xref: /netbsd-src/sys/nfs/nfs_serv.c (revision c0179c282a5968435315a82f4128c61372c68fc3)
1 /*	$NetBSD: nfs_serv.c,v 1.119 2006/11/09 09:53:57 yamt Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)nfs_serv.c	8.8 (Berkeley) 7/31/95
35  */
36 
37 /*
38  * nfs version 2 and 3 server calls to vnode ops
39  * - these routines generally have 3 phases
40  *   1 - break down and validate rpc request in mbuf list
41  *   2 - do the vnode ops for the request
42  *       (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
43  *   3 - build the rpc reply in an mbuf list
44  *   nb:
45  *	- do not mix the phases, since the nfsm_?? macros can return failures
46  *	  on a bad rpc or similar and do not do any vrele() or vput()'s
47  *
48  *      - the nfsm_reply() macro generates an nfs rpc reply with the nfs
49  *	error number iff error != 0 whereas
50  *	returning an error from the server function implies a fatal error
51  *	such as a badly constructed rpc request that should be dropped without
52  *	a reply.
53  *	For Version 3, nfsm_reply() does not return for the error case, since
54  *	most version 3 rpcs return more than the status for error cases.
55  */
56 
57 #include <sys/cdefs.h>
58 __KERNEL_RCSID(0, "$NetBSD: nfs_serv.c,v 1.119 2006/11/09 09:53:57 yamt Exp $");
59 
60 #include <sys/param.h>
61 #include <sys/systm.h>
62 #include <sys/proc.h>
63 #include <sys/file.h>
64 #include <sys/namei.h>
65 #include <sys/vnode.h>
66 #include <sys/mount.h>
67 #include <sys/socket.h>
68 #include <sys/socketvar.h>
69 #include <sys/mbuf.h>
70 #include <sys/dirent.h>
71 #include <sys/stat.h>
72 #include <sys/kernel.h>
73 #include <sys/hash.h>
74 #include <sys/kauth.h>
75 
76 #include <uvm/uvm.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 const nfstype nfsv2_type[9];
92 extern const nfstype nfsv3_type[9];
93 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
94 int nfsd_use_loan = 1;	/* use page-loan for READ OP */
95 
96 /*
97  * nfs v3 access service
98  */
99 int
100 nfsrv3_access(nfsd, slp, lwp, mrq)
101 	struct nfsrv_descript *nfsd;
102 	struct nfssvc_sock *slp;
103 	struct lwp *lwp;
104 	struct mbuf **mrq;
105 {
106 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
107 	struct mbuf *nam = nfsd->nd_nam;
108 	caddr_t dpos = nfsd->nd_dpos;
109 	kauth_cred_t cred = nfsd->nd_cr;
110 	struct vnode *vp;
111 	nfsrvfh_t nsfh;
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;
118 	struct vattr va;
119 	u_long inmode, testmode, outmode;
120 	u_quad_t frev;
121 
122 	nfsm_srvmtofh(&nsfh);
123 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
124 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly,
125 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
126 	if (error) {
127 		nfsm_reply(NFSX_UNSIGNED);
128 		nfsm_srvpostop_attr(1, (struct vattr *)0);
129 		return (0);
130 	}
131 	inmode = fxdr_unsigned(u_int32_t, *tl);
132 	outmode = 0;
133 	if ((inmode & NFSV3ACCESS_READ) &&
134 	    nfsrv_access(vp, VREAD, cred, rdonly, lwp, 0) == 0)
135 		outmode |= NFSV3ACCESS_READ;
136 	if (vp->v_type != VDIR) {
137 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
138 		if (testmode &&
139 		    nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0)
140 			outmode |= testmode;
141 		if ((inmode & NFSV3ACCESS_EXECUTE) &&
142 		    nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0)
143 			outmode |= NFSV3ACCESS_EXECUTE;
144 	} else {
145 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
146 		    NFSV3ACCESS_DELETE);
147 		if (testmode &&
148 		    nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 0) == 0)
149 			outmode |= testmode;
150 		if ((inmode & NFSV3ACCESS_LOOKUP) &&
151 		    nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0) == 0)
152 			outmode |= NFSV3ACCESS_LOOKUP;
153 	}
154 	getret = VOP_GETATTR(vp, &va, cred, lwp);
155 	vput(vp);
156 	nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
157 	nfsm_srvpostop_attr(getret, &va);
158 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
159 	*tl = txdr_unsigned(outmode);
160 	nfsm_srvdone;
161 }
162 
163 /*
164  * nfs getattr service
165  */
166 int
167 nfsrv_getattr(nfsd, slp, lwp, mrq)
168 	struct nfsrv_descript *nfsd;
169 	struct nfssvc_sock *slp;
170 	struct lwp *lwp;
171 	struct mbuf **mrq;
172 {
173 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
174 	struct mbuf *nam = nfsd->nd_nam;
175 	caddr_t dpos = nfsd->nd_dpos;
176 	kauth_cred_t cred = nfsd->nd_cr;
177 	struct nfs_fattr *fp;
178 	struct vattr va;
179 	struct vnode *vp;
180 	nfsrvfh_t nsfh;
181 	u_int32_t *tl;
182 	int32_t t1;
183 	caddr_t bpos;
184 	int error = 0, rdonly, cache = 0;
185 	char *cp2;
186 	struct mbuf *mb, *mreq;
187 	u_quad_t frev;
188 
189 	nfsm_srvmtofh(&nsfh);
190 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly,
191 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
192 	if (error) {
193 		nfsm_reply(0);
194 		return (0);
195 	}
196 	nqsrv_getl(vp, ND_READ);
197 	error = VOP_GETATTR(vp, &va, cred, lwp);
198 	vput(vp);
199 	nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
200 	if (error)
201 		return (0);
202 	nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
203 	nfsm_srvfillattr(&va, fp);
204 	nfsm_srvdone;
205 }
206 
207 /*
208  * nfs setattr service
209  */
210 int
211 nfsrv_setattr(nfsd, slp, lwp, mrq)
212 	struct nfsrv_descript *nfsd;
213 	struct nfssvc_sock *slp;
214 	struct lwp *lwp;
215 	struct mbuf **mrq;
216 {
217 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
218 	struct mbuf *nam = nfsd->nd_nam;
219 	caddr_t dpos = nfsd->nd_dpos;
220 	kauth_cred_t cred = nfsd->nd_cr;
221 	struct vattr va, preat;
222 	struct nfsv2_sattr *sp;
223 	struct nfs_fattr *fp;
224 	struct vnode *vp;
225 	nfsrvfh_t nsfh;
226 	u_int32_t *tl;
227 	int32_t t1;
228 	caddr_t bpos;
229 	int error = 0, rdonly, cache = 0, preat_ret = 1, postat_ret = 1;
230 	int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
231 	char *cp2;
232 	struct mbuf *mb, *mreq;
233 	u_quad_t frev;
234 	struct timespec guard;
235 	struct mount *mp = NULL;
236 
237 	memset(&guard, 0, sizeof guard);	/* XXX gcc */
238 
239 	nfsm_srvmtofh(&nsfh);
240 	if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL)
241 		return (ESTALE);
242 	vn_start_write(NULL, &mp, V_WAIT);
243 	VATTR_NULL(&va);
244 	if (v3) {
245 		nfsm_srvsattr(&va);
246 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
247 		gcheck = fxdr_unsigned(int, *tl);
248 		if (gcheck) {
249 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
250 			fxdr_nfsv3time(tl, &guard);
251 		}
252 	} else {
253 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
254 		/*
255 		 * Nah nah nah nah na nah
256 		 * There is a bug in the Sun client that puts 0xffff in the mode
257 		 * field of sattr when it should put in 0xffffffff. The u_short
258 		 * doesn't sign extend.
259 		 * --> check the low order 2 bytes for 0xffff
260 		 */
261 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
262 			va.va_mode = nfstov_mode(sp->sa_mode);
263 		if (sp->sa_uid != nfs_xdrneg1)
264 			va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
265 		if (sp->sa_gid != nfs_xdrneg1)
266 			va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
267 		if (sp->sa_size != nfs_xdrneg1)
268 			va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
269 		if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) {
270 #ifdef notyet
271 			fxdr_nfsv2time(&sp->sa_atime, &va.va_atime);
272 #else
273 			va.va_atime.tv_sec =
274 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
275 			va.va_atime.tv_nsec = 0;
276 #endif
277 		}
278 		if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1)
279 			fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime);
280 
281 	}
282 
283 	/*
284 	 * Now that we have all the fields, lets do it.
285 	 */
286 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam, &rdonly,
287 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
288 	if (error) {
289 		nfsm_reply(2 * NFSX_UNSIGNED);
290 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
291 		vn_finished_write(mp, 0);
292 		return (0);
293 	}
294 	nqsrv_getl(vp, ND_WRITE);
295 	if (v3) {
296 		error = preat_ret = VOP_GETATTR(vp, &preat, cred, lwp);
297 		if (!error && gcheck &&
298 			(preat.va_ctime.tv_sec != guard.tv_sec ||
299 			 preat.va_ctime.tv_nsec != guard.tv_nsec))
300 			error = NFSERR_NOT_SYNC;
301 		if (error) {
302 			vput(vp);
303 			nfsm_reply(NFSX_WCCDATA(v3));
304 			nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
305 			vn_finished_write(mp, 0);
306 			return (0);
307 		}
308 	}
309 
310 	/*
311 	 * If the size is being changed write acces is required, otherwise
312 	 * just check for a read only file system.
313 	 */
314 	if (va.va_size == ((u_quad_t)((quad_t) -1))) {
315 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
316 			error = EROFS;
317 			goto out;
318 		}
319 	} else {
320 		if (vp->v_type == VDIR) {
321 			error = EISDIR;
322 			goto out;
323 		} else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
324 			lwp, 0)) != 0)
325 			goto out;
326 	}
327 	error = VOP_SETATTR(vp, &va, cred, lwp);
328 	postat_ret = VOP_GETATTR(vp, &va, cred, lwp);
329 	if (!error)
330 		error = postat_ret;
331 out:
332 	vput(vp);
333 	vn_finished_write(mp, 0);
334 	nfsm_reply(NFSX_WCCORFATTR(v3));
335 	if (v3) {
336 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
337 		return (0);
338 	} else {
339 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
340 		nfsm_srvfillattr(&va, fp);
341 	}
342 	nfsm_srvdone;
343 }
344 
345 /*
346  * nfs lookup rpc
347  */
348 int
349 nfsrv_lookup(nfsd, slp, lwp, mrq)
350 	struct nfsrv_descript *nfsd;
351 	struct nfssvc_sock *slp;
352 	struct lwp *lwp;
353 	struct mbuf **mrq;
354 {
355 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
356 	struct mbuf *nam = nfsd->nd_nam;
357 	caddr_t dpos = nfsd->nd_dpos;
358 	kauth_cred_t cred = nfsd->nd_cr;
359 	struct nfs_fattr *fp;
360 	struct nameidata nd, ind, *ndp = &nd;
361 	struct vnode *vp, *dirp;
362 	nfsrvfh_t nsfh;
363 	caddr_t cp;
364 	u_int32_t *tl;
365 	int32_t t1;
366 	caddr_t bpos;
367 	int error = 0, cache = 0, dirattr_ret = 1;
368 	uint32_t len;
369 	int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
370 	char *cp2;
371 	struct mbuf *mb, *mreq;
372 	struct vattr va, dirattr;
373 	u_quad_t frev;
374 
375 	nfsm_srvmtofh(&nsfh);
376 	nfsm_srvnamesiz(len);
377 
378 	pubflag = nfs_ispublicfh(&nsfh);
379 
380 	nd.ni_cnd.cn_cred = cred;
381 	nd.ni_cnd.cn_nameiop = LOOKUP;
382 	nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART;
383 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
384 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
385 
386 	if (!error && pubflag) {
387 		if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
388 			/*
389 			 * Setup call to lookup() to see if we can find
390 			 * the index file. Arguably, this doesn't belong
391 			 * in a kernel.. Ugh.
392 			 */
393 			ind = nd;
394 			VOP_UNLOCK(nd.ni_vp, 0);
395 			ind.ni_pathlen = strlen(nfs_pub.np_index);
396 			ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
397 			    nfs_pub.np_index;
398 			ind.ni_startdir = nd.ni_vp;
399 			VREF(ind.ni_startdir);
400 			error = lookup(&ind);
401 			if (!error) {
402 				/*
403 				 * Found an index file. Get rid of
404 				 * the old references.
405 				 */
406 				if (dirp)
407 					vrele(dirp);
408 				dirp = nd.ni_vp;
409 				vrele(nd.ni_startdir);
410 				ndp = &ind;
411 			} else
412 				error = 0;
413 		}
414 		/*
415 		 * If the public filehandle was used, check that this lookup
416 		 * didn't result in a filehandle outside the publicly exported
417 		 * filesystem.
418 		 */
419 
420 		if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) {
421 			vput(nd.ni_vp);
422 			error = EPERM;
423 		}
424 	}
425 
426 	if (dirp) {
427 		if (v3)
428 			dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred, lwp);
429 		vrele(dirp);
430 	}
431 
432 	if (error) {
433 		nfsm_reply(NFSX_POSTOPATTR(v3));
434 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
435 		return (0);
436 	}
437 
438 	nqsrv_getl(ndp->ni_startdir, ND_READ);
439 	vrele(ndp->ni_startdir);
440 	PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
441 	vp = ndp->ni_vp;
442 	error = nfsrv_composefh(vp, &nsfh, v3);
443 	if (!error)
444 		error = VOP_GETATTR(vp, &va, cred, lwp);
445 	vput(vp);
446 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPORFATTR(v3) +
447 	    NFSX_POSTOPATTR(v3));
448 	if (error) {
449 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
450 		return (0);
451 	}
452 	nfsm_srvfhtom(&nsfh, v3);
453 	if (v3) {
454 		nfsm_srvpostop_attr(0, &va);
455 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
456 	} else {
457 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
458 		nfsm_srvfillattr(&va, fp);
459 	}
460 	nfsm_srvdone;
461 }
462 
463 /*
464  * nfs readlink service
465  */
466 int
467 nfsrv_readlink(nfsd, slp, lwp, mrq)
468 	struct nfsrv_descript *nfsd;
469 	struct nfssvc_sock *slp;
470 	struct lwp *lwp;
471 	struct mbuf **mrq;
472 {
473 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
474 	struct mbuf *nam = nfsd->nd_nam;
475 	caddr_t dpos = nfsd->nd_dpos;
476 	kauth_cred_t cred = nfsd->nd_cr;
477 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
478 	struct iovec *ivp = iv;
479 	struct mbuf *mp;
480 	u_int32_t *tl;
481 	int32_t t1;
482 	caddr_t bpos;
483 	int error = 0, rdonly, cache = 0, i, padlen, getret;
484 	uint32_t len;
485 	int v3 = (nfsd->nd_flag & ND_NFSV3);
486 	char *cp2;
487 	struct mbuf *mb, *mp2 = NULL, *mp3 = NULL, *mreq;
488 	struct vnode *vp;
489 	struct vattr attr;
490 	nfsrvfh_t nsfh;
491 	struct uio io, *uiop = &io;
492 	u_quad_t frev;
493 
494 	nfsm_srvmtofh(&nsfh);
495 	len = 0;
496 	i = 0;
497 	while (len < NFS_MAXPATHLEN) {
498 		mp = m_get(M_WAIT, MT_DATA);
499 		MCLAIM(mp, &nfs_mowner);
500 		m_clget(mp, M_WAIT);
501 		mp->m_len = NFSMSIZ(mp);
502 		if (len == 0)
503 			mp3 = mp2 = mp;
504 		else {
505 			mp2->m_next = mp;
506 			mp2 = mp;
507 		}
508 		if ((len+mp->m_len) > NFS_MAXPATHLEN) {
509 			mp->m_len = NFS_MAXPATHLEN-len;
510 			len = NFS_MAXPATHLEN;
511 		} else
512 			len += mp->m_len;
513 		ivp->iov_base = mtod(mp, caddr_t);
514 		ivp->iov_len = mp->m_len;
515 		i++;
516 		ivp++;
517 	}
518 	uiop->uio_iov = iv;
519 	uiop->uio_iovcnt = i;
520 	uiop->uio_offset = 0;
521 	uiop->uio_resid = len;
522 	uiop->uio_rw = UIO_READ;
523 	UIO_SETUP_SYSSPACE(uiop);
524 	error = nfsrv_fhtovp(&nsfh, 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, lwp);
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 	len -= uiop->uio_resid;
553 	padlen = nfsm_padlen(len);
554 	if (uiop->uio_resid || padlen)
555 		nfs_zeropad(mp3, uiop->uio_resid, padlen);
556 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
557 	*tl = txdr_unsigned(len);
558 	mb->m_next = mp3;
559 	nfsm_srvdone;
560 }
561 
562 /*
563  * nfs read service
564  */
565 int
566 nfsrv_read(nfsd, slp, lwp, mrq)
567 	struct nfsrv_descript *nfsd;
568 	struct nfssvc_sock *slp;
569 	struct lwp *lwp;
570 	struct mbuf **mrq;
571 {
572 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
573 	struct mbuf *nam = nfsd->nd_nam;
574 	caddr_t dpos = nfsd->nd_dpos;
575 	kauth_cred_t cred = nfsd->nd_cr;
576 	struct mbuf *m;
577 	struct nfs_fattr *fp;
578 	u_int32_t *tl;
579 	int32_t t1;
580 	int i;
581 	caddr_t bpos;
582 	int error = 0, rdonly, cache = 0, getret;
583 	int v3 = (nfsd->nd_flag & ND_NFSV3);
584 	uint32_t reqlen, len, cnt, left;
585 	int padlen;
586 	char *cp2;
587 	struct mbuf *mb, *mreq;
588 	struct vnode *vp;
589 	nfsrvfh_t nsfh;
590 	struct uio io, *uiop = &io;
591 	struct vattr va;
592 	off_t off;
593 	u_quad_t frev;
594 
595 	nfsm_srvmtofh(&nsfh);
596 	if (v3) {
597 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
598 		off = fxdr_hyper(tl);
599 	} else {
600 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
601 		off = (off_t)fxdr_unsigned(u_int32_t, *tl);
602 	}
603 	nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
604 	reqlen = fxdr_unsigned(uint32_t, *tl);
605 	reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd));
606 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
607 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
608 	if (error) {
609 		nfsm_reply(2 * NFSX_UNSIGNED);
610 		nfsm_srvpostop_attr(1, (struct vattr *)0);
611 		return (0);
612 	}
613 	if (vp->v_type != VREG) {
614 		if (v3)
615 			error = EINVAL;
616 		else
617 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
618 	}
619 	if (!error) {
620 	    nqsrv_getl(vp, ND_READ);
621 	    if ((error = nfsrv_access(vp, VREAD, cred, rdonly, lwp, 1)) != 0)
622 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 1);
623 	}
624 	getret = VOP_GETATTR(vp, &va, cred, lwp);
625 	if (!error)
626 		error = getret;
627 	if (error) {
628 		vput(vp);
629 		nfsm_reply(NFSX_POSTOPATTR(v3));
630 		nfsm_srvpostop_attr(getret, &va);
631 		return (0);
632 	}
633 	if (off >= va.va_size)
634 		cnt = 0;
635 	else if ((off + reqlen) > va.va_size)
636 		cnt = va.va_size - off;
637 	else
638 		cnt = reqlen;
639 	nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
640 	if (v3) {
641 		nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
642 		*tl++ = nfs_true;
643 		fp = (struct nfs_fattr *)tl;
644 		tl += (NFSX_V3FATTR / sizeof (u_int32_t));
645 	} else {
646 		nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
647 		fp = (struct nfs_fattr *)tl;
648 		tl += (NFSX_V2FATTR / sizeof (u_int32_t));
649 	}
650 	len = left = cnt;
651 	if (cnt > 0) {
652 		if (nfsd_use_loan) {
653 			struct vm_page **pgpp;
654 			voff_t pgoff = trunc_page(off);
655 			int npages;
656 			vaddr_t lva;
657 
658 			npages = (round_page(off + cnt) - pgoff) >> PAGE_SHIFT;
659 			KASSERT(npages <= M_EXT_MAXPAGES); /* XXX */
660 
661 			/* allocate kva for mbuf data */
662 			lva = sokvaalloc(npages << PAGE_SHIFT, slp->ns_so);
663 			if (lva == 0) {
664 				/* fall back to VOP_READ */
665 				goto loan_fail;
666 			}
667 
668 			/* allocate mbuf */
669 			m = m_get(M_WAIT, MT_DATA);
670 			MCLAIM(m, &nfs_mowner);
671 			pgpp = m->m_ext.ext_pgs;
672 
673 			/* loan pages */
674 			error = uvm_loanuobjpages(&vp->v_uobj, pgoff, npages,
675 			    pgpp);
676 			if (error) {
677 				sokvafree(lva, npages << PAGE_SHIFT);
678 				m_free(m);
679 				if (error == EBUSY)
680 					goto loan_fail;
681 				goto read_error;
682 			}
683 
684 			/* associate kva to mbuf */
685 			MEXTADD(m, (void *)(lva + ((vaddr_t)off & PAGE_MASK)),
686 			    cnt, M_MBUF, soloanfree, slp->ns_so);
687 			m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
688 			m->m_len = cnt;
689 
690 			/* map pages */
691 			for (i = 0; i < npages; i++) {
692 				pmap_kenter_pa(lva, VM_PAGE_TO_PHYS(pgpp[i]),
693 				    VM_PROT_READ);
694 				lva += PAGE_SIZE;
695 			}
696 
697 			pmap_update(pmap_kernel());
698 
699 			mb->m_next = m;
700 			mb = m;
701 			error = 0;
702 			uiop->uio_resid = 0;
703 		} else {
704 			struct iovec *iv;
705 			struct iovec *iv2;
706 			struct mbuf *m2;
707 			int siz;
708 loan_fail:
709 			/*
710 			 * Generate the mbuf list with the uio_iov ref. to it.
711 			 */
712 			i = 0;
713 			m = m2 = mb;
714 			while (left > 0) {
715 				siz = min(M_TRAILINGSPACE(m), left);
716 				if (siz > 0) {
717 					left -= siz;
718 					i++;
719 				}
720 				if (left > 0) {
721 					m = m_get(M_WAIT, MT_DATA);
722 					MCLAIM(m, &nfs_mowner);
723 					m_clget(m, M_WAIT);
724 					m->m_len = 0;
725 					m2->m_next = m;
726 					m2 = m;
727 				}
728 			}
729 			iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK);
730 			uiop->uio_iov = iv2 = iv;
731 			m = mb;
732 			left = cnt;
733 			i = 0;
734 			while (left > 0) {
735 				if (m == NULL)
736 					panic("nfsrv_read iov");
737 				siz = min(M_TRAILINGSPACE(m), left);
738 				if (siz > 0) {
739 					iv->iov_base = mtod(m, caddr_t) +
740 					    m->m_len;
741 					iv->iov_len = siz;
742 					m->m_len += siz;
743 					left -= siz;
744 					iv++;
745 					i++;
746 				}
747 				m = m->m_next;
748 			}
749 			uiop->uio_iovcnt = i;
750 			uiop->uio_offset = off;
751 			uiop->uio_resid = cnt;
752 			uiop->uio_rw = UIO_READ;
753 			UIO_SETUP_SYSSPACE(uiop);
754 			error = VOP_READ(vp, uiop, IO_NODELOCKED, cred);
755 			free((caddr_t)iv2, M_TEMP);
756 		}
757 read_error:
758 		if (error || (getret = VOP_GETATTR(vp, &va, cred, lwp)) != 0){
759 			if (!error)
760 				error = getret;
761 			m_freem(mreq);
762 			vput(vp);
763 			nfsm_reply(NFSX_POSTOPATTR(v3));
764 			nfsm_srvpostop_attr(getret, &va);
765 			return (0);
766 		}
767 	} else {
768 		uiop->uio_resid = 0;
769 	}
770 	vput(vp);
771 	nfsm_srvfillattr(&va, fp);
772 	len -= uiop->uio_resid;
773 	padlen = nfsm_padlen(len);
774 	if (uiop->uio_resid || padlen)
775 		nfs_zeropad(mb, uiop->uio_resid, padlen);
776 	if (v3) {
777 		/* count */
778 		*tl++ = txdr_unsigned(len);
779 		/* eof */
780 		if (off + len >= va.va_size)
781 			*tl++ = nfs_true;
782 		else
783 			*tl++ = nfs_false;
784 	}
785 	*tl = txdr_unsigned(len);
786 	nfsm_srvdone;
787 }
788 
789 /*
790  * nfs write service
791  */
792 int
793 nfsrv_write(nfsd, slp, lwp, mrq)
794 	struct nfsrv_descript *nfsd;
795 	struct nfssvc_sock *slp;
796 	struct lwp *lwp;
797 	struct mbuf **mrq;
798 {
799 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
800 	struct mbuf *nam = nfsd->nd_nam;
801 	caddr_t dpos = nfsd->nd_dpos;
802 	kauth_cred_t cred = nfsd->nd_cr;
803 	struct iovec *ivp;
804 	int i, cnt;
805 	struct mbuf *mp;
806 	struct nfs_fattr *fp;
807 	struct iovec *iv;
808 	struct vattr va, forat;
809 	u_int32_t *tl;
810 	int32_t t1;
811 	caddr_t bpos;
812 	int error = 0, rdonly, cache = 0, len, forat_ret = 1;
813 	int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
814 	int stable = NFSV3WRITE_FILESYNC;
815 	int v3 = (nfsd->nd_flag & ND_NFSV3);
816 	char *cp2;
817 	struct mbuf *mb, *mreq;
818 	struct vnode *vp;
819 	nfsrvfh_t nsfh;
820 	struct uio io, *uiop = &io;
821 	off_t off;
822 	u_quad_t frev;
823 	struct mount *mntp = NULL;
824 
825 	if (mrep == NULL) {
826 		*mrq = NULL;
827 		return (0);
828 	}
829 	nfsm_srvmtofh(&nsfh);
830 	if ((mntp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL)
831 		return (ESTALE);
832 	vn_start_write(NULL, &mntp, V_WAIT);
833 	if (v3) {
834 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
835 		off = fxdr_hyper(tl);
836 		tl += 3;
837 		stable = fxdr_unsigned(int, *tl++);
838 	} else {
839 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
840 		off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
841 		tl += 2;
842 	}
843 	retlen = len = fxdr_unsigned(int32_t, *tl);
844 	cnt = i = 0;
845 
846 	/*
847 	 * For NFS Version 2, it is not obvious what a write of zero length
848 	 * should do, but I might as well be consistent with Version 3,
849 	 * which is to return ok so long as there are no permission problems.
850 	 */
851 	if (len > 0) {
852 		zeroing = 1;
853 		mp = mrep;
854 		while (mp) {
855 			if (mp == md) {
856 				zeroing = 0;
857 				adjust = dpos - mtod(mp, caddr_t);
858 				mp->m_len -= adjust;
859 				if (mp->m_len > 0 && adjust > 0)
860 					NFSMADV(mp, adjust);
861 			}
862 			if (zeroing)
863 				mp->m_len = 0;
864 			else if (mp->m_len > 0) {
865 				i += mp->m_len;
866 				if (i > len) {
867 					mp->m_len -= (i - len);
868 					zeroing	= 1;
869 				}
870 				if (mp->m_len > 0)
871 					cnt++;
872 			}
873 			mp = mp->m_next;
874 		}
875 	}
876 	if (len > NFS_MAXDATA || len < 0 || i < len) {
877 		error = EIO;
878 		nfsm_reply(2 * NFSX_UNSIGNED);
879 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
880 		vn_finished_write(mntp, 0);
881 		return (0);
882 	}
883 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
884 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
885 	if (error) {
886 		nfsm_reply(2 * NFSX_UNSIGNED);
887 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
888 		vn_finished_write(mntp, 0);
889 		return (0);
890 	}
891 	if (v3)
892 		forat_ret = VOP_GETATTR(vp, &forat, cred, lwp);
893 	if (vp->v_type != VREG) {
894 		if (v3)
895 			error = EINVAL;
896 		else
897 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
898 	}
899 	if (!error) {
900 		nqsrv_getl(vp, ND_WRITE);
901 		error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1);
902 	}
903 	if (error) {
904 		vput(vp);
905 		nfsm_reply(NFSX_WCCDATA(v3));
906 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
907 		vn_finished_write(mntp, 0);
908 		return (0);
909 	}
910 
911 	if (len > 0) {
912 		ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK);
913 		uiop->uio_iov = iv = ivp;
914 		uiop->uio_iovcnt = cnt;
915 		mp = mrep;
916 		while (mp) {
917 			if (mp->m_len > 0) {
918 				ivp->iov_base = mtod(mp, caddr_t);
919 				ivp->iov_len = mp->m_len;
920 				ivp++;
921 			}
922 			mp = mp->m_next;
923 		}
924 
925 		/*
926 		 * XXX
927 		 * The IO_METASYNC flag indicates that all metadata (and not
928 		 * just enough to ensure data integrity) must be written to
929 		 * stable storage synchronously.
930 		 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
931 		 */
932 		if (stable == NFSV3WRITE_UNSTABLE)
933 			ioflags = IO_NODELOCKED;
934 		else if (stable == NFSV3WRITE_DATASYNC)
935 			ioflags = (IO_SYNC | IO_NODELOCKED);
936 		else
937 			ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
938 		uiop->uio_resid = len;
939 		uiop->uio_rw = UIO_WRITE;
940 		uiop->uio_offset = off;
941 		UIO_SETUP_SYSSPACE(uiop);
942 		error = VOP_WRITE(vp, uiop, ioflags, cred);
943 		nfsstats.srvvop_writes++;
944 		free(iv, M_TEMP);
945 	}
946 	aftat_ret = VOP_GETATTR(vp, &va, cred, lwp);
947 	vput(vp);
948 	vn_finished_write(mntp, 0);
949 	if (!error)
950 		error = aftat_ret;
951 	nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
952 		2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
953 	if (v3) {
954 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
955 		if (error)
956 			return (0);
957 		nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
958 		*tl++ = txdr_unsigned(retlen);
959 		if (stable == NFSV3WRITE_UNSTABLE)
960 			*tl++ = txdr_unsigned(stable);
961 		else
962 			*tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
963 		/*
964 		 * Actually, there is no need to txdr these fields,
965 		 * but it may make the values more human readable,
966 		 * for debugging purposes.
967 		 */
968 		*tl++ = txdr_unsigned(boottime.tv_sec);
969 		*tl = txdr_unsigned(boottime.tv_usec);
970 	} else {
971 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
972 		nfsm_srvfillattr(&va, fp);
973 	}
974 	nfsm_srvdone;
975 }
976 
977 /*
978  * XXX elad: the original NFSW_SAMECRED() macro also made sure the
979  *	     two nd_flag fields of the descriptors contained
980  *	     ND_KERBAUTH.
981  */
982 static int
983 nfsrv_samecred(kauth_cred_t cred1, kauth_cred_t cred2)
984 {
985 	int i, do_ngroups;
986 
987 	if (kauth_cred_geteuid(cred1) != kauth_cred_geteuid(cred2))
988 		return (0);
989 	if (kauth_cred_ngroups(cred1) != kauth_cred_ngroups(cred2))
990 		return (0);
991 	do_ngroups = kauth_cred_ngroups(cred1);
992 	for (i = 0; i < do_ngroups; i++)
993 		if (kauth_cred_group(cred1, i) !=
994 		    kauth_cred_group(cred2, i))
995 			return (0);
996 
997 	return (1);
998 }
999 
1000 static struct nfsrvw_delayhash *
1001 nfsrv_nwdelayhash(struct nfssvc_sock *slp, const nfsrvfh_t *nsfh)
1002 {
1003 	uint32_t hash;
1004 
1005 	hash = hash32_buf(NFSRVFH_DATA(nsfh), NFSRVFH_SIZE(nsfh),
1006 	    HASH32_BUF_INIT);
1007 	return &slp->ns_wdelayhashtbl[hash % NFS_WDELAYHASHSIZ];
1008 }
1009 
1010 /*
1011  * NFS write service with write gathering support. Called when
1012  * nfsrvw_procrastinate > 0.
1013  * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1014  * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1015  * Jan. 1994.
1016  */
1017 int
1018 nfsrv_writegather(ndp, slp, lwp, mrq)
1019 	struct nfsrv_descript **ndp;
1020 	struct nfssvc_sock *slp;
1021 	struct lwp *lwp;
1022 	struct mbuf **mrq;
1023 {
1024 	struct timeval now;
1025 	struct iovec *ivp;
1026 	struct mbuf *mp;
1027 	struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1028 	struct nfs_fattr *fp;
1029 	int i = 0;
1030 	struct iovec *iov;
1031 	struct nfsrvw_delayhash *wpp;
1032 	kauth_cred_t cred;
1033 	struct vattr va, forat;
1034 	u_int32_t *tl;
1035 	int32_t t1;
1036 	caddr_t bpos, dpos;
1037 	int error = 0, rdonly, cache = 0, len = 0, forat_ret = 1;
1038 	int ioflags, aftat_ret = 1, s, adjust, v3, zeroing;
1039 	char *cp2;
1040 	struct mbuf *mb, *mreq, *mrep, *md;
1041 	struct vnode *vp;
1042 	struct uio io, *uiop = &io;
1043 	u_quad_t frev, cur_usec;
1044 	struct mount *mntp = NULL;
1045 
1046 	*mrq = NULL;
1047 	if (*ndp) {
1048 	    nfsd = *ndp;
1049 	    *ndp = NULL;
1050 	    mrep = nfsd->nd_mrep;
1051 	    md = nfsd->nd_md;
1052 	    dpos = nfsd->nd_dpos;
1053 	    cred = nfsd->nd_cr;
1054 	    v3 = (nfsd->nd_flag & ND_NFSV3);
1055 	    LIST_INIT(&nfsd->nd_coalesce);
1056 	    nfsd->nd_mreq = NULL;
1057 	    nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1058 	    getmicrotime(&now);
1059 	    cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec;
1060 	    nfsd->nd_time = cur_usec + nfsrvw_procrastinate;
1061 
1062 	    /*
1063 	     * Now, get the write header..
1064 	     */
1065 	    nfsm_srvmtofh(&nfsd->nd_fh);
1066 	    if (v3) {
1067 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1068 		nfsd->nd_off = fxdr_hyper(tl);
1069 		tl += 3;
1070 		nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1071 	    } else {
1072 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1073 		nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1074 		tl += 2;
1075 	    }
1076 	    len = fxdr_unsigned(int32_t, *tl);
1077 	    nfsd->nd_len = len;
1078 	    nfsd->nd_eoff = nfsd->nd_off + len;
1079 
1080 	    /*
1081 	     * Trim the header out of the mbuf list and trim off any trailing
1082 	     * junk so that the mbuf list has only the write data.
1083 	     */
1084 	    zeroing = 1;
1085 	    i = 0;
1086 	    mp = mrep;
1087 	    while (mp) {
1088 		if (mp == md) {
1089 		    zeroing = 0;
1090 		    adjust = dpos - mtod(mp, caddr_t);
1091 		    mp->m_len -= adjust;
1092 		    if (mp->m_len > 0 && adjust > 0)
1093 			NFSMADV(mp, adjust);
1094 		}
1095 		if (zeroing)
1096 		    mp->m_len = 0;
1097 		else {
1098 		    i += mp->m_len;
1099 		    if (i > len) {
1100 			mp->m_len -= (i - len);
1101 			zeroing = 1;
1102 		    }
1103 		}
1104 		mp = mp->m_next;
1105 	    }
1106 	    if (len > NFS_MAXDATA || len < 0  || i < len) {
1107 nfsmout:
1108 		m_freem(mrep);
1109 		error = EIO;
1110 		nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1111 		if (v3)
1112 		    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1113 		nfsd->nd_mreq = mreq;
1114 		nfsd->nd_mrep = NULL;
1115 		nfsd->nd_time = 0;
1116 	    }
1117 
1118 	    /*
1119 	     * Add this entry to the hash and time queues.
1120 	     */
1121 	    s = splsoftclock();
1122 	    owp = NULL;
1123 	    wp = LIST_FIRST(&slp->ns_tq);
1124 	    while (wp && wp->nd_time < nfsd->nd_time) {
1125 		owp = wp;
1126 		wp = LIST_NEXT(wp, nd_tq);
1127 	    }
1128 	    if (owp) {
1129 		LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1130 	    } else {
1131 		LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1132 	    }
1133 	    if (nfsd->nd_mrep) {
1134 		wpp = nfsrv_nwdelayhash(slp, &nfsd->nd_fh);
1135 		owp = NULL;
1136 		wp = LIST_FIRST(wpp);
1137 		while (wp && nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) {
1138 		    owp = wp;
1139 		    wp = LIST_NEXT(wp, nd_hash);
1140 		}
1141 		while (wp && wp->nd_off < nfsd->nd_off &&
1142 		    !nfsrv_comparefh(&nfsd->nd_fh, &wp->nd_fh)) {
1143 		    owp = wp;
1144 		    wp = LIST_NEXT(wp, nd_hash);
1145 		}
1146 		if (owp) {
1147 		    LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1148 
1149 		    /*
1150 		     * Search the hash list for overlapping entries and
1151 		     * coalesce.
1152 		     */
1153 		    for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1154 			wp = LIST_NEXT(nfsd, nd_hash);
1155 			if (nfsrv_samecred(owp->nd_cr, nfsd->nd_cr))
1156 			    nfsrvw_coalesce(owp, nfsd);
1157 		    }
1158 		} else {
1159 		    LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1160 		}
1161 	    }
1162 	    splx(s);
1163 	}
1164 
1165 	/*
1166 	 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1167 	 * and generate the associated reply mbuf list(s).
1168 	 */
1169 loop1:
1170 	getmicrotime(&now);
1171 	cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec;
1172 	s = splsoftclock();
1173 	for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) {
1174 		owp = LIST_NEXT(nfsd, nd_tq);
1175 		if (nfsd->nd_time > cur_usec)
1176 		    break;
1177 		if (nfsd->nd_mreq)
1178 		    continue;
1179 		LIST_REMOVE(nfsd, nd_tq);
1180 		LIST_REMOVE(nfsd, nd_hash);
1181 		splx(s);
1182 		mrep = nfsd->nd_mrep;
1183 		nfsd->nd_mrep = NULL;
1184 		cred = nfsd->nd_cr;
1185 		v3 = (nfsd->nd_flag & ND_NFSV3);
1186 		forat_ret = aftat_ret = 1;
1187 		error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1188 		    nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH),
1189 		    FALSE);
1190 		if (!error) {
1191 		    if (v3)
1192 			forat_ret = VOP_GETATTR(vp, &forat, cred, lwp);
1193 		    if (vp->v_type != VREG) {
1194 			if (v3)
1195 			    error = EINVAL;
1196 			else
1197 			    error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1198 		    }
1199 		} else
1200 		    vp = NULL;
1201 		if (!error) {
1202 		    nqsrv_getl(vp, ND_WRITE);
1203 		    error = nfsrv_access(vp, VWRITE, cred, rdonly, lwp, 1);
1204 		}
1205 
1206 		if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1207 		    ioflags = IO_NODELOCKED;
1208 		else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1209 		    ioflags = (IO_SYNC | IO_NODELOCKED);
1210 		else
1211 		    ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1212 		uiop->uio_rw = UIO_WRITE;
1213 		uiop->uio_offset = nfsd->nd_off;
1214 		uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1215 		UIO_SETUP_SYSSPACE(uiop);
1216 		if (uiop->uio_resid > 0) {
1217 		    mp = mrep;
1218 		    i = 0;
1219 		    while (mp) {
1220 			if (mp->m_len > 0)
1221 			    i++;
1222 			mp = mp->m_next;
1223 		    }
1224 		    uiop->uio_iovcnt = i;
1225 		    iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
1226 		    uiop->uio_iov = ivp = iov;
1227 		    mp = mrep;
1228 		    while (mp) {
1229 			if (mp->m_len > 0) {
1230 			    ivp->iov_base = mtod(mp, caddr_t);
1231 			    ivp->iov_len = mp->m_len;
1232 			    ivp++;
1233 			}
1234 			mp = mp->m_next;
1235 		    }
1236 		    if (!error) {
1237 			if (vn_start_write(vp, &mntp, V_NOWAIT) != 0) {
1238 			    VOP_UNLOCK(vp, 0);
1239 			    vn_start_write(NULL, &mntp, V_WAIT);
1240 			    vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1241 			}
1242 			if (!error) {
1243 			    error = VOP_WRITE(vp, uiop, ioflags, cred);
1244 			    nfsstats.srvvop_writes++;
1245 			    vn_finished_write(mntp, 0);
1246 			}
1247 		    }
1248 		    free((caddr_t)iov, M_TEMP);
1249 		}
1250 		m_freem(mrep);
1251 		if (vp) {
1252 		    aftat_ret = VOP_GETATTR(vp, &va, cred, lwp);
1253 		    vput(vp);
1254 		}
1255 
1256 		/*
1257 		 * Loop around generating replies for all write rpcs that have
1258 		 * now been completed.
1259 		 */
1260 		swp = nfsd;
1261 		do {
1262 		    if (error) {
1263 			nfsm_writereply(NFSX_WCCDATA(v3), v3);
1264 			if (v3) {
1265 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1266 			}
1267 		    } else {
1268 			nfsm_writereply(NFSX_PREOPATTR(v3) +
1269 			    NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1270 			    NFSX_WRITEVERF(v3), v3);
1271 			if (v3) {
1272 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1273 			    nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1274 			    *tl++ = txdr_unsigned(nfsd->nd_len);
1275 			    *tl++ = txdr_unsigned(swp->nd_stable);
1276 			    /*
1277 			     * Actually, there is no need to txdr these fields,
1278 			     * but it may make the values more human readable,
1279 			     * for debugging purposes.
1280 			     */
1281 			    *tl++ = txdr_unsigned(boottime.tv_sec);
1282 			    *tl = txdr_unsigned(boottime.tv_usec);
1283 			} else {
1284 			    nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1285 			    nfsm_srvfillattr(&va, fp);
1286 			}
1287 		    }
1288 		    nfsd->nd_mreq = mreq;
1289 		    if (nfsd->nd_mrep)
1290 			panic("nfsrv_write: nd_mrep not free");
1291 
1292 		    /*
1293 		     * Done. Put it at the head of the timer queue so that
1294 		     * the final phase can return the reply.
1295 		     */
1296 		    s = splsoftclock();
1297 		    if (nfsd != swp) {
1298 			nfsd->nd_time = 0;
1299 			LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1300 		    }
1301 		    nfsd = LIST_FIRST(&swp->nd_coalesce);
1302 		    if (nfsd) {
1303 			LIST_REMOVE(nfsd, nd_tq);
1304 		    }
1305 		    splx(s);
1306 		} while (nfsd);
1307 		s = splsoftclock();
1308 		swp->nd_time = 0;
1309 		LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1310 		splx(s);
1311 		goto loop1;
1312 	}
1313 	splx(s);
1314 
1315 	/*
1316 	 * Search for a reply to return.
1317 	 */
1318 	s = splsoftclock();
1319 	LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) {
1320 		if (nfsd->nd_mreq) {
1321 		    LIST_REMOVE(nfsd, nd_tq);
1322 		    *mrq = nfsd->nd_mreq;
1323 		    *ndp = nfsd;
1324 		    break;
1325 		}
1326 	}
1327 	splx(s);
1328 	return (0);
1329 }
1330 
1331 /*
1332  * Coalesce the write request nfsd into owp. To do this we must:
1333  * - remove nfsd from the queues
1334  * - merge nfsd->nd_mrep into owp->nd_mrep
1335  * - update the nd_eoff and nd_stable for owp
1336  * - put nfsd on owp's nd_coalesce list
1337  * NB: Must be called at splsoftclock().
1338  */
1339 void
1340 nfsrvw_coalesce(owp, nfsd)
1341         struct nfsrv_descript *owp;
1342         struct nfsrv_descript *nfsd;
1343 {
1344         int overlap;
1345         struct mbuf *mp;
1346 	struct nfsrv_descript *m;
1347 
1348         LIST_REMOVE(nfsd, nd_hash);
1349         LIST_REMOVE(nfsd, nd_tq);
1350         if (owp->nd_eoff < nfsd->nd_eoff) {
1351             overlap = owp->nd_eoff - nfsd->nd_off;
1352             if (overlap < 0)
1353                 panic("nfsrv_coalesce: bad off");
1354             if (overlap > 0)
1355                 m_adj(nfsd->nd_mrep, overlap);
1356             mp = owp->nd_mrep;
1357             while (mp->m_next)
1358                 mp = mp->m_next;
1359             mp->m_next = nfsd->nd_mrep;
1360             owp->nd_eoff = nfsd->nd_eoff;
1361         } else
1362             m_freem(nfsd->nd_mrep);
1363         nfsd->nd_mrep = NULL;
1364         if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1365             owp->nd_stable = NFSV3WRITE_FILESYNC;
1366         else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1367             owp->nd_stable == NFSV3WRITE_UNSTABLE)
1368             owp->nd_stable = NFSV3WRITE_DATASYNC;
1369         LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1370  	/*
1371  	 * nfsd might hold coalesce elements! Move them to owp.
1372  	 * Otherwise, requests may be lost and clients will be stuck.
1373  	 */
1374 	while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) {
1375 		LIST_REMOVE(m, nd_tq);
1376 		LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq);
1377 	}
1378 }
1379 
1380 /*
1381  * nfs create service
1382  * now does a truncate to 0 length via. setattr if it already exists
1383  */
1384 int
1385 nfsrv_create(nfsd, slp, lwp, mrq)
1386 	struct nfsrv_descript *nfsd;
1387 	struct nfssvc_sock *slp;
1388 	struct lwp *lwp;
1389 	struct mbuf **mrq;
1390 {
1391 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1392 	struct mbuf *nam = nfsd->nd_nam;
1393 	caddr_t dpos = nfsd->nd_dpos;
1394 	kauth_cred_t cred = nfsd->nd_cr;
1395 	struct nfs_fattr *fp;
1396 	struct vattr va, dirfor, diraft;
1397 	struct nfsv2_sattr *sp;
1398 	u_int32_t *tl;
1399 	struct nameidata nd;
1400 	caddr_t cp;
1401 	int32_t t1;
1402 	caddr_t bpos;
1403 	int error = 0, cache = 0, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1404 	int rdev = 0;
1405 	int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1406 	char *cp2;
1407 	struct mbuf *mb, *mreq;
1408 	struct vnode *vp = NULL, *dirp = NULL;
1409 	nfsrvfh_t nsfh;
1410 	u_quad_t frev, tempsize;
1411 	u_char cverf[NFSX_V3CREATEVERF];
1412 	struct mount *mp = NULL;
1413 
1414 	nd.ni_cnd.cn_nameiop = 0;
1415 	nfsm_srvmtofh(&nsfh);
1416 	if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL)
1417 		return (ESTALE);
1418 	vn_start_write(NULL, &mp, V_WAIT);
1419 	nfsm_srvnamesiz(len);
1420 	nd.ni_cnd.cn_cred = cred;
1421 	nd.ni_cnd.cn_nameiop = CREATE;
1422 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1423 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1424 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1425 	if (dirp) {
1426 		if (v3)
1427 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp);
1428 		else {
1429 			vrele(dirp);
1430 			dirp = (struct vnode *)0;
1431 		}
1432 	}
1433 	if (error) {
1434 		nfsm_reply(NFSX_WCCDATA(v3));
1435 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1436 		if (dirp)
1437 			vrele(dirp);
1438 		vn_finished_write(mp, 0);
1439 		return (0);
1440 	}
1441 	VATTR_NULL(&va);
1442 	if (v3) {
1443 		va.va_mode = 0;
1444 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1445 		how = fxdr_unsigned(int, *tl);
1446 		switch (how) {
1447 		case NFSV3CREATE_GUARDED:
1448 			if (nd.ni_vp) {
1449 				error = EEXIST;
1450 				break;
1451 			}
1452 		case NFSV3CREATE_UNCHECKED:
1453 			nfsm_srvsattr(&va);
1454 			break;
1455 		case NFSV3CREATE_EXCLUSIVE:
1456 			nfsm_dissect(cp, caddr_t, NFSX_V3CREATEVERF);
1457 			memcpy(cverf, cp, NFSX_V3CREATEVERF);
1458 			exclusive_flag = 1;
1459 			break;
1460 		};
1461 		va.va_type = VREG;
1462 	} else {
1463 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1464 		va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1465 		if (va.va_type == VNON)
1466 			va.va_type = VREG;
1467 		va.va_mode = nfstov_mode(sp->sa_mode);
1468 		switch (va.va_type) {
1469 		case VREG:
1470 			tsize = fxdr_unsigned(int32_t, sp->sa_size);
1471 			if (tsize != -1)
1472 				va.va_size = (u_quad_t)tsize;
1473 			break;
1474 		case VCHR:
1475 		case VBLK:
1476 		case VFIFO:
1477 			rdev = fxdr_unsigned(int32_t, sp->sa_size);
1478 			break;
1479 		default:
1480 			break;
1481 		};
1482 	}
1483 
1484 	/*
1485 	 * Iff doesn't exist, create it
1486 	 * otherwise just truncate to 0 length
1487 	 *   should I set the mode too ??
1488 	 */
1489 	if (nd.ni_vp == NULL) {
1490 		if (va.va_type == VREG || va.va_type == VSOCK) {
1491 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1492 			error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1493 			if (!error) {
1494 				if (exclusive_flag) {
1495 					exclusive_flag = 0;
1496 					VATTR_NULL(&va);
1497 					/*
1498 					 * XXX
1499 					 * assuming NFSX_V3CREATEVERF
1500 					 * == sizeof(nfstime3)
1501 					 */
1502 					fxdr_nfsv3time(cverf, &va.va_atime);
1503 					error = VOP_SETATTR(nd.ni_vp, &va, cred,
1504 						lwp);
1505 				}
1506 			}
1507 		} else if (va.va_type == VCHR || va.va_type == VBLK ||
1508 			va.va_type == VFIFO) {
1509 			if (va.va_type == VCHR && rdev == 0xffffffff)
1510 				va.va_type = VFIFO;
1511 			if (va.va_type != VFIFO &&
1512 			    (error = kauth_authorize_generic(cred,
1513 			    KAUTH_GENERIC_ISSUSER, (u_short *)0))) {
1514 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1515 				vput(nd.ni_dvp);
1516 				nfsm_reply(0);
1517 				vn_finished_write(mp, 0);
1518 				return (error);
1519 			} else
1520 				va.va_rdev = (dev_t)rdev;
1521 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1522 			error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd,
1523 			    &va);
1524 			if (error) {
1525 				nfsm_reply(0);
1526 			}
1527 			if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1528 				vrele(nd.ni_dvp);
1529 				vput(nd.ni_vp);
1530 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1531 				error = EINVAL;
1532 				nfsm_reply(0);
1533 			}
1534 		} else {
1535 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1536 			vput(nd.ni_dvp);
1537 			error = ENXIO;
1538 		}
1539 		vp = nd.ni_vp;
1540 	} else {
1541 		vp = nd.ni_vp;
1542 		if (nd.ni_dvp == vp)
1543 			vrele(nd.ni_dvp);
1544 		else
1545 			vput(nd.ni_dvp);
1546 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1547 		if (!error && va.va_size != -1) {
1548 			error = nfsrv_access(vp, VWRITE, cred,
1549 			    (nd.ni_cnd.cn_flags & RDONLY), lwp, 0);
1550 			if (!error) {
1551 				nqsrv_getl(vp, ND_WRITE);
1552 				tempsize = va.va_size;
1553 				VATTR_NULL(&va);
1554 				va.va_size = tempsize;
1555 				error = VOP_SETATTR(vp, &va, cred, lwp);
1556 			}
1557 		}
1558 		if (error)
1559 			vput(vp);
1560 	}
1561 	if (!error) {
1562 		error = nfsrv_composefh(vp, &nsfh, v3);
1563 		if (!error)
1564 			error = VOP_GETATTR(vp, &va, cred, lwp);
1565 		vput(vp);
1566 	}
1567 	if (v3) {
1568 		if (exclusive_flag && !error) {
1569 			/*
1570 			 * XXX assuming NFSX_V3CREATEVERF == sizeof(nfstime3)
1571 			 */
1572 			char oldverf[NFSX_V3CREATEVERF];
1573 
1574 			txdr_nfsv3time(&va.va_atime, oldverf);
1575 			if (memcmp(cverf, oldverf, NFSX_V3CREATEVERF))
1576 				error = EEXIST;
1577 		}
1578 		if (dirp) {
1579 			diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp);
1580 			vrele(dirp);
1581 		}
1582 	}
1583 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1584 	if (v3) {
1585 		if (!error) {
1586 			nfsm_srvpostop_fh(&nsfh);
1587 			nfsm_srvpostop_attr(0, &va);
1588 		}
1589 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1590 	} else {
1591 		nfsm_srvfhtom(&nsfh, v3);
1592 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1593 		nfsm_srvfillattr(&va, fp);
1594 	}
1595 	vn_finished_write(mp, 0);
1596 	return (0);
1597 nfsmout:
1598 	if (dirp)
1599 		vrele(dirp);
1600 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1601 	if (nd.ni_dvp == nd.ni_vp)
1602 		vrele(nd.ni_dvp);
1603 	else
1604 		vput(nd.ni_dvp);
1605 	if (nd.ni_vp)
1606 		vput(nd.ni_vp);
1607 	vn_finished_write(mp, 0);
1608 	return (error);
1609 }
1610 
1611 /*
1612  * nfs v3 mknod service
1613  */
1614 int
1615 nfsrv_mknod(nfsd, slp, lwp, mrq)
1616 	struct nfsrv_descript *nfsd;
1617 	struct nfssvc_sock *slp;
1618 	struct lwp *lwp;
1619 	struct mbuf **mrq;
1620 {
1621 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1622 	struct mbuf *nam = nfsd->nd_nam;
1623 	caddr_t dpos = nfsd->nd_dpos;
1624 	kauth_cred_t cred = nfsd->nd_cr;
1625 	struct vattr va, dirfor, diraft;
1626 	u_int32_t *tl;
1627 	struct nameidata nd;
1628 	int32_t t1;
1629 	caddr_t bpos;
1630 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
1631 	u_int32_t major, minor;
1632 	enum vtype vtyp;
1633 	char *cp2;
1634 	struct mbuf *mb, *mreq;
1635 	struct vnode *vp, *dirp = (struct vnode *)0;
1636 	nfsrvfh_t nsfh;
1637 	u_quad_t frev;
1638 	struct mount *mp = NULL;
1639 
1640 	nd.ni_cnd.cn_nameiop = 0;
1641 	nfsm_srvmtofh(&nsfh);
1642 	if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL)
1643 		return (ESTALE);
1644 	vn_start_write(NULL, &mp, V_WAIT);
1645 	nfsm_srvnamesiz(len);
1646 	nd.ni_cnd.cn_cred = cred;
1647 	nd.ni_cnd.cn_nameiop = CREATE;
1648 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1649 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1650 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1651 	if (dirp)
1652 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp);
1653 	if (error) {
1654 		nfsm_reply(NFSX_WCCDATA(1));
1655 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1656 		if (dirp)
1657 			vrele(dirp);
1658 		vn_finished_write(mp, 0);
1659 		return (0);
1660 	}
1661 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1662 	vtyp = nfsv3tov_type(*tl);
1663 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1664 		error = NFSERR_BADTYPE;
1665 		goto abort;
1666 	}
1667 	VATTR_NULL(&va);
1668 	va.va_mode = 0;
1669 	nfsm_srvsattr(&va);
1670 	if (vtyp == VCHR || vtyp == VBLK) {
1671 		dev_t rdev;
1672 
1673 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1674 		major = fxdr_unsigned(u_int32_t, *tl++);
1675 		minor = fxdr_unsigned(u_int32_t, *tl);
1676 		rdev = makedev(major, minor);
1677 		if (major(rdev) != major || minor(rdev) != minor) {
1678 			error = EINVAL;
1679 			goto abort;
1680 		}
1681 		va.va_rdev = rdev;
1682 	}
1683 
1684 	/*
1685 	 * Iff doesn't exist, create it.
1686 	 */
1687 	if (nd.ni_vp) {
1688 		error = EEXIST;
1689 abort:
1690 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1691 		if (nd.ni_dvp == nd.ni_vp)
1692 			vrele(nd.ni_dvp);
1693 		else
1694 			vput(nd.ni_dvp);
1695 		if (nd.ni_vp)
1696 			vput(nd.ni_vp);
1697 		goto out;
1698 	}
1699 	va.va_type = vtyp;
1700 	if (vtyp == VSOCK) {
1701 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1702 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1703 	} else {
1704 		if (va.va_type != VFIFO &&
1705 		    (error = kauth_authorize_generic(cred,
1706 		    KAUTH_GENERIC_ISSUSER, (u_short *)0))) {
1707 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1708 			vput(nd.ni_dvp);
1709 			goto out;
1710 		}
1711 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1712 		error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1713 		if (error)
1714 			goto out;
1715 		if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1716 			vput(nd.ni_vp);
1717 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1718 			error = EINVAL;
1719 		}
1720 	}
1721 out:
1722 	vp = nd.ni_vp;
1723 	if (!error) {
1724 		error = nfsrv_composefh(vp, &nsfh, TRUE);
1725 		if (!error)
1726 			error = VOP_GETATTR(vp, &va, cred, lwp);
1727 		vput(vp);
1728 	}
1729 	if (dirp) {
1730 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp);
1731 		vrele(dirp);
1732 	}
1733 	nfsm_reply(NFSX_SRVFH(&nsfh, TRUE) + NFSX_POSTOPATTR(1) +
1734 	    NFSX_WCCDATA(1));
1735 	if (!error) {
1736 		nfsm_srvpostop_fh(&nsfh);
1737 		nfsm_srvpostop_attr(0, &va);
1738 	}
1739 	nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1740 	vn_finished_write(mp, 0);
1741 	return (0);
1742 nfsmout:
1743 	if (dirp)
1744 		vrele(dirp);
1745 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1746 	if (nd.ni_dvp == nd.ni_vp)
1747 		vrele(nd.ni_dvp);
1748 	else
1749 		vput(nd.ni_dvp);
1750 	if (nd.ni_vp)
1751 		vput(nd.ni_vp);
1752 	vn_finished_write(mp, 0);
1753 	return (error);
1754 }
1755 
1756 /*
1757  * nfs remove service
1758  */
1759 int
1760 nfsrv_remove(nfsd, slp, lwp, mrq)
1761 	struct nfsrv_descript *nfsd;
1762 	struct nfssvc_sock *slp;
1763 	struct lwp *lwp;
1764 	struct mbuf **mrq;
1765 {
1766 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1767 	struct mbuf *nam = nfsd->nd_nam;
1768 	caddr_t dpos = nfsd->nd_dpos;
1769 	kauth_cred_t cred = nfsd->nd_cr;
1770 	struct nameidata nd;
1771 	u_int32_t *tl;
1772 	int32_t t1;
1773 	caddr_t bpos;
1774 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
1775 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1776 	char *cp2;
1777 	struct mbuf *mb, *mreq;
1778 	struct vnode *vp, *dirp;
1779 	struct vattr dirfor, diraft;
1780 	nfsrvfh_t nsfh;
1781 	u_quad_t frev;
1782 	struct mount *mp = NULL;
1783 
1784 #ifndef nolint
1785 	vp = (struct vnode *)0;
1786 #endif
1787 	nfsm_srvmtofh(&nsfh);
1788 	if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL)
1789 		return (ESTALE);
1790 	vn_start_write(NULL, &mp, V_WAIT);
1791 	nfsm_srvnamesiz(len);
1792 	nd.ni_cnd.cn_cred = cred;
1793 	nd.ni_cnd.cn_nameiop = DELETE;
1794 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1795 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
1796 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1797 	if (dirp) {
1798 		if (v3)
1799 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp);
1800 		else
1801 			vrele(dirp);
1802 	}
1803 	if (!error) {
1804 		vp = nd.ni_vp;
1805 		if (vp->v_type == VDIR &&
1806 		    (error = kauth_authorize_generic(cred,
1807 		    KAUTH_GENERIC_ISSUSER, (u_short *)0)) != 0)
1808 			goto out;
1809 		/*
1810 		 * The root of a mounted filesystem cannot be deleted.
1811 		 */
1812 		if (vp->v_flag & VROOT) {
1813 			error = EBUSY;
1814 			goto out;
1815 		}
1816 out:
1817 		if (!error) {
1818 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1819 			nqsrv_getl(vp, ND_WRITE);
1820 			error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
1821 		} else {
1822 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1823 			if (nd.ni_dvp == vp)
1824 				vrele(nd.ni_dvp);
1825 			else
1826 				vput(nd.ni_dvp);
1827 			vput(vp);
1828 		}
1829 	}
1830 	if (dirp && v3) {
1831 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp);
1832 		vrele(dirp);
1833 	}
1834 	nfsm_reply(NFSX_WCCDATA(v3));
1835 	if (v3) {
1836 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1837 		vn_finished_write(mp, 0);
1838 		return (0);
1839 	}
1840 	vn_finished_write(mp, 0);
1841 	nfsm_srvdone;
1842 }
1843 
1844 /*
1845  * nfs rename service
1846  */
1847 int
1848 nfsrv_rename(nfsd, slp, lwp, mrq)
1849 	struct nfsrv_descript *nfsd;
1850 	struct nfssvc_sock *slp;
1851 	struct lwp *lwp;
1852 	struct mbuf **mrq;
1853 {
1854 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1855 	struct mbuf *nam = nfsd->nd_nam;
1856 	caddr_t dpos = nfsd->nd_dpos;
1857 	kauth_cred_t cred = nfsd->nd_cr;
1858 	u_int32_t *tl;
1859 	int32_t t1;
1860 	caddr_t bpos;
1861 	int error = 0, cache = 0, fdirfor_ret = 1, fdiraft_ret = 1;
1862 	uint32_t len, len2;
1863 	int tdirfor_ret = 1, tdiraft_ret = 1;
1864 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1865 	char *cp2;
1866 	struct mbuf *mb, *mreq;
1867 	struct nameidata fromnd, tond;
1868 	struct vnode *fvp, *tvp, *tdvp, *fdirp = (struct vnode *)0;
1869 	struct vnode *tdirp = (struct vnode *)0;
1870 	struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
1871 	nfsrvfh_t fnsfh, tnsfh;
1872 	u_quad_t frev;
1873 	uid_t saved_uid;
1874 	struct mount *mp = NULL;
1875 
1876 #ifndef nolint
1877 	fvp = (struct vnode *)0;
1878 #endif
1879 	fromnd.ni_cnd.cn_nameiop = 0;
1880 	tond.ni_cnd.cn_nameiop = 0;
1881 	nfsm_srvmtofh(&fnsfh);
1882 	if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&fnsfh)->fh_fsid)) == NULL)
1883 		return (ESTALE);
1884 	vn_start_write(NULL, &mp, V_WAIT);
1885 	nfsm_srvnamesiz(len);
1886 	/*
1887 	 * Remember our original uid so that we can reset cr_uid before
1888 	 * the second nfs_namei() call, in case it is remapped.
1889 	 */
1890 	saved_uid = kauth_cred_geteuid(cred);
1891 	fromnd.ni_cnd.cn_cred = cred;
1892 	fromnd.ni_cnd.cn_nameiop = DELETE;
1893 	fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART;
1894 	error = nfs_namei(&fromnd, &fnsfh, len, slp, nam, &md,
1895 		&dpos, &fdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1896 	if (fdirp) {
1897 		if (v3)
1898 			fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred, lwp);
1899 		else {
1900 			vrele(fdirp);
1901 			fdirp = (struct vnode *)0;
1902 		}
1903 	}
1904 	if (error) {
1905 		nfsm_reply(2 * NFSX_WCCDATA(v3));
1906 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1907 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1908 		if (fdirp)
1909 			vrele(fdirp);
1910 		vn_finished_write(mp, 0);
1911 		return (0);
1912 	}
1913 	fvp = fromnd.ni_vp;
1914 	nfsm_srvmtofh(&tnsfh);
1915 	if (v3) {
1916 		nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1917 		len2 = fxdr_unsigned(uint32_t, *tl);
1918 		/* len2 will be checked by nfs_namei */
1919 	}
1920 	else {
1921 		/* NFSv2 */
1922 		nfsm_strsiz(len2, NFS_MAXNAMLEN);
1923 	}
1924 	kauth_cred_seteuid(cred, saved_uid);
1925 	tond.ni_cnd.cn_cred = cred;
1926 	tond.ni_cnd.cn_nameiop = RENAME;
1927 	tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART;
1928 	error = nfs_namei(&tond, &tnsfh, len2, slp, nam, &md,
1929 		&dpos, &tdirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1930 	if (tdirp) {
1931 		if (v3)
1932 			tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred, lwp);
1933 		else {
1934 			vrele(tdirp);
1935 			tdirp = (struct vnode *)0;
1936 		}
1937 	}
1938 	if (error) {
1939 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1940 		vrele(fromnd.ni_dvp);
1941 		vrele(fvp);
1942 		goto out1;
1943 	}
1944 	tdvp = tond.ni_dvp;
1945 	tvp = tond.ni_vp;
1946 	if (tvp != NULL) {
1947 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1948 			if (v3)
1949 				error = EEXIST;
1950 			else
1951 				error = EISDIR;
1952 			goto out;
1953 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1954 			if (v3)
1955 				error = EEXIST;
1956 			else
1957 				error = ENOTDIR;
1958 			goto out;
1959 		}
1960 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1961 			if (v3)
1962 				error = EXDEV;
1963 			else
1964 				error = ENOTEMPTY;
1965 			goto out;
1966 		}
1967 	}
1968 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1969 		if (v3)
1970 			error = EXDEV;
1971 		else
1972 			error = ENOTEMPTY;
1973 		goto out;
1974 	}
1975 	if (fvp->v_mount != tdvp->v_mount) {
1976 		if (v3)
1977 			error = EXDEV;
1978 		else
1979 			error = ENOTEMPTY;
1980 		goto out;
1981 	}
1982 	if (fvp == tdvp) {
1983 		if (v3)
1984 			error = EINVAL;
1985 		else
1986 			error = ENOTEMPTY;
1987 	}
1988 	/*
1989 	 * If source is the same as the destination (that is the
1990 	 * same vnode with the same name in the same directory),
1991 	 * then there is nothing to do.
1992 	 */
1993 	if (fvp == tvp && fromnd.ni_dvp == tdvp &&
1994 	    fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
1995 	    !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
1996 	      fromnd.ni_cnd.cn_namelen))
1997 		error = -1;
1998 out:
1999 	if (!error) {
2000 		nqsrv_getl(fromnd.ni_dvp, ND_WRITE);
2001 		nqsrv_getl(tdvp, ND_WRITE);
2002 		if (tvp) {
2003 			nqsrv_getl(tvp, ND_WRITE);
2004 		}
2005 		error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
2006 				   tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
2007 	} else {
2008 		VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd);
2009 		if (tdvp == tvp)
2010 			vrele(tdvp);
2011 		else
2012 			vput(tdvp);
2013 		if (tvp)
2014 			vput(tvp);
2015 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2016 		vrele(fromnd.ni_dvp);
2017 		vrele(fvp);
2018 		if (error == -1)
2019 			error = 0;
2020 	}
2021 	vrele(tond.ni_startdir);
2022 	PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
2023 out1:
2024 	if (fdirp) {
2025 		fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, lwp);
2026 		vrele(fdirp);
2027 	}
2028 	if (tdirp) {
2029 		tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, lwp);
2030 		vrele(tdirp);
2031 	}
2032 	vrele(fromnd.ni_startdir);
2033 	PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
2034 	nfsm_reply(2 * NFSX_WCCDATA(v3));
2035 	if (v3) {
2036 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2037 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2038 	}
2039 	vn_finished_write(mp, 0);
2040 	return (0);
2041 
2042 nfsmout:
2043 	if (fdirp)
2044 		vrele(fdirp);
2045 #ifdef notdef
2046 	if (tdirp)
2047 		vrele(tdirp);
2048 #endif
2049 	if (tond.ni_cnd.cn_nameiop) {
2050 		vrele(tond.ni_startdir);
2051 		PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
2052 	}
2053 	if (fromnd.ni_cnd.cn_nameiop) {
2054 		vrele(fromnd.ni_startdir);
2055 		PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
2056 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2057 		vrele(fromnd.ni_dvp);
2058 		vrele(fvp);
2059 	}
2060 	return (error);
2061 }
2062 
2063 /*
2064  * nfs link service
2065  */
2066 int
2067 nfsrv_link(nfsd, slp, lwp, mrq)
2068 	struct nfsrv_descript *nfsd;
2069 	struct nfssvc_sock *slp;
2070 	struct lwp *lwp;
2071 	struct mbuf **mrq;
2072 {
2073 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2074 	struct mbuf *nam = nfsd->nd_nam;
2075 	caddr_t dpos = nfsd->nd_dpos;
2076 	kauth_cred_t cred = nfsd->nd_cr;
2077 	struct nameidata nd;
2078 	u_int32_t *tl;
2079 	int32_t t1;
2080 	caddr_t bpos;
2081 	int error = 0, rdonly, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2082 	int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2083 	char *cp2;
2084 	struct mbuf *mb, *mreq;
2085 	struct vnode *vp, *xp, *dirp = (struct vnode *)0;
2086 	struct vattr dirfor, diraft, at;
2087 	nfsrvfh_t nsfh, dnsfh;
2088 	u_quad_t frev;
2089 	struct mount *mp = NULL;
2090 
2091 	nfsm_srvmtofh(&nsfh);
2092 	if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL)
2093 		return (ESTALE);
2094 	vn_start_write(NULL, &mp, V_WAIT);
2095 	nfsm_srvmtofh(&dnsfh);
2096 	nfsm_srvnamesiz(len);
2097 	error = nfsrv_fhtovp(&nsfh, FALSE, &vp, cred, slp, nam,
2098 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2099 	if (error) {
2100 		nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2101 		nfsm_srvpostop_attr(getret, &at);
2102 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2103 		vn_finished_write(mp, 0);
2104 		return (0);
2105 	}
2106 	if (vp->v_type == VDIR && (error = kauth_authorize_generic(cred,
2107 	    KAUTH_GENERIC_ISSUSER, (u_short *)0)) != 0)
2108 		goto out1;
2109 	nd.ni_cnd.cn_cred = cred;
2110 	nd.ni_cnd.cn_nameiop = CREATE;
2111 	nd.ni_cnd.cn_flags = LOCKPARENT;
2112 	error = nfs_namei(&nd, &dnsfh, len, slp, nam, &md, &dpos,
2113 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2114 	if (dirp) {
2115 		if (v3)
2116 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp);
2117 		else {
2118 			vrele(dirp);
2119 			dirp = (struct vnode *)0;
2120 		}
2121 	}
2122 	if (error)
2123 		goto out1;
2124 	xp = nd.ni_vp;
2125 	if (xp != NULL) {
2126 		error = EEXIST;
2127 		goto out;
2128 	}
2129 	xp = nd.ni_dvp;
2130 	if (vp->v_mount != xp->v_mount)
2131 		error = EXDEV;
2132 out:
2133 	if (!error) {
2134 		nqsrv_getl(vp, ND_WRITE);
2135 		nqsrv_getl(xp, ND_WRITE);
2136 		error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2137 	} else {
2138 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2139 		if (nd.ni_dvp == nd.ni_vp)
2140 			vrele(nd.ni_dvp);
2141 		else
2142 			vput(nd.ni_dvp);
2143 		if (nd.ni_vp)
2144 			vrele(nd.ni_vp);
2145 	}
2146 out1:
2147 	if (v3)
2148 		getret = VOP_GETATTR(vp, &at, cred, lwp);
2149 	if (dirp) {
2150 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp);
2151 		vrele(dirp);
2152 	}
2153 	vrele(vp);
2154 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2155 	if (v3) {
2156 		nfsm_srvpostop_attr(getret, &at);
2157 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2158 		vn_finished_write(mp, 0);
2159 		return (0);
2160 	}
2161 	vn_finished_write(mp, 0);
2162 	nfsm_srvdone;
2163 }
2164 
2165 /*
2166  * nfs symbolic link service
2167  */
2168 int
2169 nfsrv_symlink(nfsd, slp, lwp, mrq)
2170 	struct nfsrv_descript *nfsd;
2171 	struct nfssvc_sock *slp;
2172 	struct lwp *lwp;
2173 	struct mbuf **mrq;
2174 {
2175 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2176 	struct mbuf *nam = nfsd->nd_nam;
2177 	caddr_t dpos = nfsd->nd_dpos;
2178 	kauth_cred_t cred = nfsd->nd_cr;
2179 	struct vattr va, dirfor, diraft;
2180 	struct nameidata nd;
2181 	u_int32_t *tl;
2182 	int32_t t1;
2183 	struct nfsv2_sattr *sp;
2184 	char *bpos, *pathcp = NULL, *cp2;
2185 	struct uio io;
2186 	struct iovec iv;
2187 	int error = 0, cache = 0, dirfor_ret = 1, diraft_ret = 1;
2188 	uint32_t len, len2;
2189 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2190 	struct mbuf *mb, *mreq;
2191 	struct vnode *dirp = (struct vnode *)0;
2192 	nfsrvfh_t nsfh;
2193 	u_quad_t frev;
2194 	struct mount *mp = NULL;
2195 
2196 	nd.ni_cnd.cn_nameiop = 0;
2197 	nfsm_srvmtofh(&nsfh);
2198 	if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL)
2199 		return (ESTALE);
2200 	vn_start_write(NULL, &mp, V_WAIT);
2201 	nfsm_srvnamesiz(len);
2202 	nd.ni_cnd.cn_cred = cred;
2203 	nd.ni_cnd.cn_nameiop = CREATE;
2204 	nd.ni_cnd.cn_flags = LOCKPARENT;
2205 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2206 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2207 	if (dirp) {
2208 		if (v3)
2209 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp);
2210 		else {
2211 			vrele(dirp);
2212 			dirp = (struct vnode *)0;
2213 		}
2214 	}
2215 	if (error)
2216 		goto out;
2217 	VATTR_NULL(&va);
2218 	va.va_type = VLNK;
2219 	if (v3) {
2220 		va.va_mode = 0;
2221 		nfsm_srvsattr(&va);
2222 		nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
2223 		len2 = fxdr_unsigned(uint32_t, *tl);
2224 		if (len2 > PATH_MAX) {
2225 			/* XXX should check _PC_NO_TRUNC */
2226 			error = ENAMETOOLONG;
2227 			goto abortop;
2228 		}
2229 	}
2230 	else {
2231 		/* NFSv2 */
2232 		nfsm_strsiz(len2, NFS_MAXPATHLEN);
2233 	}
2234 	pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK);
2235 	iv.iov_base = pathcp;
2236 	iv.iov_len = len2;
2237 	io.uio_resid = len2;
2238 	io.uio_offset = 0;
2239 	io.uio_iov = &iv;
2240 	io.uio_iovcnt = 1;
2241 	io.uio_rw = UIO_READ;
2242 	UIO_SETUP_SYSSPACE(&io);
2243 	nfsm_mtouio(&io, len2);
2244 	if (!v3) {
2245 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2246 		va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
2247 	}
2248 	*(pathcp + len2) = '\0';
2249 	if (nd.ni_vp) {
2250 		error = EEXIST;
2251 abortop:
2252 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2253 		if (nd.ni_dvp == nd.ni_vp)
2254 			vrele(nd.ni_dvp);
2255 		else
2256 			vput(nd.ni_dvp);
2257 		if (nd.ni_vp)
2258 			vrele(nd.ni_vp);
2259 		goto out;
2260 	}
2261 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2262 	error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp);
2263 	if (!error) {
2264 	    if (v3) {
2265 		error = nfsrv_composefh(nd.ni_vp, &nsfh, v3);
2266 		if (!error)
2267 		    error = VOP_GETATTR(nd.ni_vp, &va, cred, lwp);
2268 		vput(nd.ni_vp);
2269 	    } else {
2270 		vput(nd.ni_vp);
2271 	    }
2272 	}
2273 out:
2274 	if (pathcp)
2275 		free(pathcp, M_TEMP);
2276 	if (dirp) {
2277 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp);
2278 		vrele(dirp);
2279 	}
2280 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) +
2281 	    NFSX_WCCDATA(v3));
2282 	if (v3) {
2283 		if (!error) {
2284 			nfsm_srvpostop_fh(&nsfh);
2285 			nfsm_srvpostop_attr(0, &va);
2286 		}
2287 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2288 	}
2289 	vn_finished_write(mp, 0);
2290 	return (0);
2291 nfsmout:
2292 	if (dirp)
2293 		vrele(dirp);
2294 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2295 	if (nd.ni_dvp == nd.ni_vp)
2296 		vrele(nd.ni_dvp);
2297 	else
2298 		vput(nd.ni_dvp);
2299 	if (nd.ni_vp)
2300 		vrele(nd.ni_vp);
2301 	if (pathcp)
2302 		free(pathcp, M_TEMP);
2303 	vn_finished_write(mp, 0);
2304 	return (error);
2305 }
2306 
2307 /*
2308  * nfs mkdir service
2309  */
2310 int
2311 nfsrv_mkdir(nfsd, slp, lwp, mrq)
2312 	struct nfsrv_descript *nfsd;
2313 	struct nfssvc_sock *slp;
2314 	struct lwp *lwp;
2315 	struct mbuf **mrq;
2316 {
2317 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2318 	struct mbuf *nam = nfsd->nd_nam;
2319 	caddr_t dpos = nfsd->nd_dpos;
2320 	kauth_cred_t cred = nfsd->nd_cr;
2321 	struct vattr va, dirfor, diraft;
2322 	struct nfs_fattr *fp;
2323 	struct nameidata nd;
2324 	caddr_t cp;
2325 	u_int32_t *tl;
2326 	int32_t t1;
2327 	caddr_t bpos;
2328 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2329 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2330 	char *cp2;
2331 	struct mbuf *mb, *mreq;
2332 	struct vnode *vp, *dirp = (struct vnode *)0;
2333 	nfsrvfh_t nsfh;
2334 	u_quad_t frev;
2335 	struct mount *mp = NULL;
2336 
2337 	nfsm_srvmtofh(&nsfh);
2338 	if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL)
2339 		return (ESTALE);
2340 	vn_start_write(NULL, &mp, V_WAIT);
2341 	nfsm_srvnamesiz(len);
2342 	nd.ni_cnd.cn_cred = cred;
2343 	nd.ni_cnd.cn_nameiop = CREATE;
2344 	nd.ni_cnd.cn_flags = LOCKPARENT;
2345 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2346 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2347 	if (dirp) {
2348 		if (v3)
2349 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp);
2350 		else {
2351 			vrele(dirp);
2352 			dirp = (struct vnode *)0;
2353 		}
2354 	}
2355 	if (error) {
2356 		nfsm_reply(NFSX_WCCDATA(v3));
2357 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2358 		if (dirp)
2359 			vrele(dirp);
2360 		vn_finished_write(mp, 0);
2361 		return (0);
2362 	}
2363 	VATTR_NULL(&va);
2364 	if (v3) {
2365 		va.va_mode = 0;
2366 		nfsm_srvsattr(&va);
2367 	} else {
2368 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2369 		va.va_mode = nfstov_mode(*tl++);
2370 	}
2371 	va.va_type = VDIR;
2372 	vp = nd.ni_vp;
2373 	if (vp != NULL) {
2374 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2375 		if (nd.ni_dvp == vp)
2376 			vrele(nd.ni_dvp);
2377 		else
2378 			vput(nd.ni_dvp);
2379 		vrele(vp);
2380 		error = EEXIST;
2381 		goto out;
2382 	}
2383 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2384 	error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
2385 	if (!error) {
2386 		vp = nd.ni_vp;
2387 		error = nfsrv_composefh(vp, &nsfh, v3);
2388 		if (!error)
2389 			error = VOP_GETATTR(vp, &va, cred, lwp);
2390 		vput(vp);
2391 	}
2392 out:
2393 	if (dirp) {
2394 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp);
2395 		vrele(dirp);
2396 	}
2397 	nfsm_reply(NFSX_SRVFH(&nsfh, v3) + NFSX_POSTOPATTR(v3) +
2398 	    NFSX_WCCDATA(v3));
2399 	if (v3) {
2400 		if (!error) {
2401 			nfsm_srvpostop_fh(&nsfh);
2402 			nfsm_srvpostop_attr(0, &va);
2403 		}
2404 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2405 	} else {
2406 		nfsm_srvfhtom(&nsfh, v3);
2407 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2408 		nfsm_srvfillattr(&va, fp);
2409 	}
2410 	vn_finished_write(mp, 0);
2411 	return (0);
2412 nfsmout:
2413 	if (dirp)
2414 		vrele(dirp);
2415 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2416 	if (nd.ni_dvp == nd.ni_vp)
2417 		vrele(nd.ni_dvp);
2418 	else
2419 		vput(nd.ni_dvp);
2420 	if (nd.ni_vp)
2421 		vrele(nd.ni_vp);
2422 	vn_finished_write(mp, 0);
2423 	return (error);
2424 }
2425 
2426 /*
2427  * nfs rmdir service
2428  */
2429 int
2430 nfsrv_rmdir(nfsd, slp, lwp, mrq)
2431 	struct nfsrv_descript *nfsd;
2432 	struct nfssvc_sock *slp;
2433 	struct lwp *lwp;
2434 	struct mbuf **mrq;
2435 {
2436 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2437 	struct mbuf *nam = nfsd->nd_nam;
2438 	caddr_t dpos = nfsd->nd_dpos;
2439 	kauth_cred_t cred = nfsd->nd_cr;
2440 	u_int32_t *tl;
2441 	int32_t t1;
2442 	caddr_t bpos;
2443 	int error = 0, cache = 0, len, dirfor_ret = 1, diraft_ret = 1;
2444 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2445 	char *cp2;
2446 	struct mbuf *mb, *mreq;
2447 	struct vnode *vp, *dirp = (struct vnode *)0;
2448 	struct vattr dirfor, diraft;
2449 	nfsrvfh_t nsfh;
2450 	struct nameidata nd;
2451 	u_quad_t frev;
2452 	struct mount *mp = NULL;
2453 
2454 	nfsm_srvmtofh(&nsfh);
2455 	if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL)
2456 		return (ESTALE);
2457 	vn_start_write(NULL, &mp, V_WAIT);
2458 	nfsm_srvnamesiz(len);
2459 	nd.ni_cnd.cn_cred = cred;
2460 	nd.ni_cnd.cn_nameiop = DELETE;
2461 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
2462 	error = nfs_namei(&nd, &nsfh, len, slp, nam, &md, &dpos,
2463 		&dirp, lwp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2464 	if (dirp) {
2465 		if (v3)
2466 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, lwp);
2467 		else {
2468 			vrele(dirp);
2469 			dirp = (struct vnode *)0;
2470 		}
2471 	}
2472 	if (error) {
2473 		nfsm_reply(NFSX_WCCDATA(v3));
2474 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2475 		if (dirp)
2476 			vrele(dirp);
2477 		vn_finished_write(mp, 0);
2478 		return (0);
2479 	}
2480 	vp = nd.ni_vp;
2481 	if (vp->v_type != VDIR) {
2482 		error = ENOTDIR;
2483 		goto out;
2484 	}
2485 	/*
2486 	 * No rmdir "." please.
2487 	 */
2488 	if (nd.ni_dvp == vp) {
2489 		error = EINVAL;
2490 		goto out;
2491 	}
2492 	/*
2493 	 * The root of a mounted filesystem cannot be deleted.
2494 	 */
2495 	if (vp->v_flag & VROOT)
2496 		error = EBUSY;
2497 out:
2498 	if (!error) {
2499 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
2500 		nqsrv_getl(vp, ND_WRITE);
2501 		error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2502 	} else {
2503 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2504 		if (nd.ni_dvp == nd.ni_vp)
2505 			vrele(nd.ni_dvp);
2506 		else
2507 			vput(nd.ni_dvp);
2508 		vput(vp);
2509 	}
2510 	if (dirp) {
2511 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, lwp);
2512 		vrele(dirp);
2513 	}
2514 	nfsm_reply(NFSX_WCCDATA(v3));
2515 	if (v3) {
2516 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2517 		vn_finished_write(mp, 0);
2518 		return (0);
2519 	}
2520 	vn_finished_write(mp, 0);
2521 	nfsm_srvdone;
2522 }
2523 
2524 /*
2525  * nfs readdir service
2526  * - mallocs what it thinks is enough to read
2527  *	count rounded up to a multiple of NFS_SRVDIRBLKSIZ <= NFS_MAXREADDIR
2528  * - calls VOP_READDIR()
2529  * - loops around building the reply
2530  *	if the output generated exceeds count break out of loop
2531  *	The nfsm_clget macro is used here so that the reply will be packed
2532  *	tightly in mbuf clusters.
2533  * - it only knows that it has encountered eof when the VOP_READDIR()
2534  *	reads nothing
2535  * - as such one readdir rpc will return eof false although you are there
2536  *	and then the next will return eof
2537  * - it trims out records with d_fileno == 0
2538  *	this doesn't matter for Unix clients, but they might confuse clients
2539  *	for other os'.
2540  * - it trims out records with d_type == DT_WHT
2541  *	these cannot be seen through NFS (unless we extend the protocol)
2542  * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2543  *	than requested, but this may not apply to all filesystems. For
2544  *	example, client NFS does not { although it is never remote mounted
2545  *	anyhow }
2546  *     The alternate call nfsrv_readdirplus() does lookups as well.
2547  * PS: The NFS protocol spec. does not clarify what the "count" byte
2548  *	argument is a count of.. just name strings and file id's or the
2549  *	entire reply rpc or ...
2550  *	I tried just file name and id sizes and it confused the Sun client,
2551  *	so I am using the full rpc size now. The "paranoia.." comment refers
2552  *	to including the status longwords that are not a part of the dir.
2553  *	"entry" structures, but are in the rpc.
2554  */
2555 
2556 #define	NFS_SRVDIRBLKSIZ	1024
2557 
2558 struct flrep {
2559 	nfsuint64 fl_off;
2560 	u_int32_t fl_postopok;
2561 	u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2562 	u_int32_t fl_fhok;
2563 	u_int32_t fl_fhsize;
2564 };
2565 
2566 int
2567 nfsrv_readdir(nfsd, slp, lwp, mrq)
2568 	struct nfsrv_descript *nfsd;
2569 	struct nfssvc_sock *slp;
2570 	struct lwp *lwp;
2571 	struct mbuf **mrq;
2572 {
2573 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2574 	struct mbuf *nam = nfsd->nd_nam;
2575 	caddr_t dpos = nfsd->nd_dpos;
2576 	kauth_cred_t cred = nfsd->nd_cr;
2577 	char *bp, *be;
2578 	struct mbuf *mp;
2579 	struct dirent *dp;
2580 	caddr_t cp;
2581 	u_int32_t *tl;
2582 	int32_t t1;
2583 	caddr_t bpos;
2584 	struct mbuf *mb, *mreq, *mp2;
2585 	char *cpos, *cend, *cp2, *rbuf;
2586 	struct vnode *vp;
2587 	struct vattr at;
2588 	nfsrvfh_t nsfh;
2589 	struct uio io;
2590 	struct iovec iv;
2591 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2592 	int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, ncookies;
2593 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2594 	u_quad_t frev, off, toff, verf;
2595 	off_t *cookies = NULL, *cookiep;
2596 	nfsuint64 jar;
2597 
2598 	nfsm_srvmtofh(&nsfh);
2599 	if (v3) {
2600 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2601 		toff = fxdr_hyper(tl);
2602 		tl += 2;
2603 		verf = fxdr_hyper(tl);
2604 		tl += 2;
2605 	} else {
2606 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2607 		toff = fxdr_unsigned(u_quad_t, *tl++);
2608 	}
2609 	off = toff;
2610 	cnt = fxdr_unsigned(int, *tl);
2611 	siz = ((cnt + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1));
2612 	xfer = NFS_SRVMAXDATA(nfsd);
2613 	if (siz > xfer)
2614 		siz = xfer;
2615 	fullsiz = siz;
2616 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
2617 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2618 	if (!error && vp->v_type != VDIR) {
2619 		error = ENOTDIR;
2620 		vput(vp);
2621 	}
2622 	if (error) {
2623 		nfsm_reply(NFSX_UNSIGNED);
2624 		nfsm_srvpostop_attr(getret, &at);
2625 		return (0);
2626 	}
2627 	nqsrv_getl(vp, ND_READ);
2628 	if (v3) {
2629 		error = getret = VOP_GETATTR(vp, &at, cred, lwp);
2630 #ifdef NFS3_STRICTVERF
2631 		/*
2632 		 * XXX This check is too strict for Solaris 2.5 clients.
2633 		 */
2634 		if (!error && toff && verf != at.va_filerev)
2635 			error = NFSERR_BAD_COOKIE;
2636 #endif
2637 	}
2638 	if (!error)
2639 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0);
2640 	if (error) {
2641 		vput(vp);
2642 		nfsm_reply(NFSX_POSTOPATTR(v3));
2643 		nfsm_srvpostop_attr(getret, &at);
2644 		return (0);
2645 	}
2646 	VOP_UNLOCK(vp, 0);
2647 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2648 again:
2649 	iv.iov_base = rbuf;
2650 	iv.iov_len = fullsiz;
2651 	io.uio_iov = &iv;
2652 	io.uio_iovcnt = 1;
2653 	io.uio_offset = (off_t)off;
2654 	io.uio_resid = fullsiz;
2655 	io.uio_rw = UIO_READ;
2656 	UIO_SETUP_SYSSPACE(&io);
2657 	eofflag = 0;
2658 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2659 
2660 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2661 
2662 	off = (off_t)io.uio_offset;
2663 	if (!cookies && !error)
2664 		error = NFSERR_PERM;
2665 	if (v3) {
2666 		getret = VOP_GETATTR(vp, &at, cred, lwp);
2667 		if (!error)
2668 			error = getret;
2669 	}
2670 
2671 	VOP_UNLOCK(vp, 0);
2672 	if (error) {
2673 		vrele(vp);
2674 		free((caddr_t)rbuf, M_TEMP);
2675 		if (cookies)
2676 			free((caddr_t)cookies, M_TEMP);
2677 		nfsm_reply(NFSX_POSTOPATTR(v3));
2678 		nfsm_srvpostop_attr(getret, &at);
2679 		return (0);
2680 	}
2681 	if (io.uio_resid) {
2682 		siz -= io.uio_resid;
2683 
2684 		/*
2685 		 * If nothing read, return eof
2686 		 * rpc reply
2687 		 */
2688 		if (siz == 0) {
2689 			vrele(vp);
2690 			nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2691 				2 * NFSX_UNSIGNED);
2692 			if (v3) {
2693 				nfsm_srvpostop_attr(getret, &at);
2694 				nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2695 				txdr_hyper(at.va_filerev, tl);
2696 				tl += 2;
2697 			} else
2698 				nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2699 			*tl++ = nfs_false;
2700 			*tl = nfs_true;
2701 			free((caddr_t)rbuf, M_TEMP);
2702 			free((caddr_t)cookies, M_TEMP);
2703 			return (0);
2704 		}
2705 	}
2706 
2707 	/*
2708 	 * Check for degenerate cases of nothing useful read.
2709 	 * If so go try again
2710 	 */
2711 	cpos = rbuf;
2712 	cend = rbuf + siz;
2713 	dp = (struct dirent *)cpos;
2714 	cookiep = cookies;
2715 
2716 	while (cpos < cend && ncookies > 0 &&
2717 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2718 		cpos += dp->d_reclen;
2719 		dp = (struct dirent *)cpos;
2720 		cookiep++;
2721 		ncookies--;
2722 	}
2723 	if (cpos >= cend || ncookies == 0) {
2724 		toff = off;
2725 		siz = fullsiz;
2726 		free(cookies, M_TEMP);
2727 		goto again;
2728 	}
2729 
2730 	len = 3 * NFSX_UNSIGNED;	/* paranoia, probably can be 0 */
2731 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2732 	if (v3) {
2733 		nfsm_srvpostop_attr(getret, &at);
2734 		nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2735 		txdr_hyper(at.va_filerev, tl);
2736 	}
2737 	mp = mp2 = mb;
2738 	bp = bpos;
2739 	be = bp + M_TRAILINGSPACE(mp);
2740 
2741 	/* Loop through the records and build reply */
2742 	while (cpos < cend && ncookies > 0) {
2743 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2744 			nlen = dp->d_namlen;
2745 			rem = nfsm_rndup(nlen)-nlen;
2746 			len += (4 * NFSX_UNSIGNED + nlen + rem);
2747 			if (v3)
2748 				len += 2 * NFSX_UNSIGNED;
2749 			if (len > cnt) {
2750 				eofflag = 0;
2751 				break;
2752 			}
2753 			/*
2754 			 * Build the directory record xdr from
2755 			 * the dirent entry.
2756 			 */
2757 			nfsm_clget;
2758 			*tl = nfs_true;
2759 			bp += NFSX_UNSIGNED;
2760 			if (v3) {
2761 				nfsm_clget;
2762 				*tl = txdr_unsigned(dp->d_fileno >> 32);
2763 				bp += NFSX_UNSIGNED;
2764 			}
2765 			nfsm_clget;
2766 			*tl = txdr_unsigned(dp->d_fileno);
2767 			bp += NFSX_UNSIGNED;
2768 			nfsm_clget;
2769 			*tl = txdr_unsigned(nlen);
2770 			bp += NFSX_UNSIGNED;
2771 
2772 			/* And loop around copying the name */
2773 			xfer = nlen;
2774 			cp = dp->d_name;
2775 			while (xfer > 0) {
2776 				nfsm_clget;
2777 				if ((bp+xfer) > be)
2778 					tsiz = be-bp;
2779 				else
2780 					tsiz = xfer;
2781 				memcpy(bp, cp, tsiz);
2782 				bp += tsiz;
2783 				xfer -= tsiz;
2784 				if (xfer > 0)
2785 					cp += tsiz;
2786 			}
2787 			/* And null pad to an int32_t boundary */
2788 			for (i = 0; i < rem; i++)
2789 				*bp++ = '\0';
2790 			nfsm_clget;
2791 
2792 			/* Finish off the record */
2793 			txdr_hyper(*cookiep, &jar);
2794 			if (v3) {
2795 				*tl = jar.nfsuquad[0];
2796 				bp += NFSX_UNSIGNED;
2797 				nfsm_clget;
2798 			}
2799 			*tl = jar.nfsuquad[1];
2800 			bp += NFSX_UNSIGNED;
2801 		}
2802 		cpos += dp->d_reclen;
2803 		dp = (struct dirent *)cpos;
2804 		cookiep++;
2805 		ncookies--;
2806 	}
2807 	vrele(vp);
2808 	nfsm_clget;
2809 	*tl = nfs_false;
2810 	bp += NFSX_UNSIGNED;
2811 	nfsm_clget;
2812 	if (eofflag)
2813 		*tl = nfs_true;
2814 	else
2815 		*tl = nfs_false;
2816 	bp += NFSX_UNSIGNED;
2817 	if (mp != mb) {
2818 		if (bp < be)
2819 			mp->m_len = bp - mtod(mp, caddr_t);
2820 	} else
2821 		mp->m_len += bp - bpos;
2822 	free((caddr_t)rbuf, M_TEMP);
2823 	free((caddr_t)cookies, M_TEMP);
2824 	nfsm_srvdone;
2825 }
2826 
2827 int
2828 nfsrv_readdirplus(nfsd, slp, lwp, mrq)
2829 	struct nfsrv_descript *nfsd;
2830 	struct nfssvc_sock *slp;
2831 	struct lwp *lwp;
2832 	struct mbuf **mrq;
2833 {
2834 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2835 	struct mbuf *nam = nfsd->nd_nam;
2836 	caddr_t dpos = nfsd->nd_dpos;
2837 	kauth_cred_t cred = nfsd->nd_cr;
2838 	char *bp, *be;
2839 	struct mbuf *mp;
2840 	struct dirent *dp;
2841 	caddr_t cp;
2842 	u_int32_t *tl;
2843 	int32_t t1;
2844 	caddr_t bpos;
2845 	struct mbuf *mb, *mreq, *mp2;
2846 	char *cpos, *cend, *cp2, *rbuf;
2847 	struct vnode *vp, *nvp;
2848 	struct flrep fl;
2849 	nfsrvfh_t nsfh;
2850 	struct uio io;
2851 	struct iovec iv;
2852 	struct vattr va, at, *vap = &va;
2853 	struct nfs_fattr *fp;
2854 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2855 	int siz, cnt, fullsiz, eofflag, rdonly, cache = 0, dirlen, ncookies;
2856 	u_quad_t frev, off, toff, verf;
2857 	off_t *cookies = NULL, *cookiep;
2858 
2859 	nfsm_srvmtofh(&nsfh);
2860 	nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2861 	toff = fxdr_hyper(tl);
2862 	tl += 2;
2863 	verf = fxdr_hyper(tl);
2864 	tl += 2;
2865 	siz = fxdr_unsigned(int, *tl++);
2866 	cnt = fxdr_unsigned(int, *tl);
2867 	off = toff;
2868 	siz = ((siz + NFS_SRVDIRBLKSIZ - 1) & ~(NFS_SRVDIRBLKSIZ - 1));
2869 	xfer = NFS_SRVMAXDATA(nfsd);
2870 	if (siz > xfer)
2871 		siz = xfer;
2872 	fullsiz = siz;
2873 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
2874 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2875 	if (!error && vp->v_type != VDIR) {
2876 		error = ENOTDIR;
2877 		vput(vp);
2878 	}
2879 	if (error) {
2880 		nfsm_reply(NFSX_UNSIGNED);
2881 		nfsm_srvpostop_attr(getret, &at);
2882 		return (0);
2883 	}
2884 	error = getret = VOP_GETATTR(vp, &at, cred, lwp);
2885 #ifdef NFS3_STRICTVERF
2886 	/*
2887 	 * XXX This check is too strict for Solaris 2.5 clients.
2888 	 */
2889 	if (!error && toff && verf != at.va_filerev)
2890 		error = NFSERR_BAD_COOKIE;
2891 #endif
2892 	if (!error) {
2893 		nqsrv_getl(vp, ND_READ);
2894 		error = nfsrv_access(vp, VEXEC, cred, rdonly, lwp, 0);
2895 	}
2896 	if (error) {
2897 		vput(vp);
2898 		nfsm_reply(NFSX_V3POSTOPATTR);
2899 		nfsm_srvpostop_attr(getret, &at);
2900 		return (0);
2901 	}
2902 	VOP_UNLOCK(vp, 0);
2903 
2904 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2905 again:
2906 	iv.iov_base = rbuf;
2907 	iv.iov_len = fullsiz;
2908 	io.uio_iov = &iv;
2909 	io.uio_iovcnt = 1;
2910 	io.uio_offset = (off_t)off;
2911 	io.uio_resid = fullsiz;
2912 	io.uio_rw = UIO_READ;
2913 	UIO_SETUP_SYSSPACE(&io);
2914 	eofflag = 0;
2915 
2916 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2917 
2918 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2919 
2920 	off = (u_quad_t)io.uio_offset;
2921 	getret = VOP_GETATTR(vp, &at, cred, lwp);
2922 
2923 	VOP_UNLOCK(vp, 0);
2924 
2925 	/*
2926 	 * If the VGET operation doesn't work for this filesystem,
2927 	 * we can't support readdirplus. Returning NOTSUPP should
2928 	 * make clients fall back to plain readdir.
2929 	 * There's no need to check for VPTOFH as well, we wouldn't
2930 	 * even be here otherwise.
2931 	 */
2932 	if (!getret) {
2933 		if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp)))
2934 			getret = (getret == EOPNOTSUPP) ?
2935 				NFSERR_NOTSUPP : NFSERR_IO;
2936 		else
2937 			vput(nvp);
2938 	}
2939 
2940 	if (!cookies && !error)
2941 		error = NFSERR_PERM;
2942 	if (!error)
2943 		error = getret;
2944 	if (error) {
2945 		vrele(vp);
2946 		if (cookies)
2947 			free((caddr_t)cookies, M_TEMP);
2948 		free((caddr_t)rbuf, M_TEMP);
2949 		nfsm_reply(NFSX_V3POSTOPATTR);
2950 		nfsm_srvpostop_attr(getret, &at);
2951 		return (0);
2952 	}
2953 	if (io.uio_resid) {
2954 		siz -= io.uio_resid;
2955 
2956 		/*
2957 		 * If nothing read, return eof
2958 		 * rpc reply
2959 		 */
2960 		if (siz == 0) {
2961 			vrele(vp);
2962 			nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
2963 				2 * NFSX_UNSIGNED);
2964 			nfsm_srvpostop_attr(getret, &at);
2965 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2966 			txdr_hyper(at.va_filerev, tl);
2967 			tl += 2;
2968 			*tl++ = nfs_false;
2969 			*tl = nfs_true;
2970 			free((caddr_t)cookies, M_TEMP);
2971 			free((caddr_t)rbuf, M_TEMP);
2972 			return (0);
2973 		}
2974 	}
2975 
2976 	/*
2977 	 * Check for degenerate cases of nothing useful read.
2978 	 * If so go try again
2979 	 */
2980 	cpos = rbuf;
2981 	cend = rbuf + siz;
2982 	dp = (struct dirent *)cpos;
2983 	cookiep = cookies;
2984 
2985 	while (cpos < cend && ncookies > 0 &&
2986 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2987 		cpos += dp->d_reclen;
2988 		dp = (struct dirent *)cpos;
2989 		cookiep++;
2990 		ncookies--;
2991 	}
2992 	if (cpos >= cend || ncookies == 0) {
2993 		toff = off;
2994 		siz = fullsiz;
2995 		free(cookies, M_TEMP);
2996 		goto again;
2997 	}
2998 
2999 	dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
3000 	nfsm_reply(cnt);
3001 	nfsm_srvpostop_attr(getret, &at);
3002 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3003 	txdr_hyper(at.va_filerev, tl);
3004 	mp = mp2 = mb;
3005 	bp = bpos;
3006 	be = bp + M_TRAILINGSPACE(mp);
3007 
3008 	/* Loop through the records and build reply */
3009 	while (cpos < cend && ncookies > 0) {
3010 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3011 			nfsrvfh_t nnsfh;
3012 
3013 			nlen = dp->d_namlen;
3014 			rem = nfsm_rndup(nlen)-nlen;
3015 
3016 			/*
3017 			 * For readdir_and_lookup get the vnode using
3018 			 * the file number.
3019 			 */
3020 			if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp))
3021 				goto invalid;
3022 			if (nfsrv_composefh(nvp, &nnsfh, TRUE)) {
3023 				vput(nvp);
3024 				goto invalid;
3025 			}
3026 			if (VOP_GETATTR(nvp, vap, cred, lwp)) {
3027 				vput(nvp);
3028 				goto invalid;
3029 			}
3030 			vput(nvp);
3031 
3032 			/*
3033 			 * If either the dircount or maxcount will be
3034 			 * exceeded, get out now. Both of these lengths
3035 			 * are calculated conservatively, including all
3036 			 * XDR overheads.
3037 			 */
3038 			len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3039 				NFSX_V3POSTOPATTR);
3040 			dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3041 			if (len > cnt || dirlen > fullsiz) {
3042 				eofflag = 0;
3043 				break;
3044 			}
3045 
3046 			/*
3047 			 * Build the directory record xdr from
3048 			 * the dirent entry.
3049 			 */
3050 			fp = (struct nfs_fattr *)&fl.fl_fattr;
3051 			nfsm_srvfillattr(vap, fp);
3052 			fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3053 			fl.fl_fhok = nfs_true;
3054 			fl.fl_postopok = nfs_true;
3055 			txdr_hyper(*cookiep, fl.fl_off.nfsuquad);
3056 
3057 			nfsm_clget;
3058 			*tl = nfs_true;
3059 			bp += NFSX_UNSIGNED;
3060 			nfsm_clget;
3061 			*tl = txdr_unsigned(dp->d_fileno >> 32);
3062 			bp += NFSX_UNSIGNED;
3063 			nfsm_clget;
3064 			*tl = txdr_unsigned(dp->d_fileno);
3065 			bp += NFSX_UNSIGNED;
3066 			nfsm_clget;
3067 			*tl = txdr_unsigned(nlen);
3068 			bp += NFSX_UNSIGNED;
3069 
3070 			/* And loop around copying the name */
3071 			xfer = nlen;
3072 			cp = dp->d_name;
3073 			while (xfer > 0) {
3074 				nfsm_clget;
3075 				if ((bp + xfer) > be)
3076 					tsiz = be - bp;
3077 				else
3078 					tsiz = xfer;
3079 				memcpy(bp, cp, tsiz);
3080 				bp += tsiz;
3081 				xfer -= tsiz;
3082 				if (xfer > 0)
3083 					cp += tsiz;
3084 			}
3085 			/* And null pad to an int32_t boundary */
3086 			for (i = 0; i < rem; i++)
3087 				*bp++ = '\0';
3088 
3089 			/*
3090 			 * Now copy the flrep structure out.
3091 			 */
3092 			xfer = sizeof(struct flrep);
3093 			cp = (caddr_t)&fl;
3094 			while (xfer > 0) {
3095 				nfsm_clget;
3096 				if ((bp + xfer) > be)
3097 					tsiz = be - bp;
3098 				else
3099 					tsiz = xfer;
3100 				memcpy(bp, cp, tsiz);
3101 				bp += tsiz;
3102 				xfer -= tsiz;
3103 				if (xfer > 0)
3104 					cp += tsiz;
3105 			}
3106 
3107 			/*
3108 			 * ... and filehandle.
3109 			 */
3110 			xfer = NFSRVFH_SIZE(&nnsfh);
3111 			cp = NFSRVFH_DATA(&nnsfh);
3112 			while (xfer > 0) {
3113 				nfsm_clget;
3114 				if ((bp + xfer) > be)
3115 					tsiz = be - bp;
3116 				else
3117 					tsiz = xfer;
3118 				memcpy(bp, cp, tsiz);
3119 				bp += tsiz;
3120 				xfer -= tsiz;
3121 				if (xfer > 0)
3122 					cp += tsiz;
3123 			}
3124 		}
3125 invalid:
3126 		cpos += dp->d_reclen;
3127 		dp = (struct dirent *)cpos;
3128 		cookiep++;
3129 		ncookies--;
3130 	}
3131 	vrele(vp);
3132 	nfsm_clget;
3133 	*tl = nfs_false;
3134 	bp += NFSX_UNSIGNED;
3135 	nfsm_clget;
3136 	if (eofflag)
3137 		*tl = nfs_true;
3138 	else
3139 		*tl = nfs_false;
3140 	bp += NFSX_UNSIGNED;
3141 	if (mp != mb) {
3142 		if (bp < be)
3143 			mp->m_len = bp - mtod(mp, caddr_t);
3144 	} else
3145 		mp->m_len += bp - bpos;
3146 	free((caddr_t)cookies, M_TEMP);
3147 	free((caddr_t)rbuf, M_TEMP);
3148 	nfsm_srvdone;
3149 }
3150 
3151 /*
3152  * nfs commit service
3153  */
3154 int
3155 nfsrv_commit(nfsd, slp, lwp, mrq)
3156 	struct nfsrv_descript *nfsd;
3157 	struct nfssvc_sock *slp;
3158 	struct lwp *lwp;
3159 	struct mbuf **mrq;
3160 {
3161 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3162 	struct mbuf *nam = nfsd->nd_nam;
3163 	caddr_t dpos = nfsd->nd_dpos;
3164 	kauth_cred_t cred = nfsd->nd_cr;
3165 	struct vattr bfor, aft;
3166 	struct vnode *vp;
3167 	nfsrvfh_t nsfh;
3168 	u_int32_t *tl;
3169 	int32_t t1;
3170 	caddr_t bpos;
3171 	int error = 0, rdonly, for_ret = 1, aft_ret = 1, cache = 0;
3172 	uint32_t cnt;
3173 	char *cp2;
3174 	struct mbuf *mb, *mreq;
3175 	u_quad_t frev, off, end;
3176 	struct mount *mp = NULL;
3177 
3178 	nfsm_srvmtofh(&nsfh);
3179 	if ((mp = vfs_getvfs(&NFSRVFH_FHANDLE(&nsfh)->fh_fsid)) == NULL)
3180 		return (ESTALE);
3181 	vn_start_write(NULL, &mp, V_WAIT);
3182 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3183 
3184 	off = fxdr_hyper(tl);
3185 	tl += 2;
3186 	cnt = fxdr_unsigned(uint32_t, *tl);
3187 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3188 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3189 	if (error) {
3190 		nfsm_reply(2 * NFSX_UNSIGNED);
3191 		nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3192 		vn_finished_write(mp, 0);
3193 		return (0);
3194 	}
3195 	for_ret = VOP_GETATTR(vp, &bfor, cred, lwp);
3196 	end = (cnt > 0) ? off + cnt : vp->v_size;
3197 	if (end < off || end > vp->v_size)
3198 		end = vp->v_size;
3199 	if (off < vp->v_size)
3200 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end, lwp);
3201 	/* else error == 0, from nfsrv_fhtovp() */
3202 	aft_ret = VOP_GETATTR(vp, &aft, cred, lwp);
3203 	vput(vp);
3204 	nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3205 	nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3206 	if (!error) {
3207 		nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3208 		*tl++ = txdr_unsigned(boottime.tv_sec);
3209 		*tl = txdr_unsigned(boottime.tv_usec);
3210 	} else {
3211 		vn_finished_write(mp, 0);
3212 		return (0);
3213 	}
3214 	vn_finished_write(mp, 0);
3215 	nfsm_srvdone;
3216 }
3217 
3218 /*
3219  * nfs statfs service
3220  */
3221 int
3222 nfsrv_statfs(nfsd, slp, lwp, mrq)
3223 	struct nfsrv_descript *nfsd;
3224 	struct nfssvc_sock *slp;
3225 	struct lwp *lwp;
3226 	struct mbuf **mrq;
3227 {
3228 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3229 	struct mbuf *nam = nfsd->nd_nam;
3230 	caddr_t dpos = nfsd->nd_dpos;
3231 	kauth_cred_t cred = nfsd->nd_cr;
3232 	struct statvfs *sf = NULL;
3233 	struct nfs_statfs *sfp;
3234 	u_int32_t *tl;
3235 	int32_t t1;
3236 	caddr_t bpos;
3237 	int error = 0, rdonly, cache = 0, getret = 1;
3238 	int v3 = (nfsd->nd_flag & ND_NFSV3);
3239 	char *cp2;
3240 	struct mbuf *mb, *mreq;
3241 	struct vnode *vp;
3242 	struct vattr at;
3243 	nfsrvfh_t nsfh;
3244 	u_quad_t frev, tval;
3245 
3246 	nfsm_srvmtofh(&nsfh);
3247 	error = nfsrv_fhtovp(&nsfh, 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 	sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK);
3255 	error = VFS_STATVFS(vp->v_mount, sf, lwp);
3256 	getret = VOP_GETATTR(vp, &at, cred, lwp);
3257 	vput(vp);
3258 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3259 	if (v3)
3260 		nfsm_srvpostop_attr(getret, &at);
3261 	if (error) {
3262 		free(sf, M_TEMP);
3263 		return (0);
3264 	}
3265 	nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3266 	if (v3) {
3267 		tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_frsize);
3268 		txdr_hyper(tval, &sfp->sf_tbytes);
3269 		tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_frsize);
3270 		txdr_hyper(tval, &sfp->sf_fbytes);
3271 		tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_frsize);
3272 		txdr_hyper(tval, &sfp->sf_abytes);
3273 		tval = (u_quad_t)sf->f_files;
3274 		txdr_hyper(tval, &sfp->sf_tfiles);
3275 		tval = (u_quad_t)sf->f_ffree;
3276 		txdr_hyper(tval, &sfp->sf_ffiles);
3277 		txdr_hyper(tval, &sfp->sf_afiles);
3278 		sfp->sf_invarsec = 0;
3279 	} else {
3280 		sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3281 		sfp->sf_bsize = txdr_unsigned(sf->f_frsize);
3282 		sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3283 		sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3284 		sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3285 	}
3286 nfsmout:
3287 	if (sf)
3288 	    free(sf, M_TEMP);
3289 	return error;
3290 }
3291 
3292 /*
3293  * nfs fsinfo service
3294  */
3295 int
3296 nfsrv_fsinfo(nfsd, slp, lwp, mrq)
3297 	struct nfsrv_descript *nfsd;
3298 	struct nfssvc_sock *slp;
3299 	struct lwp *lwp;
3300 	struct mbuf **mrq;
3301 {
3302 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3303 	struct mbuf *nam = nfsd->nd_nam;
3304 	caddr_t dpos = nfsd->nd_dpos;
3305 	kauth_cred_t cred = nfsd->nd_cr;
3306 	u_int32_t *tl;
3307 	struct nfsv3_fsinfo *sip;
3308 	int32_t t1;
3309 	caddr_t bpos;
3310 	int error = 0, rdonly, cache = 0, getret = 1;
3311 	uint32_t maxdata;
3312 	char *cp2;
3313 	struct mbuf *mb, *mreq;
3314 	struct vnode *vp;
3315 	struct vattr at;
3316 	nfsrvfh_t nsfh;
3317 	u_quad_t frev, maxfsize;
3318 	struct statvfs *sb;
3319 
3320 	nfsm_srvmtofh(&nsfh);
3321 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3322 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3323 	if (error) {
3324 		nfsm_reply(NFSX_UNSIGNED);
3325 		nfsm_srvpostop_attr(getret, &at);
3326 		return (0);
3327 	}
3328 
3329 	/* XXX Try to make a guess on the max file size. */
3330 	sb = malloc(sizeof(*sb), M_TEMP, M_WAITOK);
3331 	VFS_STATVFS(vp->v_mount, sb, (struct lwp *)0);
3332 	maxfsize = (u_quad_t)0x80000000 * sb->f_frsize - 1;
3333 	free(sb, M_TEMP);
3334 
3335 	getret = VOP_GETATTR(vp, &at, cred, lwp);
3336 	vput(vp);
3337 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3338 	nfsm_srvpostop_attr(getret, &at);
3339 	nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3340 
3341 	/*
3342 	 * XXX
3343 	 * There should be file system VFS OP(s) to get this information.
3344 	 * For now, assume ufs.
3345 	 */
3346 	if (slp->ns_so->so_type == SOCK_DGRAM)
3347 		maxdata = NFS_MAXDGRAMDATA;
3348 	else
3349 		maxdata = NFS_MAXDATA;
3350 	sip->fs_rtmax = txdr_unsigned(maxdata);
3351 	sip->fs_rtpref = txdr_unsigned(maxdata);
3352 	sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3353 	sip->fs_wtmax = txdr_unsigned(maxdata);
3354 	sip->fs_wtpref = txdr_unsigned(maxdata);
3355 	sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3356 	sip->fs_dtpref = txdr_unsigned(maxdata);
3357 	txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3358 	sip->fs_timedelta.nfsv3_sec = 0;
3359 	sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3360 	sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3361 		NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3362 		NFSV3FSINFO_CANSETTIME);
3363 	nfsm_srvdone;
3364 }
3365 
3366 /*
3367  * nfs pathconf service
3368  */
3369 int
3370 nfsrv_pathconf(nfsd, slp, lwp, mrq)
3371 	struct nfsrv_descript *nfsd;
3372 	struct nfssvc_sock *slp;
3373 	struct lwp *lwp;
3374 	struct mbuf **mrq;
3375 {
3376 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3377 	struct mbuf *nam = nfsd->nd_nam;
3378 	caddr_t dpos = nfsd->nd_dpos;
3379 	kauth_cred_t cred = nfsd->nd_cr;
3380 	u_int32_t *tl;
3381 	struct nfsv3_pathconf *pc;
3382 	int32_t t1;
3383 	caddr_t bpos;
3384 	int error = 0, rdonly, cache = 0, getret = 1;
3385 	register_t linkmax, namemax, chownres, notrunc;
3386 	char *cp2;
3387 	struct mbuf *mb, *mreq;
3388 	struct vnode *vp;
3389 	struct vattr at;
3390 	nfsrvfh_t nsfh;
3391 	u_quad_t frev;
3392 
3393 	nfsm_srvmtofh(&nsfh);
3394 	error = nfsrv_fhtovp(&nsfh, 1, &vp, cred, slp, nam,
3395 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3396 	if (error) {
3397 		nfsm_reply(NFSX_UNSIGNED);
3398 		nfsm_srvpostop_attr(getret, &at);
3399 		return (0);
3400 	}
3401 	error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3402 	if (!error)
3403 		error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3404 	if (!error)
3405 		error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3406 	if (!error)
3407 		error = VOP_PATHCONF(vp, _PC_NO_TRUNC, &notrunc);
3408 	getret = VOP_GETATTR(vp, &at, cred, lwp);
3409 	vput(vp);
3410 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3411 	nfsm_srvpostop_attr(getret, &at);
3412 	if (error)
3413 		return (0);
3414 	nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3415 
3416 	pc->pc_linkmax = txdr_unsigned(linkmax);
3417 	pc->pc_namemax = txdr_unsigned(namemax);
3418 	pc->pc_notrunc = txdr_unsigned(notrunc);
3419 	pc->pc_chownrestricted = txdr_unsigned(chownres);
3420 
3421 	/*
3422 	 * These should probably be supported by VOP_PATHCONF(), but
3423 	 * until msdosfs is exportable (why would you want to?), the
3424 	 * Unix defaults should be ok.
3425 	 */
3426 	pc->pc_caseinsensitive = nfs_false;
3427 	pc->pc_casepreserving = nfs_true;
3428 	nfsm_srvdone;
3429 }
3430 
3431 /*
3432  * Null operation, used by clients to ping server
3433  */
3434 /* ARGSUSED */
3435 int
3436 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3437     struct lwp *lwp, struct mbuf **mrq)
3438 {
3439 	struct mbuf *mrep = nfsd->nd_mrep;
3440 	caddr_t bpos;
3441 	int error = NFSERR_RETVOID, cache = 0;
3442 	struct mbuf *mb, *mreq;
3443 	u_quad_t frev;
3444 
3445 	nfsm_reply(0);
3446 	return (0);
3447 }
3448 
3449 /*
3450  * No operation, used for obsolete procedures
3451  */
3452 /* ARGSUSED */
3453 int
3454 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3455     struct lwp *lwp, struct mbuf **mrq)
3456 {
3457 	struct mbuf *mrep = nfsd->nd_mrep;
3458 	caddr_t bpos;
3459 	int error, cache = 0;
3460 	struct mbuf *mb, *mreq;
3461 	u_quad_t frev;
3462 
3463 	if (nfsd->nd_repstat)
3464 		error = nfsd->nd_repstat;
3465 	else
3466 		error = EPROCUNAVAIL;
3467 	nfsm_reply(0);
3468 	return (0);
3469 }
3470 
3471 /*
3472  * Perform access checking for vnodes obtained from file handles that would
3473  * refer to files already opened by a Unix client. You cannot just use
3474  * vn_writechk() and VOP_ACCESS() for two reasons.
3475  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3476  * 2 - The owner is to be given access irrespective of mode bits for some
3477  *     operations, so that processes that chmod after opening a file don't
3478  *     break. I don't like this because it opens a security hole, but since
3479  *     the nfs server opens a security hole the size of a barn door anyhow,
3480  *     what the heck.
3481  *
3482  * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3483  * will return EPERM instead of EACCESS. EPERM is always an error.
3484  */
3485 int
3486 nfsrv_access(vp, flags, cred, rdonly, lwp, override)
3487 	struct vnode *vp;
3488 	int flags;
3489 	kauth_cred_t cred;
3490 	int rdonly;
3491 	struct lwp *lwp;
3492 	int override;
3493 {
3494 	struct vattr vattr;
3495 	int error;
3496 	if (flags & VWRITE) {
3497 		/* Just vn_writechk() changed to check rdonly */
3498 		/*
3499 		 * Disallow write attempts on read-only file systems;
3500 		 * unless the file is a socket or a block or character
3501 		 * device resident on the file system.
3502 		 */
3503 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3504 			switch (vp->v_type) {
3505 			case VREG:
3506 			case VDIR:
3507 			case VLNK:
3508 				return (EROFS);
3509 			default:
3510 				break;
3511 			}
3512 		}
3513 
3514 		/*
3515 		 * If the vnode is in use as a process's text,
3516 		 * we can't allow writing.
3517 		 */
3518 		if (vp->v_flag & VTEXT)
3519 			return (ETXTBSY);
3520 	}
3521 	error = VOP_GETATTR(vp, &vattr, cred, lwp);
3522 	if (error)
3523 		return (error);
3524 	error = VOP_ACCESS(vp, flags, cred, lwp);
3525 	/*
3526 	 * Allow certain operations for the owner (reads and writes
3527 	 * on files that are already open).
3528 	 */
3529 	if (override && error == EACCES && kauth_cred_geteuid(cred) == vattr.va_uid)
3530 		error = 0;
3531 	return error;
3532 }
3533