xref: /netbsd-src/sys/nfs/nfs_subs.c (revision 4472dbe5e3bd91ef2540bada7a7ca7384627ff9b)
1 /*	$NetBSD: nfs_subs.c,v 1.75 2000/03/30 12:51:16 augustss 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. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)nfs_subs.c	8.8 (Berkeley) 5/22/95
39  */
40 
41 #include "fs_nfs.h"
42 #include "opt_nfsserver.h"
43 #include "opt_iso.h"
44 
45 /*
46  * These functions support the macros and help fiddle mbuf chains for
47  * the nfs op functions. They do things like create the rpc header and
48  * copy data between mbuf chains and uio lists.
49  */
50 #include <sys/param.h>
51 #include <sys/proc.h>
52 #include <sys/systm.h>
53 #include <sys/kernel.h>
54 #include <sys/mount.h>
55 #include <sys/vnode.h>
56 #include <sys/namei.h>
57 #include <sys/mbuf.h>
58 #include <sys/socket.h>
59 #include <sys/stat.h>
60 #include <sys/malloc.h>
61 #include <sys/time.h>
62 #include <sys/dirent.h>
63 
64 #include <vm/vm.h>
65 
66 #include <uvm/uvm_extern.h>
67 
68 #include <nfs/rpcv2.h>
69 #include <nfs/nfsproto.h>
70 #include <nfs/nfsnode.h>
71 #include <nfs/nfs.h>
72 #include <nfs/xdr_subs.h>
73 #include <nfs/nfsm_subs.h>
74 #include <nfs/nfsmount.h>
75 #include <nfs/nqnfs.h>
76 #include <nfs/nfsrtt.h>
77 #include <nfs/nfs_var.h>
78 
79 #include <miscfs/specfs/specdev.h>
80 
81 #include <vm/vm.h>
82 
83 #include <netinet/in.h>
84 #ifdef ISO
85 #include <netiso/iso.h>
86 #endif
87 
88 /*
89  * Data items converted to xdr at startup, since they are constant
90  * This is kinda hokey, but may save a little time doing byte swaps
91  */
92 u_int32_t nfs_xdrneg1;
93 u_int32_t rpc_call, rpc_vers, rpc_reply, rpc_msgdenied, rpc_autherr,
94 	rpc_mismatch, rpc_auth_unix, rpc_msgaccepted,
95 	rpc_auth_kerb;
96 u_int32_t nfs_prog, nqnfs_prog, nfs_true, nfs_false;
97 
98 /* And other global data */
99 static u_int32_t nfs_xid = 0;
100 nfstype nfsv2_type[9] = { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON,
101 		      NFCHR, NFNON };
102 nfstype nfsv3_type[9] = { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK,
103 		      NFFIFO, NFNON };
104 enum vtype nv2tov_type[8] = { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VNON, VNON };
105 enum vtype nv3tov_type[8]={ VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO };
106 int nfs_ticks;
107 
108 /* NFS client/server stats. */
109 struct nfsstats nfsstats;
110 
111 /*
112  * Mapping of old NFS Version 2 RPC numbers to generic numbers.
113  */
114 int nfsv3_procid[NFS_NPROCS] = {
115 	NFSPROC_NULL,
116 	NFSPROC_GETATTR,
117 	NFSPROC_SETATTR,
118 	NFSPROC_NOOP,
119 	NFSPROC_LOOKUP,
120 	NFSPROC_READLINK,
121 	NFSPROC_READ,
122 	NFSPROC_NOOP,
123 	NFSPROC_WRITE,
124 	NFSPROC_CREATE,
125 	NFSPROC_REMOVE,
126 	NFSPROC_RENAME,
127 	NFSPROC_LINK,
128 	NFSPROC_SYMLINK,
129 	NFSPROC_MKDIR,
130 	NFSPROC_RMDIR,
131 	NFSPROC_READDIR,
132 	NFSPROC_FSSTAT,
133 	NFSPROC_NOOP,
134 	NFSPROC_NOOP,
135 	NFSPROC_NOOP,
136 	NFSPROC_NOOP,
137 	NFSPROC_NOOP,
138 	NFSPROC_NOOP,
139 	NFSPROC_NOOP,
140 	NFSPROC_NOOP
141 };
142 
143 /*
144  * and the reverse mapping from generic to Version 2 procedure numbers
145  */
146 int nfsv2_procid[NFS_NPROCS] = {
147 	NFSV2PROC_NULL,
148 	NFSV2PROC_GETATTR,
149 	NFSV2PROC_SETATTR,
150 	NFSV2PROC_LOOKUP,
151 	NFSV2PROC_NOOP,
152 	NFSV2PROC_READLINK,
153 	NFSV2PROC_READ,
154 	NFSV2PROC_WRITE,
155 	NFSV2PROC_CREATE,
156 	NFSV2PROC_MKDIR,
157 	NFSV2PROC_SYMLINK,
158 	NFSV2PROC_CREATE,
159 	NFSV2PROC_REMOVE,
160 	NFSV2PROC_RMDIR,
161 	NFSV2PROC_RENAME,
162 	NFSV2PROC_LINK,
163 	NFSV2PROC_READDIR,
164 	NFSV2PROC_NOOP,
165 	NFSV2PROC_STATFS,
166 	NFSV2PROC_NOOP,
167 	NFSV2PROC_NOOP,
168 	NFSV2PROC_NOOP,
169 	NFSV2PROC_NOOP,
170 	NFSV2PROC_NOOP,
171 	NFSV2PROC_NOOP,
172 	NFSV2PROC_NOOP,
173 };
174 
175 /*
176  * Maps errno values to nfs error numbers.
177  * Use NFSERR_IO as the catch all for ones not specifically defined in
178  * RFC 1094.
179  */
180 static u_char nfsrv_v2errmap[ELAST] = {
181   NFSERR_PERM,	NFSERR_NOENT,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
182   NFSERR_NXIO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
183   NFSERR_IO,	NFSERR_IO,	NFSERR_ACCES,	NFSERR_IO,	NFSERR_IO,
184   NFSERR_IO,	NFSERR_EXIST,	NFSERR_IO,	NFSERR_NODEV,	NFSERR_NOTDIR,
185   NFSERR_ISDIR,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
186   NFSERR_IO,	NFSERR_FBIG,	NFSERR_NOSPC,	NFSERR_IO,	NFSERR_ROFS,
187   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
188   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
189   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
190   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
191   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
192   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
193   NFSERR_IO,	NFSERR_IO,	NFSERR_NAMETOL,	NFSERR_IO,	NFSERR_IO,
194   NFSERR_NOTEMPTY, NFSERR_IO,	NFSERR_IO,	NFSERR_DQUOT,	NFSERR_STALE,
195   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
196   NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,	NFSERR_IO,
197   NFSERR_IO,	NFSERR_IO,
198 };
199 
200 /*
201  * Maps errno values to nfs error numbers.
202  * Although it is not obvious whether or not NFS clients really care if
203  * a returned error value is in the specified list for the procedure, the
204  * safest thing to do is filter them appropriately. For Version 2, the
205  * X/Open XNFS document is the only specification that defines error values
206  * for each RPC (The RFC simply lists all possible error values for all RPCs),
207  * so I have decided to not do this for Version 2.
208  * The first entry is the default error return and the rest are the valid
209  * errors for that RPC in increasing numeric order.
210  */
211 static short nfsv3err_null[] = {
212 	0,
213 	0,
214 };
215 
216 static short nfsv3err_getattr[] = {
217 	NFSERR_IO,
218 	NFSERR_IO,
219 	NFSERR_STALE,
220 	NFSERR_BADHANDLE,
221 	NFSERR_SERVERFAULT,
222 	0,
223 };
224 
225 static short nfsv3err_setattr[] = {
226 	NFSERR_IO,
227 	NFSERR_PERM,
228 	NFSERR_IO,
229 	NFSERR_ACCES,
230 	NFSERR_INVAL,
231 	NFSERR_NOSPC,
232 	NFSERR_ROFS,
233 	NFSERR_DQUOT,
234 	NFSERR_STALE,
235 	NFSERR_BADHANDLE,
236 	NFSERR_NOT_SYNC,
237 	NFSERR_SERVERFAULT,
238 	0,
239 };
240 
241 static short nfsv3err_lookup[] = {
242 	NFSERR_IO,
243 	NFSERR_NOENT,
244 	NFSERR_IO,
245 	NFSERR_ACCES,
246 	NFSERR_NOTDIR,
247 	NFSERR_NAMETOL,
248 	NFSERR_STALE,
249 	NFSERR_BADHANDLE,
250 	NFSERR_SERVERFAULT,
251 	0,
252 };
253 
254 static short nfsv3err_access[] = {
255 	NFSERR_IO,
256 	NFSERR_IO,
257 	NFSERR_STALE,
258 	NFSERR_BADHANDLE,
259 	NFSERR_SERVERFAULT,
260 	0,
261 };
262 
263 static short nfsv3err_readlink[] = {
264 	NFSERR_IO,
265 	NFSERR_IO,
266 	NFSERR_ACCES,
267 	NFSERR_INVAL,
268 	NFSERR_STALE,
269 	NFSERR_BADHANDLE,
270 	NFSERR_NOTSUPP,
271 	NFSERR_SERVERFAULT,
272 	0,
273 };
274 
275 static short nfsv3err_read[] = {
276 	NFSERR_IO,
277 	NFSERR_IO,
278 	NFSERR_NXIO,
279 	NFSERR_ACCES,
280 	NFSERR_INVAL,
281 	NFSERR_STALE,
282 	NFSERR_BADHANDLE,
283 	NFSERR_SERVERFAULT,
284 	NFSERR_JUKEBOX,
285 	0,
286 };
287 
288 static short nfsv3err_write[] = {
289 	NFSERR_IO,
290 	NFSERR_IO,
291 	NFSERR_ACCES,
292 	NFSERR_INVAL,
293 	NFSERR_FBIG,
294 	NFSERR_NOSPC,
295 	NFSERR_ROFS,
296 	NFSERR_DQUOT,
297 	NFSERR_STALE,
298 	NFSERR_BADHANDLE,
299 	NFSERR_SERVERFAULT,
300 	NFSERR_JUKEBOX,
301 	0,
302 };
303 
304 static short nfsv3err_create[] = {
305 	NFSERR_IO,
306 	NFSERR_IO,
307 	NFSERR_ACCES,
308 	NFSERR_EXIST,
309 	NFSERR_NOTDIR,
310 	NFSERR_NOSPC,
311 	NFSERR_ROFS,
312 	NFSERR_NAMETOL,
313 	NFSERR_DQUOT,
314 	NFSERR_STALE,
315 	NFSERR_BADHANDLE,
316 	NFSERR_NOTSUPP,
317 	NFSERR_SERVERFAULT,
318 	0,
319 };
320 
321 static short nfsv3err_mkdir[] = {
322 	NFSERR_IO,
323 	NFSERR_IO,
324 	NFSERR_ACCES,
325 	NFSERR_EXIST,
326 	NFSERR_NOTDIR,
327 	NFSERR_NOSPC,
328 	NFSERR_ROFS,
329 	NFSERR_NAMETOL,
330 	NFSERR_DQUOT,
331 	NFSERR_STALE,
332 	NFSERR_BADHANDLE,
333 	NFSERR_NOTSUPP,
334 	NFSERR_SERVERFAULT,
335 	0,
336 };
337 
338 static short nfsv3err_symlink[] = {
339 	NFSERR_IO,
340 	NFSERR_IO,
341 	NFSERR_ACCES,
342 	NFSERR_EXIST,
343 	NFSERR_NOTDIR,
344 	NFSERR_NOSPC,
345 	NFSERR_ROFS,
346 	NFSERR_NAMETOL,
347 	NFSERR_DQUOT,
348 	NFSERR_STALE,
349 	NFSERR_BADHANDLE,
350 	NFSERR_NOTSUPP,
351 	NFSERR_SERVERFAULT,
352 	0,
353 };
354 
355 static short nfsv3err_mknod[] = {
356 	NFSERR_IO,
357 	NFSERR_IO,
358 	NFSERR_ACCES,
359 	NFSERR_EXIST,
360 	NFSERR_NOTDIR,
361 	NFSERR_NOSPC,
362 	NFSERR_ROFS,
363 	NFSERR_NAMETOL,
364 	NFSERR_DQUOT,
365 	NFSERR_STALE,
366 	NFSERR_BADHANDLE,
367 	NFSERR_NOTSUPP,
368 	NFSERR_SERVERFAULT,
369 	NFSERR_BADTYPE,
370 	0,
371 };
372 
373 static short nfsv3err_remove[] = {
374 	NFSERR_IO,
375 	NFSERR_NOENT,
376 	NFSERR_IO,
377 	NFSERR_ACCES,
378 	NFSERR_NOTDIR,
379 	NFSERR_ROFS,
380 	NFSERR_NAMETOL,
381 	NFSERR_STALE,
382 	NFSERR_BADHANDLE,
383 	NFSERR_SERVERFAULT,
384 	0,
385 };
386 
387 static short nfsv3err_rmdir[] = {
388 	NFSERR_IO,
389 	NFSERR_NOENT,
390 	NFSERR_IO,
391 	NFSERR_ACCES,
392 	NFSERR_EXIST,
393 	NFSERR_NOTDIR,
394 	NFSERR_INVAL,
395 	NFSERR_ROFS,
396 	NFSERR_NAMETOL,
397 	NFSERR_NOTEMPTY,
398 	NFSERR_STALE,
399 	NFSERR_BADHANDLE,
400 	NFSERR_NOTSUPP,
401 	NFSERR_SERVERFAULT,
402 	0,
403 };
404 
405 static short nfsv3err_rename[] = {
406 	NFSERR_IO,
407 	NFSERR_NOENT,
408 	NFSERR_IO,
409 	NFSERR_ACCES,
410 	NFSERR_EXIST,
411 	NFSERR_XDEV,
412 	NFSERR_NOTDIR,
413 	NFSERR_ISDIR,
414 	NFSERR_INVAL,
415 	NFSERR_NOSPC,
416 	NFSERR_ROFS,
417 	NFSERR_MLINK,
418 	NFSERR_NAMETOL,
419 	NFSERR_NOTEMPTY,
420 	NFSERR_DQUOT,
421 	NFSERR_STALE,
422 	NFSERR_BADHANDLE,
423 	NFSERR_NOTSUPP,
424 	NFSERR_SERVERFAULT,
425 	0,
426 };
427 
428 static short nfsv3err_link[] = {
429 	NFSERR_IO,
430 	NFSERR_IO,
431 	NFSERR_ACCES,
432 	NFSERR_EXIST,
433 	NFSERR_XDEV,
434 	NFSERR_NOTDIR,
435 	NFSERR_INVAL,
436 	NFSERR_NOSPC,
437 	NFSERR_ROFS,
438 	NFSERR_MLINK,
439 	NFSERR_NAMETOL,
440 	NFSERR_DQUOT,
441 	NFSERR_STALE,
442 	NFSERR_BADHANDLE,
443 	NFSERR_NOTSUPP,
444 	NFSERR_SERVERFAULT,
445 	0,
446 };
447 
448 static short nfsv3err_readdir[] = {
449 	NFSERR_IO,
450 	NFSERR_IO,
451 	NFSERR_ACCES,
452 	NFSERR_NOTDIR,
453 	NFSERR_STALE,
454 	NFSERR_BADHANDLE,
455 	NFSERR_BAD_COOKIE,
456 	NFSERR_TOOSMALL,
457 	NFSERR_SERVERFAULT,
458 	0,
459 };
460 
461 static short nfsv3err_readdirplus[] = {
462 	NFSERR_IO,
463 	NFSERR_IO,
464 	NFSERR_ACCES,
465 	NFSERR_NOTDIR,
466 	NFSERR_STALE,
467 	NFSERR_BADHANDLE,
468 	NFSERR_BAD_COOKIE,
469 	NFSERR_NOTSUPP,
470 	NFSERR_TOOSMALL,
471 	NFSERR_SERVERFAULT,
472 	0,
473 };
474 
475 static short nfsv3err_fsstat[] = {
476 	NFSERR_IO,
477 	NFSERR_IO,
478 	NFSERR_STALE,
479 	NFSERR_BADHANDLE,
480 	NFSERR_SERVERFAULT,
481 	0,
482 };
483 
484 static short nfsv3err_fsinfo[] = {
485 	NFSERR_STALE,
486 	NFSERR_STALE,
487 	NFSERR_BADHANDLE,
488 	NFSERR_SERVERFAULT,
489 	0,
490 };
491 
492 static short nfsv3err_pathconf[] = {
493 	NFSERR_STALE,
494 	NFSERR_STALE,
495 	NFSERR_BADHANDLE,
496 	NFSERR_SERVERFAULT,
497 	0,
498 };
499 
500 static short nfsv3err_commit[] = {
501 	NFSERR_IO,
502 	NFSERR_IO,
503 	NFSERR_STALE,
504 	NFSERR_BADHANDLE,
505 	NFSERR_SERVERFAULT,
506 	0,
507 };
508 
509 static short *nfsrv_v3errmap[] = {
510 	nfsv3err_null,
511 	nfsv3err_getattr,
512 	nfsv3err_setattr,
513 	nfsv3err_lookup,
514 	nfsv3err_access,
515 	nfsv3err_readlink,
516 	nfsv3err_read,
517 	nfsv3err_write,
518 	nfsv3err_create,
519 	nfsv3err_mkdir,
520 	nfsv3err_symlink,
521 	nfsv3err_mknod,
522 	nfsv3err_remove,
523 	nfsv3err_rmdir,
524 	nfsv3err_rename,
525 	nfsv3err_link,
526 	nfsv3err_readdir,
527 	nfsv3err_readdirplus,
528 	nfsv3err_fsstat,
529 	nfsv3err_fsinfo,
530 	nfsv3err_pathconf,
531 	nfsv3err_commit,
532 };
533 
534 extern struct nfsrtt nfsrtt;
535 extern time_t nqnfsstarttime;
536 extern int nqsrv_clockskew;
537 extern int nqsrv_writeslack;
538 extern int nqsrv_maxlease;
539 extern int nqnfs_piggy[NFS_NPROCS];
540 extern struct nfsnodehashhead *nfsnodehashtbl;
541 extern u_long nfsnodehash;
542 
543 LIST_HEAD(nfsnodehashhead, nfsnode);
544 u_long nfsdirhashmask;
545 
546 int nfs_webnamei __P((struct nameidata *, struct vnode *, struct proc *));
547 
548 /*
549  * Create the header for an rpc request packet
550  * The hsiz is the size of the rest of the nfs request header.
551  * (just used to decide if a cluster is a good idea)
552  */
553 struct mbuf *
554 nfsm_reqh(vp, procid, hsiz, bposp)
555 	struct vnode *vp;
556 	u_long procid;
557 	int hsiz;
558 	caddr_t *bposp;
559 {
560 	struct mbuf *mb;
561 	u_int32_t *tl;
562 	caddr_t bpos;
563 	struct mbuf *mb2;
564 	struct nfsmount *nmp;
565 	int nqflag;
566 
567 	MGET(mb, M_WAIT, MT_DATA);
568 	if (hsiz >= MINCLSIZE)
569 		MCLGET(mb, M_WAIT);
570 	mb->m_len = 0;
571 	bpos = mtod(mb, caddr_t);
572 
573 	/*
574 	 * For NQNFS, add lease request.
575 	 */
576 	if (vp) {
577 		nmp = VFSTONFS(vp->v_mount);
578 		if (nmp->nm_flag & NFSMNT_NQNFS) {
579 			nqflag = NQNFS_NEEDLEASE(vp, procid);
580 			if (nqflag) {
581 				nfsm_build(tl, u_int32_t *, 2*NFSX_UNSIGNED);
582 				*tl++ = txdr_unsigned(nqflag);
583 				*tl = txdr_unsigned(nmp->nm_leaseterm);
584 			} else {
585 				nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
586 				*tl = 0;
587 			}
588 		}
589 	}
590 	/* Finally, return values */
591 	*bposp = bpos;
592 	return (mb);
593 }
594 
595 /*
596  * Build the RPC header and fill in the authorization info.
597  * The authorization string argument is only used when the credentials
598  * come from outside of the kernel.
599  * Returns the head of the mbuf list.
600  */
601 struct mbuf *
602 nfsm_rpchead(cr, nmflag, procid, auth_type, auth_len, auth_str, verf_len,
603 	verf_str, mrest, mrest_len, mbp, xidp)
604 	struct ucred *cr;
605 	int nmflag;
606 	int procid;
607 	int auth_type;
608 	int auth_len;
609 	char *auth_str;
610 	int verf_len;
611 	char *verf_str;
612 	struct mbuf *mrest;
613 	int mrest_len;
614 	struct mbuf **mbp;
615 	u_int32_t *xidp;
616 {
617 	struct mbuf *mb;
618 	u_int32_t *tl;
619 	caddr_t bpos;
620 	int i;
621 	struct mbuf *mreq, *mb2;
622 	int siz, grpsiz, authsiz;
623 	struct timeval tv;
624 	static u_int32_t base;
625 
626 	authsiz = nfsm_rndup(auth_len);
627 	MGETHDR(mb, M_WAIT, MT_DATA);
628 	if ((authsiz + 10 * NFSX_UNSIGNED) >= MINCLSIZE) {
629 		MCLGET(mb, M_WAIT);
630 	} else if ((authsiz + 10 * NFSX_UNSIGNED) < MHLEN) {
631 		MH_ALIGN(mb, authsiz + 10 * NFSX_UNSIGNED);
632 	} else {
633 		MH_ALIGN(mb, 8 * NFSX_UNSIGNED);
634 	}
635 	mb->m_len = 0;
636 	mreq = mb;
637 	bpos = mtod(mb, caddr_t);
638 
639 	/*
640 	 * First the RPC header.
641 	 */
642 	nfsm_build(tl, u_int32_t *, 8 * NFSX_UNSIGNED);
643 
644 	/*
645 	 * derive initial xid from system time
646 	 * XXX time is invalid if root not yet mounted
647 	 */
648 	if (!base && (rootvp)) {
649 		microtime(&tv);
650 		base = tv.tv_sec << 12;
651 		nfs_xid = base;
652 	}
653 	/*
654 	 * Skip zero xid if it should ever happen.
655 	 */
656 	if (++nfs_xid == 0)
657 		nfs_xid++;
658 
659 	*tl++ = *xidp = txdr_unsigned(nfs_xid);
660 	*tl++ = rpc_call;
661 	*tl++ = rpc_vers;
662 	if (nmflag & NFSMNT_NQNFS) {
663 		*tl++ = txdr_unsigned(NQNFS_PROG);
664 		*tl++ = txdr_unsigned(NQNFS_VER3);
665 	} else {
666 		*tl++ = txdr_unsigned(NFS_PROG);
667 		if (nmflag & NFSMNT_NFSV3)
668 			*tl++ = txdr_unsigned(NFS_VER3);
669 		else
670 			*tl++ = txdr_unsigned(NFS_VER2);
671 	}
672 	if (nmflag & NFSMNT_NFSV3)
673 		*tl++ = txdr_unsigned(procid);
674 	else
675 		*tl++ = txdr_unsigned(nfsv2_procid[procid]);
676 
677 	/*
678 	 * And then the authorization cred.
679 	 */
680 	*tl++ = txdr_unsigned(auth_type);
681 	*tl = txdr_unsigned(authsiz);
682 	switch (auth_type) {
683 	case RPCAUTH_UNIX:
684 		nfsm_build(tl, u_int32_t *, auth_len);
685 		*tl++ = 0;		/* stamp ?? */
686 		*tl++ = 0;		/* NULL hostname */
687 		*tl++ = txdr_unsigned(cr->cr_uid);
688 		*tl++ = txdr_unsigned(cr->cr_gid);
689 		grpsiz = (auth_len >> 2) - 5;
690 		*tl++ = txdr_unsigned(grpsiz);
691 		for (i = 0; i < grpsiz; i++)
692 			*tl++ = txdr_unsigned(cr->cr_groups[i]);
693 		break;
694 	case RPCAUTH_KERB4:
695 		siz = auth_len;
696 		while (siz > 0) {
697 			if (M_TRAILINGSPACE(mb) == 0) {
698 				MGET(mb2, M_WAIT, MT_DATA);
699 				if (siz >= MINCLSIZE)
700 					MCLGET(mb2, M_WAIT);
701 				mb->m_next = mb2;
702 				mb = mb2;
703 				mb->m_len = 0;
704 				bpos = mtod(mb, caddr_t);
705 			}
706 			i = min(siz, M_TRAILINGSPACE(mb));
707 			memcpy(bpos, auth_str, i);
708 			mb->m_len += i;
709 			auth_str += i;
710 			bpos += i;
711 			siz -= i;
712 		}
713 		if ((siz = (nfsm_rndup(auth_len) - auth_len)) > 0) {
714 			for (i = 0; i < siz; i++)
715 				*bpos++ = '\0';
716 			mb->m_len += siz;
717 		}
718 		break;
719 	};
720 
721 	/*
722 	 * And the verifier...
723 	 */
724 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
725 	if (verf_str) {
726 		*tl++ = txdr_unsigned(RPCAUTH_KERB4);
727 		*tl = txdr_unsigned(verf_len);
728 		siz = verf_len;
729 		while (siz > 0) {
730 			if (M_TRAILINGSPACE(mb) == 0) {
731 				MGET(mb2, M_WAIT, MT_DATA);
732 				if (siz >= MINCLSIZE)
733 					MCLGET(mb2, M_WAIT);
734 				mb->m_next = mb2;
735 				mb = mb2;
736 				mb->m_len = 0;
737 				bpos = mtod(mb, caddr_t);
738 			}
739 			i = min(siz, M_TRAILINGSPACE(mb));
740 			memcpy(bpos, verf_str, i);
741 			mb->m_len += i;
742 			verf_str += i;
743 			bpos += i;
744 			siz -= i;
745 		}
746 		if ((siz = (nfsm_rndup(verf_len) - verf_len)) > 0) {
747 			for (i = 0; i < siz; i++)
748 				*bpos++ = '\0';
749 			mb->m_len += siz;
750 		}
751 	} else {
752 		*tl++ = txdr_unsigned(RPCAUTH_NULL);
753 		*tl = 0;
754 	}
755 	mb->m_next = mrest;
756 	mreq->m_pkthdr.len = authsiz + 10 * NFSX_UNSIGNED + mrest_len;
757 	mreq->m_pkthdr.rcvif = (struct ifnet *)0;
758 	*mbp = mb;
759 	return (mreq);
760 }
761 
762 /*
763  * copies mbuf chain to the uio scatter/gather list
764  */
765 int
766 nfsm_mbuftouio(mrep, uiop, siz, dpos)
767 	struct mbuf **mrep;
768 	struct uio *uiop;
769 	int siz;
770 	caddr_t *dpos;
771 {
772 	char *mbufcp, *uiocp;
773 	int xfer, left, len;
774 	struct mbuf *mp;
775 	long uiosiz, rem;
776 	int error = 0;
777 
778 	mp = *mrep;
779 	mbufcp = *dpos;
780 	len = mtod(mp, caddr_t)+mp->m_len-mbufcp;
781 	rem = nfsm_rndup(siz)-siz;
782 	while (siz > 0) {
783 		if (uiop->uio_iovcnt <= 0 || uiop->uio_iov == NULL)
784 			return (EFBIG);
785 		left = uiop->uio_iov->iov_len;
786 		uiocp = uiop->uio_iov->iov_base;
787 		if (left > siz)
788 			left = siz;
789 		uiosiz = left;
790 		while (left > 0) {
791 			while (len == 0) {
792 				mp = mp->m_next;
793 				if (mp == NULL)
794 					return (EBADRPC);
795 				mbufcp = mtod(mp, caddr_t);
796 				len = mp->m_len;
797 			}
798 			xfer = (left > len) ? len : left;
799 #ifdef notdef
800 			/* Not Yet.. */
801 			if (uiop->uio_iov->iov_op != NULL)
802 				(*(uiop->uio_iov->iov_op))
803 				(mbufcp, uiocp, xfer);
804 			else
805 #endif
806 			if (uiop->uio_segflg == UIO_SYSSPACE)
807 				memcpy(uiocp, mbufcp, xfer);
808 			else
809 				copyout(mbufcp, uiocp, xfer);
810 			left -= xfer;
811 			len -= xfer;
812 			mbufcp += xfer;
813 			uiocp += xfer;
814 			uiop->uio_offset += xfer;
815 			uiop->uio_resid -= xfer;
816 		}
817 		if (uiop->uio_iov->iov_len <= siz) {
818 			uiop->uio_iovcnt--;
819 			uiop->uio_iov++;
820 		} else {
821 			(caddr_t)uiop->uio_iov->iov_base += uiosiz;
822 			uiop->uio_iov->iov_len -= uiosiz;
823 		}
824 		siz -= uiosiz;
825 	}
826 	*dpos = mbufcp;
827 	*mrep = mp;
828 	if (rem > 0) {
829 		if (len < rem)
830 			error = nfs_adv(mrep, dpos, rem, len);
831 		else
832 			*dpos += rem;
833 	}
834 	return (error);
835 }
836 
837 /*
838  * copies a uio scatter/gather list to an mbuf chain.
839  * NOTE: can ony handle iovcnt == 1
840  */
841 int
842 nfsm_uiotombuf(uiop, mq, siz, bpos)
843 	struct uio *uiop;
844 	struct mbuf **mq;
845 	int siz;
846 	caddr_t *bpos;
847 {
848 	char *uiocp;
849 	struct mbuf *mp, *mp2;
850 	int xfer, left, mlen;
851 	int uiosiz, clflg, rem;
852 	char *cp;
853 
854 #ifdef DIAGNOSTIC
855 	if (uiop->uio_iovcnt != 1)
856 		panic("nfsm_uiotombuf: iovcnt != 1");
857 #endif
858 
859 	if (siz > MLEN)		/* or should it >= MCLBYTES ?? */
860 		clflg = 1;
861 	else
862 		clflg = 0;
863 	rem = nfsm_rndup(siz)-siz;
864 	mp = mp2 = *mq;
865 	while (siz > 0) {
866 		left = uiop->uio_iov->iov_len;
867 		uiocp = uiop->uio_iov->iov_base;
868 		if (left > siz)
869 			left = siz;
870 		uiosiz = left;
871 		while (left > 0) {
872 			mlen = M_TRAILINGSPACE(mp);
873 			if (mlen == 0) {
874 				MGET(mp, M_WAIT, MT_DATA);
875 				if (clflg)
876 					MCLGET(mp, M_WAIT);
877 				mp->m_len = 0;
878 				mp2->m_next = mp;
879 				mp2 = mp;
880 				mlen = M_TRAILINGSPACE(mp);
881 			}
882 			xfer = (left > mlen) ? mlen : left;
883 #ifdef notdef
884 			/* Not Yet.. */
885 			if (uiop->uio_iov->iov_op != NULL)
886 				(*(uiop->uio_iov->iov_op))
887 				(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
888 			else
889 #endif
890 			if (uiop->uio_segflg == UIO_SYSSPACE)
891 				memcpy(mtod(mp, caddr_t)+mp->m_len, uiocp, xfer);
892 			else
893 				copyin(uiocp, mtod(mp, caddr_t)+mp->m_len, xfer);
894 			mp->m_len += xfer;
895 			left -= xfer;
896 			uiocp += xfer;
897 			uiop->uio_offset += xfer;
898 			uiop->uio_resid -= xfer;
899 		}
900 		(caddr_t)uiop->uio_iov->iov_base += uiosiz;
901 		uiop->uio_iov->iov_len -= uiosiz;
902 		siz -= uiosiz;
903 	}
904 	if (rem > 0) {
905 		if (rem > M_TRAILINGSPACE(mp)) {
906 			MGET(mp, M_WAIT, MT_DATA);
907 			mp->m_len = 0;
908 			mp2->m_next = mp;
909 		}
910 		cp = mtod(mp, caddr_t)+mp->m_len;
911 		for (left = 0; left < rem; left++)
912 			*cp++ = '\0';
913 		mp->m_len += rem;
914 		*bpos = cp;
915 	} else
916 		*bpos = mtod(mp, caddr_t)+mp->m_len;
917 	*mq = mp;
918 	return (0);
919 }
920 
921 /*
922  * Get at least "siz" bytes of correctly aligned data.
923  * When called the mbuf pointers are not necessarily correct,
924  * dsosp points to what ought to be in m_data and left contains
925  * what ought to be in m_len.
926  * This is used by the macros nfsm_dissect and nfsm_dissecton for tough
927  * cases. (The macros use the vars. dpos and dpos2)
928  */
929 int
930 nfsm_disct(mdp, dposp, siz, left, cp2)
931 	struct mbuf **mdp;
932 	caddr_t *dposp;
933 	int siz;
934 	int left;
935 	caddr_t *cp2;
936 {
937 	struct mbuf *m1, *m2;
938 	struct mbuf *havebuf = NULL;
939 	caddr_t src = *dposp;
940 	caddr_t dst;
941 	int len;
942 
943 #ifdef DEBUG
944 	if (left < 0)
945 		panic("nfsm_disct: left < 0");
946 #endif
947 	m1 = *mdp;
948 	/*
949 	 * Skip through the mbuf chain looking for an mbuf with
950 	 * some data. If the first mbuf found has enough data
951 	 * and it is correctly aligned return it.
952 	 */
953 	while (left == 0) {
954 		havebuf = m1;
955 		*mdp = m1 = m1->m_next;
956 		if (m1 == NULL)
957 			return (EBADRPC);
958 		src = mtod(m1, caddr_t);
959 		left = m1->m_len;
960 		/*
961 		 * If we start a new mbuf and it is big enough
962 		 * and correctly aligned just return it, don't
963 		 * do any pull up.
964 		 */
965 		if (left >= siz && nfsm_aligned(src)) {
966 			*cp2 = src;
967 			*dposp = src + siz;
968 			return (0);
969 		}
970 	}
971 	if (m1->m_flags & M_EXT) {
972 		if (havebuf) {
973 			/* If the first mbuf with data has external data
974 			 * and there is a previous empty mbuf use it
975 			 * to move the data into.
976 			 */
977 			m2 = m1;
978 			*mdp = m1 = havebuf;
979 			if (m1->m_flags & M_EXT) {
980 				MEXTREMOVE(m1);
981 			}
982 		} else {
983 			/*
984 			 * If the first mbuf has a external data
985 			 * and there is no previous empty mbuf
986 			 * allocate a new mbuf and move the external
987 			 * data to the new mbuf. Also make the first
988 			 * mbuf look empty.
989 			 */
990 			m2 = m_get(M_WAIT, MT_DATA);
991 			m2->m_ext = m1->m_ext;
992 			m2->m_data = src;
993 			m2->m_len = left;
994 			MCLADDREFERENCE(m1, m2);
995 			MEXTREMOVE(m1);
996 			m2->m_next = m1->m_next;
997 			m1->m_next = m2;
998 		}
999 		m1->m_len = 0;
1000 		dst = m1->m_dat;
1001 	} else {
1002 		/*
1003 		 * If the first mbuf has no external data
1004 		 * move the data to the front of the mbuf.
1005 		 */
1006 		if ((dst = m1->m_dat) != src)
1007 			memmove(dst, src, left);
1008 		dst += left;
1009 		m1->m_len = left;
1010 		m2 = m1->m_next;
1011 	}
1012 	m1->m_flags &= ~M_PKTHDR;
1013 	*cp2 = m1->m_data = m1->m_dat;   /* data is at beginning of buffer */
1014 	*dposp = mtod(m1, caddr_t) + siz;
1015 	/*
1016 	 * Loop through mbufs pulling data up into first mbuf until
1017 	 * the first mbuf is full or there is no more data to
1018 	 * pullup.
1019 	 */
1020 	while ((len = (MLEN - m1->m_len)) != 0 && m2) {
1021 		if ((len = min(len, m2->m_len)) != 0)
1022 			memcpy(dst, m2->m_data, len);
1023 		m1->m_len += len;
1024 		dst += len;
1025 		m2->m_data += len;
1026 		m2->m_len -= len;
1027 		m2 = m2->m_next;
1028 	}
1029 	if (m1->m_len < siz)
1030 		return (EBADRPC);
1031 	return (0);
1032 }
1033 
1034 /*
1035  * Advance the position in the mbuf chain.
1036  */
1037 int
1038 nfs_adv(mdp, dposp, offs, left)
1039 	struct mbuf **mdp;
1040 	caddr_t *dposp;
1041 	int offs;
1042 	int left;
1043 {
1044 	struct mbuf *m;
1045 	int s;
1046 
1047 	m = *mdp;
1048 	s = left;
1049 	while (s < offs) {
1050 		offs -= s;
1051 		m = m->m_next;
1052 		if (m == NULL)
1053 			return (EBADRPC);
1054 		s = m->m_len;
1055 	}
1056 	*mdp = m;
1057 	*dposp = mtod(m, caddr_t)+offs;
1058 	return (0);
1059 }
1060 
1061 /*
1062  * Copy a string into mbufs for the hard cases...
1063  */
1064 int
1065 nfsm_strtmbuf(mb, bpos, cp, siz)
1066 	struct mbuf **mb;
1067 	char **bpos;
1068 	const char *cp;
1069 	long siz;
1070 {
1071 	struct mbuf *m1 = NULL, *m2;
1072 	long left, xfer, len, tlen;
1073 	u_int32_t *tl;
1074 	int putsize;
1075 
1076 	putsize = 1;
1077 	m2 = *mb;
1078 	left = M_TRAILINGSPACE(m2);
1079 	if (left > 0) {
1080 		tl = ((u_int32_t *)(*bpos));
1081 		*tl++ = txdr_unsigned(siz);
1082 		putsize = 0;
1083 		left -= NFSX_UNSIGNED;
1084 		m2->m_len += NFSX_UNSIGNED;
1085 		if (left > 0) {
1086 			memcpy((caddr_t) tl, cp, left);
1087 			siz -= left;
1088 			cp += left;
1089 			m2->m_len += left;
1090 			left = 0;
1091 		}
1092 	}
1093 	/* Loop around adding mbufs */
1094 	while (siz > 0) {
1095 		MGET(m1, M_WAIT, MT_DATA);
1096 		if (siz > MLEN)
1097 			MCLGET(m1, M_WAIT);
1098 		m1->m_len = NFSMSIZ(m1);
1099 		m2->m_next = m1;
1100 		m2 = m1;
1101 		tl = mtod(m1, u_int32_t *);
1102 		tlen = 0;
1103 		if (putsize) {
1104 			*tl++ = txdr_unsigned(siz);
1105 			m1->m_len -= NFSX_UNSIGNED;
1106 			tlen = NFSX_UNSIGNED;
1107 			putsize = 0;
1108 		}
1109 		if (siz < m1->m_len) {
1110 			len = nfsm_rndup(siz);
1111 			xfer = siz;
1112 			if (xfer < len)
1113 				*(tl+(xfer>>2)) = 0;
1114 		} else {
1115 			xfer = len = m1->m_len;
1116 		}
1117 		memcpy((caddr_t) tl, cp, xfer);
1118 		m1->m_len = len+tlen;
1119 		siz -= xfer;
1120 		cp += xfer;
1121 	}
1122 	*mb = m1;
1123 	*bpos = mtod(m1, caddr_t)+m1->m_len;
1124 	return (0);
1125 }
1126 
1127 /*
1128  * Directory caching routines. They work as follows:
1129  * - a cache is maintained per VDIR nfsnode.
1130  * - for each offset cookie that is exported to userspace, and can
1131  *   thus be thrown back at us as an offset to VOP_READDIR, store
1132  *   information in the cache.
1133  * - cached are:
1134  *   - cookie itself
1135  *   - blocknumber (essentially just a search key in the buffer cache)
1136  *   - entry number in block.
1137  *   - offset cookie of block in which this entry is stored
1138  *   - 32 bit cookie if NFSMNT_XLATECOOKIE is used.
1139  * - entries are looked up in a hash table
1140  * - also maintained is an LRU list of entries, used to determine
1141  *   which ones to delete if the cache grows too large.
1142  * - if 32 <-> 64 translation mode is requested for a filesystem,
1143  *   the cache also functions as a translation table
1144  * - in the translation case, invalidating the cache does not mean
1145  *   flushing it, but just marking entries as invalid, except for
1146  *   the <64bit cookie, 32bitcookie> pair which is still valid, to
1147  *   still be able to use the cache as a translation table.
1148  * - 32 bit cookies are uniquely created by combining the hash table
1149  *   entry value, and one generation count per hash table entry,
1150  *   incremented each time an entry is appended to the chain.
1151  * - the cache is invalidated each time a direcory is modified
1152  * - sanity checks are also done; if an entry in a block turns
1153  *   out not to have a matching cookie, the cache is invalidated
1154  *   and a new block starting from the wanted offset is fetched from
1155  *   the server.
1156  * - directory entries as read from the server are extended to contain
1157  *   the 64bit and, optionally, the 32bit cookies, for sanity checking
1158  *   the cache and exporting them to userspace through the cookie
1159  *   argument to VOP_READDIR.
1160  */
1161 
1162 u_long
1163 nfs_dirhash(off)
1164 	off_t off;
1165 {
1166 	int i;
1167 	char *cp = (char *)&off;
1168 	u_long sum = 0L;
1169 
1170 	for (i = 0 ; i < sizeof (off); i++)
1171 		sum += *cp++;
1172 
1173 	return sum;
1174 }
1175 
1176 void
1177 nfs_initdircache(vp)
1178 	struct vnode *vp;
1179 {
1180 	struct nfsnode *np = VTONFS(vp);
1181 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1182 
1183 	np->n_dircachesize = 0;
1184 	np->n_dblkno = 1;
1185 	np->n_dircache =
1186 	    hashinit(NFS_DIRHASHSIZ, M_NFSDIROFF, M_WAITOK, &nfsdirhashmask);
1187 	TAILQ_INIT(&np->n_dirchain);
1188 	if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1189 		MALLOC(np->n_dirgens, unsigned *,
1190 		    NFS_DIRHASHSIZ * sizeof (unsigned), M_NFSDIROFF,
1191 		    M_WAITOK);
1192 		memset((caddr_t)np->n_dirgens, 0,
1193 		    NFS_DIRHASHSIZ * sizeof (unsigned));
1194 	}
1195 }
1196 
1197 static struct nfsdircache dzero = {0, 0, {0, 0}, {0, 0}, 0, 0, 0};
1198 
1199 struct nfsdircache *
1200 nfs_searchdircache(vp, off, do32, hashent)
1201 	struct vnode *vp;
1202 	off_t off;
1203 	int do32;
1204 	int *hashent;
1205 {
1206 	struct nfsdirhashhead *ndhp;
1207 	struct nfsdircache *ndp = NULL;
1208 	struct nfsnode *np = VTONFS(vp);
1209 	unsigned ent;
1210 
1211 	/*
1212 	 * Zero is always a valid cookie.
1213 	 */
1214 	if (off == 0)
1215 		return &dzero;
1216 
1217 	/*
1218 	 * We use a 32bit cookie as search key, directly reconstruct
1219 	 * the hashentry. Else use the hashfunction.
1220 	 */
1221 	if (do32) {
1222 		ent = (u_int32_t)off >> 24;
1223 		if (ent >= NFS_DIRHASHSIZ)
1224 			return NULL;
1225 		ndhp = &np->n_dircache[ent];
1226 	} else {
1227 		ndhp = NFSDIRHASH(np, off);
1228 	}
1229 
1230 	if (hashent)
1231 		*hashent = (int)(ndhp - np->n_dircache);
1232 	if (do32) {
1233 		for (ndp = ndhp->lh_first; ndp; ndp = ndp->dc_hash.le_next) {
1234 			if (ndp->dc_cookie32 == (u_int32_t)off) {
1235 				/*
1236 				 * An invalidated entry will become the
1237 				 * start of a new block fetched from
1238 				 * the server.
1239 				 */
1240 				if (ndp->dc_blkno == -1) {
1241 					ndp->dc_blkcookie = ndp->dc_cookie;
1242 					ndp->dc_blkno = np->n_dblkno++;
1243 					ndp->dc_entry = 0;
1244 				}
1245 				break;
1246 			}
1247 		}
1248 	} else {
1249 		for (ndp = ndhp->lh_first; ndp; ndp = ndp->dc_hash.le_next)
1250 			if (ndp->dc_cookie == off)
1251 				break;
1252 	}
1253 	return ndp;
1254 }
1255 
1256 
1257 struct nfsdircache *
1258 nfs_enterdircache(vp, off, blkoff, en, blkno)
1259 	struct vnode *vp;
1260 	off_t off, blkoff;
1261 	daddr_t blkno;
1262 	int en;
1263 {
1264 	struct nfsnode *np = VTONFS(vp);
1265 	struct nfsdirhashhead *ndhp;
1266 	struct nfsdircache *ndp = NULL, *first;
1267 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1268 	int hashent, gen, overwrite;
1269 
1270 	if (!np->n_dircache)
1271 		/*
1272 		 * XXX would like to do this in nfs_nget but vtype
1273 		 * isn't known at that time.
1274 		 */
1275 		nfs_initdircache(vp);
1276 
1277 	/*
1278 	 * XXX refuse entries for offset 0. amd(8) erroneously sets
1279 	 * cookie 0 for the '.' entry, making this necessary. This
1280 	 * isn't so bad, as 0 is a special case anyway.
1281 	 */
1282 	if (off == 0)
1283 		return &dzero;
1284 
1285 	ndp = nfs_searchdircache(vp, off, 0, &hashent);
1286 
1287 	if (ndp && ndp->dc_blkno != -1) {
1288 		/*
1289 		 * Overwriting an old entry. Check if it's the same.
1290 		 * If so, just return. If not, remove the old entry.
1291 		 */
1292 		if (ndp->dc_blkcookie == blkoff && ndp->dc_entry == en)
1293 			return ndp;
1294 		TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1295 		LIST_REMOVE(ndp, dc_hash);
1296 		FREE(ndp, M_NFSDIROFF);
1297 		ndp = 0;
1298 	}
1299 
1300 	ndhp = &np->n_dircache[hashent];
1301 
1302 	if (!ndp) {
1303 		MALLOC(ndp, struct nfsdircache *, sizeof (*ndp), M_NFSDIROFF,
1304 		    M_WAITOK);
1305 		overwrite = 0;
1306 		if (nmp->nm_flag & NFSMNT_XLATECOOKIE) {
1307 			/*
1308 			 * We're allocating a new entry, so bump the
1309 			 * generation number.
1310 			 */
1311 			gen = ++np->n_dirgens[hashent];
1312 			if (gen == 0) {
1313 				np->n_dirgens[hashent]++;
1314 				gen++;
1315 			}
1316 			ndp->dc_cookie32 = (hashent << 24) | (gen & 0xffffff);
1317 		}
1318 	} else
1319 		overwrite = 1;
1320 
1321 	/*
1322 	 * If the entry number is 0, we are at the start of a new block, so
1323 	 * allocate a new blocknumber.
1324 	 */
1325 	if (en == 0)
1326 		ndp->dc_blkno = np->n_dblkno++;
1327 	else
1328 		ndp->dc_blkno = blkno;
1329 
1330 	ndp->dc_cookie = off;
1331 	ndp->dc_blkcookie = blkoff;
1332 	ndp->dc_entry = en;
1333 
1334 	if (overwrite)
1335 		return ndp;
1336 
1337 	/*
1338 	 * If the maximum directory cookie cache size has been reached
1339 	 * for this node, take one off the front. The idea is that
1340 	 * directories are typically read front-to-back once, so that
1341 	 * the oldest entries can be thrown away without much performance
1342 	 * loss.
1343 	 */
1344 	if (np->n_dircachesize == NFS_MAXDIRCACHE) {
1345 		first = np->n_dirchain.tqh_first;
1346 		TAILQ_REMOVE(&np->n_dirchain, first, dc_chain);
1347 		LIST_REMOVE(first, dc_hash);
1348 		FREE(first, M_NFSDIROFF);
1349 	} else
1350 		np->n_dircachesize++;
1351 
1352 	LIST_INSERT_HEAD(ndhp, ndp, dc_hash);
1353 	TAILQ_INSERT_TAIL(&np->n_dirchain, ndp, dc_chain);
1354 	return ndp;
1355 }
1356 
1357 void
1358 nfs_invaldircache(vp, forcefree)
1359 	struct vnode *vp;
1360 	int forcefree;
1361 {
1362 	struct nfsnode *np = VTONFS(vp);
1363 	struct nfsdircache *ndp = NULL;
1364 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1365 
1366 #ifdef DIAGNOSTIC
1367 	if (vp->v_type != VDIR)
1368 		panic("nfs: invaldircache: not dir");
1369 #endif
1370 
1371 	if (!np->n_dircache)
1372 		return;
1373 
1374 	if (!(nmp->nm_flag & NFSMNT_XLATECOOKIE) || forcefree) {
1375 		while ((ndp = np->n_dirchain.tqh_first)) {
1376 			TAILQ_REMOVE(&np->n_dirchain, ndp, dc_chain);
1377 			LIST_REMOVE(ndp, dc_hash);
1378 			FREE(ndp, M_NFSDIROFF);
1379 		}
1380 		np->n_dircachesize = 0;
1381 		if (forcefree && np->n_dirgens) {
1382 			FREE(np->n_dirgens, M_NFSDIROFF);
1383 		}
1384 	} else {
1385 		for (ndp = np->n_dirchain.tqh_first; ndp;
1386 		    ndp = ndp->dc_chain.tqe_next)
1387 			ndp->dc_blkno = -1;
1388 	}
1389 
1390 	np->n_dblkno = 1;
1391 }
1392 
1393 /*
1394  * Called once before VFS init to initialize shared and
1395  * server-specific data structures.
1396  */
1397 void
1398 nfs_init()
1399 {
1400 
1401 #if !defined(alpha) && defined(DIAGNOSTIC)
1402 	/*
1403 	 * Check to see if major data structures haven't bloated.
1404 	 */
1405 	if (sizeof (struct nfsnode) > NFS_NODEALLOC) {
1406 		printf("struct nfsnode bloated (> %dbytes)\n", NFS_NODEALLOC);
1407 		printf("Try reducing NFS_SMALLFH\n");
1408 	}
1409 	if (sizeof (struct nfssvc_sock) > NFS_SVCALLOC) {
1410 		printf("struct nfssvc_sock bloated (> %dbytes)\n",NFS_SVCALLOC);
1411 		printf("Try reducing NFS_UIDHASHSIZ\n");
1412 	}
1413 	if (sizeof (struct nfsuid) > NFS_UIDALLOC) {
1414 		printf("struct nfsuid bloated (> %dbytes)\n",NFS_UIDALLOC);
1415 		printf("Try unionizing the nu_nickname and nu_flag fields\n");
1416 	}
1417 #endif
1418 
1419 	nfsrtt.pos = 0;
1420 	rpc_vers = txdr_unsigned(RPC_VER2);
1421 	rpc_call = txdr_unsigned(RPC_CALL);
1422 	rpc_reply = txdr_unsigned(RPC_REPLY);
1423 	rpc_msgdenied = txdr_unsigned(RPC_MSGDENIED);
1424 	rpc_msgaccepted = txdr_unsigned(RPC_MSGACCEPTED);
1425 	rpc_mismatch = txdr_unsigned(RPC_MISMATCH);
1426 	rpc_autherr = txdr_unsigned(RPC_AUTHERR);
1427 	rpc_auth_unix = txdr_unsigned(RPCAUTH_UNIX);
1428 	rpc_auth_kerb = txdr_unsigned(RPCAUTH_KERB4);
1429 	nfs_prog = txdr_unsigned(NFS_PROG);
1430 	nqnfs_prog = txdr_unsigned(NQNFS_PROG);
1431 	nfs_true = txdr_unsigned(TRUE);
1432 	nfs_false = txdr_unsigned(FALSE);
1433 	nfs_xdrneg1 = txdr_unsigned(-1);
1434 	nfs_ticks = (hz * NFS_TICKINTVL + 500) / 1000;
1435 	if (nfs_ticks < 1)
1436 		nfs_ticks = 1;
1437 #ifdef NFSSERVER
1438 	nfsrv_init(0);			/* Init server data structures */
1439 	nfsrv_initcache();		/* Init the server request cache */
1440 #endif /* NFSSERVER */
1441 
1442 	/*
1443 	 * Initialize the nqnfs data structures.
1444 	 */
1445 	if (nqnfsstarttime == 0) {
1446 		nqnfsstarttime = boottime.tv_sec + nqsrv_maxlease
1447 			+ nqsrv_clockskew + nqsrv_writeslack;
1448 		NQLOADNOVRAM(nqnfsstarttime);
1449 		CIRCLEQ_INIT(&nqtimerhead);
1450 		nqfhhashtbl = hashinit(NQLCHSZ, M_NQLEASE, M_WAITOK, &nqfhhash);
1451 	}
1452 
1453 	/*
1454 	 * Initialize reply list and start timer
1455 	 */
1456 	TAILQ_INIT(&nfs_reqq);
1457 	nfs_timer(NULL);
1458 }
1459 
1460 #ifdef NFS
1461 /*
1462  * Called once at VFS init to initialize client-specific data structures.
1463  */
1464 void
1465 nfs_vfs_init()
1466 {
1467 	int i;
1468 
1469 	/* Ensure async daemons disabled */
1470 	for (i = 0; i < NFS_MAXASYNCDAEMON; i++) {
1471 		nfs_iodwant[i] = (struct proc *)0;
1472 		nfs_iodmount[i] = (struct nfsmount *)0;
1473 	}
1474 	nfs_nhinit();			/* Init the nfsnode table */
1475 }
1476 
1477 void
1478 nfs_vfs_done()
1479 {
1480 	nfs_nhdone();
1481 }
1482 
1483 /*
1484  * Attribute cache routines.
1485  * nfs_loadattrcache() - loads or updates the cache contents from attributes
1486  *	that are on the mbuf list
1487  * nfs_getattrcache() - returns valid attributes if found in cache, returns
1488  *	error otherwise
1489  */
1490 
1491 /*
1492  * Load the attribute cache (that lives in the nfsnode entry) with
1493  * the values on the mbuf list and
1494  * Iff vap not NULL
1495  *    copy the attributes to *vaper
1496  */
1497 int
1498 nfsm_loadattrcache(vpp, mdp, dposp, vaper)
1499 	struct vnode **vpp;
1500 	struct mbuf **mdp;
1501 	caddr_t *dposp;
1502 	struct vattr *vaper;
1503 {
1504 	int32_t t1;
1505 	caddr_t cp2;
1506 	int error = 0;
1507 	struct mbuf *md;
1508 	int v3 = NFS_ISV3(*vpp);
1509 
1510 	md = *mdp;
1511 	t1 = (mtod(md, caddr_t) + md->m_len) - *dposp;
1512 	error = nfsm_disct(mdp, dposp, NFSX_FATTR(v3), t1, &cp2);
1513 	if (error)
1514 		return (error);
1515 	return nfs_loadattrcache(vpp, (struct nfs_fattr *)cp2, vaper);
1516 }
1517 
1518 int
1519 nfs_loadattrcache(vpp, fp, vaper)
1520 	struct vnode **vpp;
1521 	struct nfs_fattr *fp;
1522 	struct vattr *vaper;
1523 {
1524 	struct vnode *vp = *vpp;
1525 	struct vattr *vap;
1526 	int v3 = NFS_ISV3(vp);
1527 	enum vtype vtyp;
1528 	u_short vmode;
1529 	struct timespec mtime;
1530 	struct vnode *nvp;
1531 	int32_t rdev;
1532 	struct nfsnode *np;
1533 	extern int (**spec_nfsv2nodeop_p) __P((void *));
1534 
1535 	if (v3) {
1536 		vtyp = nfsv3tov_type(fp->fa_type);
1537 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1538 		rdev = makedev(fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata1),
1539 			fxdr_unsigned(u_int32_t, fp->fa3_rdev.specdata2));
1540 		fxdr_nfsv3time(&fp->fa3_mtime, &mtime);
1541 	} else {
1542 		vtyp = nfsv2tov_type(fp->fa_type);
1543 		vmode = fxdr_unsigned(u_short, fp->fa_mode);
1544 		if (vtyp == VNON || vtyp == VREG)
1545 			vtyp = IFTOVT(vmode);
1546 		rdev = fxdr_unsigned(int32_t, fp->fa2_rdev);
1547 		fxdr_nfsv2time(&fp->fa2_mtime, &mtime);
1548 
1549 		/*
1550 		 * Really ugly NFSv2 kludge.
1551 		 */
1552 		if (vtyp == VCHR && rdev == 0xffffffff)
1553 			vtyp = VFIFO;
1554 	}
1555 
1556 	/*
1557 	 * If v_type == VNON it is a new node, so fill in the v_type,
1558 	 * n_mtime fields. Check to see if it represents a special
1559 	 * device, and if so, check for a possible alias. Once the
1560 	 * correct vnode has been obtained, fill in the rest of the
1561 	 * information.
1562 	 */
1563 	np = VTONFS(vp);
1564 	if (vp->v_type != vtyp) {
1565 		vp->v_type = vtyp;
1566 		if (vp->v_type == VFIFO) {
1567 			extern int (**fifo_nfsv2nodeop_p) __P((void *));
1568 			vp->v_op = fifo_nfsv2nodeop_p;
1569 		}
1570 		if (vp->v_type == VCHR || vp->v_type == VBLK) {
1571 			vp->v_op = spec_nfsv2nodeop_p;
1572 			nvp = checkalias(vp, (dev_t)rdev, vp->v_mount);
1573 			if (nvp) {
1574 				/*
1575 				 * Discard unneeded vnode, but save its nfsnode.
1576 				 * Since the nfsnode does not have a lock, its
1577 				 * vnode lock has to be carried over.
1578 				 */
1579 				nvp->v_data = vp->v_data;
1580 				vp->v_data = NULL;
1581 				vp->v_op = spec_vnodeop_p;
1582 				vput(vp);
1583 				vgone(vp);
1584 				/*
1585 				 * XXX When nfs starts locking, we need to
1586 				 * lock the new node here.
1587 				 */
1588 				/*
1589 				 * Reinitialize aliased node.
1590 				 */
1591 				np->n_vnode = nvp;
1592 				*vpp = vp = nvp;
1593 			}
1594 		}
1595 		np->n_mtime = mtime.tv_sec;
1596 	}
1597 	vap = np->n_vattr;
1598 	vap->va_type = vtyp;
1599 	vap->va_mode = vmode & ALLPERMS;
1600 	vap->va_rdev = (dev_t)rdev;
1601 	vap->va_mtime = mtime;
1602 	vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
1603 	switch (vtyp) {
1604 	case VDIR:
1605 		vap->va_blocksize = NFS_DIRFRAGSIZ;
1606 		break;
1607 	case VBLK:
1608 		vap->va_blocksize = BLKDEV_IOSIZE;
1609 		break;
1610 	case VCHR:
1611 		vap->va_blocksize = MAXBSIZE;
1612 		break;
1613 	default:
1614 		vap->va_blocksize = v3 ? vp->v_mount->mnt_stat.f_iosize :
1615 		    fxdr_unsigned(int32_t, fp->fa2_blocksize);
1616 		break;
1617 	}
1618 	if (v3) {
1619 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1620 		vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
1621 		vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
1622 		vap->va_size = fxdr_hyper(&fp->fa3_size);
1623 		vap->va_bytes = fxdr_hyper(&fp->fa3_used);
1624 		vap->va_fileid = fxdr_unsigned(int32_t,
1625 		    fp->fa3_fileid.nfsuquad[1]);
1626 		fxdr_nfsv3time(&fp->fa3_atime, &vap->va_atime);
1627 		fxdr_nfsv3time(&fp->fa3_ctime, &vap->va_ctime);
1628 		vap->va_flags = 0;
1629 		vap->va_filerev = 0;
1630 	} else {
1631 		vap->va_nlink = fxdr_unsigned(u_short, fp->fa_nlink);
1632 		vap->va_uid = fxdr_unsigned(uid_t, fp->fa_uid);
1633 		vap->va_gid = fxdr_unsigned(gid_t, fp->fa_gid);
1634 		vap->va_size = fxdr_unsigned(u_int32_t, fp->fa2_size);
1635 		vap->va_bytes = fxdr_unsigned(int32_t, fp->fa2_blocks)
1636 		    * NFS_FABLKSIZE;
1637 		vap->va_fileid = fxdr_unsigned(int32_t, fp->fa2_fileid);
1638 		fxdr_nfsv2time(&fp->fa2_atime, &vap->va_atime);
1639 		vap->va_flags = 0;
1640 		vap->va_ctime.tv_sec = fxdr_unsigned(u_int32_t,
1641 		    fp->fa2_ctime.nfsv2_sec);
1642 		vap->va_ctime.tv_nsec = 0;
1643 		vap->va_gen = fxdr_unsigned(u_int32_t,fp->fa2_ctime.nfsv2_usec);
1644 		vap->va_filerev = 0;
1645 	}
1646 	if (vap->va_size != np->n_size) {
1647 		if (vap->va_type == VREG) {
1648 			if (np->n_flag & NMODIFIED) {
1649 				if (vap->va_size < np->n_size)
1650 					vap->va_size = np->n_size;
1651 				else
1652 					np->n_size = vap->va_size;
1653 			} else
1654 				np->n_size = vap->va_size;
1655 			uvm_vnp_setsize(vp, np->n_size);
1656 		} else
1657 			np->n_size = vap->va_size;
1658 	}
1659 	np->n_attrstamp = time.tv_sec;
1660 	if (vaper != NULL) {
1661 		memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(*vap));
1662 		if (np->n_flag & NCHG) {
1663 			if (np->n_flag & NACC)
1664 				vaper->va_atime = np->n_atim;
1665 			if (np->n_flag & NUPD)
1666 				vaper->va_mtime = np->n_mtim;
1667 		}
1668 	}
1669 	return (0);
1670 }
1671 
1672 /*
1673  * Check the time stamp
1674  * If the cache is valid, copy contents to *vap and return 0
1675  * otherwise return an error
1676  */
1677 int
1678 nfs_getattrcache(vp, vaper)
1679 	struct vnode *vp;
1680 	struct vattr *vaper;
1681 {
1682 	struct nfsnode *np = VTONFS(vp);
1683 	struct vattr *vap;
1684 
1685 	if ((time.tv_sec - np->n_attrstamp) >= NFS_ATTRTIMEO(np)) {
1686 		nfsstats.attrcache_misses++;
1687 		return (ENOENT);
1688 	}
1689 	nfsstats.attrcache_hits++;
1690 	vap = np->n_vattr;
1691 	if (vap->va_size != np->n_size) {
1692 		if (vap->va_type == VREG) {
1693 			if (np->n_flag & NMODIFIED) {
1694 				if (vap->va_size < np->n_size)
1695 					vap->va_size = np->n_size;
1696 				else
1697 					np->n_size = vap->va_size;
1698 			} else
1699 				np->n_size = vap->va_size;
1700 			uvm_vnp_setsize(vp, np->n_size);
1701 		} else
1702 			np->n_size = vap->va_size;
1703 	}
1704 	memcpy((caddr_t)vaper, (caddr_t)vap, sizeof(struct vattr));
1705 	if (np->n_flag & NCHG) {
1706 		if (np->n_flag & NACC)
1707 			vaper->va_atime = np->n_atim;
1708 		if (np->n_flag & NUPD)
1709 			vaper->va_mtime = np->n_mtim;
1710 	}
1711 	return (0);
1712 }
1713 
1714 /*
1715  * Heuristic to see if the server XDR encodes directory cookies or not.
1716  * it is not supposed to, but a lot of servers may do this. Also, since
1717  * most/all servers will implement V2 as well, it is expected that they
1718  * may return just 32 bits worth of cookie information, so we need to
1719  * find out in which 32 bits this information is available. We do this
1720  * to avoid trouble with emulated binaries that can't handle 64 bit
1721  * directory offsets.
1722  */
1723 
1724 void
1725 nfs_cookieheuristic(vp, flagp, p, cred)
1726 	struct vnode *vp;
1727 	int *flagp;
1728 	struct proc *p;
1729 	struct ucred *cred;
1730 {
1731 	struct uio auio;
1732 	struct iovec aiov;
1733 	caddr_t buf, cp;
1734 	struct dirent *dp;
1735 	off_t *cookies = NULL, *cop;
1736 	int error, eof, nc, len;
1737 
1738 	MALLOC(buf, caddr_t, NFS_DIRFRAGSIZ, M_TEMP, M_WAITOK);
1739 
1740 	aiov.iov_base = buf;
1741 	aiov.iov_len = NFS_DIRFRAGSIZ;
1742 	auio.uio_iov = &aiov;
1743 	auio.uio_iovcnt = 1;
1744 	auio.uio_rw = UIO_READ;
1745 	auio.uio_segflg = UIO_SYSSPACE;
1746 	auio.uio_procp = p;
1747 	auio.uio_resid = NFS_DIRFRAGSIZ;
1748 	auio.uio_offset = 0;
1749 
1750 	error = VOP_READDIR(vp, &auio, cred, &eof, &cookies, &nc);
1751 
1752 	len = NFS_DIRFRAGSIZ - auio.uio_resid;
1753 	if (error || len == 0) {
1754 		FREE(buf, M_TEMP);
1755 		if (cookies)
1756 			FREE(cookies, M_TEMP);
1757 		return;
1758 	}
1759 
1760 	/*
1761 	 * Find the first valid entry and look at its offset cookie.
1762 	 */
1763 
1764 	cp = buf;
1765 	for (cop = cookies; len > 0; len -= dp->d_reclen) {
1766 		dp = (struct dirent *)cp;
1767 		if (dp->d_fileno != 0 && len >= dp->d_reclen) {
1768 			if ((*cop >> 32) != 0 && (*cop & 0xffffffffLL) == 0) {
1769 				*flagp |= NFSMNT_SWAPCOOKIE;
1770 				nfs_invaldircache(vp, 0);
1771 				nfs_vinvalbuf(vp, 0, cred, p, 1);
1772 			}
1773 			break;
1774 		}
1775 		cop++;
1776 		cp += dp->d_reclen;
1777 	}
1778 
1779 	FREE(buf, M_TEMP);
1780 	FREE(cookies, M_TEMP);
1781 }
1782 #endif /* NFS */
1783 
1784 /*
1785  * Set up nameidata for a lookup() call and do it.
1786  *
1787  * If pubflag is set, this call is done for a lookup operation on the
1788  * public filehandle. In that case we allow crossing mountpoints and
1789  * absolute pathnames. However, the caller is expected to check that
1790  * the lookup result is within the public fs, and deny access if
1791  * it is not.
1792  */
1793 int
1794 nfs_namei(ndp, fhp, len, slp, nam, mdp, dposp, retdirp, p, kerbflag, pubflag)
1795 	struct nameidata *ndp;
1796 	fhandle_t *fhp;
1797 	int len;
1798 	struct nfssvc_sock *slp;
1799 	struct mbuf *nam;
1800 	struct mbuf **mdp;
1801 	caddr_t *dposp;
1802 	struct vnode **retdirp;
1803 	struct proc *p;
1804 	int kerbflag, pubflag;
1805 {
1806 	int i, rem;
1807 	struct mbuf *md;
1808 	char *fromcp, *tocp, *cp;
1809 	struct iovec aiov;
1810 	struct uio auio;
1811 	struct vnode *dp;
1812 	int error, rdonly, linklen;
1813 	struct componentname *cnp = &ndp->ni_cnd;
1814 
1815 	*retdirp = (struct vnode *)0;
1816 	MALLOC(cnp->cn_pnbuf, char *, len + 1, M_NAMEI, M_WAITOK);
1817 	/*
1818 	 * Copy the name from the mbuf list to ndp->ni_pnbuf
1819 	 * and set the various ndp fields appropriately.
1820 	 */
1821 	fromcp = *dposp;
1822 	tocp = cnp->cn_pnbuf;
1823 	md = *mdp;
1824 	rem = mtod(md, caddr_t) + md->m_len - fromcp;
1825 	for (i = 0; i < len; i++) {
1826 		while (rem == 0) {
1827 			md = md->m_next;
1828 			if (md == NULL) {
1829 				error = EBADRPC;
1830 				goto out;
1831 			}
1832 			fromcp = mtod(md, caddr_t);
1833 			rem = md->m_len;
1834 		}
1835 		if (*fromcp == '\0' || (!pubflag && *fromcp == '/')) {
1836 			error = EACCES;
1837 			goto out;
1838 		}
1839 		*tocp++ = *fromcp++;
1840 		rem--;
1841 	}
1842 	*tocp = '\0';
1843 	*mdp = md;
1844 	*dposp = fromcp;
1845 	len = nfsm_rndup(len)-len;
1846 	if (len > 0) {
1847 		if (rem >= len)
1848 			*dposp += len;
1849 		else if ((error = nfs_adv(mdp, dposp, len, rem)) != 0)
1850 			goto out;
1851 	}
1852 
1853 	/*
1854 	 * Extract and set starting directory.
1855 	 */
1856 	error = nfsrv_fhtovp(fhp, FALSE, &dp, ndp->ni_cnd.cn_cred, slp,
1857 	    nam, &rdonly, kerbflag, pubflag);
1858 	if (error)
1859 		goto out;
1860 	if (dp->v_type != VDIR) {
1861 		vrele(dp);
1862 		error = ENOTDIR;
1863 		goto out;
1864 	}
1865 
1866 	if (rdonly)
1867 		cnp->cn_flags |= RDONLY;
1868 
1869 	*retdirp = dp;
1870 
1871 	if (pubflag) {
1872 		/*
1873 		 * Oh joy. For WebNFS, handle those pesky '%' escapes,
1874 		 * and the 'native path' indicator.
1875 		 */
1876 		MALLOC(cp, char *, MAXPATHLEN, M_NAMEI, M_WAITOK);
1877 		fromcp = cnp->cn_pnbuf;
1878 		tocp = cp;
1879 		if ((unsigned char)*fromcp >= WEBNFS_SPECCHAR_START) {
1880 			switch ((unsigned char)*fromcp) {
1881 			case WEBNFS_NATIVE_CHAR:
1882 				/*
1883 				 * 'Native' path for us is the same
1884 				 * as a path according to the NFS spec,
1885 				 * just skip the escape char.
1886 				 */
1887 				fromcp++;
1888 				break;
1889 			/*
1890 			 * More may be added in the future, range 0x80-0xff
1891 			 */
1892 			default:
1893 				error = EIO;
1894 				FREE(cp, M_NAMEI);
1895 				goto out;
1896 			}
1897 		}
1898 		/*
1899 		 * Translate the '%' escapes, URL-style.
1900 		 */
1901 		while (*fromcp != '\0') {
1902 			if (*fromcp == WEBNFS_ESC_CHAR) {
1903 				if (fromcp[1] != '\0' && fromcp[2] != '\0') {
1904 					fromcp++;
1905 					*tocp++ = HEXSTRTOI(fromcp);
1906 					fromcp += 2;
1907 					continue;
1908 				} else {
1909 					error = ENOENT;
1910 					FREE(cp, M_NAMEI);
1911 					goto out;
1912 				}
1913 			} else
1914 				*tocp++ = *fromcp++;
1915 		}
1916 		*tocp = '\0';
1917 		FREE(cnp->cn_pnbuf, M_NAMEI);
1918 		cnp->cn_pnbuf = cp;
1919 	}
1920 
1921 	ndp->ni_pathlen = (tocp - cnp->cn_pnbuf) + 1;
1922 	ndp->ni_segflg = UIO_SYSSPACE;
1923 
1924 	if (pubflag) {
1925 		ndp->ni_rootdir = rootvnode;
1926 		ndp->ni_loopcnt = 0;
1927 		if (cnp->cn_pnbuf[0] == '/')
1928 			dp = rootvnode;
1929 	} else {
1930 		cnp->cn_flags |= NOCROSSMOUNT;
1931 	}
1932 
1933 	cnp->cn_proc = p;
1934 	VREF(dp);
1935 
1936     for (;;) {
1937 	cnp->cn_nameptr = cnp->cn_pnbuf;
1938 	ndp->ni_startdir = dp;
1939 	/*
1940 	 * And call lookup() to do the real work
1941 	 */
1942 	error = lookup(ndp);
1943 	if (error)
1944 		break;
1945 	/*
1946 	 * Check for encountering a symbolic link
1947 	 */
1948 	if ((cnp->cn_flags & ISSYMLINK) == 0) {
1949 		if (cnp->cn_flags & (SAVENAME | SAVESTART)) {
1950 			cnp->cn_flags |= HASBUF;
1951 			return (0);
1952 		}
1953 		break;
1954 	} else {
1955 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
1956 			VOP_UNLOCK(ndp->ni_dvp, 0);
1957 		if (!pubflag) {
1958 			vrele(ndp->ni_dvp);
1959 			vput(ndp->ni_vp);
1960 			ndp->ni_vp = NULL;
1961 			error = EINVAL;
1962 			break;
1963 		}
1964 
1965 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
1966 			error = ELOOP;
1967 			break;
1968 		}
1969 		if (ndp->ni_pathlen > 1)
1970 			MALLOC(cp, char *, MAXPATHLEN, M_NAMEI, M_WAITOK);
1971 		else
1972 			cp = cnp->cn_pnbuf;
1973 		aiov.iov_base = cp;
1974 		aiov.iov_len = MAXPATHLEN;
1975 		auio.uio_iov = &aiov;
1976 		auio.uio_iovcnt = 1;
1977 		auio.uio_offset = 0;
1978 		auio.uio_rw = UIO_READ;
1979 		auio.uio_segflg = UIO_SYSSPACE;
1980 		auio.uio_procp = (struct proc *)0;
1981 		auio.uio_resid = MAXPATHLEN;
1982 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
1983 		if (error) {
1984 		badlink:
1985 			if (ndp->ni_pathlen > 1)
1986 				FREE(cp, M_NAMEI);
1987 			break;
1988 		}
1989 		linklen = MAXPATHLEN - auio.uio_resid;
1990 		if (linklen == 0) {
1991 			error = ENOENT;
1992 			goto badlink;
1993 		}
1994 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
1995 			error = ENAMETOOLONG;
1996 			goto badlink;
1997 		}
1998 		if (ndp->ni_pathlen > 1) {
1999 			memcpy(cp + linklen, ndp->ni_next, ndp->ni_pathlen);
2000 			FREE(cnp->cn_pnbuf, M_NAMEI);
2001 			cnp->cn_pnbuf = cp;
2002 		} else
2003 			cnp->cn_pnbuf[linklen] = '\0';
2004 		ndp->ni_pathlen += linklen;
2005 		vput(ndp->ni_vp);
2006 		dp = ndp->ni_dvp;
2007 		/*
2008 		 * Check if root directory should replace current directory.
2009 		 */
2010 		if (cnp->cn_pnbuf[0] == '/') {
2011 			vrele(dp);
2012 			dp = ndp->ni_rootdir;
2013 			VREF(dp);
2014 		}
2015 	}
2016    }
2017 out:
2018 	FREE(cnp->cn_pnbuf, M_NAMEI);
2019 	return (error);
2020 }
2021 
2022 /*
2023  * A fiddled version of m_adj() that ensures null fill to a long
2024  * boundary and only trims off the back end
2025  */
2026 void
2027 nfsm_adj(mp, len, nul)
2028 	struct mbuf *mp;
2029 	int len;
2030 	int nul;
2031 {
2032 	struct mbuf *m;
2033 	int count, i;
2034 	char *cp;
2035 
2036 	/*
2037 	 * Trim from tail.  Scan the mbuf chain,
2038 	 * calculating its length and finding the last mbuf.
2039 	 * If the adjustment only affects this mbuf, then just
2040 	 * adjust and return.  Otherwise, rescan and truncate
2041 	 * after the remaining size.
2042 	 */
2043 	count = 0;
2044 	m = mp;
2045 	for (;;) {
2046 		count += m->m_len;
2047 		if (m->m_next == (struct mbuf *)0)
2048 			break;
2049 		m = m->m_next;
2050 	}
2051 	if (m->m_len > len) {
2052 		m->m_len -= len;
2053 		if (nul > 0) {
2054 			cp = mtod(m, caddr_t)+m->m_len-nul;
2055 			for (i = 0; i < nul; i++)
2056 				*cp++ = '\0';
2057 		}
2058 		return;
2059 	}
2060 	count -= len;
2061 	if (count < 0)
2062 		count = 0;
2063 	/*
2064 	 * Correct length for chain is "count".
2065 	 * Find the mbuf with last data, adjust its length,
2066 	 * and toss data from remaining mbufs on chain.
2067 	 */
2068 	for (m = mp; m; m = m->m_next) {
2069 		if (m->m_len >= count) {
2070 			m->m_len = count;
2071 			if (nul > 0) {
2072 				cp = mtod(m, caddr_t)+m->m_len-nul;
2073 				for (i = 0; i < nul; i++)
2074 					*cp++ = '\0';
2075 			}
2076 			break;
2077 		}
2078 		count -= m->m_len;
2079 	}
2080 	for (m = m->m_next;m;m = m->m_next)
2081 		m->m_len = 0;
2082 }
2083 
2084 /*
2085  * Make these functions instead of macros, so that the kernel text size
2086  * doesn't get too big...
2087  */
2088 void
2089 nfsm_srvwcc(nfsd, before_ret, before_vap, after_ret, after_vap, mbp, bposp)
2090 	struct nfsrv_descript *nfsd;
2091 	int before_ret;
2092 	struct vattr *before_vap;
2093 	int after_ret;
2094 	struct vattr *after_vap;
2095 	struct mbuf **mbp;
2096 	char **bposp;
2097 {
2098 	struct mbuf *mb = *mbp, *mb2;
2099 	char *bpos = *bposp;
2100 	u_int32_t *tl;
2101 
2102 	if (before_ret) {
2103 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2104 		*tl = nfs_false;
2105 	} else {
2106 		nfsm_build(tl, u_int32_t *, 7 * NFSX_UNSIGNED);
2107 		*tl++ = nfs_true;
2108 		txdr_hyper(before_vap->va_size, tl);
2109 		tl += 2;
2110 		txdr_nfsv3time(&(before_vap->va_mtime), tl);
2111 		tl += 2;
2112 		txdr_nfsv3time(&(before_vap->va_ctime), tl);
2113 	}
2114 	*bposp = bpos;
2115 	*mbp = mb;
2116 	nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp);
2117 }
2118 
2119 void
2120 nfsm_srvpostopattr(nfsd, after_ret, after_vap, mbp, bposp)
2121 	struct nfsrv_descript *nfsd;
2122 	int after_ret;
2123 	struct vattr *after_vap;
2124 	struct mbuf **mbp;
2125 	char **bposp;
2126 {
2127 	struct mbuf *mb = *mbp, *mb2;
2128 	char *bpos = *bposp;
2129 	u_int32_t *tl;
2130 	struct nfs_fattr *fp;
2131 
2132 	if (after_ret) {
2133 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
2134 		*tl = nfs_false;
2135 	} else {
2136 		nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_V3FATTR);
2137 		*tl++ = nfs_true;
2138 		fp = (struct nfs_fattr *)tl;
2139 		nfsm_srvfattr(nfsd, after_vap, fp);
2140 	}
2141 	*mbp = mb;
2142 	*bposp = bpos;
2143 }
2144 
2145 void
2146 nfsm_srvfattr(nfsd, vap, fp)
2147 	struct nfsrv_descript *nfsd;
2148 	struct vattr *vap;
2149 	struct nfs_fattr *fp;
2150 {
2151 
2152 	fp->fa_nlink = txdr_unsigned(vap->va_nlink);
2153 	fp->fa_uid = txdr_unsigned(vap->va_uid);
2154 	fp->fa_gid = txdr_unsigned(vap->va_gid);
2155 	if (nfsd->nd_flag & ND_NFSV3) {
2156 		fp->fa_type = vtonfsv3_type(vap->va_type);
2157 		fp->fa_mode = vtonfsv3_mode(vap->va_mode);
2158 		txdr_hyper(vap->va_size, &fp->fa3_size);
2159 		txdr_hyper(vap->va_bytes, &fp->fa3_used);
2160 		fp->fa3_rdev.specdata1 = txdr_unsigned(major(vap->va_rdev));
2161 		fp->fa3_rdev.specdata2 = txdr_unsigned(minor(vap->va_rdev));
2162 		fp->fa3_fsid.nfsuquad[0] = 0;
2163 		fp->fa3_fsid.nfsuquad[1] = txdr_unsigned(vap->va_fsid);
2164 		fp->fa3_fileid.nfsuquad[0] = 0;
2165 		fp->fa3_fileid.nfsuquad[1] = txdr_unsigned(vap->va_fileid);
2166 		txdr_nfsv3time(&vap->va_atime, &fp->fa3_atime);
2167 		txdr_nfsv3time(&vap->va_mtime, &fp->fa3_mtime);
2168 		txdr_nfsv3time(&vap->va_ctime, &fp->fa3_ctime);
2169 	} else {
2170 		fp->fa_type = vtonfsv2_type(vap->va_type);
2171 		fp->fa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
2172 		fp->fa2_size = txdr_unsigned(vap->va_size);
2173 		fp->fa2_blocksize = txdr_unsigned(vap->va_blocksize);
2174 		if (vap->va_type == VFIFO)
2175 			fp->fa2_rdev = 0xffffffff;
2176 		else
2177 			fp->fa2_rdev = txdr_unsigned(vap->va_rdev);
2178 		fp->fa2_blocks = txdr_unsigned(vap->va_bytes / NFS_FABLKSIZE);
2179 		fp->fa2_fsid = txdr_unsigned(vap->va_fsid);
2180 		fp->fa2_fileid = txdr_unsigned(vap->va_fileid);
2181 		txdr_nfsv2time(&vap->va_atime, &fp->fa2_atime);
2182 		txdr_nfsv2time(&vap->va_mtime, &fp->fa2_mtime);
2183 		txdr_nfsv2time(&vap->va_ctime, &fp->fa2_ctime);
2184 	}
2185 }
2186 
2187 /*
2188  * nfsrv_fhtovp() - convert a fh to a vnode ptr (optionally locked)
2189  * 	- look up fsid in mount list (if not found ret error)
2190  *	- get vp and export rights by calling VFS_FHTOVP()
2191  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2192  *	- if not lockflag unlock it with VOP_UNLOCK()
2193  */
2194 int
2195 nfsrv_fhtovp(fhp, lockflag, vpp, cred, slp, nam, rdonlyp, kerbflag, pubflag)
2196 	fhandle_t *fhp;
2197 	int lockflag;
2198 	struct vnode **vpp;
2199 	struct ucred *cred;
2200 	struct nfssvc_sock *slp;
2201 	struct mbuf *nam;
2202 	int *rdonlyp;
2203 	int kerbflag;
2204 {
2205 	struct mount *mp;
2206 	int i;
2207 	struct ucred *credanon;
2208 	int error, exflags;
2209 	struct sockaddr_in *saddr;
2210 
2211 	*vpp = (struct vnode *)0;
2212 
2213 	if (nfs_ispublicfh(fhp)) {
2214 		if (!pubflag || !nfs_pub.np_valid)
2215 			return (ESTALE);
2216 		fhp = &nfs_pub.np_handle;
2217 	}
2218 
2219 	mp = vfs_getvfs(&fhp->fh_fsid);
2220 	if (!mp)
2221 		return (ESTALE);
2222 	error = VFS_CHECKEXP(mp, nam, &exflags, &credanon);
2223 	if (error)
2224 		return (error);
2225 	error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
2226 	if (error)
2227 		return (error);
2228 
2229 	if (!(exflags & (MNT_EXNORESPORT|MNT_EXPUBLIC))) {
2230 		saddr = mtod(nam, struct sockaddr_in *);
2231 		if (saddr->sin_family == AF_INET &&
2232 		    ntohs(saddr->sin_port) >= IPPORT_RESERVED) {
2233 			vput(*vpp);
2234 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2235 		}
2236 	}
2237 	/*
2238 	 * Check/setup credentials.
2239 	 */
2240 	if (exflags & MNT_EXKERB) {
2241 		if (!kerbflag) {
2242 			vput(*vpp);
2243 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2244 		}
2245 	} else if (kerbflag) {
2246 		vput(*vpp);
2247 		return (NFSERR_AUTHERR | AUTH_TOOWEAK);
2248 	} else if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) {
2249 		cred->cr_uid = credanon->cr_uid;
2250 		cred->cr_gid = credanon->cr_gid;
2251 		for (i = 0; i < credanon->cr_ngroups && i < NGROUPS; i++)
2252 			cred->cr_groups[i] = credanon->cr_groups[i];
2253 		cred->cr_ngroups = i;
2254 	}
2255 	if (exflags & MNT_EXRDONLY)
2256 		*rdonlyp = 1;
2257 	else
2258 		*rdonlyp = 0;
2259 	if (!lockflag)
2260 		VOP_UNLOCK(*vpp, 0);
2261 	return (0);
2262 }
2263 
2264 /*
2265  * WebNFS: check if a filehandle is a public filehandle. For v3, this
2266  * means a length of 0, for v2 it means all zeroes. nfsm_srvmtofh has
2267  * transformed this to all zeroes in both cases, so check for it.
2268  */
2269 int
2270 nfs_ispublicfh(fhp)
2271 	fhandle_t *fhp;
2272 {
2273 	char *cp = (char *)fhp;
2274 	int i;
2275 
2276 	for (i = 0; i < NFSX_V3FH; i++)
2277 		if (*cp++ != 0)
2278 			return (FALSE);
2279 	return (TRUE);
2280 }
2281 
2282 /*
2283  * This function compares two net addresses by family and returns TRUE
2284  * if they are the same host.
2285  * If there is any doubt, return FALSE.
2286  * The AF_INET family is handled as a special case so that address mbufs
2287  * don't need to be saved to store "struct in_addr", which is only 4 bytes.
2288  */
2289 int
2290 netaddr_match(family, haddr, nam)
2291 	int family;
2292 	union nethostaddr *haddr;
2293 	struct mbuf *nam;
2294 {
2295 	struct sockaddr_in *inetaddr;
2296 
2297 	switch (family) {
2298 	case AF_INET:
2299 		inetaddr = mtod(nam, struct sockaddr_in *);
2300 		if (inetaddr->sin_family == AF_INET &&
2301 		    inetaddr->sin_addr.s_addr == haddr->had_inetaddr)
2302 			return (1);
2303 		break;
2304 #ifdef ISO
2305 	case AF_ISO:
2306 	    {
2307 		struct sockaddr_iso *isoaddr1, *isoaddr2;
2308 
2309 		isoaddr1 = mtod(nam, struct sockaddr_iso *);
2310 		isoaddr2 = mtod(haddr->had_nam, struct sockaddr_iso *);
2311 		if (isoaddr1->siso_family == AF_ISO &&
2312 		    isoaddr1->siso_nlen > 0 &&
2313 		    isoaddr1->siso_nlen == isoaddr2->siso_nlen &&
2314 		    SAME_ISOADDR(isoaddr1, isoaddr2))
2315 			return (1);
2316 		break;
2317 	    }
2318 #endif	/* ISO */
2319 	default:
2320 		break;
2321 	};
2322 	return (0);
2323 }
2324 
2325 
2326 /*
2327  * The write verifier has changed (probably due to a server reboot), so all
2328  * B_NEEDCOMMIT blocks will have to be written again. Since they are on the
2329  * dirty block list as B_DELWRI, all this takes is clearing the B_NEEDCOMMIT
2330  * flag. Once done the new write verifier can be set for the mount point.
2331  */
2332 void
2333 nfs_clearcommit(mp)
2334 	struct mount *mp;
2335 {
2336 	struct vnode *vp, *nvp;
2337 	struct buf *bp, *nbp;
2338 	int s;
2339 
2340 	s = splbio();
2341 loop:
2342 	for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
2343 		if (vp->v_mount != mp)	/* Paranoia */
2344 			goto loop;
2345 		nvp = vp->v_mntvnodes.le_next;
2346 		for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = nbp) {
2347 			nbp = bp->b_vnbufs.le_next;
2348 			if ((bp->b_flags & (B_BUSY | B_DELWRI | B_NEEDCOMMIT))
2349 				== (B_DELWRI | B_NEEDCOMMIT))
2350 				bp->b_flags &= ~B_NEEDCOMMIT;
2351 		}
2352 	}
2353 	splx(s);
2354 }
2355 
2356 /*
2357  * Map errnos to NFS error numbers. For Version 3 also filter out error
2358  * numbers not specified for the associated procedure.
2359  */
2360 int
2361 nfsrv_errmap(nd, err)
2362 	struct nfsrv_descript *nd;
2363 	int err;
2364 {
2365 	short *defaulterrp, *errp;
2366 
2367 	if (nd->nd_flag & ND_NFSV3) {
2368 	    if (nd->nd_procnum <= NFSPROC_COMMIT) {
2369 		errp = defaulterrp = nfsrv_v3errmap[nd->nd_procnum];
2370 		while (*++errp) {
2371 			if (*errp == err)
2372 				return (err);
2373 			else if (*errp > err)
2374 				break;
2375 		}
2376 		return ((int)*defaulterrp);
2377 	    } else
2378 		return (err & 0xffff);
2379 	}
2380 	if (err <= ELAST)
2381 		return ((int)nfsrv_v2errmap[err - 1]);
2382 	return (NFSERR_IO);
2383 }
2384 
2385 /*
2386  * Sort the group list in increasing numerical order.
2387  * (Insertion sort by Chris Torek, who was grossed out by the bubble sort
2388  *  that used to be here.)
2389  */
2390 void
2391 nfsrvw_sort(list, num)
2392         gid_t *list;
2393         int num;
2394 {
2395 	int i, j;
2396 	gid_t v;
2397 
2398 	/* Insertion sort. */
2399 	for (i = 1; i < num; i++) {
2400 		v = list[i];
2401 		/* find correct slot for value v, moving others up */
2402 		for (j = i; --j >= 0 && v < list[j];)
2403 			list[j + 1] = list[j];
2404 		list[j + 1] = v;
2405 	}
2406 }
2407 
2408 /*
2409  * copy credentials making sure that the result can be compared with memcmp().
2410  */
2411 void
2412 nfsrv_setcred(incred, outcred)
2413 	struct ucred *incred, *outcred;
2414 {
2415 	int i;
2416 
2417 	memset((caddr_t)outcred, 0, sizeof (struct ucred));
2418 	outcred->cr_ref = 1;
2419 	outcred->cr_uid = incred->cr_uid;
2420 	outcred->cr_gid = incred->cr_gid;
2421 	outcred->cr_ngroups = incred->cr_ngroups;
2422 	for (i = 0; i < incred->cr_ngroups; i++)
2423 		outcred->cr_groups[i] = incred->cr_groups[i];
2424 	nfsrvw_sort(outcred->cr_groups, outcred->cr_ngroups);
2425 }
2426