xref: /netbsd-src/sys/nfs/nfs_serv.c (revision da5f4674a3fc214be3572d358b66af40ab9401e7)
1 /*	$NetBSD: nfs_serv.c,v 1.81 2003/08/07 16:33:51 agc 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 1.81 2003/08/07 16:33:51 agc 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 <ufs/ufs/dir.h>
74 
75 #include <uvm/uvm.h>
76 
77 #include <nfs/nfsproto.h>
78 #include <nfs/rpcv2.h>
79 #include <nfs/nfs.h>
80 #include <nfs/xdr_subs.h>
81 #include <nfs/nfsm_subs.h>
82 #include <nfs/nqnfs.h>
83 #include <nfs/nfs_var.h>
84 
85 /* Global vars */
86 extern u_int32_t nfs_xdrneg1;
87 extern u_int32_t nfs_false, nfs_true;
88 extern enum vtype nv3tov_type[8];
89 extern struct nfsstats nfsstats;
90 extern const nfstype nfsv2_type[9];
91 extern const nfstype nfsv3_type[9];
92 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
93 int nfsd_use_loan = 1;	/* use page-loan for READ OP */
94 
95 /*
96  * nfs v3 access service
97  */
98 int
99 nfsrv3_access(nfsd, slp, procp, mrq)
100 	struct nfsrv_descript *nfsd;
101 	struct nfssvc_sock *slp;
102 	struct proc *procp;
103 	struct mbuf **mrq;
104 {
105 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
106 	struct mbuf *nam = nfsd->nd_nam;
107 	caddr_t dpos = nfsd->nd_dpos;
108 	struct ucred *cred = &nfsd->nd_cr;
109 	struct vnode *vp;
110 	nfsfh_t nfh;
111 	fhandle_t *fhp;
112 	u_int32_t *tl;
113 	int32_t t1;
114 	caddr_t bpos;
115 	int error = 0, rdonly, cache = 0, getret;
116 	char *cp2;
117 	struct mbuf *mb, *mreq;
118 	struct vattr va;
119 	u_long inmode, testmode, outmode;
120 	u_quad_t frev;
121 
122 	fhp = &nfh.fh_generic;
123 	nfsm_srvmtofh(fhp);
124 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
125 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
126 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
127 	if (error) {
128 		nfsm_reply(NFSX_UNSIGNED);
129 		nfsm_srvpostop_attr(1, (struct vattr *)0);
130 		return (0);
131 	}
132 	inmode = fxdr_unsigned(u_int32_t, *tl);
133 	outmode = 0;
134 	if ((inmode & NFSV3ACCESS_READ) &&
135 	    nfsrv_access(vp, VREAD, cred, rdonly, procp, 0) == 0)
136 		outmode |= NFSV3ACCESS_READ;
137 	if (vp->v_type != VDIR) {
138 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
139 		if (testmode &&
140 		    nfsrv_access(vp, VWRITE, cred, rdonly, procp, 0) == 0)
141 			outmode |= testmode;
142 		if ((inmode & NFSV3ACCESS_EXECUTE) &&
143 		    nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0) == 0)
144 			outmode |= NFSV3ACCESS_EXECUTE;
145 	} else {
146 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
147 		    NFSV3ACCESS_DELETE);
148 		if (testmode &&
149 		    nfsrv_access(vp, VWRITE, cred, rdonly, procp, 0) == 0)
150 			outmode |= testmode;
151 		if ((inmode & NFSV3ACCESS_LOOKUP) &&
152 		    nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0) == 0)
153 			outmode |= NFSV3ACCESS_LOOKUP;
154 	}
155 	getret = VOP_GETATTR(vp, &va, cred, procp);
156 	vput(vp);
157 	nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
158 	nfsm_srvpostop_attr(getret, &va);
159 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
160 	*tl = txdr_unsigned(outmode);
161 	nfsm_srvdone;
162 }
163 
164 /*
165  * nfs getattr service
166  */
167 int
168 nfsrv_getattr(nfsd, slp, procp, mrq)
169 	struct nfsrv_descript *nfsd;
170 	struct nfssvc_sock *slp;
171 	struct proc *procp;
172 	struct mbuf **mrq;
173 {
174 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
175 	struct mbuf *nam = nfsd->nd_nam;
176 	caddr_t dpos = nfsd->nd_dpos;
177 	struct ucred *cred = &nfsd->nd_cr;
178 	struct nfs_fattr *fp;
179 	struct vattr va;
180 	struct vnode *vp;
181 	nfsfh_t nfh;
182 	fhandle_t *fhp;
183 	u_int32_t *tl;
184 	int32_t t1;
185 	caddr_t bpos;
186 	int error = 0, rdonly, cache;
187 	char *cp2;
188 	struct mbuf *mb, *mreq;
189 	u_quad_t frev;
190 
191 	fhp = &nfh.fh_generic;
192 	nfsm_srvmtofh(fhp);
193 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
194 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
195 	if (error) {
196 		nfsm_reply(0);
197 		return (0);
198 	}
199 	nqsrv_getl(vp, ND_READ);
200 	error = VOP_GETATTR(vp, &va, cred, procp);
201 	vput(vp);
202 	nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
203 	if (error)
204 		return (0);
205 	nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
206 	nfsm_srvfillattr(&va, fp);
207 	nfsm_srvdone;
208 }
209 
210 /*
211  * nfs setattr service
212  */
213 int
214 nfsrv_setattr(nfsd, slp, procp, mrq)
215 	struct nfsrv_descript *nfsd;
216 	struct nfssvc_sock *slp;
217 	struct proc *procp;
218 	struct mbuf **mrq;
219 {
220 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
221 	struct mbuf *nam = nfsd->nd_nam;
222 	caddr_t dpos = nfsd->nd_dpos;
223 	struct ucred *cred = &nfsd->nd_cr;
224 	struct vattr va, preat;
225 	struct nfsv2_sattr *sp;
226 	struct nfs_fattr *fp;
227 	struct vnode *vp;
228 	nfsfh_t nfh;
229 	fhandle_t *fhp;
230 	u_int32_t *tl;
231 	int32_t t1;
232 	caddr_t bpos;
233 	int error = 0, rdonly, cache, preat_ret = 1, postat_ret = 1;
234 	int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
235 	char *cp2;
236 	struct mbuf *mb, *mreq;
237 	u_quad_t frev;
238 	struct timespec guard;
239 
240 	fhp = &nfh.fh_generic;
241 	nfsm_srvmtofh(fhp);
242 	VATTR_NULL(&va);
243 	if (v3) {
244 		nfsm_srvsattr(&va);
245 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
246 		gcheck = fxdr_unsigned(int, *tl);
247 		if (gcheck) {
248 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
249 			fxdr_nfsv3time(tl, &guard);
250 		}
251 	} else {
252 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
253 		/*
254 		 * Nah nah nah nah na nah
255 		 * There is a bug in the Sun client that puts 0xffff in the mode
256 		 * field of sattr when it should put in 0xffffffff. The u_short
257 		 * doesn't sign extend.
258 		 * --> check the low order 2 bytes for 0xffff
259 		 */
260 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
261 			va.va_mode = nfstov_mode(sp->sa_mode);
262 		if (sp->sa_uid != nfs_xdrneg1)
263 			va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
264 		if (sp->sa_gid != nfs_xdrneg1)
265 			va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
266 		if (sp->sa_size != nfs_xdrneg1)
267 			va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
268 		if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) {
269 #ifdef notyet
270 			fxdr_nfsv2time(&sp->sa_atime, &va.va_atime);
271 #else
272 			va.va_atime.tv_sec =
273 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
274 			va.va_atime.tv_nsec = 0;
275 #endif
276 		}
277 		if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1)
278 			fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime);
279 
280 	}
281 
282 	/*
283 	 * Now that we have all the fields, lets do it.
284 	 */
285 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
286 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
287 	if (error) {
288 		nfsm_reply(2 * NFSX_UNSIGNED);
289 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
290 		return (0);
291 	}
292 	nqsrv_getl(vp, ND_WRITE);
293 	if (v3) {
294 		error = preat_ret = VOP_GETATTR(vp, &preat, cred, procp);
295 		if (!error && gcheck &&
296 			(preat.va_ctime.tv_sec != guard.tv_sec ||
297 			 preat.va_ctime.tv_nsec != guard.tv_nsec))
298 			error = NFSERR_NOT_SYNC;
299 		if (error) {
300 			vput(vp);
301 			nfsm_reply(NFSX_WCCDATA(v3));
302 			nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
303 			return (0);
304 		}
305 	}
306 
307 	/*
308 	 * If the size is being changed write acces is required, otherwise
309 	 * just check for a read only file system.
310 	 */
311 	if (va.va_size == ((u_quad_t)((quad_t) -1))) {
312 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
313 			error = EROFS;
314 			goto out;
315 		}
316 	} else {
317 		if (vp->v_type == VDIR) {
318 			error = EISDIR;
319 			goto out;
320 		} else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
321 			procp, 0)) != 0)
322 			goto out;
323 	}
324 	error = VOP_SETATTR(vp, &va, cred, procp);
325 	postat_ret = VOP_GETATTR(vp, &va, cred, procp);
326 	if (!error)
327 		error = postat_ret;
328 out:
329 	vput(vp);
330 	nfsm_reply(NFSX_WCCORFATTR(v3));
331 	if (v3) {
332 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
333 		return (0);
334 	} else {
335 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
336 		nfsm_srvfillattr(&va, fp);
337 	}
338 	nfsm_srvdone;
339 }
340 
341 /*
342  * nfs lookup rpc
343  */
344 int
345 nfsrv_lookup(nfsd, slp, procp, mrq)
346 	struct nfsrv_descript *nfsd;
347 	struct nfssvc_sock *slp;
348 	struct proc *procp;
349 	struct mbuf **mrq;
350 {
351 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
352 	struct mbuf *nam = nfsd->nd_nam;
353 	caddr_t dpos = nfsd->nd_dpos;
354 	struct ucred *cred = &nfsd->nd_cr;
355 	struct nfs_fattr *fp;
356 	struct nameidata nd, ind, *ndp = &nd;
357 	struct vnode *vp, *dirp;
358 	nfsfh_t nfh;
359 	fhandle_t *fhp;
360 	caddr_t cp;
361 	u_int32_t *tl;
362 	int32_t t1;
363 	caddr_t bpos;
364 	int error = 0, cache, dirattr_ret = 1;
365 	uint32_t len;
366 	int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
367 	char *cp2;
368 	struct mbuf *mb, *mreq;
369 	struct vattr va, dirattr;
370 	u_quad_t frev;
371 
372 	fhp = &nfh.fh_generic;
373 	nfsm_srvmtofh(fhp);
374 	nfsm_srvnamesiz(len);
375 
376 	pubflag = nfs_ispublicfh(fhp);
377 
378 	nd.ni_cnd.cn_cred = cred;
379 	nd.ni_cnd.cn_nameiop = LOOKUP;
380 	nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART;
381 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
382 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
383 
384 	if (!error && pubflag) {
385 		if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
386 			/*
387 			 * Setup call to lookup() to see if we can find
388 			 * the index file. Arguably, this doesn't belong
389 			 * in a kernel.. Ugh.
390 			 */
391 			ind = nd;
392 			VOP_UNLOCK(nd.ni_vp, 0);
393 			ind.ni_pathlen = strlen(nfs_pub.np_index);
394 			ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
395 			    nfs_pub.np_index;
396 			ind.ni_startdir = nd.ni_vp;
397 			VREF(ind.ni_startdir);
398 			error = lookup(&ind);
399 			if (!error) {
400 				/*
401 				 * Found an index file. Get rid of
402 				 * the old references.
403 				 */
404 				if (dirp)
405 					vrele(dirp);
406 				dirp = nd.ni_vp;
407 				vrele(nd.ni_startdir);
408 				ndp = &ind;
409 			} else
410 				error = 0;
411 		}
412 		/*
413 		 * If the public filehandle was used, check that this lookup
414 		 * didn't result in a filehandle outside the publicly exported
415 		 * filesystem.
416 		 */
417 
418 		if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) {
419 			vput(nd.ni_vp);
420 			error = EPERM;
421 		}
422 	}
423 
424 	if (dirp) {
425 		if (v3)
426 			dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred,
427 				procp);
428 		vrele(dirp);
429 	}
430 
431 	if (error) {
432 		nfsm_reply(NFSX_POSTOPATTR(v3));
433 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
434 		return (0);
435 	}
436 
437 	nqsrv_getl(ndp->ni_startdir, ND_READ);
438 	vrele(ndp->ni_startdir);
439 	PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
440 	vp = ndp->ni_vp;
441 	memset((caddr_t)fhp, 0, sizeof(nfh));
442 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
443 	error = VFS_VPTOFH(vp, &fhp->fh_fid);
444 	if (!error)
445 		error = VOP_GETATTR(vp, &va, cred, procp);
446 	vput(vp);
447 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
448 	if (error) {
449 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
450 		return (0);
451 	}
452 	nfsm_srvfhtom(fhp, 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, procp, mrq)
468 	struct nfsrv_descript *nfsd;
469 	struct nfssvc_sock *slp;
470 	struct proc *procp;
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 	struct ucred *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, 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 	nfsfh_t nfh;
491 	fhandle_t *fhp;
492 	struct uio io, *uiop = &io;
493 	u_quad_t frev;
494 
495 	fhp = &nfh.fh_generic;
496 	nfsm_srvmtofh(fhp);
497 	len = 0;
498 	i = 0;
499 	while (len < NFS_MAXPATHLEN) {
500 		mp = m_get(M_WAIT, MT_DATA);
501 		MCLAIM(mp, &nfs_mowner);
502 		m_clget(mp, M_WAIT);
503 		mp->m_len = NFSMSIZ(mp);
504 		if (len == 0)
505 			mp3 = mp2 = mp;
506 		else {
507 			mp2->m_next = mp;
508 			mp2 = mp;
509 		}
510 		if ((len+mp->m_len) > NFS_MAXPATHLEN) {
511 			mp->m_len = NFS_MAXPATHLEN-len;
512 			len = NFS_MAXPATHLEN;
513 		} else
514 			len += mp->m_len;
515 		ivp->iov_base = mtod(mp, caddr_t);
516 		ivp->iov_len = mp->m_len;
517 		i++;
518 		ivp++;
519 	}
520 	uiop->uio_iov = iv;
521 	uiop->uio_iovcnt = i;
522 	uiop->uio_offset = 0;
523 	uiop->uio_resid = len;
524 	uiop->uio_rw = UIO_READ;
525 	uiop->uio_segflg = UIO_SYSSPACE;
526 	uiop->uio_procp = (struct proc *)0;
527 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
528 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
529 	if (error) {
530 		m_freem(mp3);
531 		nfsm_reply(2 * NFSX_UNSIGNED);
532 		nfsm_srvpostop_attr(1, (struct vattr *)0);
533 		return (0);
534 	}
535 	if (vp->v_type != VLNK) {
536 		if (v3)
537 			error = EINVAL;
538 		else
539 			error = ENXIO;
540 		goto out;
541 	}
542 	nqsrv_getl(vp, ND_READ);
543 	error = VOP_READLINK(vp, uiop, cred);
544 out:
545 	getret = VOP_GETATTR(vp, &attr, cred, procp);
546 	vput(vp);
547 	if (error)
548 		m_freem(mp3);
549 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
550 	if (v3) {
551 		nfsm_srvpostop_attr(getret, &attr);
552 		if (error)
553 			return (0);
554 	}
555 	len -= uiop->uio_resid;
556 	padlen = nfsm_padlen(len);
557 	if (uiop->uio_resid || padlen)
558 		nfs_zeropad(mp3, uiop->uio_resid, padlen);
559 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
560 	*tl = txdr_unsigned(len);
561 	mb->m_next = mp3;
562 	nfsm_srvdone;
563 }
564 
565 /*
566  * XXX UBC temp limit
567  * maximum number of pages we can do VOP_GETPAGES and loan-out at once.
568  * should be <= MAX_READ_AHEAD in genfs_vnops.c
569  */
570 #define	NFSD_READ_GETPAGES_CHUNK	16
571 
572 /*
573  * nfs read service
574  */
575 int
576 nfsrv_read(nfsd, slp, procp, mrq)
577 	struct nfsrv_descript *nfsd;
578 	struct nfssvc_sock *slp;
579 	struct proc *procp;
580 	struct mbuf **mrq;
581 {
582 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
583 	struct mbuf *nam = nfsd->nd_nam;
584 	caddr_t dpos = nfsd->nd_dpos;
585 	struct ucred *cred = &nfsd->nd_cr;
586 	struct mbuf *m;
587 	struct nfs_fattr *fp;
588 	u_int32_t *tl;
589 	int32_t t1;
590 	int i;
591 	caddr_t bpos;
592 	int error = 0, rdonly, cache, getret;
593 	int v3 = (nfsd->nd_flag & ND_NFSV3);
594 	uint32_t reqlen, len, cnt, left;
595 	int padlen;
596 	char *cp2;
597 	struct mbuf *mb, *mreq;
598 	struct vnode *vp;
599 	nfsfh_t nfh;
600 	fhandle_t *fhp;
601 	struct uio io, *uiop = &io;
602 	struct vattr va;
603 	off_t off;
604 	u_quad_t frev;
605 
606 	fhp = &nfh.fh_generic;
607 	nfsm_srvmtofh(fhp);
608 	if (v3) {
609 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
610 		off = fxdr_hyper(tl);
611 	} else {
612 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
613 		off = (off_t)fxdr_unsigned(u_int32_t, *tl);
614 	}
615 	nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
616 	reqlen = fxdr_unsigned(uint32_t, *tl);
617 	reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd));
618 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
619 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
620 	if (error) {
621 		nfsm_reply(2 * NFSX_UNSIGNED);
622 		nfsm_srvpostop_attr(1, (struct vattr *)0);
623 		return (0);
624 	}
625 	if (vp->v_type != VREG) {
626 		if (v3)
627 			error = EINVAL;
628 		else
629 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
630 	}
631 	if (!error) {
632 	    nqsrv_getl(vp, ND_READ);
633 	    if ((error = nfsrv_access(vp, VREAD, cred, rdonly, procp, 1)) != 0)
634 		error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 1);
635 	}
636 	getret = VOP_GETATTR(vp, &va, cred, procp);
637 	if (!error)
638 		error = getret;
639 	if (error) {
640 		vput(vp);
641 		nfsm_reply(NFSX_POSTOPATTR(v3));
642 		nfsm_srvpostop_attr(getret, &va);
643 		return (0);
644 	}
645 	if (off >= va.va_size)
646 		cnt = 0;
647 	else if ((off + reqlen) > va.va_size)
648 		cnt = va.va_size - off;
649 	else
650 		cnt = reqlen;
651 	nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
652 	if (v3) {
653 		nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
654 		*tl++ = nfs_true;
655 		fp = (struct nfs_fattr *)tl;
656 		tl += (NFSX_V3FATTR / sizeof (u_int32_t));
657 	} else {
658 		nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
659 		fp = (struct nfs_fattr *)tl;
660 		tl += (NFSX_V2FATTR / sizeof (u_int32_t));
661 	}
662 	len = left = cnt;
663 	if (cnt > 0) {
664 		if (nfsd_use_loan) {
665 			struct vm_page **pgpp;
666 			voff_t pgoff = trunc_page(off);
667 			int orignpages, nleftpages;
668 			vaddr_t lva, curlva;
669 
670 			orignpages = (round_page(off + cnt) - pgoff)
671 			    >> PAGE_SHIFT;
672 			KASSERT(orignpages <= M_EXT_MAXPAGES); /* XXX */
673 
674 			lva = sokvaalloc(orignpages << PAGE_SHIFT, slp->ns_so);
675 			if (lva == 0) {
676 				/* fall back to VOP_READ */
677 				goto loan_fail;
678 			}
679 
680 			m = m_get(M_WAIT, MT_DATA);
681 			pgpp = m->m_ext.ext_pgs;
682 
683 			curlva = lva;
684 			nleftpages = orignpages;
685 			while (nleftpages > 0) {
686 				int npages = nleftpages;
687 				if (npages > NFSD_READ_GETPAGES_CHUNK)
688 					npages = NFSD_READ_GETPAGES_CHUNK;
689 again:
690 				simple_lock(&vp->v_interlock);
691 				error = VOP_GETPAGES(vp, pgoff, pgpp, &npages,
692 				    0, VM_PROT_READ, 0, PGO_SYNCIO);
693 				if (error == EAGAIN) {
694 					tsleep(&lbolt, PVM, "nfsread", 0);
695 					goto again;
696 				}
697 				if (error) {
698 					uvm_unloan(m->m_ext.ext_pgs,
699 					    orignpages - nleftpages,
700 					    UVM_LOAN_TOPAGE);
701 					sokvafree(lva,
702 					    orignpages << PAGE_SHIFT);
703 					m_free(m);
704 					goto read_error;
705 				}
706 
707 				/* loan and unbusy pages */
708 				simple_lock(&vp->v_interlock);
709 				for (i = 0; i < npages; i++) {
710 					if (pgpp[i]->flags & PG_RELEASED) {
711 						uvm_lock_pageq();
712 						uvm_page_unbusy(pgpp, npages);
713 						uvm_unlock_pageq();
714 						simple_unlock(&vp->v_interlock);
715 						continue;
716 					}
717 				}
718 				uvm_loanuobjpages(pgpp, npages);
719 				simple_unlock(&vp->v_interlock);
720 
721 				/* map pages */
722 				for (i = 0; i < npages; i++) {
723 					pmap_kenter_pa(curlva,
724 					    VM_PAGE_TO_PHYS(pgpp[i]),
725 					    VM_PROT_READ);
726 					curlva += PAGE_SIZE;
727 				}
728 				nleftpages -= npages;
729 				pgpp += npages;
730 				pgoff += npages << PAGE_SHIFT;
731 			}
732 
733 			lva += off & PAGE_MASK;
734 
735 			MCLAIM(m, &nfs_mowner);
736 			MEXTADD(m, (void *)lva, cnt, M_MBUF, soloanfree,
737 			    slp->ns_so);
738 			m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
739 			m->m_len = cnt;
740 
741 			pmap_update(pmap_kernel());
742 			mb->m_next = m;
743 			mb = m;
744 			error = 0;
745 			uiop->uio_resid = 0;
746 		} else {
747 			struct iovec *iv;
748 			struct iovec *iv2;
749 			struct mbuf *m2;
750 			int siz;
751 loan_fail:
752 			/*
753 			 * Generate the mbuf list with the uio_iov ref. to it.
754 			 */
755 			i = 0;
756 			m = m2 = mb;
757 			while (left > 0) {
758 				siz = min(M_TRAILINGSPACE(m), left);
759 				if (siz > 0) {
760 					left -= siz;
761 					i++;
762 				}
763 				if (left > 0) {
764 					m = m_get(M_WAIT, MT_DATA);
765 					MCLAIM(m, &nfs_mowner);
766 					m_clget(m, M_WAIT);
767 					m->m_len = 0;
768 					m2->m_next = m;
769 					m2 = m;
770 				}
771 			}
772 			iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK);
773 			uiop->uio_iov = iv2 = iv;
774 			m = mb;
775 			left = cnt;
776 			i = 0;
777 			while (left > 0) {
778 				if (m == NULL)
779 					panic("nfsrv_read iov");
780 				siz = min(M_TRAILINGSPACE(m), left);
781 				if (siz > 0) {
782 					iv->iov_base = mtod(m, caddr_t) +
783 					    m->m_len;
784 					iv->iov_len = siz;
785 					m->m_len += siz;
786 					left -= siz;
787 					iv++;
788 					i++;
789 				}
790 				m = m->m_next;
791 			}
792 			uiop->uio_iovcnt = i;
793 			uiop->uio_offset = off;
794 			uiop->uio_resid = cnt;
795 			uiop->uio_rw = UIO_READ;
796 			uiop->uio_segflg = UIO_SYSSPACE;
797 			error = VOP_READ(vp, uiop, IO_NODELOCKED, cred);
798 			off = uiop->uio_offset;
799 			free((caddr_t)iv2, M_TEMP);
800 		}
801 read_error:
802 		if (error || (getret = VOP_GETATTR(vp, &va, cred, procp)) != 0){
803 			if (!error)
804 				error = getret;
805 			m_freem(mreq);
806 			vput(vp);
807 			nfsm_reply(NFSX_POSTOPATTR(v3));
808 			nfsm_srvpostop_attr(getret, &va);
809 			return (0);
810 		}
811 	} else {
812 		uiop->uio_resid = 0;
813 	}
814 	vput(vp);
815 	nfsm_srvfillattr(&va, fp);
816 	len -= uiop->uio_resid;
817 	padlen = nfsm_padlen(len);
818 	if (uiop->uio_resid || padlen)
819 		nfs_zeropad(mb, uiop->uio_resid, padlen);
820 	if (v3) {
821 		*tl++ = txdr_unsigned(len);
822 		if (len < reqlen)
823 			*tl++ = nfs_true;
824 		else
825 			*tl++ = nfs_false;
826 	}
827 	*tl = txdr_unsigned(len);
828 	nfsm_srvdone;
829 }
830 
831 /*
832  * nfs write service
833  */
834 int
835 nfsrv_write(nfsd, slp, procp, mrq)
836 	struct nfsrv_descript *nfsd;
837 	struct nfssvc_sock *slp;
838 	struct proc *procp;
839 	struct mbuf **mrq;
840 {
841 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
842 	struct mbuf *nam = nfsd->nd_nam;
843 	caddr_t dpos = nfsd->nd_dpos;
844 	struct ucred *cred = &nfsd->nd_cr;
845 	struct iovec *ivp;
846 	int i, cnt;
847 	struct mbuf *mp;
848 	struct nfs_fattr *fp;
849 	struct iovec *iv;
850 	struct vattr va, forat;
851 	u_int32_t *tl;
852 	int32_t t1;
853 	caddr_t bpos;
854 	int error = 0, rdonly, cache, len, forat_ret = 1;
855 	int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
856 	int stable = NFSV3WRITE_FILESYNC;
857 	int v3 = (nfsd->nd_flag & ND_NFSV3);
858 	char *cp2;
859 	struct mbuf *mb, *mreq;
860 	struct vnode *vp;
861 	nfsfh_t nfh;
862 	fhandle_t *fhp;
863 	struct uio io, *uiop = &io;
864 	off_t off;
865 	u_quad_t frev;
866 
867 	if (mrep == NULL) {
868 		*mrq = NULL;
869 		return (0);
870 	}
871 	fhp = &nfh.fh_generic;
872 	nfsm_srvmtofh(fhp);
873 	if (v3) {
874 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
875 		off = fxdr_hyper(tl);
876 		tl += 3;
877 		stable = fxdr_unsigned(int, *tl++);
878 	} else {
879 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
880 		off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
881 		tl += 2;
882 	}
883 	retlen = len = fxdr_unsigned(int32_t, *tl);
884 	cnt = i = 0;
885 
886 	/*
887 	 * For NFS Version 2, it is not obvious what a write of zero length
888 	 * should do, but I might as well be consistent with Version 3,
889 	 * which is to return ok so long as there are no permission problems.
890 	 */
891 	if (len > 0) {
892 		zeroing = 1;
893 		mp = mrep;
894 		while (mp) {
895 			if (mp == md) {
896 				zeroing = 0;
897 				adjust = dpos - mtod(mp, caddr_t);
898 				mp->m_len -= adjust;
899 				if (mp->m_len > 0 && adjust > 0)
900 					NFSMADV(mp, adjust);
901 			}
902 			if (zeroing)
903 				mp->m_len = 0;
904 			else if (mp->m_len > 0) {
905 				i += mp->m_len;
906 				if (i > len) {
907 					mp->m_len -= (i - len);
908 					zeroing	= 1;
909 				}
910 				if (mp->m_len > 0)
911 					cnt++;
912 			}
913 			mp = mp->m_next;
914 		}
915 	}
916 	if (len > NFS_MAXDATA || len < 0 || i < len) {
917 		error = EIO;
918 		nfsm_reply(2 * NFSX_UNSIGNED);
919 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
920 		return (0);
921 	}
922 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
923 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
924 	if (error) {
925 		nfsm_reply(2 * NFSX_UNSIGNED);
926 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
927 		return (0);
928 	}
929 	if (v3)
930 		forat_ret = VOP_GETATTR(vp, &forat, cred, procp);
931 	if (vp->v_type != VREG) {
932 		if (v3)
933 			error = EINVAL;
934 		else
935 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
936 	}
937 	if (!error) {
938 		nqsrv_getl(vp, ND_WRITE);
939 		error = nfsrv_access(vp, VWRITE, cred, rdonly, procp, 1);
940 	}
941 	if (error) {
942 		vput(vp);
943 		nfsm_reply(NFSX_WCCDATA(v3));
944 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
945 		return (0);
946 	}
947 
948 	if (len > 0) {
949 		ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK);
950 		uiop->uio_iov = iv = ivp;
951 		uiop->uio_iovcnt = cnt;
952 		mp = mrep;
953 		while (mp) {
954 			if (mp->m_len > 0) {
955 				ivp->iov_base = mtod(mp, caddr_t);
956 				ivp->iov_len = mp->m_len;
957 				ivp++;
958 			}
959 			mp = mp->m_next;
960 		}
961 
962 		/*
963 		 * XXX
964 		 * The IO_METASYNC flag indicates that all metadata (and not
965 		 * just enough to ensure data integrity) must be written to
966 		 * stable storage synchronously.
967 		 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
968 		 */
969 		if (stable == NFSV3WRITE_UNSTABLE)
970 			ioflags = IO_NODELOCKED;
971 		else if (stable == NFSV3WRITE_DATASYNC)
972 			ioflags = (IO_SYNC | IO_NODELOCKED);
973 		else
974 			ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
975 		uiop->uio_resid = len;
976 		uiop->uio_rw = UIO_WRITE;
977 		uiop->uio_segflg = UIO_SYSSPACE;
978 		uiop->uio_procp = (struct proc *)0;
979 		uiop->uio_offset = off;
980 		error = VOP_WRITE(vp, uiop, ioflags, cred);
981 		nfsstats.srvvop_writes++;
982 		free(iv, M_TEMP);
983 	}
984 	aftat_ret = VOP_GETATTR(vp, &va, cred, procp);
985 	vput(vp);
986 	if (!error)
987 		error = aftat_ret;
988 	nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
989 		2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
990 	if (v3) {
991 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
992 		if (error)
993 			return (0);
994 		nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
995 		*tl++ = txdr_unsigned(retlen);
996 		if (stable == NFSV3WRITE_UNSTABLE)
997 			*tl++ = txdr_unsigned(stable);
998 		else
999 			*tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1000 		/*
1001 		 * Actually, there is no need to txdr these fields,
1002 		 * but it may make the values more human readable,
1003 		 * for debugging purposes.
1004 		 */
1005 		*tl++ = txdr_unsigned(boottime.tv_sec);
1006 		*tl = txdr_unsigned(boottime.tv_usec);
1007 	} else {
1008 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1009 		nfsm_srvfillattr(&va, fp);
1010 	}
1011 	nfsm_srvdone;
1012 }
1013 
1014 /*
1015  * NFS write service with write gathering support. Called when
1016  * nfsrvw_procrastinate > 0.
1017  * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1018  * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1019  * Jan. 1994.
1020  */
1021 int
1022 nfsrv_writegather(ndp, slp, procp, mrq)
1023 	struct nfsrv_descript **ndp;
1024 	struct nfssvc_sock *slp;
1025 	struct proc *procp;
1026 	struct mbuf **mrq;
1027 {
1028 	struct iovec *ivp;
1029 	struct mbuf *mp;
1030 	struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1031 	struct nfs_fattr *fp;
1032 	int i = 0;
1033 	struct iovec *iov;
1034 	struct nfsrvw_delayhash *wpp;
1035 	struct ucred *cred;
1036 	struct vattr va, forat;
1037 	u_int32_t *tl;
1038 	int32_t t1;
1039 	caddr_t bpos, dpos;
1040 	int error = 0, rdonly, cache, len = 0, forat_ret = 1;
1041 	int ioflags, aftat_ret = 1, s, adjust, v3, zeroing;
1042 	char *cp2;
1043 	struct mbuf *mb, *mreq, *mrep, *md;
1044 	struct vnode *vp;
1045 	struct uio io, *uiop = &io;
1046 	u_quad_t frev, cur_usec;
1047 
1048 	*mrq = NULL;
1049 	if (*ndp) {
1050 	    nfsd = *ndp;
1051 	    *ndp = NULL;
1052 	    mrep = nfsd->nd_mrep;
1053 	    md = nfsd->nd_md;
1054 	    dpos = nfsd->nd_dpos;
1055 	    cred = &nfsd->nd_cr;
1056 	    v3 = (nfsd->nd_flag & ND_NFSV3);
1057 	    LIST_INIT(&nfsd->nd_coalesce);
1058 	    nfsd->nd_mreq = NULL;
1059 	    nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1060 	    cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1061 	    nfsd->nd_time = cur_usec + nfsrvw_procrastinate;
1062 
1063 	    /*
1064 	     * Now, get the write header..
1065 	     */
1066 	    nfsm_srvmtofh(&nfsd->nd_fh);
1067 	    if (v3) {
1068 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1069 		nfsd->nd_off = fxdr_hyper(tl);
1070 		tl += 3;
1071 		nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1072 	    } else {
1073 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1074 		nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1075 		tl += 2;
1076 	    }
1077 	    len = fxdr_unsigned(int32_t, *tl);
1078 	    nfsd->nd_len = len;
1079 	    nfsd->nd_eoff = nfsd->nd_off + len;
1080 
1081 	    /*
1082 	     * Trim the header out of the mbuf list and trim off any trailing
1083 	     * junk so that the mbuf list has only the write data.
1084 	     */
1085 	    zeroing = 1;
1086 	    i = 0;
1087 	    mp = mrep;
1088 	    while (mp) {
1089 		if (mp == md) {
1090 		    zeroing = 0;
1091 		    adjust = dpos - mtod(mp, caddr_t);
1092 		    mp->m_len -= adjust;
1093 		    if (mp->m_len > 0 && adjust > 0)
1094 			NFSMADV(mp, adjust);
1095 		}
1096 		if (zeroing)
1097 		    mp->m_len = 0;
1098 		else {
1099 		    i += mp->m_len;
1100 		    if (i > len) {
1101 			mp->m_len -= (i - len);
1102 			zeroing = 1;
1103 		    }
1104 		}
1105 		mp = mp->m_next;
1106 	    }
1107 	    if (len > NFS_MAXDATA || len < 0  || i < len) {
1108 nfsmout:
1109 		m_freem(mrep);
1110 		error = EIO;
1111 		nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1112 		if (v3)
1113 		    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1114 		nfsd->nd_mreq = mreq;
1115 		nfsd->nd_mrep = NULL;
1116 		nfsd->nd_time = 0;
1117 	    }
1118 
1119 	    /*
1120 	     * Add this entry to the hash and time queues.
1121 	     */
1122 	    s = splsoftclock();
1123 	    owp = NULL;
1124 	    wp = LIST_FIRST(&slp->ns_tq);
1125 	    while (wp && wp->nd_time < nfsd->nd_time) {
1126 		owp = wp;
1127 		wp = LIST_NEXT(wp, nd_tq);
1128 	    }
1129 	    if (owp) {
1130 		LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1131 	    } else {
1132 		LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1133 	    }
1134 	    if (nfsd->nd_mrep) {
1135 		wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1136 		owp = NULL;
1137 		wp = LIST_FIRST(wpp);
1138 		while (wp &&
1139 		    memcmp((caddr_t)&nfsd->nd_fh, (caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1140 		    owp = wp;
1141 		    wp = LIST_NEXT(wp, nd_hash);
1142 		}
1143 		while (wp && wp->nd_off < nfsd->nd_off &&
1144 		    !memcmp((caddr_t)&nfsd->nd_fh, (caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1145 		    owp = wp;
1146 		    wp = LIST_NEXT(wp, nd_hash);
1147 		}
1148 		if (owp) {
1149 		    LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1150 
1151 		    /*
1152 		     * Search the hash list for overlapping entries and
1153 		     * coalesce.
1154 		     */
1155 		    for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1156 			wp = LIST_NEXT(nfsd, nd_hash);
1157 			if (NFSW_SAMECRED(owp, nfsd))
1158 			    nfsrvw_coalesce(owp, nfsd);
1159 		    }
1160 		} else {
1161 		    LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1162 		}
1163 	    }
1164 	    splx(s);
1165 	}
1166 
1167 	/*
1168 	 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1169 	 * and generate the associated reply mbuf list(s).
1170 	 */
1171 loop1:
1172 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1173 	s = splsoftclock();
1174 	for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) {
1175 		owp = LIST_NEXT(nfsd, nd_tq);
1176 		if (nfsd->nd_time > cur_usec)
1177 		    break;
1178 		if (nfsd->nd_mreq)
1179 		    continue;
1180 		LIST_REMOVE(nfsd, nd_tq);
1181 		LIST_REMOVE(nfsd, nd_hash);
1182 		splx(s);
1183 		mrep = nfsd->nd_mrep;
1184 		nfsd->nd_mrep = NULL;
1185 		cred = &nfsd->nd_cr;
1186 		v3 = (nfsd->nd_flag & ND_NFSV3);
1187 		forat_ret = aftat_ret = 1;
1188 		error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1189 		    nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH),
1190 		    FALSE);
1191 		if (!error) {
1192 		    if (v3)
1193 			forat_ret = VOP_GETATTR(vp, &forat, cred, procp);
1194 		    if (vp->v_type != VREG) {
1195 			if (v3)
1196 			    error = EINVAL;
1197 			else
1198 			    error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1199 		    }
1200 		} else
1201 		    vp = NULL;
1202 		if (!error) {
1203 		    nqsrv_getl(vp, ND_WRITE);
1204 		    error = nfsrv_access(vp, VWRITE, cred, rdonly, procp, 1);
1205 		}
1206 
1207 		if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1208 		    ioflags = IO_NODELOCKED;
1209 		else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1210 		    ioflags = (IO_SYNC | IO_NODELOCKED);
1211 		else
1212 		    ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1213 		uiop->uio_rw = UIO_WRITE;
1214 		uiop->uio_segflg = UIO_SYSSPACE;
1215 		uiop->uio_procp = (struct proc *)0;
1216 		uiop->uio_offset = nfsd->nd_off;
1217 		uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1218 		if (uiop->uio_resid > 0) {
1219 		    mp = mrep;
1220 		    i = 0;
1221 		    while (mp) {
1222 			if (mp->m_len > 0)
1223 			    i++;
1224 			mp = mp->m_next;
1225 		    }
1226 		    uiop->uio_iovcnt = i;
1227 		    iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
1228 		    uiop->uio_iov = ivp = iov;
1229 		    mp = mrep;
1230 		    while (mp) {
1231 			if (mp->m_len > 0) {
1232 			    ivp->iov_base = mtod(mp, caddr_t);
1233 			    ivp->iov_len = mp->m_len;
1234 			    ivp++;
1235 			}
1236 			mp = mp->m_next;
1237 		    }
1238 		    if (!error) {
1239 			error = VOP_WRITE(vp, uiop, ioflags, cred);
1240 			nfsstats.srvvop_writes++;
1241 		    }
1242 		    free((caddr_t)iov, M_TEMP);
1243 		}
1244 		m_freem(mrep);
1245 		if (vp) {
1246 		    aftat_ret = VOP_GETATTR(vp, &va, cred, procp);
1247 		    vput(vp);
1248 		}
1249 
1250 		/*
1251 		 * Loop around generating replies for all write rpcs that have
1252 		 * now been completed.
1253 		 */
1254 		swp = nfsd;
1255 		do {
1256 		    if (error) {
1257 			nfsm_writereply(NFSX_WCCDATA(v3), v3);
1258 			if (v3) {
1259 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1260 			}
1261 		    } else {
1262 			nfsm_writereply(NFSX_PREOPATTR(v3) +
1263 			    NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1264 			    NFSX_WRITEVERF(v3), v3);
1265 			if (v3) {
1266 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1267 			    nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1268 			    *tl++ = txdr_unsigned(nfsd->nd_len);
1269 			    *tl++ = txdr_unsigned(swp->nd_stable);
1270 			    /*
1271 			     * Actually, there is no need to txdr these fields,
1272 			     * but it may make the values more human readable,
1273 			     * for debugging purposes.
1274 			     */
1275 			    *tl++ = txdr_unsigned(boottime.tv_sec);
1276 			    *tl = txdr_unsigned(boottime.tv_usec);
1277 			} else {
1278 			    nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1279 			    nfsm_srvfillattr(&va, fp);
1280 			}
1281 		    }
1282 		    nfsd->nd_mreq = mreq;
1283 		    if (nfsd->nd_mrep)
1284 			panic("nfsrv_write: nd_mrep not free");
1285 
1286 		    /*
1287 		     * Done. Put it at the head of the timer queue so that
1288 		     * the final phase can return the reply.
1289 		     */
1290 		    s = splsoftclock();
1291 		    if (nfsd != swp) {
1292 			nfsd->nd_time = 0;
1293 			LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1294 		    }
1295 		    nfsd = LIST_FIRST(&swp->nd_coalesce);
1296 		    if (nfsd) {
1297 			LIST_REMOVE(nfsd, nd_tq);
1298 		    }
1299 		    splx(s);
1300 		} while (nfsd);
1301 		s = splsoftclock();
1302 		swp->nd_time = 0;
1303 		LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1304 		splx(s);
1305 		goto loop1;
1306 	}
1307 	splx(s);
1308 
1309 	/*
1310 	 * Search for a reply to return.
1311 	 */
1312 	s = splsoftclock();
1313 	LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) {
1314 		if (nfsd->nd_mreq) {
1315 		    LIST_REMOVE(nfsd, nd_tq);
1316 		    *mrq = nfsd->nd_mreq;
1317 		    *ndp = nfsd;
1318 		    break;
1319 		}
1320 	}
1321 	splx(s);
1322 	return (0);
1323 }
1324 
1325 /*
1326  * Coalesce the write request nfsd into owp. To do this we must:
1327  * - remove nfsd from the queues
1328  * - merge nfsd->nd_mrep into owp->nd_mrep
1329  * - update the nd_eoff and nd_stable for owp
1330  * - put nfsd on owp's nd_coalesce list
1331  * NB: Must be called at splsoftclock().
1332  */
1333 void
1334 nfsrvw_coalesce(owp, nfsd)
1335         struct nfsrv_descript *owp;
1336         struct nfsrv_descript *nfsd;
1337 {
1338         int overlap;
1339         struct mbuf *mp;
1340 	struct nfsrv_descript *m;
1341 
1342         LIST_REMOVE(nfsd, nd_hash);
1343         LIST_REMOVE(nfsd, nd_tq);
1344         if (owp->nd_eoff < nfsd->nd_eoff) {
1345             overlap = owp->nd_eoff - nfsd->nd_off;
1346             if (overlap < 0)
1347                 panic("nfsrv_coalesce: bad off");
1348             if (overlap > 0)
1349                 m_adj(nfsd->nd_mrep, overlap);
1350             mp = owp->nd_mrep;
1351             while (mp->m_next)
1352                 mp = mp->m_next;
1353             mp->m_next = nfsd->nd_mrep;
1354             owp->nd_eoff = nfsd->nd_eoff;
1355         } else
1356             m_freem(nfsd->nd_mrep);
1357         nfsd->nd_mrep = NULL;
1358         if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1359             owp->nd_stable = NFSV3WRITE_FILESYNC;
1360         else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1361             owp->nd_stable == NFSV3WRITE_UNSTABLE)
1362             owp->nd_stable = NFSV3WRITE_DATASYNC;
1363         LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1364  	/*
1365  	 * nfsd might hold coalesce elements! Move them to owp.
1366  	 * Otherwise, requests may be lost and clients will be stuck.
1367  	 */
1368 	while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) {
1369 		LIST_REMOVE(m, nd_tq);
1370 		LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq);
1371 	}
1372 }
1373 
1374 /*
1375  * nfs create service
1376  * now does a truncate to 0 length via. setattr if it already exists
1377  */
1378 int
1379 nfsrv_create(nfsd, slp, procp, mrq)
1380 	struct nfsrv_descript *nfsd;
1381 	struct nfssvc_sock *slp;
1382 	struct proc *procp;
1383 	struct mbuf **mrq;
1384 {
1385 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1386 	struct mbuf *nam = nfsd->nd_nam;
1387 	caddr_t dpos = nfsd->nd_dpos;
1388 	struct ucred *cred = &nfsd->nd_cr;
1389 	struct nfs_fattr *fp;
1390 	struct vattr va, dirfor, diraft;
1391 	struct nfsv2_sattr *sp;
1392 	u_int32_t *tl;
1393 	struct nameidata nd;
1394 	caddr_t cp;
1395 	int32_t t1;
1396 	caddr_t bpos;
1397 	int error = 0, cache, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1398 	int rdev = 0;
1399 	int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1400 	char *cp2;
1401 	struct mbuf *mb, *mreq;
1402 	struct vnode *vp = NULL, *dirp = NULL;
1403 	nfsfh_t nfh;
1404 	fhandle_t *fhp;
1405 	u_quad_t frev, tempsize;
1406 	u_char cverf[NFSX_V3CREATEVERF];
1407 
1408 	nd.ni_cnd.cn_nameiop = 0;
1409 	fhp = &nfh.fh_generic;
1410 	nfsm_srvmtofh(fhp);
1411 	nfsm_srvnamesiz(len);
1412 	nd.ni_cnd.cn_cred = cred;
1413 	nd.ni_cnd.cn_nameiop = CREATE;
1414 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1415 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1416 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1417 	if (dirp) {
1418 		if (v3)
1419 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
1420 				procp);
1421 		else {
1422 			vrele(dirp);
1423 			dirp = (struct vnode *)0;
1424 		}
1425 	}
1426 	if (error) {
1427 		nfsm_reply(NFSX_WCCDATA(v3));
1428 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1429 		if (dirp)
1430 			vrele(dirp);
1431 		return (0);
1432 	}
1433 	VATTR_NULL(&va);
1434 	if (v3) {
1435 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1436 		how = fxdr_unsigned(int, *tl);
1437 		switch (how) {
1438 		case NFSV3CREATE_GUARDED:
1439 			if (nd.ni_vp) {
1440 				error = EEXIST;
1441 				break;
1442 			}
1443 		case NFSV3CREATE_UNCHECKED:
1444 			nfsm_srvsattr(&va);
1445 			break;
1446 		case NFSV3CREATE_EXCLUSIVE:
1447 			nfsm_dissect(cp, caddr_t, NFSX_V3CREATEVERF);
1448 			memcpy(cverf, cp, NFSX_V3CREATEVERF);
1449 			exclusive_flag = 1;
1450 			if (nd.ni_vp == NULL)
1451 				va.va_mode = 0;
1452 			break;
1453 		};
1454 		va.va_type = VREG;
1455 	} else {
1456 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1457 		va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1458 		if (va.va_type == VNON)
1459 			va.va_type = VREG;
1460 		va.va_mode = nfstov_mode(sp->sa_mode);
1461 		switch (va.va_type) {
1462 		case VREG:
1463 			tsize = fxdr_unsigned(int32_t, sp->sa_size);
1464 			if (tsize != -1)
1465 				va.va_size = (u_quad_t)tsize;
1466 			break;
1467 		case VCHR:
1468 		case VBLK:
1469 		case VFIFO:
1470 			rdev = fxdr_unsigned(int32_t, sp->sa_size);
1471 			break;
1472 		default:
1473 			break;
1474 		};
1475 	}
1476 
1477 	/*
1478 	 * Iff doesn't exist, create it
1479 	 * otherwise just truncate to 0 length
1480 	 *   should I set the mode too ??
1481 	 */
1482 	if (nd.ni_vp == NULL) {
1483 		if (va.va_type == VREG || va.va_type == VSOCK) {
1484 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1485 			error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1486 			if (!error) {
1487 				if (exclusive_flag) {
1488 					exclusive_flag = 0;
1489 					VATTR_NULL(&va);
1490 					memcpy((caddr_t)&va.va_atime, cverf,
1491 						NFSX_V3CREATEVERF);
1492 					error = VOP_SETATTR(nd.ni_vp, &va, cred,
1493 						procp);
1494 				}
1495 			}
1496 		} else if (va.va_type == VCHR || va.va_type == VBLK ||
1497 			va.va_type == VFIFO) {
1498 			if (va.va_type == VCHR && rdev == 0xffffffff)
1499 				va.va_type = VFIFO;
1500 			if (va.va_type != VFIFO &&
1501 			    (error = suser(cred, (u_short *)0))) {
1502 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1503 				vput(nd.ni_dvp);
1504 				nfsm_reply(0);
1505 				return (error);
1506 			} else
1507 				va.va_rdev = (dev_t)rdev;
1508 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1509 			error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd,
1510 			    &va);
1511 			if (error) {
1512 				nfsm_reply(0);
1513 			}
1514 			if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1515 				vrele(nd.ni_dvp);
1516 				vput(nd.ni_vp);
1517 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1518 				error = EINVAL;
1519 				nfsm_reply(0);
1520 			}
1521 		} else {
1522 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1523 			vput(nd.ni_dvp);
1524 			error = ENXIO;
1525 		}
1526 		vp = nd.ni_vp;
1527 	} else {
1528 		vp = nd.ni_vp;
1529 		if (nd.ni_dvp == vp)
1530 			vrele(nd.ni_dvp);
1531 		else
1532 			vput(nd.ni_dvp);
1533 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1534 		if (!error && va.va_size != -1) {
1535 			error = nfsrv_access(vp, VWRITE, cred,
1536 			    (nd.ni_cnd.cn_flags & RDONLY), procp, 0);
1537 			if (!error) {
1538 				nqsrv_getl(vp, ND_WRITE);
1539 				tempsize = va.va_size;
1540 				VATTR_NULL(&va);
1541 				va.va_size = tempsize;
1542 				error = VOP_SETATTR(vp, &va, cred,
1543 					 procp);
1544 			}
1545 		}
1546 		if (error)
1547 			vput(vp);
1548 	}
1549 	if (!error) {
1550 		memset((caddr_t)fhp, 0, sizeof(nfh));
1551 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1552 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
1553 		if (!error)
1554 			error = VOP_GETATTR(vp, &va, cred, procp);
1555 		vput(vp);
1556 	}
1557 	if (v3) {
1558 		if (exclusive_flag && !error &&
1559 			memcmp(cverf, (caddr_t)&va.va_atime, NFSX_V3CREATEVERF))
1560 			error = EEXIST;
1561 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1562 		vrele(dirp);
1563 	}
1564 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1565 	if (v3) {
1566 		if (!error) {
1567 			nfsm_srvpostop_fh(fhp);
1568 			nfsm_srvpostop_attr(0, &va);
1569 		}
1570 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1571 	} else {
1572 		nfsm_srvfhtom(fhp, v3);
1573 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1574 		nfsm_srvfillattr(&va, fp);
1575 	}
1576 	return (0);
1577 nfsmout:
1578 	if (dirp)
1579 		vrele(dirp);
1580 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1581 	if (nd.ni_dvp == nd.ni_vp)
1582 		vrele(nd.ni_dvp);
1583 	else
1584 		vput(nd.ni_dvp);
1585 	if (nd.ni_vp)
1586 		vput(nd.ni_vp);
1587 	return (error);
1588 }
1589 
1590 /*
1591  * nfs v3 mknod service
1592  */
1593 int
1594 nfsrv_mknod(nfsd, slp, procp, mrq)
1595 	struct nfsrv_descript *nfsd;
1596 	struct nfssvc_sock *slp;
1597 	struct proc *procp;
1598 	struct mbuf **mrq;
1599 {
1600 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1601 	struct mbuf *nam = nfsd->nd_nam;
1602 	caddr_t dpos = nfsd->nd_dpos;
1603 	struct ucred *cred = &nfsd->nd_cr;
1604 	struct vattr va, dirfor, diraft;
1605 	u_int32_t *tl;
1606 	struct nameidata nd;
1607 	int32_t t1;
1608 	caddr_t bpos;
1609 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
1610 	u_int32_t major, minor;
1611 	enum vtype vtyp;
1612 	char *cp2;
1613 	struct mbuf *mb, *mreq;
1614 	struct vnode *vp, *dirp = (struct vnode *)0;
1615 	nfsfh_t nfh;
1616 	fhandle_t *fhp;
1617 	u_quad_t frev;
1618 
1619 	nd.ni_cnd.cn_nameiop = 0;
1620 	fhp = &nfh.fh_generic;
1621 	nfsm_srvmtofh(fhp);
1622 	nfsm_srvnamesiz(len);
1623 	nd.ni_cnd.cn_cred = cred;
1624 	nd.ni_cnd.cn_nameiop = CREATE;
1625 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1626 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1627 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1628 	if (dirp)
1629 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, procp);
1630 	if (error) {
1631 		nfsm_reply(NFSX_WCCDATA(1));
1632 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1633 		if (dirp)
1634 			vrele(dirp);
1635 		return (0);
1636 	}
1637 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1638 	vtyp = nfsv3tov_type(*tl);
1639 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1640 		error = NFSERR_BADTYPE;
1641 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1642 		if (nd.ni_dvp == nd.ni_vp)
1643 			vrele(nd.ni_dvp);
1644 		else
1645 			vput(nd.ni_dvp);
1646 		if (nd.ni_vp)
1647 			vput(nd.ni_vp);
1648 		goto out;
1649 	}
1650 	VATTR_NULL(&va);
1651 	nfsm_srvsattr(&va);
1652 	if (vtyp == VCHR || vtyp == VBLK) {
1653 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1654 		major = fxdr_unsigned(u_int32_t, *tl++);
1655 		minor = fxdr_unsigned(u_int32_t, *tl);
1656 		va.va_rdev = makedev(major, minor);
1657 	}
1658 
1659 	/*
1660 	 * Iff doesn't exist, create it.
1661 	 */
1662 	if (nd.ni_vp) {
1663 		error = EEXIST;
1664 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1665 		if (nd.ni_dvp == nd.ni_vp)
1666 			vrele(nd.ni_dvp);
1667 		else
1668 			vput(nd.ni_dvp);
1669 		vput(nd.ni_vp);
1670 		goto out;
1671 	}
1672 	va.va_type = vtyp;
1673 	if (vtyp == VSOCK) {
1674 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1675 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1676 	} else {
1677 		if (va.va_type != VFIFO &&
1678 		    (error = suser(cred, (u_short *)0))) {
1679 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1680 			vput(nd.ni_dvp);
1681 			goto out;
1682 		}
1683 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1684 		error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1685 		if (error) {
1686 			goto out;
1687 		}
1688 		if (error)
1689 			goto out;
1690 		if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1691 			vput(nd.ni_vp);
1692 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1693 			error = EINVAL;
1694 		}
1695 	}
1696 out:
1697 	vp = nd.ni_vp;
1698 	if (!error) {
1699 		memset((caddr_t)fhp, 0, sizeof(nfh));
1700 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1701 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
1702 		if (!error)
1703 			error = VOP_GETATTR(vp, &va, cred, procp);
1704 		vput(vp);
1705 	}
1706 	diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1707 	vrele(dirp);
1708 	nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
1709 	if (!error) {
1710 		nfsm_srvpostop_fh(fhp);
1711 		nfsm_srvpostop_attr(0, &va);
1712 	}
1713 	nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1714 	return (0);
1715 nfsmout:
1716 	if (dirp)
1717 		vrele(dirp);
1718 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1719 	if (nd.ni_dvp == nd.ni_vp)
1720 		vrele(nd.ni_dvp);
1721 	else
1722 		vput(nd.ni_dvp);
1723 	if (nd.ni_vp)
1724 		vput(nd.ni_vp);
1725 	return (error);
1726 }
1727 
1728 /*
1729  * nfs remove service
1730  */
1731 int
1732 nfsrv_remove(nfsd, slp, procp, mrq)
1733 	struct nfsrv_descript *nfsd;
1734 	struct nfssvc_sock *slp;
1735 	struct proc *procp;
1736 	struct mbuf **mrq;
1737 {
1738 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1739 	struct mbuf *nam = nfsd->nd_nam;
1740 	caddr_t dpos = nfsd->nd_dpos;
1741 	struct ucred *cred = &nfsd->nd_cr;
1742 	struct nameidata nd;
1743 	u_int32_t *tl;
1744 	int32_t t1;
1745 	caddr_t bpos;
1746 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
1747 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1748 	char *cp2;
1749 	struct mbuf *mb, *mreq;
1750 	struct vnode *vp, *dirp;
1751 	struct vattr dirfor, diraft;
1752 	nfsfh_t nfh;
1753 	fhandle_t *fhp;
1754 	u_quad_t frev;
1755 
1756 #ifndef nolint
1757 	vp = (struct vnode *)0;
1758 #endif
1759 	fhp = &nfh.fh_generic;
1760 	nfsm_srvmtofh(fhp);
1761 	nfsm_srvnamesiz(len);
1762 	nd.ni_cnd.cn_cred = cred;
1763 	nd.ni_cnd.cn_nameiop = DELETE;
1764 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1765 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1766 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1767 	if (dirp) {
1768 		if (v3)
1769 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
1770 				procp);
1771 		else
1772 			vrele(dirp);
1773 	}
1774 	if (!error) {
1775 		vp = nd.ni_vp;
1776 		if (vp->v_type == VDIR &&
1777 		    (error = suser(cred, (u_short *)0)) != 0)
1778 			goto out;
1779 		/*
1780 		 * The root of a mounted filesystem cannot be deleted.
1781 		 */
1782 		if (vp->v_flag & VROOT) {
1783 			error = EBUSY;
1784 			goto out;
1785 		}
1786 out:
1787 		if (!error) {
1788 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1789 			nqsrv_getl(vp, ND_WRITE);
1790 			error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
1791 		} else {
1792 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1793 			if (nd.ni_dvp == vp)
1794 				vrele(nd.ni_dvp);
1795 			else
1796 				vput(nd.ni_dvp);
1797 			vput(vp);
1798 		}
1799 	}
1800 	if (dirp && v3) {
1801 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1802 		vrele(dirp);
1803 	}
1804 	nfsm_reply(NFSX_WCCDATA(v3));
1805 	if (v3) {
1806 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1807 		return (0);
1808 	}
1809 	nfsm_srvdone;
1810 }
1811 
1812 /*
1813  * nfs rename service
1814  */
1815 int
1816 nfsrv_rename(nfsd, slp, procp, mrq)
1817 	struct nfsrv_descript *nfsd;
1818 	struct nfssvc_sock *slp;
1819 	struct proc *procp;
1820 	struct mbuf **mrq;
1821 {
1822 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1823 	struct mbuf *nam = nfsd->nd_nam;
1824 	caddr_t dpos = nfsd->nd_dpos;
1825 	struct ucred *cred = &nfsd->nd_cr;
1826 	u_int32_t *tl;
1827 	int32_t t1;
1828 	caddr_t bpos;
1829 	int error = 0, cache, fdirfor_ret = 1, fdiraft_ret = 1;
1830 	uint32_t len, len2;
1831 	int tdirfor_ret = 1, tdiraft_ret = 1;
1832 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1833 	char *cp2;
1834 	struct mbuf *mb, *mreq;
1835 	struct nameidata fromnd, tond;
1836 	struct vnode *fvp, *tvp, *tdvp, *fdirp = (struct vnode *)0;
1837 	struct vnode *tdirp = (struct vnode *)0;
1838 	struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
1839 	nfsfh_t fnfh, tnfh;
1840 	fhandle_t *ffhp, *tfhp;
1841 	u_quad_t frev;
1842 	uid_t saved_uid;
1843 
1844 #ifndef nolint
1845 	fvp = (struct vnode *)0;
1846 #endif
1847 	ffhp = &fnfh.fh_generic;
1848 	tfhp = &tnfh.fh_generic;
1849 	fromnd.ni_cnd.cn_nameiop = 0;
1850 	tond.ni_cnd.cn_nameiop = 0;
1851 	nfsm_srvmtofh(ffhp);
1852 	nfsm_srvnamesiz(len);
1853 	/*
1854 	 * Remember our original uid so that we can reset cr_uid before
1855 	 * the second nfs_namei() call, in case it is remapped.
1856 	 */
1857 	saved_uid = cred->cr_uid;
1858 	fromnd.ni_cnd.cn_cred = cred;
1859 	fromnd.ni_cnd.cn_nameiop = DELETE;
1860 	fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART;
1861 	error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md,
1862 		&dpos, &fdirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1863 	if (fdirp) {
1864 		if (v3)
1865 			fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred,
1866 				procp);
1867 		else {
1868 			vrele(fdirp);
1869 			fdirp = (struct vnode *)0;
1870 		}
1871 	}
1872 	if (error) {
1873 		nfsm_reply(2 * NFSX_WCCDATA(v3));
1874 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1875 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1876 		if (fdirp)
1877 			vrele(fdirp);
1878 		return (0);
1879 	}
1880 	fvp = fromnd.ni_vp;
1881 	nfsm_srvmtofh(tfhp);
1882 	if (v3) {
1883 		nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1884 		len2 = fxdr_unsigned(uint32_t, *tl);
1885 		/* len2 will be checked by nfs_namei */
1886 	}
1887 	else {
1888 		/* NFSv2 */
1889 		nfsm_strsiz(len2, NFS_MAXNAMLEN);
1890 	}
1891 	cred->cr_uid = saved_uid;
1892 	tond.ni_cnd.cn_cred = cred;
1893 	tond.ni_cnd.cn_nameiop = RENAME;
1894 	tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART;
1895 	error = nfs_namei(&tond, tfhp, len2, slp, nam, &md,
1896 		&dpos, &tdirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1897 	if (tdirp) {
1898 		if (v3)
1899 			tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred,
1900 				procp);
1901 		else {
1902 			vrele(tdirp);
1903 			tdirp = (struct vnode *)0;
1904 		}
1905 	}
1906 	if (error) {
1907 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1908 		vrele(fromnd.ni_dvp);
1909 		vrele(fvp);
1910 		goto out1;
1911 	}
1912 	tdvp = tond.ni_dvp;
1913 	tvp = tond.ni_vp;
1914 	if (tvp != NULL) {
1915 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1916 			if (v3)
1917 				error = EEXIST;
1918 			else
1919 				error = EISDIR;
1920 			goto out;
1921 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1922 			if (v3)
1923 				error = EEXIST;
1924 			else
1925 				error = ENOTDIR;
1926 			goto out;
1927 		}
1928 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1929 			if (v3)
1930 				error = EXDEV;
1931 			else
1932 				error = ENOTEMPTY;
1933 			goto out;
1934 		}
1935 	}
1936 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1937 		if (v3)
1938 			error = EXDEV;
1939 		else
1940 			error = ENOTEMPTY;
1941 		goto out;
1942 	}
1943 	if (fvp->v_mount != tdvp->v_mount) {
1944 		if (v3)
1945 			error = EXDEV;
1946 		else
1947 			error = ENOTEMPTY;
1948 		goto out;
1949 	}
1950 	if (fvp == tdvp) {
1951 		if (v3)
1952 			error = EINVAL;
1953 		else
1954 			error = ENOTEMPTY;
1955 	}
1956 	/*
1957 	 * If source is the same as the destination (that is the
1958 	 * same vnode with the same name in the same directory),
1959 	 * then there is nothing to do.
1960 	 */
1961 	if (fvp == tvp && fromnd.ni_dvp == tdvp &&
1962 	    fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
1963 	    !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
1964 	      fromnd.ni_cnd.cn_namelen))
1965 		error = -1;
1966 out:
1967 	if (!error) {
1968 		nqsrv_getl(fromnd.ni_dvp, ND_WRITE);
1969 		nqsrv_getl(tdvp, ND_WRITE);
1970 		if (tvp) {
1971 			nqsrv_getl(tvp, ND_WRITE);
1972 		}
1973 		error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
1974 				   tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
1975 	} else {
1976 		VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd);
1977 		if (tdvp == tvp)
1978 			vrele(tdvp);
1979 		else
1980 			vput(tdvp);
1981 		if (tvp)
1982 			vput(tvp);
1983 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1984 		vrele(fromnd.ni_dvp);
1985 		vrele(fvp);
1986 		if (error == -1)
1987 			error = 0;
1988 	}
1989 	vrele(tond.ni_startdir);
1990 	PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
1991 out1:
1992 	if (fdirp) {
1993 		fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, procp);
1994 		vrele(fdirp);
1995 	}
1996 	if (tdirp) {
1997 		tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, procp);
1998 		vrele(tdirp);
1999 	}
2000 	vrele(fromnd.ni_startdir);
2001 	PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
2002 	nfsm_reply(2 * NFSX_WCCDATA(v3));
2003 	if (v3) {
2004 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2005 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2006 	}
2007 	return (0);
2008 
2009 nfsmout:
2010 	if (fdirp)
2011 		vrele(fdirp);
2012 	if (tdirp)
2013 		vrele(tdirp);
2014 	if (tond.ni_cnd.cn_nameiop) {
2015 		vrele(tond.ni_startdir);
2016 		PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
2017 	}
2018 	if (fromnd.ni_cnd.cn_nameiop) {
2019 		vrele(fromnd.ni_startdir);
2020 		PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
2021 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2022 		vrele(fromnd.ni_dvp);
2023 		vrele(fvp);
2024 	}
2025 	return (error);
2026 }
2027 
2028 /*
2029  * nfs link service
2030  */
2031 int
2032 nfsrv_link(nfsd, slp, procp, mrq)
2033 	struct nfsrv_descript *nfsd;
2034 	struct nfssvc_sock *slp;
2035 	struct proc *procp;
2036 	struct mbuf **mrq;
2037 {
2038 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2039 	struct mbuf *nam = nfsd->nd_nam;
2040 	caddr_t dpos = nfsd->nd_dpos;
2041 	struct ucred *cred = &nfsd->nd_cr;
2042 	struct nameidata nd;
2043 	u_int32_t *tl;
2044 	int32_t t1;
2045 	caddr_t bpos;
2046 	int error = 0, rdonly, cache, len, dirfor_ret = 1, diraft_ret = 1;
2047 	int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2048 	char *cp2;
2049 	struct mbuf *mb, *mreq;
2050 	struct vnode *vp, *xp, *dirp = (struct vnode *)0;
2051 	struct vattr dirfor, diraft, at;
2052 	nfsfh_t nfh, dnfh;
2053 	fhandle_t *fhp, *dfhp;
2054 	u_quad_t frev;
2055 
2056 	fhp = &nfh.fh_generic;
2057 	dfhp = &dnfh.fh_generic;
2058 	nfsm_srvmtofh(fhp);
2059 	nfsm_srvmtofh(dfhp);
2060 	nfsm_srvnamesiz(len);
2061 	error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, slp, nam,
2062 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2063 	if (error) {
2064 		nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2065 		nfsm_srvpostop_attr(getret, &at);
2066 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2067 		return (0);
2068 	}
2069 	if (vp->v_type == VDIR && (error = suser(cred, (u_short *)0)) != 0)
2070 		goto out1;
2071 	nd.ni_cnd.cn_cred = cred;
2072 	nd.ni_cnd.cn_nameiop = CREATE;
2073 	nd.ni_cnd.cn_flags = LOCKPARENT;
2074 	error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos,
2075 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2076 	if (dirp) {
2077 		if (v3)
2078 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2079 				procp);
2080 		else {
2081 			vrele(dirp);
2082 			dirp = (struct vnode *)0;
2083 		}
2084 	}
2085 	if (error)
2086 		goto out1;
2087 	xp = nd.ni_vp;
2088 	if (xp != NULL) {
2089 		error = EEXIST;
2090 		goto out;
2091 	}
2092 	xp = nd.ni_dvp;
2093 	if (vp->v_mount != xp->v_mount)
2094 		error = EXDEV;
2095 out:
2096 	if (!error) {
2097 		nqsrv_getl(vp, ND_WRITE);
2098 		nqsrv_getl(xp, ND_WRITE);
2099 		error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2100 	} else {
2101 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2102 		if (nd.ni_dvp == nd.ni_vp)
2103 			vrele(nd.ni_dvp);
2104 		else
2105 			vput(nd.ni_dvp);
2106 		if (nd.ni_vp)
2107 			vrele(nd.ni_vp);
2108 	}
2109 out1:
2110 	if (v3)
2111 		getret = VOP_GETATTR(vp, &at, cred, procp);
2112 	if (dirp) {
2113 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2114 		vrele(dirp);
2115 	}
2116 	vrele(vp);
2117 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2118 	if (v3) {
2119 		nfsm_srvpostop_attr(getret, &at);
2120 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2121 		return (0);
2122 	}
2123 	nfsm_srvdone;
2124 }
2125 
2126 /*
2127  * nfs symbolic link service
2128  */
2129 int
2130 nfsrv_symlink(nfsd, slp, procp, mrq)
2131 	struct nfsrv_descript *nfsd;
2132 	struct nfssvc_sock *slp;
2133 	struct proc *procp;
2134 	struct mbuf **mrq;
2135 {
2136 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2137 	struct mbuf *nam = nfsd->nd_nam;
2138 	caddr_t dpos = nfsd->nd_dpos;
2139 	struct ucred *cred = &nfsd->nd_cr;
2140 	struct vattr va, dirfor, diraft;
2141 	struct nameidata nd;
2142 	u_int32_t *tl;
2143 	int32_t t1;
2144 	struct nfsv2_sattr *sp;
2145 	char *bpos, *pathcp = NULL, *cp2;
2146 	struct uio io;
2147 	struct iovec iv;
2148 	int error = 0, cache, dirfor_ret = 1, diraft_ret = 1;
2149 	uint32_t len, len2;
2150 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2151 	struct mbuf *mb, *mreq;
2152 	struct vnode *dirp = (struct vnode *)0;
2153 	nfsfh_t nfh;
2154 	fhandle_t *fhp;
2155 	u_quad_t frev;
2156 
2157 	nd.ni_cnd.cn_nameiop = 0;
2158 	fhp = &nfh.fh_generic;
2159 	nfsm_srvmtofh(fhp);
2160 	nfsm_srvnamesiz(len);
2161 	nd.ni_cnd.cn_cred = cred;
2162 	nd.ni_cnd.cn_nameiop = CREATE;
2163 	nd.ni_cnd.cn_flags = LOCKPARENT;
2164 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2165 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2166 	if (dirp) {
2167 		if (v3)
2168 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2169 				procp);
2170 		else {
2171 			vrele(dirp);
2172 			dirp = (struct vnode *)0;
2173 		}
2174 	}
2175 	if (error)
2176 		goto out;
2177 	VATTR_NULL(&va);
2178 	if (v3) {
2179 		nfsm_srvsattr(&va);
2180 		nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
2181 		len2 = fxdr_unsigned(uint32_t, *tl);
2182 		if (len2 > PATH_MAX) {
2183 			/* XXX should check _PC_NO_TRUNC */
2184 			error = ENAMETOOLONG;
2185 			goto abortop;
2186 		}
2187 	}
2188 	else {
2189 		/* NFSv2 */
2190 		nfsm_strsiz(len2, NFS_MAXPATHLEN);
2191 	}
2192 	pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK);
2193 	iv.iov_base = pathcp;
2194 	iv.iov_len = len2;
2195 	io.uio_resid = len2;
2196 	io.uio_offset = 0;
2197 	io.uio_iov = &iv;
2198 	io.uio_iovcnt = 1;
2199 	io.uio_segflg = UIO_SYSSPACE;
2200 	io.uio_rw = UIO_READ;
2201 	io.uio_procp = (struct proc *)0;
2202 	nfsm_mtouio(&io, len2);
2203 	if (!v3) {
2204 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2205 		va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
2206 	}
2207 	*(pathcp + len2) = '\0';
2208 	if (nd.ni_vp) {
2209 		error = EEXIST;
2210 abortop:
2211 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2212 		if (nd.ni_dvp == nd.ni_vp)
2213 			vrele(nd.ni_dvp);
2214 		else
2215 			vput(nd.ni_dvp);
2216 		if (nd.ni_vp)
2217 			vrele(nd.ni_vp);
2218 		goto out;
2219 	}
2220 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2221 	error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp);
2222 	if (!error) {
2223 	    if (v3) {
2224 		memset((caddr_t)fhp, 0, sizeof(nfh));
2225 		fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2226 		error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2227 		if (!error)
2228 		    error = VOP_GETATTR(nd.ni_vp, &va, cred,
2229 					procp);
2230 		vput(nd.ni_vp);
2231 	    } else {
2232 		vput(nd.ni_vp);
2233 	    }
2234 	}
2235 out:
2236 	if (pathcp)
2237 		free(pathcp, M_TEMP);
2238 	if (dirp) {
2239 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2240 		vrele(dirp);
2241 	}
2242 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2243 	if (v3) {
2244 		if (!error) {
2245 			nfsm_srvpostop_fh(fhp);
2246 			nfsm_srvpostop_attr(0, &va);
2247 		}
2248 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2249 	}
2250 	return (0);
2251 nfsmout:
2252 	if (dirp)
2253 		vrele(dirp);
2254 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2255 	if (nd.ni_dvp == nd.ni_vp)
2256 		vrele(nd.ni_dvp);
2257 	else
2258 		vput(nd.ni_dvp);
2259 	if (nd.ni_vp)
2260 		vrele(nd.ni_vp);
2261 	if (pathcp)
2262 		free(pathcp, M_TEMP);
2263 	return (error);
2264 }
2265 
2266 /*
2267  * nfs mkdir service
2268  */
2269 int
2270 nfsrv_mkdir(nfsd, slp, procp, mrq)
2271 	struct nfsrv_descript *nfsd;
2272 	struct nfssvc_sock *slp;
2273 	struct proc *procp;
2274 	struct mbuf **mrq;
2275 {
2276 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2277 	struct mbuf *nam = nfsd->nd_nam;
2278 	caddr_t dpos = nfsd->nd_dpos;
2279 	struct ucred *cred = &nfsd->nd_cr;
2280 	struct vattr va, dirfor, diraft;
2281 	struct nfs_fattr *fp;
2282 	struct nameidata nd;
2283 	caddr_t cp;
2284 	u_int32_t *tl;
2285 	int32_t t1;
2286 	caddr_t bpos;
2287 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2288 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2289 	char *cp2;
2290 	struct mbuf *mb, *mreq;
2291 	struct vnode *vp, *dirp = (struct vnode *)0;
2292 	nfsfh_t nfh;
2293 	fhandle_t *fhp;
2294 	u_quad_t frev;
2295 
2296 	fhp = &nfh.fh_generic;
2297 	nfsm_srvmtofh(fhp);
2298 	nfsm_srvnamesiz(len);
2299 	nd.ni_cnd.cn_cred = cred;
2300 	nd.ni_cnd.cn_nameiop = CREATE;
2301 	nd.ni_cnd.cn_flags = LOCKPARENT;
2302 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2303 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2304 	if (dirp) {
2305 		if (v3)
2306 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2307 				procp);
2308 		else {
2309 			vrele(dirp);
2310 			dirp = (struct vnode *)0;
2311 		}
2312 	}
2313 	if (error) {
2314 		nfsm_reply(NFSX_WCCDATA(v3));
2315 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2316 		if (dirp)
2317 			vrele(dirp);
2318 		return (0);
2319 	}
2320 	VATTR_NULL(&va);
2321 	if (v3) {
2322 		nfsm_srvsattr(&va);
2323 	} else {
2324 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2325 		va.va_mode = nfstov_mode(*tl++);
2326 	}
2327 	va.va_type = VDIR;
2328 	vp = nd.ni_vp;
2329 	if (vp != NULL) {
2330 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2331 		if (nd.ni_dvp == vp)
2332 			vrele(nd.ni_dvp);
2333 		else
2334 			vput(nd.ni_dvp);
2335 		vrele(vp);
2336 		error = EEXIST;
2337 		goto out;
2338 	}
2339 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2340 	error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
2341 	if (!error) {
2342 		vp = nd.ni_vp;
2343 		memset((caddr_t)fhp, 0, sizeof(nfh));
2344 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2345 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
2346 		if (!error)
2347 			error = VOP_GETATTR(vp, &va, cred, procp);
2348 		vput(vp);
2349 	}
2350 out:
2351 	if (dirp) {
2352 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2353 		vrele(dirp);
2354 	}
2355 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2356 	if (v3) {
2357 		if (!error) {
2358 			nfsm_srvpostop_fh(fhp);
2359 			nfsm_srvpostop_attr(0, &va);
2360 		}
2361 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2362 	} else {
2363 		nfsm_srvfhtom(fhp, v3);
2364 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2365 		nfsm_srvfillattr(&va, fp);
2366 	}
2367 	return (0);
2368 nfsmout:
2369 	if (dirp)
2370 		vrele(dirp);
2371 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2372 	if (nd.ni_dvp == nd.ni_vp)
2373 		vrele(nd.ni_dvp);
2374 	else
2375 		vput(nd.ni_dvp);
2376 	if (nd.ni_vp)
2377 		vrele(nd.ni_vp);
2378 	return (error);
2379 }
2380 
2381 /*
2382  * nfs rmdir service
2383  */
2384 int
2385 nfsrv_rmdir(nfsd, slp, procp, mrq)
2386 	struct nfsrv_descript *nfsd;
2387 	struct nfssvc_sock *slp;
2388 	struct proc *procp;
2389 	struct mbuf **mrq;
2390 {
2391 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2392 	struct mbuf *nam = nfsd->nd_nam;
2393 	caddr_t dpos = nfsd->nd_dpos;
2394 	struct ucred *cred = &nfsd->nd_cr;
2395 	u_int32_t *tl;
2396 	int32_t t1;
2397 	caddr_t bpos;
2398 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2399 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2400 	char *cp2;
2401 	struct mbuf *mb, *mreq;
2402 	struct vnode *vp, *dirp = (struct vnode *)0;
2403 	struct vattr dirfor, diraft;
2404 	nfsfh_t nfh;
2405 	fhandle_t *fhp;
2406 	struct nameidata nd;
2407 	u_quad_t frev;
2408 
2409 	fhp = &nfh.fh_generic;
2410 	nfsm_srvmtofh(fhp);
2411 	nfsm_srvnamesiz(len);
2412 	nd.ni_cnd.cn_cred = cred;
2413 	nd.ni_cnd.cn_nameiop = DELETE;
2414 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
2415 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2416 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2417 	if (dirp) {
2418 		if (v3)
2419 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2420 				procp);
2421 		else {
2422 			vrele(dirp);
2423 			dirp = (struct vnode *)0;
2424 		}
2425 	}
2426 	if (error) {
2427 		nfsm_reply(NFSX_WCCDATA(v3));
2428 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2429 		if (dirp)
2430 			vrele(dirp);
2431 		return (0);
2432 	}
2433 	vp = nd.ni_vp;
2434 	if (vp->v_type != VDIR) {
2435 		error = ENOTDIR;
2436 		goto out;
2437 	}
2438 	/*
2439 	 * No rmdir "." please.
2440 	 */
2441 	if (nd.ni_dvp == vp) {
2442 		error = EINVAL;
2443 		goto out;
2444 	}
2445 	/*
2446 	 * The root of a mounted filesystem cannot be deleted.
2447 	 */
2448 	if (vp->v_flag & VROOT)
2449 		error = EBUSY;
2450 out:
2451 	if (!error) {
2452 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
2453 		nqsrv_getl(vp, ND_WRITE);
2454 		error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2455 	} else {
2456 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2457 		if (nd.ni_dvp == nd.ni_vp)
2458 			vrele(nd.ni_dvp);
2459 		else
2460 			vput(nd.ni_dvp);
2461 		vput(vp);
2462 	}
2463 	if (dirp) {
2464 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2465 		vrele(dirp);
2466 	}
2467 	nfsm_reply(NFSX_WCCDATA(v3));
2468 	if (v3) {
2469 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2470 		return (0);
2471 	}
2472 	nfsm_srvdone;
2473 }
2474 
2475 /*
2476  * nfs readdir service
2477  * - mallocs what it thinks is enough to read
2478  *	count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
2479  * - calls VOP_READDIR()
2480  * - loops around building the reply
2481  *	if the output generated exceeds count break out of loop
2482  *	The nfsm_clget macro is used here so that the reply will be packed
2483  *	tightly in mbuf clusters.
2484  * - it only knows that it has encountered eof when the VOP_READDIR()
2485  *	reads nothing
2486  * - as such one readdir rpc will return eof false although you are there
2487  *	and then the next will return eof
2488  * - it trims out records with d_fileno == 0
2489  *	this doesn't matter for Unix clients, but they might confuse clients
2490  *	for other os'.
2491  * - it trims out records with d_type == DT_WHT
2492  *	these cannot be seen through NFS (unless we extend the protocol)
2493  * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2494  *	than requested, but this may not apply to all filesystems. For
2495  *	example, client NFS does not { although it is never remote mounted
2496  *	anyhow }
2497  *     The alternate call nfsrv_readdirplus() does lookups as well.
2498  * PS: The NFS protocol spec. does not clarify what the "count" byte
2499  *	argument is a count of.. just name strings and file id's or the
2500  *	entire reply rpc or ...
2501  *	I tried just file name and id sizes and it confused the Sun client,
2502  *	so I am using the full rpc size now. The "paranoia.." comment refers
2503  *	to including the status longwords that are not a part of the dir.
2504  *	"entry" structures, but are in the rpc.
2505  */
2506 struct flrep {
2507 	nfsuint64 fl_off;
2508 	u_int32_t fl_postopok;
2509 	u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2510 	u_int32_t fl_fhok;
2511 	u_int32_t fl_fhsize;
2512 	u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
2513 };
2514 
2515 int
2516 nfsrv_readdir(nfsd, slp, procp, mrq)
2517 	struct nfsrv_descript *nfsd;
2518 	struct nfssvc_sock *slp;
2519 	struct proc *procp;
2520 	struct mbuf **mrq;
2521 {
2522 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2523 	struct mbuf *nam = nfsd->nd_nam;
2524 	caddr_t dpos = nfsd->nd_dpos;
2525 	struct ucred *cred = &nfsd->nd_cr;
2526 	char *bp, *be;
2527 	struct mbuf *mp;
2528 	struct dirent *dp;
2529 	caddr_t cp;
2530 	u_int32_t *tl;
2531 	int32_t t1;
2532 	caddr_t bpos;
2533 	struct mbuf *mb, *mreq, *mp2;
2534 	char *cpos, *cend, *cp2, *rbuf;
2535 	struct vnode *vp;
2536 	struct vattr at;
2537 	nfsfh_t nfh;
2538 	fhandle_t *fhp;
2539 	struct uio io;
2540 	struct iovec iv;
2541 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2542 	int siz, cnt, fullsiz, eofflag, rdonly, cache, ncookies;
2543 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2544 	u_quad_t frev, off, toff, verf;
2545 	off_t *cookies = NULL, *cookiep;
2546 	nfsuint64 jar;
2547 
2548 	fhp = &nfh.fh_generic;
2549 	nfsm_srvmtofh(fhp);
2550 	if (v3) {
2551 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2552 		toff = fxdr_hyper(tl);
2553 		tl += 2;
2554 		verf = fxdr_hyper(tl);
2555 		tl += 2;
2556 	} else {
2557 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2558 		toff = fxdr_unsigned(u_quad_t, *tl++);
2559 	}
2560 	off = toff;
2561 	cnt = fxdr_unsigned(int, *tl);
2562 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2563 	xfer = NFS_SRVMAXDATA(nfsd);
2564 	if (siz > xfer)
2565 		siz = xfer;
2566 	fullsiz = siz;
2567 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
2568 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2569 	if (!error && vp->v_type != VDIR) {
2570 		error = ENOTDIR;
2571 		vput(vp);
2572 	}
2573 	if (error) {
2574 		nfsm_reply(NFSX_UNSIGNED);
2575 		nfsm_srvpostop_attr(getret, &at);
2576 		return (0);
2577 	}
2578 	nqsrv_getl(vp, ND_READ);
2579 	if (v3) {
2580 		error = getret = VOP_GETATTR(vp, &at, cred, procp);
2581 #ifdef NFS3_STRICTVERF
2582 		/*
2583 		 * XXX This check is too strict for Solaris 2.5 clients.
2584 		 */
2585 		if (!error && toff && verf != at.va_filerev)
2586 			error = NFSERR_BAD_COOKIE;
2587 #endif
2588 	}
2589 	if (!error)
2590 		error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0);
2591 	if (error) {
2592 		vput(vp);
2593 		nfsm_reply(NFSX_POSTOPATTR(v3));
2594 		nfsm_srvpostop_attr(getret, &at);
2595 		return (0);
2596 	}
2597 	VOP_UNLOCK(vp, 0);
2598 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2599 again:
2600 	iv.iov_base = rbuf;
2601 	iv.iov_len = fullsiz;
2602 	io.uio_iov = &iv;
2603 	io.uio_iovcnt = 1;
2604 	io.uio_offset = (off_t)off;
2605 	io.uio_resid = fullsiz;
2606 	io.uio_segflg = UIO_SYSSPACE;
2607 	io.uio_rw = UIO_READ;
2608 	io.uio_procp = (struct proc *)0;
2609 	eofflag = 0;
2610 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2611 
2612 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2613 
2614 	off = (off_t)io.uio_offset;
2615 	if (!cookies && !error)
2616 		error = NFSERR_PERM;
2617 	if (v3) {
2618 		getret = VOP_GETATTR(vp, &at, cred, procp);
2619 		if (!error)
2620 			error = getret;
2621 	}
2622 
2623 	VOP_UNLOCK(vp, 0);
2624 	if (error) {
2625 		vrele(vp);
2626 		free((caddr_t)rbuf, M_TEMP);
2627 		if (cookies)
2628 			free((caddr_t)cookies, M_TEMP);
2629 		nfsm_reply(NFSX_POSTOPATTR(v3));
2630 		nfsm_srvpostop_attr(getret, &at);
2631 		return (0);
2632 	}
2633 	if (io.uio_resid) {
2634 		siz -= io.uio_resid;
2635 
2636 		/*
2637 		 * If nothing read, return eof
2638 		 * rpc reply
2639 		 */
2640 		if (siz == 0) {
2641 			vrele(vp);
2642 			nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2643 				2 * NFSX_UNSIGNED);
2644 			if (v3) {
2645 				nfsm_srvpostop_attr(getret, &at);
2646 				nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2647 				txdr_hyper(at.va_filerev, tl);
2648 				tl += 2;
2649 			} else
2650 				nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2651 			*tl++ = nfs_false;
2652 			*tl = nfs_true;
2653 			free((caddr_t)rbuf, M_TEMP);
2654 			free((caddr_t)cookies, M_TEMP);
2655 			return (0);
2656 		}
2657 	}
2658 
2659 	/*
2660 	 * Check for degenerate cases of nothing useful read.
2661 	 * If so go try again
2662 	 */
2663 	cpos = rbuf;
2664 	cend = rbuf + siz;
2665 	dp = (struct dirent *)cpos;
2666 	cookiep = cookies;
2667 
2668 	while (cpos < cend && ncookies > 0 &&
2669 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2670 		cpos += dp->d_reclen;
2671 		dp = (struct dirent *)cpos;
2672 		cookiep++;
2673 		ncookies--;
2674 	}
2675 	if (cpos >= cend || ncookies == 0) {
2676 		toff = off;
2677 		siz = fullsiz;
2678 		goto again;
2679 	}
2680 
2681 	len = 3 * NFSX_UNSIGNED;	/* paranoia, probably can be 0 */
2682 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2683 	if (v3) {
2684 		nfsm_srvpostop_attr(getret, &at);
2685 		nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2686 		txdr_hyper(at.va_filerev, tl);
2687 	}
2688 	mp = mp2 = mb;
2689 	bp = bpos;
2690 	be = bp + M_TRAILINGSPACE(mp);
2691 
2692 	/* Loop through the records and build reply */
2693 	while (cpos < cend && ncookies > 0) {
2694 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2695 			nlen = dp->d_namlen;
2696 			rem = nfsm_rndup(nlen)-nlen;
2697 			len += (4 * NFSX_UNSIGNED + nlen + rem);
2698 			if (v3)
2699 				len += 2 * NFSX_UNSIGNED;
2700 			if (len > cnt) {
2701 				eofflag = 0;
2702 				break;
2703 			}
2704 			/*
2705 			 * Build the directory record xdr from
2706 			 * the dirent entry.
2707 			 */
2708 			nfsm_clget;
2709 			*tl = nfs_true;
2710 			bp += NFSX_UNSIGNED;
2711 			if (v3) {
2712 				nfsm_clget;
2713 				*tl = 0;
2714 				bp += NFSX_UNSIGNED;
2715 			}
2716 			nfsm_clget;
2717 			*tl = txdr_unsigned(dp->d_fileno);
2718 			bp += NFSX_UNSIGNED;
2719 			nfsm_clget;
2720 			*tl = txdr_unsigned(nlen);
2721 			bp += NFSX_UNSIGNED;
2722 
2723 			/* And loop around copying the name */
2724 			xfer = nlen;
2725 			cp = dp->d_name;
2726 			while (xfer > 0) {
2727 				nfsm_clget;
2728 				if ((bp+xfer) > be)
2729 					tsiz = be-bp;
2730 				else
2731 					tsiz = xfer;
2732 				memcpy(bp, cp, tsiz);
2733 				bp += tsiz;
2734 				xfer -= tsiz;
2735 				if (xfer > 0)
2736 					cp += tsiz;
2737 			}
2738 			/* And null pad to an int32_t boundary */
2739 			for (i = 0; i < rem; i++)
2740 				*bp++ = '\0';
2741 			nfsm_clget;
2742 
2743 			/* Finish off the record */
2744 			txdr_hyper(*cookiep, &jar);
2745 			if (v3) {
2746 				*tl = jar.nfsuquad[0];
2747 				bp += NFSX_UNSIGNED;
2748 				nfsm_clget;
2749 			}
2750 			*tl = jar.nfsuquad[1];
2751 			bp += NFSX_UNSIGNED;
2752 		}
2753 		cpos += dp->d_reclen;
2754 		dp = (struct dirent *)cpos;
2755 		cookiep++;
2756 		ncookies--;
2757 	}
2758 	vrele(vp);
2759 	nfsm_clget;
2760 	*tl = nfs_false;
2761 	bp += NFSX_UNSIGNED;
2762 	nfsm_clget;
2763 	if (eofflag)
2764 		*tl = nfs_true;
2765 	else
2766 		*tl = nfs_false;
2767 	bp += NFSX_UNSIGNED;
2768 	if (mp != mb) {
2769 		if (bp < be)
2770 			mp->m_len = bp - mtod(mp, caddr_t);
2771 	} else
2772 		mp->m_len += bp - bpos;
2773 	free((caddr_t)rbuf, M_TEMP);
2774 	free((caddr_t)cookies, M_TEMP);
2775 	nfsm_srvdone;
2776 }
2777 
2778 int
2779 nfsrv_readdirplus(nfsd, slp, procp, mrq)
2780 	struct nfsrv_descript *nfsd;
2781 	struct nfssvc_sock *slp;
2782 	struct proc *procp;
2783 	struct mbuf **mrq;
2784 {
2785 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2786 	struct mbuf *nam = nfsd->nd_nam;
2787 	caddr_t dpos = nfsd->nd_dpos;
2788 	struct ucred *cred = &nfsd->nd_cr;
2789 	char *bp, *be;
2790 	struct mbuf *mp;
2791 	struct dirent *dp;
2792 	caddr_t cp;
2793 	u_int32_t *tl;
2794 	int32_t t1;
2795 	caddr_t bpos;
2796 	struct mbuf *mb, *mreq, *mp2;
2797 	char *cpos, *cend, *cp2, *rbuf;
2798 	struct vnode *vp, *nvp;
2799 	struct flrep fl;
2800 	nfsfh_t nfh;
2801 	fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
2802 	struct uio io;
2803 	struct iovec iv;
2804 	struct vattr va, at, *vap = &va;
2805 	struct nfs_fattr *fp;
2806 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2807 	int siz, cnt, fullsiz, eofflag, rdonly, cache, dirlen, ncookies;
2808 	u_quad_t frev, off, toff, verf;
2809 	off_t *cookies = NULL, *cookiep;
2810 
2811 	fhp = &nfh.fh_generic;
2812 	nfsm_srvmtofh(fhp);
2813 	nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2814 	toff = fxdr_hyper(tl);
2815 	tl += 2;
2816 	verf = fxdr_hyper(tl);
2817 	tl += 2;
2818 	siz = fxdr_unsigned(int, *tl++);
2819 	cnt = fxdr_unsigned(int, *tl);
2820 	off = toff;
2821 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2822 	xfer = NFS_SRVMAXDATA(nfsd);
2823 	if (siz > xfer)
2824 		siz = xfer;
2825 	fullsiz = siz;
2826 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
2827 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2828 	if (!error && vp->v_type != VDIR) {
2829 		error = ENOTDIR;
2830 		vput(vp);
2831 	}
2832 	if (error) {
2833 		nfsm_reply(NFSX_UNSIGNED);
2834 		nfsm_srvpostop_attr(getret, &at);
2835 		return (0);
2836 	}
2837 	error = getret = VOP_GETATTR(vp, &at, cred, procp);
2838 #ifdef NFS3_STRICTVERF
2839 	/*
2840 	 * XXX This check is too strict for Solaris 2.5 clients.
2841 	 */
2842 	if (!error && toff && verf != at.va_filerev)
2843 		error = NFSERR_BAD_COOKIE;
2844 #endif
2845 	if (!error) {
2846 		nqsrv_getl(vp, ND_READ);
2847 		error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0);
2848 	}
2849 	if (error) {
2850 		vput(vp);
2851 		nfsm_reply(NFSX_V3POSTOPATTR);
2852 		nfsm_srvpostop_attr(getret, &at);
2853 		return (0);
2854 	}
2855 	VOP_UNLOCK(vp, 0);
2856 
2857 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2858 again:
2859 	iv.iov_base = rbuf;
2860 	iv.iov_len = fullsiz;
2861 	io.uio_iov = &iv;
2862 	io.uio_iovcnt = 1;
2863 	io.uio_offset = (off_t)off;
2864 	io.uio_resid = fullsiz;
2865 	io.uio_segflg = UIO_SYSSPACE;
2866 	io.uio_rw = UIO_READ;
2867 	io.uio_procp = (struct proc *)0;
2868 	eofflag = 0;
2869 
2870 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2871 
2872 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2873 
2874 	off = (u_quad_t)io.uio_offset;
2875 	getret = VOP_GETATTR(vp, &at, cred, procp);
2876 
2877 	VOP_UNLOCK(vp, 0);
2878 
2879 	/*
2880 	 * If the VGET operation doesn't work for this filesystem,
2881 	 * we can't support readdirplus. Returning NOTSUPP should
2882 	 * make clients fall back to plain readdir.
2883 	 * There's no need to check for VPTOFH as well, we wouldn't
2884 	 * even be here otherwise.
2885 	 */
2886 	if (!getret) {
2887 		if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp)))
2888 			getret = (getret == EOPNOTSUPP) ?
2889 				NFSERR_NOTSUPP : NFSERR_IO;
2890 		else
2891 			vput(nvp);
2892 	}
2893 
2894 	if (!cookies && !error)
2895 		error = NFSERR_PERM;
2896 	if (!error)
2897 		error = getret;
2898 	if (error) {
2899 		vrele(vp);
2900 		if (cookies)
2901 			free((caddr_t)cookies, M_TEMP);
2902 		free((caddr_t)rbuf, M_TEMP);
2903 		nfsm_reply(NFSX_V3POSTOPATTR);
2904 		nfsm_srvpostop_attr(getret, &at);
2905 		return (0);
2906 	}
2907 	if (io.uio_resid) {
2908 		siz -= io.uio_resid;
2909 
2910 		/*
2911 		 * If nothing read, return eof
2912 		 * rpc reply
2913 		 */
2914 		if (siz == 0) {
2915 			vrele(vp);
2916 			nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
2917 				2 * NFSX_UNSIGNED);
2918 			nfsm_srvpostop_attr(getret, &at);
2919 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2920 			txdr_hyper(at.va_filerev, tl);
2921 			tl += 2;
2922 			*tl++ = nfs_false;
2923 			*tl = nfs_true;
2924 			free((caddr_t)cookies, M_TEMP);
2925 			free((caddr_t)rbuf, M_TEMP);
2926 			return (0);
2927 		}
2928 	}
2929 
2930 	/*
2931 	 * Check for degenerate cases of nothing useful read.
2932 	 * If so go try again
2933 	 */
2934 	cpos = rbuf;
2935 	cend = rbuf + siz;
2936 	dp = (struct dirent *)cpos;
2937 	cookiep = cookies;
2938 
2939 	while (cpos < cend && ncookies > 0 &&
2940 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2941 		cpos += dp->d_reclen;
2942 		dp = (struct dirent *)cpos;
2943 		cookiep++;
2944 		ncookies--;
2945 	}
2946 	if (cpos >= cend || ncookies == 0) {
2947 		toff = off;
2948 		siz = fullsiz;
2949 		goto again;
2950 	}
2951 
2952 	dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
2953 	nfsm_reply(cnt);
2954 	nfsm_srvpostop_attr(getret, &at);
2955 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2956 	txdr_hyper(at.va_filerev, tl);
2957 	mp = mp2 = mb;
2958 	bp = bpos;
2959 	be = bp + M_TRAILINGSPACE(mp);
2960 
2961 	/* Loop through the records and build reply */
2962 	while (cpos < cend && ncookies > 0) {
2963 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2964 			nlen = dp->d_namlen;
2965 			rem = nfsm_rndup(nlen)-nlen;
2966 
2967 			/*
2968 			 * For readdir_and_lookup get the vnode using
2969 			 * the file number.
2970 			 */
2971 			if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp))
2972 				goto invalid;
2973 			memset((caddr_t)nfhp, 0, NFSX_V3FH);
2974 			nfhp->fh_fsid =
2975 				nvp->v_mount->mnt_stat.f_fsid;
2976 			if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) {
2977 				vput(nvp);
2978 				goto invalid;
2979 			}
2980 			if (VOP_GETATTR(nvp, vap, cred, procp)) {
2981 				vput(nvp);
2982 				goto invalid;
2983 			}
2984 			vput(nvp);
2985 
2986 			/*
2987 			 * If either the dircount or maxcount will be
2988 			 * exceeded, get out now. Both of these lengths
2989 			 * are calculated conservatively, including all
2990 			 * XDR overheads.
2991 			 */
2992 			len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
2993 				NFSX_V3POSTOPATTR);
2994 			dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
2995 			if (len > cnt || dirlen > fullsiz) {
2996 				eofflag = 0;
2997 				break;
2998 			}
2999 
3000 			/*
3001 			 * Build the directory record xdr from
3002 			 * the dirent entry.
3003 			 */
3004 			fp = (struct nfs_fattr *)&fl.fl_fattr;
3005 			nfsm_srvfillattr(vap, fp);
3006 			fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3007 			fl.fl_fhok = nfs_true;
3008 			fl.fl_postopok = nfs_true;
3009 			txdr_hyper(*cookiep, fl.fl_off.nfsuquad);
3010 
3011 			nfsm_clget;
3012 			*tl = nfs_true;
3013 			bp += NFSX_UNSIGNED;
3014 			nfsm_clget;
3015 			*tl = 0;
3016 			bp += NFSX_UNSIGNED;
3017 			nfsm_clget;
3018 			*tl = txdr_unsigned(dp->d_fileno);
3019 			bp += NFSX_UNSIGNED;
3020 			nfsm_clget;
3021 			*tl = txdr_unsigned(nlen);
3022 			bp += NFSX_UNSIGNED;
3023 
3024 			/* And loop around copying the name */
3025 			xfer = nlen;
3026 			cp = dp->d_name;
3027 			while (xfer > 0) {
3028 				nfsm_clget;
3029 				if ((bp + xfer) > be)
3030 					tsiz = be - bp;
3031 				else
3032 					tsiz = xfer;
3033 				memcpy(bp, cp, tsiz);
3034 				bp += tsiz;
3035 				xfer -= tsiz;
3036 				if (xfer > 0)
3037 					cp += tsiz;
3038 			}
3039 			/* And null pad to an int32_t boundary */
3040 			for (i = 0; i < rem; i++)
3041 				*bp++ = '\0';
3042 
3043 			/*
3044 			 * Now copy the flrep structure out.
3045 			 */
3046 			xfer = sizeof (struct flrep);
3047 			cp = (caddr_t)&fl;
3048 			while (xfer > 0) {
3049 				nfsm_clget;
3050 				if ((bp + xfer) > be)
3051 					tsiz = be - bp;
3052 				else
3053 					tsiz = xfer;
3054 				memcpy(bp, cp, tsiz);
3055 				bp += tsiz;
3056 				xfer -= tsiz;
3057 				if (xfer > 0)
3058 					cp += tsiz;
3059 			}
3060 		}
3061 invalid:
3062 		cpos += dp->d_reclen;
3063 		dp = (struct dirent *)cpos;
3064 		cookiep++;
3065 		ncookies--;
3066 	}
3067 	vrele(vp);
3068 	nfsm_clget;
3069 	*tl = nfs_false;
3070 	bp += NFSX_UNSIGNED;
3071 	nfsm_clget;
3072 	if (eofflag)
3073 		*tl = nfs_true;
3074 	else
3075 		*tl = nfs_false;
3076 	bp += NFSX_UNSIGNED;
3077 	if (mp != mb) {
3078 		if (bp < be)
3079 			mp->m_len = bp - mtod(mp, caddr_t);
3080 	} else
3081 		mp->m_len += bp - bpos;
3082 	free((caddr_t)cookies, M_TEMP);
3083 	free((caddr_t)rbuf, M_TEMP);
3084 	nfsm_srvdone;
3085 }
3086 
3087 /*
3088  * nfs commit service
3089  */
3090 int
3091 nfsrv_commit(nfsd, slp, procp, mrq)
3092 	struct nfsrv_descript *nfsd;
3093 	struct nfssvc_sock *slp;
3094 	struct proc *procp;
3095 	struct mbuf **mrq;
3096 {
3097 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3098 	struct mbuf *nam = nfsd->nd_nam;
3099 	caddr_t dpos = nfsd->nd_dpos;
3100 	struct ucred *cred = &nfsd->nd_cr;
3101 	struct vattr bfor, aft;
3102 	struct vnode *vp;
3103 	nfsfh_t nfh;
3104 	fhandle_t *fhp;
3105 	u_int32_t *tl;
3106 	int32_t t1;
3107 	caddr_t bpos;
3108 	int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt, cache;
3109 	char *cp2;
3110 	struct mbuf *mb, *mreq;
3111 	u_quad_t frev, off, end;
3112 
3113 #ifndef nolint
3114 	cache = 0;
3115 #endif
3116 	fhp = &nfh.fh_generic;
3117 	nfsm_srvmtofh(fhp);
3118 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3119 
3120 	off = fxdr_hyper(tl);
3121 	tl += 2;
3122 	cnt = fxdr_unsigned(int, *tl);
3123 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3124 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3125 	if (error) {
3126 		nfsm_reply(2 * NFSX_UNSIGNED);
3127 		nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3128 		return (0);
3129 	}
3130 	for_ret = VOP_GETATTR(vp, &bfor, cred, procp);
3131 	end = (cnt > 0) ? off + cnt : vp->v_size;
3132 	if (end < off || end > vp->v_size)
3133 		end = vp->v_size;
3134 	if (off < vp->v_size)
3135 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end, procp);
3136 	/* else error == 0, from nfsrv_fhtovp() */
3137 	aft_ret = VOP_GETATTR(vp, &aft, cred, procp);
3138 	vput(vp);
3139 	nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3140 	nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3141 	if (!error) {
3142 		nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3143 		*tl++ = txdr_unsigned(boottime.tv_sec);
3144 		*tl = txdr_unsigned(boottime.tv_usec);
3145 	} else
3146 		return (0);
3147 	nfsm_srvdone;
3148 }
3149 
3150 /*
3151  * nfs statfs service
3152  */
3153 int
3154 nfsrv_statfs(nfsd, slp, procp, mrq)
3155 	struct nfsrv_descript *nfsd;
3156 	struct nfssvc_sock *slp;
3157 	struct proc *procp;
3158 	struct mbuf **mrq;
3159 {
3160 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3161 	struct mbuf *nam = nfsd->nd_nam;
3162 	caddr_t dpos = nfsd->nd_dpos;
3163 	struct ucred *cred = &nfsd->nd_cr;
3164 	struct statfs *sf;
3165 	struct nfs_statfs *sfp;
3166 	u_int32_t *tl;
3167 	int32_t t1;
3168 	caddr_t bpos;
3169 	int error = 0, rdonly, cache, getret = 1;
3170 	int v3 = (nfsd->nd_flag & ND_NFSV3);
3171 	char *cp2;
3172 	struct mbuf *mb, *mreq;
3173 	struct vnode *vp;
3174 	struct vattr at;
3175 	nfsfh_t nfh;
3176 	fhandle_t *fhp;
3177 	struct statfs statfs;
3178 	u_quad_t frev, tval;
3179 
3180 #ifndef nolint
3181 	cache = 0;
3182 #endif
3183 	fhp = &nfh.fh_generic;
3184 	nfsm_srvmtofh(fhp);
3185 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3186 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3187 	if (error) {
3188 		nfsm_reply(NFSX_UNSIGNED);
3189 		nfsm_srvpostop_attr(getret, &at);
3190 		return (0);
3191 	}
3192 	sf = &statfs;
3193 	error = VFS_STATFS(vp->v_mount, sf, procp);
3194 	getret = VOP_GETATTR(vp, &at, cred, procp);
3195 	vput(vp);
3196 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3197 	if (v3)
3198 		nfsm_srvpostop_attr(getret, &at);
3199 	if (error)
3200 		return (0);
3201 	nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3202 	if (v3) {
3203 		tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_bsize);
3204 		txdr_hyper(tval, &sfp->sf_tbytes);
3205 		tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_bsize);
3206 		txdr_hyper(tval, &sfp->sf_fbytes);
3207 		tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_bsize);
3208 		txdr_hyper(tval, &sfp->sf_abytes);
3209 		tval = (u_quad_t)sf->f_files;
3210 		txdr_hyper(tval, &sfp->sf_tfiles);
3211 		tval = (u_quad_t)sf->f_ffree;
3212 		txdr_hyper(tval, &sfp->sf_ffiles);
3213 		txdr_hyper(tval, &sfp->sf_afiles);
3214 		sfp->sf_invarsec = 0;
3215 	} else {
3216 		sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3217 		sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
3218 		sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3219 		sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3220 		sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3221 	}
3222 	nfsm_srvdone;
3223 }
3224 
3225 /*
3226  * nfs fsinfo service
3227  */
3228 int
3229 nfsrv_fsinfo(nfsd, slp, procp, mrq)
3230 	struct nfsrv_descript *nfsd;
3231 	struct nfssvc_sock *slp;
3232 	struct proc *procp;
3233 	struct mbuf **mrq;
3234 {
3235 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3236 	struct mbuf *nam = nfsd->nd_nam;
3237 	caddr_t dpos = nfsd->nd_dpos;
3238 	struct ucred *cred = &nfsd->nd_cr;
3239 	u_int32_t *tl;
3240 	struct nfsv3_fsinfo *sip;
3241 	int32_t t1;
3242 	caddr_t bpos;
3243 	int error = 0, rdonly, cache, getret = 1;
3244 	uint32_t maxdata;
3245 	char *cp2;
3246 	struct mbuf *mb, *mreq;
3247 	struct vnode *vp;
3248 	struct vattr at;
3249 	nfsfh_t nfh;
3250 	fhandle_t *fhp;
3251 	u_quad_t frev, maxfsize;
3252 	struct statfs sb;
3253 
3254 #ifndef nolint
3255 	cache = 0;
3256 #endif
3257 	fhp = &nfh.fh_generic;
3258 	nfsm_srvmtofh(fhp);
3259 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3260 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3261 	if (error) {
3262 		nfsm_reply(NFSX_UNSIGNED);
3263 		nfsm_srvpostop_attr(getret, &at);
3264 		return (0);
3265 	}
3266 
3267 	/* XXX Try to make a guess on the max file size. */
3268 	VFS_STATFS(vp->v_mount, &sb, (struct proc *)0);
3269 	maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
3270 
3271 	getret = VOP_GETATTR(vp, &at, cred, procp);
3272 	vput(vp);
3273 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3274 	nfsm_srvpostop_attr(getret, &at);
3275 	nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3276 
3277 	/*
3278 	 * XXX
3279 	 * There should be file system VFS OP(s) to get this information.
3280 	 * For now, assume ufs.
3281 	 */
3282 	if (slp->ns_so->so_type == SOCK_DGRAM)
3283 		maxdata = NFS_MAXDGRAMDATA;
3284 	else
3285 		maxdata = NFS_MAXDATA;
3286 	sip->fs_rtmax = txdr_unsigned(maxdata);
3287 	sip->fs_rtpref = txdr_unsigned(maxdata);
3288 	sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3289 	sip->fs_wtmax = txdr_unsigned(maxdata);
3290 	sip->fs_wtpref = txdr_unsigned(maxdata);
3291 	sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3292 	sip->fs_dtpref = txdr_unsigned(maxdata);
3293 	txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3294 	sip->fs_timedelta.nfsv3_sec = 0;
3295 	sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3296 	sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3297 		NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3298 		NFSV3FSINFO_CANSETTIME);
3299 	nfsm_srvdone;
3300 }
3301 
3302 /*
3303  * nfs pathconf service
3304  */
3305 int
3306 nfsrv_pathconf(nfsd, slp, procp, mrq)
3307 	struct nfsrv_descript *nfsd;
3308 	struct nfssvc_sock *slp;
3309 	struct proc *procp;
3310 	struct mbuf **mrq;
3311 {
3312 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3313 	struct mbuf *nam = nfsd->nd_nam;
3314 	caddr_t dpos = nfsd->nd_dpos;
3315 	struct ucred *cred = &nfsd->nd_cr;
3316 	u_int32_t *tl;
3317 	struct nfsv3_pathconf *pc;
3318 	int32_t t1;
3319 	caddr_t bpos;
3320 	int error = 0, rdonly, cache, getret = 1;
3321 	register_t linkmax, namemax, chownres, notrunc;
3322 	char *cp2;
3323 	struct mbuf *mb, *mreq;
3324 	struct vnode *vp;
3325 	struct vattr at;
3326 	nfsfh_t nfh;
3327 	fhandle_t *fhp;
3328 	u_quad_t frev;
3329 
3330 #ifndef nolint
3331 	cache = 0;
3332 #endif
3333 	fhp = &nfh.fh_generic;
3334 	nfsm_srvmtofh(fhp);
3335 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3336 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3337 	if (error) {
3338 		nfsm_reply(NFSX_UNSIGNED);
3339 		nfsm_srvpostop_attr(getret, &at);
3340 		return (0);
3341 	}
3342 	error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3343 	if (!error)
3344 		error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3345 	if (!error)
3346 		error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3347 	if (!error)
3348 		error = VOP_PATHCONF(vp, _PC_NO_TRUNC, &notrunc);
3349 	getret = VOP_GETATTR(vp, &at, cred, procp);
3350 	vput(vp);
3351 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3352 	nfsm_srvpostop_attr(getret, &at);
3353 	if (error)
3354 		return (0);
3355 	nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3356 
3357 	pc->pc_linkmax = txdr_unsigned(linkmax);
3358 	pc->pc_namemax = txdr_unsigned(namemax);
3359 	pc->pc_notrunc = txdr_unsigned(notrunc);
3360 	pc->pc_chownrestricted = txdr_unsigned(chownres);
3361 
3362 	/*
3363 	 * These should probably be supported by VOP_PATHCONF(), but
3364 	 * until msdosfs is exportable (why would you want to?), the
3365 	 * Unix defaults should be ok.
3366 	 */
3367 	pc->pc_caseinsensitive = nfs_false;
3368 	pc->pc_casepreserving = nfs_true;
3369 	nfsm_srvdone;
3370 }
3371 
3372 /*
3373  * Null operation, used by clients to ping server
3374  */
3375 /* ARGSUSED */
3376 int
3377 nfsrv_null(nfsd, slp, procp, mrq)
3378 	struct nfsrv_descript *nfsd;
3379 	struct nfssvc_sock *slp;
3380 	struct proc *procp;
3381 	struct mbuf **mrq;
3382 {
3383 	struct mbuf *mrep = nfsd->nd_mrep;
3384 	caddr_t bpos;
3385 	int error = NFSERR_RETVOID, cache = 0;
3386 	struct mbuf *mb, *mreq;
3387 	u_quad_t frev;
3388 
3389 	nfsm_reply(0);
3390 	return (0);
3391 }
3392 
3393 /*
3394  * No operation, used for obsolete procedures
3395  */
3396 /* ARGSUSED */
3397 int
3398 nfsrv_noop(nfsd, slp, procp, mrq)
3399 	struct nfsrv_descript *nfsd;
3400 	struct nfssvc_sock *slp;
3401 	struct proc *procp;
3402 	struct mbuf **mrq;
3403 {
3404 	struct mbuf *mrep = nfsd->nd_mrep;
3405 	caddr_t bpos;
3406 	int error, cache = 0;
3407 	struct mbuf *mb, *mreq;
3408 	u_quad_t frev;
3409 
3410 	if (nfsd->nd_repstat)
3411 		error = nfsd->nd_repstat;
3412 	else
3413 		error = EPROCUNAVAIL;
3414 	nfsm_reply(0);
3415 	return (0);
3416 }
3417 
3418 /*
3419  * Perform access checking for vnodes obtained from file handles that would
3420  * refer to files already opened by a Unix client. You cannot just use
3421  * vn_writechk() and VOP_ACCESS() for two reasons.
3422  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3423  * 2 - The owner is to be given access irrespective of mode bits for some
3424  *     operations, so that processes that chmod after opening a file don't
3425  *     break. I don't like this because it opens a security hole, but since
3426  *     the nfs server opens a security hole the size of a barn door anyhow,
3427  *     what the heck.
3428  *
3429  * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3430  * will return EPERM instead of EACCESS. EPERM is always an error.
3431  */
3432 int
3433 nfsrv_access(vp, flags, cred, rdonly, p, override)
3434 	struct vnode *vp;
3435 	int flags;
3436 	struct ucred *cred;
3437 	int rdonly;
3438 	struct proc *p;
3439 {
3440 	struct vattr vattr;
3441 	int error;
3442 	if (flags & VWRITE) {
3443 		/* Just vn_writechk() changed to check rdonly */
3444 		/*
3445 		 * Disallow write attempts on read-only file systems;
3446 		 * unless the file is a socket or a block or character
3447 		 * device resident on the file system.
3448 		 */
3449 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3450 			switch (vp->v_type) {
3451 			case VREG:
3452 			case VDIR:
3453 			case VLNK:
3454 				return (EROFS);
3455 			default:
3456 				break;
3457 			}
3458 		}
3459 
3460 		/*
3461 		 * If the vnode is in use as a process's text,
3462 		 * we can't allow writing.
3463 		 */
3464 		if (vp->v_flag & VTEXT)
3465 			return (ETXTBSY);
3466 	}
3467 	error = VOP_GETATTR(vp, &vattr, cred, p);
3468 	if (error)
3469 		return (error);
3470 	error = VOP_ACCESS(vp, flags, cred, p);
3471 	/*
3472 	 * Allow certain operations for the owner (reads and writes
3473 	 * on files that are already open).
3474 	 */
3475 	if (override && error == EACCES && cred->cr_uid == vattr.va_uid)
3476 		error = 0;
3477 	return error;
3478 }
3479