xref: /dflybsd-src/sys/vfs/nfs/nfs_serv.c (revision 3cf94ecf50c1958c7b10de185222fd2816ccfff0)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)nfs_serv.c  8.8 (Berkeley) 7/31/95
37  * $FreeBSD: src/sys/nfs/nfs_serv.c,v 1.93.2.6 2002/12/29 18:19:53 dillon Exp $
38  * $DragonFly: src/sys/vfs/nfs/nfs_serv.c,v 1.23 2005/05/29 10:08:36 hsu Exp $
39  */
40 
41 /*
42  * nfs version 2 and 3 server calls to vnode ops
43  * - these routines generally have 3 phases
44  *   1 - break down and validate rpc request in mbuf list
45  *   2 - do the vnode ops for the request
46  *       (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
47  *   3 - build the rpc reply in an mbuf list
48  *   nb:
49  *	- do not mix the phases, since the nfsm_?? macros can return failures
50  *	  on a bad rpc or similar and do not do any vrele() or vput()'s
51  *
52  *      - the nfsm_reply() macro generates an nfs rpc reply with the nfs
53  *	error number iff error != 0 whereas
54  *	returning an error from the server function implies a fatal error
55  *	such as a badly constructed rpc request that should be dropped without
56  *	a reply.
57  *	For Version 3, nfsm_reply() does not return for the error case, since
58  *	most version 3 rpcs return more than the status for error cases.
59  *
60  * Other notes:
61  *	Warning: always pay careful attention to resource cleanup on return
62  *	and note that nfsm_*() macros can terminate a procedure on certain
63  *	errors.
64  */
65 
66 #include <sys/param.h>
67 #include <sys/systm.h>
68 #include <sys/proc.h>
69 #include <sys/nlookup.h>
70 #include <sys/namei.h>
71 #include <sys/unistd.h>
72 #include <sys/vnode.h>
73 #include <sys/mount.h>
74 #include <sys/socket.h>
75 #include <sys/socketvar.h>
76 #include <sys/malloc.h>
77 #include <sys/mbuf.h>
78 #include <sys/dirent.h>
79 #include <sys/stat.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
82 #include <sys/buf.h>
83 
84 #include <vm/vm.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_zone.h>
87 #include <vm/vm_object.h>
88 
89 #include <sys/buf2.h>
90 
91 #include "nfsproto.h"
92 #include "rpcv2.h"
93 #include "nfs.h"
94 #include "xdr_subs.h"
95 #include "nfsm_subs.h"
96 #include "nqnfs.h"
97 
98 #ifdef NFSRV_DEBUG
99 #define nfsdbprintf(info)	printf info
100 #else
101 #define nfsdbprintf(info)
102 #endif
103 
104 #define MAX_COMMIT_COUNT	(1024 * 1024)
105 
106 #define NUM_HEURISTIC		1017
107 #define NHUSE_INIT		64
108 #define NHUSE_INC		16
109 #define NHUSE_MAX		2048
110 
111 static struct nfsheur {
112     struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
113     off_t nh_nextr;		/* next offset for sequential detection */
114     int nh_use;			/* use count for selection */
115     int nh_seqcount;		/* heuristic */
116 } nfsheur[NUM_HEURISTIC];
117 
118 nfstype nfsv3_type[9] = { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK,
119 		      NFFIFO, NFNON };
120 #ifndef NFS_NOSERVER
121 nfstype nfsv2_type[9] = { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON,
122 		      NFCHR, NFNON };
123 /* Global vars */
124 extern u_int32_t nfs_xdrneg1;
125 extern u_int32_t nfs_false, nfs_true;
126 extern enum vtype nv3tov_type[8];
127 extern struct nfsstats nfsstats;
128 
129 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
130 int nfsrvw_procrastinate_v3 = 0;
131 
132 static struct timespec	nfsver;
133 
134 SYSCTL_DECL(_vfs_nfs);
135 
136 static int nfs_async;
137 SYSCTL_INT(_vfs_nfs, OID_AUTO, async, CTLFLAG_RW, &nfs_async, 0, "");
138 static int nfs_commit_blks;
139 static int nfs_commit_miss;
140 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0, "");
141 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, "");
142 
143 static int nfsrv_access (struct vnode *,int,struct ucred *,int,
144 		struct thread *, int);
145 static void nfsrvw_coalesce (struct nfsrv_descript *,
146 		struct nfsrv_descript *);
147 
148 /*
149  * nfs v3 access service
150  */
151 int
152 nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
153 	      struct thread *td, struct mbuf **mrq)
154 {
155 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
156 	struct sockaddr *nam = nfsd->nd_nam;
157 	caddr_t dpos = nfsd->nd_dpos;
158 	struct ucred *cred = &nfsd->nd_cr;
159 	struct vnode *vp = NULL;
160 	nfsfh_t nfh;
161 	fhandle_t *fhp;
162 	u_int32_t *tl;
163 	int32_t t1;
164 	caddr_t bpos;
165 	int error = 0, rdonly, cache, getret;
166 	char *cp2;
167 	struct mbuf *mb, *mreq, *mb2;
168 	struct vattr vattr, *vap = &vattr;
169 	u_long testmode, nfsmode;
170 	u_quad_t frev;
171 
172 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
173 #ifndef nolint
174 	cache = 0;
175 #endif
176 	fhp = &nfh.fh_generic;
177 	nfsm_srvmtofh(fhp);
178 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
179 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
180 	    (nfsd->nd_flag & ND_KERBAUTH), TRUE);
181 	if (error) {
182 		nfsm_reply(NFSX_UNSIGNED);
183 		nfsm_srvpostop_attr(1, (struct vattr *)0);
184 		error = 0;
185 		goto nfsmout;
186 	}
187 	nfsmode = fxdr_unsigned(u_int32_t, *tl);
188 	if ((nfsmode & NFSV3ACCESS_READ) &&
189 		nfsrv_access(vp, VREAD, cred, rdonly, td, 0))
190 		nfsmode &= ~NFSV3ACCESS_READ;
191 	if (vp->v_type == VDIR)
192 		testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
193 			NFSV3ACCESS_DELETE);
194 	else
195 		testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
196 	if ((nfsmode & testmode) &&
197 		nfsrv_access(vp, VWRITE, cred, rdonly, td, 0))
198 		nfsmode &= ~testmode;
199 	if (vp->v_type == VDIR)
200 		testmode = NFSV3ACCESS_LOOKUP;
201 	else
202 		testmode = NFSV3ACCESS_EXECUTE;
203 	if ((nfsmode & testmode) &&
204 		nfsrv_access(vp, VEXEC, cred, rdonly, td, 0))
205 		nfsmode &= ~testmode;
206 	getret = VOP_GETATTR(vp, vap, td);
207 	vput(vp);
208 	vp = NULL;
209 	nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
210 	nfsm_srvpostop_attr(getret, vap);
211 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
212 	*tl = txdr_unsigned(nfsmode);
213 nfsmout:
214 	if (vp)
215 		vput(vp);
216 	return(error);
217 }
218 
219 /*
220  * nfs getattr service
221  */
222 int
223 nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
224 	      struct thread *td, struct mbuf **mrq)
225 {
226 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
227 	struct sockaddr *nam = nfsd->nd_nam;
228 	caddr_t dpos = nfsd->nd_dpos;
229 	struct ucred *cred = &nfsd->nd_cr;
230 	struct nfs_fattr *fp;
231 	struct vattr va;
232 	struct vattr *vap = &va;
233 	struct vnode *vp = NULL;
234 	nfsfh_t nfh;
235 	fhandle_t *fhp;
236 	u_int32_t *tl;
237 	int32_t t1;
238 	caddr_t bpos;
239 	int error = 0, rdonly, cache;
240 	char *cp2;
241 	struct mbuf *mb, *mb2, *mreq;
242 	u_quad_t frev;
243 
244 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
245 	fhp = &nfh.fh_generic;
246 	nfsm_srvmtofh(fhp);
247 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
248 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
249 	if (error) {
250 		nfsm_reply(0);
251 		error = 0;
252 		goto nfsmout;
253 	}
254 	nqsrv_getl(vp, ND_READ);
255 	error = VOP_GETATTR(vp, vap, td);
256 	vput(vp);
257 	vp = NULL;
258 	nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
259 	if (error) {
260 		error = 0;
261 		goto nfsmout;
262 	}
263 	nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
264 	nfsm_srvfillattr(vap, fp);
265 	/* fall through */
266 
267 nfsmout:
268 	if (vp)
269 		vput(vp);
270 	return(error);
271 }
272 
273 /*
274  * nfs setattr service
275  */
276 int
277 nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
278 	      struct thread *td, struct mbuf **mrq)
279 {
280 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
281 	struct sockaddr *nam = nfsd->nd_nam;
282 	caddr_t dpos = nfsd->nd_dpos;
283 	struct ucred *cred = &nfsd->nd_cr;
284 	struct vattr va, preat;
285 	struct vattr *vap = &va;
286 	struct nfsv2_sattr *sp;
287 	struct nfs_fattr *fp;
288 	struct vnode *vp = NULL;
289 	nfsfh_t nfh;
290 	fhandle_t *fhp;
291 	u_int32_t *tl;
292 	int32_t t1;
293 	caddr_t bpos;
294 	int error = 0, rdonly, cache, preat_ret = 1, postat_ret = 1;
295 	int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
296 	char *cp2;
297 	struct mbuf *mb, *mb2, *mreq;
298 	u_quad_t frev;
299 	struct timespec guard;
300 
301 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
302 	fhp = &nfh.fh_generic;
303 	nfsm_srvmtofh(fhp);
304 	VATTR_NULL(vap);
305 	if (v3) {
306 		nfsm_srvsattr(vap);
307 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
308 		gcheck = fxdr_unsigned(int, *tl);
309 		if (gcheck) {
310 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
311 			fxdr_nfsv3time(tl, &guard);
312 		}
313 	} else {
314 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
315 		/*
316 		 * Nah nah nah nah na nah
317 		 * There is a bug in the Sun client that puts 0xffff in the mode
318 		 * field of sattr when it should put in 0xffffffff. The u_short
319 		 * doesn't sign extend.
320 		 * --> check the low order 2 bytes for 0xffff
321 		 */
322 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
323 			vap->va_mode = nfstov_mode(sp->sa_mode);
324 		if (sp->sa_uid != nfs_xdrneg1)
325 			vap->va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
326 		if (sp->sa_gid != nfs_xdrneg1)
327 			vap->va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
328 		if (sp->sa_size != nfs_xdrneg1)
329 			vap->va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
330 		if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) {
331 #ifdef notyet
332 			fxdr_nfsv2time(&sp->sa_atime, &vap->va_atime);
333 #else
334 			vap->va_atime.tv_sec =
335 				fxdr_unsigned(int32_t, sp->sa_atime.nfsv2_sec);
336 			vap->va_atime.tv_nsec = 0;
337 #endif
338 		}
339 		if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1)
340 			fxdr_nfsv2time(&sp->sa_mtime, &vap->va_mtime);
341 
342 	}
343 
344 	/*
345 	 * Now that we have all the fields, lets do it.
346 	 */
347 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
348 		(nfsd->nd_flag & ND_KERBAUTH), TRUE);
349 	if (error) {
350 		nfsm_reply(2 * NFSX_UNSIGNED);
351 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
352 		error = 0;
353 		goto nfsmout;
354 	}
355 
356 	/*
357 	 * vp now an active resource, pay careful attention to cleanup
358 	 */
359 
360 	nqsrv_getl(vp, ND_WRITE);
361 	if (v3) {
362 		error = preat_ret = VOP_GETATTR(vp, &preat, td);
363 		if (!error && gcheck &&
364 			(preat.va_ctime.tv_sec != guard.tv_sec ||
365 			 preat.va_ctime.tv_nsec != guard.tv_nsec))
366 			error = NFSERR_NOT_SYNC;
367 		if (error) {
368 			vput(vp);
369 			vp = NULL;
370 			nfsm_reply(NFSX_WCCDATA(v3));
371 			nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
372 			error = 0;
373 			goto nfsmout;
374 		}
375 	}
376 
377 	/*
378 	 * If the size is being changed write acces is required, otherwise
379 	 * just check for a read only file system.
380 	 */
381 	if (vap->va_size == ((u_quad_t)((quad_t) -1))) {
382 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
383 			error = EROFS;
384 			goto out;
385 		}
386 	} else {
387 		if (vp->v_type == VDIR) {
388 			error = EISDIR;
389 			goto out;
390 		} else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
391 			    td, 0)) != 0){
392 			goto out;
393 		}
394 	}
395 	error = VOP_SETATTR(vp, vap, cred, td);
396 	postat_ret = VOP_GETATTR(vp, vap, td);
397 	if (!error)
398 		error = postat_ret;
399 out:
400 	vput(vp);
401 	vp = NULL;
402 	nfsm_reply(NFSX_WCCORFATTR(v3));
403 	if (v3) {
404 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
405 		error = 0;
406 		goto nfsmout;
407 	} else {
408 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
409 		nfsm_srvfillattr(vap, fp);
410 	}
411 	/* fall through */
412 
413 nfsmout:
414 	if (vp)
415 		vput(vp);
416 	return(error);
417 }
418 
419 /*
420  * nfs lookup rpc
421  */
422 int
423 nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
424 	     struct thread *td, struct mbuf **mrq)
425 {
426 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
427 	struct sockaddr *nam = nfsd->nd_nam;
428 	caddr_t dpos = nfsd->nd_dpos;
429 	struct ucred *cred = &nfsd->nd_cr;
430 	struct nfs_fattr *fp;
431 	struct nlookupdata nd;
432 	struct vnode *vp;
433 	struct vnode *dirp;
434 	struct namecache *ncp;
435 	nfsfh_t nfh;
436 	fhandle_t *fhp;
437 	caddr_t cp;
438 	u_int32_t *tl;
439 	int32_t t1;
440 	caddr_t bpos;
441 	int error = 0, cache = 0, len, dirattr_ret = 1;
442 	int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
443 	char *cp2;
444 	struct mbuf *mb, *mb2, *mreq;
445 	struct vattr va, dirattr, *vap = &va;
446 	u_quad_t frev;
447 
448 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
449 	nlookup_zero(&nd);
450 	dirp = NULL;
451 	vp = NULL;
452 
453 	fhp = &nfh.fh_generic;
454 	nfsm_srvmtofh(fhp);
455 	nfsm_srvnamesiz(len);
456 
457 	pubflag = nfs_ispublicfh(fhp);
458 
459 	error = nfs_namei(&nd, cred, NAMEI_LOOKUP, NULL, &vp,
460 		fhp, len, slp, nam, &md, &dpos,
461 		&dirp, td, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
462 
463 	/*
464 	 * namei failure, only dirp to cleanup.  Clear out garbarge from
465 	 * structure in case macros jump to nfsmout.
466 	 */
467 
468 	if (error) {
469 		if (dirp) {
470 			if (v3)
471 				dirattr_ret = VOP_GETATTR(dirp, &dirattr, td);
472 			vrele(dirp);
473 			dirp = NULL;
474 		}
475 		nfsm_reply(NFSX_POSTOPATTR(v3));
476 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
477 		error = 0;
478 		goto nfsmout;
479 	}
480 
481 	/*
482 	 * Locate index file for public filehandle
483 	 *
484 	 * error is 0 on entry and 0 on exit from this block.
485 	 */
486 
487 	if (pubflag) {
488 		if (vp->v_type == VDIR && nfs_pub.np_index != NULL) {
489 			/*
490 			 * Setup call to lookup() to see if we can find
491 			 * the index file. Arguably, this doesn't belong
492 			 * in a kernel.. Ugh.  If an error occurs, do not
493 			 * try to install an index file and then clear the
494 			 * error.
495 			 *
496 			 * When we replace nd with ind and redirect ndp,
497 			 * maintenance of ni_startdir and ni_vp shift to
498 			 * ind and we have to clean them up in the old nd.
499 			 * However, the cnd resource continues to be maintained
500 			 * via the original nd.  Confused?  You aren't alone!
501 			 */
502 			VOP_UNLOCK(vp, 0, td);
503 			ncp = cache_hold(nd.nl_ncp);
504 			nlookup_done(&nd);
505 			error = nlookup_init_raw(&nd, nfs_pub.np_index,
506 						UIO_SYSSPACE, 0, cred, ncp);
507 			cache_drop(ncp);
508 			if (error == 0)
509 				error = nlookup(&nd);
510 
511 			if (error == 0) {
512 				/*
513 				 * Found an index file. Get rid of
514 				 * the old references.  transfer vp and
515 				 * load up the new vp.  Fortunately we do
516 				 * not have to deal with dvp, that would be
517 				 * a huge mess.
518 				 */
519 				if (dirp)
520 					vrele(dirp);
521 				dirp = vp;
522 				vp = NULL;
523 				error = cache_vget(nd.nl_ncp, nd.nl_cred,
524 							LK_EXCLUSIVE, &vp);
525 				KKASSERT(error == 0);
526 			}
527 			error = 0;
528 		}
529 		/*
530 		 * If the public filehandle was used, check that this lookup
531 		 * didn't result in a filehandle outside the publicly exported
532 		 * filesystem.  We clear the poor vp here to avoid lockups due
533 		 * to NFS I/O.
534 		 */
535 
536 		if (vp->v_mount != nfs_pub.np_mount) {
537 			vput(vp);
538 			vp = NULL;
539 			error = EPERM;
540 		}
541 	}
542 
543 	if (dirp) {
544 		if (v3)
545 			dirattr_ret = VOP_GETATTR(dirp, &dirattr, td);
546 		vrele(dirp);
547 		dirp = NULL;
548 	}
549 
550 	/*
551 	 * Resources at this point:
552 	 *	ndp->ni_vp	may not be NULL
553 	 *
554 	 */
555 
556 	if (error) {
557 		nfsm_reply(NFSX_POSTOPATTR(v3));
558 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
559 		error = 0;
560 		goto nfsmout;
561 	}
562 
563 #if 0
564 	/* XXX not sure how to deal with this */
565 	nqsrv_getl(ndp->ni_startdir, ND_READ);
566 #endif
567 
568 	/*
569 	 * Clear out some resources prior to potentially blocking.  This
570 	 * is not as critical as ni_dvp resources in other routines, but
571 	 * it helps.
572 	 */
573 	nlookup_done(&nd);
574 
575 	/*
576 	 * Get underlying attribute, then release remaining resources ( for
577 	 * the same potential blocking reason ) and reply.
578 	 */
579 	bzero((caddr_t)fhp, sizeof(nfh));
580 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
581 	error = VFS_VPTOFH(vp, &fhp->fh_fid);
582 	if (!error)
583 		error = VOP_GETATTR(vp, vap, td);
584 
585 	vput(vp);
586 	vp = NULL;
587 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
588 	if (error) {
589 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
590 		error = 0;
591 		goto nfsmout;
592 	}
593 	nfsm_srvfhtom(fhp, v3);
594 	if (v3) {
595 		nfsm_srvpostop_attr(0, vap);
596 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
597 	} else {
598 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
599 		nfsm_srvfillattr(vap, fp);
600 	}
601 
602 nfsmout:
603 	if (dirp)
604 		vrele(dirp);
605 	nlookup_done(&nd);		/* may be called twice */
606 	if (vp)
607 		vput(vp);
608 	return (error);
609 }
610 
611 /*
612  * nfs readlink service
613  */
614 int
615 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
616 	       struct thread *td, struct mbuf **mrq)
617 {
618 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
619 	struct sockaddr *nam = nfsd->nd_nam;
620 	caddr_t dpos = nfsd->nd_dpos;
621 	struct ucred *cred = &nfsd->nd_cr;
622 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
623 	struct iovec *ivp = iv;
624 	struct mbuf *mp;
625 	u_int32_t *tl;
626 	int32_t t1;
627 	caddr_t bpos;
628 	int error = 0, rdonly, cache, i, tlen, len, getret;
629 	int v3 = (nfsd->nd_flag & ND_NFSV3);
630 	char *cp2;
631 	struct mbuf *mb, *mb2, *mp2, *mp3, *mreq;
632 	struct vnode *vp = NULL;
633 	struct vattr attr;
634 	nfsfh_t nfh;
635 	fhandle_t *fhp;
636 	struct uio io, *uiop = &io;
637 	u_quad_t frev;
638 
639 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
640 #ifndef nolint
641 	mp2 = (struct mbuf *)0;
642 #endif
643 	mp3 = NULL;
644 	fhp = &nfh.fh_generic;
645 	nfsm_srvmtofh(fhp);
646 	len = 0;
647 	i = 0;
648 	while (len < NFS_MAXPATHLEN) {
649 		mp = m_getcl(MB_WAIT, MT_DATA, 0);
650 		mp->m_len = MCLBYTES;
651 		if (len == 0)
652 			mp3 = mp2 = mp;
653 		else {
654 			mp2->m_next = mp;
655 			mp2 = mp;
656 		}
657 		if ((len+mp->m_len) > NFS_MAXPATHLEN) {
658 			mp->m_len = NFS_MAXPATHLEN-len;
659 			len = NFS_MAXPATHLEN;
660 		} else
661 			len += mp->m_len;
662 		ivp->iov_base = mtod(mp, caddr_t);
663 		ivp->iov_len = mp->m_len;
664 		i++;
665 		ivp++;
666 	}
667 	uiop->uio_iov = iv;
668 	uiop->uio_iovcnt = i;
669 	uiop->uio_offset = 0;
670 	uiop->uio_resid = len;
671 	uiop->uio_rw = UIO_READ;
672 	uiop->uio_segflg = UIO_SYSSPACE;
673 	uiop->uio_td = NULL;
674 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
675 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
676 	if (error) {
677 		nfsm_reply(2 * NFSX_UNSIGNED);
678 		nfsm_srvpostop_attr(1, (struct vattr *)0);
679 		error = 0;
680 		goto nfsmout;
681 	}
682 	if (vp->v_type != VLNK) {
683 		if (v3)
684 			error = EINVAL;
685 		else
686 			error = ENXIO;
687 		goto out;
688 	}
689 	nqsrv_getl(vp, ND_READ);
690 	error = VOP_READLINK(vp, uiop, cred);
691 out:
692 	getret = VOP_GETATTR(vp, &attr, td);
693 	vput(vp);
694 	vp = NULL;
695 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
696 	if (v3) {
697 		nfsm_srvpostop_attr(getret, &attr);
698 		if (error) {
699 			error = 0;
700 			goto nfsmout;
701 		}
702 	}
703 	if (uiop->uio_resid > 0) {
704 		len -= uiop->uio_resid;
705 		tlen = nfsm_rndup(len);
706 		nfsm_adj(mp3, NFS_MAXPATHLEN-tlen, tlen-len);
707 	}
708 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
709 	*tl = txdr_unsigned(len);
710 	mb->m_next = mp3;
711 	mp3 = NULL;
712 nfsmout:
713 	if (mp3)
714 		m_freem(mp3);
715 	if (vp)
716 		vput(vp);
717 	return(error);
718 }
719 
720 /*
721  * nfs read service
722  */
723 int
724 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
725 	   struct thread *td, struct mbuf **mrq)
726 {
727 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
728 	struct sockaddr *nam = nfsd->nd_nam;
729 	caddr_t dpos = nfsd->nd_dpos;
730 	struct ucred *cred = &nfsd->nd_cr;
731 	struct iovec *iv;
732 	struct iovec *iv2;
733 	struct mbuf *m;
734 	struct nfs_fattr *fp;
735 	u_int32_t *tl;
736 	int32_t t1;
737 	int i;
738 	caddr_t bpos;
739 	int error = 0, rdonly, cache, cnt, len, left, siz, tlen, getret;
740 	int v3 = (nfsd->nd_flag & ND_NFSV3), reqlen;
741 	char *cp2;
742 	struct mbuf *mb, *mb2, *mreq;
743 	struct mbuf *m2;
744 	struct vnode *vp = NULL;
745 	nfsfh_t nfh;
746 	fhandle_t *fhp;
747 	struct uio io, *uiop = &io;
748 	struct vattr va, *vap = &va;
749 	struct nfsheur *nh;
750 	off_t off;
751 	int ioflag = 0;
752 	u_quad_t frev;
753 
754 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
755 	fhp = &nfh.fh_generic;
756 	nfsm_srvmtofh(fhp);
757 	if (v3) {
758 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
759 		off = fxdr_hyper(tl);
760 	} else {
761 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
762 		off = (off_t)fxdr_unsigned(u_int32_t, *tl);
763 	}
764 	nfsm_srvstrsiz(reqlen, NFS_SRVMAXDATA(nfsd));
765 
766 	/*
767 	 * Reference vp.  If an error occurs, vp will be invalid, but we
768 	 * have to NULL it just in case.  The macros might goto nfsmout
769 	 * as well.
770 	 */
771 
772 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
773 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
774 	if (error) {
775 		vp = NULL;
776 		nfsm_reply(2 * NFSX_UNSIGNED);
777 		nfsm_srvpostop_attr(1, (struct vattr *)0);
778 		error = 0;
779 		goto nfsmout;
780 	}
781 
782 	if (vp->v_type != VREG) {
783 		if (v3)
784 			error = EINVAL;
785 		else
786 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
787 	}
788 	if (!error) {
789 	    nqsrv_getl(vp, ND_READ);
790 	    if ((error = nfsrv_access(vp, VREAD, cred, rdonly, td, 1)) != 0)
791 		error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 1);
792 	}
793 	getret = VOP_GETATTR(vp, vap, td);
794 	if (!error)
795 		error = getret;
796 	if (error) {
797 		vput(vp);
798 		vp = NULL;
799 		nfsm_reply(NFSX_POSTOPATTR(v3));
800 		nfsm_srvpostop_attr(getret, vap);
801 		error = 0;
802 		goto nfsmout;
803 	}
804 
805 	/*
806 	 * Calculate byte count to read
807 	 */
808 
809 	if (off >= vap->va_size)
810 		cnt = 0;
811 	else if ((off + reqlen) > vap->va_size)
812 		cnt = vap->va_size - off;
813 	else
814 		cnt = reqlen;
815 
816 	/*
817 	 * Calculate seqcount for heuristic
818 	 */
819 
820 	{
821 		int hi;
822 		int try = 32;
823 
824 		/*
825 		 * Locate best candidate
826 		 */
827 
828 		hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
829 		nh = &nfsheur[hi];
830 
831 		while (try--) {
832 			if (nfsheur[hi].nh_vp == vp) {
833 				nh = &nfsheur[hi];
834 				break;
835 			}
836 			if (nfsheur[hi].nh_use > 0)
837 				--nfsheur[hi].nh_use;
838 			hi = (hi + 1) % NUM_HEURISTIC;
839 			if (nfsheur[hi].nh_use < nh->nh_use)
840 				nh = &nfsheur[hi];
841 		}
842 
843 		if (nh->nh_vp != vp) {
844 			nh->nh_vp = vp;
845 			nh->nh_nextr = off;
846 			nh->nh_use = NHUSE_INIT;
847 			if (off == 0)
848 				nh->nh_seqcount = 4;
849 			else
850 				nh->nh_seqcount = 1;
851 		}
852 
853 		/*
854 		 * Calculate heuristic
855 		 */
856 
857 		if ((off == 0 && nh->nh_seqcount > 0) || off == nh->nh_nextr) {
858 			if (++nh->nh_seqcount > IO_SEQMAX)
859 				nh->nh_seqcount = IO_SEQMAX;
860 		} else if (nh->nh_seqcount > 1) {
861 			nh->nh_seqcount = 1;
862 		} else {
863 			nh->nh_seqcount = 0;
864 		}
865 		nh->nh_use += NHUSE_INC;
866 		if (nh->nh_use > NHUSE_MAX)
867 			nh->nh_use = NHUSE_MAX;
868 		ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
869         }
870 
871 	nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
872 	if (v3) {
873 		nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
874 		*tl++ = nfs_true;
875 		fp = (struct nfs_fattr *)tl;
876 		tl += (NFSX_V3FATTR / sizeof (u_int32_t));
877 	} else {
878 		nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
879 		fp = (struct nfs_fattr *)tl;
880 		tl += (NFSX_V2FATTR / sizeof (u_int32_t));
881 	}
882 	len = left = nfsm_rndup(cnt);
883 	if (cnt > 0) {
884 		/*
885 		 * Generate the mbuf list with the uio_iov ref. to it.
886 		 */
887 		i = 0;
888 		m = m2 = mb;
889 		while (left > 0) {
890 			siz = min(M_TRAILINGSPACE(m), left);
891 			if (siz > 0) {
892 				left -= siz;
893 				i++;
894 			}
895 			if (left > 0) {
896 				m = m_getcl(MB_WAIT, MT_DATA, 0);
897 				m->m_len = 0;
898 				m2->m_next = m;
899 				m2 = m;
900 			}
901 		}
902 		MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
903 		       M_TEMP, M_WAITOK);
904 		uiop->uio_iov = iv2 = iv;
905 		m = mb;
906 		left = len;
907 		i = 0;
908 		while (left > 0) {
909 			if (m == NULL)
910 				panic("nfsrv_read iov");
911 			siz = min(M_TRAILINGSPACE(m), left);
912 			if (siz > 0) {
913 				iv->iov_base = mtod(m, caddr_t) + m->m_len;
914 				iv->iov_len = siz;
915 				m->m_len += siz;
916 				left -= siz;
917 				iv++;
918 				i++;
919 			}
920 			m = m->m_next;
921 		}
922 		uiop->uio_iovcnt = i;
923 		uiop->uio_offset = off;
924 		uiop->uio_resid = len;
925 		uiop->uio_rw = UIO_READ;
926 		uiop->uio_segflg = UIO_SYSSPACE;
927 		error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
928 		off = uiop->uio_offset;
929 		nh->nh_nextr = off;
930 		FREE((caddr_t)iv2, M_TEMP);
931 		if (error || (getret = VOP_GETATTR(vp, vap, td))) {
932 			if (!error)
933 				error = getret;
934 			m_freem(mreq);
935 			vput(vp);
936 			vp = NULL;
937 			nfsm_reply(NFSX_POSTOPATTR(v3));
938 			nfsm_srvpostop_attr(getret, vap);
939 			error = 0;
940 			goto nfsmout;
941 		}
942 	} else {
943 		uiop->uio_resid = 0;
944 	}
945 	vput(vp);
946 	vp = NULL;
947 	nfsm_srvfillattr(vap, fp);
948 	tlen = len - uiop->uio_resid;
949 	cnt = cnt < tlen ? cnt : tlen;
950 	tlen = nfsm_rndup(cnt);
951 	if (len != tlen || tlen != cnt)
952 		nfsm_adj(mb, len - tlen, tlen - cnt);
953 	if (v3) {
954 		*tl++ = txdr_unsigned(cnt);
955 		if (len < reqlen)
956 			*tl++ = nfs_true;
957 		else
958 			*tl++ = nfs_false;
959 	}
960 	*tl = txdr_unsigned(cnt);
961 nfsmout:
962 	if (vp)
963 		vput(vp);
964 	return(error);
965 }
966 
967 /*
968  * nfs write service
969  */
970 int
971 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
972 	    struct thread *td, struct mbuf **mrq)
973 {
974 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
975 	struct sockaddr *nam = nfsd->nd_nam;
976 	caddr_t dpos = nfsd->nd_dpos;
977 	struct ucred *cred = &nfsd->nd_cr;
978 	struct iovec *ivp;
979 	int i, cnt;
980 	struct mbuf *mp;
981 	struct nfs_fattr *fp;
982 	struct iovec *iv;
983 	struct vattr va, forat;
984 	struct vattr *vap = &va;
985 	u_int32_t *tl;
986 	int32_t t1;
987 	caddr_t bpos;
988 	int error = 0, rdonly, cache, len, forat_ret = 1;
989 	int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
990 	int stable = NFSV3WRITE_FILESYNC;
991 	int v3 = (nfsd->nd_flag & ND_NFSV3);
992 	char *cp2;
993 	struct mbuf *mb, *mb2, *mreq;
994 	struct vnode *vp = NULL;
995 	nfsfh_t nfh;
996 	fhandle_t *fhp;
997 	struct uio io, *uiop = &io;
998 	off_t off;
999 	u_quad_t frev;
1000 
1001 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1002 	if (mrep == NULL) {
1003 		*mrq = NULL;
1004 		error = 0;
1005 		goto nfsmout;
1006 	}
1007 	fhp = &nfh.fh_generic;
1008 	nfsm_srvmtofh(fhp);
1009 	if (v3) {
1010 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1011 		off = fxdr_hyper(tl);
1012 		tl += 3;
1013 		stable = fxdr_unsigned(int, *tl++);
1014 	} else {
1015 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1016 		off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1017 		tl += 2;
1018 		if (nfs_async)
1019 	    		stable = NFSV3WRITE_UNSTABLE;
1020 	}
1021 	retlen = len = fxdr_unsigned(int32_t, *tl);
1022 	cnt = i = 0;
1023 
1024 	/*
1025 	 * For NFS Version 2, it is not obvious what a write of zero length
1026 	 * should do, but I might as well be consistent with Version 3,
1027 	 * which is to return ok so long as there are no permission problems.
1028 	 */
1029 	if (len > 0) {
1030 	    zeroing = 1;
1031 	    mp = mrep;
1032 	    while (mp) {
1033 		if (mp == md) {
1034 			zeroing = 0;
1035 			adjust = dpos - mtod(mp, caddr_t);
1036 			mp->m_len -= adjust;
1037 			if (mp->m_len > 0 && adjust > 0)
1038 				NFSMADV(mp, adjust);
1039 		}
1040 		if (zeroing)
1041 			mp->m_len = 0;
1042 		else if (mp->m_len > 0) {
1043 			i += mp->m_len;
1044 			if (i > len) {
1045 				mp->m_len -= (i - len);
1046 				zeroing	= 1;
1047 			}
1048 			if (mp->m_len > 0)
1049 				cnt++;
1050 		}
1051 		mp = mp->m_next;
1052 	    }
1053 	}
1054 	if (len > NFS_MAXDATA || len < 0 || i < len) {
1055 		error = EIO;
1056 		nfsm_reply(2 * NFSX_UNSIGNED);
1057 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1058 		error = 0;
1059 		goto nfsmout;
1060 	}
1061 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
1062 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
1063 	if (error) {
1064 		vp = NULL;
1065 		nfsm_reply(2 * NFSX_UNSIGNED);
1066 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1067 		error = 0;
1068 		goto nfsmout;
1069 	}
1070 	if (v3)
1071 		forat_ret = VOP_GETATTR(vp, &forat, td);
1072 	if (vp->v_type != VREG) {
1073 		if (v3)
1074 			error = EINVAL;
1075 		else
1076 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1077 	}
1078 	if (!error) {
1079 		nqsrv_getl(vp, ND_WRITE);
1080 		error = nfsrv_access(vp, VWRITE, cred, rdonly, td, 1);
1081 	}
1082 	if (error) {
1083 		vput(vp);
1084 		vp = NULL;
1085 		nfsm_reply(NFSX_WCCDATA(v3));
1086 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1087 		error = 0;
1088 		goto nfsmout;
1089 	}
1090 
1091 	if (len > 0) {
1092 	    MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
1093 		M_WAITOK);
1094 	    uiop->uio_iov = iv = ivp;
1095 	    uiop->uio_iovcnt = cnt;
1096 	    mp = mrep;
1097 	    while (mp) {
1098 		if (mp->m_len > 0) {
1099 			ivp->iov_base = mtod(mp, caddr_t);
1100 			ivp->iov_len = mp->m_len;
1101 			ivp++;
1102 		}
1103 		mp = mp->m_next;
1104 	    }
1105 
1106 	    /*
1107 	     * XXX
1108 	     * The IO_METASYNC flag indicates that all metadata (and not just
1109 	     * enough to ensure data integrity) mus be written to stable storage
1110 	     * synchronously.
1111 	     * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
1112 	     */
1113 	    if (stable == NFSV3WRITE_UNSTABLE)
1114 		ioflags = IO_NODELOCKED;
1115 	    else if (stable == NFSV3WRITE_DATASYNC)
1116 		ioflags = (IO_SYNC | IO_NODELOCKED);
1117 	    else
1118 		ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1119 	    uiop->uio_resid = len;
1120 	    uiop->uio_rw = UIO_WRITE;
1121 	    uiop->uio_segflg = UIO_SYSSPACE;
1122 	    uiop->uio_td = NULL;
1123 	    uiop->uio_offset = off;
1124 	    error = VOP_WRITE(vp, uiop, ioflags, cred);
1125 	    nfsstats.srvvop_writes++;
1126 	    FREE((caddr_t)iv, M_TEMP);
1127 	}
1128 	aftat_ret = VOP_GETATTR(vp, vap, td);
1129 	vput(vp);
1130 	vp = NULL;
1131 	if (!error)
1132 		error = aftat_ret;
1133 	nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
1134 		2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
1135 	if (v3) {
1136 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1137 		if (error) {
1138 			error = 0;
1139 			goto nfsmout;
1140 		}
1141 		nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1142 		*tl++ = txdr_unsigned(retlen);
1143 		/*
1144 		 * If nfs_async is set, then pretend the write was FILESYNC.
1145 		 */
1146 		if (stable == NFSV3WRITE_UNSTABLE && !nfs_async)
1147 			*tl++ = txdr_unsigned(stable);
1148 		else
1149 			*tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1150 		/*
1151 		 * Actually, there is no need to txdr these fields,
1152 		 * but it may make the values more human readable,
1153 		 * for debugging purposes.
1154 		 */
1155 		if (nfsver.tv_sec == 0)
1156 			nfsver = boottime;
1157 		*tl++ = txdr_unsigned(nfsver.tv_sec);
1158 		*tl = txdr_unsigned(nfsver.tv_nsec / 1000);
1159 	} else {
1160 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1161 		nfsm_srvfillattr(vap, fp);
1162 	}
1163 nfsmout:
1164 	if (vp)
1165 		vput(vp);
1166 	return(error);
1167 }
1168 
1169 /*
1170  * NFS write service with write gathering support. Called when
1171  * nfsrvw_procrastinate > 0.
1172  * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1173  * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1174  * Jan. 1994.
1175  */
1176 int
1177 nfsrv_writegather(struct nfsrv_descript **ndp, struct nfssvc_sock *slp,
1178 		  struct thread *td, struct mbuf **mrq)
1179 {
1180 	struct iovec *ivp;
1181 	struct mbuf *mp;
1182 	struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1183 	struct nfs_fattr *fp;
1184 	int i;
1185 	struct iovec *iov;
1186 	struct nfsrvw_delayhash *wpp;
1187 	struct ucred *cred;
1188 	struct vattr va, forat;
1189 	u_int32_t *tl;
1190 	int32_t t1;
1191 	caddr_t bpos, dpos;
1192 	int error = 0, rdonly, cache, len, forat_ret = 1;
1193 	int ioflags, aftat_ret = 1, s, adjust, v3, zeroing;
1194 	char *cp2;
1195 	struct mbuf *mb, *mb2, *mreq, *mrep, *md;
1196 	struct vnode *vp = NULL;
1197 	struct uio io, *uiop = &io;
1198 	u_quad_t frev, cur_usec;
1199 
1200 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1201 #ifndef nolint
1202 	i = 0;
1203 	len = 0;
1204 #endif
1205 	*mrq = NULL;
1206 	if (*ndp) {
1207 	    nfsd = *ndp;
1208 	    *ndp = NULL;
1209 	    mrep = nfsd->nd_mrep;
1210 	    md = nfsd->nd_md;
1211 	    dpos = nfsd->nd_dpos;
1212 	    cred = &nfsd->nd_cr;
1213 	    v3 = (nfsd->nd_flag & ND_NFSV3);
1214 	    LIST_INIT(&nfsd->nd_coalesce);
1215 	    nfsd->nd_mreq = NULL;
1216 	    nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1217 	    cur_usec = nfs_curusec();
1218 	    nfsd->nd_time = cur_usec +
1219 		(v3 ? nfsrvw_procrastinate_v3 : nfsrvw_procrastinate);
1220 
1221 	    /*
1222 	     * Now, get the write header..
1223 	     */
1224 	    nfsm_srvmtofh(&nfsd->nd_fh);
1225 	    if (v3) {
1226 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1227 		nfsd->nd_off = fxdr_hyper(tl);
1228 		tl += 3;
1229 		nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1230 	    } else {
1231 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1232 		nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1233 		tl += 2;
1234 		if (nfs_async)
1235 			nfsd->nd_stable = NFSV3WRITE_UNSTABLE;
1236 	    }
1237 	    len = fxdr_unsigned(int32_t, *tl);
1238 	    nfsd->nd_len = len;
1239 	    nfsd->nd_eoff = nfsd->nd_off + len;
1240 
1241 	    /*
1242 	     * Trim the header out of the mbuf list and trim off any trailing
1243 	     * junk so that the mbuf list has only the write data.
1244 	     */
1245 	    zeroing = 1;
1246 	    i = 0;
1247 	    mp = mrep;
1248 	    while (mp) {
1249 		if (mp == md) {
1250 		    zeroing = 0;
1251 		    adjust = dpos - mtod(mp, caddr_t);
1252 		    mp->m_len -= adjust;
1253 		    if (mp->m_len > 0 && adjust > 0)
1254 			NFSMADV(mp, adjust);
1255 		}
1256 		if (zeroing)
1257 		    mp->m_len = 0;
1258 		else {
1259 		    i += mp->m_len;
1260 		    if (i > len) {
1261 			mp->m_len -= (i - len);
1262 			zeroing = 1;
1263 		    }
1264 		}
1265 		mp = mp->m_next;
1266 	    }
1267 	    if (len > NFS_MAXDATA || len < 0  || i < len) {
1268 nfsmout:
1269 		m_freem(mrep);
1270 		error = EIO;
1271 		nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1272 		if (v3)
1273 		    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1274 		nfsd->nd_mreq = mreq;
1275 		nfsd->nd_mrep = NULL;
1276 		nfsd->nd_time = 0;
1277 	    }
1278 
1279 	    /*
1280 	     * Add this entry to the hash and time queues.
1281 	     */
1282 	    s = splsoftclock();
1283 	    owp = NULL;
1284 	    wp = slp->ns_tq.lh_first;
1285 	    while (wp && wp->nd_time < nfsd->nd_time) {
1286 		owp = wp;
1287 		wp = wp->nd_tq.le_next;
1288 	    }
1289 	    NFS_DPF(WG, ("Q%03x", nfsd->nd_retxid & 0xfff));
1290 	    if (owp) {
1291 		LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1292 	    } else {
1293 		LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1294 	    }
1295 	    if (nfsd->nd_mrep) {
1296 		wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1297 		owp = NULL;
1298 		wp = wpp->lh_first;
1299 		while (wp &&
1300 		    bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh,NFSX_V3FH)) {
1301 		    owp = wp;
1302 		    wp = wp->nd_hash.le_next;
1303 		}
1304 		while (wp && wp->nd_off < nfsd->nd_off &&
1305 		    !bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh,NFSX_V3FH)) {
1306 		    owp = wp;
1307 		    wp = wp->nd_hash.le_next;
1308 		}
1309 		if (owp) {
1310 		    LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1311 
1312 		    /*
1313 		     * Search the hash list for overlapping entries and
1314 		     * coalesce.
1315 		     */
1316 		    for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1317 			wp = nfsd->nd_hash.le_next;
1318 			if (NFSW_SAMECRED(owp, nfsd))
1319 			    nfsrvw_coalesce(owp, nfsd);
1320 		    }
1321 		} else {
1322 		    LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1323 		}
1324 	    }
1325 	    splx(s);
1326 	}
1327 
1328 	/*
1329 	 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1330 	 * and generate the associated reply mbuf list(s).
1331 	 */
1332 loop1:
1333 	cur_usec = nfs_curusec();
1334 	s = splsoftclock();
1335 	for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = owp) {
1336 		owp = nfsd->nd_tq.le_next;
1337 		if (nfsd->nd_time > cur_usec)
1338 		    break;
1339 		if (nfsd->nd_mreq)
1340 		    continue;
1341 		NFS_DPF(WG, ("P%03x", nfsd->nd_retxid & 0xfff));
1342 		LIST_REMOVE(nfsd, nd_tq);
1343 		LIST_REMOVE(nfsd, nd_hash);
1344 		splx(s);
1345 		mrep = nfsd->nd_mrep;
1346 		nfsd->nd_mrep = NULL;
1347 		cred = &nfsd->nd_cr;
1348 		v3 = (nfsd->nd_flag & ND_NFSV3);
1349 		forat_ret = aftat_ret = 1;
1350 		error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1351 		    nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
1352 		if (!error) {
1353 		    if (v3)
1354 			forat_ret = VOP_GETATTR(vp, &forat, td);
1355 		    if (vp->v_type != VREG) {
1356 			if (v3)
1357 			    error = EINVAL;
1358 			else
1359 			    error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1360 		    }
1361 		} else {
1362 		    vp = NULL;
1363 		}
1364 		if (!error) {
1365 		    nqsrv_getl(vp, ND_WRITE);
1366 		    error = nfsrv_access(vp, VWRITE, cred, rdonly, td, 1);
1367 		}
1368 
1369 		if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1370 		    ioflags = IO_NODELOCKED;
1371 		else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1372 		    ioflags = (IO_SYNC | IO_NODELOCKED);
1373 		else
1374 		    ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1375 		uiop->uio_rw = UIO_WRITE;
1376 		uiop->uio_segflg = UIO_SYSSPACE;
1377 		uiop->uio_td = NULL;
1378 		uiop->uio_offset = nfsd->nd_off;
1379 		uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1380 		if (uiop->uio_resid > 0) {
1381 		    mp = mrep;
1382 		    i = 0;
1383 		    while (mp) {
1384 			if (mp->m_len > 0)
1385 			    i++;
1386 			mp = mp->m_next;
1387 		    }
1388 		    uiop->uio_iovcnt = i;
1389 		    MALLOC(iov, struct iovec *, i * sizeof (struct iovec),
1390 			M_TEMP, M_WAITOK);
1391 		    uiop->uio_iov = ivp = iov;
1392 		    mp = mrep;
1393 		    while (mp) {
1394 			if (mp->m_len > 0) {
1395 			    ivp->iov_base = mtod(mp, caddr_t);
1396 			    ivp->iov_len = mp->m_len;
1397 			    ivp++;
1398 			}
1399 			mp = mp->m_next;
1400 		    }
1401 		    if (!error) {
1402 			error = VOP_WRITE(vp, uiop, ioflags, cred);
1403 			nfsstats.srvvop_writes++;
1404 		    }
1405 		    FREE((caddr_t)iov, M_TEMP);
1406 		}
1407 		m_freem(mrep);
1408 		if (vp) {
1409 		    aftat_ret = VOP_GETATTR(vp, &va, td);
1410 		    vput(vp);
1411 		    vp = NULL;
1412 		}
1413 
1414 		/*
1415 		 * Loop around generating replies for all write rpcs that have
1416 		 * now been completed.
1417 		 */
1418 		swp = nfsd;
1419 		do {
1420 		    NFS_DPF(WG, ("R%03x", nfsd->nd_retxid & 0xfff));
1421 		    if (error) {
1422 			nfsm_writereply(NFSX_WCCDATA(v3), v3);
1423 			if (v3) {
1424 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1425 			}
1426 		    } else {
1427 			nfsm_writereply(NFSX_PREOPATTR(v3) +
1428 			    NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1429 			    NFSX_WRITEVERF(v3), v3);
1430 			if (v3) {
1431 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1432 			    nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1433 			    *tl++ = txdr_unsigned(nfsd->nd_len);
1434 			    *tl++ = txdr_unsigned(swp->nd_stable);
1435 			    /*
1436 			     * Actually, there is no need to txdr these fields,
1437 			     * but it may make the values more human readable,
1438 			     * for debugging purposes.
1439 			     */
1440 			    if (nfsver.tv_sec == 0)
1441 				    nfsver = boottime;
1442 			    *tl++ = txdr_unsigned(nfsver.tv_sec);
1443 			    *tl = txdr_unsigned(nfsver.tv_nsec / 1000);
1444 			} else {
1445 			    nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1446 			    nfsm_srvfillattr(&va, fp);
1447 			}
1448 		    }
1449 		    nfsd->nd_mreq = mreq;
1450 		    if (nfsd->nd_mrep)
1451 			panic("nfsrv_write: nd_mrep not free");
1452 
1453 		    /*
1454 		     * Done. Put it at the head of the timer queue so that
1455 		     * the final phase can return the reply.
1456 		     */
1457 		    s = splsoftclock();
1458 		    if (nfsd != swp) {
1459 			nfsd->nd_time = 0;
1460 			LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1461 		    }
1462 		    nfsd = swp->nd_coalesce.lh_first;
1463 		    if (nfsd) {
1464 			LIST_REMOVE(nfsd, nd_tq);
1465 		    }
1466 		    splx(s);
1467 		} while (nfsd);
1468 		s = splsoftclock();
1469 		swp->nd_time = 0;
1470 		LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1471 		splx(s);
1472 		goto loop1;
1473 	}
1474 	splx(s);
1475 
1476 	/*
1477 	 * Search for a reply to return.
1478 	 */
1479 	s = splsoftclock();
1480 	for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = nfsd->nd_tq.le_next)
1481 		if (nfsd->nd_mreq) {
1482 		    NFS_DPF(WG, ("X%03x", nfsd->nd_retxid & 0xfff));
1483 		    LIST_REMOVE(nfsd, nd_tq);
1484 		    *mrq = nfsd->nd_mreq;
1485 		    *ndp = nfsd;
1486 		    break;
1487 		}
1488 	splx(s);
1489 	return (0);
1490 }
1491 
1492 /*
1493  * Coalesce the write request nfsd into owp. To do this we must:
1494  * - remove nfsd from the queues
1495  * - merge nfsd->nd_mrep into owp->nd_mrep
1496  * - update the nd_eoff and nd_stable for owp
1497  * - put nfsd on owp's nd_coalesce list
1498  * NB: Must be called at splsoftclock().
1499  */
1500 static void
1501 nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
1502 {
1503         int overlap;
1504         struct mbuf *mp;
1505 	struct nfsrv_descript *p;
1506 
1507 	NFS_DPF(WG, ("C%03x-%03x",
1508 		     nfsd->nd_retxid & 0xfff, owp->nd_retxid & 0xfff));
1509         LIST_REMOVE(nfsd, nd_hash);
1510         LIST_REMOVE(nfsd, nd_tq);
1511         if (owp->nd_eoff < nfsd->nd_eoff) {
1512             overlap = owp->nd_eoff - nfsd->nd_off;
1513             if (overlap < 0)
1514                 panic("nfsrv_coalesce: bad off");
1515             if (overlap > 0)
1516                 m_adj(nfsd->nd_mrep, overlap);
1517             mp = owp->nd_mrep;
1518             while (mp->m_next)
1519                 mp = mp->m_next;
1520             mp->m_next = nfsd->nd_mrep;
1521             owp->nd_eoff = nfsd->nd_eoff;
1522         } else
1523             m_freem(nfsd->nd_mrep);
1524         nfsd->nd_mrep = NULL;
1525         if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1526             owp->nd_stable = NFSV3WRITE_FILESYNC;
1527         else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1528             owp->nd_stable == NFSV3WRITE_UNSTABLE)
1529             owp->nd_stable = NFSV3WRITE_DATASYNC;
1530         LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1531 
1532 	/*
1533 	 * If nfsd had anything else coalesced into it, transfer them
1534 	 * to owp, otherwise their replies will never get sent.
1535 	 */
1536 	for (p = nfsd->nd_coalesce.lh_first; p;
1537 	     p = nfsd->nd_coalesce.lh_first) {
1538 	    LIST_REMOVE(p, nd_tq);
1539 	    LIST_INSERT_HEAD(&owp->nd_coalesce, p, nd_tq);
1540 	}
1541 }
1542 
1543 /*
1544  * nfs create service
1545  * now does a truncate to 0 length via. setattr if it already exists
1546  */
1547 int
1548 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1549 	     struct thread *td, struct mbuf **mrq)
1550 {
1551 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1552 	struct sockaddr *nam = nfsd->nd_nam;
1553 	caddr_t dpos = nfsd->nd_dpos;
1554 	struct ucred *cred = &nfsd->nd_cr;
1555 	struct nfs_fattr *fp;
1556 	struct vattr va, dirfor, diraft;
1557 	struct vattr *vap = &va;
1558 	struct nfsv2_sattr *sp;
1559 	u_int32_t *tl;
1560 	struct nlookupdata nd;
1561 	int32_t t1;
1562 	caddr_t bpos;
1563 	int error = 0, rdev, cache, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1564 	int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1565 	caddr_t cp;
1566 	char *cp2;
1567 	struct mbuf *mb, *mb2, *mreq;
1568 	struct vnode *dirp;
1569 	struct vnode *dvp;
1570 	struct vnode *vp;
1571 	nfsfh_t nfh;
1572 	fhandle_t *fhp;
1573 	u_quad_t frev, tempsize;
1574 	u_char cverf[NFSX_V3CREATEVERF];
1575 
1576 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1577 #ifndef nolint
1578 	rdev = 0;
1579 #endif
1580 	nlookup_zero(&nd);
1581 	dirp = NULL;
1582 	dvp = NULL;
1583 	vp = NULL;
1584 
1585 	fhp = &nfh.fh_generic;
1586 	nfsm_srvmtofh(fhp);
1587 	nfsm_srvnamesiz(len);
1588 
1589 	/*
1590 	 * Call namei and do initial cleanup to get a few things
1591 	 * out of the way.  If we get an initial error we cleanup
1592 	 * and return here to avoid special-casing the invalid nd
1593 	 * structure through the rest of the case.  dirp may be
1594 	 * set even if an error occurs, but the nd structure will not
1595 	 * be valid at all if an error occurs so we have to invalidate it
1596 	 * prior to calling nfsm_reply ( which might goto nfsmout ).
1597 	 */
1598 	error = nfs_namei(&nd, cred, NAMEI_CREATE, &dvp, &vp,
1599 			  fhp, len, slp, nam, &md, &dpos, &dirp,
1600 			  td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1601 	if (dirp) {
1602 		if (v3) {
1603 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
1604 		} else {
1605 			vrele(dirp);
1606 			dirp = NULL;
1607 		}
1608 	}
1609 	if (error) {
1610 		nfsm_reply(NFSX_WCCDATA(v3));
1611 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1612 		error = 0;
1613 		goto nfsmout;
1614 	}
1615 
1616 	/*
1617 	 * No error.  Continue.  State:
1618 	 *
1619 	 *	dirp 		may be valid
1620 	 *	vp		may be valid or NULL if the target does not
1621 	 *			exist.
1622 	 *	dvp		is valid
1623 	 *
1624 	 * The error state is set through the code and we may also do some
1625 	 * opportunistic releasing of vnodes to avoid holding locks through
1626 	 * NFS I/O.  The cleanup at the end is a catch-all
1627 	 */
1628 
1629 	VATTR_NULL(vap);
1630 	if (v3) {
1631 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1632 		how = fxdr_unsigned(int, *tl);
1633 		switch (how) {
1634 		case NFSV3CREATE_GUARDED:
1635 			if (vp) {
1636 				error = EEXIST;
1637 				break;
1638 			}
1639 			/* fall through */
1640 		case NFSV3CREATE_UNCHECKED:
1641 			nfsm_srvsattr(vap);
1642 			break;
1643 		case NFSV3CREATE_EXCLUSIVE:
1644 			nfsm_dissect(cp, caddr_t, NFSX_V3CREATEVERF);
1645 			bcopy(cp, cverf, NFSX_V3CREATEVERF);
1646 			exclusive_flag = 1;
1647 			break;
1648 		};
1649 		vap->va_type = VREG;
1650 	} else {
1651 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1652 		vap->va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1653 		if (vap->va_type == VNON)
1654 			vap->va_type = VREG;
1655 		vap->va_mode = nfstov_mode(sp->sa_mode);
1656 		switch (vap->va_type) {
1657 		case VREG:
1658 			tsize = fxdr_unsigned(int32_t, sp->sa_size);
1659 			if (tsize != -1)
1660 				vap->va_size = (u_quad_t)tsize;
1661 			break;
1662 		case VCHR:
1663 		case VBLK:
1664 		case VFIFO:
1665 			rdev = fxdr_unsigned(long, sp->sa_size);
1666 			break;
1667 		default:
1668 			break;
1669 		};
1670 	}
1671 
1672 	/*
1673 	 * Iff doesn't exist, create it
1674 	 * otherwise just truncate to 0 length
1675 	 *   should I set the mode too ?
1676 	 *
1677 	 * The only possible error we can have at this point is EEXIST.
1678 	 * nd.ni_vp will also be non-NULL in that case.
1679 	 */
1680 	if (vp == NULL) {
1681 		if (vap->va_mode == (mode_t)VNOVAL)
1682 			vap->va_mode = 0;
1683 		if (vap->va_type == VREG || vap->va_type == VSOCK) {
1684 			nqsrv_getl(dvp, ND_WRITE);
1685 			error = VOP_NCREATE(nd.nl_ncp, &vp, nd.nl_cred, vap);
1686 			if (error == 0) {
1687 			    	nfsrv_object_create(vp);
1688 				if (exclusive_flag) {
1689 					exclusive_flag = 0;
1690 					VATTR_NULL(vap);
1691 					bcopy(cverf, (caddr_t)&vap->va_atime,
1692 						NFSX_V3CREATEVERF);
1693 					error = VOP_SETATTR(vp, vap, cred, td);
1694 				}
1695 			}
1696 		} else if (
1697 			vap->va_type == VCHR ||
1698 			vap->va_type == VBLK ||
1699 			vap->va_type == VFIFO
1700 		) {
1701 			/*
1702 			 * Handle SysV FIFO node special cases.  All other
1703 			 * devices require super user to access.
1704 			 */
1705 			if (vap->va_type == VCHR && rdev == 0xffffffff)
1706 				vap->va_type = VFIFO;
1707                         if (vap->va_type != VFIFO &&
1708                             (error = suser_cred(cred, 0))) {
1709 				goto nfsmreply0;
1710                         }
1711 			vap->va_rdev = rdev;
1712 			nqsrv_getl(dvp, ND_WRITE);
1713 
1714 			error = VOP_NMKNOD(nd.nl_ncp, &vp, nd.nl_cred, vap);
1715 			if (error)
1716 				goto nfsmreply0;
1717 #if 0
1718 			/*
1719 			 * XXX what is this junk supposed to do ?
1720 			 */
1721 
1722 			vput(vp);
1723 			vp = NULL;
1724 
1725 			/*
1726 			 * release dvp prior to lookup
1727 			 */
1728 			vput(dvp);
1729 			dvp = NULL;
1730 
1731 			/*
1732 			 * Setup for lookup.
1733 			 *
1734 			 * Even though LOCKPARENT was cleared, ni_dvp may
1735 			 * be garbage.
1736 			 */
1737 			nd.ni_cnd.cn_nameiop = NAMEI_LOOKUP;
1738 			nd.ni_cnd.cn_flags &= ~(CNP_LOCKPARENT);
1739 			nd.ni_cnd.cn_td = td;
1740 			nd.ni_cnd.cn_cred = cred;
1741 
1742 			error = lookup(&nd);
1743 			nd.ni_dvp = NULL;
1744 
1745 			if (error != 0) {
1746 				nfsm_reply(0);
1747 				/* fall through on certain errors */
1748 			}
1749 			nfsrv_object_create(nd.ni_vp);
1750 			if (nd.ni_cnd.cn_flags & CNP_ISSYMLINK) {
1751 				error = EINVAL;
1752 				goto nfsmreply0;
1753 			}
1754 #endif
1755 		} else {
1756 			error = ENXIO;
1757 		}
1758 	} else {
1759 		if (vap->va_size != -1) {
1760 			error = nfsrv_access(vp, VWRITE, cred,
1761 			    (nd.nl_flags & NLC_NFS_RDONLY), td, 0);
1762 			if (!error) {
1763 				nqsrv_getl(vp, ND_WRITE);
1764 				tempsize = vap->va_size;
1765 				VATTR_NULL(vap);
1766 				vap->va_size = tempsize;
1767 				error = VOP_SETATTR(vp, vap, cred, td);
1768 			}
1769 		}
1770 	}
1771 
1772 	if (!error) {
1773 		bzero((caddr_t)fhp, sizeof(nfh));
1774 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1775 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
1776 		if (!error)
1777 			error = VOP_GETATTR(vp, vap, td);
1778 	}
1779 	if (v3) {
1780 		if (exclusive_flag && !error &&
1781 			bcmp(cverf, (caddr_t)&vap->va_atime, NFSX_V3CREATEVERF))
1782 			error = EEXIST;
1783 		diraft_ret = VOP_GETATTR(dirp, &diraft, td);
1784 		vrele(dirp);
1785 		dirp = NULL;
1786 	}
1787 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1788 	if (v3) {
1789 		if (!error) {
1790 			nfsm_srvpostop_fh(fhp);
1791 			nfsm_srvpostop_attr(0, vap);
1792 		}
1793 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1794 		error = 0;
1795 	} else {
1796 		nfsm_srvfhtom(fhp, v3);
1797 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1798 		nfsm_srvfillattr(vap, fp);
1799 	}
1800 	goto nfsmout;
1801 
1802 nfsmreply0:
1803 	nfsm_reply(0);
1804 	error = 0;
1805 	/* fall through */
1806 
1807 nfsmout:
1808 	if (dirp)
1809 		vrele(dirp);
1810 	nlookup_done(&nd);
1811 	if (dvp) {
1812 		if (dvp == vp)
1813 			vrele(dvp);
1814 		else
1815 			vput(dvp);
1816 	}
1817 	if (vp)
1818 		vput(vp);
1819 	return (error);
1820 }
1821 
1822 /*
1823  * nfs v3 mknod service
1824  */
1825 int
1826 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1827 	    struct thread *td, struct mbuf **mrq)
1828 {
1829 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1830 	struct sockaddr *nam = nfsd->nd_nam;
1831 	caddr_t dpos = nfsd->nd_dpos;
1832 	struct ucred *cred = &nfsd->nd_cr;
1833 	struct vattr va, dirfor, diraft;
1834 	struct vattr *vap = &va;
1835 	u_int32_t *tl;
1836 	struct nlookupdata nd;
1837 	int32_t t1;
1838 	caddr_t bpos;
1839 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
1840 	u_int32_t major, minor;
1841 	enum vtype vtyp;
1842 	char *cp2;
1843 	struct mbuf *mb, *mb2, *mreq;
1844 	struct vnode *dirp;
1845 	struct vnode *dvp;
1846 	struct vnode *vp;
1847 	nfsfh_t nfh;
1848 	fhandle_t *fhp;
1849 	u_quad_t frev;
1850 
1851 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1852 	nlookup_zero(&nd);
1853 	dirp = NULL;
1854 	dvp = NULL;
1855 	vp = NULL;
1856 
1857 	fhp = &nfh.fh_generic;
1858 	nfsm_srvmtofh(fhp);
1859 	nfsm_srvnamesiz(len);
1860 
1861 	/*
1862 	 * Handle nfs_namei() call.  If an error occurs, the nd structure
1863 	 * is not valid.  However, nfsm_*() routines may still jump to
1864 	 * nfsmout.
1865 	 */
1866 
1867 	error = nfs_namei(&nd, cred, NAMEI_CREATE, &dvp, &vp,
1868 			  fhp, len, slp, nam, &md, &dpos, &dirp,
1869 			  td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1870 	if (dirp)
1871 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
1872 	if (error) {
1873 		nfsm_reply(NFSX_WCCDATA(1));
1874 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1875 		error = 0;
1876 		goto nfsmout;
1877 	}
1878 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1879 	vtyp = nfsv3tov_type(*tl);
1880 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1881 		error = NFSERR_BADTYPE;
1882 		goto out;
1883 	}
1884 	VATTR_NULL(vap);
1885 	nfsm_srvsattr(vap);
1886 	if (vtyp == VCHR || vtyp == VBLK) {
1887 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1888 		major = fxdr_unsigned(u_int32_t, *tl++);
1889 		minor = fxdr_unsigned(u_int32_t, *tl);
1890 		vap->va_rdev = makeudev(major, minor);
1891 	}
1892 
1893 	/*
1894 	 * Iff doesn't exist, create it.
1895 	 */
1896 	if (vp) {
1897 		error = EEXIST;
1898 		goto out;
1899 	}
1900 	vap->va_type = vtyp;
1901 	if (vap->va_mode == (mode_t)VNOVAL)
1902 		vap->va_mode = 0;
1903 	if (vtyp == VSOCK) {
1904 		nqsrv_getl(dvp, ND_WRITE);
1905 		error = VOP_NCREATE(nd.nl_ncp, &vp, nd.nl_cred, vap);
1906 	} else {
1907 		if (vtyp != VFIFO && (error = suser_cred(cred, 0)))
1908 			goto out;
1909 		nqsrv_getl(dvp, ND_WRITE);
1910 
1911 		error = VOP_NMKNOD(nd.nl_ncp, &vp, nd.nl_cred, vap);
1912 		if (error)
1913 			goto out;
1914 
1915 #if 0
1916 		vput(vp);
1917 		vp = NULL;
1918 
1919 		/*
1920 		 * Release dvp prior to lookup
1921 		 */
1922 		vput(dvp);
1923 		dvp = NULL;
1924 
1925 		/*
1926 		 * XXX what is this stuff for?
1927 		 */
1928 		KKASSERT(td->td_proc);
1929 		nd.ni_cnd.cn_nameiop = NAMEI_LOOKUP;
1930 		nd.ni_cnd.cn_flags &= ~(CNP_LOCKPARENT);
1931 		nd.ni_cnd.cn_td = td;
1932 		nd.ni_cnd.cn_cred = td->td_proc->p_ucred;
1933 
1934 		error = lookup(&nd);
1935 		nd.ni_dvp = NULL;
1936 
1937 		if (error)
1938 			goto out;
1939 		if (nd.ni_cnd.cn_flags & CNP_ISSYMLINK)
1940 			error = EINVAL;
1941 #endif
1942 	}
1943 
1944 	/*
1945 	 * send response, cleanup, return.
1946 	 */
1947 out:
1948 	nlookup_done(&nd);
1949 	if (dvp) {
1950 		if (dvp == vp)
1951 			vrele(dvp);
1952 		else
1953 			vput(dvp);
1954 		dvp = NULL;
1955 	}
1956 	if (!error) {
1957 		bzero((caddr_t)fhp, sizeof(nfh));
1958 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1959 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
1960 		if (!error)
1961 			error = VOP_GETATTR(vp, vap, td);
1962 	}
1963 	if (vp) {
1964 		vput(vp);
1965 		vp = NULL;
1966 	}
1967 	diraft_ret = VOP_GETATTR(dirp, &diraft, td);
1968 	if (dirp) {
1969 		vrele(dirp);
1970 		dirp = NULL;
1971 	}
1972 	nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
1973 	if (!error) {
1974 		nfsm_srvpostop_fh(fhp);
1975 		nfsm_srvpostop_attr(0, vap);
1976 	}
1977 	nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1978 	return (0);
1979 nfsmout:
1980 	if (dirp)
1981 		vrele(dirp);
1982 	nlookup_done(&nd);
1983 	if (dvp) {
1984 		if (dvp == vp)
1985 			vrele(dvp);
1986 		else
1987 			vput(dvp);
1988 	}
1989 	if (vp)
1990 		vput(vp);
1991 	return (error);
1992 }
1993 
1994 /*
1995  * nfs remove service
1996  */
1997 int
1998 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1999 	     struct thread *td, struct mbuf **mrq)
2000 {
2001 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2002 	struct sockaddr *nam = nfsd->nd_nam;
2003 	caddr_t dpos = nfsd->nd_dpos;
2004 	struct ucred *cred = &nfsd->nd_cr;
2005 	struct nlookupdata nd;
2006 	u_int32_t *tl;
2007 	int32_t t1;
2008 	caddr_t bpos;
2009 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2010 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2011 	char *cp2;
2012 	struct mbuf *mb, *mreq;
2013 	struct vnode *dirp;
2014 	struct vnode *dvp;
2015 	struct vnode *vp;
2016 	struct vattr dirfor, diraft;
2017 	nfsfh_t nfh;
2018 	fhandle_t *fhp;
2019 	u_quad_t frev;
2020 
2021 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2022 	nlookup_zero(&nd);
2023 	dirp = NULL;
2024 	dvp = NULL;
2025 	vp = NULL;
2026 
2027 	fhp = &nfh.fh_generic;
2028 	nfsm_srvmtofh(fhp);
2029 	nfsm_srvnamesiz(len);
2030 
2031 	error = nfs_namei(&nd, cred, NAMEI_DELETE, &dvp, &vp,
2032 			  fhp, len, slp, nam, &md, &dpos, &dirp,
2033 			  td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2034 	if (dirp) {
2035 		if (v3)
2036 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
2037 	}
2038 	if (error == 0) {
2039 		if (vp->v_type == VDIR) {
2040 			error = EPERM;		/* POSIX */
2041 			goto out;
2042 		}
2043 		/*
2044 		 * The root of a mounted filesystem cannot be deleted.
2045 		 */
2046 		if (vp->v_flag & VROOT) {
2047 			error = EBUSY;
2048 			goto out;
2049 		}
2050 out:
2051 		if (!error) {
2052 			nqsrv_getl(dvp, ND_WRITE);
2053 			nqsrv_getl(vp, ND_WRITE);
2054 			error = VOP_NREMOVE(nd.nl_ncp, nd.nl_cred);
2055 		}
2056 	}
2057 	if (dirp && v3)
2058 		diraft_ret = VOP_GETATTR(dirp, &diraft, td);
2059 	nfsm_reply(NFSX_WCCDATA(v3));
2060 	if (v3) {
2061 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2062 		error = 0;
2063 	}
2064 nfsmout:
2065 	nlookup_done(&nd);
2066 	if (dirp)
2067 		vrele(dirp);
2068 	if (dvp) {
2069 		if (dvp == vp)
2070 			vrele(dvp);
2071 		else
2072 			vput(dvp);
2073 	}
2074 	if (vp)
2075 		vput(vp);
2076 	return(error);
2077 }
2078 
2079 /*
2080  * nfs rename service
2081  */
2082 int
2083 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2084 	     struct thread *td, struct mbuf **mrq)
2085 {
2086 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2087 	struct sockaddr *nam = nfsd->nd_nam;
2088 	caddr_t dpos = nfsd->nd_dpos;
2089 	struct ucred *cred = &nfsd->nd_cr;
2090 	u_int32_t *tl;
2091 	int32_t t1;
2092 	caddr_t bpos;
2093 	int error = 0, cache = 0, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
2094 	int tdirfor_ret = 1, tdiraft_ret = 1;
2095 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2096 	char *cp2;
2097 	struct mbuf *mb, *mreq;
2098 	struct nlookupdata fromnd, tond;
2099 	struct vnode *fvp, *fdirp;
2100 	struct vnode *tvp, *tdirp;
2101 	struct namecache *ncp;
2102 	struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
2103 	nfsfh_t fnfh, tnfh;
2104 	fhandle_t *ffhp, *tfhp;
2105 	u_quad_t frev;
2106 	uid_t saved_uid;
2107 
2108 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2109 #ifndef nolint
2110 	fvp = (struct vnode *)0;
2111 #endif
2112 	ffhp = &fnfh.fh_generic;
2113 	tfhp = &tnfh.fh_generic;
2114 
2115 	/*
2116 	 * Clear fields incase goto nfsmout occurs from macro.
2117 	 */
2118 
2119 	nlookup_zero(&fromnd);
2120 	nlookup_zero(&tond);
2121 	fdirp = NULL;
2122 	tdirp = NULL;
2123 
2124 	nfsm_srvmtofh(ffhp);
2125 	nfsm_srvnamesiz(len);
2126 	/*
2127 	 * Remember our original uid so that we can reset cr_uid before
2128 	 * the second nfs_namei() call, in case it is remapped.
2129 	 */
2130 	saved_uid = cred->cr_uid;
2131 	error = nfs_namei(&fromnd, cred, NAMEI_DELETE, NULL, NULL,
2132 			  ffhp, len, slp, nam, &md, &dpos, &fdirp,
2133 			  td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2134 	if (fdirp) {
2135 		if (v3)
2136 			fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, td);
2137 	}
2138 	if (error) {
2139 		nfsm_reply(2 * NFSX_WCCDATA(v3));
2140 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2141 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2142 		error = 0;
2143 		goto nfsmout;
2144 	}
2145 
2146 	/*
2147 	 * We have to unlock the from ncp before we can safely lookup
2148 	 * the target ncp.
2149 	 */
2150 	KKASSERT(fromnd.nl_flags & NLC_NCPISLOCKED);
2151 	cache_unlock(fromnd.nl_ncp);
2152 	fromnd.nl_flags &= ~NLC_NCPISLOCKED;
2153 	nfsm_srvmtofh(tfhp);
2154 	nfsm_strsiz(len2, NFS_MAXNAMLEN);
2155 	cred->cr_uid = saved_uid;
2156 
2157 	error = nfs_namei(&tond, cred, NAMEI_RENAME, NULL, NULL,
2158 			  tfhp, len2, slp, nam, &md, &dpos, &tdirp,
2159 			  td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2160 	if (tdirp) {
2161 		if (v3)
2162 			tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, td);
2163 	}
2164 	if (error)
2165 		goto out1;
2166 
2167 	/*
2168 	 * relock the source
2169 	 */
2170 	if (cache_lock_nonblock(fromnd.nl_ncp) == 0) {
2171 		cache_resolve(fromnd.nl_ncp, fromnd.nl_cred);
2172 	} else if (fromnd.nl_ncp > tond.nl_ncp) {
2173 		cache_lock(fromnd.nl_ncp);
2174 		cache_resolve(fromnd.nl_ncp, fromnd.nl_cred);
2175 	} else {
2176 		cache_unlock(tond.nl_ncp);
2177 		cache_lock(fromnd.nl_ncp);
2178 		cache_resolve(fromnd.nl_ncp, fromnd.nl_cred);
2179 		cache_lock(tond.nl_ncp);
2180 		cache_resolve(tond.nl_ncp, tond.nl_cred);
2181 	}
2182 	fromnd.nl_flags |= NLC_NCPISLOCKED;
2183 
2184 	tvp = tond.nl_ncp->nc_vp;
2185 	fvp = fromnd.nl_ncp->nc_vp;
2186 
2187 	if (tvp != NULL) {
2188 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
2189 			if (v3)
2190 				error = EEXIST;
2191 			else
2192 				error = EISDIR;
2193 			goto out;
2194 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
2195 			if (v3)
2196 				error = EEXIST;
2197 			else
2198 				error = ENOTDIR;
2199 			goto out;
2200 		}
2201 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
2202 			if (v3)
2203 				error = EXDEV;
2204 			else
2205 				error = ENOTEMPTY;
2206 			goto out;
2207 		}
2208 	}
2209 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
2210 		if (v3)
2211 			error = EXDEV;
2212 		else
2213 			error = ENOTEMPTY;
2214 		goto out;
2215 	}
2216 	if (fromnd.nl_ncp->nc_mount != tond.nl_ncp->nc_mount) {
2217 		if (v3)
2218 			error = EXDEV;
2219 		else
2220 			error = ENOTEMPTY;
2221 		goto out;
2222 	}
2223 	if (fromnd.nl_ncp == tond.nl_ncp->nc_parent) {
2224 		if (v3)
2225 			error = EINVAL;
2226 		else
2227 			error = ENOTEMPTY;
2228 	}
2229 
2230 	/*
2231 	 * You cannot rename a source into itself or a subdirectory of itself.
2232 	 * We check this by travsering the target directory upwards looking
2233 	 * for a match against the source.
2234 	 */
2235 	if (error == 0) {
2236 		for (ncp = tond.nl_ncp; ncp; ncp = ncp->nc_parent) {
2237 			if (fromnd.nl_ncp == ncp) {
2238 				error = EINVAL;
2239 				break;
2240 			}
2241 		}
2242 	}
2243 
2244 	/*
2245 	 * If source is the same as the destination (that is the
2246 	 * same vnode with the same name in the same directory),
2247 	 * then there is nothing to do.
2248 	 */
2249 	if (fromnd.nl_ncp == tond.nl_ncp)
2250 		error = -1;
2251 out:
2252 	if (!error) {
2253 		/*
2254 		 * The VOP_NRENAME function releases all vnode references &
2255 		 * locks prior to returning so we need to clear the pointers
2256 		 * to bypass cleanup code later on.
2257 		 */
2258 #if 0
2259 		/* XXX should be handled by NRENAME */
2260 		nqsrv_getl(fdvp, ND_WRITE);
2261 		nqsrv_getl(tdvp, ND_WRITE);
2262 		if (tvp) {
2263 			nqsrv_getl(tvp, ND_WRITE);
2264 		}
2265 #endif
2266 		error = VOP_NRENAME(fromnd.nl_ncp, tond.nl_ncp, tond.nl_cred);
2267 	} else {
2268 		if (error == -1)
2269 			error = 0;
2270 	}
2271 	/* fall through */
2272 
2273 out1:
2274 	if (fdirp)
2275 		fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, td);
2276 	if (tdirp)
2277 		tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, td);
2278 	nfsm_reply(2 * NFSX_WCCDATA(v3));
2279 	if (v3) {
2280 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2281 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2282 	}
2283 	error = 0;
2284 	/* fall through */
2285 
2286 nfsmout:
2287 	if (tdirp)
2288 		vrele(tdirp);
2289 	nlookup_done(&tond);
2290 	if (fdirp)
2291 		vrele(fdirp);
2292 	nlookup_done(&fromnd);
2293 	return (error);
2294 }
2295 
2296 /*
2297  * nfs link service
2298  */
2299 int
2300 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2301 	   struct thread *td, struct mbuf **mrq)
2302 {
2303 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2304 	struct sockaddr *nam = nfsd->nd_nam;
2305 	caddr_t dpos = nfsd->nd_dpos;
2306 	struct ucred *cred = &nfsd->nd_cr;
2307 	struct nlookupdata nd;
2308 	u_int32_t *tl;
2309 	int32_t t1;
2310 	caddr_t bpos;
2311 	int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2312 	int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2313 	char *cp2;
2314 	struct mbuf *mb, *mreq;
2315 	struct vnode *dirp;
2316 	struct vnode *dvp;
2317 	struct vnode *vp;
2318 	struct vnode *xp;
2319 	struct vattr dirfor, diraft, at;
2320 	nfsfh_t nfh, dnfh;
2321 	fhandle_t *fhp, *dfhp;
2322 	u_quad_t frev;
2323 
2324 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2325 	nlookup_zero(&nd);
2326 	dirp = dvp = vp = xp = NULL;
2327 
2328 	fhp = &nfh.fh_generic;
2329 	dfhp = &dnfh.fh_generic;
2330 	nfsm_srvmtofh(fhp);
2331 	nfsm_srvmtofh(dfhp);
2332 	nfsm_srvnamesiz(len);
2333 
2334 	error = nfsrv_fhtovp(fhp, FALSE, &xp, cred, slp, nam,
2335 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
2336 	if (error) {
2337 		nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2338 		nfsm_srvpostop_attr(getret, &at);
2339 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2340 		xp = NULL;
2341 		error = 0;
2342 		goto nfsmout;
2343 	}
2344 	if (xp->v_type == VDIR) {
2345 		error = EPERM;		/* POSIX */
2346 		goto out1;
2347 	}
2348 
2349 	error = nfs_namei(&nd, cred, NAMEI_CREATE, &dvp, &vp,
2350 			  dfhp, len, slp, nam, &md, &dpos, &dirp,
2351 			  td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2352 	if (dirp) {
2353 		if (v3)
2354 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
2355 	}
2356 	if (error)
2357 		goto out1;
2358 
2359 	if (vp != NULL) {
2360 		error = EEXIST;
2361 		goto out;
2362 	}
2363 	if (xp->v_mount != dvp->v_mount)
2364 		error = EXDEV;
2365 out:
2366 	if (!error) {
2367 #if 0
2368 		nqsrv_getl(xp, ND_WRITE);
2369 		nqsrv_getl(vp, ND_WRITE);
2370 #endif
2371 		error = VOP_NLINK(nd.nl_ncp, xp, nd.nl_cred);
2372 	}
2373 	/* fall through */
2374 
2375 out1:
2376 	if (v3)
2377 		getret = VOP_GETATTR(xp, &at, td);
2378 	if (dirp)
2379 		diraft_ret = VOP_GETATTR(dirp, &diraft, td);
2380 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2381 	if (v3) {
2382 		nfsm_srvpostop_attr(getret, &at);
2383 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2384 		error = 0;
2385 	}
2386 	/* fall through */
2387 
2388 nfsmout:
2389 	nlookup_done(&nd);
2390 	if (dirp)
2391 		vrele(dirp);
2392 	if (xp)
2393 		vrele(xp);
2394 	if (dvp) {
2395 		if (dvp == vp)
2396 			vrele(dvp);
2397 		else
2398 			vput(dvp);
2399 	}
2400 	if (vp)
2401 		vput(vp);
2402 	return(error);
2403 }
2404 
2405 /*
2406  * nfs symbolic link service
2407  */
2408 int
2409 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2410 	      struct thread *td, struct mbuf **mrq)
2411 {
2412 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2413 	struct sockaddr *nam = nfsd->nd_nam;
2414 	caddr_t dpos = nfsd->nd_dpos;
2415 	struct ucred *cred = &nfsd->nd_cr;
2416 	struct vattr va, dirfor, diraft;
2417 	struct nlookupdata nd;
2418 	struct vattr *vap = &va;
2419 	u_int32_t *tl;
2420 	int32_t t1;
2421 	struct nfsv2_sattr *sp;
2422 	char *bpos, *pathcp = (char *)0, *cp2;
2423 	struct uio io;
2424 	struct iovec iv;
2425 	int error = 0, cache = 0, len, len2, dirfor_ret = 1, diraft_ret = 1;
2426 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2427 	struct mbuf *mb, *mreq, *mb2;
2428 	struct vnode *dirp;
2429 	struct vnode *vp;
2430 	nfsfh_t nfh;
2431 	fhandle_t *fhp;
2432 	u_quad_t frev;
2433 
2434 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2435 	nlookup_zero(&nd);
2436 	dirp = vp = NULL;
2437 
2438 	fhp = &nfh.fh_generic;
2439 	nfsm_srvmtofh(fhp);
2440 	nfsm_srvnamesiz(len);
2441 
2442 	error = nfs_namei(&nd, cred, NAMEI_CREATE, NULL, &vp,
2443 			fhp, len, slp, nam, &md, &dpos, &dirp,
2444 			td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2445 	if (dirp) {
2446 		if (v3)
2447 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
2448 	}
2449 	if (error)
2450 		goto out;
2451 
2452 	VATTR_NULL(vap);
2453 	if (v3)
2454 		nfsm_srvsattr(vap);
2455 	nfsm_strsiz(len2, NFS_MAXPATHLEN);
2456 	MALLOC(pathcp, caddr_t, len2 + 1, M_TEMP, M_WAITOK);
2457 	iv.iov_base = pathcp;
2458 	iv.iov_len = len2;
2459 	io.uio_resid = len2;
2460 	io.uio_offset = 0;
2461 	io.uio_iov = &iv;
2462 	io.uio_iovcnt = 1;
2463 	io.uio_segflg = UIO_SYSSPACE;
2464 	io.uio_rw = UIO_READ;
2465 	io.uio_td = NULL;
2466 	nfsm_mtouio(&io, len2);
2467 	if (!v3) {
2468 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2469 		vap->va_mode = nfstov_mode(sp->sa_mode);
2470 	}
2471 	*(pathcp + len2) = '\0';
2472 	if (vp) {
2473 		error = EEXIST;
2474 		goto out;
2475 	}
2476 
2477 	if (vap->va_mode == (mode_t)VNOVAL)
2478 		vap->va_mode = 0;
2479 #if 0
2480 	nqsrv_getl(dvp, ND_WRITE);
2481 #endif
2482 	error = VOP_NSYMLINK(nd.nl_ncp, &vp, nd.nl_cred, vap, pathcp);
2483 	if (error == 0) {
2484 		bzero((caddr_t)fhp, sizeof(nfh));
2485 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2486 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
2487 		if (!error)
2488 			error = VOP_GETATTR(vp, vap, td);
2489 	}
2490 
2491 #if 0
2492 	/*
2493 	 * We have a vp in hand from the new API call, we do not have to
2494 	 * look it up again.
2495 	 */
2496 	if (error == 0) {
2497 	    if (v3) {
2498 		/*
2499 		 * Issue lookup.  Leave SAVESTART set so we can easily free
2500 		 * the name buffer later on.
2501 		 *
2502 		 * since LOCKPARENT is not set, ni_dvp will be garbage on
2503 		 * return whether an error occurs or not.
2504 		 */
2505 		nd.ni_cnd.cn_nameiop = NAMEI_LOOKUP;
2506 		nd.ni_cnd.cn_flags &= ~(CNP_LOCKPARENT | CNP_FOLLOW);
2507 		nd.ni_cnd.cn_td = td;
2508 		nd.ni_cnd.cn_cred = cred;
2509 
2510 		error = lookup(&nd);
2511 		nd.ni_dvp = NULL;
2512 
2513 		if (error == 0) {
2514 			bzero((caddr_t)fhp, sizeof(nfh));
2515 			fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2516 			error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2517 			if (!error)
2518 				error = VOP_GETATTR(nd.ni_vp, vap, td);
2519 			vput(nd.ni_vp);
2520 			nd.ni_vp = NULL;
2521 		}
2522 	    }
2523 	}
2524 #endif
2525 out:
2526 	if (vp) {
2527 		vput(vp);
2528 		vp = NULL;
2529 	}
2530 	if (pathcp) {
2531 		FREE(pathcp, M_TEMP);
2532 		pathcp = NULL;
2533 	}
2534 	if (dirp) {
2535 		diraft_ret = VOP_GETATTR(dirp, &diraft, td);
2536 		vrele(dirp);
2537 		dirp = NULL;
2538 	}
2539 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2540 	if (v3) {
2541 		if (!error) {
2542 			nfsm_srvpostop_fh(fhp);
2543 			nfsm_srvpostop_attr(0, vap);
2544 		}
2545 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2546 	}
2547 	error = 0;
2548 	/* fall through */
2549 
2550 nfsmout:
2551 	nlookup_done(&nd);
2552 	if (vp)
2553 		vput(vp);
2554 	if (dirp)
2555 		vrele(dirp);
2556 	if (pathcp)
2557 		FREE(pathcp, M_TEMP);
2558 	return (error);
2559 }
2560 
2561 /*
2562  * nfs mkdir service
2563  */
2564 int
2565 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2566 	    struct thread *td, struct mbuf **mrq)
2567 {
2568 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2569 	struct sockaddr *nam = nfsd->nd_nam;
2570 	caddr_t dpos = nfsd->nd_dpos;
2571 	struct ucred *cred = &nfsd->nd_cr;
2572 	struct vattr va, dirfor, diraft;
2573 	struct vattr *vap = &va;
2574 	struct nfs_fattr *fp;
2575 	struct nlookupdata nd;
2576 	caddr_t cp;
2577 	u_int32_t *tl;
2578 	int32_t t1;
2579 	caddr_t bpos;
2580 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2581 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2582 	char *cp2;
2583 	struct mbuf *mb, *mb2, *mreq;
2584 	struct vnode *dirp;
2585 	struct vnode *vp;
2586 	nfsfh_t nfh;
2587 	fhandle_t *fhp;
2588 	u_quad_t frev;
2589 
2590 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2591 	nlookup_zero(&nd);
2592 	dirp = NULL;
2593 	vp = NULL;
2594 
2595 	fhp = &nfh.fh_generic;
2596 	nfsm_srvmtofh(fhp);
2597 	nfsm_srvnamesiz(len);
2598 
2599 	error = nfs_namei(&nd, cred, NAMEI_CREATE, NULL, &vp,
2600 			  fhp, len, slp, nam, &md, &dpos, &dirp,
2601 			  td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2602 	if (dirp) {
2603 		if (v3)
2604 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
2605 	}
2606 	if (error) {
2607 		nfsm_reply(NFSX_WCCDATA(v3));
2608 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2609 		error = 0;
2610 		goto nfsmout;
2611 	}
2612 	VATTR_NULL(vap);
2613 	if (v3) {
2614 		nfsm_srvsattr(vap);
2615 	} else {
2616 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2617 		vap->va_mode = nfstov_mode(*tl++);
2618 	}
2619 
2620 	/*
2621 	 * At this point nd.ni_dvp is referenced and exclusively locked and
2622 	 * nd.ni_vp, if it exists, is referenced but not locked.
2623 	 */
2624 
2625 	vap->va_type = VDIR;
2626 	if (vp != NULL) {
2627 		error = EEXIST;
2628 		goto out;
2629 	}
2630 
2631 	/*
2632 	 * Issue mkdir op.  Since SAVESTART is not set, the pathname
2633 	 * component is freed by the VOP call.  This will fill-in
2634 	 * nd.ni_vp, reference, and exclusively lock it.
2635 	 */
2636 	if (vap->va_mode == (mode_t)VNOVAL)
2637 		vap->va_mode = 0;
2638 #if 0
2639 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2640 #endif
2641 	error = VOP_NMKDIR(nd.nl_ncp, &vp, nd.nl_cred, vap);
2642 
2643 	if (error == 0) {
2644 		bzero((caddr_t)fhp, sizeof(nfh));
2645 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2646 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
2647 		if (error == 0)
2648 			error = VOP_GETATTR(vp, vap, td);
2649 	}
2650 out:
2651 	if (dirp)
2652 		diraft_ret = VOP_GETATTR(dirp, &diraft, td);
2653 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2654 	if (v3) {
2655 		if (!error) {
2656 			nfsm_srvpostop_fh(fhp);
2657 			nfsm_srvpostop_attr(0, vap);
2658 		}
2659 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2660 	} else {
2661 		nfsm_srvfhtom(fhp, v3);
2662 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2663 		nfsm_srvfillattr(vap, fp);
2664 	}
2665 	error = 0;
2666 	/* fall through */
2667 
2668 nfsmout:
2669 	nlookup_done(&nd);
2670 	if (dirp)
2671 		vrele(dirp);
2672 	if (vp)
2673 		vput(vp);
2674 	return (error);
2675 }
2676 
2677 /*
2678  * nfs rmdir service
2679  */
2680 int
2681 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2682 	    struct thread *td, struct mbuf **mrq)
2683 {
2684 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2685 	struct sockaddr *nam = nfsd->nd_nam;
2686 	caddr_t dpos = nfsd->nd_dpos;
2687 	struct ucred *cred = &nfsd->nd_cr;
2688 	u_int32_t *tl;
2689 	int32_t t1;
2690 	caddr_t bpos;
2691 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2692 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2693 	char *cp2;
2694 	struct mbuf *mb, *mreq;
2695 	struct vnode *dirp;
2696 	struct vnode *vp;
2697 	struct vattr dirfor, diraft;
2698 	nfsfh_t nfh;
2699 	fhandle_t *fhp;
2700 	struct nlookupdata nd;
2701 	u_quad_t frev;
2702 
2703 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2704 	nlookup_zero(&nd);
2705 	dirp = NULL;
2706 	vp = NULL;
2707 
2708 	fhp = &nfh.fh_generic;
2709 	nfsm_srvmtofh(fhp);
2710 	nfsm_srvnamesiz(len);
2711 
2712 	error = nfs_namei(&nd, cred, NAMEI_DELETE, NULL, &vp,
2713 			  fhp, len, slp, nam, &md, &dpos, &dirp,
2714 			  td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2715 	if (dirp) {
2716 		if (v3)
2717 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
2718 	}
2719 	if (error) {
2720 		nfsm_reply(NFSX_WCCDATA(v3));
2721 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2722 		error = 0;
2723 		goto nfsmout;
2724 	}
2725 	if (vp->v_type != VDIR) {
2726 		error = ENOTDIR;
2727 		goto out;
2728 	}
2729 
2730 	/*
2731 	 * The root of a mounted filesystem cannot be deleted.
2732 	 */
2733 	if (vp->v_flag & VROOT)
2734 		error = EBUSY;
2735 out:
2736 	/*
2737 	 * Issue or abort op.  Since SAVESTART is not set, path name
2738 	 * component is freed by the VOP after either.
2739 	 */
2740 	if (!error) {
2741 #if 0
2742 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
2743 		nqsrv_getl(vp, ND_WRITE);
2744 #endif
2745 		vput(vp);
2746 		vp = NULL;
2747 		error = VOP_NRMDIR(nd.nl_ncp, nd.nl_cred);
2748 	}
2749 	nlookup_done(&nd);
2750 
2751 	if (dirp)
2752 		diraft_ret = VOP_GETATTR(dirp, &diraft, td);
2753 	nfsm_reply(NFSX_WCCDATA(v3));
2754 	if (v3) {
2755 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2756 		error = 0;
2757 	}
2758 	/* fall through */
2759 
2760 nfsmout:
2761 	nlookup_done(&nd);
2762 	if (dirp)
2763 		vrele(dirp);
2764 	if (vp)
2765 		vput(vp);
2766 	return(error);
2767 }
2768 
2769 /*
2770  * nfs readdir service
2771  * - mallocs what it thinks is enough to read
2772  *	count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
2773  * - calls VOP_READDIR()
2774  * - loops around building the reply
2775  *	if the output generated exceeds count break out of loop
2776  *	The nfsm_clget macro is used here so that the reply will be packed
2777  *	tightly in mbuf clusters.
2778  * - it only knows that it has encountered eof when the VOP_READDIR()
2779  *	reads nothing
2780  * - as such one readdir rpc will return eof false although you are there
2781  *	and then the next will return eof
2782  * - it trims out records with d_fileno == 0
2783  *	this doesn't matter for Unix clients, but they might confuse clients
2784  *	for other os'.
2785  * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2786  *	than requested, but this may not apply to all filesystems. For
2787  *	example, client NFS does not { although it is never remote mounted
2788  *	anyhow }
2789  *     The alternate call nfsrv_readdirplus() does lookups as well.
2790  * PS: The NFS protocol spec. does not clarify what the "count" byte
2791  *	argument is a count of.. just name strings and file id's or the
2792  *	entire reply rpc or ...
2793  *	I tried just file name and id sizes and it confused the Sun client,
2794  *	so I am using the full rpc size now. The "paranoia.." comment refers
2795  *	to including the status longwords that are not a part of the dir.
2796  *	"entry" structures, but are in the rpc.
2797  */
2798 struct flrep {
2799 	nfsuint64	fl_off;
2800 	u_int32_t	fl_postopok;
2801 	u_int32_t	fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2802 	u_int32_t	fl_fhok;
2803 	u_int32_t	fl_fhsize;
2804 	u_int32_t	fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
2805 };
2806 
2807 int
2808 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2809 	      struct thread *td, struct mbuf **mrq)
2810 {
2811 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2812 	struct sockaddr *nam = nfsd->nd_nam;
2813 	caddr_t dpos = nfsd->nd_dpos;
2814 	struct ucred *cred = &nfsd->nd_cr;
2815 	char *bp, *be;
2816 	struct mbuf *mp;
2817 	struct dirent *dp;
2818 	caddr_t cp;
2819 	u_int32_t *tl;
2820 	int32_t t1;
2821 	caddr_t bpos;
2822 	struct mbuf *mb, *mb2, *mreq, *mp2;
2823 	char *cpos, *cend, *cp2, *rbuf;
2824 	struct vnode *vp = NULL;
2825 	struct vattr at;
2826 	nfsfh_t nfh;
2827 	fhandle_t *fhp;
2828 	struct uio io;
2829 	struct iovec iv;
2830 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2831 	int siz, cnt, fullsiz, eofflag, rdonly, cache, ncookies;
2832 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2833 	u_quad_t frev, off, toff, verf;
2834 	u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
2835 
2836 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2837 	fhp = &nfh.fh_generic;
2838 	nfsm_srvmtofh(fhp);
2839 	if (v3) {
2840 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2841 		toff = fxdr_hyper(tl);
2842 		tl += 2;
2843 		verf = fxdr_hyper(tl);
2844 		tl += 2;
2845 	} else {
2846 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2847 		toff = fxdr_unsigned(u_quad_t, *tl++);
2848 		verf = 0;	/* shut up gcc */
2849 	}
2850 	off = toff;
2851 	cnt = fxdr_unsigned(int, *tl);
2852 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2853 	xfer = NFS_SRVMAXDATA(nfsd);
2854 	if (cnt > xfer)
2855 		cnt = xfer;
2856 	if (siz > xfer)
2857 		siz = xfer;
2858 	fullsiz = siz;
2859 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
2860 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
2861 	if (!error && vp->v_type != VDIR) {
2862 		error = ENOTDIR;
2863 		vput(vp);
2864 		vp = NULL;
2865 	}
2866 	if (error) {
2867 		nfsm_reply(NFSX_UNSIGNED);
2868 		nfsm_srvpostop_attr(getret, &at);
2869 		error = 0;
2870 		goto nfsmout;
2871 	}
2872 
2873 	/*
2874 	 * Obtain lock on vnode for this section of the code
2875 	 */
2876 
2877 	nqsrv_getl(vp, ND_READ);
2878 	if (v3) {
2879 		error = getret = VOP_GETATTR(vp, &at, td);
2880 #if 0
2881 		/*
2882 		 * XXX This check may be too strict for Solaris 2.5 clients.
2883 		 */
2884 		if (!error && toff && verf && verf != at.va_filerev)
2885 			error = NFSERR_BAD_COOKIE;
2886 #endif
2887 	}
2888 	if (!error)
2889 		error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
2890 	if (error) {
2891 		vput(vp);
2892 		vp = NULL;
2893 		nfsm_reply(NFSX_POSTOPATTR(v3));
2894 		nfsm_srvpostop_attr(getret, &at);
2895 		error = 0;
2896 		goto nfsmout;
2897 	}
2898 	VOP_UNLOCK(vp, 0, td);
2899 
2900 	/*
2901 	 * end section.  Allocate rbuf and continue
2902 	 */
2903 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
2904 again:
2905 	iv.iov_base = rbuf;
2906 	iv.iov_len = fullsiz;
2907 	io.uio_iov = &iv;
2908 	io.uio_iovcnt = 1;
2909 	io.uio_offset = (off_t)off;
2910 	io.uio_resid = fullsiz;
2911 	io.uio_segflg = UIO_SYSSPACE;
2912 	io.uio_rw = UIO_READ;
2913 	io.uio_td = NULL;
2914 	eofflag = 0;
2915 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
2916 	if (cookies) {
2917 		free((caddr_t)cookies, M_TEMP);
2918 		cookies = NULL;
2919 	}
2920 	error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
2921 	off = (off_t)io.uio_offset;
2922 	if (!cookies && !error)
2923 		error = NFSERR_PERM;
2924 	if (v3) {
2925 		getret = VOP_GETATTR(vp, &at, td);
2926 		if (!error)
2927 			error = getret;
2928 	}
2929 	VOP_UNLOCK(vp, 0, td);
2930 	if (error) {
2931 		vrele(vp);
2932 		vp = NULL;
2933 		free((caddr_t)rbuf, M_TEMP);
2934 		if (cookies)
2935 			free((caddr_t)cookies, M_TEMP);
2936 		nfsm_reply(NFSX_POSTOPATTR(v3));
2937 		nfsm_srvpostop_attr(getret, &at);
2938 		error = 0;
2939 		goto nfsmout;
2940 	}
2941 	if (io.uio_resid) {
2942 		siz -= io.uio_resid;
2943 
2944 		/*
2945 		 * If nothing read, return eof
2946 		 * rpc reply
2947 		 */
2948 		if (siz == 0) {
2949 			vrele(vp);
2950 			vp = NULL;
2951 			nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2952 				2 * NFSX_UNSIGNED);
2953 			if (v3) {
2954 				nfsm_srvpostop_attr(getret, &at);
2955 				nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2956 				txdr_hyper(at.va_filerev, tl);
2957 				tl += 2;
2958 			} else
2959 				nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2960 			*tl++ = nfs_false;
2961 			*tl = nfs_true;
2962 			FREE((caddr_t)rbuf, M_TEMP);
2963 			FREE((caddr_t)cookies, M_TEMP);
2964 			error = 0;
2965 			goto nfsmout;
2966 		}
2967 	}
2968 
2969 	/*
2970 	 * Check for degenerate cases of nothing useful read.
2971 	 * If so go try again
2972 	 */
2973 	cpos = rbuf;
2974 	cend = rbuf + siz;
2975 	dp = (struct dirent *)cpos;
2976 	cookiep = cookies;
2977 	/*
2978 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2979 	 * directory offset up to a block boundary, so it is necessary to
2980 	 * skip over the records that preceed the requested offset. This
2981 	 * requires the assumption that file offset cookies monotonically
2982 	 * increase.
2983 	 */
2984 	while (cpos < cend && ncookies > 0 &&
2985 		(dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2986 		 ((u_quad_t)(*cookiep)) <= toff)) {
2987 		cpos += dp->d_reclen;
2988 		dp = (struct dirent *)cpos;
2989 		cookiep++;
2990 		ncookies--;
2991 	}
2992 	if (cpos >= cend || ncookies == 0) {
2993 		toff = off;
2994 		siz = fullsiz;
2995 		goto again;
2996 	}
2997 
2998 	len = 3 * NFSX_UNSIGNED;	/* paranoia, probably can be 0 */
2999 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
3000 	if (v3) {
3001 		nfsm_srvpostop_attr(getret, &at);
3002 		nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3003 		txdr_hyper(at.va_filerev, tl);
3004 	}
3005 	mp = mp2 = mb;
3006 	bp = bpos;
3007 	be = bp + M_TRAILINGSPACE(mp);
3008 
3009 	/* Loop through the records and build reply */
3010 	while (cpos < cend && ncookies > 0) {
3011 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3012 			nlen = dp->d_namlen;
3013 			rem = nfsm_rndup(nlen) - nlen;
3014 			len += (4 * NFSX_UNSIGNED + nlen + rem);
3015 			if (v3)
3016 				len += 2 * NFSX_UNSIGNED;
3017 			if (len > cnt) {
3018 				eofflag = 0;
3019 				break;
3020 			}
3021 			/*
3022 			 * Build the directory record xdr from
3023 			 * the dirent entry.
3024 			 */
3025 			nfsm_clget;
3026 			*tl = nfs_true;
3027 			bp += NFSX_UNSIGNED;
3028 			if (v3) {
3029 				nfsm_clget;
3030 				*tl = 0;
3031 				bp += NFSX_UNSIGNED;
3032 			}
3033 			nfsm_clget;
3034 			*tl = txdr_unsigned(dp->d_fileno);
3035 			bp += NFSX_UNSIGNED;
3036 			nfsm_clget;
3037 			*tl = txdr_unsigned(nlen);
3038 			bp += NFSX_UNSIGNED;
3039 
3040 			/* And loop around copying the name */
3041 			xfer = nlen;
3042 			cp = dp->d_name;
3043 			while (xfer > 0) {
3044 				nfsm_clget;
3045 				if ((bp+xfer) > be)
3046 					tsiz = be-bp;
3047 				else
3048 					tsiz = xfer;
3049 				bcopy(cp, bp, tsiz);
3050 				bp += tsiz;
3051 				xfer -= tsiz;
3052 				if (xfer > 0)
3053 					cp += tsiz;
3054 			}
3055 			/* And null pad to a int32_t boundary */
3056 			for (i = 0; i < rem; i++)
3057 				*bp++ = '\0';
3058 			nfsm_clget;
3059 
3060 			/* Finish off the record */
3061 			if (v3) {
3062 				*tl = 0;
3063 				bp += NFSX_UNSIGNED;
3064 				nfsm_clget;
3065 			}
3066 			*tl = txdr_unsigned(*cookiep);
3067 			bp += NFSX_UNSIGNED;
3068 		}
3069 		cpos += dp->d_reclen;
3070 		dp = (struct dirent *)cpos;
3071 		cookiep++;
3072 		ncookies--;
3073 	}
3074 	vrele(vp);
3075 	vp = NULL;
3076 	nfsm_clget;
3077 	*tl = nfs_false;
3078 	bp += NFSX_UNSIGNED;
3079 	nfsm_clget;
3080 	if (eofflag)
3081 		*tl = nfs_true;
3082 	else
3083 		*tl = nfs_false;
3084 	bp += NFSX_UNSIGNED;
3085 	if (mp != mb) {
3086 		if (bp < be)
3087 			mp->m_len = bp - mtod(mp, caddr_t);
3088 	} else
3089 		mp->m_len += bp - bpos;
3090 	FREE((caddr_t)rbuf, M_TEMP);
3091 	FREE((caddr_t)cookies, M_TEMP);
3092 
3093 nfsmout:
3094 	if (vp)
3095 		vrele(vp);
3096 	return(error);
3097 }
3098 
3099 int
3100 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3101 		  struct thread *td, struct mbuf **mrq)
3102 {
3103 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3104 	struct sockaddr *nam = nfsd->nd_nam;
3105 	caddr_t dpos = nfsd->nd_dpos;
3106 	struct ucred *cred = &nfsd->nd_cr;
3107 	char *bp, *be;
3108 	struct mbuf *mp;
3109 	struct dirent *dp;
3110 	caddr_t cp;
3111 	u_int32_t *tl;
3112 	int32_t t1;
3113 	caddr_t bpos;
3114 	struct mbuf *mb, *mb2, *mreq, *mp2;
3115 	char *cpos, *cend, *cp2, *rbuf;
3116 	struct vnode *vp = NULL, *nvp;
3117 	struct flrep fl;
3118 	nfsfh_t nfh;
3119 	fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
3120 	struct uio io;
3121 	struct iovec iv;
3122 	struct vattr va, at, *vap = &va;
3123 	struct nfs_fattr *fp;
3124 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3125 	int siz, cnt, fullsiz, eofflag, rdonly, cache, dirlen, ncookies;
3126 	u_quad_t frev, off, toff, verf;
3127 	u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3128 
3129 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3130 	fhp = &nfh.fh_generic;
3131 	nfsm_srvmtofh(fhp);
3132 	nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
3133 	toff = fxdr_hyper(tl);
3134 	tl += 2;
3135 	verf = fxdr_hyper(tl);
3136 	tl += 2;
3137 	siz = fxdr_unsigned(int, *tl++);
3138 	cnt = fxdr_unsigned(int, *tl);
3139 	off = toff;
3140 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3141 	xfer = NFS_SRVMAXDATA(nfsd);
3142 	if (cnt > xfer)
3143 		cnt = xfer;
3144 	if (siz > xfer)
3145 		siz = xfer;
3146 	fullsiz = siz;
3147 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3148 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3149 	if (!error && vp->v_type != VDIR) {
3150 		error = ENOTDIR;
3151 		vput(vp);
3152 		vp = NULL;
3153 	}
3154 	if (error) {
3155 		nfsm_reply(NFSX_UNSIGNED);
3156 		nfsm_srvpostop_attr(getret, &at);
3157 		error = 0;
3158 		goto nfsmout;
3159 	}
3160 	error = getret = VOP_GETATTR(vp, &at, td);
3161 #if 0
3162 	/*
3163 	 * XXX This check may be too strict for Solaris 2.5 clients.
3164 	 */
3165 	if (!error && toff && verf && verf != at.va_filerev)
3166 		error = NFSERR_BAD_COOKIE;
3167 #endif
3168 	if (!error) {
3169 		nqsrv_getl(vp, ND_READ);
3170 		error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
3171 	}
3172 	if (error) {
3173 		vput(vp);
3174 		vp = NULL;
3175 		nfsm_reply(NFSX_V3POSTOPATTR);
3176 		nfsm_srvpostop_attr(getret, &at);
3177 		error = 0;
3178 		goto nfsmout;
3179 	}
3180 	VOP_UNLOCK(vp, 0, td);
3181 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3182 again:
3183 	iv.iov_base = rbuf;
3184 	iv.iov_len = fullsiz;
3185 	io.uio_iov = &iv;
3186 	io.uio_iovcnt = 1;
3187 	io.uio_offset = (off_t)off;
3188 	io.uio_resid = fullsiz;
3189 	io.uio_segflg = UIO_SYSSPACE;
3190 	io.uio_rw = UIO_READ;
3191 	io.uio_td = NULL;
3192 	eofflag = 0;
3193 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
3194 	if (cookies) {
3195 		free((caddr_t)cookies, M_TEMP);
3196 		cookies = NULL;
3197 	}
3198 	error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3199 	off = (u_quad_t)io.uio_offset;
3200 	getret = VOP_GETATTR(vp, &at, td);
3201 	VOP_UNLOCK(vp, 0, td);
3202 	if (!cookies && !error)
3203 		error = NFSERR_PERM;
3204 	if (!error)
3205 		error = getret;
3206 	if (error) {
3207 		vrele(vp);
3208 		vp = NULL;
3209 		if (cookies)
3210 			free((caddr_t)cookies, M_TEMP);
3211 		free((caddr_t)rbuf, M_TEMP);
3212 		nfsm_reply(NFSX_V3POSTOPATTR);
3213 		nfsm_srvpostop_attr(getret, &at);
3214 		error = 0;
3215 		goto nfsmout;
3216 	}
3217 	if (io.uio_resid) {
3218 		siz -= io.uio_resid;
3219 
3220 		/*
3221 		 * If nothing read, return eof
3222 		 * rpc reply
3223 		 */
3224 		if (siz == 0) {
3225 			vrele(vp);
3226 			vp = NULL;
3227 			nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3228 				2 * NFSX_UNSIGNED);
3229 			nfsm_srvpostop_attr(getret, &at);
3230 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3231 			txdr_hyper(at.va_filerev, tl);
3232 			tl += 2;
3233 			*tl++ = nfs_false;
3234 			*tl = nfs_true;
3235 			FREE((caddr_t)cookies, M_TEMP);
3236 			FREE((caddr_t)rbuf, M_TEMP);
3237 			error = 0;
3238 			goto nfsmout;
3239 		}
3240 	}
3241 
3242 	/*
3243 	 * Check for degenerate cases of nothing useful read.
3244 	 * If so go try again
3245 	 */
3246 	cpos = rbuf;
3247 	cend = rbuf + siz;
3248 	dp = (struct dirent *)cpos;
3249 	cookiep = cookies;
3250 	/*
3251 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
3252 	 * directory offset up to a block boundary, so it is necessary to
3253 	 * skip over the records that preceed the requested offset. This
3254 	 * requires the assumption that file offset cookies monotonically
3255 	 * increase.
3256 	 */
3257 	while (cpos < cend && ncookies > 0 &&
3258 		(dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3259 		 ((u_quad_t)(*cookiep)) <= toff)) {
3260 		cpos += dp->d_reclen;
3261 		dp = (struct dirent *)cpos;
3262 		cookiep++;
3263 		ncookies--;
3264 	}
3265 	if (cpos >= cend || ncookies == 0) {
3266 		toff = off;
3267 		siz = fullsiz;
3268 		goto again;
3269 	}
3270 
3271 	/*
3272 	 * Probe one of the directory entries to see if the filesystem
3273 	 * supports VGET.
3274 	 */
3275 	if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp) == EOPNOTSUPP) {
3276 		error = NFSERR_NOTSUPP;
3277 		vrele(vp);
3278 		vp = NULL;
3279 		free((caddr_t)cookies, M_TEMP);
3280 		free((caddr_t)rbuf, M_TEMP);
3281 		nfsm_reply(NFSX_V3POSTOPATTR);
3282 		nfsm_srvpostop_attr(getret, &at);
3283 		error = 0;
3284 		goto nfsmout;
3285 	}
3286 	if (nvp) {
3287 		vput(nvp);
3288 		nvp = NULL;
3289 	}
3290 
3291 	dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
3292 	nfsm_reply(cnt);
3293 	nfsm_srvpostop_attr(getret, &at);
3294 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3295 	txdr_hyper(at.va_filerev, tl);
3296 	mp = mp2 = mb;
3297 	bp = bpos;
3298 	be = bp + M_TRAILINGSPACE(mp);
3299 
3300 	/* Loop through the records and build reply */
3301 	while (cpos < cend && ncookies > 0) {
3302 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3303 			nlen = dp->d_namlen;
3304 			rem = nfsm_rndup(nlen)-nlen;
3305 
3306 			/*
3307 			 * For readdir_and_lookup get the vnode using
3308 			 * the file number.
3309 			 */
3310 			if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp))
3311 				goto invalid;
3312 			bzero((caddr_t)nfhp, NFSX_V3FH);
3313 			nfhp->fh_fsid =
3314 				nvp->v_mount->mnt_stat.f_fsid;
3315 			if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) {
3316 				vput(nvp);
3317 				nvp = NULL;
3318 				goto invalid;
3319 			}
3320 			if (VOP_GETATTR(nvp, vap, td)) {
3321 				vput(nvp);
3322 				nvp = NULL;
3323 				goto invalid;
3324 			}
3325 			vput(nvp);
3326 			nvp = NULL;
3327 
3328 			/*
3329 			 * If either the dircount or maxcount will be
3330 			 * exceeded, get out now. Both of these lengths
3331 			 * are calculated conservatively, including all
3332 			 * XDR overheads.
3333 			 */
3334 			len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3335 				NFSX_V3POSTOPATTR);
3336 			dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3337 			if (len > cnt || dirlen > fullsiz) {
3338 				eofflag = 0;
3339 				break;
3340 			}
3341 
3342 			/*
3343 			 * Build the directory record xdr from
3344 			 * the dirent entry.
3345 			 */
3346 			fp = (struct nfs_fattr *)&fl.fl_fattr;
3347 			nfsm_srvfillattr(vap, fp);
3348 			fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3349 			fl.fl_fhok = nfs_true;
3350 			fl.fl_postopok = nfs_true;
3351 			fl.fl_off.nfsuquad[0] = 0;
3352 			fl.fl_off.nfsuquad[1] = txdr_unsigned(*cookiep);
3353 
3354 			nfsm_clget;
3355 			*tl = nfs_true;
3356 			bp += NFSX_UNSIGNED;
3357 			nfsm_clget;
3358 			*tl = 0;
3359 			bp += NFSX_UNSIGNED;
3360 			nfsm_clget;
3361 			*tl = txdr_unsigned(dp->d_fileno);
3362 			bp += NFSX_UNSIGNED;
3363 			nfsm_clget;
3364 			*tl = txdr_unsigned(nlen);
3365 			bp += NFSX_UNSIGNED;
3366 
3367 			/* And loop around copying the name */
3368 			xfer = nlen;
3369 			cp = dp->d_name;
3370 			while (xfer > 0) {
3371 				nfsm_clget;
3372 				if ((bp + xfer) > be)
3373 					tsiz = be - bp;
3374 				else
3375 					tsiz = xfer;
3376 				bcopy(cp, bp, tsiz);
3377 				bp += tsiz;
3378 				xfer -= tsiz;
3379 				if (xfer > 0)
3380 					cp += tsiz;
3381 			}
3382 			/* And null pad to a int32_t boundary */
3383 			for (i = 0; i < rem; i++)
3384 				*bp++ = '\0';
3385 
3386 			/*
3387 			 * Now copy the flrep structure out.
3388 			 */
3389 			xfer = sizeof (struct flrep);
3390 			cp = (caddr_t)&fl;
3391 			while (xfer > 0) {
3392 				nfsm_clget;
3393 				if ((bp + xfer) > be)
3394 					tsiz = be - bp;
3395 				else
3396 					tsiz = xfer;
3397 				bcopy(cp, bp, tsiz);
3398 				bp += tsiz;
3399 				xfer -= tsiz;
3400 				if (xfer > 0)
3401 					cp += tsiz;
3402 			}
3403 		}
3404 invalid:
3405 		cpos += dp->d_reclen;
3406 		dp = (struct dirent *)cpos;
3407 		cookiep++;
3408 		ncookies--;
3409 	}
3410 	vrele(vp);
3411 	vp = NULL;
3412 	nfsm_clget;
3413 	*tl = nfs_false;
3414 	bp += NFSX_UNSIGNED;
3415 	nfsm_clget;
3416 	if (eofflag)
3417 		*tl = nfs_true;
3418 	else
3419 		*tl = nfs_false;
3420 	bp += NFSX_UNSIGNED;
3421 	if (mp != mb) {
3422 		if (bp < be)
3423 			mp->m_len = bp - mtod(mp, caddr_t);
3424 	} else
3425 		mp->m_len += bp - bpos;
3426 	FREE((caddr_t)cookies, M_TEMP);
3427 	FREE((caddr_t)rbuf, M_TEMP);
3428 nfsmout:
3429 	if (vp)
3430 		vrele(vp);
3431 	return(error);
3432 }
3433 
3434 /*
3435  * nfs commit service
3436  */
3437 int
3438 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3439 	     struct thread *td, struct mbuf **mrq)
3440 {
3441 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3442 	struct sockaddr *nam = nfsd->nd_nam;
3443 	caddr_t dpos = nfsd->nd_dpos;
3444 	struct ucred *cred = &nfsd->nd_cr;
3445 	struct vattr bfor, aft;
3446 	struct vnode *vp = NULL;
3447 	nfsfh_t nfh;
3448 	fhandle_t *fhp;
3449 	u_int32_t *tl;
3450 	int32_t t1;
3451 	caddr_t bpos;
3452 	int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt, cache;
3453 	char *cp2;
3454 	struct mbuf *mb, *mb2, *mreq;
3455 	u_quad_t frev, off;
3456 
3457 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3458 #ifndef nolint
3459 	cache = 0;
3460 #endif
3461 	fhp = &nfh.fh_generic;
3462 	nfsm_srvmtofh(fhp);
3463 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3464 
3465 	/*
3466 	 * XXX At this time VOP_FSYNC() does not accept offset and byte
3467 	 * count parameters, so these arguments are useless (someday maybe).
3468 	 */
3469 	off = fxdr_hyper(tl);
3470 	tl += 2;
3471 	cnt = fxdr_unsigned(int, *tl);
3472 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3473 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3474 	if (error) {
3475 		nfsm_reply(2 * NFSX_UNSIGNED);
3476 		nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3477 		error = 0;
3478 		goto nfsmout;
3479 	}
3480 	for_ret = VOP_GETATTR(vp, &bfor, td);
3481 
3482 	if (cnt > MAX_COMMIT_COUNT) {
3483 		/*
3484 		 * Give up and do the whole thing
3485 		 */
3486 		if (vp->v_object &&
3487 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3488 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3489 		}
3490 		error = VOP_FSYNC(vp, MNT_WAIT, td);
3491 	} else {
3492 		/*
3493 		 * Locate and synchronously write any buffers that fall
3494 		 * into the requested range.  Note:  we are assuming that
3495 		 * f_iosize is a power of 2.
3496 		 */
3497 		int iosize = vp->v_mount->mnt_stat.f_iosize;
3498 		int iomask = iosize - 1;
3499 		int s;
3500 		daddr_t lblkno;
3501 
3502 		/*
3503 		 * Align to iosize boundry, super-align to page boundry.
3504 		 */
3505 		if (off & iomask) {
3506 			cnt += off & iomask;
3507 			off &= ~(u_quad_t)iomask;
3508 		}
3509 		if (off & PAGE_MASK) {
3510 			cnt += off & PAGE_MASK;
3511 			off &= ~(u_quad_t)PAGE_MASK;
3512 		}
3513 		lblkno = off / iosize;
3514 
3515 		if (vp->v_object &&
3516 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3517 			vm_object_page_clean(vp->v_object, off / PAGE_SIZE, (cnt + PAGE_MASK) / PAGE_SIZE, OBJPC_SYNC);
3518 		}
3519 
3520 		s = splbio();
3521 		while (cnt > 0) {
3522 			struct buf *bp;
3523 
3524 			/*
3525 			 * If we have a buffer and it is marked B_DELWRI we
3526 			 * have to lock and write it.  Otherwise the prior
3527 			 * write is assumed to have already been committed.
3528 			 */
3529 			if ((bp = gbincore(vp, lblkno)) != NULL && (bp->b_flags & B_DELWRI)) {
3530 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
3531 					BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL);
3532 					continue; /* retry */
3533 				}
3534 				bremfree(bp);
3535 				bp->b_flags &= ~B_ASYNC;
3536 				VOP_BWRITE(bp->b_vp, bp);
3537 				++nfs_commit_miss;
3538 			}
3539 			++nfs_commit_blks;
3540 			if (cnt < iosize)
3541 				break;
3542 			cnt -= iosize;
3543 			++lblkno;
3544 		}
3545 		splx(s);
3546 	}
3547 
3548 	aft_ret = VOP_GETATTR(vp, &aft, td);
3549 	vput(vp);
3550 	vp = NULL;
3551 	nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3552 	nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3553 	if (!error) {
3554 		nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3555 		if (nfsver.tv_sec == 0)
3556 			nfsver = boottime;
3557 		*tl++ = txdr_unsigned(nfsver.tv_sec);
3558 		*tl = txdr_unsigned(nfsver.tv_nsec / 1000);
3559 	} else {
3560 		error = 0;
3561 	}
3562 nfsmout:
3563 	if (vp)
3564 		vput(vp);
3565 	return(error);
3566 }
3567 
3568 /*
3569  * nfs statfs service
3570  */
3571 int
3572 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3573 	     struct thread *td, struct mbuf **mrq)
3574 {
3575 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3576 	struct sockaddr *nam = nfsd->nd_nam;
3577 	caddr_t dpos = nfsd->nd_dpos;
3578 	struct ucred *cred = &nfsd->nd_cr;
3579 	struct statfs *sf;
3580 	struct nfs_statfs *sfp;
3581 	u_int32_t *tl;
3582 	int32_t t1;
3583 	caddr_t bpos;
3584 	int error = 0, rdonly, cache, getret = 1;
3585 	int v3 = (nfsd->nd_flag & ND_NFSV3);
3586 	char *cp2;
3587 	struct mbuf *mb, *mb2, *mreq;
3588 	struct vnode *vp = NULL;
3589 	struct vattr at;
3590 	nfsfh_t nfh;
3591 	fhandle_t *fhp;
3592 	struct statfs statfs;
3593 	u_quad_t frev, tval;
3594 
3595 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3596 #ifndef nolint
3597 	cache = 0;
3598 #endif
3599 	fhp = &nfh.fh_generic;
3600 	nfsm_srvmtofh(fhp);
3601 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3602 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3603 	if (error) {
3604 		nfsm_reply(NFSX_UNSIGNED);
3605 		nfsm_srvpostop_attr(getret, &at);
3606 		error = 0;
3607 		goto nfsmout;
3608 	}
3609 	sf = &statfs;
3610 	error = VFS_STATFS(vp->v_mount, sf, td);
3611 	getret = VOP_GETATTR(vp, &at, td);
3612 	vput(vp);
3613 	vp = NULL;
3614 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3615 	if (v3)
3616 		nfsm_srvpostop_attr(getret, &at);
3617 	if (error) {
3618 		error = 0;
3619 		goto nfsmout;
3620 	}
3621 	nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3622 	if (v3) {
3623 		tval = (u_quad_t)sf->f_blocks;
3624 		tval *= (u_quad_t)sf->f_bsize;
3625 		txdr_hyper(tval, &sfp->sf_tbytes);
3626 		tval = (u_quad_t)sf->f_bfree;
3627 		tval *= (u_quad_t)sf->f_bsize;
3628 		txdr_hyper(tval, &sfp->sf_fbytes);
3629 		tval = (u_quad_t)sf->f_bavail;
3630 		tval *= (u_quad_t)sf->f_bsize;
3631 		txdr_hyper(tval, &sfp->sf_abytes);
3632 		sfp->sf_tfiles.nfsuquad[0] = 0;
3633 		sfp->sf_tfiles.nfsuquad[1] = txdr_unsigned(sf->f_files);
3634 		sfp->sf_ffiles.nfsuquad[0] = 0;
3635 		sfp->sf_ffiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3636 		sfp->sf_afiles.nfsuquad[0] = 0;
3637 		sfp->sf_afiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3638 		sfp->sf_invarsec = 0;
3639 	} else {
3640 		sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3641 		sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
3642 		sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3643 		sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3644 		sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3645 	}
3646 nfsmout:
3647 	if (vp)
3648 		vput(vp);
3649 	return(error);
3650 }
3651 
3652 /*
3653  * nfs fsinfo service
3654  */
3655 int
3656 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3657 	     struct thread *td, struct mbuf **mrq)
3658 {
3659 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3660 	struct sockaddr *nam = nfsd->nd_nam;
3661 	caddr_t dpos = nfsd->nd_dpos;
3662 	struct ucred *cred = &nfsd->nd_cr;
3663 	u_int32_t *tl;
3664 	struct nfsv3_fsinfo *sip;
3665 	int32_t t1;
3666 	caddr_t bpos;
3667 	int error = 0, rdonly, cache, getret = 1, pref;
3668 	char *cp2;
3669 	struct mbuf *mb, *mb2, *mreq;
3670 	struct vnode *vp = NULL;
3671 	struct vattr at;
3672 	nfsfh_t nfh;
3673 	fhandle_t *fhp;
3674 	u_quad_t frev, maxfsize;
3675 	struct statfs sb;
3676 
3677 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3678 #ifndef nolint
3679 	cache = 0;
3680 #endif
3681 	fhp = &nfh.fh_generic;
3682 	nfsm_srvmtofh(fhp);
3683 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3684 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3685 	if (error) {
3686 		nfsm_reply(NFSX_UNSIGNED);
3687 		nfsm_srvpostop_attr(getret, &at);
3688 		error = 0;
3689 		goto nfsmout;
3690 	}
3691 
3692 	/* XXX Try to make a guess on the max file size. */
3693 	VFS_STATFS(vp->v_mount, &sb, td);
3694 	maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
3695 
3696 	getret = VOP_GETATTR(vp, &at, td);
3697 	vput(vp);
3698 	vp = NULL;
3699 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3700 	nfsm_srvpostop_attr(getret, &at);
3701 	nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3702 
3703 	/*
3704 	 * XXX
3705 	 * There should be file system VFS OP(s) to get this information.
3706 	 * For now, assume ufs.
3707 	 */
3708 	if (slp->ns_so->so_type == SOCK_DGRAM)
3709 		pref = NFS_MAXDGRAMDATA;
3710 	else
3711 		pref = NFS_MAXDATA;
3712 	sip->fs_rtmax = txdr_unsigned(NFS_MAXDATA);
3713 	sip->fs_rtpref = txdr_unsigned(pref);
3714 	sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3715 	sip->fs_wtmax = txdr_unsigned(NFS_MAXDATA);
3716 	sip->fs_wtpref = txdr_unsigned(pref);
3717 	sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3718 	sip->fs_dtpref = txdr_unsigned(pref);
3719 	txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3720 	sip->fs_timedelta.nfsv3_sec = 0;
3721 	sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3722 	sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3723 		NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3724 		NFSV3FSINFO_CANSETTIME);
3725 nfsmout:
3726 	if (vp)
3727 		vput(vp);
3728 	return(error);
3729 }
3730 
3731 /*
3732  * nfs pathconf service
3733  */
3734 int
3735 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3736 	       struct thread *td, struct mbuf **mrq)
3737 {
3738 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3739 	struct sockaddr *nam = nfsd->nd_nam;
3740 	caddr_t dpos = nfsd->nd_dpos;
3741 	struct ucred *cred = &nfsd->nd_cr;
3742 	u_int32_t *tl;
3743 	struct nfsv3_pathconf *pc;
3744 	int32_t t1;
3745 	caddr_t bpos;
3746 	int error = 0, rdonly, cache, getret = 1;
3747 	register_t linkmax, namemax, chownres, notrunc;
3748 	char *cp2;
3749 	struct mbuf *mb, *mb2, *mreq;
3750 	struct vnode *vp = NULL;
3751 	struct vattr at;
3752 	nfsfh_t nfh;
3753 	fhandle_t *fhp;
3754 	u_quad_t frev;
3755 
3756 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3757 #ifndef nolint
3758 	cache = 0;
3759 #endif
3760 	fhp = &nfh.fh_generic;
3761 	nfsm_srvmtofh(fhp);
3762 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3763 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3764 	if (error) {
3765 		nfsm_reply(NFSX_UNSIGNED);
3766 		nfsm_srvpostop_attr(getret, &at);
3767 		error = 0;
3768 		goto nfsmout;
3769 	}
3770 	error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3771 	if (!error)
3772 		error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3773 	if (!error)
3774 		error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3775 	if (!error)
3776 		error = VOP_PATHCONF(vp, _PC_NO_TRUNC, &notrunc);
3777 	getret = VOP_GETATTR(vp, &at, td);
3778 	vput(vp);
3779 	vp = NULL;
3780 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3781 	nfsm_srvpostop_attr(getret, &at);
3782 	if (error) {
3783 		error = 0;
3784 		goto nfsmout;
3785 	}
3786 	nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3787 
3788 	pc->pc_linkmax = txdr_unsigned(linkmax);
3789 	pc->pc_namemax = txdr_unsigned(namemax);
3790 	pc->pc_notrunc = txdr_unsigned(notrunc);
3791 	pc->pc_chownrestricted = txdr_unsigned(chownres);
3792 
3793 	/*
3794 	 * These should probably be supported by VOP_PATHCONF(), but
3795 	 * until msdosfs is exportable (why would you want to?), the
3796 	 * Unix defaults should be ok.
3797 	 */
3798 	pc->pc_caseinsensitive = nfs_false;
3799 	pc->pc_casepreserving = nfs_true;
3800 nfsmout:
3801 	if (vp)
3802 		vput(vp);
3803 	return(error);
3804 }
3805 
3806 /*
3807  * Null operation, used by clients to ping server
3808  */
3809 /* ARGSUSED */
3810 int
3811 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3812 	   struct thread *td, struct mbuf **mrq)
3813 {
3814 	struct mbuf *mrep = nfsd->nd_mrep;
3815 	caddr_t bpos;
3816 	int error = NFSERR_RETVOID, cache;
3817 	struct mbuf *mb, *mreq;
3818 	u_quad_t frev;
3819 
3820 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3821 #ifndef nolint
3822 	cache = 0;
3823 #endif
3824 	nfsm_reply(0);
3825 	nfsm_srvdone;
3826 }
3827 
3828 /*
3829  * No operation, used for obsolete procedures
3830  */
3831 /* ARGSUSED */
3832 int
3833 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3834 	   struct thread *td, struct mbuf **mrq)
3835 {
3836 	struct mbuf *mrep = nfsd->nd_mrep;
3837 	caddr_t bpos;
3838 	int error, cache;
3839 	struct mbuf *mb, *mreq;
3840 	u_quad_t frev;
3841 
3842 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3843 #ifndef nolint
3844 	cache = 0;
3845 #endif
3846 	if (nfsd->nd_repstat)
3847 		error = nfsd->nd_repstat;
3848 	else
3849 		error = EPROCUNAVAIL;
3850 	nfsm_reply(0);
3851 	error = 0;
3852 	nfsm_srvdone;
3853 }
3854 
3855 /*
3856  * Perform access checking for vnodes obtained from file handles that would
3857  * refer to files already opened by a Unix client. You cannot just use
3858  * vn_writechk() and VOP_ACCESS() for two reasons.
3859  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3860  * 2 - The owner is to be given access irrespective of mode bits for some
3861  *     operations, so that processes that chmod after opening a file don't
3862  *     break. I don't like this because it opens a security hole, but since
3863  *     the nfs server opens a security hole the size of a barn door anyhow,
3864  *     what the heck.
3865  *
3866  * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3867  * will return EPERM instead of EACCESS. EPERM is always an error.
3868  */
3869 static int
3870 nfsrv_access(struct vnode *vp, int flags, struct ucred *cred,
3871 	     int rdonly, struct thread *td, int override)
3872 {
3873 	struct vattr vattr;
3874 	int error;
3875 
3876 	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3877 	if (flags & VWRITE) {
3878 		/* Just vn_writechk() changed to check rdonly */
3879 		/*
3880 		 * Disallow write attempts on read-only file systems;
3881 		 * unless the file is a socket or a block or character
3882 		 * device resident on the file system.
3883 		 */
3884 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3885 			switch (vp->v_type) {
3886 			case VREG:
3887 			case VDIR:
3888 			case VLNK:
3889 				return (EROFS);
3890 			default:
3891 				break;
3892 			}
3893 		}
3894 		/*
3895 		 * If there's shared text associated with
3896 		 * the inode, we can't allow writing.
3897 		 */
3898 		if (vp->v_flag & VTEXT)
3899 			return (ETXTBSY);
3900 	}
3901 	error = VOP_GETATTR(vp, &vattr, td);
3902 	if (error)
3903 		return (error);
3904 	error = VOP_ACCESS(vp, flags, cred, td);
3905 	/*
3906 	 * Allow certain operations for the owner (reads and writes
3907 	 * on files that are already open).
3908 	 */
3909 	if (override && error == EACCES && cred->cr_uid == vattr.va_uid)
3910 		error = 0;
3911 	return error;
3912 }
3913 #endif /* NFS_NOSERVER */
3914 
3915