xref: /netbsd/sys/nfs/nfs_serv.c (revision bf9ec67e)
1 /*	$NetBSD: nfs_serv.c,v 1.62 2001/11/10 10:59:09 lukem Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)nfs_serv.c	8.8 (Berkeley) 7/31/95
39  */
40 
41 /*
42  * nfs version 2 and 3 server calls to vnode ops
43  * - these routines generally have 3 phases
44  *   1 - break down and validate rpc request in mbuf list
45  *   2 - do the vnode ops for the request
46  *       (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
47  *   3 - build the rpc reply in an mbuf list
48  *   nb:
49  *	- do not mix the phases, since the nfsm_?? macros can return failures
50  *	  on a bad rpc or similar and do not do any vrele() or vput()'s
51  *
52  *      - the nfsm_reply() macro generates an nfs rpc reply with the nfs
53  *	error number iff error != 0 whereas
54  *	returning an error from the server function implies a fatal error
55  *	such as a badly constructed rpc request that should be dropped without
56  *	a reply.
57  *	For Version 3, nfsm_reply() does not return for the error case, since
58  *	most version 3 rpcs return more than the status for error cases.
59  */
60 
61 #include <sys/cdefs.h>
62 __KERNEL_RCSID(0, "$NetBSD: nfs_serv.c,v 1.62 2001/11/10 10:59:09 lukem Exp $");
63 
64 #include <sys/param.h>
65 #include <sys/systm.h>
66 #include <sys/proc.h>
67 #include <sys/file.h>
68 #include <sys/namei.h>
69 #include <sys/vnode.h>
70 #include <sys/mount.h>
71 #include <sys/socket.h>
72 #include <sys/socketvar.h>
73 #include <sys/mbuf.h>
74 #include <sys/dirent.h>
75 #include <sys/stat.h>
76 #include <sys/kernel.h>
77 #include <ufs/ufs/dir.h>
78 
79 #include <uvm/uvm_extern.h>
80 
81 #include <nfs/nfsproto.h>
82 #include <nfs/rpcv2.h>
83 #include <nfs/nfs.h>
84 #include <nfs/xdr_subs.h>
85 #include <nfs/nfsm_subs.h>
86 #include <nfs/nqnfs.h>
87 #include <nfs/nfs_var.h>
88 
89 /* Global vars */
90 extern u_int32_t nfs_xdrneg1;
91 extern u_int32_t nfs_false, nfs_true;
92 extern enum vtype nv3tov_type[8];
93 extern struct nfsstats nfsstats;
94 extern nfstype nfsv2_type[9];
95 extern nfstype nfsv3_type[9];
96 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
97 
98 /*
99  * nfs v3 access service
100  */
101 int
102 nfsrv3_access(nfsd, slp, procp, mrq)
103 	struct nfsrv_descript *nfsd;
104 	struct nfssvc_sock *slp;
105 	struct proc *procp;
106 	struct mbuf **mrq;
107 {
108 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
109 	struct mbuf *nam = nfsd->nd_nam;
110 	caddr_t dpos = nfsd->nd_dpos;
111 	struct ucred *cred = &nfsd->nd_cr;
112 	struct vnode *vp;
113 	nfsfh_t nfh;
114 	fhandle_t *fhp;
115 	u_int32_t *tl;
116 	int32_t t1;
117 	caddr_t bpos;
118 	int error = 0, rdonly, cache = 0, getret;
119 	char *cp2;
120 	struct mbuf *mb, *mreq, *mb2;
121 	struct vattr va;
122 	u_long inmode, testmode, outmode;
123 	u_quad_t frev;
124 
125 	fhp = &nfh.fh_generic;
126 	nfsm_srvmtofh(fhp);
127 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
128 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
129 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
130 	if (error) {
131 		nfsm_reply(NFSX_UNSIGNED);
132 		nfsm_srvpostop_attr(1, (struct vattr *)0);
133 		return (0);
134 	}
135 	inmode = fxdr_unsigned(u_int32_t, *tl);
136 	outmode = 0;
137 	if ((inmode & NFSV3ACCESS_READ) &&
138 	    nfsrv_access(vp, VREAD, cred, rdonly, procp, 0) == 0)
139 		outmode |= NFSV3ACCESS_READ;
140 	if (vp->v_type != VDIR) {
141 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
142 		if (testmode &&
143 		    nfsrv_access(vp, VWRITE, cred, rdonly, procp, 0) == 0)
144 			outmode |= testmode;
145 		if ((inmode & NFSV3ACCESS_EXECUTE) &&
146 		    nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0) == 0)
147 			outmode |= NFSV3ACCESS_EXECUTE;
148 	} else {
149 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
150 		    NFSV3ACCESS_DELETE);
151 		if (testmode &&
152 		    nfsrv_access(vp, VWRITE, cred, rdonly, procp, 0) == 0)
153 			outmode |= testmode;
154 		if ((inmode & NFSV3ACCESS_LOOKUP) &&
155 		    nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0) == 0)
156 			outmode |= NFSV3ACCESS_LOOKUP;
157 	}
158 	getret = VOP_GETATTR(vp, &va, cred, procp);
159 	vput(vp);
160 	nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
161 	nfsm_srvpostop_attr(getret, &va);
162 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
163 	*tl = txdr_unsigned(outmode);
164 	nfsm_srvdone;
165 }
166 
167 /*
168  * nfs getattr service
169  */
170 int
171 nfsrv_getattr(nfsd, slp, procp, mrq)
172 	struct nfsrv_descript *nfsd;
173 	struct nfssvc_sock *slp;
174 	struct proc *procp;
175 	struct mbuf **mrq;
176 {
177 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
178 	struct mbuf *nam = nfsd->nd_nam;
179 	caddr_t dpos = nfsd->nd_dpos;
180 	struct ucred *cred = &nfsd->nd_cr;
181 	struct nfs_fattr *fp;
182 	struct vattr va;
183 	struct vnode *vp;
184 	nfsfh_t nfh;
185 	fhandle_t *fhp;
186 	u_int32_t *tl;
187 	int32_t t1;
188 	caddr_t bpos;
189 	int error = 0, rdonly, cache;
190 	char *cp2;
191 	struct mbuf *mb, *mb2, *mreq;
192 	u_quad_t frev;
193 
194 	fhp = &nfh.fh_generic;
195 	nfsm_srvmtofh(fhp);
196 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
197 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
198 	if (error) {
199 		nfsm_reply(0);
200 		return (0);
201 	}
202 	nqsrv_getl(vp, ND_READ);
203 	error = VOP_GETATTR(vp, &va, cred, procp);
204 	vput(vp);
205 	nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
206 	if (error)
207 		return (0);
208 	nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
209 	nfsm_srvfillattr(&va, fp);
210 	nfsm_srvdone;
211 }
212 
213 /*
214  * nfs setattr service
215  */
216 int
217 nfsrv_setattr(nfsd, slp, procp, mrq)
218 	struct nfsrv_descript *nfsd;
219 	struct nfssvc_sock *slp;
220 	struct proc *procp;
221 	struct mbuf **mrq;
222 {
223 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
224 	struct mbuf *nam = nfsd->nd_nam;
225 	caddr_t dpos = nfsd->nd_dpos;
226 	struct ucred *cred = &nfsd->nd_cr;
227 	struct vattr va, preat;
228 	struct nfsv2_sattr *sp;
229 	struct nfs_fattr *fp;
230 	struct vnode *vp;
231 	nfsfh_t nfh;
232 	fhandle_t *fhp;
233 	u_int32_t *tl;
234 	int32_t t1;
235 	caddr_t bpos;
236 	int error = 0, rdonly, cache, preat_ret = 1, postat_ret = 1;
237 	int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
238 	char *cp2;
239 	struct mbuf *mb, *mb2, *mreq;
240 	u_quad_t frev;
241 	struct timespec guard;
242 
243 	fhp = &nfh.fh_generic;
244 	nfsm_srvmtofh(fhp);
245 	VATTR_NULL(&va);
246 	if (v3) {
247 		nfsm_srvsattr(&va);
248 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
249 		gcheck = fxdr_unsigned(int, *tl);
250 		if (gcheck) {
251 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
252 			fxdr_nfsv3time(tl, &guard);
253 		}
254 	} else {
255 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
256 		/*
257 		 * Nah nah nah nah na nah
258 		 * There is a bug in the Sun client that puts 0xffff in the mode
259 		 * field of sattr when it should put in 0xffffffff. The u_short
260 		 * doesn't sign extend.
261 		 * --> check the low order 2 bytes for 0xffff
262 		 */
263 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
264 			va.va_mode = nfstov_mode(sp->sa_mode);
265 		if (sp->sa_uid != nfs_xdrneg1)
266 			va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
267 		if (sp->sa_gid != nfs_xdrneg1)
268 			va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
269 		if (sp->sa_size != nfs_xdrneg1)
270 			va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
271 		if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) {
272 #ifdef notyet
273 			fxdr_nfsv2time(&sp->sa_atime, &va.va_atime);
274 #else
275 			va.va_atime.tv_sec =
276 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
277 			va.va_atime.tv_nsec = 0;
278 #endif
279 		}
280 		if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1)
281 			fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime);
282 
283 	}
284 
285 	/*
286 	 * Now that we have all the fields, lets do it.
287 	 */
288 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
289 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
290 	if (error) {
291 		nfsm_reply(2 * NFSX_UNSIGNED);
292 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
293 		return (0);
294 	}
295 	nqsrv_getl(vp, ND_WRITE);
296 	if (v3) {
297 		error = preat_ret = VOP_GETATTR(vp, &preat, cred, procp);
298 		if (!error && gcheck &&
299 			(preat.va_ctime.tv_sec != guard.tv_sec ||
300 			 preat.va_ctime.tv_nsec != guard.tv_nsec))
301 			error = NFSERR_NOT_SYNC;
302 		if (error) {
303 			vput(vp);
304 			nfsm_reply(NFSX_WCCDATA(v3));
305 			nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
306 			return (0);
307 		}
308 	}
309 
310 	/*
311 	 * If the size is being changed write acces is required, otherwise
312 	 * just check for a read only file system.
313 	 */
314 	if (va.va_size == ((u_quad_t)((quad_t) -1))) {
315 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
316 			error = EROFS;
317 			goto out;
318 		}
319 	} else {
320 		if (vp->v_type == VDIR) {
321 			error = EISDIR;
322 			goto out;
323 		} else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
324 			procp, 0)) != 0)
325 			goto out;
326 	}
327 	error = VOP_SETATTR(vp, &va, cred, procp);
328 	postat_ret = VOP_GETATTR(vp, &va, cred, procp);
329 	if (!error)
330 		error = postat_ret;
331 out:
332 	vput(vp);
333 	nfsm_reply(NFSX_WCCORFATTR(v3));
334 	if (v3) {
335 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
336 		return (0);
337 	} else {
338 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
339 		nfsm_srvfillattr(&va, fp);
340 	}
341 	nfsm_srvdone;
342 }
343 
344 /*
345  * nfs lookup rpc
346  */
347 int
348 nfsrv_lookup(nfsd, slp, procp, mrq)
349 	struct nfsrv_descript *nfsd;
350 	struct nfssvc_sock *slp;
351 	struct proc *procp;
352 	struct mbuf **mrq;
353 {
354 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
355 	struct mbuf *nam = nfsd->nd_nam;
356 	caddr_t dpos = nfsd->nd_dpos;
357 	struct ucred *cred = &nfsd->nd_cr;
358 	struct nfs_fattr *fp;
359 	struct nameidata nd, ind, *ndp = &nd;
360 	struct vnode *vp, *dirp;
361 	nfsfh_t nfh;
362 	fhandle_t *fhp;
363 	caddr_t cp;
364 	u_int32_t *tl;
365 	int32_t t1;
366 	caddr_t bpos;
367 	int error = 0, cache, len, dirattr_ret = 1;
368 	int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
369 	char *cp2;
370 	struct mbuf *mb, *mb2, *mreq;
371 	struct vattr va, dirattr;
372 	u_quad_t frev;
373 
374 	fhp = &nfh.fh_generic;
375 	nfsm_srvmtofh(fhp);
376 	nfsm_srvnamesiz(len);
377 
378 	pubflag = nfs_ispublicfh(fhp);
379 
380 	nd.ni_cnd.cn_cred = cred;
381 	nd.ni_cnd.cn_nameiop = LOOKUP;
382 	nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART;
383 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
384 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
385 
386 	if (!error && pubflag) {
387 		if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
388 			/*
389 			 * Setup call to lookup() to see if we can find
390 			 * the index file. Arguably, this doesn't belong
391 			 * in a kernel.. Ugh.
392 			 */
393 			ind = nd;
394 			VOP_UNLOCK(nd.ni_vp, 0);
395 			ind.ni_pathlen = strlen(nfs_pub.np_index);
396 			ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
397 			    nfs_pub.np_index;
398 			ind.ni_startdir = nd.ni_vp;
399 			VREF(ind.ni_startdir);
400 			error = lookup(&ind);
401 			if (!error) {
402 				/*
403 				 * Found an index file. Get rid of
404 				 * the old references.
405 				 */
406 				if (dirp)
407 					vrele(dirp);
408 				dirp = nd.ni_vp;
409 				vrele(nd.ni_startdir);
410 				ndp = &ind;
411 			} else
412 				error = 0;
413 		}
414 		/*
415 		 * If the public filehandle was used, check that this lookup
416 		 * didn't result in a filehandle outside the publicly exported
417 		 * filesystem.
418 		 */
419 
420 		if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) {
421 			vput(nd.ni_vp);
422 			error = EPERM;
423 		}
424 	}
425 
426 	if (dirp) {
427 		if (v3)
428 			dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred,
429 				procp);
430 		vrele(dirp);
431 	}
432 
433 	if (error) {
434 		nfsm_reply(NFSX_POSTOPATTR(v3));
435 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
436 		return (0);
437 	}
438 
439 	nqsrv_getl(ndp->ni_startdir, ND_READ);
440 	vrele(ndp->ni_startdir);
441 	PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
442 	vp = ndp->ni_vp;
443 	memset((caddr_t)fhp, 0, sizeof(nfh));
444 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
445 	error = VFS_VPTOFH(vp, &fhp->fh_fid);
446 	if (!error)
447 		error = VOP_GETATTR(vp, &va, cred, procp);
448 	vput(vp);
449 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
450 	if (error) {
451 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
452 		return (0);
453 	}
454 	nfsm_srvfhtom(fhp, v3);
455 	if (v3) {
456 		nfsm_srvpostop_attr(0, &va);
457 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
458 	} else {
459 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
460 		nfsm_srvfillattr(&va, fp);
461 	}
462 	nfsm_srvdone;
463 }
464 
465 /*
466  * nfs readlink service
467  */
468 int
469 nfsrv_readlink(nfsd, slp, procp, mrq)
470 	struct nfsrv_descript *nfsd;
471 	struct nfssvc_sock *slp;
472 	struct proc *procp;
473 	struct mbuf **mrq;
474 {
475 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
476 	struct mbuf *nam = nfsd->nd_nam;
477 	caddr_t dpos = nfsd->nd_dpos;
478 	struct ucred *cred = &nfsd->nd_cr;
479 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
480 	struct iovec *ivp = iv;
481 	struct mbuf *mp;
482 	u_int32_t *tl;
483 	int32_t t1;
484 	caddr_t bpos;
485 	int error = 0, rdonly, cache, i, tlen, len, getret;
486 	int v3 = (nfsd->nd_flag & ND_NFSV3);
487 	char *cp2;
488 	struct mbuf *mb, *mb2, *mp2 = NULL, *mp3 = NULL, *mreq;
489 	struct vnode *vp;
490 	struct vattr attr;
491 	nfsfh_t nfh;
492 	fhandle_t *fhp;
493 	struct uio io, *uiop = &io;
494 	u_quad_t frev;
495 
496 	fhp = &nfh.fh_generic;
497 	nfsm_srvmtofh(fhp);
498 	len = 0;
499 	i = 0;
500 	while (len < NFS_MAXPATHLEN) {
501 		MGET(mp, M_WAIT, MT_DATA);
502 		MCLGET(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 	if (uiop->uio_resid > 0) {
556 		len -= uiop->uio_resid;
557 		tlen = nfsm_rndup(len);
558 		nfsm_adj(mp3, NFS_MAXPATHLEN-tlen, tlen-len);
559 	}
560 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
561 	*tl = txdr_unsigned(len);
562 	mb->m_next = mp3;
563 	nfsm_srvdone;
564 }
565 
566 /*
567  * nfs read service
568  */
569 int
570 nfsrv_read(nfsd, slp, procp, mrq)
571 	struct nfsrv_descript *nfsd;
572 	struct nfssvc_sock *slp;
573 	struct proc *procp;
574 	struct mbuf **mrq;
575 {
576 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
577 	struct mbuf *nam = nfsd->nd_nam;
578 	caddr_t dpos = nfsd->nd_dpos;
579 	struct ucred *cred = &nfsd->nd_cr;
580 	struct iovec *iv;
581 	struct iovec *iv2;
582 	struct mbuf *m;
583 	struct nfs_fattr *fp;
584 	u_int32_t *tl;
585 	int32_t t1;
586 	int i;
587 	caddr_t bpos;
588 	int error = 0, rdonly, cache, cnt, len, left, siz, tlen, getret;
589 	int v3 = (nfsd->nd_flag & ND_NFSV3), reqlen;
590 	char *cp2;
591 	struct mbuf *mb, *mb2, *mreq;
592 	struct mbuf *m2;
593 	struct vnode *vp;
594 	nfsfh_t nfh;
595 	fhandle_t *fhp;
596 	struct uio io, *uiop = &io;
597 	struct vattr va;
598 	off_t off;
599 	u_quad_t frev;
600 
601 	fhp = &nfh.fh_generic;
602 	nfsm_srvmtofh(fhp);
603 	if (v3) {
604 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
605 		off = fxdr_hyper(tl);
606 	} else {
607 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
608 		off = (off_t)fxdr_unsigned(u_int32_t, *tl);
609 	}
610 	nfsm_srvstrsiz(reqlen, NFS_SRVMAXDATA(nfsd));
611 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
612 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
613 	if (error) {
614 		nfsm_reply(2 * NFSX_UNSIGNED);
615 		nfsm_srvpostop_attr(1, (struct vattr *)0);
616 		return (0);
617 	}
618 	if (vp->v_type != VREG) {
619 		if (v3)
620 			error = EINVAL;
621 		else
622 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
623 	}
624 	if (!error) {
625 	    nqsrv_getl(vp, ND_READ);
626 	    if ((error = nfsrv_access(vp, VREAD, cred, rdonly, procp, 1)) != 0)
627 		error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 1);
628 	}
629 	getret = VOP_GETATTR(vp, &va, cred, procp);
630 	if (!error)
631 		error = getret;
632 	if (error) {
633 		vput(vp);
634 		nfsm_reply(NFSX_POSTOPATTR(v3));
635 		nfsm_srvpostop_attr(getret, &va);
636 		return (0);
637 	}
638 	if (off >= va.va_size)
639 		cnt = 0;
640 	else if ((off + reqlen) > va.va_size)
641 		cnt = nfsm_rndup(va.va_size - off);
642 	else
643 		cnt = reqlen;
644 	nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
645 	if (v3) {
646 		nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
647 		*tl++ = nfs_true;
648 		fp = (struct nfs_fattr *)tl;
649 		tl += (NFSX_V3FATTR / sizeof (u_int32_t));
650 	} else {
651 		nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
652 		fp = (struct nfs_fattr *)tl;
653 		tl += (NFSX_V2FATTR / sizeof (u_int32_t));
654 	}
655 	len = left = cnt;
656 	if (cnt > 0) {
657 		/*
658 		 * Generate the mbuf list with the uio_iov ref. to it.
659 		 */
660 		i = 0;
661 		m = m2 = mb;
662 		while (left > 0) {
663 			siz = min(M_TRAILINGSPACE(m), left);
664 			if (siz > 0) {
665 				left -= siz;
666 				i++;
667 			}
668 			if (left > 0) {
669 				MGET(m, M_WAIT, MT_DATA);
670 				MCLGET(m, M_WAIT);
671 				m->m_len = 0;
672 				m2->m_next = m;
673 				m2 = m;
674 			}
675 		}
676 		iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
677 		uiop->uio_iov = iv2 = iv;
678 		m = mb;
679 		left = cnt;
680 		i = 0;
681 		while (left > 0) {
682 			if (m == NULL)
683 				panic("nfsrv_read iov");
684 			siz = min(M_TRAILINGSPACE(m), left);
685 			if (siz > 0) {
686 				iv->iov_base = mtod(m, caddr_t) + m->m_len;
687 				iv->iov_len = siz;
688 				m->m_len += siz;
689 				left -= siz;
690 				iv++;
691 				i++;
692 			}
693 			m = m->m_next;
694 		}
695 		uiop->uio_iovcnt = i;
696 		uiop->uio_offset = off;
697 		uiop->uio_resid = cnt;
698 		uiop->uio_rw = UIO_READ;
699 		uiop->uio_segflg = UIO_SYSSPACE;
700 		error = VOP_READ(vp, uiop, IO_NODELOCKED, cred);
701 		off = uiop->uio_offset;
702 		free((caddr_t)iv2, M_TEMP);
703 		if (error || (getret = VOP_GETATTR(vp, &va, cred, procp)) != 0){
704 			if (!error)
705 				error = getret;
706 			m_freem(mreq);
707 			vput(vp);
708 			nfsm_reply(NFSX_POSTOPATTR(v3));
709 			nfsm_srvpostop_attr(getret, &va);
710 			return (0);
711 		}
712 	} else
713 		uiop->uio_resid = 0;
714 	vput(vp);
715 	nfsm_srvfillattr(&va, fp);
716 	len -= uiop->uio_resid;
717 	tlen = nfsm_rndup(len);
718 	if (cnt != tlen || tlen != len)
719 		nfsm_adj(mb, cnt - tlen, tlen - len);
720 	if (v3) {
721 		*tl++ = txdr_unsigned(len);
722 		if (len < reqlen)
723 			*tl++ = nfs_true;
724 		else
725 			*tl++ = nfs_false;
726 	}
727 	*tl = txdr_unsigned(len);
728 	nfsm_srvdone;
729 }
730 
731 /*
732  * nfs write service
733  */
734 int
735 nfsrv_write(nfsd, slp, procp, mrq)
736 	struct nfsrv_descript *nfsd;
737 	struct nfssvc_sock *slp;
738 	struct proc *procp;
739 	struct mbuf **mrq;
740 {
741 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
742 	struct mbuf *nam = nfsd->nd_nam;
743 	caddr_t dpos = nfsd->nd_dpos;
744 	struct ucred *cred = &nfsd->nd_cr;
745 	struct iovec *ivp;
746 	int i, cnt;
747 	struct mbuf *mp;
748 	struct nfs_fattr *fp;
749 	struct iovec *iv;
750 	struct vattr va, forat;
751 	u_int32_t *tl;
752 	int32_t t1;
753 	caddr_t bpos;
754 	int error = 0, rdonly, cache, len, forat_ret = 1;
755 	int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
756 	int stable = NFSV3WRITE_FILESYNC;
757 	int v3 = (nfsd->nd_flag & ND_NFSV3);
758 	char *cp2;
759 	struct mbuf *mb, *mb2, *mreq;
760 	struct vnode *vp;
761 	nfsfh_t nfh;
762 	fhandle_t *fhp;
763 	struct uio io, *uiop = &io;
764 	off_t off;
765 	u_quad_t frev;
766 
767 	if (mrep == NULL) {
768 		*mrq = NULL;
769 		return (0);
770 	}
771 	fhp = &nfh.fh_generic;
772 	nfsm_srvmtofh(fhp);
773 	if (v3) {
774 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
775 		off = fxdr_hyper(tl);
776 		tl += 3;
777 		stable = fxdr_unsigned(int, *tl++);
778 	} else {
779 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
780 		off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
781 		tl += 2;
782 	}
783 	retlen = len = fxdr_unsigned(int32_t, *tl);
784 	cnt = i = 0;
785 
786 	/*
787 	 * For NFS Version 2, it is not obvious what a write of zero length
788 	 * should do, but I might as well be consistent with Version 3,
789 	 * which is to return ok so long as there are no permission problems.
790 	 */
791 	if (len > 0) {
792 	    zeroing = 1;
793 	    mp = mrep;
794 	    while (mp) {
795 		if (mp == md) {
796 			zeroing = 0;
797 			adjust = dpos - mtod(mp, caddr_t);
798 			mp->m_len -= adjust;
799 			if (mp->m_len > 0 && adjust > 0)
800 				NFSMADV(mp, adjust);
801 		}
802 		if (zeroing)
803 			mp->m_len = 0;
804 		else if (mp->m_len > 0) {
805 			i += mp->m_len;
806 			if (i > len) {
807 				mp->m_len -= (i - len);
808 				zeroing	= 1;
809 			}
810 			if (mp->m_len > 0)
811 				cnt++;
812 		}
813 		mp = mp->m_next;
814 	    }
815 	}
816 	if (len > NFS_MAXDATA || len < 0 || i < len) {
817 		error = EIO;
818 		nfsm_reply(2 * NFSX_UNSIGNED);
819 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
820 		return (0);
821 	}
822 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
823 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
824 	if (error) {
825 		nfsm_reply(2 * NFSX_UNSIGNED);
826 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
827 		return (0);
828 	}
829 	if (v3)
830 		forat_ret = VOP_GETATTR(vp, &forat, cred, procp);
831 	if (vp->v_type != VREG) {
832 		if (v3)
833 			error = EINVAL;
834 		else
835 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
836 	}
837 	if (!error) {
838 		nqsrv_getl(vp, ND_WRITE);
839 		error = nfsrv_access(vp, VWRITE, cred, rdonly, procp, 1);
840 	}
841 	if (error) {
842 		vput(vp);
843 		nfsm_reply(NFSX_WCCDATA(v3));
844 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
845 		return (0);
846 	}
847 
848 	if (len > 0) {
849 	    ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK);
850 	    uiop->uio_iov = iv = ivp;
851 	    uiop->uio_iovcnt = cnt;
852 	    mp = mrep;
853 	    while (mp) {
854 		if (mp->m_len > 0) {
855 			ivp->iov_base = mtod(mp, caddr_t);
856 			ivp->iov_len = mp->m_len;
857 			ivp++;
858 		}
859 		mp = mp->m_next;
860 	    }
861 
862 	    /*
863 	     * XXX
864 	     * The IO_METASYNC flag indicates that all metadata (and not just
865 	     * enough to ensure data integrity) mus be written to stable storage
866 	     * synchronously.
867 	     * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
868 	     */
869 	    if (stable == NFSV3WRITE_UNSTABLE)
870 		ioflags = IO_NODELOCKED;
871 	    else if (stable == NFSV3WRITE_DATASYNC)
872 		ioflags = (IO_SYNC | IO_NODELOCKED);
873 	    else
874 		ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
875 	    uiop->uio_resid = len;
876 	    uiop->uio_rw = UIO_WRITE;
877 	    uiop->uio_segflg = UIO_SYSSPACE;
878 	    uiop->uio_procp = (struct proc *)0;
879 	    uiop->uio_offset = off;
880 	    error = VOP_WRITE(vp, uiop, ioflags, cred);
881 	    nfsstats.srvvop_writes++;
882 	    free((caddr_t)iv, M_TEMP);
883 	}
884 	aftat_ret = VOP_GETATTR(vp, &va, cred, procp);
885 	vput(vp);
886 	if (!error)
887 		error = aftat_ret;
888 	nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
889 		2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
890 	if (v3) {
891 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
892 		if (error)
893 			return (0);
894 		nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
895 		*tl++ = txdr_unsigned(retlen);
896 		if (stable == NFSV3WRITE_UNSTABLE)
897 			*tl++ = txdr_unsigned(stable);
898 		else
899 			*tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
900 		/*
901 		 * Actually, there is no need to txdr these fields,
902 		 * but it may make the values more human readable,
903 		 * for debugging purposes.
904 		 */
905 		*tl++ = txdr_unsigned(boottime.tv_sec);
906 		*tl = txdr_unsigned(boottime.tv_usec);
907 	} else {
908 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
909 		nfsm_srvfillattr(&va, fp);
910 	}
911 	nfsm_srvdone;
912 }
913 
914 /*
915  * NFS write service with write gathering support. Called when
916  * nfsrvw_procrastinate > 0.
917  * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
918  * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
919  * Jan. 1994.
920  */
921 int
922 nfsrv_writegather(ndp, slp, procp, mrq)
923 	struct nfsrv_descript **ndp;
924 	struct nfssvc_sock *slp;
925 	struct proc *procp;
926 	struct mbuf **mrq;
927 {
928 	struct iovec *ivp;
929 	struct mbuf *mp;
930 	struct nfsrv_descript *wp, *nfsd, *owp, *swp;
931 	struct nfs_fattr *fp;
932 	int i = 0;
933 	struct iovec *iov;
934 	struct nfsrvw_delayhash *wpp;
935 	struct ucred *cred;
936 	struct vattr va, forat;
937 	u_int32_t *tl;
938 	int32_t t1;
939 	caddr_t bpos, dpos;
940 	int error = 0, rdonly, cache, len = 0, forat_ret = 1;
941 	int ioflags, aftat_ret = 1, s, adjust, v3, zeroing;
942 	char *cp2;
943 	struct mbuf *mb, *mb2, *mreq, *mrep, *md;
944 	struct vnode *vp;
945 	struct uio io, *uiop = &io;
946 	u_quad_t frev, cur_usec;
947 
948 	*mrq = NULL;
949 	if (*ndp) {
950 	    nfsd = *ndp;
951 	    *ndp = NULL;
952 	    mrep = nfsd->nd_mrep;
953 	    md = nfsd->nd_md;
954 	    dpos = nfsd->nd_dpos;
955 	    cred = &nfsd->nd_cr;
956 	    v3 = (nfsd->nd_flag & ND_NFSV3);
957 	    LIST_INIT(&nfsd->nd_coalesce);
958 	    nfsd->nd_mreq = NULL;
959 	    nfsd->nd_stable = NFSV3WRITE_FILESYNC;
960 	    cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
961 	    nfsd->nd_time = cur_usec + nfsrvw_procrastinate;
962 
963 	    /*
964 	     * Now, get the write header..
965 	     */
966 	    nfsm_srvmtofh(&nfsd->nd_fh);
967 	    if (v3) {
968 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
969 		nfsd->nd_off = fxdr_hyper(tl);
970 		tl += 3;
971 		nfsd->nd_stable = fxdr_unsigned(int, *tl++);
972 	    } else {
973 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
974 		nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
975 		tl += 2;
976 	    }
977 	    len = fxdr_unsigned(int32_t, *tl);
978 	    nfsd->nd_len = len;
979 	    nfsd->nd_eoff = nfsd->nd_off + len;
980 
981 	    /*
982 	     * Trim the header out of the mbuf list and trim off any trailing
983 	     * junk so that the mbuf list has only the write data.
984 	     */
985 	    zeroing = 1;
986 	    i = 0;
987 	    mp = mrep;
988 	    while (mp) {
989 		if (mp == md) {
990 		    zeroing = 0;
991 		    adjust = dpos - mtod(mp, caddr_t);
992 		    mp->m_len -= adjust;
993 		    if (mp->m_len > 0 && adjust > 0)
994 			NFSMADV(mp, adjust);
995 		}
996 		if (zeroing)
997 		    mp->m_len = 0;
998 		else {
999 		    i += mp->m_len;
1000 		    if (i > len) {
1001 			mp->m_len -= (i - len);
1002 			zeroing = 1;
1003 		    }
1004 		}
1005 		mp = mp->m_next;
1006 	    }
1007 	    if (len > NFS_MAXDATA || len < 0  || i < len) {
1008 nfsmout:
1009 		m_freem(mrep);
1010 		error = EIO;
1011 		nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1012 		if (v3)
1013 		    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1014 		nfsd->nd_mreq = mreq;
1015 		nfsd->nd_mrep = NULL;
1016 		nfsd->nd_time = 0;
1017 	    }
1018 
1019 	    /*
1020 	     * Add this entry to the hash and time queues.
1021 	     */
1022 	    s = splsoftclock();
1023 	    owp = NULL;
1024 	    wp = slp->ns_tq.lh_first;
1025 	    while (wp && wp->nd_time < nfsd->nd_time) {
1026 		owp = wp;
1027 		wp = wp->nd_tq.le_next;
1028 	    }
1029 	    if (owp) {
1030 		LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1031 	    } else {
1032 		LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1033 	    }
1034 	    if (nfsd->nd_mrep) {
1035 		wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1036 		owp = NULL;
1037 		wp = wpp->lh_first;
1038 		while (wp &&
1039 		    memcmp((caddr_t)&nfsd->nd_fh, (caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1040 		    owp = wp;
1041 		    wp = wp->nd_hash.le_next;
1042 		}
1043 		while (wp && wp->nd_off < nfsd->nd_off &&
1044 		    !memcmp((caddr_t)&nfsd->nd_fh, (caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1045 		    owp = wp;
1046 		    wp = wp->nd_hash.le_next;
1047 		}
1048 		if (owp) {
1049 		    LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1050 
1051 		    /*
1052 		     * Search the hash list for overlapping entries and
1053 		     * coalesce.
1054 		     */
1055 		    for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1056 			wp = nfsd->nd_hash.le_next;
1057 			if (NFSW_SAMECRED(owp, nfsd))
1058 			    nfsrvw_coalesce(owp, nfsd);
1059 		    }
1060 		} else {
1061 		    LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1062 		}
1063 	    }
1064 	    splx(s);
1065 	}
1066 
1067 	/*
1068 	 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1069 	 * and generate the associated reply mbuf list(s).
1070 	 */
1071 loop1:
1072 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1073 	s = splsoftclock();
1074 	for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = owp) {
1075 		owp = nfsd->nd_tq.le_next;
1076 		if (nfsd->nd_time > cur_usec)
1077 		    break;
1078 		if (nfsd->nd_mreq)
1079 		    continue;
1080 		LIST_REMOVE(nfsd, nd_tq);
1081 		LIST_REMOVE(nfsd, nd_hash);
1082 		splx(s);
1083 		mrep = nfsd->nd_mrep;
1084 		nfsd->nd_mrep = NULL;
1085 		cred = &nfsd->nd_cr;
1086 		v3 = (nfsd->nd_flag & ND_NFSV3);
1087 		forat_ret = aftat_ret = 1;
1088 		error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1089 		    nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH),
1090 		    FALSE);
1091 		if (!error) {
1092 		    if (v3)
1093 			forat_ret = VOP_GETATTR(vp, &forat, cred, procp);
1094 		    if (vp->v_type != VREG) {
1095 			if (v3)
1096 			    error = EINVAL;
1097 			else
1098 			    error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1099 		    }
1100 		} else
1101 		    vp = NULL;
1102 		if (!error) {
1103 		    nqsrv_getl(vp, ND_WRITE);
1104 		    error = nfsrv_access(vp, VWRITE, cred, rdonly, procp, 1);
1105 		}
1106 
1107 		if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1108 		    ioflags = IO_NODELOCKED;
1109 		else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1110 		    ioflags = (IO_SYNC | IO_NODELOCKED);
1111 		else
1112 		    ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1113 		uiop->uio_rw = UIO_WRITE;
1114 		uiop->uio_segflg = UIO_SYSSPACE;
1115 		uiop->uio_procp = (struct proc *)0;
1116 		uiop->uio_offset = nfsd->nd_off;
1117 		uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1118 		if (uiop->uio_resid > 0) {
1119 		    mp = mrep;
1120 		    i = 0;
1121 		    while (mp) {
1122 			if (mp->m_len > 0)
1123 			    i++;
1124 			mp = mp->m_next;
1125 		    }
1126 		    uiop->uio_iovcnt = i;
1127 		    iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
1128 		    uiop->uio_iov = ivp = iov;
1129 		    mp = mrep;
1130 		    while (mp) {
1131 			if (mp->m_len > 0) {
1132 			    ivp->iov_base = mtod(mp, caddr_t);
1133 			    ivp->iov_len = mp->m_len;
1134 			    ivp++;
1135 			}
1136 			mp = mp->m_next;
1137 		    }
1138 		    if (!error) {
1139 			error = VOP_WRITE(vp, uiop, ioflags, cred);
1140 			nfsstats.srvvop_writes++;
1141 		    }
1142 		    free((caddr_t)iov, M_TEMP);
1143 		}
1144 		m_freem(mrep);
1145 		if (vp) {
1146 		    aftat_ret = VOP_GETATTR(vp, &va, cred, procp);
1147 		    vput(vp);
1148 		}
1149 
1150 		/*
1151 		 * Loop around generating replies for all write rpcs that have
1152 		 * now been completed.
1153 		 */
1154 		swp = nfsd;
1155 		do {
1156 		    if (error) {
1157 			nfsm_writereply(NFSX_WCCDATA(v3), v3);
1158 			if (v3) {
1159 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1160 			}
1161 		    } else {
1162 			nfsm_writereply(NFSX_PREOPATTR(v3) +
1163 			    NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1164 			    NFSX_WRITEVERF(v3), v3);
1165 			if (v3) {
1166 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1167 			    nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1168 			    *tl++ = txdr_unsigned(nfsd->nd_len);
1169 			    *tl++ = txdr_unsigned(swp->nd_stable);
1170 			    /*
1171 			     * Actually, there is no need to txdr these fields,
1172 			     * but it may make the values more human readable,
1173 			     * for debugging purposes.
1174 			     */
1175 			    *tl++ = txdr_unsigned(boottime.tv_sec);
1176 			    *tl = txdr_unsigned(boottime.tv_usec);
1177 			} else {
1178 			    nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1179 			    nfsm_srvfillattr(&va, fp);
1180 			}
1181 		    }
1182 		    nfsd->nd_mreq = mreq;
1183 		    if (nfsd->nd_mrep)
1184 			panic("nfsrv_write: nd_mrep not free");
1185 
1186 		    /*
1187 		     * Done. Put it at the head of the timer queue so that
1188 		     * the final phase can return the reply.
1189 		     */
1190 		    s = splsoftclock();
1191 		    if (nfsd != swp) {
1192 			nfsd->nd_time = 0;
1193 			LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1194 		    }
1195 		    nfsd = swp->nd_coalesce.lh_first;
1196 		    if (nfsd) {
1197 			LIST_REMOVE(nfsd, nd_tq);
1198 		    }
1199 		    splx(s);
1200 		} while (nfsd);
1201 		s = splsoftclock();
1202 		swp->nd_time = 0;
1203 		LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1204 		splx(s);
1205 		goto loop1;
1206 	}
1207 	splx(s);
1208 
1209 	/*
1210 	 * Search for a reply to return.
1211 	 */
1212 	s = splsoftclock();
1213 	for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = nfsd->nd_tq.le_next)
1214 		if (nfsd->nd_mreq) {
1215 		    LIST_REMOVE(nfsd, nd_tq);
1216 		    *mrq = nfsd->nd_mreq;
1217 		    *ndp = nfsd;
1218 		    break;
1219 		}
1220 	splx(s);
1221 	return (0);
1222 }
1223 
1224 /*
1225  * Coalesce the write request nfsd into owp. To do this we must:
1226  * - remove nfsd from the queues
1227  * - merge nfsd->nd_mrep into owp->nd_mrep
1228  * - update the nd_eoff and nd_stable for owp
1229  * - put nfsd on owp's nd_coalesce list
1230  * NB: Must be called at splsoftclock().
1231  */
1232 void
1233 nfsrvw_coalesce(owp, nfsd)
1234         struct nfsrv_descript *owp;
1235         struct nfsrv_descript *nfsd;
1236 {
1237         int overlap;
1238         struct mbuf *mp;
1239 
1240         LIST_REMOVE(nfsd, nd_hash);
1241         LIST_REMOVE(nfsd, nd_tq);
1242         if (owp->nd_eoff < nfsd->nd_eoff) {
1243             overlap = owp->nd_eoff - nfsd->nd_off;
1244             if (overlap < 0)
1245                 panic("nfsrv_coalesce: bad off");
1246             if (overlap > 0)
1247                 m_adj(nfsd->nd_mrep, overlap);
1248             mp = owp->nd_mrep;
1249             while (mp->m_next)
1250                 mp = mp->m_next;
1251             mp->m_next = nfsd->nd_mrep;
1252             owp->nd_eoff = nfsd->nd_eoff;
1253         } else
1254             m_freem(nfsd->nd_mrep);
1255         nfsd->nd_mrep = NULL;
1256         if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1257             owp->nd_stable = NFSV3WRITE_FILESYNC;
1258         else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1259             owp->nd_stable == NFSV3WRITE_UNSTABLE)
1260             owp->nd_stable = NFSV3WRITE_DATASYNC;
1261         LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1262  	/*
1263  	 * nfsd might hold coalesce elements! Move them to owp.
1264  	 * Otherwise, requests may be lost and clients will be stuck.
1265  	 */
1266  	if (nfsd->nd_coalesce.lh_first)
1267  	{
1268  		struct nfsrv_descript *m;
1269 
1270  		while ((m = nfsd->nd_coalesce.lh_first))
1271  		{
1272  			LIST_REMOVE(m, nd_tq);
1273  			LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq);
1274  		}
1275  	}
1276 }
1277 
1278 /*
1279  * nfs create service
1280  * now does a truncate to 0 length via. setattr if it already exists
1281  */
1282 int
1283 nfsrv_create(nfsd, slp, procp, mrq)
1284 	struct nfsrv_descript *nfsd;
1285 	struct nfssvc_sock *slp;
1286 	struct proc *procp;
1287 	struct mbuf **mrq;
1288 {
1289 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1290 	struct mbuf *nam = nfsd->nd_nam;
1291 	caddr_t dpos = nfsd->nd_dpos;
1292 	struct ucred *cred = &nfsd->nd_cr;
1293 	struct nfs_fattr *fp;
1294 	struct vattr va, dirfor, diraft;
1295 	struct nfsv2_sattr *sp;
1296 	u_int32_t *tl;
1297 	struct nameidata nd;
1298 	caddr_t cp;
1299 	int32_t t1;
1300 	caddr_t bpos;
1301 	int error = 0, cache, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1302 	int rdev = 0;
1303 	int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1304 	char *cp2;
1305 	struct mbuf *mb, *mb2, *mreq;
1306 	struct vnode *vp = NULL, *dirp = NULL;
1307 	nfsfh_t nfh;
1308 	fhandle_t *fhp;
1309 	u_quad_t frev, tempsize;
1310 	u_char cverf[NFSX_V3CREATEVERF];
1311 
1312 	nd.ni_cnd.cn_nameiop = 0;
1313 	fhp = &nfh.fh_generic;
1314 	nfsm_srvmtofh(fhp);
1315 	nfsm_srvnamesiz(len);
1316 	nd.ni_cnd.cn_cred = cred;
1317 	nd.ni_cnd.cn_nameiop = CREATE;
1318 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1319 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1320 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1321 	if (dirp) {
1322 		if (v3)
1323 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
1324 				procp);
1325 		else {
1326 			vrele(dirp);
1327 			dirp = (struct vnode *)0;
1328 		}
1329 	}
1330 	if (error) {
1331 		nfsm_reply(NFSX_WCCDATA(v3));
1332 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1333 		if (dirp)
1334 			vrele(dirp);
1335 		return (0);
1336 	}
1337 	VATTR_NULL(&va);
1338 	if (v3) {
1339 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1340 		how = fxdr_unsigned(int, *tl);
1341 		switch (how) {
1342 		case NFSV3CREATE_GUARDED:
1343 			if (nd.ni_vp) {
1344 				error = EEXIST;
1345 				break;
1346 			}
1347 		case NFSV3CREATE_UNCHECKED:
1348 			nfsm_srvsattr(&va);
1349 			break;
1350 		case NFSV3CREATE_EXCLUSIVE:
1351 			nfsm_dissect(cp, caddr_t, NFSX_V3CREATEVERF);
1352 			memcpy(cverf, cp, NFSX_V3CREATEVERF);
1353 			exclusive_flag = 1;
1354 			if (nd.ni_vp == NULL)
1355 				va.va_mode = 0;
1356 			break;
1357 		};
1358 		va.va_type = VREG;
1359 	} else {
1360 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1361 		va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1362 		if (va.va_type == VNON)
1363 			va.va_type = VREG;
1364 		va.va_mode = nfstov_mode(sp->sa_mode);
1365 		switch (va.va_type) {
1366 		case VREG:
1367 			tsize = fxdr_unsigned(int32_t, sp->sa_size);
1368 			if (tsize != -1)
1369 				va.va_size = (u_quad_t)tsize;
1370 			break;
1371 		case VCHR:
1372 		case VBLK:
1373 		case VFIFO:
1374 			rdev = fxdr_unsigned(int32_t, sp->sa_size);
1375 			break;
1376 		default:
1377 			break;
1378 		};
1379 	}
1380 
1381 	/*
1382 	 * Iff doesn't exist, create it
1383 	 * otherwise just truncate to 0 length
1384 	 *   should I set the mode too ??
1385 	 */
1386 	if (nd.ni_vp == NULL) {
1387 		if (va.va_type == VREG || va.va_type == VSOCK) {
1388 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1389 			error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1390 			if (!error) {
1391 				if (exclusive_flag) {
1392 					exclusive_flag = 0;
1393 					VATTR_NULL(&va);
1394 					memcpy((caddr_t)&va.va_atime, cverf,
1395 						NFSX_V3CREATEVERF);
1396 					error = VOP_SETATTR(nd.ni_vp, &va, cred,
1397 						procp);
1398 				}
1399 			}
1400 		} else if (va.va_type == VCHR || va.va_type == VBLK ||
1401 			va.va_type == VFIFO) {
1402 			if (va.va_type == VCHR && rdev == 0xffffffff)
1403 				va.va_type = VFIFO;
1404 			if (va.va_type != VFIFO &&
1405 			    (error = suser(cred, (u_short *)0))) {
1406 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1407 				vput(nd.ni_dvp);
1408 				nfsm_reply(0);
1409 				return (error);
1410 			} else
1411 				va.va_rdev = (dev_t)rdev;
1412 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1413 			error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd,
1414 			    &va);
1415 			if (error) {
1416 				nfsm_reply(0);
1417 			}
1418 			PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1419 			if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1420 				vrele(nd.ni_dvp);
1421 				vput(nd.ni_vp);
1422 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1423 				error = EINVAL;
1424 				nfsm_reply(0);
1425 			}
1426 		} else {
1427 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1428 			vput(nd.ni_dvp);
1429 			error = ENXIO;
1430 		}
1431 		vp = nd.ni_vp;
1432 	} else {
1433 		vp = nd.ni_vp;
1434 		if (nd.ni_dvp == vp)
1435 			vrele(nd.ni_dvp);
1436 		else
1437 			vput(nd.ni_dvp);
1438 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1439 		if (!error && va.va_size != -1) {
1440 			error = nfsrv_access(vp, VWRITE, cred,
1441 			    (nd.ni_cnd.cn_flags & RDONLY), procp, 0);
1442 			if (!error) {
1443 				nqsrv_getl(vp, ND_WRITE);
1444 				tempsize = va.va_size;
1445 				VATTR_NULL(&va);
1446 				va.va_size = tempsize;
1447 				error = VOP_SETATTR(vp, &va, cred,
1448 					 procp);
1449 			}
1450 		}
1451 		if (error)
1452 			vput(vp);
1453 	}
1454 	if (!error) {
1455 		memset((caddr_t)fhp, 0, sizeof(nfh));
1456 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1457 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
1458 		if (!error)
1459 			error = VOP_GETATTR(vp, &va, cred, procp);
1460 		vput(vp);
1461 	}
1462 	if (v3) {
1463 		if (exclusive_flag && !error &&
1464 			memcmp(cverf, (caddr_t)&va.va_atime, NFSX_V3CREATEVERF))
1465 			error = EEXIST;
1466 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1467 		vrele(dirp);
1468 	}
1469 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1470 	if (v3) {
1471 		if (!error) {
1472 			nfsm_srvpostop_fh(fhp);
1473 			nfsm_srvpostop_attr(0, &va);
1474 		}
1475 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1476 	} else {
1477 		nfsm_srvfhtom(fhp, v3);
1478 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1479 		nfsm_srvfillattr(&va, fp);
1480 	}
1481 	return (0);
1482 nfsmout:
1483 	if (dirp)
1484 		vrele(dirp);
1485 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1486 	if (nd.ni_dvp == nd.ni_vp)
1487 		vrele(nd.ni_dvp);
1488 	else
1489 		vput(nd.ni_dvp);
1490 	if (nd.ni_vp)
1491 		vput(nd.ni_vp);
1492 	return (error);
1493 }
1494 
1495 /*
1496  * nfs v3 mknod service
1497  */
1498 int
1499 nfsrv_mknod(nfsd, slp, procp, mrq)
1500 	struct nfsrv_descript *nfsd;
1501 	struct nfssvc_sock *slp;
1502 	struct proc *procp;
1503 	struct mbuf **mrq;
1504 {
1505 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1506 	struct mbuf *nam = nfsd->nd_nam;
1507 	caddr_t dpos = nfsd->nd_dpos;
1508 	struct ucred *cred = &nfsd->nd_cr;
1509 	struct vattr va, dirfor, diraft;
1510 	u_int32_t *tl;
1511 	struct nameidata nd;
1512 	int32_t t1;
1513 	caddr_t bpos;
1514 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
1515 	u_int32_t major, minor;
1516 	enum vtype vtyp;
1517 	char *cp2;
1518 	struct mbuf *mb, *mb2, *mreq;
1519 	struct vnode *vp, *dirp = (struct vnode *)0;
1520 	nfsfh_t nfh;
1521 	fhandle_t *fhp;
1522 	u_quad_t frev;
1523 
1524 	nd.ni_cnd.cn_nameiop = 0;
1525 	fhp = &nfh.fh_generic;
1526 	nfsm_srvmtofh(fhp);
1527 	nfsm_srvnamesiz(len);
1528 	nd.ni_cnd.cn_cred = cred;
1529 	nd.ni_cnd.cn_nameiop = CREATE;
1530 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1531 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1532 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1533 	if (dirp)
1534 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, procp);
1535 	if (error) {
1536 		nfsm_reply(NFSX_WCCDATA(1));
1537 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1538 		if (dirp)
1539 			vrele(dirp);
1540 		return (0);
1541 	}
1542 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1543 	vtyp = nfsv3tov_type(*tl);
1544 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1545 		error = NFSERR_BADTYPE;
1546 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1547 		if (nd.ni_dvp == nd.ni_vp)
1548 			vrele(nd.ni_dvp);
1549 		else
1550 			vput(nd.ni_dvp);
1551 		if (nd.ni_vp)
1552 			vput(nd.ni_vp);
1553 		goto out;
1554 	}
1555 	VATTR_NULL(&va);
1556 	nfsm_srvsattr(&va);
1557 	if (vtyp == VCHR || vtyp == VBLK) {
1558 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1559 		major = fxdr_unsigned(u_int32_t, *tl++);
1560 		minor = fxdr_unsigned(u_int32_t, *tl);
1561 		va.va_rdev = makedev(major, minor);
1562 	}
1563 
1564 	/*
1565 	 * Iff doesn't exist, create it.
1566 	 */
1567 	if (nd.ni_vp) {
1568 		error = EEXIST;
1569 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1570 		if (nd.ni_dvp == nd.ni_vp)
1571 			vrele(nd.ni_dvp);
1572 		else
1573 			vput(nd.ni_dvp);
1574 		vput(nd.ni_vp);
1575 		goto out;
1576 	}
1577 	va.va_type = vtyp;
1578 	if (vtyp == VSOCK) {
1579 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1580 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1581 	} else {
1582 		if (va.va_type != VFIFO &&
1583 		    (error = suser(cred, (u_short *)0))) {
1584 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1585 			vput(nd.ni_dvp);
1586 			goto out;
1587 		}
1588 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1589 		error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1590 		if (error) {
1591 			goto out;
1592 		}
1593 		if (error)
1594 			goto out;
1595 		if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1596 			vput(nd.ni_vp);
1597 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1598 			error = EINVAL;
1599 		}
1600 	}
1601 out:
1602 	vp = nd.ni_vp;
1603 	if (!error) {
1604 		memset((caddr_t)fhp, 0, sizeof(nfh));
1605 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1606 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
1607 		if (!error)
1608 			error = VOP_GETATTR(vp, &va, cred, procp);
1609 		vput(vp);
1610 	}
1611 	diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1612 	vrele(dirp);
1613 	nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
1614 	if (!error) {
1615 		nfsm_srvpostop_fh(fhp);
1616 		nfsm_srvpostop_attr(0, &va);
1617 	}
1618 	nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1619 	return (0);
1620 nfsmout:
1621 	if (dirp)
1622 		vrele(dirp);
1623 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1624 	if (nd.ni_dvp == nd.ni_vp)
1625 		vrele(nd.ni_dvp);
1626 	else
1627 		vput(nd.ni_dvp);
1628 	if (nd.ni_vp)
1629 		vput(nd.ni_vp);
1630 	return (error);
1631 }
1632 
1633 /*
1634  * nfs remove service
1635  */
1636 int
1637 nfsrv_remove(nfsd, slp, procp, mrq)
1638 	struct nfsrv_descript *nfsd;
1639 	struct nfssvc_sock *slp;
1640 	struct proc *procp;
1641 	struct mbuf **mrq;
1642 {
1643 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1644 	struct mbuf *nam = nfsd->nd_nam;
1645 	caddr_t dpos = nfsd->nd_dpos;
1646 	struct ucred *cred = &nfsd->nd_cr;
1647 	struct nameidata nd;
1648 	u_int32_t *tl;
1649 	int32_t t1;
1650 	caddr_t bpos;
1651 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
1652 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1653 	char *cp2;
1654 	struct mbuf *mb, *mreq;
1655 	struct vnode *vp, *dirp;
1656 	struct vattr dirfor, diraft;
1657 	nfsfh_t nfh;
1658 	fhandle_t *fhp;
1659 	u_quad_t frev;
1660 
1661 #ifndef nolint
1662 	vp = (struct vnode *)0;
1663 #endif
1664 	fhp = &nfh.fh_generic;
1665 	nfsm_srvmtofh(fhp);
1666 	nfsm_srvnamesiz(len);
1667 	nd.ni_cnd.cn_cred = cred;
1668 	nd.ni_cnd.cn_nameiop = DELETE;
1669 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1670 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1671 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1672 	if (dirp) {
1673 		if (v3)
1674 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
1675 				procp);
1676 		else
1677 			vrele(dirp);
1678 	}
1679 	if (!error) {
1680 		vp = nd.ni_vp;
1681 		if (vp->v_type == VDIR &&
1682 		    (error = suser(cred, (u_short *)0)) != 0)
1683 			goto out;
1684 		/*
1685 		 * The root of a mounted filesystem cannot be deleted.
1686 		 */
1687 		if (vp->v_flag & VROOT) {
1688 			error = EBUSY;
1689 			goto out;
1690 		}
1691 out:
1692 		if (!error) {
1693 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1694 			nqsrv_getl(vp, ND_WRITE);
1695 			error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
1696 		} else {
1697 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1698 			if (nd.ni_dvp == vp)
1699 				vrele(nd.ni_dvp);
1700 			else
1701 				vput(nd.ni_dvp);
1702 			vput(vp);
1703 		}
1704 	}
1705 	if (dirp && v3) {
1706 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1707 		vrele(dirp);
1708 	}
1709 	nfsm_reply(NFSX_WCCDATA(v3));
1710 	if (v3) {
1711 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1712 		return (0);
1713 	}
1714 	nfsm_srvdone;
1715 }
1716 
1717 /*
1718  * nfs rename service
1719  */
1720 int
1721 nfsrv_rename(nfsd, slp, procp, mrq)
1722 	struct nfsrv_descript *nfsd;
1723 	struct nfssvc_sock *slp;
1724 	struct proc *procp;
1725 	struct mbuf **mrq;
1726 {
1727 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1728 	struct mbuf *nam = nfsd->nd_nam;
1729 	caddr_t dpos = nfsd->nd_dpos;
1730 	struct ucred *cred = &nfsd->nd_cr;
1731 	u_int32_t *tl;
1732 	int32_t t1;
1733 	caddr_t bpos;
1734 	int error = 0, cache, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
1735 	int tdirfor_ret = 1, tdiraft_ret = 1;
1736 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1737 	char *cp2;
1738 	struct mbuf *mb, *mreq;
1739 	struct nameidata fromnd, tond;
1740 	struct vnode *fvp, *tvp, *tdvp, *fdirp = (struct vnode *)0;
1741 	struct vnode *tdirp = (struct vnode *)0;
1742 	struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
1743 	nfsfh_t fnfh, tnfh;
1744 	fhandle_t *ffhp, *tfhp;
1745 	u_quad_t frev;
1746 	uid_t saved_uid;
1747 
1748 #ifndef nolint
1749 	fvp = (struct vnode *)0;
1750 #endif
1751 	ffhp = &fnfh.fh_generic;
1752 	tfhp = &tnfh.fh_generic;
1753 	fromnd.ni_cnd.cn_nameiop = 0;
1754 	tond.ni_cnd.cn_nameiop = 0;
1755 	nfsm_srvmtofh(ffhp);
1756 	nfsm_srvnamesiz(len);
1757 	/*
1758 	 * Remember our original uid so that we can reset cr_uid before
1759 	 * the second nfs_namei() call, in case it is remapped.
1760 	 */
1761 	saved_uid = cred->cr_uid;
1762 	fromnd.ni_cnd.cn_cred = cred;
1763 	fromnd.ni_cnd.cn_nameiop = DELETE;
1764 	fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART;
1765 	error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md,
1766 		&dpos, &fdirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1767 	if (fdirp) {
1768 		if (v3)
1769 			fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred,
1770 				procp);
1771 		else {
1772 			vrele(fdirp);
1773 			fdirp = (struct vnode *)0;
1774 		}
1775 	}
1776 	if (error) {
1777 		nfsm_reply(2 * NFSX_WCCDATA(v3));
1778 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1779 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1780 		if (fdirp)
1781 			vrele(fdirp);
1782 		return (0);
1783 	}
1784 	fvp = fromnd.ni_vp;
1785 	nfsm_srvmtofh(tfhp);
1786 	nfsm_strsiz(len2, NFS_MAXNAMLEN);
1787 	cred->cr_uid = saved_uid;
1788 	tond.ni_cnd.cn_cred = cred;
1789 	tond.ni_cnd.cn_nameiop = RENAME;
1790 	tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART;
1791 	error = nfs_namei(&tond, tfhp, len2, slp, nam, &md,
1792 		&dpos, &tdirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1793 	if (tdirp) {
1794 		if (v3)
1795 			tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred,
1796 				procp);
1797 		else {
1798 			vrele(tdirp);
1799 			tdirp = (struct vnode *)0;
1800 		}
1801 	}
1802 	if (error) {
1803 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1804 		vrele(fromnd.ni_dvp);
1805 		vrele(fvp);
1806 		goto out1;
1807 	}
1808 	tdvp = tond.ni_dvp;
1809 	tvp = tond.ni_vp;
1810 	if (tvp != NULL) {
1811 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1812 			if (v3)
1813 				error = EEXIST;
1814 			else
1815 				error = EISDIR;
1816 			goto out;
1817 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1818 			if (v3)
1819 				error = EEXIST;
1820 			else
1821 				error = ENOTDIR;
1822 			goto out;
1823 		}
1824 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1825 			if (v3)
1826 				error = EXDEV;
1827 			else
1828 				error = ENOTEMPTY;
1829 			goto out;
1830 		}
1831 	}
1832 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1833 		if (v3)
1834 			error = EXDEV;
1835 		else
1836 			error = ENOTEMPTY;
1837 		goto out;
1838 	}
1839 	if (fvp->v_mount != tdvp->v_mount) {
1840 		if (v3)
1841 			error = EXDEV;
1842 		else
1843 			error = ENOTEMPTY;
1844 		goto out;
1845 	}
1846 	if (fvp == tdvp) {
1847 		if (v3)
1848 			error = EINVAL;
1849 		else
1850 			error = ENOTEMPTY;
1851 	}
1852 	/*
1853 	 * If source is the same as the destination (that is the
1854 	 * same vnode with the same name in the same directory),
1855 	 * then there is nothing to do.
1856 	 */
1857 	if (fvp == tvp && fromnd.ni_dvp == tdvp &&
1858 	    fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
1859 	    !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
1860 	      fromnd.ni_cnd.cn_namelen))
1861 		error = -1;
1862 out:
1863 	if (!error) {
1864 		nqsrv_getl(fromnd.ni_dvp, ND_WRITE);
1865 		nqsrv_getl(tdvp, ND_WRITE);
1866 		if (tvp) {
1867 			nqsrv_getl(tvp, ND_WRITE);
1868 		}
1869 		error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
1870 				   tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
1871 	} else {
1872 		VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd);
1873 		if (tdvp == tvp)
1874 			vrele(tdvp);
1875 		else
1876 			vput(tdvp);
1877 		if (tvp)
1878 			vput(tvp);
1879 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1880 		vrele(fromnd.ni_dvp);
1881 		vrele(fvp);
1882 		if (error == -1)
1883 			error = 0;
1884 	}
1885 	vrele(tond.ni_startdir);
1886 	PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
1887 out1:
1888 	if (fdirp) {
1889 		fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, procp);
1890 		vrele(fdirp);
1891 	}
1892 	if (tdirp) {
1893 		tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, procp);
1894 		vrele(tdirp);
1895 	}
1896 	vrele(fromnd.ni_startdir);
1897 	PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
1898 	nfsm_reply(2 * NFSX_WCCDATA(v3));
1899 	if (v3) {
1900 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1901 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1902 	}
1903 	return (0);
1904 
1905 nfsmout:
1906 	if (fdirp)
1907 		vrele(fdirp);
1908 	if (tdirp)
1909 		vrele(tdirp);
1910 	if (tond.ni_cnd.cn_nameiop) {
1911 		vrele(tond.ni_startdir);
1912 		PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
1913 	}
1914 	if (fromnd.ni_cnd.cn_nameiop) {
1915 		vrele(fromnd.ni_startdir);
1916 		PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
1917 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1918 		vrele(fromnd.ni_dvp);
1919 		vrele(fvp);
1920 	}
1921 	return (error);
1922 }
1923 
1924 /*
1925  * nfs link service
1926  */
1927 int
1928 nfsrv_link(nfsd, slp, procp, mrq)
1929 	struct nfsrv_descript *nfsd;
1930 	struct nfssvc_sock *slp;
1931 	struct proc *procp;
1932 	struct mbuf **mrq;
1933 {
1934 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1935 	struct mbuf *nam = nfsd->nd_nam;
1936 	caddr_t dpos = nfsd->nd_dpos;
1937 	struct ucred *cred = &nfsd->nd_cr;
1938 	struct nameidata nd;
1939 	u_int32_t *tl;
1940 	int32_t t1;
1941 	caddr_t bpos;
1942 	int error = 0, rdonly, cache, len, dirfor_ret = 1, diraft_ret = 1;
1943 	int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
1944 	char *cp2;
1945 	struct mbuf *mb, *mreq;
1946 	struct vnode *vp, *xp, *dirp = (struct vnode *)0;
1947 	struct vattr dirfor, diraft, at;
1948 	nfsfh_t nfh, dnfh;
1949 	fhandle_t *fhp, *dfhp;
1950 	u_quad_t frev;
1951 
1952 	fhp = &nfh.fh_generic;
1953 	dfhp = &dnfh.fh_generic;
1954 	nfsm_srvmtofh(fhp);
1955 	nfsm_srvmtofh(dfhp);
1956 	nfsm_srvnamesiz(len);
1957 	error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, slp, nam,
1958 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1959 	if (error) {
1960 		nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
1961 		nfsm_srvpostop_attr(getret, &at);
1962 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1963 		return (0);
1964 	}
1965 	if (vp->v_type == VDIR && (error = suser(cred, (u_short *)0)) != 0)
1966 		goto out1;
1967 	nd.ni_cnd.cn_cred = cred;
1968 	nd.ni_cnd.cn_nameiop = CREATE;
1969 	nd.ni_cnd.cn_flags = LOCKPARENT;
1970 	error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos,
1971 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1972 	if (dirp) {
1973 		if (v3)
1974 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
1975 				procp);
1976 		else {
1977 			vrele(dirp);
1978 			dirp = (struct vnode *)0;
1979 		}
1980 	}
1981 	if (error)
1982 		goto out1;
1983 	xp = nd.ni_vp;
1984 	if (xp != NULL) {
1985 		error = EEXIST;
1986 		goto out;
1987 	}
1988 	xp = nd.ni_dvp;
1989 	if (vp->v_mount != xp->v_mount)
1990 		error = EXDEV;
1991 out:
1992 	if (!error) {
1993 		nqsrv_getl(vp, ND_WRITE);
1994 		nqsrv_getl(xp, ND_WRITE);
1995 		error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
1996 	} else {
1997 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1998 		if (nd.ni_dvp == nd.ni_vp)
1999 			vrele(nd.ni_dvp);
2000 		else
2001 			vput(nd.ni_dvp);
2002 		if (nd.ni_vp)
2003 			vrele(nd.ni_vp);
2004 	}
2005 out1:
2006 	if (v3)
2007 		getret = VOP_GETATTR(vp, &at, cred, procp);
2008 	if (dirp) {
2009 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2010 		vrele(dirp);
2011 	}
2012 	vrele(vp);
2013 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2014 	if (v3) {
2015 		nfsm_srvpostop_attr(getret, &at);
2016 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2017 		return (0);
2018 	}
2019 	nfsm_srvdone;
2020 }
2021 
2022 /*
2023  * nfs symbolic link service
2024  */
2025 int
2026 nfsrv_symlink(nfsd, slp, procp, mrq)
2027 	struct nfsrv_descript *nfsd;
2028 	struct nfssvc_sock *slp;
2029 	struct proc *procp;
2030 	struct mbuf **mrq;
2031 {
2032 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2033 	struct mbuf *nam = nfsd->nd_nam;
2034 	caddr_t dpos = nfsd->nd_dpos;
2035 	struct ucred *cred = &nfsd->nd_cr;
2036 	struct vattr va, dirfor, diraft;
2037 	struct nameidata nd;
2038 	u_int32_t *tl;
2039 	int32_t t1;
2040 	struct nfsv2_sattr *sp;
2041 	char *bpos, *pathcp = NULL, *cp2;
2042 	struct uio io;
2043 	struct iovec iv;
2044 	int error = 0, cache, len, len2, dirfor_ret = 1, diraft_ret = 1;
2045 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2046 	struct mbuf *mb, *mreq, *mb2;
2047 	struct vnode *dirp = (struct vnode *)0;
2048 	nfsfh_t nfh;
2049 	fhandle_t *fhp;
2050 	u_quad_t frev;
2051 
2052 	nd.ni_cnd.cn_nameiop = 0;
2053 	fhp = &nfh.fh_generic;
2054 	nfsm_srvmtofh(fhp);
2055 	nfsm_srvnamesiz(len);
2056 	nd.ni_cnd.cn_cred = cred;
2057 	nd.ni_cnd.cn_nameiop = CREATE;
2058 	nd.ni_cnd.cn_flags = LOCKPARENT;
2059 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2060 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2061 	if (dirp) {
2062 		if (v3)
2063 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2064 				procp);
2065 		else {
2066 			vrele(dirp);
2067 			dirp = (struct vnode *)0;
2068 		}
2069 	}
2070 	if (error)
2071 		goto out;
2072 	VATTR_NULL(&va);
2073 	if (v3)
2074 		nfsm_srvsattr(&va);
2075 	nfsm_strsiz(len2, NFS_MAXPATHLEN);
2076 	pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK);
2077 	iv.iov_base = pathcp;
2078 	iv.iov_len = len2;
2079 	io.uio_resid = len2;
2080 	io.uio_offset = 0;
2081 	io.uio_iov = &iv;
2082 	io.uio_iovcnt = 1;
2083 	io.uio_segflg = UIO_SYSSPACE;
2084 	io.uio_rw = UIO_READ;
2085 	io.uio_procp = (struct proc *)0;
2086 	nfsm_mtouio(&io, len2);
2087 	if (!v3) {
2088 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2089 		va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
2090 	}
2091 	*(pathcp + len2) = '\0';
2092 	if (nd.ni_vp) {
2093 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2094 		if (nd.ni_dvp == nd.ni_vp)
2095 			vrele(nd.ni_dvp);
2096 		else
2097 			vput(nd.ni_dvp);
2098 		vrele(nd.ni_vp);
2099 		error = EEXIST;
2100 		goto out;
2101 	}
2102 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2103 	error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp);
2104 	if (!error) {
2105 	    if (v3) {
2106 		memset((caddr_t)fhp, 0, sizeof(nfh));
2107 		fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2108 		error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2109 		if (!error)
2110 		    error = VOP_GETATTR(nd.ni_vp, &va, cred,
2111 					procp);
2112 		vput(nd.ni_vp);
2113 	    } else {
2114 		vput(nd.ni_vp);
2115 	    }
2116 	}
2117 out:
2118 	if (pathcp)
2119 		free(pathcp, M_TEMP);
2120 	if (dirp) {
2121 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2122 		vrele(dirp);
2123 	}
2124 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2125 	if (v3) {
2126 		if (!error) {
2127 			nfsm_srvpostop_fh(fhp);
2128 			nfsm_srvpostop_attr(0, &va);
2129 		}
2130 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2131 	}
2132 	return (0);
2133 nfsmout:
2134 	if (dirp)
2135 		vrele(dirp);
2136 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2137 	if (nd.ni_dvp == nd.ni_vp)
2138 		vrele(nd.ni_dvp);
2139 	else
2140 		vput(nd.ni_dvp);
2141 	if (nd.ni_vp)
2142 		vrele(nd.ni_vp);
2143 	if (pathcp)
2144 		free(pathcp, M_TEMP);
2145 	return (error);
2146 }
2147 
2148 /*
2149  * nfs mkdir service
2150  */
2151 int
2152 nfsrv_mkdir(nfsd, slp, procp, mrq)
2153 	struct nfsrv_descript *nfsd;
2154 	struct nfssvc_sock *slp;
2155 	struct proc *procp;
2156 	struct mbuf **mrq;
2157 {
2158 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2159 	struct mbuf *nam = nfsd->nd_nam;
2160 	caddr_t dpos = nfsd->nd_dpos;
2161 	struct ucred *cred = &nfsd->nd_cr;
2162 	struct vattr va, dirfor, diraft;
2163 	struct nfs_fattr *fp;
2164 	struct nameidata nd;
2165 	caddr_t cp;
2166 	u_int32_t *tl;
2167 	int32_t t1;
2168 	caddr_t bpos;
2169 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2170 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2171 	char *cp2;
2172 	struct mbuf *mb, *mb2, *mreq;
2173 	struct vnode *vp, *dirp = (struct vnode *)0;
2174 	nfsfh_t nfh;
2175 	fhandle_t *fhp;
2176 	u_quad_t frev;
2177 
2178 	fhp = &nfh.fh_generic;
2179 	nfsm_srvmtofh(fhp);
2180 	nfsm_srvnamesiz(len);
2181 	nd.ni_cnd.cn_cred = cred;
2182 	nd.ni_cnd.cn_nameiop = CREATE;
2183 	nd.ni_cnd.cn_flags = LOCKPARENT;
2184 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2185 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2186 	if (dirp) {
2187 		if (v3)
2188 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2189 				procp);
2190 		else {
2191 			vrele(dirp);
2192 			dirp = (struct vnode *)0;
2193 		}
2194 	}
2195 	if (error) {
2196 		nfsm_reply(NFSX_WCCDATA(v3));
2197 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2198 		if (dirp)
2199 			vrele(dirp);
2200 		return (0);
2201 	}
2202 	VATTR_NULL(&va);
2203 	if (v3) {
2204 		nfsm_srvsattr(&va);
2205 	} else {
2206 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2207 		va.va_mode = nfstov_mode(*tl++);
2208 	}
2209 	va.va_type = VDIR;
2210 	vp = nd.ni_vp;
2211 	if (vp != NULL) {
2212 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2213 		if (nd.ni_dvp == vp)
2214 			vrele(nd.ni_dvp);
2215 		else
2216 			vput(nd.ni_dvp);
2217 		vrele(vp);
2218 		error = EEXIST;
2219 		goto out;
2220 	}
2221 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2222 	error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
2223 	if (!error) {
2224 		vp = nd.ni_vp;
2225 		memset((caddr_t)fhp, 0, sizeof(nfh));
2226 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2227 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
2228 		if (!error)
2229 			error = VOP_GETATTR(vp, &va, cred, procp);
2230 		vput(vp);
2231 	}
2232 out:
2233 	if (dirp) {
2234 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2235 		vrele(dirp);
2236 	}
2237 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2238 	if (v3) {
2239 		if (!error) {
2240 			nfsm_srvpostop_fh(fhp);
2241 			nfsm_srvpostop_attr(0, &va);
2242 		}
2243 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2244 	} else {
2245 		nfsm_srvfhtom(fhp, v3);
2246 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2247 		nfsm_srvfillattr(&va, fp);
2248 	}
2249 	return (0);
2250 nfsmout:
2251 	if (dirp)
2252 		vrele(dirp);
2253 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2254 	if (nd.ni_dvp == nd.ni_vp)
2255 		vrele(nd.ni_dvp);
2256 	else
2257 		vput(nd.ni_dvp);
2258 	if (nd.ni_vp)
2259 		vrele(nd.ni_vp);
2260 	return (error);
2261 }
2262 
2263 /*
2264  * nfs rmdir service
2265  */
2266 int
2267 nfsrv_rmdir(nfsd, slp, procp, mrq)
2268 	struct nfsrv_descript *nfsd;
2269 	struct nfssvc_sock *slp;
2270 	struct proc *procp;
2271 	struct mbuf **mrq;
2272 {
2273 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2274 	struct mbuf *nam = nfsd->nd_nam;
2275 	caddr_t dpos = nfsd->nd_dpos;
2276 	struct ucred *cred = &nfsd->nd_cr;
2277 	u_int32_t *tl;
2278 	int32_t t1;
2279 	caddr_t bpos;
2280 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2281 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2282 	char *cp2;
2283 	struct mbuf *mb, *mreq;
2284 	struct vnode *vp, *dirp = (struct vnode *)0;
2285 	struct vattr dirfor, diraft;
2286 	nfsfh_t nfh;
2287 	fhandle_t *fhp;
2288 	struct nameidata nd;
2289 	u_quad_t frev;
2290 
2291 	fhp = &nfh.fh_generic;
2292 	nfsm_srvmtofh(fhp);
2293 	nfsm_srvnamesiz(len);
2294 	nd.ni_cnd.cn_cred = cred;
2295 	nd.ni_cnd.cn_nameiop = DELETE;
2296 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
2297 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2298 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2299 	if (dirp) {
2300 		if (v3)
2301 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2302 				procp);
2303 		else {
2304 			vrele(dirp);
2305 			dirp = (struct vnode *)0;
2306 		}
2307 	}
2308 	if (error) {
2309 		nfsm_reply(NFSX_WCCDATA(v3));
2310 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2311 		if (dirp)
2312 			vrele(dirp);
2313 		return (0);
2314 	}
2315 	vp = nd.ni_vp;
2316 	if (vp->v_type != VDIR) {
2317 		error = ENOTDIR;
2318 		goto out;
2319 	}
2320 	/*
2321 	 * No rmdir "." please.
2322 	 */
2323 	if (nd.ni_dvp == vp) {
2324 		error = EINVAL;
2325 		goto out;
2326 	}
2327 	/*
2328 	 * The root of a mounted filesystem cannot be deleted.
2329 	 */
2330 	if (vp->v_flag & VROOT)
2331 		error = EBUSY;
2332 out:
2333 	if (!error) {
2334 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
2335 		nqsrv_getl(vp, ND_WRITE);
2336 		error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2337 	} else {
2338 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2339 		if (nd.ni_dvp == nd.ni_vp)
2340 			vrele(nd.ni_dvp);
2341 		else
2342 			vput(nd.ni_dvp);
2343 		vput(vp);
2344 	}
2345 	if (dirp) {
2346 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2347 		vrele(dirp);
2348 	}
2349 	nfsm_reply(NFSX_WCCDATA(v3));
2350 	if (v3) {
2351 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2352 		return (0);
2353 	}
2354 	nfsm_srvdone;
2355 }
2356 
2357 /*
2358  * nfs readdir service
2359  * - mallocs what it thinks is enough to read
2360  *	count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
2361  * - calls VOP_READDIR()
2362  * - loops around building the reply
2363  *	if the output generated exceeds count break out of loop
2364  *	The nfsm_clget macro is used here so that the reply will be packed
2365  *	tightly in mbuf clusters.
2366  * - it only knows that it has encountered eof when the VOP_READDIR()
2367  *	reads nothing
2368  * - as such one readdir rpc will return eof false although you are there
2369  *	and then the next will return eof
2370  * - it trims out records with d_fileno == 0
2371  *	this doesn't matter for Unix clients, but they might confuse clients
2372  *	for other os'.
2373  * - it trims out records with d_type == DT_WHT
2374  *	these cannot be seen through NFS (unless we extend the protocol)
2375  * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2376  *	than requested, but this may not apply to all filesystems. For
2377  *	example, client NFS does not { although it is never remote mounted
2378  *	anyhow }
2379  *     The alternate call nfsrv_readdirplus() does lookups as well.
2380  * PS: The NFS protocol spec. does not clarify what the "count" byte
2381  *	argument is a count of.. just name strings and file id's or the
2382  *	entire reply rpc or ...
2383  *	I tried just file name and id sizes and it confused the Sun client,
2384  *	so I am using the full rpc size now. The "paranoia.." comment refers
2385  *	to including the status longwords that are not a part of the dir.
2386  *	"entry" structures, but are in the rpc.
2387  */
2388 struct flrep {
2389 	nfsuint64 fl_off;
2390 	u_int32_t fl_postopok;
2391 	u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2392 	u_int32_t fl_fhok;
2393 	u_int32_t fl_fhsize;
2394 	u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
2395 };
2396 
2397 int
2398 nfsrv_readdir(nfsd, slp, procp, mrq)
2399 	struct nfsrv_descript *nfsd;
2400 	struct nfssvc_sock *slp;
2401 	struct proc *procp;
2402 	struct mbuf **mrq;
2403 {
2404 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2405 	struct mbuf *nam = nfsd->nd_nam;
2406 	caddr_t dpos = nfsd->nd_dpos;
2407 	struct ucred *cred = &nfsd->nd_cr;
2408 	char *bp, *be;
2409 	struct mbuf *mp;
2410 	struct dirent *dp;
2411 	caddr_t cp;
2412 	u_int32_t *tl;
2413 	int32_t t1;
2414 	caddr_t bpos;
2415 	struct mbuf *mb, *mb2, *mreq, *mp2;
2416 	char *cpos, *cend, *cp2, *rbuf;
2417 	struct vnode *vp;
2418 	struct vattr at;
2419 	nfsfh_t nfh;
2420 	fhandle_t *fhp;
2421 	struct uio io;
2422 	struct iovec iv;
2423 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2424 	int siz, cnt, fullsiz, eofflag, rdonly, cache, ncookies;
2425 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2426 	u_quad_t frev, off, toff, verf;
2427 	off_t *cookies = NULL, *cookiep;
2428 	nfsuint64 jar;
2429 
2430 	fhp = &nfh.fh_generic;
2431 	nfsm_srvmtofh(fhp);
2432 	if (v3) {
2433 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2434 		toff = fxdr_hyper(tl);
2435 		tl += 2;
2436 		verf = fxdr_hyper(tl);
2437 		tl += 2;
2438 	} else {
2439 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2440 		toff = fxdr_unsigned(u_quad_t, *tl++);
2441 	}
2442 	off = toff;
2443 	cnt = fxdr_unsigned(int, *tl);
2444 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2445 	xfer = NFS_SRVMAXDATA(nfsd);
2446 	if (siz > xfer)
2447 		siz = xfer;
2448 	fullsiz = siz;
2449 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
2450 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2451 	if (!error && vp->v_type != VDIR) {
2452 		error = ENOTDIR;
2453 		vput(vp);
2454 	}
2455 	if (error) {
2456 		nfsm_reply(NFSX_UNSIGNED);
2457 		nfsm_srvpostop_attr(getret, &at);
2458 		return (0);
2459 	}
2460 	nqsrv_getl(vp, ND_READ);
2461 	if (v3) {
2462 		error = getret = VOP_GETATTR(vp, &at, cred, procp);
2463 #ifdef NFS3_STRICTVERF
2464 		/*
2465 		 * XXX This check is too strict for Solaris 2.5 clients.
2466 		 */
2467 		if (!error && toff && verf != at.va_filerev)
2468 			error = NFSERR_BAD_COOKIE;
2469 #endif
2470 	}
2471 	if (!error)
2472 		error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0);
2473 	if (error) {
2474 		vput(vp);
2475 		nfsm_reply(NFSX_POSTOPATTR(v3));
2476 		nfsm_srvpostop_attr(getret, &at);
2477 		return (0);
2478 	}
2479 	VOP_UNLOCK(vp, 0);
2480 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2481 again:
2482 	iv.iov_base = rbuf;
2483 	iv.iov_len = fullsiz;
2484 	io.uio_iov = &iv;
2485 	io.uio_iovcnt = 1;
2486 	io.uio_offset = (off_t)off;
2487 	io.uio_resid = fullsiz;
2488 	io.uio_segflg = UIO_SYSSPACE;
2489 	io.uio_rw = UIO_READ;
2490 	io.uio_procp = (struct proc *)0;
2491 	eofflag = 0;
2492 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2493 
2494 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2495 
2496 	off = (off_t)io.uio_offset;
2497 	if (!cookies && !error)
2498 		error = NFSERR_PERM;
2499 	if (v3) {
2500 		getret = VOP_GETATTR(vp, &at, cred, procp);
2501 		if (!error)
2502 			error = getret;
2503 	}
2504 
2505 	VOP_UNLOCK(vp, 0);
2506 	if (error) {
2507 		vrele(vp);
2508 		free((caddr_t)rbuf, M_TEMP);
2509 		if (cookies)
2510 			free((caddr_t)cookies, M_TEMP);
2511 		nfsm_reply(NFSX_POSTOPATTR(v3));
2512 		nfsm_srvpostop_attr(getret, &at);
2513 		return (0);
2514 	}
2515 	if (io.uio_resid) {
2516 		siz -= io.uio_resid;
2517 
2518 		/*
2519 		 * If nothing read, return eof
2520 		 * rpc reply
2521 		 */
2522 		if (siz == 0) {
2523 			vrele(vp);
2524 			nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2525 				2 * NFSX_UNSIGNED);
2526 			if (v3) {
2527 				nfsm_srvpostop_attr(getret, &at);
2528 				nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2529 				txdr_hyper(at.va_filerev, tl);
2530 				tl += 2;
2531 			} else
2532 				nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2533 			*tl++ = nfs_false;
2534 			*tl = nfs_true;
2535 			free((caddr_t)rbuf, M_TEMP);
2536 			free((caddr_t)cookies, M_TEMP);
2537 			return (0);
2538 		}
2539 	}
2540 
2541 	/*
2542 	 * Check for degenerate cases of nothing useful read.
2543 	 * If so go try again
2544 	 */
2545 	cpos = rbuf;
2546 	cend = rbuf + siz;
2547 	dp = (struct dirent *)cpos;
2548 	cookiep = cookies;
2549 
2550 	while (cpos < cend && ncookies > 0 &&
2551 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2552 		cpos += dp->d_reclen;
2553 		dp = (struct dirent *)cpos;
2554 		cookiep++;
2555 		ncookies--;
2556 	}
2557 	if (cpos >= cend || ncookies == 0) {
2558 		toff = off;
2559 		siz = fullsiz;
2560 		goto again;
2561 	}
2562 
2563 	len = 3 * NFSX_UNSIGNED;	/* paranoia, probably can be 0 */
2564 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2565 	if (v3) {
2566 		nfsm_srvpostop_attr(getret, &at);
2567 		nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2568 		txdr_hyper(at.va_filerev, tl);
2569 	}
2570 	mp = mp2 = mb;
2571 	bp = bpos;
2572 	be = bp + M_TRAILINGSPACE(mp);
2573 
2574 	/* Loop through the records and build reply */
2575 	while (cpos < cend && ncookies > 0) {
2576 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2577 			nlen = dp->d_namlen;
2578 			rem = nfsm_rndup(nlen)-nlen;
2579 			len += (4 * NFSX_UNSIGNED + nlen + rem);
2580 			if (v3)
2581 				len += 2 * NFSX_UNSIGNED;
2582 			if (len > cnt) {
2583 				eofflag = 0;
2584 				break;
2585 			}
2586 			/*
2587 			 * Build the directory record xdr from
2588 			 * the dirent entry.
2589 			 */
2590 			nfsm_clget;
2591 			*tl = nfs_true;
2592 			bp += NFSX_UNSIGNED;
2593 			if (v3) {
2594 				nfsm_clget;
2595 				*tl = 0;
2596 				bp += NFSX_UNSIGNED;
2597 			}
2598 			nfsm_clget;
2599 			*tl = txdr_unsigned(dp->d_fileno);
2600 			bp += NFSX_UNSIGNED;
2601 			nfsm_clget;
2602 			*tl = txdr_unsigned(nlen);
2603 			bp += NFSX_UNSIGNED;
2604 
2605 			/* And loop around copying the name */
2606 			xfer = nlen;
2607 			cp = dp->d_name;
2608 			while (xfer > 0) {
2609 				nfsm_clget;
2610 				if ((bp+xfer) > be)
2611 					tsiz = be-bp;
2612 				else
2613 					tsiz = xfer;
2614 				memcpy(bp, cp, tsiz);
2615 				bp += tsiz;
2616 				xfer -= tsiz;
2617 				if (xfer > 0)
2618 					cp += tsiz;
2619 			}
2620 			/* And null pad to an int32_t boundary */
2621 			for (i = 0; i < rem; i++)
2622 				*bp++ = '\0';
2623 			nfsm_clget;
2624 
2625 			/* Finish off the record */
2626 			txdr_hyper(*cookiep, &jar);
2627 			if (v3) {
2628 				*tl = jar.nfsuquad[0];
2629 				bp += NFSX_UNSIGNED;
2630 				nfsm_clget;
2631 			}
2632 			*tl = jar.nfsuquad[1];
2633 			bp += NFSX_UNSIGNED;
2634 		}
2635 		cpos += dp->d_reclen;
2636 		dp = (struct dirent *)cpos;
2637 		cookiep++;
2638 		ncookies--;
2639 	}
2640 	vrele(vp);
2641 	nfsm_clget;
2642 	*tl = nfs_false;
2643 	bp += NFSX_UNSIGNED;
2644 	nfsm_clget;
2645 	if (eofflag)
2646 		*tl = nfs_true;
2647 	else
2648 		*tl = nfs_false;
2649 	bp += NFSX_UNSIGNED;
2650 	if (mp != mb) {
2651 		if (bp < be)
2652 			mp->m_len = bp - mtod(mp, caddr_t);
2653 	} else
2654 		mp->m_len += bp - bpos;
2655 	free((caddr_t)rbuf, M_TEMP);
2656 	free((caddr_t)cookies, M_TEMP);
2657 	nfsm_srvdone;
2658 }
2659 
2660 int
2661 nfsrv_readdirplus(nfsd, slp, procp, mrq)
2662 	struct nfsrv_descript *nfsd;
2663 	struct nfssvc_sock *slp;
2664 	struct proc *procp;
2665 	struct mbuf **mrq;
2666 {
2667 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2668 	struct mbuf *nam = nfsd->nd_nam;
2669 	caddr_t dpos = nfsd->nd_dpos;
2670 	struct ucred *cred = &nfsd->nd_cr;
2671 	char *bp, *be;
2672 	struct mbuf *mp;
2673 	struct dirent *dp;
2674 	caddr_t cp;
2675 	u_int32_t *tl;
2676 	int32_t t1;
2677 	caddr_t bpos;
2678 	struct mbuf *mb, *mb2, *mreq, *mp2;
2679 	char *cpos, *cend, *cp2, *rbuf;
2680 	struct vnode *vp, *nvp;
2681 	struct flrep fl;
2682 	nfsfh_t nfh;
2683 	fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
2684 	struct uio io;
2685 	struct iovec iv;
2686 	struct vattr va, at, *vap = &va;
2687 	struct nfs_fattr *fp;
2688 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2689 	int siz, cnt, fullsiz, eofflag, rdonly, cache, dirlen, ncookies;
2690 	u_quad_t frev, off, toff, verf;
2691 	off_t *cookies = NULL, *cookiep;
2692 
2693 	fhp = &nfh.fh_generic;
2694 	nfsm_srvmtofh(fhp);
2695 	nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2696 	toff = fxdr_hyper(tl);
2697 	tl += 2;
2698 	verf = fxdr_hyper(tl);
2699 	tl += 2;
2700 	siz = fxdr_unsigned(int, *tl++);
2701 	cnt = fxdr_unsigned(int, *tl);
2702 	off = toff;
2703 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2704 	xfer = NFS_SRVMAXDATA(nfsd);
2705 	if (siz > xfer)
2706 		siz = xfer;
2707 	fullsiz = siz;
2708 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
2709 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2710 	if (!error && vp->v_type != VDIR) {
2711 		error = ENOTDIR;
2712 		vput(vp);
2713 	}
2714 	if (error) {
2715 		nfsm_reply(NFSX_UNSIGNED);
2716 		nfsm_srvpostop_attr(getret, &at);
2717 		return (0);
2718 	}
2719 	error = getret = VOP_GETATTR(vp, &at, cred, procp);
2720 #ifdef NFS3_STRICTVERF
2721 	/*
2722 	 * XXX This check is too strict for Solaris 2.5 clients.
2723 	 */
2724 	if (!error && toff && verf != at.va_filerev)
2725 		error = NFSERR_BAD_COOKIE;
2726 #endif
2727 	if (!error) {
2728 		nqsrv_getl(vp, ND_READ);
2729 		error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0);
2730 	}
2731 	if (error) {
2732 		vput(vp);
2733 		nfsm_reply(NFSX_V3POSTOPATTR);
2734 		nfsm_srvpostop_attr(getret, &at);
2735 		return (0);
2736 	}
2737 	VOP_UNLOCK(vp, 0);
2738 
2739 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2740 again:
2741 	iv.iov_base = rbuf;
2742 	iv.iov_len = fullsiz;
2743 	io.uio_iov = &iv;
2744 	io.uio_iovcnt = 1;
2745 	io.uio_offset = (off_t)off;
2746 	io.uio_resid = fullsiz;
2747 	io.uio_segflg = UIO_SYSSPACE;
2748 	io.uio_rw = UIO_READ;
2749 	io.uio_procp = (struct proc *)0;
2750 	eofflag = 0;
2751 
2752 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2753 
2754 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2755 
2756 	off = (u_quad_t)io.uio_offset;
2757 	getret = VOP_GETATTR(vp, &at, cred, procp);
2758 
2759 	VOP_UNLOCK(vp, 0);
2760 
2761 	/*
2762 	 * If the VGET operation doesn't work for this filesystem,
2763 	 * we can't support readdirplus. Returning NOTSUPP should
2764 	 * make clients fall back to plain readdir.
2765 	 * There's no need to check for VPTOFH as well, we wouldn't
2766 	 * even be here otherwise.
2767 	 */
2768 	if (!getret) {
2769 		if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp)))
2770 			getret = (getret == EOPNOTSUPP) ?
2771 				NFSERR_NOTSUPP : NFSERR_IO;
2772 		else
2773 			vput(nvp);
2774 	}
2775 
2776 	if (!cookies && !error)
2777 		error = NFSERR_PERM;
2778 	if (!error)
2779 		error = getret;
2780 	if (error) {
2781 		vrele(vp);
2782 		if (cookies)
2783 			free((caddr_t)cookies, M_TEMP);
2784 		free((caddr_t)rbuf, M_TEMP);
2785 		nfsm_reply(NFSX_V3POSTOPATTR);
2786 		nfsm_srvpostop_attr(getret, &at);
2787 		return (0);
2788 	}
2789 	if (io.uio_resid) {
2790 		siz -= io.uio_resid;
2791 
2792 		/*
2793 		 * If nothing read, return eof
2794 		 * rpc reply
2795 		 */
2796 		if (siz == 0) {
2797 			vrele(vp);
2798 			nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
2799 				2 * NFSX_UNSIGNED);
2800 			nfsm_srvpostop_attr(getret, &at);
2801 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2802 			txdr_hyper(at.va_filerev, tl);
2803 			tl += 2;
2804 			*tl++ = nfs_false;
2805 			*tl = nfs_true;
2806 			free((caddr_t)cookies, M_TEMP);
2807 			free((caddr_t)rbuf, M_TEMP);
2808 			return (0);
2809 		}
2810 	}
2811 
2812 	/*
2813 	 * Check for degenerate cases of nothing useful read.
2814 	 * If so go try again
2815 	 */
2816 	cpos = rbuf;
2817 	cend = rbuf + siz;
2818 	dp = (struct dirent *)cpos;
2819 	cookiep = cookies;
2820 
2821 	while (cpos < cend && ncookies > 0 &&
2822 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2823 		cpos += dp->d_reclen;
2824 		dp = (struct dirent *)cpos;
2825 		cookiep++;
2826 		ncookies--;
2827 	}
2828 	if (cpos >= cend || ncookies == 0) {
2829 		toff = off;
2830 		siz = fullsiz;
2831 		goto again;
2832 	}
2833 
2834 	dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
2835 	nfsm_reply(cnt);
2836 	nfsm_srvpostop_attr(getret, &at);
2837 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2838 	txdr_hyper(at.va_filerev, tl);
2839 	mp = mp2 = mb;
2840 	bp = bpos;
2841 	be = bp + M_TRAILINGSPACE(mp);
2842 
2843 	/* Loop through the records and build reply */
2844 	while (cpos < cend && ncookies > 0) {
2845 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2846 			nlen = dp->d_namlen;
2847 			rem = nfsm_rndup(nlen)-nlen;
2848 
2849 			/*
2850 			 * For readdir_and_lookup get the vnode using
2851 			 * the file number.
2852 			 */
2853 			if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp))
2854 				goto invalid;
2855 			memset((caddr_t)nfhp, 0, NFSX_V3FH);
2856 			nfhp->fh_fsid =
2857 				nvp->v_mount->mnt_stat.f_fsid;
2858 			if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) {
2859 				vput(nvp);
2860 				goto invalid;
2861 			}
2862 			if (VOP_GETATTR(nvp, vap, cred, procp)) {
2863 				vput(nvp);
2864 				goto invalid;
2865 			}
2866 			vput(nvp);
2867 
2868 			/*
2869 			 * If either the dircount or maxcount will be
2870 			 * exceeded, get out now. Both of these lengths
2871 			 * are calculated conservatively, including all
2872 			 * XDR overheads.
2873 			 */
2874 			len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
2875 				NFSX_V3POSTOPATTR);
2876 			dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
2877 			if (len > cnt || dirlen > fullsiz) {
2878 				eofflag = 0;
2879 				break;
2880 			}
2881 
2882 			/*
2883 			 * Build the directory record xdr from
2884 			 * the dirent entry.
2885 			 */
2886 			fp = (struct nfs_fattr *)&fl.fl_fattr;
2887 			nfsm_srvfillattr(vap, fp);
2888 			fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
2889 			fl.fl_fhok = nfs_true;
2890 			fl.fl_postopok = nfs_true;
2891 			txdr_hyper(*cookiep, fl.fl_off.nfsuquad);
2892 
2893 			nfsm_clget;
2894 			*tl = nfs_true;
2895 			bp += NFSX_UNSIGNED;
2896 			nfsm_clget;
2897 			*tl = 0;
2898 			bp += NFSX_UNSIGNED;
2899 			nfsm_clget;
2900 			*tl = txdr_unsigned(dp->d_fileno);
2901 			bp += NFSX_UNSIGNED;
2902 			nfsm_clget;
2903 			*tl = txdr_unsigned(nlen);
2904 			bp += NFSX_UNSIGNED;
2905 
2906 			/* And loop around copying the name */
2907 			xfer = nlen;
2908 			cp = dp->d_name;
2909 			while (xfer > 0) {
2910 				nfsm_clget;
2911 				if ((bp + xfer) > be)
2912 					tsiz = be - bp;
2913 				else
2914 					tsiz = xfer;
2915 				memcpy(bp, cp, tsiz);
2916 				bp += tsiz;
2917 				xfer -= tsiz;
2918 				if (xfer > 0)
2919 					cp += tsiz;
2920 			}
2921 			/* And null pad to an int32_t boundary */
2922 			for (i = 0; i < rem; i++)
2923 				*bp++ = '\0';
2924 
2925 			/*
2926 			 * Now copy the flrep structure out.
2927 			 */
2928 			xfer = sizeof (struct flrep);
2929 			cp = (caddr_t)&fl;
2930 			while (xfer > 0) {
2931 				nfsm_clget;
2932 				if ((bp + xfer) > be)
2933 					tsiz = be - bp;
2934 				else
2935 					tsiz = xfer;
2936 				memcpy(bp, cp, tsiz);
2937 				bp += tsiz;
2938 				xfer -= tsiz;
2939 				if (xfer > 0)
2940 					cp += tsiz;
2941 			}
2942 		}
2943 invalid:
2944 		cpos += dp->d_reclen;
2945 		dp = (struct dirent *)cpos;
2946 		cookiep++;
2947 		ncookies--;
2948 	}
2949 	vrele(vp);
2950 	nfsm_clget;
2951 	*tl = nfs_false;
2952 	bp += NFSX_UNSIGNED;
2953 	nfsm_clget;
2954 	if (eofflag)
2955 		*tl = nfs_true;
2956 	else
2957 		*tl = nfs_false;
2958 	bp += NFSX_UNSIGNED;
2959 	if (mp != mb) {
2960 		if (bp < be)
2961 			mp->m_len = bp - mtod(mp, caddr_t);
2962 	} else
2963 		mp->m_len += bp - bpos;
2964 	free((caddr_t)cookies, M_TEMP);
2965 	free((caddr_t)rbuf, M_TEMP);
2966 	nfsm_srvdone;
2967 }
2968 
2969 /*
2970  * nfs commit service
2971  */
2972 int
2973 nfsrv_commit(nfsd, slp, procp, mrq)
2974 	struct nfsrv_descript *nfsd;
2975 	struct nfssvc_sock *slp;
2976 	struct proc *procp;
2977 	struct mbuf **mrq;
2978 {
2979 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2980 	struct mbuf *nam = nfsd->nd_nam;
2981 	caddr_t dpos = nfsd->nd_dpos;
2982 	struct ucred *cred = &nfsd->nd_cr;
2983 	struct vattr bfor, aft;
2984 	struct vnode *vp;
2985 	nfsfh_t nfh;
2986 	fhandle_t *fhp;
2987 	u_int32_t *tl;
2988 	int32_t t1;
2989 	caddr_t bpos;
2990 	int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt, cache;
2991 	char *cp2;
2992 	struct mbuf *mb, *mb2, *mreq;
2993 	u_quad_t frev, off;
2994 
2995 #ifndef nolint
2996 	cache = 0;
2997 #endif
2998 	fhp = &nfh.fh_generic;
2999 	nfsm_srvmtofh(fhp);
3000 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3001 
3002 	off = fxdr_hyper(tl);
3003 	tl += 2;
3004 	cnt = fxdr_unsigned(int, *tl);
3005 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3006 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3007 	if (error) {
3008 		nfsm_reply(2 * NFSX_UNSIGNED);
3009 		nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3010 		return (0);
3011 	}
3012 	for_ret = VOP_GETATTR(vp, &bfor, cred, procp);
3013 	error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, off + cnt, procp);
3014 	aft_ret = VOP_GETATTR(vp, &aft, cred, procp);
3015 	vput(vp);
3016 	nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3017 	nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3018 	if (!error) {
3019 		nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3020 		*tl++ = txdr_unsigned(boottime.tv_sec);
3021 		*tl = txdr_unsigned(boottime.tv_usec);
3022 	} else
3023 		return (0);
3024 	nfsm_srvdone;
3025 }
3026 
3027 /*
3028  * nfs statfs service
3029  */
3030 int
3031 nfsrv_statfs(nfsd, slp, procp, mrq)
3032 	struct nfsrv_descript *nfsd;
3033 	struct nfssvc_sock *slp;
3034 	struct proc *procp;
3035 	struct mbuf **mrq;
3036 {
3037 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3038 	struct mbuf *nam = nfsd->nd_nam;
3039 	caddr_t dpos = nfsd->nd_dpos;
3040 	struct ucred *cred = &nfsd->nd_cr;
3041 	struct statfs *sf;
3042 	struct nfs_statfs *sfp;
3043 	u_int32_t *tl;
3044 	int32_t t1;
3045 	caddr_t bpos;
3046 	int error = 0, rdonly, cache, getret = 1;
3047 	int v3 = (nfsd->nd_flag & ND_NFSV3);
3048 	char *cp2;
3049 	struct mbuf *mb, *mb2, *mreq;
3050 	struct vnode *vp;
3051 	struct vattr at;
3052 	nfsfh_t nfh;
3053 	fhandle_t *fhp;
3054 	struct statfs statfs;
3055 	u_quad_t frev, tval;
3056 
3057 #ifndef nolint
3058 	cache = 0;
3059 #endif
3060 	fhp = &nfh.fh_generic;
3061 	nfsm_srvmtofh(fhp);
3062 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3063 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3064 	if (error) {
3065 		nfsm_reply(NFSX_UNSIGNED);
3066 		nfsm_srvpostop_attr(getret, &at);
3067 		return (0);
3068 	}
3069 	sf = &statfs;
3070 	error = VFS_STATFS(vp->v_mount, sf, procp);
3071 	getret = VOP_GETATTR(vp, &at, cred, procp);
3072 	vput(vp);
3073 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3074 	if (v3)
3075 		nfsm_srvpostop_attr(getret, &at);
3076 	if (error)
3077 		return (0);
3078 	nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3079 	if (v3) {
3080 		tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_bsize);
3081 		txdr_hyper(tval, &sfp->sf_tbytes);
3082 		tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_bsize);
3083 		txdr_hyper(tval, &sfp->sf_fbytes);
3084 		tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_bsize);
3085 		txdr_hyper(tval, &sfp->sf_abytes);
3086 		tval = (u_quad_t)sf->f_files;
3087 		txdr_hyper(tval, &sfp->sf_tfiles);
3088 		tval = (u_quad_t)sf->f_ffree;
3089 		txdr_hyper(tval, &sfp->sf_ffiles);
3090 		txdr_hyper(tval, &sfp->sf_afiles);
3091 		sfp->sf_invarsec = 0;
3092 	} else {
3093 		sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3094 		sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
3095 		sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3096 		sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3097 		sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3098 	}
3099 	nfsm_srvdone;
3100 }
3101 
3102 /*
3103  * nfs fsinfo service
3104  */
3105 int
3106 nfsrv_fsinfo(nfsd, slp, procp, mrq)
3107 	struct nfsrv_descript *nfsd;
3108 	struct nfssvc_sock *slp;
3109 	struct proc *procp;
3110 	struct mbuf **mrq;
3111 {
3112 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3113 	struct mbuf *nam = nfsd->nd_nam;
3114 	caddr_t dpos = nfsd->nd_dpos;
3115 	struct ucred *cred = &nfsd->nd_cr;
3116 	u_int32_t *tl;
3117 	struct nfsv3_fsinfo *sip;
3118 	int32_t t1;
3119 	caddr_t bpos;
3120 	int error = 0, rdonly, cache, getret = 1, pref;
3121 	char *cp2;
3122 	struct mbuf *mb, *mb2, *mreq;
3123 	struct vnode *vp;
3124 	struct vattr at;
3125 	nfsfh_t nfh;
3126 	fhandle_t *fhp;
3127 	u_quad_t frev, maxfsize;
3128 	struct statfs sb;
3129 
3130 #ifndef nolint
3131 	cache = 0;
3132 #endif
3133 	fhp = &nfh.fh_generic;
3134 	nfsm_srvmtofh(fhp);
3135 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3136 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3137 	if (error) {
3138 		nfsm_reply(NFSX_UNSIGNED);
3139 		nfsm_srvpostop_attr(getret, &at);
3140 		return (0);
3141 	}
3142 
3143 	/* XXX Try to make a guess on the max file size. */
3144 	VFS_STATFS(vp->v_mount, &sb, (struct proc *)0);
3145 	maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
3146 
3147 	getret = VOP_GETATTR(vp, &at, cred, procp);
3148 	vput(vp);
3149 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3150 	nfsm_srvpostop_attr(getret, &at);
3151 	nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3152 
3153 	/*
3154 	 * XXX
3155 	 * There should be file system VFS OP(s) to get this information.
3156 	 * For now, assume ufs.
3157 	 */
3158 	if (slp->ns_so->so_type == SOCK_DGRAM)
3159 		pref = NFS_MAXDGRAMDATA;
3160 	else
3161 		pref = NFS_MAXDATA;
3162 	sip->fs_rtmax = txdr_unsigned(NFS_MAXDATA);
3163 	sip->fs_rtpref = txdr_unsigned(pref);
3164 	sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3165 	sip->fs_wtmax = txdr_unsigned(NFS_MAXDATA);
3166 	sip->fs_wtpref = txdr_unsigned(pref);
3167 	sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3168 	sip->fs_dtpref = txdr_unsigned(pref);
3169 	txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3170 	sip->fs_timedelta.nfsv3_sec = 0;
3171 	sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3172 	sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3173 		NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3174 		NFSV3FSINFO_CANSETTIME);
3175 	nfsm_srvdone;
3176 }
3177 
3178 /*
3179  * nfs pathconf service
3180  */
3181 int
3182 nfsrv_pathconf(nfsd, slp, procp, mrq)
3183 	struct nfsrv_descript *nfsd;
3184 	struct nfssvc_sock *slp;
3185 	struct proc *procp;
3186 	struct mbuf **mrq;
3187 {
3188 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3189 	struct mbuf *nam = nfsd->nd_nam;
3190 	caddr_t dpos = nfsd->nd_dpos;
3191 	struct ucred *cred = &nfsd->nd_cr;
3192 	u_int32_t *tl;
3193 	struct nfsv3_pathconf *pc;
3194 	int32_t t1;
3195 	caddr_t bpos;
3196 	int error = 0, rdonly, cache, getret = 1;
3197 	register_t linkmax, namemax, chownres, notrunc;
3198 	char *cp2;
3199 	struct mbuf *mb, *mb2, *mreq;
3200 	struct vnode *vp;
3201 	struct vattr at;
3202 	nfsfh_t nfh;
3203 	fhandle_t *fhp;
3204 	u_quad_t frev;
3205 
3206 #ifndef nolint
3207 	cache = 0;
3208 #endif
3209 	fhp = &nfh.fh_generic;
3210 	nfsm_srvmtofh(fhp);
3211 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3212 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3213 	if (error) {
3214 		nfsm_reply(NFSX_UNSIGNED);
3215 		nfsm_srvpostop_attr(getret, &at);
3216 		return (0);
3217 	}
3218 	error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3219 	if (!error)
3220 		error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3221 	if (!error)
3222 		error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3223 	if (!error)
3224 		error = VOP_PATHCONF(vp, _PC_NO_TRUNC, &notrunc);
3225 	getret = VOP_GETATTR(vp, &at, cred, procp);
3226 	vput(vp);
3227 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3228 	nfsm_srvpostop_attr(getret, &at);
3229 	if (error)
3230 		return (0);
3231 	nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3232 
3233 	pc->pc_linkmax = txdr_unsigned(linkmax);
3234 	pc->pc_namemax = txdr_unsigned(namemax);
3235 	pc->pc_notrunc = txdr_unsigned(notrunc);
3236 	pc->pc_chownrestricted = txdr_unsigned(chownres);
3237 
3238 	/*
3239 	 * These should probably be supported by VOP_PATHCONF(), but
3240 	 * until msdosfs is exportable (why would you want to?), the
3241 	 * Unix defaults should be ok.
3242 	 */
3243 	pc->pc_caseinsensitive = nfs_false;
3244 	pc->pc_casepreserving = nfs_true;
3245 	nfsm_srvdone;
3246 }
3247 
3248 /*
3249  * Null operation, used by clients to ping server
3250  */
3251 /* ARGSUSED */
3252 int
3253 nfsrv_null(nfsd, slp, procp, mrq)
3254 	struct nfsrv_descript *nfsd;
3255 	struct nfssvc_sock *slp;
3256 	struct proc *procp;
3257 	struct mbuf **mrq;
3258 {
3259 	struct mbuf *mrep = nfsd->nd_mrep;
3260 	caddr_t bpos;
3261 	int error = NFSERR_RETVOID, cache = 0;
3262 	struct mbuf *mb, *mreq;
3263 	u_quad_t frev;
3264 
3265 	nfsm_reply(0);
3266 	return (0);
3267 }
3268 
3269 /*
3270  * No operation, used for obsolete procedures
3271  */
3272 /* ARGSUSED */
3273 int
3274 nfsrv_noop(nfsd, slp, procp, mrq)
3275 	struct nfsrv_descript *nfsd;
3276 	struct nfssvc_sock *slp;
3277 	struct proc *procp;
3278 	struct mbuf **mrq;
3279 {
3280 	struct mbuf *mrep = nfsd->nd_mrep;
3281 	caddr_t bpos;
3282 	int error, cache = 0;
3283 	struct mbuf *mb, *mreq;
3284 	u_quad_t frev;
3285 
3286 	if (nfsd->nd_repstat)
3287 		error = nfsd->nd_repstat;
3288 	else
3289 		error = EPROCUNAVAIL;
3290 	nfsm_reply(0);
3291 	return (0);
3292 }
3293 
3294 /*
3295  * Perform access checking for vnodes obtained from file handles that would
3296  * refer to files already opened by a Unix client. You cannot just use
3297  * vn_writechk() and VOP_ACCESS() for two reasons.
3298  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3299  * 2 - The owner is to be given access irrespective of mode bits for some
3300  *     operations, so that processes that chmod after opening a file don't
3301  *     break. I don't like this because it opens a security hole, but since
3302  *     the nfs server opens a security hole the size of a barn door anyhow,
3303  *     what the heck.
3304  *
3305  * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3306  * will return EPERM instead of EACCESS. EPERM is always an error.
3307  */
3308 int
3309 nfsrv_access(vp, flags, cred, rdonly, p, override)
3310 	struct vnode *vp;
3311 	int flags;
3312 	struct ucred *cred;
3313 	int rdonly;
3314 	struct proc *p;
3315 {
3316 	struct vattr vattr;
3317 	int error;
3318 	if (flags & VWRITE) {
3319 		/* Just vn_writechk() changed to check rdonly */
3320 		/*
3321 		 * Disallow write attempts on read-only file systems;
3322 		 * unless the file is a socket or a block or character
3323 		 * device resident on the file system.
3324 		 */
3325 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3326 			switch (vp->v_type) {
3327 			case VREG:
3328 			case VDIR:
3329 			case VLNK:
3330 				return (EROFS);
3331 			default:
3332 				break;
3333 			}
3334 		}
3335 
3336 		/*
3337 		 * If the vnode is in use as a process's text,
3338 		 * we can't allow writing.
3339 		 */
3340 		if (vp->v_flag & VTEXT)
3341 			return (ETXTBSY);
3342 	}
3343 	error = VOP_GETATTR(vp, &vattr, cred, p);
3344 	if (error)
3345 		return (error);
3346 	error = VOP_ACCESS(vp, flags, cred, p);
3347 	/*
3348 	 * Allow certain operations for the owner (reads and writes
3349 	 * on files that are already open).
3350 	 */
3351 	if (override && error == EACCES && cred->cr_uid == vattr.va_uid)
3352 		error = 0;
3353 	return error;
3354 }
3355