment only affects this mbuf, then just
2143 	 * adjust and return.  Otherwise, rescan and truncate
2144 	 * after the remaining size.
2145 	 */
2146 	count = 0;
2147 	m = mp;
2148 	for (;;) {
2149 		count += m->m_len;
2150 		if (m->m_next == NULL)
2151 			break;
2152 		m = m->m_next;
2153 	}
2154 
2155 	KDASSERT(count >= len);
2156 
2157 	if (m->m_len >= len) {
2158 		m->m_len -= len;
2159 	} else {
2160 		count -= len;
2161 		/*
2162 		 * Correct length for chain is "count".
2163 		 * Find the mbuf with last data, adjust its length,
2164 		 * and toss data from remaining mbufs on chain.
2165 		 */
2166 		for (m = mp; m; m = m->m_next) {
2167 			if (m->m_len >= count) {
2168 				m->m_len = count;
2169 				break;
2170 			}
2171 			count -= m->m_len;
2172 		}
2173 		m_freem(m->m_next);
2174 		m->m_next = NULL;
2175 	}
2176 
2177 	KDASSERT(m->m_next == NULL);
2178 
2179 	/*
2180 	 * zero-padding.
2181 	 */
2182 	if (nul > 0) {
2183 		char *cp;
2184 		int i;
2185 
2186 		if (M_ROMAP(m) || M_TRAILINGSPACE(m) < nul) {
2187 			struct mbuf *n;
2188 
2189 			KDASSERT(MLEN >= nul);
2190 			n = m_get(M_WAIT, MT_DATA);
2191 			MCLAIM(n, &nfs_mowner);
2192 			n->m_len = nul;
2193 			n->m_next = NULL;
2194 			m->m_next = n;
2195 			cp = mtod(n, caddr_t);
2196 		} else {
2197 			cp = mtod(m, caddr_t) + m->m_len;
2198 			m->m_len += nul;
2199 		}
2200 		for (i = 0; i < nul; i++)
2201 			*cp++ = '\0';
2202 	}
2203 	return;
2204 }
2205 
2206 /*
2207  * Make these functions instead of macros, so that the kernel text size
2208  * doesn't get too big...
2209  */
2210 void
2211 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp)
2212 	struct nfsrv_descript *nfsd;
2213 	int before_ret;
2214 	struct vattr *before_vap;
2215 	int after_ret;
2216 	struct vattr *after_vap;
2217 	struct mbuf **mbp;
2218 	char **bposp;
2219 {
2220 	struct mbuf *mb = *mbp;
2221 	char *bpos = *bposp;
2222 	u_int32_t *tl;
2223 
2224 	if (before_ret) {
2225 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2226 		*tl = nfs_false;
2227 	} else {
2228 		nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
2229 		*tl++ = nfs_true;
2230 		txdr_hyper(before_vap->va_size, tl);
2231 		tl += 2;
2232 		txdr_nfsv3time(&(before_vap->va_mtime), tl);
2233 		tl += 2;
2234 		txdr_nfsv3time(&(before_vap->va_ctime), tl);
2235 	}
2236 	*bposp = bpos;
2237 	*mbp = mb;
2238 	nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
2239 }
2240 
2241 void
2242 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp)
2243 	struct nfsrv_descript *nfsd;
2244 	int after_ret;
2245 	struct vattr *after_vap;
2246 	struct mbuf **mbp;
2247 	char **bposp;
2248 {
2249 	struct mbuf *mb = *mbp;
2250 	char *bpos = *bposp;
2251 	u_int32_t *tl;
2252 	struct nfs_fattr *fp;
2253 
2254 	if (after_ret) {
2255 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2256 		*tl = nfs_false;
2257 	} else {
2258 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
2259 		*tl++ = nfs_true;
2260 		fp = (struct nfs_fattr *)tl;
2261 		nfsm_srvfattr(nfsd, after_vap, fp);
2262 	}
2263 	*mbp = mb;
2264 	*bposp = bpos;
2265 }
2266 
2267 void
2268 nfsm_srvfattr(nfsd, vap, fp)
2269 	struct nfsrv_descript *nfsd;
2270 	struct vattr *vap;
2271 	struct nfs_fattr *fp;
2272 {
2273 
2274 	fp->fa_nlink = txdr_unsigned(vap->va_nlink);
2275 	fp->fa_uid = txdr_unsigned(vap->va_uid);
2276 	fp->fa_gid = txdr_unsigned(vap->va_gid);
2277 	if (nfsd->nd_flag & ND_NFSV3) {
2278 		fp->fa_type = vtonfsv3_type(vap->va_type);
2279 		fp->fa_mode = vtonfsv3_mode(vap->va_mode);
2280 		txdr_hyper(vap->va_size, &fp->fa3_size);
2281 		txdr_hyper(vap->va_bytes, &fp->fa3_used);
2282 		fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
2283 		fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
2284 		fp->fa3_fsid.nfsuquad[0] = 0;
2285 		fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
2286 		fp->fa3_fileid.nfsuquad[0] = 0;
2287 		fp->fa3_fileid.nfsuquad[1] = txdr_unsigned(vap->va_fileid);
2288 		txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
2289 		txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
2290 		txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
2291 	} else {
2292 		fp->fa_type = vtonfsv2_type(vap->va_type);
2293 		fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
2294 		fp->fa2_size = txdr_unsigned(vap->va_size);
2295 		fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
2296 		if (vap->va_type == VFIFO)
2297 			fp->fa2_rdev = 0xffffffff;
2298 		else
2299 			fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
2300 		fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
2301 		fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
2302 		fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
2303 		txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
2304 		txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
2305 		txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
2306 	}
2307 }
2308 
2309 /*
2310  * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
2311  * 	- look up fsid in mount list (if not found ret error)
2312  *	- get vp and export rights by calling VFS_FHTOVP()
2313  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2314  *	- if not lockflag unlock it with VOP_UNLOCK()
2315  */
2316 int
2317 nfsrv_fhtovp(fhp, lockflag, vpp, cred, slp, nam, rdonlyp, kerbflag, pubflag)
2318 	fhandle_t *fhp;
2319 	int lockflag;
2320 	struct vnode **vpp;
2321 	struct ucred *cred;
2322 	struct nfssvc_sock *slp;
2323 	struct mbuf *nam;
2324 	int *rdonlyp;
2325 	int kerbflag;
2326 {
2327 	struct mount *mp;
2328 	int i;
2329 	struct ucred *credanon;
2330 	int error, exflags;
2331 	struct sockaddr_in *saddr;
2332 
2333 	*vpp = (struct vnode *)0;
2334 
2335 	if (nfs_ispublicfh(fhp)) {
2336 		if (!pubflag || !nfs_pub.np_valid)
2337 			return (ESTALE);
2338 		fhp = &nfs_pub.np_handle;
2339 	}
2340 
2341 	mp = vfs_getvfs(&fhp->fh_fsid);
2342 	if (!mp)
2343 		return (ESTALE);
2344 	error = VFS_CHECKEXP(mp, nam, &exflags, &credanon);
2345 	if (error)
2346 		return (error);
2347 	error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
2348 	if (error)
2349 		return (error);
2350 
2351 	if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
2352 		saddr = mtod(nam, struct sockaddr_in *);
2353 		if ((saddr->sin_family == AF_INET) &&
2354 		    ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
2355 			vput(*vpp);
2356 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2357 		}
2358 #ifdef INET6
2359 		if ((saddr->sin_family == AF_INET6) &&
2360 		    ntohs(saddr->sin_port) >= IPV6PORT_RESERVED) {
2361 			vput(*vpp);
2362 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2363 		}
2364 #endif
2365 	}
2366 	/*
2367 	 * Check/setup credentials.
2368 	 */
2369 	if (exflags & MNT_EXKERB) {
2370 		if (!kerbflag) {
2371 			vput(*vpp);
2372 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2373 		}
2374 	} else if (kerbflag) {
2375 		vput(*vpp);
2376 		return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2377 	} else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
2378 		cred->cr_uid = credanon->cr_uid;
2379 		cred->cr_gid = credanon->cr_gid;
2380 		for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
2381 			cred->cr_groups[i] = credanon->cr_groups[i];
2382 		cred->cr_ngroups = i;
2383 	}
2384 	if (exflags & MNT_EXRDONLY)
2385 		*rdonlyp = 1;
2386 	else
2387 		*rdonlyp = 0;
2388 	if (!lockflag)
2389 		VOP_UNLOCK(*vpp, 0);
2390 	return (0);
2391 }
2392 
2393 /*
2394  * WebNFS: check if a filehandle is a public filehandle. For v3, this
2395  * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
2396  * transformed this to all zeroes in both cases, so check for it.
2397  */
2398 int
2399 nfs_ispublicfh(fhp)
2400 	fhandle_t *fhp;
2401 {
2402 	char *cp = (char *)fhp;
2403 	int i;
2404 
2405 	for (i = 0; i < NFSX_V3FH; i++)
2406 		if (*cp++ != 0)
2407 			return (FALSE);
2408 	return (TRUE);
2409 }
2410 
2411 /*
2412  * This function compares two net addresses by family and returns TRUE
2413  * if they are the same host.
2414  * If there is any doubt, return FALSE.
2415  * The AF_INET family is handled as a special case so that address mbufs
2416  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
2417  */
2418 int
2419 netaddr_match(family, haddr, nam)
2420 	int family;
2421 	union nethostaddr *haddr;
2422 	struct mbuf *nam;
2423 {
2424 	struct sockaddr_in *inetaddr;
2425 
2426 	switch (family) {
2427 	case AF_INET:
2428 		inetaddr = mtod(nam, struct sockaddr_in *);
2429 		if (inetaddr->sin_family == AF_INET &&
2430 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
2431 			return (1);
2432 		break;
2433 #ifdef INET6
2434 	case AF_INET6:
2435 	    {
2436 		struct sockaddr_in6 *sin6_1, *sin6_2;
2437 
2438 		sin6_1 = mtod(nam, struct sockaddr_in6 *);
2439 		sin6_2 = mtod(haddr->had_nam, struct sockaddr_in6 *);
2440 		if (sin6_1->sin6_family == AF_INET6 &&
2441 		    IN6_ARE_ADDR_EQUAL(&sin6_1->sin6_addr, &sin6_2->sin6_addr))
2442 			return 1;
2443 	    }
2444 #endif
2445 #ifdef ISO
2446 	case AF_ISO:
2447 	    {
2448 		struct sockaddr_iso *isoaddr1, *isoaddr2;
2449 
2450 		isoaddr1 = mtod(nam, struct sockaddr_iso *);
2451 		isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *);
2452 		if (isoaddr1->siso_family == AF_ISO &&
2453 		    isoaddr1->siso_nlen > 0 &&
2454 		    isoaddr1->siso_nlen == isoaddr2->siso_nlen &&
2455 		    SAME_ISOADDR(isoaddr1, isoaddr2))
2456 			return (1);
2457 		break;
2458 	    }
2459 #endif	/* ISO */
2460 	default:
2461 		break;
2462 	};
2463 	return (0);
2464 }
2465 
2466 /*
2467  * The write verifier has changed (probably due to a server reboot), so all
2468  * PG_NEEDCOMMIT pages will have to be written again. Since they are marked
2469  * as dirty or are being written out just now, all this takes is clearing
2470  * the PG_NEEDCOMMIT flag. Once done the new write verifier can be set for
2471  * the mount point.
2472  */
2473 void
2474 nfs_clearcommit(mp)
2475 	struct mount *mp;
2476 {
2477 	struct vnode *vp;
2478 	struct nfsnode *np;
2479 	struct vm_page *pg;
2480 	struct nfsmount *nmp = VFSTONFS(mp);
2481 
2482 	lockmgr(&nmp->nm_writeverflock, LK_EXCLUSIVE, NULL);
2483 
2484 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2485 		KASSERT(vp->v_mount == mp);
2486 		if (vp->v_type == VNON)
2487 			continue;
2488 		np = VTONFS(vp);
2489 		np->n_pushlo = np->n_pushhi = np->n_pushedlo =
2490 		    np->n_pushedhi = 0;
2491 		np->n_commitflags &=
2492 		    ~(NFS_COMMIT_PUSH_VALID | NFS_COMMIT_PUSHED_VALID);
2493 		simple_lock(&vp->v_uobj.vmobjlock);
2494 		TAILQ_FOREACH(pg, &vp->v_uobj.memq, listq) {
2495 			pg->flags &= ~PG_NEEDCOMMIT;
2496 		}
2497 		simple_unlock(&vp->v_uobj.vmobjlock);
2498 	}
2499 	simple_lock(&nmp->nm_slock);
2500 	nmp->nm_iflag &= ~NFSMNT_STALEWRITEVERF;
2501 	simple_unlock(&nmp->nm_slock);
2502 	lockmgr(&nmp->nm_writeverflock, LK_RELEASE, NULL);
2503 }
2504 
2505 void
2506 nfs_merge_commit_ranges(vp)
2507 	struct vnode *vp;
2508 {
2509 	struct nfsnode *np = VTONFS(vp);
2510 
2511 	KASSERT(np->n_commitflags & NFS_COMMIT_PUSH_VALID);
2512 
2513 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2514 		np->n_pushedlo = np->n_pushlo;
2515 		np->n_pushedhi = np->n_pushhi;
2516 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2517 	} else {
2518 		if (np->n_pushlo < np->n_pushedlo)
2519 			np->n_pushedlo = np->n_pushlo;
2520 		if (np->n_pushhi > np->n_pushedhi)
2521 			np->n_pushedhi = np->n_pushhi;
2522 	}
2523 
2524 	np->n_pushlo = np->n_pushhi = 0;
2525 	np->n_commitflags &= ~NFS_COMMIT_PUSH_VALID;
2526 
2527 #ifdef NFS_DEBUG_COMMIT
2528 	printf("merge: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2529 	    (unsigned)np->n_pushedhi);
2530 #endif
2531 }
2532 
2533 int
2534 nfs_in_committed_range(vp, off, len)
2535 	struct vnode *vp;
2536 	off_t off, len;
2537 {
2538 	struct nfsnode *np = VTONFS(vp);
2539 	off_t lo, hi;
2540 
2541 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2542 		return 0;
2543 	lo = off;
2544 	hi = lo + len;
2545 
2546 	return (lo >= np->n_pushedlo && hi <= np->n_pushedhi);
2547 }
2548 
2549 int
2550 nfs_in_tobecommitted_range(vp, off, len)
2551 	struct vnode *vp;
2552 	off_t off, len;
2553 {
2554 	struct nfsnode *np = VTONFS(vp);
2555 	off_t lo, hi;
2556 
2557 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2558 		return 0;
2559 	lo = off;
2560 	hi = lo + len;
2561 
2562 	return (lo >= np->n_pushlo && hi <= np->n_pushhi);
2563 }
2564 
2565 void
2566 nfs_add_committed_range(vp, off, len)
2567 	struct vnode *vp;
2568 	off_t off, len;
2569 {
2570 	struct nfsnode *np = VTONFS(vp);
2571 	off_t lo, hi;
2572 
2573 	lo = off;
2574 	hi = lo + len;
2575 
2576 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID)) {
2577 		np->n_pushedlo = lo;
2578 		np->n_pushedhi = hi;
2579 		np->n_commitflags |= NFS_COMMIT_PUSHED_VALID;
2580 	} else {
2581 		if (hi > np->n_pushedhi)
2582 			np->n_pushedhi = hi;
2583 		if (lo < np->n_pushedlo)
2584 			np->n_pushedlo = lo;
2585 	}
2586 #ifdef NFS_DEBUG_COMMIT
2587 	printf("add: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2588 	    (unsigned)np->n_pushedhi);
2589 #endif
2590 }
2591 
2592 void
2593 nfs_del_committed_range(vp, off, len)
2594 	struct vnode *vp;
2595 	off_t off, len;
2596 {
2597 	struct nfsnode *np = VTONFS(vp);
2598 	off_t lo, hi;
2599 
2600 	if (!(np->n_commitflags & NFS_COMMIT_PUSHED_VALID))
2601 		return;
2602 
2603 	lo = off;
2604 	hi = lo + len;
2605 
2606 	if (lo > np->n_pushedhi || hi < np->n_pushedlo)
2607 		return;
2608 	if (lo <= np->n_pushedlo)
2609 		np->n_pushedlo = hi;
2610 	else if (hi >= np->n_pushedhi)
2611 		np->n_pushedhi = lo;
2612 	else {
2613 		/*
2614 		 * XXX There's only one range. If the deleted range
2615 		 * is in the middle, pick the largest of the
2616 		 * contiguous ranges that it leaves.
2617 		 */
2618 		if ((np->n_pushedlo - lo) > (hi - np->n_pushedhi))
2619 			np->n_pushedhi = lo;
2620 		else
2621 			np->n_pushedlo = hi;
2622 	}
2623 #ifdef NFS_DEBUG_COMMIT
2624 	printf("del: committed: %u - %u\n", (unsigned)np->n_pushedlo,
2625 	    (unsigned)np->n_pushedhi);
2626 #endif
2627 }
2628 
2629 void
2630 nfs_add_tobecommitted_range(vp, off, len)
2631 	struct vnode *vp;
2632 	off_t off, len;
2633 {
2634 	struct nfsnode *np = VTONFS(vp);
2635 	off_t lo, hi;
2636 
2637 	lo = off;
2638 	hi = lo + len;
2639 
2640 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID)) {
2641 		np->n_pushlo = lo;
2642 		np->n_pushhi = hi;
2643 		np->n_commitflags |= NFS_COMMIT_PUSH_VALID;
2644 	} else {
2645 		if (lo < np->n_pushlo)
2646 			np->n_pushlo = lo;
2647 		if (hi > np->n_pushhi)
2648 			np->n_pushhi = hi;
2649 	}
2650 #ifdef NFS_DEBUG_COMMIT
2651 	printf("add: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2652 	    (unsigned)np->n_pushhi);
2653 #endif
2654 }
2655 
2656 void
2657 nfs_del_tobecommitted_range(vp, off, len)
2658 	struct vnode *vp;
2659 	off_t off, len;
2660 {
2661 	struct nfsnode *np = VTONFS(vp);
2662 	off_t lo, hi;
2663 
2664 	if (!(np->n_commitflags & NFS_COMMIT_PUSH_VALID))
2665 		return;
2666 
2667 	lo = off;
2668 	hi = lo + len;
2669 
2670 	if (lo > np->n_pushhi || hi < np->n_pushlo)
2671 		return;
2672 
2673 	if (lo <= np->n_pushlo)
2674 		np->n_pushlo = hi;
2675 	else if (hi >= np->n_pushhi)
2676 		np->n_pushhi = lo;
2677 	else {
2678 		/*
2679 		 * XXX There's only one range. If the deleted range
2680 		 * is in the middle, pick the largest of the
2681 		 * contiguous ranges that it leaves.
2682 		 */
2683 		if ((np->n_pushlo - lo) > (hi - np->n_pushhi))
2684 			np->n_pushhi = lo;
2685 		else
2686 			np->n_pushlo = hi;
2687 	}
2688 #ifdef NFS_DEBUG_COMMIT
2689 	printf("del: tobecommitted: %u - %u\n", (unsigned)np->n_pushlo,
2690 	    (unsigned)np->n_pushhi);
2691 #endif
2692 }
2693 
2694 /*
2695  * Map errnos to NFS error numbers. For Version 3 also filter out error
2696  * numbers not specified for the associated procedure.
2697  */
2698 int
2699 nfsrv_errmap(nd, err)
2700 	struct nfsrv_descript *nd;
2701 	int err;
2702 {
2703 	const short *defaulterrp, *errp;
2704 
2705 	if (nd->nd_flag & ND_NFSV3) {
2706 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
2707 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2708 		while (*++errp) {
2709 			if (*errp == err)
2710 				return (err);
2711 			else if (*errp > err)
2712 				break;
2713 		}
2714 		return ((int)*defaulterrp);
2715 	    } else
2716 		return (err & 0xffff);
2717 	}
2718 	if (err <= ELAST)
2719 		return ((int)nfsrv_v2errmap[err - 1]);
2720 	return (NFSERR_IO);
2721 }
2722 
2723 /*
2724  * Sort the group list in increasing numerical order.
2725  * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
2726  *  that used to be here.)
2727  */
2728 void
2729 nfsrvw_sort(list, num)
2730         gid_t *list;
2731         int num;
2732 {
2733 	int i, j;
2734 	gid_t v;
2735 
2736 	/* Insertion sort. */
2737 	for (i = 1; i < num; i++) {
2738 		v = list[i];
2739 		/* find correct slot for value v, moving others up */
2740 		for (j = i; --j >= 0 && v < list[j];)
2741 			list[j + 1] = list[j];
2742 		list[j + 1] = v;
2743 	}
2744 }
2745 
2746 /*
2747  * copy credentials making sure that the result can be compared with memcmp().
2748  */
2749 void
2750 nfsrv_setcred(incred, outcred)
2751 	struct ucred *incred, *outcred;
2752 {
2753 	int i;
2754 
2755 	memset((caddr_t)outcred, 0, sizeof (struct ucred));
2756 	outcred->cr_ref = 1;
2757 	outcred->cr_uid = incred->cr_uid;
2758 	outcred->cr_gid = incred->cr_gid;
2759 	outcred->cr_ngroups = incred->cr_ngroups;
2760 	for (i = 0; i < incred->cr_ngroups; i++)
2761 		outcred->cr_groups[i] = incred->cr_groups[i];
2762 	nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
2763 }
2764 
2765 u_int32_t
2766 nfs_getxid()
2767 {
2768 	static u_int32_t base;
2769 	static u_int32_t nfs_xid = 0;
2770 	static struct simplelock nfs_xidlock = SIMPLELOCK_INITIALIZER;
2771 	u_int32_t newxid;
2772 
2773 	simple_lock(&nfs_xidlock);
2774 	/*
2775 	 * derive initial xid from system time
2776 	 * XXX time is invalid if root not yet mounted
2777 	 */
2778 	if (__predict_false(!base && (rootvp))) {
2779 		struct timeval tv;
2780 
2781 		microtime(&tv);
2782 		base = tv.tv_sec << 12;
2783 		nfs_xid = base;
2784 	}
2785 
2786 	/*
2787 	 * Skip zero xid if it should ever happen.
2788 	 */
2789 	if (__predict_false(++nfs_xid == 0))
2790 		nfs_xid++;
2791 	newxid = nfs_xid;
2792 	simple_unlock(&nfs_xidlock);
2793 
2794 	return txdr_unsigned(newxid);
2795 }
2796 
2797 /*
2798  * assign a new xid for existing request.
2799  * used for NFSERR_JUKEBOX handling.
2800  */
2801 void
2802 nfs_renewxid(struct nfsreq *req)
2803 {
2804 	u_int32_t xid;
2805 	int off;
2806 
2807 	xid = nfs_getxid();
2808 	if (req->r_nmp->nm_sotype == SOCK_STREAM)
2809 		off = sizeof(u_int32_t); /* RPC record mark */
2810 	else
2811 		off = 0;
2812 
2813 	m_copyback(req->r_mreq, off, sizeof(xid), (void *)&xid);
2814 	req->r_xid = xid;
2815 }
2816