xref: /netbsd-src/sys/nfs/nfs_serv.c (revision 3b01aba77a7a698587faaae455bbfe740923c1f5)
1 /*	$NetBSD: nfs_serv.c,v 1.60 2001/07/24 15:39:33 assar Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)nfs_serv.c	8.8 (Berkeley) 7/31/95
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 
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/proc.h>
64 #include <sys/file.h>
65 #include <sys/namei.h>
66 #include <sys/vnode.h>
67 #include <sys/mount.h>
68 #include <sys/socket.h>
69 #include <sys/socketvar.h>
70 #include <sys/mbuf.h>
71 #include <sys/dirent.h>
72 #include <sys/stat.h>
73 #include <sys/kernel.h>
74 #include <ufs/ufs/dir.h>
75 
76 #include <uvm/uvm_extern.h>
77 
78 #include <nfs/nfsproto.h>
79 #include <nfs/rpcv2.h>
80 #include <nfs/nfs.h>
81 #include <nfs/xdr_subs.h>
82 #include <nfs/nfsm_subs.h>
83 #include <nfs/nqnfs.h>
84 #include <nfs/nfs_var.h>
85 
86 /* Global vars */
87 extern u_int32_t nfs_xdrneg1;
88 extern u_int32_t nfs_false, nfs_true;
89 extern enum vtype nv3tov_type[8];
90 extern struct nfsstats nfsstats;
91 extern nfstype nfsv2_type[9];
92 extern nfstype nfsv3_type[9];
93 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
94 
95 /*
96  * nfs v3 access service
97  */
98 int
99 nfsrv3_access(nfsd, slp, procp, mrq)
100 	struct nfsrv_descript *nfsd;
101 	struct nfssvc_sock *slp;
102 	struct proc *procp;
103 	struct mbuf **mrq;
104 {
105 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
106 	struct mbuf *nam = nfsd->nd_nam;
107 	caddr_t dpos = nfsd->nd_dpos;
108 	struct ucred *cred = &nfsd->nd_cr;
109 	struct vnode *vp;
110 	nfsfh_t nfh;
111 	fhandle_t *fhp;
112 	u_int32_t *tl;
113 	int32_t t1;
114 	caddr_t bpos;
115 	int error = 0, rdonly, cache = 0, getret;
116 	char *cp2;
117 	struct mbuf *mb, *mreq, *mb2;
118 	struct vattr va;
119 	u_long inmode, testmode, outmode;
120 	u_quad_t frev;
121 
122 	fhp = &nfh.fh_generic;
123 	nfsm_srvmtofh(fhp);
124 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
125 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
126 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
127 	if (error) {
128 		nfsm_reply(NFSX_UNSIGNED);
129 		nfsm_srvpostop_attr(1, (struct vattr *)0);
130 		return (0);
131 	}
132 	inmode = fxdr_unsigned(u_int32_t, *tl);
133 	outmode = 0;
134 	if ((inmode & NFSV3ACCESS_READ) &&
135 	    nfsrv_access(vp, VREAD, cred, rdonly, procp, 0) == 0)
136 		outmode |= NFSV3ACCESS_READ;
137 	if (vp->v_type != VDIR) {
138 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
139 		if (testmode &&
140 		    nfsrv_access(vp, VWRITE, cred, rdonly, procp, 0) == 0)
141 			outmode |= testmode;
142 		if ((inmode & NFSV3ACCESS_EXECUTE) &&
143 		    nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0) == 0)
144 			outmode |= NFSV3ACCESS_EXECUTE;
145 	} else {
146 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
147 		    NFSV3ACCESS_DELETE);
148 		if (testmode &&
149 		    nfsrv_access(vp, VWRITE, cred, rdonly, procp, 0) == 0)
150 			outmode |= testmode;
151 		if ((inmode & NFSV3ACCESS_LOOKUP) &&
152 		    nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0) == 0)
153 			outmode |= NFSV3ACCESS_LOOKUP;
154 	}
155 	getret = VOP_GETATTR(vp, &va, cred, procp);
156 	vput(vp);
157 	nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
158 	nfsm_srvpostop_attr(getret, &va);
159 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
160 	*tl = txdr_unsigned(outmode);
161 	nfsm_srvdone;
162 }
163 
164 /*
165  * nfs getattr service
166  */
167 int
168 nfsrv_getattr(nfsd, slp, procp, mrq)
169 	struct nfsrv_descript *nfsd;
170 	struct nfssvc_sock *slp;
171 	struct proc *procp;
172 	struct mbuf **mrq;
173 {
174 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
175 	struct mbuf *nam = nfsd->nd_nam;
176 	caddr_t dpos = nfsd->nd_dpos;
177 	struct ucred *cred = &nfsd->nd_cr;
178 	struct nfs_fattr *fp;
179 	struct vattr va;
180 	struct vnode *vp;
181 	nfsfh_t nfh;
182 	fhandle_t *fhp;
183 	u_int32_t *tl;
184 	int32_t t1;
185 	caddr_t bpos;
186 	int error = 0, rdonly, cache;
187 	char *cp2;
188 	struct mbuf *mb, *mb2, *mreq;
189 	u_quad_t frev;
190 
191 	fhp = &nfh.fh_generic;
192 	nfsm_srvmtofh(fhp);
193 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
194 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
195 	if (error) {
196 		nfsm_reply(0);
197 		return (0);
198 	}
199 	nqsrv_getl(vp, ND_READ);
200 	error = VOP_GETATTR(vp, &va, cred, procp);
201 	vput(vp);
202 	nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
203 	if (error)
204 		return (0);
205 	nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
206 	nfsm_srvfillattr(&va, fp);
207 	nfsm_srvdone;
208 }
209 
210 /*
211  * nfs setattr service
212  */
213 int
214 nfsrv_setattr(nfsd, slp, procp, mrq)
215 	struct nfsrv_descript *nfsd;
216 	struct nfssvc_sock *slp;
217 	struct proc *procp;
218 	struct mbuf **mrq;
219 {
220 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
221 	struct mbuf *nam = nfsd->nd_nam;
222 	caddr_t dpos = nfsd->nd_dpos;
223 	struct ucred *cred = &nfsd->nd_cr;
224 	struct vattr va, preat;
225 	struct nfsv2_sattr *sp;
226 	struct nfs_fattr *fp;
227 	struct vnode *vp;
228 	nfsfh_t nfh;
229 	fhandle_t *fhp;
230 	u_int32_t *tl;
231 	int32_t t1;
232 	caddr_t bpos;
233 	int error = 0, rdonly, cache, preat_ret = 1, postat_ret = 1;
234 	int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
235 	char *cp2;
236 	struct mbuf *mb, *mb2, *mreq;
237 	u_quad_t frev;
238 	struct timespec guard;
239 
240 	fhp = &nfh.fh_generic;
241 	nfsm_srvmtofh(fhp);
242 	VATTR_NULL(&va);
243 	if (v3) {
244 		nfsm_srvsattr(&va);
245 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
246 		gcheck = fxdr_unsigned(int, *tl);
247 		if (gcheck) {
248 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
249 			fxdr_nfsv3time(tl, &guard);
250 		}
251 	} else {
252 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
253 		/*
254 		 * Nah nah nah nah na nah
255 		 * There is a bug in the Sun client that puts 0xffff in the mode
256 		 * field of sattr when it should put in 0xffffffff. The u_short
257 		 * doesn't sign extend.
258 		 * --> check the low order 2 bytes for 0xffff
259 		 */
260 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
261 			va.va_mode = nfstov_mode(sp->sa_mode);
262 		if (sp->sa_uid != nfs_xdrneg1)
263 			va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
264 		if (sp->sa_gid != nfs_xdrneg1)
265 			va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
266 		if (sp->sa_size != nfs_xdrneg1)
267 			va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
268 		if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) {
269 #ifdef notyet
270 			fxdr_nfsv2time(&sp->sa_atime, &va.va_atime);
271 #else
272 			va.va_atime.tv_sec =
273 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
274 			va.va_atime.tv_nsec = 0;
275 #endif
276 		}
277 		if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1)
278 			fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime);
279 
280 	}
281 
282 	/*
283 	 * Now that we have all the fields, lets do it.
284 	 */
285 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
286 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
287 	if (error) {
288 		nfsm_reply(2 * NFSX_UNSIGNED);
289 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
290 		return (0);
291 	}
292 	nqsrv_getl(vp, ND_WRITE);
293 	if (v3) {
294 		error = preat_ret = VOP_GETATTR(vp, &preat, cred, procp);
295 		if (!error && gcheck &&
296 			(preat.va_ctime.tv_sec != guard.tv_sec ||
297 			 preat.va_ctime.tv_nsec != guard.tv_nsec))
298 			error = NFSERR_NOT_SYNC;
299 		if (error) {
300 			vput(vp);
301 			nfsm_reply(NFSX_WCCDATA(v3));
302 			nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
303 			return (0);
304 		}
305 	}
306 
307 	/*
308 	 * If the size is being changed write acces is required, otherwise
309 	 * just check for a read only file system.
310 	 */
311 	if (va.va_size == ((u_quad_t)((quad_t) -1))) {
312 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
313 			error = EROFS;
314 			goto out;
315 		}
316 	} else {
317 		if (vp->v_type == VDIR) {
318 			error = EISDIR;
319 			goto out;
320 		} else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
321 			procp, 0)) != 0)
322 			goto out;
323 	}
324 	error = VOP_SETATTR(vp, &va, cred, procp);
325 	postat_ret = VOP_GETATTR(vp, &va, cred, procp);
326 	if (!error)
327 		error = postat_ret;
328 out:
329 	vput(vp);
330 	nfsm_reply(NFSX_WCCORFATTR(v3));
331 	if (v3) {
332 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
333 		return (0);
334 	} else {
335 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
336 		nfsm_srvfillattr(&va, fp);
337 	}
338 	nfsm_srvdone;
339 }
340 
341 /*
342  * nfs lookup rpc
343  */
344 int
345 nfsrv_lookup(nfsd, slp, procp, mrq)
346 	struct nfsrv_descript *nfsd;
347 	struct nfssvc_sock *slp;
348 	struct proc *procp;
349 	struct mbuf **mrq;
350 {
351 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
352 	struct mbuf *nam = nfsd->nd_nam;
353 	caddr_t dpos = nfsd->nd_dpos;
354 	struct ucred *cred = &nfsd->nd_cr;
355 	struct nfs_fattr *fp;
356 	struct nameidata nd, ind, *ndp = &nd;
357 	struct vnode *vp, *dirp;
358 	nfsfh_t nfh;
359 	fhandle_t *fhp;
360 	caddr_t cp;
361 	u_int32_t *tl;
362 	int32_t t1;
363 	caddr_t bpos;
364 	int error = 0, cache, len, dirattr_ret = 1;
365 	int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
366 	char *cp2;
367 	struct mbuf *mb, *mb2, *mreq;
368 	struct vattr va, dirattr;
369 	u_quad_t frev;
370 
371 	fhp = &nfh.fh_generic;
372 	nfsm_srvmtofh(fhp);
373 	nfsm_srvnamesiz(len);
374 
375 	pubflag = nfs_ispublicfh(fhp);
376 
377 	nd.ni_cnd.cn_cred = cred;
378 	nd.ni_cnd.cn_nameiop = LOOKUP;
379 	nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART;
380 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
381 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
382 
383 	if (!error && pubflag) {
384 		if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
385 			/*
386 			 * Setup call to lookup() to see if we can find
387 			 * the index file. Arguably, this doesn't belong
388 			 * in a kernel.. Ugh.
389 			 */
390 			ind = nd;
391 			VOP_UNLOCK(nd.ni_vp, 0);
392 			ind.ni_pathlen = strlen(nfs_pub.np_index);
393 			ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
394 			    nfs_pub.np_index;
395 			ind.ni_startdir = nd.ni_vp;
396 			VREF(ind.ni_startdir);
397 			error = lookup(&ind);
398 			if (!error) {
399 				/*
400 				 * Found an index file. Get rid of
401 				 * the old references.
402 				 */
403 				if (dirp)
404 					vrele(dirp);
405 				dirp = nd.ni_vp;
406 				vrele(nd.ni_startdir);
407 				ndp = &ind;
408 			} else
409 				error = 0;
410 		}
411 		/*
412 		 * If the public filehandle was used, check that this lookup
413 		 * didn't result in a filehandle outside the publicly exported
414 		 * filesystem.
415 		 */
416 
417 		if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) {
418 			vput(nd.ni_vp);
419 			error = EPERM;
420 		}
421 	}
422 
423 	if (dirp) {
424 		if (v3)
425 			dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred,
426 				procp);
427 		vrele(dirp);
428 	}
429 
430 	if (error) {
431 		nfsm_reply(NFSX_POSTOPATTR(v3));
432 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
433 		return (0);
434 	}
435 
436 	nqsrv_getl(ndp->ni_startdir, ND_READ);
437 	vrele(ndp->ni_startdir);
438 	PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
439 	vp = ndp->ni_vp;
440 	memset((caddr_t)fhp, 0, sizeof(nfh));
441 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
442 	error = VFS_VPTOFH(vp, &fhp->fh_fid);
443 	if (!error)
444 		error = VOP_GETATTR(vp, &va, cred, procp);
445 	vput(vp);
446 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
447 	if (error) {
448 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
449 		return (0);
450 	}
451 	nfsm_srvfhtom(fhp, v3);
452 	if (v3) {
453 		nfsm_srvpostop_attr(0, &va);
454 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
455 	} else {
456 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
457 		nfsm_srvfillattr(&va, fp);
458 	}
459 	nfsm_srvdone;
460 }
461 
462 /*
463  * nfs readlink service
464  */
465 int
466 nfsrv_readlink(nfsd, slp, procp, mrq)
467 	struct nfsrv_descript *nfsd;
468 	struct nfssvc_sock *slp;
469 	struct proc *procp;
470 	struct mbuf **mrq;
471 {
472 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
473 	struct mbuf *nam = nfsd->nd_nam;
474 	caddr_t dpos = nfsd->nd_dpos;
475 	struct ucred *cred = &nfsd->nd_cr;
476 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
477 	struct iovec *ivp = iv;
478 	struct mbuf *mp;
479 	u_int32_t *tl;
480 	int32_t t1;
481 	caddr_t bpos;
482 	int error = 0, rdonly, cache, i, tlen, len, getret;
483 	int v3 = (nfsd->nd_flag & ND_NFSV3);
484 	char *cp2;
485 	struct mbuf *mb, *mb2, *mp2 = NULL, *mp3 = NULL, *mreq;
486 	struct vnode *vp;
487 	struct vattr attr;
488 	nfsfh_t nfh;
489 	fhandle_t *fhp;
490 	struct uio io, *uiop = &io;
491 	u_quad_t frev;
492 
493 	fhp = &nfh.fh_generic;
494 	nfsm_srvmtofh(fhp);
495 	len = 0;
496 	i = 0;
497 	while (len < NFS_MAXPATHLEN) {
498 		MGET(mp, M_WAIT, MT_DATA);
499 		MCLGET(mp, M_WAIT);
500 		mp->m_len = NFSMSIZ(mp);
501 		if (len == 0)
502 			mp3 = mp2 = mp;
503 		else {
504 			mp2->m_next = mp;
505 			mp2 = mp;
506 		}
507 		if ((len+mp->m_len) > NFS_MAXPATHLEN) {
508 			mp->m_len = NFS_MAXPATHLEN-len;
509 			len = NFS_MAXPATHLEN;
510 		} else
511 			len += mp->m_len;
512 		ivp->iov_base = mtod(mp, caddr_t);
513 		ivp->iov_len = mp->m_len;
514 		i++;
515 		ivp++;
516 	}
517 	uiop->uio_iov = iv;
518 	uiop->uio_iovcnt = i;
519 	uiop->uio_offset = 0;
520 	uiop->uio_resid = len;
521 	uiop->uio_rw = UIO_READ;
522 	uiop->uio_segflg = UIO_SYSSPACE;
523 	uiop->uio_procp = (struct proc *)0;
524 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
525 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
526 	if (error) {
527 		m_freem(mp3);
528 		nfsm_reply(2 * NFSX_UNSIGNED);
529 		nfsm_srvpostop_attr(1, (struct vattr *)0);
530 		return (0);
531 	}
532 	if (vp->v_type != VLNK) {
533 		if (v3)
534 			error = EINVAL;
535 		else
536 			error = ENXIO;
537 		goto out;
538 	}
539 	nqsrv_getl(vp, ND_READ);
540 	error = VOP_READLINK(vp, uiop, cred);
541 out:
542 	getret = VOP_GETATTR(vp, &attr, cred, procp);
543 	vput(vp);
544 	if (error)
545 		m_freem(mp3);
546 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
547 	if (v3) {
548 		nfsm_srvpostop_attr(getret, &attr);
549 		if (error)
550 			return (0);
551 	}
552 	if (uiop->uio_resid > 0) {
553 		len -= uiop->uio_resid;
554 		tlen = nfsm_rndup(len);
555 		nfsm_adj(mp3, NFS_MAXPATHLEN-tlen, tlen-len);
556 	}
557 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
558 	*tl = txdr_unsigned(len);
559 	mb->m_next = mp3;
560 	nfsm_srvdone;
561 }
562 
563 /*
564  * nfs read service
565  */
566 int
567 nfsrv_read(nfsd, slp, procp, mrq)
568 	struct nfsrv_descript *nfsd;
569 	struct nfssvc_sock *slp;
570 	struct proc *procp;
571 	struct mbuf **mrq;
572 {
573 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
574 	struct mbuf *nam = nfsd->nd_nam;
575 	caddr_t dpos = nfsd->nd_dpos;
576 	struct ucred *cred = &nfsd->nd_cr;
577 	struct iovec *iv;
578 	struct iovec *iv2;
579 	struct mbuf *m;
580 	struct nfs_fattr *fp;
581 	u_int32_t *tl;
582 	int32_t t1;
583 	int i;
584 	caddr_t bpos;
585 	int error = 0, rdonly, cache, cnt, len, left, siz, tlen, getret;
586 	int v3 = (nfsd->nd_flag & ND_NFSV3), reqlen;
587 	char *cp2;
588 	struct mbuf *mb, *mb2, *mreq;
589 	struct mbuf *m2;
590 	struct vnode *vp;
591 	nfsfh_t nfh;
592 	fhandle_t *fhp;
593 	struct uio io, *uiop = &io;
594 	struct vattr va;
595 	off_t off;
596 	u_quad_t frev;
597 
598 	fhp = &nfh.fh_generic;
599 	nfsm_srvmtofh(fhp);
600 	if (v3) {
601 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
602 		off = fxdr_hyper(tl);
603 	} else {
604 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
605 		off = (off_t)fxdr_unsigned(u_int32_t, *tl);
606 	}
607 	nfsm_srvstrsiz(reqlen, NFS_SRVMAXDATA(nfsd));
608 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
609 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
610 	if (error) {
611 		nfsm_reply(2 * NFSX_UNSIGNED);
612 		nfsm_srvpostop_attr(1, (struct vattr *)0);
613 		return (0);
614 	}
615 	if (vp->v_type != VREG) {
616 		if (v3)
617 			error = EINVAL;
618 		else
619 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
620 	}
621 	if (!error) {
622 	    nqsrv_getl(vp, ND_READ);
623 	    if ((error = nfsrv_access(vp, VREAD, cred, rdonly, procp, 1)) != 0)
624 		error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 1);
625 	}
626 	getret = VOP_GETATTR(vp, &va, cred, procp);
627 	if (!error)
628 		error = getret;
629 	if (error) {
630 		vput(vp);
631 		nfsm_reply(NFSX_POSTOPATTR(v3));
632 		nfsm_srvpostop_attr(getret, &va);
633 		return (0);
634 	}
635 	if (off >= va.va_size)
636 		cnt = 0;
637 	else if ((off + reqlen) > va.va_size)
638 		cnt = nfsm_rndup(va.va_size - off);
639 	else
640 		cnt = reqlen;
641 	nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
642 	if (v3) {
643 		nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
644 		*tl++ = nfs_true;
645 		fp = (struct nfs_fattr *)tl;
646 		tl += (NFSX_V3FATTR / sizeof (u_int32_t));
647 	} else {
648 		nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
649 		fp = (struct nfs_fattr *)tl;
650 		tl += (NFSX_V2FATTR / sizeof (u_int32_t));
651 	}
652 	len = left = cnt;
653 	if (cnt > 0) {
654 		/*
655 		 * Generate the mbuf list with the uio_iov ref. to it.
656 		 */
657 		i = 0;
658 		m = m2 = mb;
659 		while (left > 0) {
660 			siz = min(M_TRAILINGSPACE(m), left);
661 			if (siz > 0) {
662 				left -= siz;
663 				i++;
664 			}
665 			if (left > 0) {
666 				MGET(m, M_WAIT, MT_DATA);
667 				MCLGET(m, M_WAIT);
668 				m->m_len = 0;
669 				m2->m_next = m;
670 				m2 = m;
671 			}
672 		}
673 		iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
674 		uiop->uio_iov = iv2 = iv;
675 		m = mb;
676 		left = cnt;
677 		i = 0;
678 		while (left > 0) {
679 			if (m == NULL)
680 				panic("nfsrv_read iov");
681 			siz = min(M_TRAILINGSPACE(m), left);
682 			if (siz > 0) {
683 				iv->iov_base = mtod(m, caddr_t) + m->m_len;
684 				iv->iov_len = siz;
685 				m->m_len += siz;
686 				left -= siz;
687 				iv++;
688 				i++;
689 			}
690 			m = m->m_next;
691 		}
692 		uiop->uio_iovcnt = i;
693 		uiop->uio_offset = off;
694 		uiop->uio_resid = cnt;
695 		uiop->uio_rw = UIO_READ;
696 		uiop->uio_segflg = UIO_SYSSPACE;
697 		error = VOP_READ(vp, uiop, IO_NODELOCKED, cred);
698 		off = uiop->uio_offset;
699 		free((caddr_t)iv2, M_TEMP);
700 		if (error || (getret = VOP_GETATTR(vp, &va, cred, procp)) != 0){
701 			if (!error)
702 				error = getret;
703 			m_freem(mreq);
704 			vput(vp);
705 			nfsm_reply(NFSX_POSTOPATTR(v3));
706 			nfsm_srvpostop_attr(getret, &va);
707 			return (0);
708 		}
709 	} else
710 		uiop->uio_resid = 0;
711 	vput(vp);
712 	nfsm_srvfillattr(&va, fp);
713 	len -= uiop->uio_resid;
714 	tlen = nfsm_rndup(len);
715 	if (cnt != tlen || tlen != len)
716 		nfsm_adj(mb, cnt - tlen, tlen - len);
717 	if (v3) {
718 		*tl++ = txdr_unsigned(len);
719 		if (len < reqlen)
720 			*tl++ = nfs_true;
721 		else
722 			*tl++ = nfs_false;
723 	}
724 	*tl = txdr_unsigned(len);
725 	nfsm_srvdone;
726 }
727 
728 /*
729  * nfs write service
730  */
731 int
732 nfsrv_write(nfsd, slp, procp, mrq)
733 	struct nfsrv_descript *nfsd;
734 	struct nfssvc_sock *slp;
735 	struct proc *procp;
736 	struct mbuf **mrq;
737 {
738 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
739 	struct mbuf *nam = nfsd->nd_nam;
740 	caddr_t dpos = nfsd->nd_dpos;
741 	struct ucred *cred = &nfsd->nd_cr;
742 	struct iovec *ivp;
743 	int i, cnt;
744 	struct mbuf *mp;
745 	struct nfs_fattr *fp;
746 	struct iovec *iv;
747 	struct vattr va, forat;
748 	u_int32_t *tl;
749 	int32_t t1;
750 	caddr_t bpos;
751 	int error = 0, rdonly, cache, len, forat_ret = 1;
752 	int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
753 	int stable = NFSV3WRITE_FILESYNC;
754 	int v3 = (nfsd->nd_flag & ND_NFSV3);
755 	char *cp2;
756 	struct mbuf *mb, *mb2, *mreq;
757 	struct vnode *vp;
758 	nfsfh_t nfh;
759 	fhandle_t *fhp;
760 	struct uio io, *uiop = &io;
761 	off_t off;
762 	u_quad_t frev;
763 
764 	if (mrep == NULL) {
765 		*mrq = NULL;
766 		return (0);
767 	}
768 	fhp = &nfh.fh_generic;
769 	nfsm_srvmtofh(fhp);
770 	if (v3) {
771 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
772 		off = fxdr_hyper(tl);
773 		tl += 3;
774 		stable = fxdr_unsigned(int, *tl++);
775 	} else {
776 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
777 		off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
778 		tl += 2;
779 	}
780 	retlen = len = fxdr_unsigned(int32_t, *tl);
781 	cnt = i = 0;
782 
783 	/*
784 	 * For NFS Version 2, it is not obvious what a write of zero length
785 	 * should do, but I might as well be consistent with Version 3,
786 	 * which is to return ok so long as there are no permission problems.
787 	 */
788 	if (len > 0) {
789 	    zeroing = 1;
790 	    mp = mrep;
791 	    while (mp) {
792 		if (mp == md) {
793 			zeroing = 0;
794 			adjust = dpos - mtod(mp, caddr_t);
795 			mp->m_len -= adjust;
796 			if (mp->m_len > 0 && adjust > 0)
797 				NFSMADV(mp, adjust);
798 		}
799 		if (zeroing)
800 			mp->m_len = 0;
801 		else if (mp->m_len > 0) {
802 			i += mp->m_len;
803 			if (i > len) {
804 				mp->m_len -= (i - len);
805 				zeroing	= 1;
806 			}
807 			if (mp->m_len > 0)
808 				cnt++;
809 		}
810 		mp = mp->m_next;
811 	    }
812 	}
813 	if (len > NFS_MAXDATA || len < 0 || i < len) {
814 		error = EIO;
815 		nfsm_reply(2 * NFSX_UNSIGNED);
816 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
817 		return (0);
818 	}
819 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
820 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
821 	if (error) {
822 		nfsm_reply(2 * NFSX_UNSIGNED);
823 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
824 		return (0);
825 	}
826 	if (v3)
827 		forat_ret = VOP_GETATTR(vp, &forat, cred, procp);
828 	if (vp->v_type != VREG) {
829 		if (v3)
830 			error = EINVAL;
831 		else
832 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
833 	}
834 	if (!error) {
835 		nqsrv_getl(vp, ND_WRITE);
836 		error = nfsrv_access(vp, VWRITE, cred, rdonly, procp, 1);
837 	}
838 	if (error) {
839 		vput(vp);
840 		nfsm_reply(NFSX_WCCDATA(v3));
841 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
842 		return (0);
843 	}
844 
845 	if (len > 0) {
846 	    ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK);
847 	    uiop->uio_iov = iv = ivp;
848 	    uiop->uio_iovcnt = cnt;
849 	    mp = mrep;
850 	    while (mp) {
851 		if (mp->m_len > 0) {
852 			ivp->iov_base = mtod(mp, caddr_t);
853 			ivp->iov_len = mp->m_len;
854 			ivp++;
855 		}
856 		mp = mp->m_next;
857 	    }
858 
859 	    /*
860 	     * XXX
861 	     * The IO_METASYNC flag indicates that all metadata (and not just
862 	     * enough to ensure data integrity) mus be written to stable storage
863 	     * synchronously.
864 	     * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
865 	     */
866 	    if (stable == NFSV3WRITE_UNSTABLE)
867 		ioflags = IO_NODELOCKED;
868 	    else if (stable == NFSV3WRITE_DATASYNC)
869 		ioflags = (IO_SYNC | IO_NODELOCKED);
870 	    else
871 		ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
872 	    uiop->uio_resid = len;
873 	    uiop->uio_rw = UIO_WRITE;
874 	    uiop->uio_segflg = UIO_SYSSPACE;
875 	    uiop->uio_procp = (struct proc *)0;
876 	    uiop->uio_offset = off;
877 	    error = VOP_WRITE(vp, uiop, ioflags, cred);
878 	    nfsstats.srvvop_writes++;
879 	    free((caddr_t)iv, M_TEMP);
880 	}
881 	aftat_ret = VOP_GETATTR(vp, &va, cred, procp);
882 	vput(vp);
883 	if (!error)
884 		error = aftat_ret;
885 	nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
886 		2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
887 	if (v3) {
888 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
889 		if (error)
890 			return (0);
891 		nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
892 		*tl++ = txdr_unsigned(retlen);
893 		if (stable == NFSV3WRITE_UNSTABLE)
894 			*tl++ = txdr_unsigned(stable);
895 		else
896 			*tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
897 		/*
898 		 * Actually, there is no need to txdr these fields,
899 		 * but it may make the values more human readable,
900 		 * for debugging purposes.
901 		 */
902 		*tl++ = txdr_unsigned(boottime.tv_sec);
903 		*tl = txdr_unsigned(boottime.tv_usec);
904 	} else {
905 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
906 		nfsm_srvfillattr(&va, fp);
907 	}
908 	nfsm_srvdone;
909 }
910 
911 /*
912  * NFS write service with write gathering support. Called when
913  * nfsrvw_procrastinate > 0.
914  * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
915  * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
916  * Jan. 1994.
917  */
918 int
919 nfsrv_writegather(ndp, slp, procp, mrq)
920 	struct nfsrv_descript **ndp;
921 	struct nfssvc_sock *slp;
922 	struct proc *procp;
923 	struct mbuf **mrq;
924 {
925 	struct iovec *ivp;
926 	struct mbuf *mp;
927 	struct nfsrv_descript *wp, *nfsd, *owp, *swp;
928 	struct nfs_fattr *fp;
929 	int i = 0;
930 	struct iovec *iov;
931 	struct nfsrvw_delayhash *wpp;
932 	struct ucred *cred;
933 	struct vattr va, forat;
934 	u_int32_t *tl;
935 	int32_t t1;
936 	caddr_t bpos, dpos;
937 	int error = 0, rdonly, cache, len = 0, forat_ret = 1;
938 	int ioflags, aftat_ret = 1, s, adjust, v3, zeroing;
939 	char *cp2;
940 	struct mbuf *mb, *mb2, *mreq, *mrep, *md;
941 	struct vnode *vp;
942 	struct uio io, *uiop = &io;
943 	u_quad_t frev, cur_usec;
944 
945 	*mrq = NULL;
946 	if (*ndp) {
947 	    nfsd = *ndp;
948 	    *ndp = NULL;
949 	    mrep = nfsd->nd_mrep;
950 	    md = nfsd->nd_md;
951 	    dpos = nfsd->nd_dpos;
952 	    cred = &nfsd->nd_cr;
953 	    v3 = (nfsd->nd_flag & ND_NFSV3);
954 	    LIST_INIT(&nfsd->nd_coalesce);
955 	    nfsd->nd_mreq = NULL;
956 	    nfsd->nd_stable = NFSV3WRITE_FILESYNC;
957 	    cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
958 	    nfsd->nd_time = cur_usec + nfsrvw_procrastinate;
959 
960 	    /*
961 	     * Now, get the write header..
962 	     */
963 	    nfsm_srvmtofh(&nfsd->nd_fh);
964 	    if (v3) {
965 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
966 		nfsd->nd_off = fxdr_hyper(tl);
967 		tl += 3;
968 		nfsd->nd_stable = fxdr_unsigned(int, *tl++);
969 	    } else {
970 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
971 		nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
972 		tl += 2;
973 	    }
974 	    len = fxdr_unsigned(int32_t, *tl);
975 	    nfsd->nd_len = len;
976 	    nfsd->nd_eoff = nfsd->nd_off + len;
977 
978 	    /*
979 	     * Trim the header out of the mbuf list and trim off any trailing
980 	     * junk so that the mbuf list has only the write data.
981 	     */
982 	    zeroing = 1;
983 	    i = 0;
984 	    mp = mrep;
985 	    while (mp) {
986 		if (mp == md) {
987 		    zeroing = 0;
988 		    adjust = dpos - mtod(mp, caddr_t);
989 		    mp->m_len -= adjust;
990 		    if (mp->m_len > 0 && adjust > 0)
991 			NFSMADV(mp, adjust);
992 		}
993 		if (zeroing)
994 		    mp->m_len = 0;
995 		else {
996 		    i += mp->m_len;
997 		    if (i > len) {
998 			mp->m_len -= (i - len);
999 			zeroing = 1;
1000 		    }
1001 		}
1002 		mp = mp->m_next;
1003 	    }
1004 	    if (len > NFS_MAXDATA || len < 0  || i < len) {
1005 nfsmout:
1006 		m_freem(mrep);
1007 		error = EIO;
1008 		nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1009 		if (v3)
1010 		    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1011 		nfsd->nd_mreq = mreq;
1012 		nfsd->nd_mrep = NULL;
1013 		nfsd->nd_time = 0;
1014 	    }
1015 
1016 	    /*
1017 	     * Add this entry to the hash and time queues.
1018 	     */
1019 	    s = splsoftclock();
1020 	    owp = NULL;
1021 	    wp = slp->ns_tq.lh_first;
1022 	    while (wp && wp->nd_time < nfsd->nd_time) {
1023 		owp = wp;
1024 		wp = wp->nd_tq.le_next;
1025 	    }
1026 	    if (owp) {
1027 		LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1028 	    } else {
1029 		LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1030 	    }
1031 	    if (nfsd->nd_mrep) {
1032 		wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1033 		owp = NULL;
1034 		wp = wpp->lh_first;
1035 		while (wp &&
1036 		    memcmp((caddr_t)&nfsd->nd_fh, (caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1037 		    owp = wp;
1038 		    wp = wp->nd_hash.le_next;
1039 		}
1040 		while (wp && wp->nd_off < nfsd->nd_off &&
1041 		    !memcmp((caddr_t)&nfsd->nd_fh, (caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1042 		    owp = wp;
1043 		    wp = wp->nd_hash.le_next;
1044 		}
1045 		if (owp) {
1046 		    LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1047 
1048 		    /*
1049 		     * Search the hash list for overlapping entries and
1050 		     * coalesce.
1051 		     */
1052 		    for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1053 			wp = nfsd->nd_hash.le_next;
1054 			if (NFSW_SAMECRED(owp, nfsd))
1055 			    nfsrvw_coalesce(owp, nfsd);
1056 		    }
1057 		} else {
1058 		    LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1059 		}
1060 	    }
1061 	    splx(s);
1062 	}
1063 
1064 	/*
1065 	 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1066 	 * and generate the associated reply mbuf list(s).
1067 	 */
1068 loop1:
1069 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1070 	s = splsoftclock();
1071 	for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = owp) {
1072 		owp = nfsd->nd_tq.le_next;
1073 		if (nfsd->nd_time > cur_usec)
1074 		    break;
1075 		if (nfsd->nd_mreq)
1076 		    continue;
1077 		LIST_REMOVE(nfsd, nd_tq);
1078 		LIST_REMOVE(nfsd, nd_hash);
1079 		splx(s);
1080 		mrep = nfsd->nd_mrep;
1081 		nfsd->nd_mrep = NULL;
1082 		cred = &nfsd->nd_cr;
1083 		v3 = (nfsd->nd_flag & ND_NFSV3);
1084 		forat_ret = aftat_ret = 1;
1085 		error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1086 		    nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH),
1087 		    FALSE);
1088 		if (!error) {
1089 		    if (v3)
1090 			forat_ret = VOP_GETATTR(vp, &forat, cred, procp);
1091 		    if (vp->v_type != VREG) {
1092 			if (v3)
1093 			    error = EINVAL;
1094 			else
1095 			    error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1096 		    }
1097 		} else
1098 		    vp = NULL;
1099 		if (!error) {
1100 		    nqsrv_getl(vp, ND_WRITE);
1101 		    error = nfsrv_access(vp, VWRITE, cred, rdonly, procp, 1);
1102 		}
1103 
1104 		if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1105 		    ioflags = IO_NODELOCKED;
1106 		else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1107 		    ioflags = (IO_SYNC | IO_NODELOCKED);
1108 		else
1109 		    ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1110 		uiop->uio_rw = UIO_WRITE;
1111 		uiop->uio_segflg = UIO_SYSSPACE;
1112 		uiop->uio_procp = (struct proc *)0;
1113 		uiop->uio_offset = nfsd->nd_off;
1114 		uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1115 		if (uiop->uio_resid > 0) {
1116 		    mp = mrep;
1117 		    i = 0;
1118 		    while (mp) {
1119 			if (mp->m_len > 0)
1120 			    i++;
1121 			mp = mp->m_next;
1122 		    }
1123 		    uiop->uio_iovcnt = i;
1124 		    iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
1125 		    uiop->uio_iov = ivp = iov;
1126 		    mp = mrep;
1127 		    while (mp) {
1128 			if (mp->m_len > 0) {
1129 			    ivp->iov_base = mtod(mp, caddr_t);
1130 			    ivp->iov_len = mp->m_len;
1131 			    ivp++;
1132 			}
1133 			mp = mp->m_next;
1134 		    }
1135 		    if (!error) {
1136 			error = VOP_WRITE(vp, uiop, ioflags, cred);
1137 			nfsstats.srvvop_writes++;
1138 		    }
1139 		    free((caddr_t)iov, M_TEMP);
1140 		}
1141 		m_freem(mrep);
1142 		if (vp) {
1143 		    aftat_ret = VOP_GETATTR(vp, &va, cred, procp);
1144 		    vput(vp);
1145 		}
1146 
1147 		/*
1148 		 * Loop around generating replies for all write rpcs that have
1149 		 * now been completed.
1150 		 */
1151 		swp = nfsd;
1152 		do {
1153 		    if (error) {
1154 			nfsm_writereply(NFSX_WCCDATA(v3), v3);
1155 			if (v3) {
1156 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1157 			}
1158 		    } else {
1159 			nfsm_writereply(NFSX_PREOPATTR(v3) +
1160 			    NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1161 			    NFSX_WRITEVERF(v3), v3);
1162 			if (v3) {
1163 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1164 			    nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1165 			    *tl++ = txdr_unsigned(nfsd->nd_len);
1166 			    *tl++ = txdr_unsigned(swp->nd_stable);
1167 			    /*
1168 			     * Actually, there is no need to txdr these fields,
1169 			     * but it may make the values more human readable,
1170 			     * for debugging purposes.
1171 			     */
1172 			    *tl++ = txdr_unsigned(boottime.tv_sec);
1173 			    *tl = txdr_unsigned(boottime.tv_usec);
1174 			} else {
1175 			    nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1176 			    nfsm_srvfillattr(&va, fp);
1177 			}
1178 		    }
1179 		    nfsd->nd_mreq = mreq;
1180 		    if (nfsd->nd_mrep)
1181 			panic("nfsrv_write: nd_mrep not free");
1182 
1183 		    /*
1184 		     * Done. Put it at the head of the timer queue so that
1185 		     * the final phase can return the reply.
1186 		     */
1187 		    s = splsoftclock();
1188 		    if (nfsd != swp) {
1189 			nfsd->nd_time = 0;
1190 			LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1191 		    }
1192 		    nfsd = swp->nd_coalesce.lh_first;
1193 		    if (nfsd) {
1194 			LIST_REMOVE(nfsd, nd_tq);
1195 		    }
1196 		    splx(s);
1197 		} while (nfsd);
1198 		s = splsoftclock();
1199 		swp->nd_time = 0;
1200 		LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1201 		splx(s);
1202 		goto loop1;
1203 	}
1204 	splx(s);
1205 
1206 	/*
1207 	 * Search for a reply to return.
1208 	 */
1209 	s = splsoftclock();
1210 	for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = nfsd->nd_tq.le_next)
1211 		if (nfsd->nd_mreq) {
1212 		    LIST_REMOVE(nfsd, nd_tq);
1213 		    *mrq = nfsd->nd_mreq;
1214 		    *ndp = nfsd;
1215 		    break;
1216 		}
1217 	splx(s);
1218 	return (0);
1219 }
1220 
1221 /*
1222  * Coalesce the write request nfsd into owp. To do this we must:
1223  * - remove nfsd from the queues
1224  * - merge nfsd->nd_mrep into owp->nd_mrep
1225  * - update the nd_eoff and nd_stable for owp
1226  * - put nfsd on owp's nd_coalesce list
1227  * NB: Must be called at splsoftclock().
1228  */
1229 void
1230 nfsrvw_coalesce(owp, nfsd)
1231         struct nfsrv_descript *owp;
1232         struct nfsrv_descript *nfsd;
1233 {
1234         int overlap;
1235         struct mbuf *mp;
1236 
1237         LIST_REMOVE(nfsd, nd_hash);
1238         LIST_REMOVE(nfsd, nd_tq);
1239         if (owp->nd_eoff < nfsd->nd_eoff) {
1240             overlap = owp->nd_eoff - nfsd->nd_off;
1241             if (overlap < 0)
1242                 panic("nfsrv_coalesce: bad off");
1243             if (overlap > 0)
1244                 m_adj(nfsd->nd_mrep, overlap);
1245             mp = owp->nd_mrep;
1246             while (mp->m_next)
1247                 mp = mp->m_next;
1248             mp->m_next = nfsd->nd_mrep;
1249             owp->nd_eoff = nfsd->nd_eoff;
1250         } else
1251             m_freem(nfsd->nd_mrep);
1252         nfsd->nd_mrep = NULL;
1253         if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1254             owp->nd_stable = NFSV3WRITE_FILESYNC;
1255         else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1256             owp->nd_stable == NFSV3WRITE_UNSTABLE)
1257             owp->nd_stable = NFSV3WRITE_DATASYNC;
1258         LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1259  	/*
1260  	 * nfsd might hold coalesce elements! Move them to owp.
1261  	 * Otherwise, requests may be lost and clients will be stuck.
1262  	 */
1263  	if (nfsd->nd_coalesce.lh_first)
1264  	{
1265  		struct nfsrv_descript *m;
1266 
1267  		while ((m = nfsd->nd_coalesce.lh_first))
1268  		{
1269  			LIST_REMOVE(m, nd_tq);
1270  			LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq);
1271  		}
1272  	}
1273 }
1274 
1275 /*
1276  * nfs create service
1277  * now does a truncate to 0 length via. setattr if it already exists
1278  */
1279 int
1280 nfsrv_create(nfsd, slp, procp, mrq)
1281 	struct nfsrv_descript *nfsd;
1282 	struct nfssvc_sock *slp;
1283 	struct proc *procp;
1284 	struct mbuf **mrq;
1285 {
1286 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1287 	struct mbuf *nam = nfsd->nd_nam;
1288 	caddr_t dpos = nfsd->nd_dpos;
1289 	struct ucred *cred = &nfsd->nd_cr;
1290 	struct nfs_fattr *fp;
1291 	struct vattr va, dirfor, diraft;
1292 	struct nfsv2_sattr *sp;
1293 	u_int32_t *tl;
1294 	struct nameidata nd;
1295 	caddr_t cp;
1296 	int32_t t1;
1297 	caddr_t bpos;
1298 	int error = 0, cache, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1299 	int rdev = 0;
1300 	int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1301 	char *cp2;
1302 	struct mbuf *mb, *mb2, *mreq;
1303 	struct vnode *vp = NULL, *dirp = NULL;
1304 	nfsfh_t nfh;
1305 	fhandle_t *fhp;
1306 	u_quad_t frev, tempsize;
1307 	u_char cverf[NFSX_V3CREATEVERF];
1308 
1309 	nd.ni_cnd.cn_nameiop = 0;
1310 	fhp = &nfh.fh_generic;
1311 	nfsm_srvmtofh(fhp);
1312 	nfsm_srvnamesiz(len);
1313 	nd.ni_cnd.cn_cred = cred;
1314 	nd.ni_cnd.cn_nameiop = CREATE;
1315 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART;
1316 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1317 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1318 	if (dirp) {
1319 		if (v3)
1320 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
1321 				procp);
1322 		else {
1323 			vrele(dirp);
1324 			dirp = (struct vnode *)0;
1325 		}
1326 	}
1327 	if (error) {
1328 		nfsm_reply(NFSX_WCCDATA(v3));
1329 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1330 		if (dirp)
1331 			vrele(dirp);
1332 		return (0);
1333 	}
1334 	VATTR_NULL(&va);
1335 	if (v3) {
1336 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1337 		how = fxdr_unsigned(int, *tl);
1338 		switch (how) {
1339 		case NFSV3CREATE_GUARDED:
1340 			if (nd.ni_vp) {
1341 				error = EEXIST;
1342 				break;
1343 			}
1344 		case NFSV3CREATE_UNCHECKED:
1345 			nfsm_srvsattr(&va);
1346 			break;
1347 		case NFSV3CREATE_EXCLUSIVE:
1348 			nfsm_dissect(cp, caddr_t, NFSX_V3CREATEVERF);
1349 			memcpy(cverf, cp, NFSX_V3CREATEVERF);
1350 			exclusive_flag = 1;
1351 			if (nd.ni_vp == NULL)
1352 				va.va_mode = 0;
1353 			break;
1354 		};
1355 		va.va_type = VREG;
1356 	} else {
1357 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1358 		va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1359 		if (va.va_type == VNON)
1360 			va.va_type = VREG;
1361 		va.va_mode = nfstov_mode(sp->sa_mode);
1362 		switch (va.va_type) {
1363 		case VREG:
1364 			tsize = fxdr_unsigned(int32_t, sp->sa_size);
1365 			if (tsize != -1)
1366 				va.va_size = (u_quad_t)tsize;
1367 			break;
1368 		case VCHR:
1369 		case VBLK:
1370 		case VFIFO:
1371 			rdev = fxdr_unsigned(int32_t, sp->sa_size);
1372 			break;
1373 		default:
1374 			break;
1375 		};
1376 	}
1377 
1378 	/*
1379 	 * Iff doesn't exist, create it
1380 	 * otherwise just truncate to 0 length
1381 	 *   should I set the mode too ??
1382 	 */
1383 	if (nd.ni_vp == NULL) {
1384 		if (va.va_type == VREG || va.va_type == VSOCK) {
1385 			vrele(nd.ni_startdir);
1386 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1387 			error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1388 			if (!error) {
1389 				PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1390 				if (exclusive_flag) {
1391 					exclusive_flag = 0;
1392 					VATTR_NULL(&va);
1393 					memcpy((caddr_t)&va.va_atime, cverf,
1394 						NFSX_V3CREATEVERF);
1395 					error = VOP_SETATTR(nd.ni_vp, &va, cred,
1396 						procp);
1397 				}
1398 			}
1399 		} else if (va.va_type == VCHR || va.va_type == VBLK ||
1400 			va.va_type == VFIFO) {
1401 			if (va.va_type == VCHR && rdev == 0xffffffff)
1402 				va.va_type = VFIFO;
1403 			if (va.va_type != VFIFO &&
1404 			    (error = suser(cred, (u_short *)0))) {
1405 				vrele(nd.ni_startdir);
1406 				PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1407 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1408 				vput(nd.ni_dvp);
1409 				nfsm_reply(0);
1410 				return (error);
1411 			} else
1412 				va.va_rdev = (dev_t)rdev;
1413 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1414 			error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd,
1415 			    &va);
1416 			if (error) {
1417 				vrele(nd.ni_startdir);
1418 				nfsm_reply(0);
1419 			}
1420 			PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1421 			if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1422 				vrele(nd.ni_dvp);
1423 				vput(nd.ni_vp);
1424 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1425 				error = EINVAL;
1426 				nfsm_reply(0);
1427 			}
1428 		} else {
1429 			vrele(nd.ni_startdir);
1430 			PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1431 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1432 			vput(nd.ni_dvp);
1433 			error = ENXIO;
1434 		}
1435 		vp = nd.ni_vp;
1436 	} else {
1437 		vrele(nd.ni_startdir);
1438 		PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1439 		vp = nd.ni_vp;
1440 		if (nd.ni_dvp == vp)
1441 			vrele(nd.ni_dvp);
1442 		else
1443 			vput(nd.ni_dvp);
1444 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1445 		if (!error && va.va_size != -1) {
1446 			error = nfsrv_access(vp, VWRITE, cred,
1447 			    (nd.ni_cnd.cn_flags & RDONLY), procp, 0);
1448 			if (!error) {
1449 				nqsrv_getl(vp, ND_WRITE);
1450 				tempsize = va.va_size;
1451 				VATTR_NULL(&va);
1452 				va.va_size = tempsize;
1453 				error = VOP_SETATTR(vp, &va, cred,
1454 					 procp);
1455 			}
1456 		}
1457 		if (error)
1458 			vput(vp);
1459 	}
1460 	if (!error) {
1461 		memset((caddr_t)fhp, 0, sizeof(nfh));
1462 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1463 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
1464 		if (!error)
1465 			error = VOP_GETATTR(vp, &va, cred, procp);
1466 		vput(vp);
1467 	}
1468 	if (v3) {
1469 		if (exclusive_flag && !error &&
1470 			memcmp(cverf, (caddr_t)&va.va_atime, NFSX_V3CREATEVERF))
1471 			error = EEXIST;
1472 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1473 		vrele(dirp);
1474 	}
1475 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1476 	if (v3) {
1477 		if (!error) {
1478 			nfsm_srvpostop_fh(fhp);
1479 			nfsm_srvpostop_attr(0, &va);
1480 		}
1481 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1482 	} else {
1483 		nfsm_srvfhtom(fhp, v3);
1484 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1485 		nfsm_srvfillattr(&va, fp);
1486 	}
1487 	return (0);
1488 nfsmout:
1489 	if (dirp)
1490 		vrele(dirp);
1491 	if (nd.ni_cnd.cn_nameiop) {
1492 		vrele(nd.ni_startdir);
1493 		PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1494 	}
1495 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1496 	if (nd.ni_dvp == nd.ni_vp)
1497 		vrele(nd.ni_dvp);
1498 	else
1499 		vput(nd.ni_dvp);
1500 	if (nd.ni_vp)
1501 		vput(nd.ni_vp);
1502 	return (error);
1503 }
1504 
1505 /*
1506  * nfs v3 mknod service
1507  */
1508 int
1509 nfsrv_mknod(nfsd, slp, procp, mrq)
1510 	struct nfsrv_descript *nfsd;
1511 	struct nfssvc_sock *slp;
1512 	struct proc *procp;
1513 	struct mbuf **mrq;
1514 {
1515 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1516 	struct mbuf *nam = nfsd->nd_nam;
1517 	caddr_t dpos = nfsd->nd_dpos;
1518 	struct ucred *cred = &nfsd->nd_cr;
1519 	struct vattr va, dirfor, diraft;
1520 	u_int32_t *tl;
1521 	struct nameidata nd;
1522 	int32_t t1;
1523 	caddr_t bpos;
1524 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
1525 	u_int32_t major, minor;
1526 	enum vtype vtyp;
1527 	char *cp2;
1528 	struct mbuf *mb, *mb2, *mreq;
1529 	struct vnode *vp, *dirp = (struct vnode *)0;
1530 	nfsfh_t nfh;
1531 	fhandle_t *fhp;
1532 	u_quad_t frev;
1533 
1534 	nd.ni_cnd.cn_nameiop = 0;
1535 	fhp = &nfh.fh_generic;
1536 	nfsm_srvmtofh(fhp);
1537 	nfsm_srvnamesiz(len);
1538 	nd.ni_cnd.cn_cred = cred;
1539 	nd.ni_cnd.cn_nameiop = CREATE;
1540 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART;
1541 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1542 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1543 	if (dirp)
1544 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, procp);
1545 	if (error) {
1546 		nfsm_reply(NFSX_WCCDATA(1));
1547 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1548 		if (dirp)
1549 			vrele(dirp);
1550 		return (0);
1551 	}
1552 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1553 	vtyp = nfsv3tov_type(*tl);
1554 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1555 		vrele(nd.ni_startdir);
1556 		PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1557 		error = NFSERR_BADTYPE;
1558 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1559 		if (nd.ni_dvp == nd.ni_vp)
1560 			vrele(nd.ni_dvp);
1561 		else
1562 			vput(nd.ni_dvp);
1563 		if (nd.ni_vp)
1564 			vput(nd.ni_vp);
1565 		goto out;
1566 	}
1567 	VATTR_NULL(&va);
1568 	nfsm_srvsattr(&va);
1569 	if (vtyp == VCHR || vtyp == VBLK) {
1570 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1571 		major = fxdr_unsigned(u_int32_t, *tl++);
1572 		minor = fxdr_unsigned(u_int32_t, *tl);
1573 		va.va_rdev = makedev(major, minor);
1574 	}
1575 
1576 	/*
1577 	 * Iff doesn't exist, create it.
1578 	 */
1579 	if (nd.ni_vp) {
1580 		vrele(nd.ni_startdir);
1581 		PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1582 		error = EEXIST;
1583 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1584 		if (nd.ni_dvp == nd.ni_vp)
1585 			vrele(nd.ni_dvp);
1586 		else
1587 			vput(nd.ni_dvp);
1588 		vput(nd.ni_vp);
1589 		goto out;
1590 	}
1591 	va.va_type = vtyp;
1592 	if (vtyp == VSOCK) {
1593 		vrele(nd.ni_startdir);
1594 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1595 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1596 		if (!error)
1597 			PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1598 	} else {
1599 		if (va.va_type != VFIFO &&
1600 		    (error = suser(cred, (u_short *)0))) {
1601 			vrele(nd.ni_startdir);
1602 			PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1603 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1604 			vput(nd.ni_dvp);
1605 			goto out;
1606 		}
1607 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1608 		error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1609 		if (error) {
1610 			vrele(nd.ni_startdir);
1611 			goto out;
1612 		}
1613 		PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1614 		if (error)
1615 			goto out;
1616 		if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1617 			vrele(nd.ni_dvp);
1618 			vput(nd.ni_vp);
1619 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1620 			error = EINVAL;
1621 		}
1622 	}
1623 out:
1624 	vp = nd.ni_vp;
1625 	if (!error) {
1626 		memset((caddr_t)fhp, 0, sizeof(nfh));
1627 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1628 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
1629 		if (!error)
1630 			error = VOP_GETATTR(vp, &va, cred, procp);
1631 		vput(vp);
1632 	}
1633 	diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1634 	vrele(dirp);
1635 	nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
1636 	if (!error) {
1637 		nfsm_srvpostop_fh(fhp);
1638 		nfsm_srvpostop_attr(0, &va);
1639 	}
1640 	nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1641 	return (0);
1642 nfsmout:
1643 	if (dirp)
1644 		vrele(dirp);
1645 	if (nd.ni_cnd.cn_nameiop) {
1646 		vrele(nd.ni_startdir);
1647 		PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1648 	}
1649 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1650 	if (nd.ni_dvp == nd.ni_vp)
1651 		vrele(nd.ni_dvp);
1652 	else
1653 		vput(nd.ni_dvp);
1654 	if (nd.ni_vp)
1655 		vput(nd.ni_vp);
1656 	return (error);
1657 }
1658 
1659 /*
1660  * nfs remove service
1661  */
1662 int
1663 nfsrv_remove(nfsd, slp, procp, mrq)
1664 	struct nfsrv_descript *nfsd;
1665 	struct nfssvc_sock *slp;
1666 	struct proc *procp;
1667 	struct mbuf **mrq;
1668 {
1669 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1670 	struct mbuf *nam = nfsd->nd_nam;
1671 	caddr_t dpos = nfsd->nd_dpos;
1672 	struct ucred *cred = &nfsd->nd_cr;
1673 	struct nameidata nd;
1674 	u_int32_t *tl;
1675 	int32_t t1;
1676 	caddr_t bpos;
1677 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
1678 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1679 	char *cp2;
1680 	struct mbuf *mb, *mreq;
1681 	struct vnode *vp, *dirp;
1682 	struct vattr dirfor, diraft;
1683 	nfsfh_t nfh;
1684 	fhandle_t *fhp;
1685 	u_quad_t frev;
1686 
1687 #ifndef nolint
1688 	vp = (struct vnode *)0;
1689 #endif
1690 	fhp = &nfh.fh_generic;
1691 	nfsm_srvmtofh(fhp);
1692 	nfsm_srvnamesiz(len);
1693 	nd.ni_cnd.cn_cred = cred;
1694 	nd.ni_cnd.cn_nameiop = DELETE;
1695 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1696 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1697 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1698 	if (dirp) {
1699 		if (v3)
1700 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
1701 				procp);
1702 		else
1703 			vrele(dirp);
1704 	}
1705 	if (!error) {
1706 		vp = nd.ni_vp;
1707 		if (vp->v_type == VDIR &&
1708 		    (error = suser(cred, (u_short *)0)) != 0)
1709 			goto out;
1710 		/*
1711 		 * The root of a mounted filesystem cannot be deleted.
1712 		 */
1713 		if (vp->v_flag & VROOT) {
1714 			error = EBUSY;
1715 			goto out;
1716 		}
1717 out:
1718 		if (!error) {
1719 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1720 			nqsrv_getl(vp, ND_WRITE);
1721 			error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
1722 		} else {
1723 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1724 			if (nd.ni_dvp == vp)
1725 				vrele(nd.ni_dvp);
1726 			else
1727 				vput(nd.ni_dvp);
1728 			vput(vp);
1729 		}
1730 	}
1731 	if (dirp && v3) {
1732 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1733 		vrele(dirp);
1734 	}
1735 	nfsm_reply(NFSX_WCCDATA(v3));
1736 	if (v3) {
1737 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1738 		return (0);
1739 	}
1740 	nfsm_srvdone;
1741 }
1742 
1743 /*
1744  * nfs rename service
1745  */
1746 int
1747 nfsrv_rename(nfsd, slp, procp, mrq)
1748 	struct nfsrv_descript *nfsd;
1749 	struct nfssvc_sock *slp;
1750 	struct proc *procp;
1751 	struct mbuf **mrq;
1752 {
1753 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1754 	struct mbuf *nam = nfsd->nd_nam;
1755 	caddr_t dpos = nfsd->nd_dpos;
1756 	struct ucred *cred = &nfsd->nd_cr;
1757 	u_int32_t *tl;
1758 	int32_t t1;
1759 	caddr_t bpos;
1760 	int error = 0, cache, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
1761 	int tdirfor_ret = 1, tdiraft_ret = 1;
1762 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1763 	char *cp2;
1764 	struct mbuf *mb, *mreq;
1765 	struct nameidata fromnd, tond;
1766 	struct vnode *fvp, *tvp, *tdvp, *fdirp = (struct vnode *)0;
1767 	struct vnode *tdirp = (struct vnode *)0;
1768 	struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
1769 	nfsfh_t fnfh, tnfh;
1770 	fhandle_t *ffhp, *tfhp;
1771 	u_quad_t frev;
1772 	uid_t saved_uid;
1773 
1774 #ifndef nolint
1775 	fvp = (struct vnode *)0;
1776 #endif
1777 	ffhp = &fnfh.fh_generic;
1778 	tfhp = &tnfh.fh_generic;
1779 	fromnd.ni_cnd.cn_nameiop = 0;
1780 	tond.ni_cnd.cn_nameiop = 0;
1781 	nfsm_srvmtofh(ffhp);
1782 	nfsm_srvnamesiz(len);
1783 	/*
1784 	 * Remember our original uid so that we can reset cr_uid before
1785 	 * the second nfs_namei() call, in case it is remapped.
1786 	 */
1787 	saved_uid = cred->cr_uid;
1788 	fromnd.ni_cnd.cn_cred = cred;
1789 	fromnd.ni_cnd.cn_nameiop = DELETE;
1790 	fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART;
1791 	error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md,
1792 		&dpos, &fdirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1793 	if (fdirp) {
1794 		if (v3)
1795 			fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred,
1796 				procp);
1797 		else {
1798 			vrele(fdirp);
1799 			fdirp = (struct vnode *)0;
1800 		}
1801 	}
1802 	if (error) {
1803 		nfsm_reply(2 * NFSX_WCCDATA(v3));
1804 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1805 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1806 		if (fdirp)
1807 			vrele(fdirp);
1808 		return (0);
1809 	}
1810 	fvp = fromnd.ni_vp;
1811 	nfsm_srvmtofh(tfhp);
1812 	nfsm_strsiz(len2, NFS_MAXNAMLEN);
1813 	cred->cr_uid = saved_uid;
1814 	tond.ni_cnd.cn_cred = cred;
1815 	tond.ni_cnd.cn_nameiop = RENAME;
1816 	tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART;
1817 	error = nfs_namei(&tond, tfhp, len2, slp, nam, &md,
1818 		&dpos, &tdirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1819 	if (tdirp) {
1820 		if (v3)
1821 			tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred,
1822 				procp);
1823 		else {
1824 			vrele(tdirp);
1825 			tdirp = (struct vnode *)0;
1826 		}
1827 	}
1828 	if (error) {
1829 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1830 		vrele(fromnd.ni_dvp);
1831 		vrele(fvp);
1832 		goto out1;
1833 	}
1834 	tdvp = tond.ni_dvp;
1835 	tvp = tond.ni_vp;
1836 	if (tvp != NULL) {
1837 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1838 			if (v3)
1839 				error = EEXIST;
1840 			else
1841 				error = EISDIR;
1842 			goto out;
1843 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1844 			if (v3)
1845 				error = EEXIST;
1846 			else
1847 				error = ENOTDIR;
1848 			goto out;
1849 		}
1850 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1851 			if (v3)
1852 				error = EXDEV;
1853 			else
1854 				error = ENOTEMPTY;
1855 			goto out;
1856 		}
1857 	}
1858 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1859 		if (v3)
1860 			error = EXDEV;
1861 		else
1862 			error = ENOTEMPTY;
1863 		goto out;
1864 	}
1865 	if (fvp->v_mount != tdvp->v_mount) {
1866 		if (v3)
1867 			error = EXDEV;
1868 		else
1869 			error = ENOTEMPTY;
1870 		goto out;
1871 	}
1872 	if (fvp == tdvp) {
1873 		if (v3)
1874 			error = EINVAL;
1875 		else
1876 			error = ENOTEMPTY;
1877 	}
1878 	/*
1879 	 * If source is the same as the destination (that is the
1880 	 * same vnode with the same name in the same directory),
1881 	 * then there is nothing to do.
1882 	 */
1883 	if (fvp == tvp && fromnd.ni_dvp == tdvp &&
1884 	    fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
1885 	    !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
1886 	      fromnd.ni_cnd.cn_namelen))
1887 		error = -1;
1888 out:
1889 	if (!error) {
1890 		nqsrv_getl(fromnd.ni_dvp, ND_WRITE);
1891 		nqsrv_getl(tdvp, ND_WRITE);
1892 		if (tvp) {
1893 			nqsrv_getl(tvp, ND_WRITE);
1894 		}
1895 		error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
1896 				   tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
1897 	} else {
1898 		VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd);
1899 		if (tdvp == tvp)
1900 			vrele(tdvp);
1901 		else
1902 			vput(tdvp);
1903 		if (tvp)
1904 			vput(tvp);
1905 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1906 		vrele(fromnd.ni_dvp);
1907 		vrele(fvp);
1908 		if (error == -1)
1909 			error = 0;
1910 	}
1911 	vrele(tond.ni_startdir);
1912 	PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
1913 out1:
1914 	if (fdirp) {
1915 		fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, procp);
1916 		vrele(fdirp);
1917 	}
1918 	if (tdirp) {
1919 		tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, procp);
1920 		vrele(tdirp);
1921 	}
1922 	vrele(fromnd.ni_startdir);
1923 	PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
1924 	nfsm_reply(2 * NFSX_WCCDATA(v3));
1925 	if (v3) {
1926 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1927 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1928 	}
1929 	return (0);
1930 
1931 nfsmout:
1932 	if (fdirp)
1933 		vrele(fdirp);
1934 	if (tdirp)
1935 		vrele(tdirp);
1936 	if (tond.ni_cnd.cn_nameiop) {
1937 		vrele(tond.ni_startdir);
1938 		PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
1939 	}
1940 	if (fromnd.ni_cnd.cn_nameiop) {
1941 		vrele(fromnd.ni_startdir);
1942 		PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
1943 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1944 		vrele(fromnd.ni_dvp);
1945 		vrele(fvp);
1946 	}
1947 	return (error);
1948 }
1949 
1950 /*
1951  * nfs link service
1952  */
1953 int
1954 nfsrv_link(nfsd, slp, procp, mrq)
1955 	struct nfsrv_descript *nfsd;
1956 	struct nfssvc_sock *slp;
1957 	struct proc *procp;
1958 	struct mbuf **mrq;
1959 {
1960 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1961 	struct mbuf *nam = nfsd->nd_nam;
1962 	caddr_t dpos = nfsd->nd_dpos;
1963 	struct ucred *cred = &nfsd->nd_cr;
1964 	struct nameidata nd;
1965 	u_int32_t *tl;
1966 	int32_t t1;
1967 	caddr_t bpos;
1968 	int error = 0, rdonly, cache, len, dirfor_ret = 1, diraft_ret = 1;
1969 	int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
1970 	char *cp2;
1971 	struct mbuf *mb, *mreq;
1972 	struct vnode *vp, *xp, *dirp = (struct vnode *)0;
1973 	struct vattr dirfor, diraft, at;
1974 	nfsfh_t nfh, dnfh;
1975 	fhandle_t *fhp, *dfhp;
1976 	u_quad_t frev;
1977 
1978 	fhp = &nfh.fh_generic;
1979 	dfhp = &dnfh.fh_generic;
1980 	nfsm_srvmtofh(fhp);
1981 	nfsm_srvmtofh(dfhp);
1982 	nfsm_srvnamesiz(len);
1983 	error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, slp, nam,
1984 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1985 	if (error) {
1986 		nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
1987 		nfsm_srvpostop_attr(getret, &at);
1988 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1989 		return (0);
1990 	}
1991 	if (vp->v_type == VDIR && (error = suser(cred, (u_short *)0)) != 0)
1992 		goto out1;
1993 	nd.ni_cnd.cn_cred = cred;
1994 	nd.ni_cnd.cn_nameiop = CREATE;
1995 	nd.ni_cnd.cn_flags = LOCKPARENT;
1996 	error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos,
1997 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1998 	if (dirp) {
1999 		if (v3)
2000 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2001 				procp);
2002 		else {
2003 			vrele(dirp);
2004 			dirp = (struct vnode *)0;
2005 		}
2006 	}
2007 	if (error)
2008 		goto out1;
2009 	xp = nd.ni_vp;
2010 	if (xp != NULL) {
2011 		error = EEXIST;
2012 		goto out;
2013 	}
2014 	xp = nd.ni_dvp;
2015 	if (vp->v_mount != xp->v_mount)
2016 		error = EXDEV;
2017 out:
2018 	if (!error) {
2019 		nqsrv_getl(vp, ND_WRITE);
2020 		nqsrv_getl(xp, ND_WRITE);
2021 		error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2022 	} else {
2023 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2024 		if (nd.ni_dvp == nd.ni_vp)
2025 			vrele(nd.ni_dvp);
2026 		else
2027 			vput(nd.ni_dvp);
2028 		if (nd.ni_vp)
2029 			vrele(nd.ni_vp);
2030 	}
2031 out1:
2032 	if (v3)
2033 		getret = VOP_GETATTR(vp, &at, cred, procp);
2034 	if (dirp) {
2035 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2036 		vrele(dirp);
2037 	}
2038 	vrele(vp);
2039 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2040 	if (v3) {
2041 		nfsm_srvpostop_attr(getret, &at);
2042 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2043 		return (0);
2044 	}
2045 	nfsm_srvdone;
2046 }
2047 
2048 /*
2049  * nfs symbolic link service
2050  */
2051 int
2052 nfsrv_symlink(nfsd, slp, procp, mrq)
2053 	struct nfsrv_descript *nfsd;
2054 	struct nfssvc_sock *slp;
2055 	struct proc *procp;
2056 	struct mbuf **mrq;
2057 {
2058 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2059 	struct mbuf *nam = nfsd->nd_nam;
2060 	caddr_t dpos = nfsd->nd_dpos;
2061 	struct ucred *cred = &nfsd->nd_cr;
2062 	struct vattr va, dirfor, diraft;
2063 	struct nameidata nd;
2064 	u_int32_t *tl;
2065 	int32_t t1;
2066 	struct nfsv2_sattr *sp;
2067 	char *bpos, *pathcp = NULL, *cp2;
2068 	struct uio io;
2069 	struct iovec iv;
2070 	int error = 0, cache, len, len2, dirfor_ret = 1, diraft_ret = 1;
2071 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2072 	struct mbuf *mb, *mreq, *mb2;
2073 	struct vnode *dirp = (struct vnode *)0;
2074 	nfsfh_t nfh;
2075 	fhandle_t *fhp;
2076 	u_quad_t frev;
2077 
2078 	nd.ni_cnd.cn_nameiop = 0;
2079 	fhp = &nfh.fh_generic;
2080 	nfsm_srvmtofh(fhp);
2081 	nfsm_srvnamesiz(len);
2082 	nd.ni_cnd.cn_cred = cred;
2083 	nd.ni_cnd.cn_nameiop = CREATE;
2084 	nd.ni_cnd.cn_flags = LOCKPARENT | SAVESTART;
2085 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2086 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2087 	if (dirp) {
2088 		if (v3)
2089 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2090 				procp);
2091 		else {
2092 			vrele(dirp);
2093 			dirp = (struct vnode *)0;
2094 		}
2095 	}
2096 	if (error)
2097 		goto out;
2098 	VATTR_NULL(&va);
2099 	if (v3)
2100 		nfsm_srvsattr(&va);
2101 	nfsm_strsiz(len2, NFS_MAXPATHLEN);
2102 	pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK);
2103 	iv.iov_base = pathcp;
2104 	iv.iov_len = len2;
2105 	io.uio_resid = len2;
2106 	io.uio_offset = 0;
2107 	io.uio_iov = &iv;
2108 	io.uio_iovcnt = 1;
2109 	io.uio_segflg = UIO_SYSSPACE;
2110 	io.uio_rw = UIO_READ;
2111 	io.uio_procp = (struct proc *)0;
2112 	nfsm_mtouio(&io, len2);
2113 	if (!v3) {
2114 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2115 		va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
2116 	}
2117 	*(pathcp + len2) = '\0';
2118 	if (nd.ni_vp) {
2119 		vrele(nd.ni_startdir);
2120 		PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
2121 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2122 		if (nd.ni_dvp == nd.ni_vp)
2123 			vrele(nd.ni_dvp);
2124 		else
2125 			vput(nd.ni_dvp);
2126 		vrele(nd.ni_vp);
2127 		error = EEXIST;
2128 		goto out;
2129 	}
2130 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2131 	error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp);
2132 	if (error)
2133 		vrele(nd.ni_startdir);
2134 	else {
2135 	    if (v3) {
2136 		memset((caddr_t)fhp, 0, sizeof(nfh));
2137 		fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2138 		error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2139 		if (!error)
2140 		    error = VOP_GETATTR(nd.ni_vp, &va, cred,
2141 					procp);
2142 		vput(nd.ni_vp);
2143 	    } else {
2144 		vrele(nd.ni_startdir);
2145 		vput(nd.ni_vp);
2146 	    }
2147 	    PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
2148 	}
2149 out:
2150 	if (pathcp)
2151 		free(pathcp, M_TEMP);
2152 	if (dirp) {
2153 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2154 		vrele(dirp);
2155 	}
2156 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2157 	if (v3) {
2158 		if (!error) {
2159 			nfsm_srvpostop_fh(fhp);
2160 			nfsm_srvpostop_attr(0, &va);
2161 		}
2162 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2163 	}
2164 	return (0);
2165 nfsmout:
2166 	if (nd.ni_cnd.cn_nameiop) {
2167 		vrele(nd.ni_startdir);
2168 		PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
2169 	}
2170 	if (dirp)
2171 		vrele(dirp);
2172 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2173 	if (nd.ni_dvp == nd.ni_vp)
2174 		vrele(nd.ni_dvp);
2175 	else
2176 		vput(nd.ni_dvp);
2177 	if (nd.ni_vp)
2178 		vrele(nd.ni_vp);
2179 	if (pathcp)
2180 		free(pathcp, M_TEMP);
2181 	return (error);
2182 }
2183 
2184 /*
2185  * nfs mkdir service
2186  */
2187 int
2188 nfsrv_mkdir(nfsd, slp, procp, mrq)
2189 	struct nfsrv_descript *nfsd;
2190 	struct nfssvc_sock *slp;
2191 	struct proc *procp;
2192 	struct mbuf **mrq;
2193 {
2194 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2195 	struct mbuf *nam = nfsd->nd_nam;
2196 	caddr_t dpos = nfsd->nd_dpos;
2197 	struct ucred *cred = &nfsd->nd_cr;
2198 	struct vattr va, dirfor, diraft;
2199 	struct nfs_fattr *fp;
2200 	struct nameidata nd;
2201 	caddr_t cp;
2202 	u_int32_t *tl;
2203 	int32_t t1;
2204 	caddr_t bpos;
2205 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2206 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2207 	char *cp2;
2208 	struct mbuf *mb, *mb2, *mreq;
2209 	struct vnode *vp, *dirp = (struct vnode *)0;
2210 	nfsfh_t nfh;
2211 	fhandle_t *fhp;
2212 	u_quad_t frev;
2213 
2214 	fhp = &nfh.fh_generic;
2215 	nfsm_srvmtofh(fhp);
2216 	nfsm_srvnamesiz(len);
2217 	nd.ni_cnd.cn_cred = cred;
2218 	nd.ni_cnd.cn_nameiop = CREATE;
2219 	nd.ni_cnd.cn_flags = LOCKPARENT;
2220 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2221 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2222 	if (dirp) {
2223 		if (v3)
2224 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2225 				procp);
2226 		else {
2227 			vrele(dirp);
2228 			dirp = (struct vnode *)0;
2229 		}
2230 	}
2231 	if (error) {
2232 		nfsm_reply(NFSX_WCCDATA(v3));
2233 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2234 		if (dirp)
2235 			vrele(dirp);
2236 		return (0);
2237 	}
2238 	VATTR_NULL(&va);
2239 	if (v3) {
2240 		nfsm_srvsattr(&va);
2241 	} else {
2242 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2243 		va.va_mode = nfstov_mode(*tl++);
2244 	}
2245 	va.va_type = VDIR;
2246 	vp = nd.ni_vp;
2247 	if (vp != NULL) {
2248 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2249 		if (nd.ni_dvp == vp)
2250 			vrele(nd.ni_dvp);
2251 		else
2252 			vput(nd.ni_dvp);
2253 		vrele(vp);
2254 		error = EEXIST;
2255 		goto out;
2256 	}
2257 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2258 	error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
2259 	if (!error) {
2260 		vp = nd.ni_vp;
2261 		memset((caddr_t)fhp, 0, sizeof(nfh));
2262 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2263 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
2264 		if (!error)
2265 			error = VOP_GETATTR(vp, &va, cred, procp);
2266 		vput(vp);
2267 	}
2268 out:
2269 	if (dirp) {
2270 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2271 		vrele(dirp);
2272 	}
2273 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2274 	if (v3) {
2275 		if (!error) {
2276 			nfsm_srvpostop_fh(fhp);
2277 			nfsm_srvpostop_attr(0, &va);
2278 		}
2279 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2280 	} else {
2281 		nfsm_srvfhtom(fhp, v3);
2282 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2283 		nfsm_srvfillattr(&va, fp);
2284 	}
2285 	return (0);
2286 nfsmout:
2287 	if (dirp)
2288 		vrele(dirp);
2289 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2290 	if (nd.ni_dvp == nd.ni_vp)
2291 		vrele(nd.ni_dvp);
2292 	else
2293 		vput(nd.ni_dvp);
2294 	if (nd.ni_vp)
2295 		vrele(nd.ni_vp);
2296 	return (error);
2297 }
2298 
2299 /*
2300  * nfs rmdir service
2301  */
2302 int
2303 nfsrv_rmdir(nfsd, slp, procp, mrq)
2304 	struct nfsrv_descript *nfsd;
2305 	struct nfssvc_sock *slp;
2306 	struct proc *procp;
2307 	struct mbuf **mrq;
2308 {
2309 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2310 	struct mbuf *nam = nfsd->nd_nam;
2311 	caddr_t dpos = nfsd->nd_dpos;
2312 	struct ucred *cred = &nfsd->nd_cr;
2313 	u_int32_t *tl;
2314 	int32_t t1;
2315 	caddr_t bpos;
2316 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2317 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2318 	char *cp2;
2319 	struct mbuf *mb, *mreq;
2320 	struct vnode *vp, *dirp = (struct vnode *)0;
2321 	struct vattr dirfor, diraft;
2322 	nfsfh_t nfh;
2323 	fhandle_t *fhp;
2324 	struct nameidata nd;
2325 	u_quad_t frev;
2326 
2327 	fhp = &nfh.fh_generic;
2328 	nfsm_srvmtofh(fhp);
2329 	nfsm_srvnamesiz(len);
2330 	nd.ni_cnd.cn_cred = cred;
2331 	nd.ni_cnd.cn_nameiop = DELETE;
2332 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
2333 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2334 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2335 	if (dirp) {
2336 		if (v3)
2337 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2338 				procp);
2339 		else {
2340 			vrele(dirp);
2341 			dirp = (struct vnode *)0;
2342 		}
2343 	}
2344 	if (error) {
2345 		nfsm_reply(NFSX_WCCDATA(v3));
2346 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2347 		if (dirp)
2348 			vrele(dirp);
2349 		return (0);
2350 	}
2351 	vp = nd.ni_vp;
2352 	if (vp->v_type != VDIR) {
2353 		error = ENOTDIR;
2354 		goto out;
2355 	}
2356 	/*
2357 	 * No rmdir "." please.
2358 	 */
2359 	if (nd.ni_dvp == vp) {
2360 		error = EINVAL;
2361 		goto out;
2362 	}
2363 	/*
2364 	 * The root of a mounted filesystem cannot be deleted.
2365 	 */
2366 	if (vp->v_flag & VROOT)
2367 		error = EBUSY;
2368 out:
2369 	if (!error) {
2370 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
2371 		nqsrv_getl(vp, ND_WRITE);
2372 		error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2373 	} else {
2374 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2375 		if (nd.ni_dvp == nd.ni_vp)
2376 			vrele(nd.ni_dvp);
2377 		else
2378 			vput(nd.ni_dvp);
2379 		vput(vp);
2380 	}
2381 	if (dirp) {
2382 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2383 		vrele(dirp);
2384 	}
2385 	nfsm_reply(NFSX_WCCDATA(v3));
2386 	if (v3) {
2387 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2388 		return (0);
2389 	}
2390 	nfsm_srvdone;
2391 }
2392 
2393 /*
2394  * nfs readdir service
2395  * - mallocs what it thinks is enough to read
2396  *	count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
2397  * - calls VOP_READDIR()
2398  * - loops around building the reply
2399  *	if the output generated exceeds count break out of loop
2400  *	The nfsm_clget macro is used here so that the reply will be packed
2401  *	tightly in mbuf clusters.
2402  * - it only knows that it has encountered eof when the VOP_READDIR()
2403  *	reads nothing
2404  * - as such one readdir rpc will return eof false although you are there
2405  *	and then the next will return eof
2406  * - it trims out records with d_fileno == 0
2407  *	this doesn't matter for Unix clients, but they might confuse clients
2408  *	for other os'.
2409  * - it trims out records with d_type == DT_WHT
2410  *	these cannot be seen through NFS (unless we extend the protocol)
2411  * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2412  *	than requested, but this may not apply to all filesystems. For
2413  *	example, client NFS does not { although it is never remote mounted
2414  *	anyhow }
2415  *     The alternate call nfsrv_readdirplus() does lookups as well.
2416  * PS: The NFS protocol spec. does not clarify what the "count" byte
2417  *	argument is a count of.. just name strings and file id's or the
2418  *	entire reply rpc or ...
2419  *	I tried just file name and id sizes and it confused the Sun client,
2420  *	so I am using the full rpc size now. The "paranoia.." comment refers
2421  *	to including the status longwords that are not a part of the dir.
2422  *	"entry" structures, but are in the rpc.
2423  */
2424 struct flrep {
2425 	nfsuint64 fl_off;
2426 	u_int32_t fl_postopok;
2427 	u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2428 	u_int32_t fl_fhok;
2429 	u_int32_t fl_fhsize;
2430 	u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
2431 };
2432 
2433 int
2434 nfsrv_readdir(nfsd, slp, procp, mrq)
2435 	struct nfsrv_descript *nfsd;
2436 	struct nfssvc_sock *slp;
2437 	struct proc *procp;
2438 	struct mbuf **mrq;
2439 {
2440 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2441 	struct mbuf *nam = nfsd->nd_nam;
2442 	caddr_t dpos = nfsd->nd_dpos;
2443 	struct ucred *cred = &nfsd->nd_cr;
2444 	char *bp, *be;
2445 	struct mbuf *mp;
2446 	struct dirent *dp;
2447 	caddr_t cp;
2448 	u_int32_t *tl;
2449 	int32_t t1;
2450 	caddr_t bpos;
2451 	struct mbuf *mb, *mb2, *mreq, *mp2;
2452 	char *cpos, *cend, *cp2, *rbuf;
2453 	struct vnode *vp;
2454 	struct vattr at;
2455 	nfsfh_t nfh;
2456 	fhandle_t *fhp;
2457 	struct uio io;
2458 	struct iovec iv;
2459 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2460 	int siz, cnt, fullsiz, eofflag, rdonly, cache, ncookies;
2461 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2462 	u_quad_t frev, off, toff, verf;
2463 	off_t *cookies = NULL, *cookiep;
2464 	nfsuint64 jar;
2465 
2466 	fhp = &nfh.fh_generic;
2467 	nfsm_srvmtofh(fhp);
2468 	if (v3) {
2469 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2470 		toff = fxdr_hyper(tl);
2471 		tl += 2;
2472 		verf = fxdr_hyper(tl);
2473 		tl += 2;
2474 	} else {
2475 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2476 		toff = fxdr_unsigned(u_quad_t, *tl++);
2477 	}
2478 	off = toff;
2479 	cnt = fxdr_unsigned(int, *tl);
2480 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2481 	xfer = NFS_SRVMAXDATA(nfsd);
2482 	if (siz > xfer)
2483 		siz = xfer;
2484 	fullsiz = siz;
2485 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
2486 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2487 	if (!error && vp->v_type != VDIR) {
2488 		error = ENOTDIR;
2489 		vput(vp);
2490 	}
2491 	if (error) {
2492 		nfsm_reply(NFSX_UNSIGNED);
2493 		nfsm_srvpostop_attr(getret, &at);
2494 		return (0);
2495 	}
2496 	nqsrv_getl(vp, ND_READ);
2497 	if (v3) {
2498 		error = getret = VOP_GETATTR(vp, &at, cred, procp);
2499 #ifdef NFS3_STRICTVERF
2500 		/*
2501 		 * XXX This check is too strict for Solaris 2.5 clients.
2502 		 */
2503 		if (!error && toff && verf != at.va_filerev)
2504 			error = NFSERR_BAD_COOKIE;
2505 #endif
2506 	}
2507 	if (!error)
2508 		error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0);
2509 	if (error) {
2510 		vput(vp);
2511 		nfsm_reply(NFSX_POSTOPATTR(v3));
2512 		nfsm_srvpostop_attr(getret, &at);
2513 		return (0);
2514 	}
2515 	VOP_UNLOCK(vp, 0);
2516 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2517 again:
2518 	iv.iov_base = rbuf;
2519 	iv.iov_len = fullsiz;
2520 	io.uio_iov = &iv;
2521 	io.uio_iovcnt = 1;
2522 	io.uio_offset = (off_t)off;
2523 	io.uio_resid = fullsiz;
2524 	io.uio_segflg = UIO_SYSSPACE;
2525 	io.uio_rw = UIO_READ;
2526 	io.uio_procp = (struct proc *)0;
2527 	eofflag = 0;
2528 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2529 
2530 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2531 
2532 	off = (off_t)io.uio_offset;
2533 	if (!cookies && !error)
2534 		error = NFSERR_PERM;
2535 	if (v3) {
2536 		getret = VOP_GETATTR(vp, &at, cred, procp);
2537 		if (!error)
2538 			error = getret;
2539 	}
2540 
2541 	VOP_UNLOCK(vp, 0);
2542 	if (error) {
2543 		vrele(vp);
2544 		free((caddr_t)rbuf, M_TEMP);
2545 		if (cookies)
2546 			free((caddr_t)cookies, M_TEMP);
2547 		nfsm_reply(NFSX_POSTOPATTR(v3));
2548 		nfsm_srvpostop_attr(getret, &at);
2549 		return (0);
2550 	}
2551 	if (io.uio_resid) {
2552 		siz -= io.uio_resid;
2553 
2554 		/*
2555 		 * If nothing read, return eof
2556 		 * rpc reply
2557 		 */
2558 		if (siz == 0) {
2559 			vrele(vp);
2560 			nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2561 				2 * NFSX_UNSIGNED);
2562 			if (v3) {
2563 				nfsm_srvpostop_attr(getret, &at);
2564 				nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2565 				txdr_hyper(at.va_filerev, tl);
2566 				tl += 2;
2567 			} else
2568 				nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2569 			*tl++ = nfs_false;
2570 			*tl = nfs_true;
2571 			free((caddr_t)rbuf, M_TEMP);
2572 			free((caddr_t)cookies, M_TEMP);
2573 			return (0);
2574 		}
2575 	}
2576 
2577 	/*
2578 	 * Check for degenerate cases of nothing useful read.
2579 	 * If so go try again
2580 	 */
2581 	cpos = rbuf;
2582 	cend = rbuf + siz;
2583 	dp = (struct dirent *)cpos;
2584 	cookiep = cookies;
2585 
2586 	while (cpos < cend && ncookies > 0 &&
2587 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2588 		cpos += dp->d_reclen;
2589 		dp = (struct dirent *)cpos;
2590 		cookiep++;
2591 		ncookies--;
2592 	}
2593 	if (cpos >= cend || ncookies == 0) {
2594 		toff = off;
2595 		siz = fullsiz;
2596 		goto again;
2597 	}
2598 
2599 	len = 3 * NFSX_UNSIGNED;	/* paranoia, probably can be 0 */
2600 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2601 	if (v3) {
2602 		nfsm_srvpostop_attr(getret, &at);
2603 		nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2604 		txdr_hyper(at.va_filerev, tl);
2605 	}
2606 	mp = mp2 = mb;
2607 	bp = bpos;
2608 	be = bp + M_TRAILINGSPACE(mp);
2609 
2610 	/* Loop through the records and build reply */
2611 	while (cpos < cend && ncookies > 0) {
2612 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2613 			nlen = dp->d_namlen;
2614 			rem = nfsm_rndup(nlen)-nlen;
2615 			len += (4 * NFSX_UNSIGNED + nlen + rem);
2616 			if (v3)
2617 				len += 2 * NFSX_UNSIGNED;
2618 			if (len > cnt) {
2619 				eofflag = 0;
2620 				break;
2621 			}
2622 			/*
2623 			 * Build the directory record xdr from
2624 			 * the dirent entry.
2625 			 */
2626 			nfsm_clget;
2627 			*tl = nfs_true;
2628 			bp += NFSX_UNSIGNED;
2629 			if (v3) {
2630 				nfsm_clget;
2631 				*tl = 0;
2632 				bp += NFSX_UNSIGNED;
2633 			}
2634 			nfsm_clget;
2635 			*tl = txdr_unsigned(dp->d_fileno);
2636 			bp += NFSX_UNSIGNED;
2637 			nfsm_clget;
2638 			*tl = txdr_unsigned(nlen);
2639 			bp += NFSX_UNSIGNED;
2640 
2641 			/* And loop around copying the name */
2642 			xfer = nlen;
2643 			cp = dp->d_name;
2644 			while (xfer > 0) {
2645 				nfsm_clget;
2646 				if ((bp+xfer) > be)
2647 					tsiz = be-bp;
2648 				else
2649 					tsiz = xfer;
2650 				memcpy(bp, cp, tsiz);
2651 				bp += tsiz;
2652 				xfer -= tsiz;
2653 				if (xfer > 0)
2654 					cp += tsiz;
2655 			}
2656 			/* And null pad to an int32_t boundary */
2657 			for (i = 0; i < rem; i++)
2658 				*bp++ = '\0';
2659 			nfsm_clget;
2660 
2661 			/* Finish off the record */
2662 			txdr_hyper(*cookiep, &jar);
2663 			if (v3) {
2664 				*tl = jar.nfsuquad[0];
2665 				bp += NFSX_UNSIGNED;
2666 				nfsm_clget;
2667 			}
2668 			*tl = jar.nfsuquad[1];
2669 			bp += NFSX_UNSIGNED;
2670 		}
2671 		cpos += dp->d_reclen;
2672 		dp = (struct dirent *)cpos;
2673 		cookiep++;
2674 		ncookies--;
2675 	}
2676 	vrele(vp);
2677 	nfsm_clget;
2678 	*tl = nfs_false;
2679 	bp += NFSX_UNSIGNED;
2680 	nfsm_clget;
2681 	if (eofflag)
2682 		*tl = nfs_true;
2683 	else
2684 		*tl = nfs_false;
2685 	bp += NFSX_UNSIGNED;
2686 	if (mp != mb) {
2687 		if (bp < be)
2688 			mp->m_len = bp - mtod(mp, caddr_t);
2689 	} else
2690 		mp->m_len += bp - bpos;
2691 	free((caddr_t)rbuf, M_TEMP);
2692 	free((caddr_t)cookies, M_TEMP);
2693 	nfsm_srvdone;
2694 }
2695 
2696 int
2697 nfsrv_readdirplus(nfsd, slp, procp, mrq)
2698 	struct nfsrv_descript *nfsd;
2699 	struct nfssvc_sock *slp;
2700 	struct proc *procp;
2701 	struct mbuf **mrq;
2702 {
2703 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2704 	struct mbuf *nam = nfsd->nd_nam;
2705 	caddr_t dpos = nfsd->nd_dpos;
2706 	struct ucred *cred = &nfsd->nd_cr;
2707 	char *bp, *be;
2708 	struct mbuf *mp;
2709 	struct dirent *dp;
2710 	caddr_t cp;
2711 	u_int32_t *tl;
2712 	int32_t t1;
2713 	caddr_t bpos;
2714 	struct mbuf *mb, *mb2, *mreq, *mp2;
2715 	char *cpos, *cend, *cp2, *rbuf;
2716 	struct vnode *vp, *nvp;
2717 	struct flrep fl;
2718 	nfsfh_t nfh;
2719 	fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
2720 	struct uio io;
2721 	struct iovec iv;
2722 	struct vattr va, at, *vap = &va;
2723 	struct nfs_fattr *fp;
2724 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2725 	int siz, cnt, fullsiz, eofflag, rdonly, cache, dirlen, ncookies;
2726 	u_quad_t frev, off, toff, verf;
2727 	off_t *cookies = NULL, *cookiep;
2728 
2729 	fhp = &nfh.fh_generic;
2730 	nfsm_srvmtofh(fhp);
2731 	nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2732 	toff = fxdr_hyper(tl);
2733 	tl += 2;
2734 	verf = fxdr_hyper(tl);
2735 	tl += 2;
2736 	siz = fxdr_unsigned(int, *tl++);
2737 	cnt = fxdr_unsigned(int, *tl);
2738 	off = toff;
2739 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2740 	xfer = NFS_SRVMAXDATA(nfsd);
2741 	if (siz > xfer)
2742 		siz = xfer;
2743 	fullsiz = siz;
2744 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
2745 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2746 	if (!error && vp->v_type != VDIR) {
2747 		error = ENOTDIR;
2748 		vput(vp);
2749 	}
2750 	if (error) {
2751 		nfsm_reply(NFSX_UNSIGNED);
2752 		nfsm_srvpostop_attr(getret, &at);
2753 		return (0);
2754 	}
2755 	error = getret = VOP_GETATTR(vp, &at, cred, procp);
2756 #ifdef NFS3_STRICTVERF
2757 	/*
2758 	 * XXX This check is too strict for Solaris 2.5 clients.
2759 	 */
2760 	if (!error && toff && verf != at.va_filerev)
2761 		error = NFSERR_BAD_COOKIE;
2762 #endif
2763 	if (!error) {
2764 		nqsrv_getl(vp, ND_READ);
2765 		error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0);
2766 	}
2767 	if (error) {
2768 		vput(vp);
2769 		nfsm_reply(NFSX_V3POSTOPATTR);
2770 		nfsm_srvpostop_attr(getret, &at);
2771 		return (0);
2772 	}
2773 	VOP_UNLOCK(vp, 0);
2774 
2775 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2776 again:
2777 	iv.iov_base = rbuf;
2778 	iv.iov_len = fullsiz;
2779 	io.uio_iov = &iv;
2780 	io.uio_iovcnt = 1;
2781 	io.uio_offset = (off_t)off;
2782 	io.uio_resid = fullsiz;
2783 	io.uio_segflg = UIO_SYSSPACE;
2784 	io.uio_rw = UIO_READ;
2785 	io.uio_procp = (struct proc *)0;
2786 	eofflag = 0;
2787 
2788 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2789 
2790 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2791 
2792 	off = (u_quad_t)io.uio_offset;
2793 	getret = VOP_GETATTR(vp, &at, cred, procp);
2794 
2795 	VOP_UNLOCK(vp, 0);
2796 
2797 	/*
2798 	 * If the VGET operation doesn't work for this filesystem,
2799 	 * we can't support readdirplus. Returning NOTSUPP should
2800 	 * make clients fall back to plain readdir.
2801 	 * There's no need to check for VPTOFH as well, we wouldn't
2802 	 * even be here otherwise.
2803 	 */
2804 	if (!getret) {
2805 		if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp)))
2806 			getret = (getret == EOPNOTSUPP) ?
2807 				NFSERR_NOTSUPP : NFSERR_IO;
2808 		else
2809 			vput(nvp);
2810 	}
2811 
2812 	if (!cookies && !error)
2813 		error = NFSERR_PERM;
2814 	if (!error)
2815 		error = getret;
2816 	if (error) {
2817 		vrele(vp);
2818 		if (cookies)
2819 			free((caddr_t)cookies, M_TEMP);
2820 		free((caddr_t)rbuf, M_TEMP);
2821 		nfsm_reply(NFSX_V3POSTOPATTR);
2822 		nfsm_srvpostop_attr(getret, &at);
2823 		return (0);
2824 	}
2825 	if (io.uio_resid) {
2826 		siz -= io.uio_resid;
2827 
2828 		/*
2829 		 * If nothing read, return eof
2830 		 * rpc reply
2831 		 */
2832 		if (siz == 0) {
2833 			vrele(vp);
2834 			nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
2835 				2 * NFSX_UNSIGNED);
2836 			nfsm_srvpostop_attr(getret, &at);
2837 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2838 			txdr_hyper(at.va_filerev, tl);
2839 			tl += 2;
2840 			*tl++ = nfs_false;
2841 			*tl = nfs_true;
2842 			free((caddr_t)cookies, M_TEMP);
2843 			free((caddr_t)rbuf, M_TEMP);
2844 			return (0);
2845 		}
2846 	}
2847 
2848 	/*
2849 	 * Check for degenerate cases of nothing useful read.
2850 	 * If so go try again
2851 	 */
2852 	cpos = rbuf;
2853 	cend = rbuf + siz;
2854 	dp = (struct dirent *)cpos;
2855 	cookiep = cookies;
2856 
2857 	while (cpos < cend && ncookies > 0 &&
2858 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2859 		cpos += dp->d_reclen;
2860 		dp = (struct dirent *)cpos;
2861 		cookiep++;
2862 		ncookies--;
2863 	}
2864 	if (cpos >= cend || ncookies == 0) {
2865 		toff = off;
2866 		siz = fullsiz;
2867 		goto again;
2868 	}
2869 
2870 	dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
2871 	nfsm_reply(cnt);
2872 	nfsm_srvpostop_attr(getret, &at);
2873 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2874 	txdr_hyper(at.va_filerev, tl);
2875 	mp = mp2 = mb;
2876 	bp = bpos;
2877 	be = bp + M_TRAILINGSPACE(mp);
2878 
2879 	/* Loop through the records and build reply */
2880 	while (cpos < cend && ncookies > 0) {
2881 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2882 			nlen = dp->d_namlen;
2883 			rem = nfsm_rndup(nlen)-nlen;
2884 
2885 			/*
2886 			 * For readdir_and_lookup get the vnode using
2887 			 * the file number.
2888 			 */
2889 			if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp))
2890 				goto invalid;
2891 			memset((caddr_t)nfhp, 0, NFSX_V3FH);
2892 			nfhp->fh_fsid =
2893 				nvp->v_mount->mnt_stat.f_fsid;
2894 			if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) {
2895 				vput(nvp);
2896 				goto invalid;
2897 			}
2898 			if (VOP_GETATTR(nvp, vap, cred, procp)) {
2899 				vput(nvp);
2900 				goto invalid;
2901 			}
2902 			vput(nvp);
2903 
2904 			/*
2905 			 * If either the dircount or maxcount will be
2906 			 * exceeded, get out now. Both of these lengths
2907 			 * are calculated conservatively, including all
2908 			 * XDR overheads.
2909 			 */
2910 			len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
2911 				NFSX_V3POSTOPATTR);
2912 			dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
2913 			if (len > cnt || dirlen > fullsiz) {
2914 				eofflag = 0;
2915 				break;
2916 			}
2917 
2918 			/*
2919 			 * Build the directory record xdr from
2920 			 * the dirent entry.
2921 			 */
2922 			fp = (struct nfs_fattr *)&fl.fl_fattr;
2923 			nfsm_srvfillattr(vap, fp);
2924 			fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
2925 			fl.fl_fhok = nfs_true;
2926 			fl.fl_postopok = nfs_true;
2927 			txdr_hyper(*cookiep, fl.fl_off.nfsuquad);
2928 
2929 			nfsm_clget;
2930 			*tl = nfs_true;
2931 			bp += NFSX_UNSIGNED;
2932 			nfsm_clget;
2933 			*tl = 0;
2934 			bp += NFSX_UNSIGNED;
2935 			nfsm_clget;
2936 			*tl = txdr_unsigned(dp->d_fileno);
2937 			bp += NFSX_UNSIGNED;
2938 			nfsm_clget;
2939 			*tl = txdr_unsigned(nlen);
2940 			bp += NFSX_UNSIGNED;
2941 
2942 			/* And loop around copying the name */
2943 			xfer = nlen;
2944 			cp = dp->d_name;
2945 			while (xfer > 0) {
2946 				nfsm_clget;
2947 				if ((bp + xfer) > be)
2948 					tsiz = be - bp;
2949 				else
2950 					tsiz = xfer;
2951 				memcpy(bp, cp, tsiz);
2952 				bp += tsiz;
2953 				xfer -= tsiz;
2954 				if (xfer > 0)
2955 					cp += tsiz;
2956 			}
2957 			/* And null pad to an int32_t boundary */
2958 			for (i = 0; i < rem; i++)
2959 				*bp++ = '\0';
2960 
2961 			/*
2962 			 * Now copy the flrep structure out.
2963 			 */
2964 			xfer = sizeof (struct flrep);
2965 			cp = (caddr_t)&fl;
2966 			while (xfer > 0) {
2967 				nfsm_clget;
2968 				if ((bp + xfer) > be)
2969 					tsiz = be - bp;
2970 				else
2971 					tsiz = xfer;
2972 				memcpy(bp, cp, tsiz);
2973 				bp += tsiz;
2974 				xfer -= tsiz;
2975 				if (xfer > 0)
2976 					cp += tsiz;
2977 			}
2978 		}
2979 invalid:
2980 		cpos += dp->d_reclen;
2981 		dp = (struct dirent *)cpos;
2982 		cookiep++;
2983 		ncookies--;
2984 	}
2985 	vrele(vp);
2986 	nfsm_clget;
2987 	*tl = nfs_false;
2988 	bp += NFSX_UNSIGNED;
2989 	nfsm_clget;
2990 	if (eofflag)
2991 		*tl = nfs_true;
2992 	else
2993 		*tl = nfs_false;
2994 	bp += NFSX_UNSIGNED;
2995 	if (mp != mb) {
2996 		if (bp < be)
2997 			mp->m_len = bp - mtod(mp, caddr_t);
2998 	} else
2999 		mp->m_len += bp - bpos;
3000 	free((caddr_t)cookies, M_TEMP);
3001 	free((caddr_t)rbuf, M_TEMP);
3002 	nfsm_srvdone;
3003 }
3004 
3005 /*
3006  * nfs commit service
3007  */
3008 int
3009 nfsrv_commit(nfsd, slp, procp, mrq)
3010 	struct nfsrv_descript *nfsd;
3011 	struct nfssvc_sock *slp;
3012 	struct proc *procp;
3013 	struct mbuf **mrq;
3014 {
3015 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3016 	struct mbuf *nam = nfsd->nd_nam;
3017 	caddr_t dpos = nfsd->nd_dpos;
3018 	struct ucred *cred = &nfsd->nd_cr;
3019 	struct vattr bfor, aft;
3020 	struct vnode *vp;
3021 	nfsfh_t nfh;
3022 	fhandle_t *fhp;
3023 	u_int32_t *tl;
3024 	int32_t t1;
3025 	caddr_t bpos;
3026 	int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt, cache;
3027 	char *cp2;
3028 	struct mbuf *mb, *mb2, *mreq;
3029 	u_quad_t frev, off;
3030 
3031 #ifndef nolint
3032 	cache = 0;
3033 #endif
3034 	fhp = &nfh.fh_generic;
3035 	nfsm_srvmtofh(fhp);
3036 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3037 
3038 	off = fxdr_hyper(tl);
3039 	tl += 2;
3040 	cnt = fxdr_unsigned(int, *tl);
3041 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3042 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3043 	if (error) {
3044 		nfsm_reply(2 * NFSX_UNSIGNED);
3045 		nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3046 		return (0);
3047 	}
3048 	for_ret = VOP_GETATTR(vp, &bfor, cred, procp);
3049 	error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, off + cnt, procp);
3050 	aft_ret = VOP_GETATTR(vp, &aft, cred, procp);
3051 	vput(vp);
3052 	nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3053 	nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3054 	if (!error) {
3055 		nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3056 		*tl++ = txdr_unsigned(boottime.tv_sec);
3057 		*tl = txdr_unsigned(boottime.tv_usec);
3058 	} else
3059 		return (0);
3060 	nfsm_srvdone;
3061 }
3062 
3063 /*
3064  * nfs statfs service
3065  */
3066 int
3067 nfsrv_statfs(nfsd, slp, procp, mrq)
3068 	struct nfsrv_descript *nfsd;
3069 	struct nfssvc_sock *slp;
3070 	struct proc *procp;
3071 	struct mbuf **mrq;
3072 {
3073 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3074 	struct mbuf *nam = nfsd->nd_nam;
3075 	caddr_t dpos = nfsd->nd_dpos;
3076 	struct ucred *cred = &nfsd->nd_cr;
3077 	struct statfs *sf;
3078 	struct nfs_statfs *sfp;
3079 	u_int32_t *tl;
3080 	int32_t t1;
3081 	caddr_t bpos;
3082 	int error = 0, rdonly, cache, getret = 1;
3083 	int v3 = (nfsd->nd_flag & ND_NFSV3);
3084 	char *cp2;
3085 	struct mbuf *mb, *mb2, *mreq;
3086 	struct vnode *vp;
3087 	struct vattr at;
3088 	nfsfh_t nfh;
3089 	fhandle_t *fhp;
3090 	struct statfs statfs;
3091 	u_quad_t frev, tval;
3092 
3093 #ifndef nolint
3094 	cache = 0;
3095 #endif
3096 	fhp = &nfh.fh_generic;
3097 	nfsm_srvmtofh(fhp);
3098 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3099 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3100 	if (error) {
3101 		nfsm_reply(NFSX_UNSIGNED);
3102 		nfsm_srvpostop_attr(getret, &at);
3103 		return (0);
3104 	}
3105 	sf = &statfs;
3106 	error = VFS_STATFS(vp->v_mount, sf, procp);
3107 	getret = VOP_GETATTR(vp, &at, cred, procp);
3108 	vput(vp);
3109 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3110 	if (v3)
3111 		nfsm_srvpostop_attr(getret, &at);
3112 	if (error)
3113 		return (0);
3114 	nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3115 	if (v3) {
3116 		tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_bsize);
3117 		txdr_hyper(tval, &sfp->sf_tbytes);
3118 		tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_bsize);
3119 		txdr_hyper(tval, &sfp->sf_fbytes);
3120 		tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_bsize);
3121 		txdr_hyper(tval, &sfp->sf_abytes);
3122 		tval = (u_quad_t)sf->f_files;
3123 		txdr_hyper(tval, &sfp->sf_tfiles);
3124 		tval = (u_quad_t)sf->f_ffree;
3125 		txdr_hyper(tval, &sfp->sf_ffiles);
3126 		txdr_hyper(tval, &sfp->sf_afiles);
3127 		sfp->sf_invarsec = 0;
3128 	} else {
3129 		sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3130 		sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
3131 		sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3132 		sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3133 		sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3134 	}
3135 	nfsm_srvdone;
3136 }
3137 
3138 /*
3139  * nfs fsinfo service
3140  */
3141 int
3142 nfsrv_fsinfo(nfsd, slp, procp, mrq)
3143 	struct nfsrv_descript *nfsd;
3144 	struct nfssvc_sock *slp;
3145 	struct proc *procp;
3146 	struct mbuf **mrq;
3147 {
3148 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3149 	struct mbuf *nam = nfsd->nd_nam;
3150 	caddr_t dpos = nfsd->nd_dpos;
3151 	struct ucred *cred = &nfsd->nd_cr;
3152 	u_int32_t *tl;
3153 	struct nfsv3_fsinfo *sip;
3154 	int32_t t1;
3155 	caddr_t bpos;
3156 	int error = 0, rdonly, cache, getret = 1, pref;
3157 	char *cp2;
3158 	struct mbuf *mb, *mb2, *mreq;
3159 	struct vnode *vp;
3160 	struct vattr at;
3161 	nfsfh_t nfh;
3162 	fhandle_t *fhp;
3163 	u_quad_t frev, maxfsize;
3164 	struct statfs sb;
3165 
3166 #ifndef nolint
3167 	cache = 0;
3168 #endif
3169 	fhp = &nfh.fh_generic;
3170 	nfsm_srvmtofh(fhp);
3171 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3172 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3173 	if (error) {
3174 		nfsm_reply(NFSX_UNSIGNED);
3175 		nfsm_srvpostop_attr(getret, &at);
3176 		return (0);
3177 	}
3178 
3179 	/* XXX Try to make a guess on the max file size. */
3180 	VFS_STATFS(vp->v_mount, &sb, (struct proc *)0);
3181 	maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
3182 
3183 	getret = VOP_GETATTR(vp, &at, cred, procp);
3184 	vput(vp);
3185 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3186 	nfsm_srvpostop_attr(getret, &at);
3187 	nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3188 
3189 	/*
3190 	 * XXX
3191 	 * There should be file system VFS OP(s) to get this information.
3192 	 * For now, assume ufs.
3193 	 */
3194 	if (slp->ns_so->so_type == SOCK_DGRAM)
3195 		pref = NFS_MAXDGRAMDATA;
3196 	else
3197 		pref = NFS_MAXDATA;
3198 	sip->fs_rtmax = txdr_unsigned(NFS_MAXDATA);
3199 	sip->fs_rtpref = txdr_unsigned(pref);
3200 	sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3201 	sip->fs_wtmax = txdr_unsigned(NFS_MAXDATA);
3202 	sip->fs_wtpref = txdr_unsigned(pref);
3203 	sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3204 	sip->fs_dtpref = txdr_unsigned(pref);
3205 	txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3206 	sip->fs_timedelta.nfsv3_sec = 0;
3207 	sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3208 	sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3209 		NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3210 		NFSV3FSINFO_CANSETTIME);
3211 	nfsm_srvdone;
3212 }
3213 
3214 /*
3215  * nfs pathconf service
3216  */
3217 int
3218 nfsrv_pathconf(nfsd, slp, procp, mrq)
3219 	struct nfsrv_descript *nfsd;
3220 	struct nfssvc_sock *slp;
3221 	struct proc *procp;
3222 	struct mbuf **mrq;
3223 {
3224 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3225 	struct mbuf *nam = nfsd->nd_nam;
3226 	caddr_t dpos = nfsd->nd_dpos;
3227 	struct ucred *cred = &nfsd->nd_cr;
3228 	u_int32_t *tl;
3229 	struct nfsv3_pathconf *pc;
3230 	int32_t t1;
3231 	caddr_t bpos;
3232 	int error = 0, rdonly, cache, getret = 1;
3233 	register_t linkmax, namemax, chownres, notrunc;
3234 	char *cp2;
3235 	struct mbuf *mb, *mb2, *mreq;
3236 	struct vnode *vp;
3237 	struct vattr at;
3238 	nfsfh_t nfh;
3239 	fhandle_t *fhp;
3240 	u_quad_t frev;
3241 
3242 #ifndef nolint
3243 	cache = 0;
3244 #endif
3245 	fhp = &nfh.fh_generic;
3246 	nfsm_srvmtofh(fhp);
3247 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3248 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3249 	if (error) {
3250 		nfsm_reply(NFSX_UNSIGNED);
3251 		nfsm_srvpostop_attr(getret, &at);
3252 		return (0);
3253 	}
3254 	error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3255 	if (!error)
3256 		error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3257 	if (!error)
3258 		error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3259 	if (!error)
3260 		error = VOP_PATHCONF(vp, _PC_NO_TRUNC, &notrunc);
3261 	getret = VOP_GETATTR(vp, &at, cred, procp);
3262 	vput(vp);
3263 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3264 	nfsm_srvpostop_attr(getret, &at);
3265 	if (error)
3266 		return (0);
3267 	nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3268 
3269 	pc->pc_linkmax = txdr_unsigned(linkmax);
3270 	pc->pc_namemax = txdr_unsigned(namemax);
3271 	pc->pc_notrunc = txdr_unsigned(notrunc);
3272 	pc->pc_chownrestricted = txdr_unsigned(chownres);
3273 
3274 	/*
3275 	 * These should probably be supported by VOP_PATHCONF(), but
3276 	 * until msdosfs is exportable (why would you want to?), the
3277 	 * Unix defaults should be ok.
3278 	 */
3279 	pc->pc_caseinsensitive = nfs_false;
3280 	pc->pc_casepreserving = nfs_true;
3281 	nfsm_srvdone;
3282 }
3283 
3284 /*
3285  * Null operation, used by clients to ping server
3286  */
3287 /* ARGSUSED */
3288 int
3289 nfsrv_null(nfsd, slp, procp, mrq)
3290 	struct nfsrv_descript *nfsd;
3291 	struct nfssvc_sock *slp;
3292 	struct proc *procp;
3293 	struct mbuf **mrq;
3294 {
3295 	struct mbuf *mrep = nfsd->nd_mrep;
3296 	caddr_t bpos;
3297 	int error = NFSERR_RETVOID, cache = 0;
3298 	struct mbuf *mb, *mreq;
3299 	u_quad_t frev;
3300 
3301 	nfsm_reply(0);
3302 	return (0);
3303 }
3304 
3305 /*
3306  * No operation, used for obsolete procedures
3307  */
3308 /* ARGSUSED */
3309 int
3310 nfsrv_noop(nfsd, slp, procp, mrq)
3311 	struct nfsrv_descript *nfsd;
3312 	struct nfssvc_sock *slp;
3313 	struct proc *procp;
3314 	struct mbuf **mrq;
3315 {
3316 	struct mbuf *mrep = nfsd->nd_mrep;
3317 	caddr_t bpos;
3318 	int error, cache = 0;
3319 	struct mbuf *mb, *mreq;
3320 	u_quad_t frev;
3321 
3322 	if (nfsd->nd_repstat)
3323 		error = nfsd->nd_repstat;
3324 	else
3325 		error = EPROCUNAVAIL;
3326 	nfsm_reply(0);
3327 	return (0);
3328 }
3329 
3330 /*
3331  * Perform access checking for vnodes obtained from file handles that would
3332  * refer to files already opened by a Unix client. You cannot just use
3333  * vn_writechk() and VOP_ACCESS() for two reasons.
3334  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3335  * 2 - The owner is to be given access irrespective of mode bits for some
3336  *     operations, so that processes that chmod after opening a file don't
3337  *     break. I don't like this because it opens a security hole, but since
3338  *     the nfs server opens a security hole the size of a barn door anyhow,
3339  *     what the heck.
3340  *
3341  * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3342  * will return EPERM instead of EACCESS. EPERM is always an error.
3343  */
3344 int
3345 nfsrv_access(vp, flags, cred, rdonly, p, override)
3346 	struct vnode *vp;
3347 	int flags;
3348 	struct ucred *cred;
3349 	int rdonly;
3350 	struct proc *p;
3351 {
3352 	struct vattr vattr;
3353 	int error;
3354 	if (flags & VWRITE) {
3355 		/* Just vn_writechk() changed to check rdonly */
3356 		/*
3357 		 * Disallow write attempts on read-only file systems;
3358 		 * unless the file is a socket or a block or character
3359 		 * device resident on the file system.
3360 		 */
3361 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3362 			switch (vp->v_type) {
3363 			case VREG:
3364 			case VDIR:
3365 			case VLNK:
3366 				return (EROFS);
3367 			default:
3368 				break;
3369 			}
3370 		}
3371 
3372 		/*
3373 		 * If the vnode is in use as a process's text,
3374 		 * we can't allow writing.
3375 		 */
3376 		if (vp->v_flag & VTEXT)
3377 			return (ETXTBSY);
3378 	}
3379 	error = VOP_GETATTR(vp, &vattr, cred, p);
3380 	if (error)
3381 		return (error);
3382 	error = VOP_ACCESS(vp, flags, cred, p);
3383 	/*
3384 	 * Allow certain operations for the owner (reads and writes
3385 	 * on files that are already open).
3386 	 */
3387 	if (override && error == EACCES && cred->cr_uid == vattr.va_uid)
3388 		error = 0;
3389 	return error;
3390 }
3391