xref: /netbsd-src/sys/nfs/nfs_serv.c (revision 8a8f936f250a330d54f8a24ed0e92aadf9743a7b)
1 /*	$NetBSD: nfs_serv.c,v 1.61 2001/09/23 01:37:03 chs Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)nfs_serv.c	8.8 (Berkeley) 7/31/95
39  */
40 
41 /*
42  * nfs version 2 and 3 server calls to vnode ops
43  * - these routines generally have 3 phases
44  *   1 - break down and validate rpc request in mbuf list
45  *   2 - do the vnode ops for the request
46  *       (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
47  *   3 - build the rpc reply in an mbuf list
48  *   nb:
49  *	- do not mix the phases, since the nfsm_?? macros can return failures
50  *	  on a bad rpc or similar and do not do any vrele() or vput()'s
51  *
52  *      - the nfsm_reply() macro generates an nfs rpc reply with the nfs
53  *	error number iff error != 0 whereas
54  *	returning an error from the server function implies a fatal error
55  *	such as a badly constructed rpc request that should be dropped without
56  *	a reply.
57  *	For Version 3, nfsm_reply() does not return for the error case, since
58  *	most version 3 rpcs return more than the status for error cases.
59  */
60 
61 #include <sys/param.h>
62 #include <sys/systm.h>
63 #include <sys/proc.h>
64 #include <sys/file.h>
65 #include <sys/namei.h>
66 #include <sys/vnode.h>
67 #include <sys/mount.h>
68 #include <sys/socket.h>
69 #include <sys/socketvar.h>
70 #include <sys/mbuf.h>
71 #include <sys/dirent.h>
72 #include <sys/stat.h>
73 #include <sys/kernel.h>
74 #include <ufs/ufs/dir.h>
75 
76 #include <uvm/uvm_extern.h>
77 
78 #include <nfs/nfsproto.h>
79 #include <nfs/rpcv2.h>
80 #include <nfs/nfs.h>
81 #include <nfs/xdr_subs.h>
82 #include <nfs/nfsm_subs.h>
83 #include <nfs/nqnfs.h>
84 #include <nfs/nfs_var.h>
85 
86 /* Global vars */
87 extern u_int32_t nfs_xdrneg1;
88 extern u_int32_t nfs_false, nfs_true;
89 extern enum vtype nv3tov_type[8];
90 extern struct nfsstats nfsstats;
91 extern nfstype nfsv2_type[9];
92 extern nfstype nfsv3_type[9];
93 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
94 
95 /*
96  * nfs v3 access service
97  */
98 int
99 nfsrv3_access(nfsd, slp, procp, mrq)
100 	struct nfsrv_descript *nfsd;
101 	struct nfssvc_sock *slp;
102 	struct proc *procp;
103 	struct mbuf **mrq;
104 {
105 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
106 	struct mbuf *nam = nfsd->nd_nam;
107 	caddr_t dpos = nfsd->nd_dpos;
108 	struct ucred *cred = &nfsd->nd_cr;
109 	struct vnode *vp;
110 	nfsfh_t nfh;
111 	fhandle_t *fhp;
112 	u_int32_t *tl;
113 	int32_t t1;
114 	caddr_t bpos;
115 	int error = 0, rdonly, cache = 0, getret;
116 	char *cp2;
117 	struct mbuf *mb, *mreq, *mb2;
118 	struct vattr va;
119 	u_long inmode, testmode, outmode;
120 	u_quad_t frev;
121 
122 	fhp = &nfh.fh_generic;
123 	nfsm_srvmtofh(fhp);
124 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
125 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
126 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
127 	if (error) {
128 		nfsm_reply(NFSX_UNSIGNED);
129 		nfsm_srvpostop_attr(1, (struct vattr *)0);
130 		return (0);
131 	}
132 	inmode = fxdr_unsigned(u_int32_t, *tl);
133 	outmode = 0;
134 	if ((inmode & NFSV3ACCESS_READ) &&
135 	    nfsrv_access(vp, VREAD, cred, rdonly, procp, 0) == 0)
136 		outmode |= NFSV3ACCESS_READ;
137 	if (vp->v_type != VDIR) {
138 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
139 		if (testmode &&
140 		    nfsrv_access(vp, VWRITE, cred, rdonly, procp, 0) == 0)
141 			outmode |= testmode;
142 		if ((inmode & NFSV3ACCESS_EXECUTE) &&
143 		    nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0) == 0)
144 			outmode |= NFSV3ACCESS_EXECUTE;
145 	} else {
146 		testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
147 		    NFSV3ACCESS_DELETE);
148 		if (testmode &&
149 		    nfsrv_access(vp, VWRITE, cred, rdonly, procp, 0) == 0)
150 			outmode |= testmode;
151 		if ((inmode & NFSV3ACCESS_LOOKUP) &&
152 		    nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0) == 0)
153 			outmode |= NFSV3ACCESS_LOOKUP;
154 	}
155 	getret = VOP_GETATTR(vp, &va, cred, procp);
156 	vput(vp);
157 	nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
158 	nfsm_srvpostop_attr(getret, &va);
159 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
160 	*tl = txdr_unsigned(outmode);
161 	nfsm_srvdone;
162 }
163 
164 /*
165  * nfs getattr service
166  */
167 int
168 nfsrv_getattr(nfsd, slp, procp, mrq)
169 	struct nfsrv_descript *nfsd;
170 	struct nfssvc_sock *slp;
171 	struct proc *procp;
172 	struct mbuf **mrq;
173 {
174 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
175 	struct mbuf *nam = nfsd->nd_nam;
176 	caddr_t dpos = nfsd->nd_dpos;
177 	struct ucred *cred = &nfsd->nd_cr;
178 	struct nfs_fattr *fp;
179 	struct vattr va;
180 	struct vnode *vp;
181 	nfsfh_t nfh;
182 	fhandle_t *fhp;
183 	u_int32_t *tl;
184 	int32_t t1;
185 	caddr_t bpos;
186 	int error = 0, rdonly, cache;
187 	char *cp2;
188 	struct mbuf *mb, *mb2, *mreq;
189 	u_quad_t frev;
190 
191 	fhp = &nfh.fh_generic;
192 	nfsm_srvmtofh(fhp);
193 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
194 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
195 	if (error) {
196 		nfsm_reply(0);
197 		return (0);
198 	}
199 	nqsrv_getl(vp, ND_READ);
200 	error = VOP_GETATTR(vp, &va, cred, procp);
201 	vput(vp);
202 	nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
203 	if (error)
204 		return (0);
205 	nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
206 	nfsm_srvfillattr(&va, fp);
207 	nfsm_srvdone;
208 }
209 
210 /*
211  * nfs setattr service
212  */
213 int
214 nfsrv_setattr(nfsd, slp, procp, mrq)
215 	struct nfsrv_descript *nfsd;
216 	struct nfssvc_sock *slp;
217 	struct proc *procp;
218 	struct mbuf **mrq;
219 {
220 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
221 	struct mbuf *nam = nfsd->nd_nam;
222 	caddr_t dpos = nfsd->nd_dpos;
223 	struct ucred *cred = &nfsd->nd_cr;
224 	struct vattr va, preat;
225 	struct nfsv2_sattr *sp;
226 	struct nfs_fattr *fp;
227 	struct vnode *vp;
228 	nfsfh_t nfh;
229 	fhandle_t *fhp;
230 	u_int32_t *tl;
231 	int32_t t1;
232 	caddr_t bpos;
233 	int error = 0, rdonly, cache, preat_ret = 1, postat_ret = 1;
234 	int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
235 	char *cp2;
236 	struct mbuf *mb, *mb2, *mreq;
237 	u_quad_t frev;
238 	struct timespec guard;
239 
240 	fhp = &nfh.fh_generic;
241 	nfsm_srvmtofh(fhp);
242 	VATTR_NULL(&va);
243 	if (v3) {
244 		nfsm_srvsattr(&va);
245 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
246 		gcheck = fxdr_unsigned(int, *tl);
247 		if (gcheck) {
248 			nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
249 			fxdr_nfsv3time(tl, &guard);
250 		}
251 	} else {
252 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
253 		/*
254 		 * Nah nah nah nah na nah
255 		 * There is a bug in the Sun client that puts 0xffff in the mode
256 		 * field of sattr when it should put in 0xffffffff. The u_short
257 		 * doesn't sign extend.
258 		 * --> check the low order 2 bytes for 0xffff
259 		 */
260 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
261 			va.va_mode = nfstov_mode(sp->sa_mode);
262 		if (sp->sa_uid != nfs_xdrneg1)
263 			va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
264 		if (sp->sa_gid != nfs_xdrneg1)
265 			va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
266 		if (sp->sa_size != nfs_xdrneg1)
267 			va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
268 		if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) {
269 #ifdef notyet
270 			fxdr_nfsv2time(&sp->sa_atime, &va.va_atime);
271 #else
272 			va.va_atime.tv_sec =
273 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
274 			va.va_atime.tv_nsec = 0;
275 #endif
276 		}
277 		if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1)
278 			fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime);
279 
280 	}
281 
282 	/*
283 	 * Now that we have all the fields, lets do it.
284 	 */
285 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
286 	    (nfsd->nd_flag & ND_KERBAUTH), FALSE);
287 	if (error) {
288 		nfsm_reply(2 * NFSX_UNSIGNED);
289 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
290 		return (0);
291 	}
292 	nqsrv_getl(vp, ND_WRITE);
293 	if (v3) {
294 		error = preat_ret = VOP_GETATTR(vp, &preat, cred, procp);
295 		if (!error && gcheck &&
296 			(preat.va_ctime.tv_sec != guard.tv_sec ||
297 			 preat.va_ctime.tv_nsec != guard.tv_nsec))
298 			error = NFSERR_NOT_SYNC;
299 		if (error) {
300 			vput(vp);
301 			nfsm_reply(NFSX_WCCDATA(v3));
302 			nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
303 			return (0);
304 		}
305 	}
306 
307 	/*
308 	 * If the size is being changed write acces is required, otherwise
309 	 * just check for a read only file system.
310 	 */
311 	if (va.va_size == ((u_quad_t)((quad_t) -1))) {
312 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
313 			error = EROFS;
314 			goto out;
315 		}
316 	} else {
317 		if (vp->v_type == VDIR) {
318 			error = EISDIR;
319 			goto out;
320 		} else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
321 			procp, 0)) != 0)
322 			goto out;
323 	}
324 	error = VOP_SETATTR(vp, &va, cred, procp);
325 	postat_ret = VOP_GETATTR(vp, &va, cred, procp);
326 	if (!error)
327 		error = postat_ret;
328 out:
329 	vput(vp);
330 	nfsm_reply(NFSX_WCCORFATTR(v3));
331 	if (v3) {
332 		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
333 		return (0);
334 	} else {
335 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
336 		nfsm_srvfillattr(&va, fp);
337 	}
338 	nfsm_srvdone;
339 }
340 
341 /*
342  * nfs lookup rpc
343  */
344 int
345 nfsrv_lookup(nfsd, slp, procp, mrq)
346 	struct nfsrv_descript *nfsd;
347 	struct nfssvc_sock *slp;
348 	struct proc *procp;
349 	struct mbuf **mrq;
350 {
351 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
352 	struct mbuf *nam = nfsd->nd_nam;
353 	caddr_t dpos = nfsd->nd_dpos;
354 	struct ucred *cred = &nfsd->nd_cr;
355 	struct nfs_fattr *fp;
356 	struct nameidata nd, ind, *ndp = &nd;
357 	struct vnode *vp, *dirp;
358 	nfsfh_t nfh;
359 	fhandle_t *fhp;
360 	caddr_t cp;
361 	u_int32_t *tl;
362 	int32_t t1;
363 	caddr_t bpos;
364 	int error = 0, cache, len, dirattr_ret = 1;
365 	int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
366 	char *cp2;
367 	struct mbuf *mb, *mb2, *mreq;
368 	struct vattr va, dirattr;
369 	u_quad_t frev;
370 
371 	fhp = &nfh.fh_generic;
372 	nfsm_srvmtofh(fhp);
373 	nfsm_srvnamesiz(len);
374 
375 	pubflag = nfs_ispublicfh(fhp);
376 
377 	nd.ni_cnd.cn_cred = cred;
378 	nd.ni_cnd.cn_nameiop = LOOKUP;
379 	nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART;
380 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
381 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
382 
383 	if (!error && pubflag) {
384 		if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
385 			/*
386 			 * Setup call to lookup() to see if we can find
387 			 * the index file. Arguably, this doesn't belong
388 			 * in a kernel.. Ugh.
389 			 */
390 			ind = nd;
391 			VOP_UNLOCK(nd.ni_vp, 0);
392 			ind.ni_pathlen = strlen(nfs_pub.np_index);
393 			ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
394 			    nfs_pub.np_index;
395 			ind.ni_startdir = nd.ni_vp;
396 			VREF(ind.ni_startdir);
397 			error = lookup(&ind);
398 			if (!error) {
399 				/*
400 				 * Found an index file. Get rid of
401 				 * the old references.
402 				 */
403 				if (dirp)
404 					vrele(dirp);
405 				dirp = nd.ni_vp;
406 				vrele(nd.ni_startdir);
407 				ndp = &ind;
408 			} else
409 				error = 0;
410 		}
411 		/*
412 		 * If the public filehandle was used, check that this lookup
413 		 * didn't result in a filehandle outside the publicly exported
414 		 * filesystem.
415 		 */
416 
417 		if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) {
418 			vput(nd.ni_vp);
419 			error = EPERM;
420 		}
421 	}
422 
423 	if (dirp) {
424 		if (v3)
425 			dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred,
426 				procp);
427 		vrele(dirp);
428 	}
429 
430 	if (error) {
431 		nfsm_reply(NFSX_POSTOPATTR(v3));
432 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
433 		return (0);
434 	}
435 
436 	nqsrv_getl(ndp->ni_startdir, ND_READ);
437 	vrele(ndp->ni_startdir);
438 	PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
439 	vp = ndp->ni_vp;
440 	memset((caddr_t)fhp, 0, sizeof(nfh));
441 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
442 	error = VFS_VPTOFH(vp, &fhp->fh_fid);
443 	if (!error)
444 		error = VOP_GETATTR(vp, &va, cred, procp);
445 	vput(vp);
446 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
447 	if (error) {
448 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
449 		return (0);
450 	}
451 	nfsm_srvfhtom(fhp, v3);
452 	if (v3) {
453 		nfsm_srvpostop_attr(0, &va);
454 		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
455 	} else {
456 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
457 		nfsm_srvfillattr(&va, fp);
458 	}
459 	nfsm_srvdone;
460 }
461 
462 /*
463  * nfs readlink service
464  */
465 int
466 nfsrv_readlink(nfsd, slp, procp, mrq)
467 	struct nfsrv_descript *nfsd;
468 	struct nfssvc_sock *slp;
469 	struct proc *procp;
470 	struct mbuf **mrq;
471 {
472 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
473 	struct mbuf *nam = nfsd->nd_nam;
474 	caddr_t dpos = nfsd->nd_dpos;
475 	struct ucred *cred = &nfsd->nd_cr;
476 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
477 	struct iovec *ivp = iv;
478 	struct mbuf *mp;
479 	u_int32_t *tl;
480 	int32_t t1;
481 	caddr_t bpos;
482 	int error = 0, rdonly, cache, i, tlen, len, getret;
483 	int v3 = (nfsd->nd_flag & ND_NFSV3);
484 	char *cp2;
485 	struct mbuf *mb, *mb2, *mp2 = NULL, *mp3 = NULL, *mreq;
486 	struct vnode *vp;
487 	struct vattr attr;
488 	nfsfh_t nfh;
489 	fhandle_t *fhp;
490 	struct uio io, *uiop = &io;
491 	u_quad_t frev;
492 
493 	fhp = &nfh.fh_generic;
494 	nfsm_srvmtofh(fhp);
495 	len = 0;
496 	i = 0;
497 	while (len < NFS_MAXPATHLEN) {
498 		MGET(mp, M_WAIT, MT_DATA);
499 		MCLGET(mp, M_WAIT);
500 		mp->m_len = NFSMSIZ(mp);
501 		if (len == 0)
502 			mp3 = mp2 = mp;
503 		else {
504 			mp2->m_next = mp;
505 			mp2 = mp;
506 		}
507 		if ((len+mp->m_len) > NFS_MAXPATHLEN) {
508 			mp->m_len = NFS_MAXPATHLEN-len;
509 			len = NFS_MAXPATHLEN;
510 		} else
511 			len += mp->m_len;
512 		ivp->iov_base = mtod(mp, caddr_t);
513 		ivp->iov_len = mp->m_len;
514 		i++;
515 		ivp++;
516 	}
517 	uiop->uio_iov = iv;
518 	uiop->uio_iovcnt = i;
519 	uiop->uio_offset = 0;
520 	uiop->uio_resid = len;
521 	uiop->uio_rw = UIO_READ;
522 	uiop->uio_segflg = UIO_SYSSPACE;
523 	uiop->uio_procp = (struct proc *)0;
524 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
525 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
526 	if (error) {
527 		m_freem(mp3);
528 		nfsm_reply(2 * NFSX_UNSIGNED);
529 		nfsm_srvpostop_attr(1, (struct vattr *)0);
530 		return (0);
531 	}
532 	if (vp->v_type != VLNK) {
533 		if (v3)
534 			error = EINVAL;
535 		else
536 			error = ENXIO;
537 		goto out;
538 	}
539 	nqsrv_getl(vp, ND_READ);
540 	error = VOP_READLINK(vp, uiop, cred);
541 out:
542 	getret = VOP_GETATTR(vp, &attr, cred, procp);
543 	vput(vp);
544 	if (error)
545 		m_freem(mp3);
546 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
547 	if (v3) {
548 		nfsm_srvpostop_attr(getret, &attr);
549 		if (error)
550 			return (0);
551 	}
552 	if (uiop->uio_resid > 0) {
553 		len -= uiop->uio_resid;
554 		tlen = nfsm_rndup(len);
555 		nfsm_adj(mp3, NFS_MAXPATHLEN-tlen, tlen-len);
556 	}
557 	nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
558 	*tl = txdr_unsigned(len);
559 	mb->m_next = mp3;
560 	nfsm_srvdone;
561 }
562 
563 /*
564  * nfs read service
565  */
566 int
567 nfsrv_read(nfsd, slp, procp, mrq)
568 	struct nfsrv_descript *nfsd;
569 	struct nfssvc_sock *slp;
570 	struct proc *procp;
571 	struct mbuf **mrq;
572 {
573 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
574 	struct mbuf *nam = nfsd->nd_nam;
575 	caddr_t dpos = nfsd->nd_dpos;
576 	struct ucred *cred = &nfsd->nd_cr;
577 	struct iovec *iv;
578 	struct iovec *iv2;
579 	struct mbuf *m;
580 	struct nfs_fattr *fp;
581 	u_int32_t *tl;
582 	int32_t t1;
583 	int i;
584 	caddr_t bpos;
585 	int error = 0, rdonly, cache, cnt, len, left, siz, tlen, getret;
586 	int v3 = (nfsd->nd_flag & ND_NFSV3), reqlen;
587 	char *cp2;
588 	struct mbuf *mb, *mb2, *mreq;
589 	struct mbuf *m2;
590 	struct vnode *vp;
591 	nfsfh_t nfh;
592 	fhandle_t *fhp;
593 	struct uio io, *uiop = &io;
594 	struct vattr va;
595 	off_t off;
596 	u_quad_t frev;
597 
598 	fhp = &nfh.fh_generic;
599 	nfsm_srvmtofh(fhp);
600 	if (v3) {
601 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
602 		off = fxdr_hyper(tl);
603 	} else {
604 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
605 		off = (off_t)fxdr_unsigned(u_int32_t, *tl);
606 	}
607 	nfsm_srvstrsiz(reqlen, NFS_SRVMAXDATA(nfsd));
608 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
609 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
610 	if (error) {
611 		nfsm_reply(2 * NFSX_UNSIGNED);
612 		nfsm_srvpostop_attr(1, (struct vattr *)0);
613 		return (0);
614 	}
615 	if (vp->v_type != VREG) {
616 		if (v3)
617 			error = EINVAL;
618 		else
619 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
620 	}
621 	if (!error) {
622 	    nqsrv_getl(vp, ND_READ);
623 	    if ((error = nfsrv_access(vp, VREAD, cred, rdonly, procp, 1)) != 0)
624 		error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 1);
625 	}
626 	getret = VOP_GETATTR(vp, &va, cred, procp);
627 	if (!error)
628 		error = getret;
629 	if (error) {
630 		vput(vp);
631 		nfsm_reply(NFSX_POSTOPATTR(v3));
632 		nfsm_srvpostop_attr(getret, &va);
633 		return (0);
634 	}
635 	if (off >= va.va_size)
636 		cnt = 0;
637 	else if ((off + reqlen) > va.va_size)
638 		cnt = nfsm_rndup(va.va_size - off);
639 	else
640 		cnt = reqlen;
641 	nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
642 	if (v3) {
643 		nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
644 		*tl++ = nfs_true;
645 		fp = (struct nfs_fattr *)tl;
646 		tl += (NFSX_V3FATTR / sizeof (u_int32_t));
647 	} else {
648 		nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
649 		fp = (struct nfs_fattr *)tl;
650 		tl += (NFSX_V2FATTR / sizeof (u_int32_t));
651 	}
652 	len = left = cnt;
653 	if (cnt > 0) {
654 		/*
655 		 * Generate the mbuf list with the uio_iov ref. to it.
656 		 */
657 		i = 0;
658 		m = m2 = mb;
659 		while (left > 0) {
660 			siz = min(M_TRAILINGSPACE(m), left);
661 			if (siz > 0) {
662 				left -= siz;
663 				i++;
664 			}
665 			if (left > 0) {
666 				MGET(m, M_WAIT, MT_DATA);
667 				MCLGET(m, M_WAIT);
668 				m->m_len = 0;
669 				m2->m_next = m;
670 				m2 = m;
671 			}
672 		}
673 		iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
674 		uiop->uio_iov = iv2 = iv;
675 		m = mb;
676 		left = cnt;
677 		i = 0;
678 		while (left > 0) {
679 			if (m == NULL)
680 				panic("nfsrv_read iov");
681 			siz = min(M_TRAILINGSPACE(m), left);
682 			if (siz > 0) {
683 				iv->iov_base = mtod(m, caddr_t) + m->m_len;
684 				iv->iov_len = siz;
685 				m->m_len += siz;
686 				left -= siz;
687 				iv++;
688 				i++;
689 			}
690 			m = m->m_next;
691 		}
692 		uiop->uio_iovcnt = i;
693 		uiop->uio_offset = off;
694 		uiop->uio_resid = cnt;
695 		uiop->uio_rw = UIO_READ;
696 		uiop->uio_segflg = UIO_SYSSPACE;
697 		error = VOP_READ(vp, uiop, IO_NODELOCKED, cred);
698 		off = uiop->uio_offset;
699 		free((caddr_t)iv2, M_TEMP);
700 		if (error || (getret = VOP_GETATTR(vp, &va, cred, procp)) != 0){
701 			if (!error)
702 				error = getret;
703 			m_freem(mreq);
704 			vput(vp);
705 			nfsm_reply(NFSX_POSTOPATTR(v3));
706 			nfsm_srvpostop_attr(getret, &va);
707 			return (0);
708 		}
709 	} else
710 		uiop->uio_resid = 0;
711 	vput(vp);
712 	nfsm_srvfillattr(&va, fp);
713 	len -= uiop->uio_resid;
714 	tlen = nfsm_rndup(len);
715 	if (cnt != tlen || tlen != len)
716 		nfsm_adj(mb, cnt - tlen, tlen - len);
717 	if (v3) {
718 		*tl++ = txdr_unsigned(len);
719 		if (len < reqlen)
720 			*tl++ = nfs_true;
721 		else
722 			*tl++ = nfs_false;
723 	}
724 	*tl = txdr_unsigned(len);
725 	nfsm_srvdone;
726 }
727 
728 /*
729  * nfs write service
730  */
731 int
732 nfsrv_write(nfsd, slp, procp, mrq)
733 	struct nfsrv_descript *nfsd;
734 	struct nfssvc_sock *slp;
735 	struct proc *procp;
736 	struct mbuf **mrq;
737 {
738 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
739 	struct mbuf *nam = nfsd->nd_nam;
740 	caddr_t dpos = nfsd->nd_dpos;
741 	struct ucred *cred = &nfsd->nd_cr;
742 	struct iovec *ivp;
743 	int i, cnt;
744 	struct mbuf *mp;
745 	struct nfs_fattr *fp;
746 	struct iovec *iv;
747 	struct vattr va, forat;
748 	u_int32_t *tl;
749 	int32_t t1;
750 	caddr_t bpos;
751 	int error = 0, rdonly, cache, len, forat_ret = 1;
752 	int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
753 	int stable = NFSV3WRITE_FILESYNC;
754 	int v3 = (nfsd->nd_flag & ND_NFSV3);
755 	char *cp2;
756 	struct mbuf *mb, *mb2, *mreq;
757 	struct vnode *vp;
758 	nfsfh_t nfh;
759 	fhandle_t *fhp;
760 	struct uio io, *uiop = &io;
761 	off_t off;
762 	u_quad_t frev;
763 
764 	if (mrep == NULL) {
765 		*mrq = NULL;
766 		return (0);
767 	}
768 	fhp = &nfh.fh_generic;
769 	nfsm_srvmtofh(fhp);
770 	if (v3) {
771 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
772 		off = fxdr_hyper(tl);
773 		tl += 3;
774 		stable = fxdr_unsigned(int, *tl++);
775 	} else {
776 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
777 		off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
778 		tl += 2;
779 	}
780 	retlen = len = fxdr_unsigned(int32_t, *tl);
781 	cnt = i = 0;
782 
783 	/*
784 	 * For NFS Version 2, it is not obvious what a write of zero length
785 	 * should do, but I might as well be consistent with Version 3,
786 	 * which is to return ok so long as there are no permission problems.
787 	 */
788 	if (len > 0) {
789 	    zeroing = 1;
790 	    mp = mrep;
791 	    while (mp) {
792 		if (mp == md) {
793 			zeroing = 0;
794 			adjust = dpos - mtod(mp, caddr_t);
795 			mp->m_len -= adjust;
796 			if (mp->m_len > 0 && adjust > 0)
797 				NFSMADV(mp, adjust);
798 		}
799 		if (zeroing)
800 			mp->m_len = 0;
801 		else if (mp->m_len > 0) {
802 			i += mp->m_len;
803 			if (i > len) {
804 				mp->m_len -= (i - len);
805 				zeroing	= 1;
806 			}
807 			if (mp->m_len > 0)
808 				cnt++;
809 		}
810 		mp = mp->m_next;
811 	    }
812 	}
813 	if (len > NFS_MAXDATA || len < 0 || i < len) {
814 		error = EIO;
815 		nfsm_reply(2 * NFSX_UNSIGNED);
816 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
817 		return (0);
818 	}
819 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
820 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
821 	if (error) {
822 		nfsm_reply(2 * NFSX_UNSIGNED);
823 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
824 		return (0);
825 	}
826 	if (v3)
827 		forat_ret = VOP_GETATTR(vp, &forat, cred, procp);
828 	if (vp->v_type != VREG) {
829 		if (v3)
830 			error = EINVAL;
831 		else
832 			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
833 	}
834 	if (!error) {
835 		nqsrv_getl(vp, ND_WRITE);
836 		error = nfsrv_access(vp, VWRITE, cred, rdonly, procp, 1);
837 	}
838 	if (error) {
839 		vput(vp);
840 		nfsm_reply(NFSX_WCCDATA(v3));
841 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
842 		return (0);
843 	}
844 
845 	if (len > 0) {
846 	    ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK);
847 	    uiop->uio_iov = iv = ivp;
848 	    uiop->uio_iovcnt = cnt;
849 	    mp = mrep;
850 	    while (mp) {
851 		if (mp->m_len > 0) {
852 			ivp->iov_base = mtod(mp, caddr_t);
853 			ivp->iov_len = mp->m_len;
854 			ivp++;
855 		}
856 		mp = mp->m_next;
857 	    }
858 
859 	    /*
860 	     * XXX
861 	     * The IO_METASYNC flag indicates that all metadata (and not just
862 	     * enough to ensure data integrity) mus be written to stable storage
863 	     * synchronously.
864 	     * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
865 	     */
866 	    if (stable == NFSV3WRITE_UNSTABLE)
867 		ioflags = IO_NODELOCKED;
868 	    else if (stable == NFSV3WRITE_DATASYNC)
869 		ioflags = (IO_SYNC | IO_NODELOCKED);
870 	    else
871 		ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
872 	    uiop->uio_resid = len;
873 	    uiop->uio_rw = UIO_WRITE;
874 	    uiop->uio_segflg = UIO_SYSSPACE;
875 	    uiop->uio_procp = (struct proc *)0;
876 	    uiop->uio_offset = off;
877 	    error = VOP_WRITE(vp, uiop, ioflags, cred);
878 	    nfsstats.srvvop_writes++;
879 	    free((caddr_t)iv, M_TEMP);
880 	}
881 	aftat_ret = VOP_GETATTR(vp, &va, cred, procp);
882 	vput(vp);
883 	if (!error)
884 		error = aftat_ret;
885 	nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
886 		2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
887 	if (v3) {
888 		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
889 		if (error)
890 			return (0);
891 		nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
892 		*tl++ = txdr_unsigned(retlen);
893 		if (stable == NFSV3WRITE_UNSTABLE)
894 			*tl++ = txdr_unsigned(stable);
895 		else
896 			*tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
897 		/*
898 		 * Actually, there is no need to txdr these fields,
899 		 * but it may make the values more human readable,
900 		 * for debugging purposes.
901 		 */
902 		*tl++ = txdr_unsigned(boottime.tv_sec);
903 		*tl = txdr_unsigned(boottime.tv_usec);
904 	} else {
905 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
906 		nfsm_srvfillattr(&va, fp);
907 	}
908 	nfsm_srvdone;
909 }
910 
911 /*
912  * NFS write service with write gathering support. Called when
913  * nfsrvw_procrastinate > 0.
914  * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
915  * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
916  * Jan. 1994.
917  */
918 int
919 nfsrv_writegather(ndp, slp, procp, mrq)
920 	struct nfsrv_descript **ndp;
921 	struct nfssvc_sock *slp;
922 	struct proc *procp;
923 	struct mbuf **mrq;
924 {
925 	struct iovec *ivp;
926 	struct mbuf *mp;
927 	struct nfsrv_descript *wp, *nfsd, *owp, *swp;
928 	struct nfs_fattr *fp;
929 	int i = 0;
930 	struct iovec *iov;
931 	struct nfsrvw_delayhash *wpp;
932 	struct ucred *cred;
933 	struct vattr va, forat;
934 	u_int32_t *tl;
935 	int32_t t1;
936 	caddr_t bpos, dpos;
937 	int error = 0, rdonly, cache, len = 0, forat_ret = 1;
938 	int ioflags, aftat_ret = 1, s, adjust, v3, zeroing;
939 	char *cp2;
940 	struct mbuf *mb, *mb2, *mreq, *mrep, *md;
941 	struct vnode *vp;
942 	struct uio io, *uiop = &io;
943 	u_quad_t frev, cur_usec;
944 
945 	*mrq = NULL;
946 	if (*ndp) {
947 	    nfsd = *ndp;
948 	    *ndp = NULL;
949 	    mrep = nfsd->nd_mrep;
950 	    md = nfsd->nd_md;
951 	    dpos = nfsd->nd_dpos;
952 	    cred = &nfsd->nd_cr;
953 	    v3 = (nfsd->nd_flag & ND_NFSV3);
954 	    LIST_INIT(&nfsd->nd_coalesce);
955 	    nfsd->nd_mreq = NULL;
956 	    nfsd->nd_stable = NFSV3WRITE_FILESYNC;
957 	    cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
958 	    nfsd->nd_time = cur_usec + nfsrvw_procrastinate;
959 
960 	    /*
961 	     * Now, get the write header..
962 	     */
963 	    nfsm_srvmtofh(&nfsd->nd_fh);
964 	    if (v3) {
965 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
966 		nfsd->nd_off = fxdr_hyper(tl);
967 		tl += 3;
968 		nfsd->nd_stable = fxdr_unsigned(int, *tl++);
969 	    } else {
970 		nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
971 		nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
972 		tl += 2;
973 	    }
974 	    len = fxdr_unsigned(int32_t, *tl);
975 	    nfsd->nd_len = len;
976 	    nfsd->nd_eoff = nfsd->nd_off + len;
977 
978 	    /*
979 	     * Trim the header out of the mbuf list and trim off any trailing
980 	     * junk so that the mbuf list has only the write data.
981 	     */
982 	    zeroing = 1;
983 	    i = 0;
984 	    mp = mrep;
985 	    while (mp) {
986 		if (mp == md) {
987 		    zeroing = 0;
988 		    adjust = dpos - mtod(mp, caddr_t);
989 		    mp->m_len -= adjust;
990 		    if (mp->m_len > 0 && adjust > 0)
991 			NFSMADV(mp, adjust);
992 		}
993 		if (zeroing)
994 		    mp->m_len = 0;
995 		else {
996 		    i += mp->m_len;
997 		    if (i > len) {
998 			mp->m_len -= (i - len);
999 			zeroing = 1;
1000 		    }
1001 		}
1002 		mp = mp->m_next;
1003 	    }
1004 	    if (len > NFS_MAXDATA || len < 0  || i < len) {
1005 nfsmout:
1006 		m_freem(mrep);
1007 		error = EIO;
1008 		nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1009 		if (v3)
1010 		    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1011 		nfsd->nd_mreq = mreq;
1012 		nfsd->nd_mrep = NULL;
1013 		nfsd->nd_time = 0;
1014 	    }
1015 
1016 	    /*
1017 	     * Add this entry to the hash and time queues.
1018 	     */
1019 	    s = splsoftclock();
1020 	    owp = NULL;
1021 	    wp = slp->ns_tq.lh_first;
1022 	    while (wp && wp->nd_time < nfsd->nd_time) {
1023 		owp = wp;
1024 		wp = wp->nd_tq.le_next;
1025 	    }
1026 	    if (owp) {
1027 		LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1028 	    } else {
1029 		LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1030 	    }
1031 	    if (nfsd->nd_mrep) {
1032 		wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1033 		owp = NULL;
1034 		wp = wpp->lh_first;
1035 		while (wp &&
1036 		    memcmp((caddr_t)&nfsd->nd_fh, (caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1037 		    owp = wp;
1038 		    wp = wp->nd_hash.le_next;
1039 		}
1040 		while (wp && wp->nd_off < nfsd->nd_off &&
1041 		    !memcmp((caddr_t)&nfsd->nd_fh, (caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1042 		    owp = wp;
1043 		    wp = wp->nd_hash.le_next;
1044 		}
1045 		if (owp) {
1046 		    LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1047 
1048 		    /*
1049 		     * Search the hash list for overlapping entries and
1050 		     * coalesce.
1051 		     */
1052 		    for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1053 			wp = nfsd->nd_hash.le_next;
1054 			if (NFSW_SAMECRED(owp, nfsd))
1055 			    nfsrvw_coalesce(owp, nfsd);
1056 		    }
1057 		} else {
1058 		    LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1059 		}
1060 	    }
1061 	    splx(s);
1062 	}
1063 
1064 	/*
1065 	 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1066 	 * and generate the associated reply mbuf list(s).
1067 	 */
1068 loop1:
1069 	cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1070 	s = splsoftclock();
1071 	for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = owp) {
1072 		owp = nfsd->nd_tq.le_next;
1073 		if (nfsd->nd_time > cur_usec)
1074 		    break;
1075 		if (nfsd->nd_mreq)
1076 		    continue;
1077 		LIST_REMOVE(nfsd, nd_tq);
1078 		LIST_REMOVE(nfsd, nd_hash);
1079 		splx(s);
1080 		mrep = nfsd->nd_mrep;
1081 		nfsd->nd_mrep = NULL;
1082 		cred = &nfsd->nd_cr;
1083 		v3 = (nfsd->nd_flag & ND_NFSV3);
1084 		forat_ret = aftat_ret = 1;
1085 		error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1086 		    nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH),
1087 		    FALSE);
1088 		if (!error) {
1089 		    if (v3)
1090 			forat_ret = VOP_GETATTR(vp, &forat, cred, procp);
1091 		    if (vp->v_type != VREG) {
1092 			if (v3)
1093 			    error = EINVAL;
1094 			else
1095 			    error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1096 		    }
1097 		} else
1098 		    vp = NULL;
1099 		if (!error) {
1100 		    nqsrv_getl(vp, ND_WRITE);
1101 		    error = nfsrv_access(vp, VWRITE, cred, rdonly, procp, 1);
1102 		}
1103 
1104 		if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1105 		    ioflags = IO_NODELOCKED;
1106 		else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1107 		    ioflags = (IO_SYNC | IO_NODELOCKED);
1108 		else
1109 		    ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1110 		uiop->uio_rw = UIO_WRITE;
1111 		uiop->uio_segflg = UIO_SYSSPACE;
1112 		uiop->uio_procp = (struct proc *)0;
1113 		uiop->uio_offset = nfsd->nd_off;
1114 		uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1115 		if (uiop->uio_resid > 0) {
1116 		    mp = mrep;
1117 		    i = 0;
1118 		    while (mp) {
1119 			if (mp->m_len > 0)
1120 			    i++;
1121 			mp = mp->m_next;
1122 		    }
1123 		    uiop->uio_iovcnt = i;
1124 		    iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
1125 		    uiop->uio_iov = ivp = iov;
1126 		    mp = mrep;
1127 		    while (mp) {
1128 			if (mp->m_len > 0) {
1129 			    ivp->iov_base = mtod(mp, caddr_t);
1130 			    ivp->iov_len = mp->m_len;
1131 			    ivp++;
1132 			}
1133 			mp = mp->m_next;
1134 		    }
1135 		    if (!error) {
1136 			error = VOP_WRITE(vp, uiop, ioflags, cred);
1137 			nfsstats.srvvop_writes++;
1138 		    }
1139 		    free((caddr_t)iov, M_TEMP);
1140 		}
1141 		m_freem(mrep);
1142 		if (vp) {
1143 		    aftat_ret = VOP_GETATTR(vp, &va, cred, procp);
1144 		    vput(vp);
1145 		}
1146 
1147 		/*
1148 		 * Loop around generating replies for all write rpcs that have
1149 		 * now been completed.
1150 		 */
1151 		swp = nfsd;
1152 		do {
1153 		    if (error) {
1154 			nfsm_writereply(NFSX_WCCDATA(v3), v3);
1155 			if (v3) {
1156 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1157 			}
1158 		    } else {
1159 			nfsm_writereply(NFSX_PREOPATTR(v3) +
1160 			    NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1161 			    NFSX_WRITEVERF(v3), v3);
1162 			if (v3) {
1163 			    nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1164 			    nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1165 			    *tl++ = txdr_unsigned(nfsd->nd_len);
1166 			    *tl++ = txdr_unsigned(swp->nd_stable);
1167 			    /*
1168 			     * Actually, there is no need to txdr these fields,
1169 			     * but it may make the values more human readable,
1170 			     * for debugging purposes.
1171 			     */
1172 			    *tl++ = txdr_unsigned(boottime.tv_sec);
1173 			    *tl = txdr_unsigned(boottime.tv_usec);
1174 			} else {
1175 			    nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1176 			    nfsm_srvfillattr(&va, fp);
1177 			}
1178 		    }
1179 		    nfsd->nd_mreq = mreq;
1180 		    if (nfsd->nd_mrep)
1181 			panic("nfsrv_write: nd_mrep not free");
1182 
1183 		    /*
1184 		     * Done. Put it at the head of the timer queue so that
1185 		     * the final phase can return the reply.
1186 		     */
1187 		    s = splsoftclock();
1188 		    if (nfsd != swp) {
1189 			nfsd->nd_time = 0;
1190 			LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1191 		    }
1192 		    nfsd = swp->nd_coalesce.lh_first;
1193 		    if (nfsd) {
1194 			LIST_REMOVE(nfsd, nd_tq);
1195 		    }
1196 		    splx(s);
1197 		} while (nfsd);
1198 		s = splsoftclock();
1199 		swp->nd_time = 0;
1200 		LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1201 		splx(s);
1202 		goto loop1;
1203 	}
1204 	splx(s);
1205 
1206 	/*
1207 	 * Search for a reply to return.
1208 	 */
1209 	s = splsoftclock();
1210 	for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = nfsd->nd_tq.le_next)
1211 		if (nfsd->nd_mreq) {
1212 		    LIST_REMOVE(nfsd, nd_tq);
1213 		    *mrq = nfsd->nd_mreq;
1214 		    *ndp = nfsd;
1215 		    break;
1216 		}
1217 	splx(s);
1218 	return (0);
1219 }
1220 
1221 /*
1222  * Coalesce the write request nfsd into owp. To do this we must:
1223  * - remove nfsd from the queues
1224  * - merge nfsd->nd_mrep into owp->nd_mrep
1225  * - update the nd_eoff and nd_stable for owp
1226  * - put nfsd on owp's nd_coalesce list
1227  * NB: Must be called at splsoftclock().
1228  */
1229 void
1230 nfsrvw_coalesce(owp, nfsd)
1231         struct nfsrv_descript *owp;
1232         struct nfsrv_descript *nfsd;
1233 {
1234         int overlap;
1235         struct mbuf *mp;
1236 
1237         LIST_REMOVE(nfsd, nd_hash);
1238         LIST_REMOVE(nfsd, nd_tq);
1239         if (owp->nd_eoff < nfsd->nd_eoff) {
1240             overlap = owp->nd_eoff - nfsd->nd_off;
1241             if (overlap < 0)
1242                 panic("nfsrv_coalesce: bad off");
1243             if (overlap > 0)
1244                 m_adj(nfsd->nd_mrep, overlap);
1245             mp = owp->nd_mrep;
1246             while (mp->m_next)
1247                 mp = mp->m_next;
1248             mp->m_next = nfsd->nd_mrep;
1249             owp->nd_eoff = nfsd->nd_eoff;
1250         } else
1251             m_freem(nfsd->nd_mrep);
1252         nfsd->nd_mrep = NULL;
1253         if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1254             owp->nd_stable = NFSV3WRITE_FILESYNC;
1255         else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1256             owp->nd_stable == NFSV3WRITE_UNSTABLE)
1257             owp->nd_stable = NFSV3WRITE_DATASYNC;
1258         LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1259  	/*
1260  	 * nfsd might hold coalesce elements! Move them to owp.
1261  	 * Otherwise, requests may be lost and clients will be stuck.
1262  	 */
1263  	if (nfsd->nd_coalesce.lh_first)
1264  	{
1265  		struct nfsrv_descript *m;
1266 
1267  		while ((m = nfsd->nd_coalesce.lh_first))
1268  		{
1269  			LIST_REMOVE(m, nd_tq);
1270  			LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq);
1271  		}
1272  	}
1273 }
1274 
1275 /*
1276  * nfs create service
1277  * now does a truncate to 0 length via. setattr if it already exists
1278  */
1279 int
1280 nfsrv_create(nfsd, slp, procp, mrq)
1281 	struct nfsrv_descript *nfsd;
1282 	struct nfssvc_sock *slp;
1283 	struct proc *procp;
1284 	struct mbuf **mrq;
1285 {
1286 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1287 	struct mbuf *nam = nfsd->nd_nam;
1288 	caddr_t dpos = nfsd->nd_dpos;
1289 	struct ucred *cred = &nfsd->nd_cr;
1290 	struct nfs_fattr *fp;
1291 	struct vattr va, dirfor, diraft;
1292 	struct nfsv2_sattr *sp;
1293 	u_int32_t *tl;
1294 	struct nameidata nd;
1295 	caddr_t cp;
1296 	int32_t t1;
1297 	caddr_t bpos;
1298 	int error = 0, cache, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1299 	int rdev = 0;
1300 	int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1301 	char *cp2;
1302 	struct mbuf *mb, *mb2, *mreq;
1303 	struct vnode *vp = NULL, *dirp = NULL;
1304 	nfsfh_t nfh;
1305 	fhandle_t *fhp;
1306 	u_quad_t frev, tempsize;
1307 	u_char cverf[NFSX_V3CREATEVERF];
1308 
1309 	nd.ni_cnd.cn_nameiop = 0;
1310 	fhp = &nfh.fh_generic;
1311 	nfsm_srvmtofh(fhp);
1312 	nfsm_srvnamesiz(len);
1313 	nd.ni_cnd.cn_cred = cred;
1314 	nd.ni_cnd.cn_nameiop = CREATE;
1315 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1316 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1317 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1318 	if (dirp) {
1319 		if (v3)
1320 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
1321 				procp);
1322 		else {
1323 			vrele(dirp);
1324 			dirp = (struct vnode *)0;
1325 		}
1326 	}
1327 	if (error) {
1328 		nfsm_reply(NFSX_WCCDATA(v3));
1329 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1330 		if (dirp)
1331 			vrele(dirp);
1332 		return (0);
1333 	}
1334 	VATTR_NULL(&va);
1335 	if (v3) {
1336 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1337 		how = fxdr_unsigned(int, *tl);
1338 		switch (how) {
1339 		case NFSV3CREATE_GUARDED:
1340 			if (nd.ni_vp) {
1341 				error = EEXIST;
1342 				break;
1343 			}
1344 		case NFSV3CREATE_UNCHECKED:
1345 			nfsm_srvsattr(&va);
1346 			break;
1347 		case NFSV3CREATE_EXCLUSIVE:
1348 			nfsm_dissect(cp, caddr_t, NFSX_V3CREATEVERF);
1349 			memcpy(cverf, cp, NFSX_V3CREATEVERF);
1350 			exclusive_flag = 1;
1351 			if (nd.ni_vp == NULL)
1352 				va.va_mode = 0;
1353 			break;
1354 		};
1355 		va.va_type = VREG;
1356 	} else {
1357 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1358 		va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1359 		if (va.va_type == VNON)
1360 			va.va_type = VREG;
1361 		va.va_mode = nfstov_mode(sp->sa_mode);
1362 		switch (va.va_type) {
1363 		case VREG:
1364 			tsize = fxdr_unsigned(int32_t, sp->sa_size);
1365 			if (tsize != -1)
1366 				va.va_size = (u_quad_t)tsize;
1367 			break;
1368 		case VCHR:
1369 		case VBLK:
1370 		case VFIFO:
1371 			rdev = fxdr_unsigned(int32_t, sp->sa_size);
1372 			break;
1373 		default:
1374 			break;
1375 		};
1376 	}
1377 
1378 	/*
1379 	 * Iff doesn't exist, create it
1380 	 * otherwise just truncate to 0 length
1381 	 *   should I set the mode too ??
1382 	 */
1383 	if (nd.ni_vp == NULL) {
1384 		if (va.va_type == VREG || va.va_type == VSOCK) {
1385 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1386 			error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1387 			if (!error) {
1388 				if (exclusive_flag) {
1389 					exclusive_flag = 0;
1390 					VATTR_NULL(&va);
1391 					memcpy((caddr_t)&va.va_atime, cverf,
1392 						NFSX_V3CREATEVERF);
1393 					error = VOP_SETATTR(nd.ni_vp, &va, cred,
1394 						procp);
1395 				}
1396 			}
1397 		} else if (va.va_type == VCHR || va.va_type == VBLK ||
1398 			va.va_type == VFIFO) {
1399 			if (va.va_type == VCHR && rdev == 0xffffffff)
1400 				va.va_type = VFIFO;
1401 			if (va.va_type != VFIFO &&
1402 			    (error = suser(cred, (u_short *)0))) {
1403 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1404 				vput(nd.ni_dvp);
1405 				nfsm_reply(0);
1406 				return (error);
1407 			} else
1408 				va.va_rdev = (dev_t)rdev;
1409 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1410 			error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd,
1411 			    &va);
1412 			if (error) {
1413 				nfsm_reply(0);
1414 			}
1415 			PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
1416 			if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1417 				vrele(nd.ni_dvp);
1418 				vput(nd.ni_vp);
1419 				VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1420 				error = EINVAL;
1421 				nfsm_reply(0);
1422 			}
1423 		} else {
1424 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1425 			vput(nd.ni_dvp);
1426 			error = ENXIO;
1427 		}
1428 		vp = nd.ni_vp;
1429 	} else {
1430 		vp = nd.ni_vp;
1431 		if (nd.ni_dvp == vp)
1432 			vrele(nd.ni_dvp);
1433 		else
1434 			vput(nd.ni_dvp);
1435 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1436 		if (!error && va.va_size != -1) {
1437 			error = nfsrv_access(vp, VWRITE, cred,
1438 			    (nd.ni_cnd.cn_flags & RDONLY), procp, 0);
1439 			if (!error) {
1440 				nqsrv_getl(vp, ND_WRITE);
1441 				tempsize = va.va_size;
1442 				VATTR_NULL(&va);
1443 				va.va_size = tempsize;
1444 				error = VOP_SETATTR(vp, &va, cred,
1445 					 procp);
1446 			}
1447 		}
1448 		if (error)
1449 			vput(vp);
1450 	}
1451 	if (!error) {
1452 		memset((caddr_t)fhp, 0, sizeof(nfh));
1453 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1454 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
1455 		if (!error)
1456 			error = VOP_GETATTR(vp, &va, cred, procp);
1457 		vput(vp);
1458 	}
1459 	if (v3) {
1460 		if (exclusive_flag && !error &&
1461 			memcmp(cverf, (caddr_t)&va.va_atime, NFSX_V3CREATEVERF))
1462 			error = EEXIST;
1463 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1464 		vrele(dirp);
1465 	}
1466 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1467 	if (v3) {
1468 		if (!error) {
1469 			nfsm_srvpostop_fh(fhp);
1470 			nfsm_srvpostop_attr(0, &va);
1471 		}
1472 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1473 	} else {
1474 		nfsm_srvfhtom(fhp, v3);
1475 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1476 		nfsm_srvfillattr(&va, fp);
1477 	}
1478 	return (0);
1479 nfsmout:
1480 	if (dirp)
1481 		vrele(dirp);
1482 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1483 	if (nd.ni_dvp == nd.ni_vp)
1484 		vrele(nd.ni_dvp);
1485 	else
1486 		vput(nd.ni_dvp);
1487 	if (nd.ni_vp)
1488 		vput(nd.ni_vp);
1489 	return (error);
1490 }
1491 
1492 /*
1493  * nfs v3 mknod service
1494  */
1495 int
1496 nfsrv_mknod(nfsd, slp, procp, mrq)
1497 	struct nfsrv_descript *nfsd;
1498 	struct nfssvc_sock *slp;
1499 	struct proc *procp;
1500 	struct mbuf **mrq;
1501 {
1502 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1503 	struct mbuf *nam = nfsd->nd_nam;
1504 	caddr_t dpos = nfsd->nd_dpos;
1505 	struct ucred *cred = &nfsd->nd_cr;
1506 	struct vattr va, dirfor, diraft;
1507 	u_int32_t *tl;
1508 	struct nameidata nd;
1509 	int32_t t1;
1510 	caddr_t bpos;
1511 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
1512 	u_int32_t major, minor;
1513 	enum vtype vtyp;
1514 	char *cp2;
1515 	struct mbuf *mb, *mb2, *mreq;
1516 	struct vnode *vp, *dirp = (struct vnode *)0;
1517 	nfsfh_t nfh;
1518 	fhandle_t *fhp;
1519 	u_quad_t frev;
1520 
1521 	nd.ni_cnd.cn_nameiop = 0;
1522 	fhp = &nfh.fh_generic;
1523 	nfsm_srvmtofh(fhp);
1524 	nfsm_srvnamesiz(len);
1525 	nd.ni_cnd.cn_cred = cred;
1526 	nd.ni_cnd.cn_nameiop = CREATE;
1527 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1528 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1529 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1530 	if (dirp)
1531 		dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, procp);
1532 	if (error) {
1533 		nfsm_reply(NFSX_WCCDATA(1));
1534 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1535 		if (dirp)
1536 			vrele(dirp);
1537 		return (0);
1538 	}
1539 	nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1540 	vtyp = nfsv3tov_type(*tl);
1541 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1542 		error = NFSERR_BADTYPE;
1543 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1544 		if (nd.ni_dvp == nd.ni_vp)
1545 			vrele(nd.ni_dvp);
1546 		else
1547 			vput(nd.ni_dvp);
1548 		if (nd.ni_vp)
1549 			vput(nd.ni_vp);
1550 		goto out;
1551 	}
1552 	VATTR_NULL(&va);
1553 	nfsm_srvsattr(&va);
1554 	if (vtyp == VCHR || vtyp == VBLK) {
1555 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1556 		major = fxdr_unsigned(u_int32_t, *tl++);
1557 		minor = fxdr_unsigned(u_int32_t, *tl);
1558 		va.va_rdev = makedev(major, minor);
1559 	}
1560 
1561 	/*
1562 	 * Iff doesn't exist, create it.
1563 	 */
1564 	if (nd.ni_vp) {
1565 		error = EEXIST;
1566 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1567 		if (nd.ni_dvp == nd.ni_vp)
1568 			vrele(nd.ni_dvp);
1569 		else
1570 			vput(nd.ni_dvp);
1571 		vput(nd.ni_vp);
1572 		goto out;
1573 	}
1574 	va.va_type = vtyp;
1575 	if (vtyp == VSOCK) {
1576 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1577 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1578 	} else {
1579 		if (va.va_type != VFIFO &&
1580 		    (error = suser(cred, (u_short *)0))) {
1581 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1582 			vput(nd.ni_dvp);
1583 			goto out;
1584 		}
1585 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
1586 		error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1587 		if (error) {
1588 			goto out;
1589 		}
1590 		if (error)
1591 			goto out;
1592 		if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1593 			vput(nd.ni_vp);
1594 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1595 			error = EINVAL;
1596 		}
1597 	}
1598 out:
1599 	vp = nd.ni_vp;
1600 	if (!error) {
1601 		memset((caddr_t)fhp, 0, sizeof(nfh));
1602 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1603 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
1604 		if (!error)
1605 			error = VOP_GETATTR(vp, &va, cred, procp);
1606 		vput(vp);
1607 	}
1608 	diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1609 	vrele(dirp);
1610 	nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
1611 	if (!error) {
1612 		nfsm_srvpostop_fh(fhp);
1613 		nfsm_srvpostop_attr(0, &va);
1614 	}
1615 	nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1616 	return (0);
1617 nfsmout:
1618 	if (dirp)
1619 		vrele(dirp);
1620 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1621 	if (nd.ni_dvp == nd.ni_vp)
1622 		vrele(nd.ni_dvp);
1623 	else
1624 		vput(nd.ni_dvp);
1625 	if (nd.ni_vp)
1626 		vput(nd.ni_vp);
1627 	return (error);
1628 }
1629 
1630 /*
1631  * nfs remove service
1632  */
1633 int
1634 nfsrv_remove(nfsd, slp, procp, mrq)
1635 	struct nfsrv_descript *nfsd;
1636 	struct nfssvc_sock *slp;
1637 	struct proc *procp;
1638 	struct mbuf **mrq;
1639 {
1640 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1641 	struct mbuf *nam = nfsd->nd_nam;
1642 	caddr_t dpos = nfsd->nd_dpos;
1643 	struct ucred *cred = &nfsd->nd_cr;
1644 	struct nameidata nd;
1645 	u_int32_t *tl;
1646 	int32_t t1;
1647 	caddr_t bpos;
1648 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
1649 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1650 	char *cp2;
1651 	struct mbuf *mb, *mreq;
1652 	struct vnode *vp, *dirp;
1653 	struct vattr dirfor, diraft;
1654 	nfsfh_t nfh;
1655 	fhandle_t *fhp;
1656 	u_quad_t frev;
1657 
1658 #ifndef nolint
1659 	vp = (struct vnode *)0;
1660 #endif
1661 	fhp = &nfh.fh_generic;
1662 	nfsm_srvmtofh(fhp);
1663 	nfsm_srvnamesiz(len);
1664 	nd.ni_cnd.cn_cred = cred;
1665 	nd.ni_cnd.cn_nameiop = DELETE;
1666 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1667 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1668 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1669 	if (dirp) {
1670 		if (v3)
1671 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
1672 				procp);
1673 		else
1674 			vrele(dirp);
1675 	}
1676 	if (!error) {
1677 		vp = nd.ni_vp;
1678 		if (vp->v_type == VDIR &&
1679 		    (error = suser(cred, (u_short *)0)) != 0)
1680 			goto out;
1681 		/*
1682 		 * The root of a mounted filesystem cannot be deleted.
1683 		 */
1684 		if (vp->v_flag & VROOT) {
1685 			error = EBUSY;
1686 			goto out;
1687 		}
1688 out:
1689 		if (!error) {
1690 			nqsrv_getl(nd.ni_dvp, ND_WRITE);
1691 			nqsrv_getl(vp, ND_WRITE);
1692 			error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
1693 		} else {
1694 			VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1695 			if (nd.ni_dvp == vp)
1696 				vrele(nd.ni_dvp);
1697 			else
1698 				vput(nd.ni_dvp);
1699 			vput(vp);
1700 		}
1701 	}
1702 	if (dirp && v3) {
1703 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1704 		vrele(dirp);
1705 	}
1706 	nfsm_reply(NFSX_WCCDATA(v3));
1707 	if (v3) {
1708 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1709 		return (0);
1710 	}
1711 	nfsm_srvdone;
1712 }
1713 
1714 /*
1715  * nfs rename service
1716  */
1717 int
1718 nfsrv_rename(nfsd, slp, procp, mrq)
1719 	struct nfsrv_descript *nfsd;
1720 	struct nfssvc_sock *slp;
1721 	struct proc *procp;
1722 	struct mbuf **mrq;
1723 {
1724 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1725 	struct mbuf *nam = nfsd->nd_nam;
1726 	caddr_t dpos = nfsd->nd_dpos;
1727 	struct ucred *cred = &nfsd->nd_cr;
1728 	u_int32_t *tl;
1729 	int32_t t1;
1730 	caddr_t bpos;
1731 	int error = 0, cache, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
1732 	int tdirfor_ret = 1, tdiraft_ret = 1;
1733 	int v3 = (nfsd->nd_flag & ND_NFSV3);
1734 	char *cp2;
1735 	struct mbuf *mb, *mreq;
1736 	struct nameidata fromnd, tond;
1737 	struct vnode *fvp, *tvp, *tdvp, *fdirp = (struct vnode *)0;
1738 	struct vnode *tdirp = (struct vnode *)0;
1739 	struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
1740 	nfsfh_t fnfh, tnfh;
1741 	fhandle_t *ffhp, *tfhp;
1742 	u_quad_t frev;
1743 	uid_t saved_uid;
1744 
1745 #ifndef nolint
1746 	fvp = (struct vnode *)0;
1747 #endif
1748 	ffhp = &fnfh.fh_generic;
1749 	tfhp = &tnfh.fh_generic;
1750 	fromnd.ni_cnd.cn_nameiop = 0;
1751 	tond.ni_cnd.cn_nameiop = 0;
1752 	nfsm_srvmtofh(ffhp);
1753 	nfsm_srvnamesiz(len);
1754 	/*
1755 	 * Remember our original uid so that we can reset cr_uid before
1756 	 * the second nfs_namei() call, in case it is remapped.
1757 	 */
1758 	saved_uid = cred->cr_uid;
1759 	fromnd.ni_cnd.cn_cred = cred;
1760 	fromnd.ni_cnd.cn_nameiop = DELETE;
1761 	fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART;
1762 	error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md,
1763 		&dpos, &fdirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1764 	if (fdirp) {
1765 		if (v3)
1766 			fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred,
1767 				procp);
1768 		else {
1769 			vrele(fdirp);
1770 			fdirp = (struct vnode *)0;
1771 		}
1772 	}
1773 	if (error) {
1774 		nfsm_reply(2 * NFSX_WCCDATA(v3));
1775 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1776 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1777 		if (fdirp)
1778 			vrele(fdirp);
1779 		return (0);
1780 	}
1781 	fvp = fromnd.ni_vp;
1782 	nfsm_srvmtofh(tfhp);
1783 	nfsm_strsiz(len2, NFS_MAXNAMLEN);
1784 	cred->cr_uid = saved_uid;
1785 	tond.ni_cnd.cn_cred = cred;
1786 	tond.ni_cnd.cn_nameiop = RENAME;
1787 	tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART;
1788 	error = nfs_namei(&tond, tfhp, len2, slp, nam, &md,
1789 		&dpos, &tdirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1790 	if (tdirp) {
1791 		if (v3)
1792 			tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred,
1793 				procp);
1794 		else {
1795 			vrele(tdirp);
1796 			tdirp = (struct vnode *)0;
1797 		}
1798 	}
1799 	if (error) {
1800 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1801 		vrele(fromnd.ni_dvp);
1802 		vrele(fvp);
1803 		goto out1;
1804 	}
1805 	tdvp = tond.ni_dvp;
1806 	tvp = tond.ni_vp;
1807 	if (tvp != NULL) {
1808 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1809 			if (v3)
1810 				error = EEXIST;
1811 			else
1812 				error = EISDIR;
1813 			goto out;
1814 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1815 			if (v3)
1816 				error = EEXIST;
1817 			else
1818 				error = ENOTDIR;
1819 			goto out;
1820 		}
1821 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1822 			if (v3)
1823 				error = EXDEV;
1824 			else
1825 				error = ENOTEMPTY;
1826 			goto out;
1827 		}
1828 	}
1829 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1830 		if (v3)
1831 			error = EXDEV;
1832 		else
1833 			error = ENOTEMPTY;
1834 		goto out;
1835 	}
1836 	if (fvp->v_mount != tdvp->v_mount) {
1837 		if (v3)
1838 			error = EXDEV;
1839 		else
1840 			error = ENOTEMPTY;
1841 		goto out;
1842 	}
1843 	if (fvp == tdvp) {
1844 		if (v3)
1845 			error = EINVAL;
1846 		else
1847 			error = ENOTEMPTY;
1848 	}
1849 	/*
1850 	 * If source is the same as the destination (that is the
1851 	 * same vnode with the same name in the same directory),
1852 	 * then there is nothing to do.
1853 	 */
1854 	if (fvp == tvp && fromnd.ni_dvp == tdvp &&
1855 	    fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
1856 	    !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
1857 	      fromnd.ni_cnd.cn_namelen))
1858 		error = -1;
1859 out:
1860 	if (!error) {
1861 		nqsrv_getl(fromnd.ni_dvp, ND_WRITE);
1862 		nqsrv_getl(tdvp, ND_WRITE);
1863 		if (tvp) {
1864 			nqsrv_getl(tvp, ND_WRITE);
1865 		}
1866 		error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
1867 				   tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
1868 	} else {
1869 		VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd);
1870 		if (tdvp == tvp)
1871 			vrele(tdvp);
1872 		else
1873 			vput(tdvp);
1874 		if (tvp)
1875 			vput(tvp);
1876 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1877 		vrele(fromnd.ni_dvp);
1878 		vrele(fvp);
1879 		if (error == -1)
1880 			error = 0;
1881 	}
1882 	vrele(tond.ni_startdir);
1883 	PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
1884 out1:
1885 	if (fdirp) {
1886 		fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, procp);
1887 		vrele(fdirp);
1888 	}
1889 	if (tdirp) {
1890 		tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, procp);
1891 		vrele(tdirp);
1892 	}
1893 	vrele(fromnd.ni_startdir);
1894 	PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
1895 	nfsm_reply(2 * NFSX_WCCDATA(v3));
1896 	if (v3) {
1897 		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1898 		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1899 	}
1900 	return (0);
1901 
1902 nfsmout:
1903 	if (fdirp)
1904 		vrele(fdirp);
1905 	if (tdirp)
1906 		vrele(tdirp);
1907 	if (tond.ni_cnd.cn_nameiop) {
1908 		vrele(tond.ni_startdir);
1909 		PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
1910 	}
1911 	if (fromnd.ni_cnd.cn_nameiop) {
1912 		vrele(fromnd.ni_startdir);
1913 		PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
1914 		VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1915 		vrele(fromnd.ni_dvp);
1916 		vrele(fvp);
1917 	}
1918 	return (error);
1919 }
1920 
1921 /*
1922  * nfs link service
1923  */
1924 int
1925 nfsrv_link(nfsd, slp, procp, mrq)
1926 	struct nfsrv_descript *nfsd;
1927 	struct nfssvc_sock *slp;
1928 	struct proc *procp;
1929 	struct mbuf **mrq;
1930 {
1931 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1932 	struct mbuf *nam = nfsd->nd_nam;
1933 	caddr_t dpos = nfsd->nd_dpos;
1934 	struct ucred *cred = &nfsd->nd_cr;
1935 	struct nameidata nd;
1936 	u_int32_t *tl;
1937 	int32_t t1;
1938 	caddr_t bpos;
1939 	int error = 0, rdonly, cache, len, dirfor_ret = 1, diraft_ret = 1;
1940 	int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
1941 	char *cp2;
1942 	struct mbuf *mb, *mreq;
1943 	struct vnode *vp, *xp, *dirp = (struct vnode *)0;
1944 	struct vattr dirfor, diraft, at;
1945 	nfsfh_t nfh, dnfh;
1946 	fhandle_t *fhp, *dfhp;
1947 	u_quad_t frev;
1948 
1949 	fhp = &nfh.fh_generic;
1950 	dfhp = &dnfh.fh_generic;
1951 	nfsm_srvmtofh(fhp);
1952 	nfsm_srvmtofh(dfhp);
1953 	nfsm_srvnamesiz(len);
1954 	error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, slp, nam,
1955 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1956 	if (error) {
1957 		nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
1958 		nfsm_srvpostop_attr(getret, &at);
1959 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1960 		return (0);
1961 	}
1962 	if (vp->v_type == VDIR && (error = suser(cred, (u_short *)0)) != 0)
1963 		goto out1;
1964 	nd.ni_cnd.cn_cred = cred;
1965 	nd.ni_cnd.cn_nameiop = CREATE;
1966 	nd.ni_cnd.cn_flags = LOCKPARENT;
1967 	error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos,
1968 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1969 	if (dirp) {
1970 		if (v3)
1971 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
1972 				procp);
1973 		else {
1974 			vrele(dirp);
1975 			dirp = (struct vnode *)0;
1976 		}
1977 	}
1978 	if (error)
1979 		goto out1;
1980 	xp = nd.ni_vp;
1981 	if (xp != NULL) {
1982 		error = EEXIST;
1983 		goto out;
1984 	}
1985 	xp = nd.ni_dvp;
1986 	if (vp->v_mount != xp->v_mount)
1987 		error = EXDEV;
1988 out:
1989 	if (!error) {
1990 		nqsrv_getl(vp, ND_WRITE);
1991 		nqsrv_getl(xp, ND_WRITE);
1992 		error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
1993 	} else {
1994 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1995 		if (nd.ni_dvp == nd.ni_vp)
1996 			vrele(nd.ni_dvp);
1997 		else
1998 			vput(nd.ni_dvp);
1999 		if (nd.ni_vp)
2000 			vrele(nd.ni_vp);
2001 	}
2002 out1:
2003 	if (v3)
2004 		getret = VOP_GETATTR(vp, &at, cred, procp);
2005 	if (dirp) {
2006 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2007 		vrele(dirp);
2008 	}
2009 	vrele(vp);
2010 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2011 	if (v3) {
2012 		nfsm_srvpostop_attr(getret, &at);
2013 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2014 		return (0);
2015 	}
2016 	nfsm_srvdone;
2017 }
2018 
2019 /*
2020  * nfs symbolic link service
2021  */
2022 int
2023 nfsrv_symlink(nfsd, slp, procp, mrq)
2024 	struct nfsrv_descript *nfsd;
2025 	struct nfssvc_sock *slp;
2026 	struct proc *procp;
2027 	struct mbuf **mrq;
2028 {
2029 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2030 	struct mbuf *nam = nfsd->nd_nam;
2031 	caddr_t dpos = nfsd->nd_dpos;
2032 	struct ucred *cred = &nfsd->nd_cr;
2033 	struct vattr va, dirfor, diraft;
2034 	struct nameidata nd;
2035 	u_int32_t *tl;
2036 	int32_t t1;
2037 	struct nfsv2_sattr *sp;
2038 	char *bpos, *pathcp = NULL, *cp2;
2039 	struct uio io;
2040 	struct iovec iv;
2041 	int error = 0, cache, len, len2, dirfor_ret = 1, diraft_ret = 1;
2042 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2043 	struct mbuf *mb, *mreq, *mb2;
2044 	struct vnode *dirp = (struct vnode *)0;
2045 	nfsfh_t nfh;
2046 	fhandle_t *fhp;
2047 	u_quad_t frev;
2048 
2049 	nd.ni_cnd.cn_nameiop = 0;
2050 	fhp = &nfh.fh_generic;
2051 	nfsm_srvmtofh(fhp);
2052 	nfsm_srvnamesiz(len);
2053 	nd.ni_cnd.cn_cred = cred;
2054 	nd.ni_cnd.cn_nameiop = CREATE;
2055 	nd.ni_cnd.cn_flags = LOCKPARENT;
2056 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2057 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2058 	if (dirp) {
2059 		if (v3)
2060 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2061 				procp);
2062 		else {
2063 			vrele(dirp);
2064 			dirp = (struct vnode *)0;
2065 		}
2066 	}
2067 	if (error)
2068 		goto out;
2069 	VATTR_NULL(&va);
2070 	if (v3)
2071 		nfsm_srvsattr(&va);
2072 	nfsm_strsiz(len2, NFS_MAXPATHLEN);
2073 	pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK);
2074 	iv.iov_base = pathcp;
2075 	iv.iov_len = len2;
2076 	io.uio_resid = len2;
2077 	io.uio_offset = 0;
2078 	io.uio_iov = &iv;
2079 	io.uio_iovcnt = 1;
2080 	io.uio_segflg = UIO_SYSSPACE;
2081 	io.uio_rw = UIO_READ;
2082 	io.uio_procp = (struct proc *)0;
2083 	nfsm_mtouio(&io, len2);
2084 	if (!v3) {
2085 		nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2086 		va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
2087 	}
2088 	*(pathcp + len2) = '\0';
2089 	if (nd.ni_vp) {
2090 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2091 		if (nd.ni_dvp == nd.ni_vp)
2092 			vrele(nd.ni_dvp);
2093 		else
2094 			vput(nd.ni_dvp);
2095 		vrele(nd.ni_vp);
2096 		error = EEXIST;
2097 		goto out;
2098 	}
2099 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2100 	error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp);
2101 	if (!error) {
2102 	    if (v3) {
2103 		memset((caddr_t)fhp, 0, sizeof(nfh));
2104 		fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2105 		error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2106 		if (!error)
2107 		    error = VOP_GETATTR(nd.ni_vp, &va, cred,
2108 					procp);
2109 		vput(nd.ni_vp);
2110 	    } else {
2111 		vput(nd.ni_vp);
2112 	    }
2113 	}
2114 out:
2115 	if (pathcp)
2116 		free(pathcp, M_TEMP);
2117 	if (dirp) {
2118 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2119 		vrele(dirp);
2120 	}
2121 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2122 	if (v3) {
2123 		if (!error) {
2124 			nfsm_srvpostop_fh(fhp);
2125 			nfsm_srvpostop_attr(0, &va);
2126 		}
2127 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2128 	}
2129 	return (0);
2130 nfsmout:
2131 	if (dirp)
2132 		vrele(dirp);
2133 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2134 	if (nd.ni_dvp == nd.ni_vp)
2135 		vrele(nd.ni_dvp);
2136 	else
2137 		vput(nd.ni_dvp);
2138 	if (nd.ni_vp)
2139 		vrele(nd.ni_vp);
2140 	if (pathcp)
2141 		free(pathcp, M_TEMP);
2142 	return (error);
2143 }
2144 
2145 /*
2146  * nfs mkdir service
2147  */
2148 int
2149 nfsrv_mkdir(nfsd, slp, procp, mrq)
2150 	struct nfsrv_descript *nfsd;
2151 	struct nfssvc_sock *slp;
2152 	struct proc *procp;
2153 	struct mbuf **mrq;
2154 {
2155 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2156 	struct mbuf *nam = nfsd->nd_nam;
2157 	caddr_t dpos = nfsd->nd_dpos;
2158 	struct ucred *cred = &nfsd->nd_cr;
2159 	struct vattr va, dirfor, diraft;
2160 	struct nfs_fattr *fp;
2161 	struct nameidata nd;
2162 	caddr_t cp;
2163 	u_int32_t *tl;
2164 	int32_t t1;
2165 	caddr_t bpos;
2166 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2167 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2168 	char *cp2;
2169 	struct mbuf *mb, *mb2, *mreq;
2170 	struct vnode *vp, *dirp = (struct vnode *)0;
2171 	nfsfh_t nfh;
2172 	fhandle_t *fhp;
2173 	u_quad_t frev;
2174 
2175 	fhp = &nfh.fh_generic;
2176 	nfsm_srvmtofh(fhp);
2177 	nfsm_srvnamesiz(len);
2178 	nd.ni_cnd.cn_cred = cred;
2179 	nd.ni_cnd.cn_nameiop = CREATE;
2180 	nd.ni_cnd.cn_flags = LOCKPARENT;
2181 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2182 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2183 	if (dirp) {
2184 		if (v3)
2185 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2186 				procp);
2187 		else {
2188 			vrele(dirp);
2189 			dirp = (struct vnode *)0;
2190 		}
2191 	}
2192 	if (error) {
2193 		nfsm_reply(NFSX_WCCDATA(v3));
2194 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2195 		if (dirp)
2196 			vrele(dirp);
2197 		return (0);
2198 	}
2199 	VATTR_NULL(&va);
2200 	if (v3) {
2201 		nfsm_srvsattr(&va);
2202 	} else {
2203 		nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2204 		va.va_mode = nfstov_mode(*tl++);
2205 	}
2206 	va.va_type = VDIR;
2207 	vp = nd.ni_vp;
2208 	if (vp != NULL) {
2209 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2210 		if (nd.ni_dvp == vp)
2211 			vrele(nd.ni_dvp);
2212 		else
2213 			vput(nd.ni_dvp);
2214 		vrele(vp);
2215 		error = EEXIST;
2216 		goto out;
2217 	}
2218 	nqsrv_getl(nd.ni_dvp, ND_WRITE);
2219 	error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
2220 	if (!error) {
2221 		vp = nd.ni_vp;
2222 		memset((caddr_t)fhp, 0, sizeof(nfh));
2223 		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2224 		error = VFS_VPTOFH(vp, &fhp->fh_fid);
2225 		if (!error)
2226 			error = VOP_GETATTR(vp, &va, cred, procp);
2227 		vput(vp);
2228 	}
2229 out:
2230 	if (dirp) {
2231 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2232 		vrele(dirp);
2233 	}
2234 	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2235 	if (v3) {
2236 		if (!error) {
2237 			nfsm_srvpostop_fh(fhp);
2238 			nfsm_srvpostop_attr(0, &va);
2239 		}
2240 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2241 	} else {
2242 		nfsm_srvfhtom(fhp, v3);
2243 		nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2244 		nfsm_srvfillattr(&va, fp);
2245 	}
2246 	return (0);
2247 nfsmout:
2248 	if (dirp)
2249 		vrele(dirp);
2250 	VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2251 	if (nd.ni_dvp == nd.ni_vp)
2252 		vrele(nd.ni_dvp);
2253 	else
2254 		vput(nd.ni_dvp);
2255 	if (nd.ni_vp)
2256 		vrele(nd.ni_vp);
2257 	return (error);
2258 }
2259 
2260 /*
2261  * nfs rmdir service
2262  */
2263 int
2264 nfsrv_rmdir(nfsd, slp, procp, mrq)
2265 	struct nfsrv_descript *nfsd;
2266 	struct nfssvc_sock *slp;
2267 	struct proc *procp;
2268 	struct mbuf **mrq;
2269 {
2270 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2271 	struct mbuf *nam = nfsd->nd_nam;
2272 	caddr_t dpos = nfsd->nd_dpos;
2273 	struct ucred *cred = &nfsd->nd_cr;
2274 	u_int32_t *tl;
2275 	int32_t t1;
2276 	caddr_t bpos;
2277 	int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2278 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2279 	char *cp2;
2280 	struct mbuf *mb, *mreq;
2281 	struct vnode *vp, *dirp = (struct vnode *)0;
2282 	struct vattr dirfor, diraft;
2283 	nfsfh_t nfh;
2284 	fhandle_t *fhp;
2285 	struct nameidata nd;
2286 	u_quad_t frev;
2287 
2288 	fhp = &nfh.fh_generic;
2289 	nfsm_srvmtofh(fhp);
2290 	nfsm_srvnamesiz(len);
2291 	nd.ni_cnd.cn_cred = cred;
2292 	nd.ni_cnd.cn_nameiop = DELETE;
2293 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
2294 	error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2295 		&dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2296 	if (dirp) {
2297 		if (v3)
2298 			dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2299 				procp);
2300 		else {
2301 			vrele(dirp);
2302 			dirp = (struct vnode *)0;
2303 		}
2304 	}
2305 	if (error) {
2306 		nfsm_reply(NFSX_WCCDATA(v3));
2307 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2308 		if (dirp)
2309 			vrele(dirp);
2310 		return (0);
2311 	}
2312 	vp = nd.ni_vp;
2313 	if (vp->v_type != VDIR) {
2314 		error = ENOTDIR;
2315 		goto out;
2316 	}
2317 	/*
2318 	 * No rmdir "." please.
2319 	 */
2320 	if (nd.ni_dvp == vp) {
2321 		error = EINVAL;
2322 		goto out;
2323 	}
2324 	/*
2325 	 * The root of a mounted filesystem cannot be deleted.
2326 	 */
2327 	if (vp->v_flag & VROOT)
2328 		error = EBUSY;
2329 out:
2330 	if (!error) {
2331 		nqsrv_getl(nd.ni_dvp, ND_WRITE);
2332 		nqsrv_getl(vp, ND_WRITE);
2333 		error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2334 	} else {
2335 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2336 		if (nd.ni_dvp == nd.ni_vp)
2337 			vrele(nd.ni_dvp);
2338 		else
2339 			vput(nd.ni_dvp);
2340 		vput(vp);
2341 	}
2342 	if (dirp) {
2343 		diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2344 		vrele(dirp);
2345 	}
2346 	nfsm_reply(NFSX_WCCDATA(v3));
2347 	if (v3) {
2348 		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2349 		return (0);
2350 	}
2351 	nfsm_srvdone;
2352 }
2353 
2354 /*
2355  * nfs readdir service
2356  * - mallocs what it thinks is enough to read
2357  *	count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
2358  * - calls VOP_READDIR()
2359  * - loops around building the reply
2360  *	if the output generated exceeds count break out of loop
2361  *	The nfsm_clget macro is used here so that the reply will be packed
2362  *	tightly in mbuf clusters.
2363  * - it only knows that it has encountered eof when the VOP_READDIR()
2364  *	reads nothing
2365  * - as such one readdir rpc will return eof false although you are there
2366  *	and then the next will return eof
2367  * - it trims out records with d_fileno == 0
2368  *	this doesn't matter for Unix clients, but they might confuse clients
2369  *	for other os'.
2370  * - it trims out records with d_type == DT_WHT
2371  *	these cannot be seen through NFS (unless we extend the protocol)
2372  * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2373  *	than requested, but this may not apply to all filesystems. For
2374  *	example, client NFS does not { although it is never remote mounted
2375  *	anyhow }
2376  *     The alternate call nfsrv_readdirplus() does lookups as well.
2377  * PS: The NFS protocol spec. does not clarify what the "count" byte
2378  *	argument is a count of.. just name strings and file id's or the
2379  *	entire reply rpc or ...
2380  *	I tried just file name and id sizes and it confused the Sun client,
2381  *	so I am using the full rpc size now. The "paranoia.." comment refers
2382  *	to including the status longwords that are not a part of the dir.
2383  *	"entry" structures, but are in the rpc.
2384  */
2385 struct flrep {
2386 	nfsuint64 fl_off;
2387 	u_int32_t fl_postopok;
2388 	u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2389 	u_int32_t fl_fhok;
2390 	u_int32_t fl_fhsize;
2391 	u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
2392 };
2393 
2394 int
2395 nfsrv_readdir(nfsd, slp, procp, mrq)
2396 	struct nfsrv_descript *nfsd;
2397 	struct nfssvc_sock *slp;
2398 	struct proc *procp;
2399 	struct mbuf **mrq;
2400 {
2401 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2402 	struct mbuf *nam = nfsd->nd_nam;
2403 	caddr_t dpos = nfsd->nd_dpos;
2404 	struct ucred *cred = &nfsd->nd_cr;
2405 	char *bp, *be;
2406 	struct mbuf *mp;
2407 	struct dirent *dp;
2408 	caddr_t cp;
2409 	u_int32_t *tl;
2410 	int32_t t1;
2411 	caddr_t bpos;
2412 	struct mbuf *mb, *mb2, *mreq, *mp2;
2413 	char *cpos, *cend, *cp2, *rbuf;
2414 	struct vnode *vp;
2415 	struct vattr at;
2416 	nfsfh_t nfh;
2417 	fhandle_t *fhp;
2418 	struct uio io;
2419 	struct iovec iv;
2420 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2421 	int siz, cnt, fullsiz, eofflag, rdonly, cache, ncookies;
2422 	int v3 = (nfsd->nd_flag & ND_NFSV3);
2423 	u_quad_t frev, off, toff, verf;
2424 	off_t *cookies = NULL, *cookiep;
2425 	nfsuint64 jar;
2426 
2427 	fhp = &nfh.fh_generic;
2428 	nfsm_srvmtofh(fhp);
2429 	if (v3) {
2430 		nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2431 		toff = fxdr_hyper(tl);
2432 		tl += 2;
2433 		verf = fxdr_hyper(tl);
2434 		tl += 2;
2435 	} else {
2436 		nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2437 		toff = fxdr_unsigned(u_quad_t, *tl++);
2438 	}
2439 	off = toff;
2440 	cnt = fxdr_unsigned(int, *tl);
2441 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2442 	xfer = NFS_SRVMAXDATA(nfsd);
2443 	if (siz > xfer)
2444 		siz = xfer;
2445 	fullsiz = siz;
2446 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
2447 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2448 	if (!error && vp->v_type != VDIR) {
2449 		error = ENOTDIR;
2450 		vput(vp);
2451 	}
2452 	if (error) {
2453 		nfsm_reply(NFSX_UNSIGNED);
2454 		nfsm_srvpostop_attr(getret, &at);
2455 		return (0);
2456 	}
2457 	nqsrv_getl(vp, ND_READ);
2458 	if (v3) {
2459 		error = getret = VOP_GETATTR(vp, &at, cred, procp);
2460 #ifdef NFS3_STRICTVERF
2461 		/*
2462 		 * XXX This check is too strict for Solaris 2.5 clients.
2463 		 */
2464 		if (!error && toff && verf != at.va_filerev)
2465 			error = NFSERR_BAD_COOKIE;
2466 #endif
2467 	}
2468 	if (!error)
2469 		error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0);
2470 	if (error) {
2471 		vput(vp);
2472 		nfsm_reply(NFSX_POSTOPATTR(v3));
2473 		nfsm_srvpostop_attr(getret, &at);
2474 		return (0);
2475 	}
2476 	VOP_UNLOCK(vp, 0);
2477 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2478 again:
2479 	iv.iov_base = rbuf;
2480 	iv.iov_len = fullsiz;
2481 	io.uio_iov = &iv;
2482 	io.uio_iovcnt = 1;
2483 	io.uio_offset = (off_t)off;
2484 	io.uio_resid = fullsiz;
2485 	io.uio_segflg = UIO_SYSSPACE;
2486 	io.uio_rw = UIO_READ;
2487 	io.uio_procp = (struct proc *)0;
2488 	eofflag = 0;
2489 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2490 
2491 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2492 
2493 	off = (off_t)io.uio_offset;
2494 	if (!cookies && !error)
2495 		error = NFSERR_PERM;
2496 	if (v3) {
2497 		getret = VOP_GETATTR(vp, &at, cred, procp);
2498 		if (!error)
2499 			error = getret;
2500 	}
2501 
2502 	VOP_UNLOCK(vp, 0);
2503 	if (error) {
2504 		vrele(vp);
2505 		free((caddr_t)rbuf, M_TEMP);
2506 		if (cookies)
2507 			free((caddr_t)cookies, M_TEMP);
2508 		nfsm_reply(NFSX_POSTOPATTR(v3));
2509 		nfsm_srvpostop_attr(getret, &at);
2510 		return (0);
2511 	}
2512 	if (io.uio_resid) {
2513 		siz -= io.uio_resid;
2514 
2515 		/*
2516 		 * If nothing read, return eof
2517 		 * rpc reply
2518 		 */
2519 		if (siz == 0) {
2520 			vrele(vp);
2521 			nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2522 				2 * NFSX_UNSIGNED);
2523 			if (v3) {
2524 				nfsm_srvpostop_attr(getret, &at);
2525 				nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2526 				txdr_hyper(at.va_filerev, tl);
2527 				tl += 2;
2528 			} else
2529 				nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2530 			*tl++ = nfs_false;
2531 			*tl = nfs_true;
2532 			free((caddr_t)rbuf, M_TEMP);
2533 			free((caddr_t)cookies, M_TEMP);
2534 			return (0);
2535 		}
2536 	}
2537 
2538 	/*
2539 	 * Check for degenerate cases of nothing useful read.
2540 	 * If so go try again
2541 	 */
2542 	cpos = rbuf;
2543 	cend = rbuf + siz;
2544 	dp = (struct dirent *)cpos;
2545 	cookiep = cookies;
2546 
2547 	while (cpos < cend && ncookies > 0 &&
2548 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2549 		cpos += dp->d_reclen;
2550 		dp = (struct dirent *)cpos;
2551 		cookiep++;
2552 		ncookies--;
2553 	}
2554 	if (cpos >= cend || ncookies == 0) {
2555 		toff = off;
2556 		siz = fullsiz;
2557 		goto again;
2558 	}
2559 
2560 	len = 3 * NFSX_UNSIGNED;	/* paranoia, probably can be 0 */
2561 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2562 	if (v3) {
2563 		nfsm_srvpostop_attr(getret, &at);
2564 		nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2565 		txdr_hyper(at.va_filerev, tl);
2566 	}
2567 	mp = mp2 = mb;
2568 	bp = bpos;
2569 	be = bp + M_TRAILINGSPACE(mp);
2570 
2571 	/* Loop through the records and build reply */
2572 	while (cpos < cend && ncookies > 0) {
2573 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2574 			nlen = dp->d_namlen;
2575 			rem = nfsm_rndup(nlen)-nlen;
2576 			len += (4 * NFSX_UNSIGNED + nlen + rem);
2577 			if (v3)
2578 				len += 2 * NFSX_UNSIGNED;
2579 			if (len > cnt) {
2580 				eofflag = 0;
2581 				break;
2582 			}
2583 			/*
2584 			 * Build the directory record xdr from
2585 			 * the dirent entry.
2586 			 */
2587 			nfsm_clget;
2588 			*tl = nfs_true;
2589 			bp += NFSX_UNSIGNED;
2590 			if (v3) {
2591 				nfsm_clget;
2592 				*tl = 0;
2593 				bp += NFSX_UNSIGNED;
2594 			}
2595 			nfsm_clget;
2596 			*tl = txdr_unsigned(dp->d_fileno);
2597 			bp += NFSX_UNSIGNED;
2598 			nfsm_clget;
2599 			*tl = txdr_unsigned(nlen);
2600 			bp += NFSX_UNSIGNED;
2601 
2602 			/* And loop around copying the name */
2603 			xfer = nlen;
2604 			cp = dp->d_name;
2605 			while (xfer > 0) {
2606 				nfsm_clget;
2607 				if ((bp+xfer) > be)
2608 					tsiz = be-bp;
2609 				else
2610 					tsiz = xfer;
2611 				memcpy(bp, cp, tsiz);
2612 				bp += tsiz;
2613 				xfer -= tsiz;
2614 				if (xfer > 0)
2615 					cp += tsiz;
2616 			}
2617 			/* And null pad to an int32_t boundary */
2618 			for (i = 0; i < rem; i++)
2619 				*bp++ = '\0';
2620 			nfsm_clget;
2621 
2622 			/* Finish off the record */
2623 			txdr_hyper(*cookiep, &jar);
2624 			if (v3) {
2625 				*tl = jar.nfsuquad[0];
2626 				bp += NFSX_UNSIGNED;
2627 				nfsm_clget;
2628 			}
2629 			*tl = jar.nfsuquad[1];
2630 			bp += NFSX_UNSIGNED;
2631 		}
2632 		cpos += dp->d_reclen;
2633 		dp = (struct dirent *)cpos;
2634 		cookiep++;
2635 		ncookies--;
2636 	}
2637 	vrele(vp);
2638 	nfsm_clget;
2639 	*tl = nfs_false;
2640 	bp += NFSX_UNSIGNED;
2641 	nfsm_clget;
2642 	if (eofflag)
2643 		*tl = nfs_true;
2644 	else
2645 		*tl = nfs_false;
2646 	bp += NFSX_UNSIGNED;
2647 	if (mp != mb) {
2648 		if (bp < be)
2649 			mp->m_len = bp - mtod(mp, caddr_t);
2650 	} else
2651 		mp->m_len += bp - bpos;
2652 	free((caddr_t)rbuf, M_TEMP);
2653 	free((caddr_t)cookies, M_TEMP);
2654 	nfsm_srvdone;
2655 }
2656 
2657 int
2658 nfsrv_readdirplus(nfsd, slp, procp, mrq)
2659 	struct nfsrv_descript *nfsd;
2660 	struct nfssvc_sock *slp;
2661 	struct proc *procp;
2662 	struct mbuf **mrq;
2663 {
2664 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2665 	struct mbuf *nam = nfsd->nd_nam;
2666 	caddr_t dpos = nfsd->nd_dpos;
2667 	struct ucred *cred = &nfsd->nd_cr;
2668 	char *bp, *be;
2669 	struct mbuf *mp;
2670 	struct dirent *dp;
2671 	caddr_t cp;
2672 	u_int32_t *tl;
2673 	int32_t t1;
2674 	caddr_t bpos;
2675 	struct mbuf *mb, *mb2, *mreq, *mp2;
2676 	char *cpos, *cend, *cp2, *rbuf;
2677 	struct vnode *vp, *nvp;
2678 	struct flrep fl;
2679 	nfsfh_t nfh;
2680 	fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
2681 	struct uio io;
2682 	struct iovec iv;
2683 	struct vattr va, at, *vap = &va;
2684 	struct nfs_fattr *fp;
2685 	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2686 	int siz, cnt, fullsiz, eofflag, rdonly, cache, dirlen, ncookies;
2687 	u_quad_t frev, off, toff, verf;
2688 	off_t *cookies = NULL, *cookiep;
2689 
2690 	fhp = &nfh.fh_generic;
2691 	nfsm_srvmtofh(fhp);
2692 	nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2693 	toff = fxdr_hyper(tl);
2694 	tl += 2;
2695 	verf = fxdr_hyper(tl);
2696 	tl += 2;
2697 	siz = fxdr_unsigned(int, *tl++);
2698 	cnt = fxdr_unsigned(int, *tl);
2699 	off = toff;
2700 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2701 	xfer = NFS_SRVMAXDATA(nfsd);
2702 	if (siz > xfer)
2703 		siz = xfer;
2704 	fullsiz = siz;
2705 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
2706 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2707 	if (!error && vp->v_type != VDIR) {
2708 		error = ENOTDIR;
2709 		vput(vp);
2710 	}
2711 	if (error) {
2712 		nfsm_reply(NFSX_UNSIGNED);
2713 		nfsm_srvpostop_attr(getret, &at);
2714 		return (0);
2715 	}
2716 	error = getret = VOP_GETATTR(vp, &at, cred, procp);
2717 #ifdef NFS3_STRICTVERF
2718 	/*
2719 	 * XXX This check is too strict for Solaris 2.5 clients.
2720 	 */
2721 	if (!error && toff && verf != at.va_filerev)
2722 		error = NFSERR_BAD_COOKIE;
2723 #endif
2724 	if (!error) {
2725 		nqsrv_getl(vp, ND_READ);
2726 		error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0);
2727 	}
2728 	if (error) {
2729 		vput(vp);
2730 		nfsm_reply(NFSX_V3POSTOPATTR);
2731 		nfsm_srvpostop_attr(getret, &at);
2732 		return (0);
2733 	}
2734 	VOP_UNLOCK(vp, 0);
2735 
2736 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2737 again:
2738 	iv.iov_base = rbuf;
2739 	iv.iov_len = fullsiz;
2740 	io.uio_iov = &iv;
2741 	io.uio_iovcnt = 1;
2742 	io.uio_offset = (off_t)off;
2743 	io.uio_resid = fullsiz;
2744 	io.uio_segflg = UIO_SYSSPACE;
2745 	io.uio_rw = UIO_READ;
2746 	io.uio_procp = (struct proc *)0;
2747 	eofflag = 0;
2748 
2749 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2750 
2751 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2752 
2753 	off = (u_quad_t)io.uio_offset;
2754 	getret = VOP_GETATTR(vp, &at, cred, procp);
2755 
2756 	VOP_UNLOCK(vp, 0);
2757 
2758 	/*
2759 	 * If the VGET operation doesn't work for this filesystem,
2760 	 * we can't support readdirplus. Returning NOTSUPP should
2761 	 * make clients fall back to plain readdir.
2762 	 * There's no need to check for VPTOFH as well, we wouldn't
2763 	 * even be here otherwise.
2764 	 */
2765 	if (!getret) {
2766 		if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp)))
2767 			getret = (getret == EOPNOTSUPP) ?
2768 				NFSERR_NOTSUPP : NFSERR_IO;
2769 		else
2770 			vput(nvp);
2771 	}
2772 
2773 	if (!cookies && !error)
2774 		error = NFSERR_PERM;
2775 	if (!error)
2776 		error = getret;
2777 	if (error) {
2778 		vrele(vp);
2779 		if (cookies)
2780 			free((caddr_t)cookies, M_TEMP);
2781 		free((caddr_t)rbuf, M_TEMP);
2782 		nfsm_reply(NFSX_V3POSTOPATTR);
2783 		nfsm_srvpostop_attr(getret, &at);
2784 		return (0);
2785 	}
2786 	if (io.uio_resid) {
2787 		siz -= io.uio_resid;
2788 
2789 		/*
2790 		 * If nothing read, return eof
2791 		 * rpc reply
2792 		 */
2793 		if (siz == 0) {
2794 			vrele(vp);
2795 			nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
2796 				2 * NFSX_UNSIGNED);
2797 			nfsm_srvpostop_attr(getret, &at);
2798 			nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2799 			txdr_hyper(at.va_filerev, tl);
2800 			tl += 2;
2801 			*tl++ = nfs_false;
2802 			*tl = nfs_true;
2803 			free((caddr_t)cookies, M_TEMP);
2804 			free((caddr_t)rbuf, M_TEMP);
2805 			return (0);
2806 		}
2807 	}
2808 
2809 	/*
2810 	 * Check for degenerate cases of nothing useful read.
2811 	 * If so go try again
2812 	 */
2813 	cpos = rbuf;
2814 	cend = rbuf + siz;
2815 	dp = (struct dirent *)cpos;
2816 	cookiep = cookies;
2817 
2818 	while (cpos < cend && ncookies > 0 &&
2819 		(dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2820 		cpos += dp->d_reclen;
2821 		dp = (struct dirent *)cpos;
2822 		cookiep++;
2823 		ncookies--;
2824 	}
2825 	if (cpos >= cend || ncookies == 0) {
2826 		toff = off;
2827 		siz = fullsiz;
2828 		goto again;
2829 	}
2830 
2831 	dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
2832 	nfsm_reply(cnt);
2833 	nfsm_srvpostop_attr(getret, &at);
2834 	nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2835 	txdr_hyper(at.va_filerev, tl);
2836 	mp = mp2 = mb;
2837 	bp = bpos;
2838 	be = bp + M_TRAILINGSPACE(mp);
2839 
2840 	/* Loop through the records and build reply */
2841 	while (cpos < cend && ncookies > 0) {
2842 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2843 			nlen = dp->d_namlen;
2844 			rem = nfsm_rndup(nlen)-nlen;
2845 
2846 			/*
2847 			 * For readdir_and_lookup get the vnode using
2848 			 * the file number.
2849 			 */
2850 			if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp))
2851 				goto invalid;
2852 			memset((caddr_t)nfhp, 0, NFSX_V3FH);
2853 			nfhp->fh_fsid =
2854 				nvp->v_mount->mnt_stat.f_fsid;
2855 			if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) {
2856 				vput(nvp);
2857 				goto invalid;
2858 			}
2859 			if (VOP_GETATTR(nvp, vap, cred, procp)) {
2860 				vput(nvp);
2861 				goto invalid;
2862 			}
2863 			vput(nvp);
2864 
2865 			/*
2866 			 * If either the dircount or maxcount will be
2867 			 * exceeded, get out now. Both of these lengths
2868 			 * are calculated conservatively, including all
2869 			 * XDR overheads.
2870 			 */
2871 			len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
2872 				NFSX_V3POSTOPATTR);
2873 			dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
2874 			if (len > cnt || dirlen > fullsiz) {
2875 				eofflag = 0;
2876 				break;
2877 			}
2878 
2879 			/*
2880 			 * Build the directory record xdr from
2881 			 * the dirent entry.
2882 			 */
2883 			fp = (struct nfs_fattr *)&fl.fl_fattr;
2884 			nfsm_srvfillattr(vap, fp);
2885 			fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
2886 			fl.fl_fhok = nfs_true;
2887 			fl.fl_postopok = nfs_true;
2888 			txdr_hyper(*cookiep, fl.fl_off.nfsuquad);
2889 
2890 			nfsm_clget;
2891 			*tl = nfs_true;
2892 			bp += NFSX_UNSIGNED;
2893 			nfsm_clget;
2894 			*tl = 0;
2895 			bp += NFSX_UNSIGNED;
2896 			nfsm_clget;
2897 			*tl = txdr_unsigned(dp->d_fileno);
2898 			bp += NFSX_UNSIGNED;
2899 			nfsm_clget;
2900 			*tl = txdr_unsigned(nlen);
2901 			bp += NFSX_UNSIGNED;
2902 
2903 			/* And loop around copying the name */
2904 			xfer = nlen;
2905 			cp = dp->d_name;
2906 			while (xfer > 0) {
2907 				nfsm_clget;
2908 				if ((bp + xfer) > be)
2909 					tsiz = be - bp;
2910 				else
2911 					tsiz = xfer;
2912 				memcpy(bp, cp, tsiz);
2913 				bp += tsiz;
2914 				xfer -= tsiz;
2915 				if (xfer > 0)
2916 					cp += tsiz;
2917 			}
2918 			/* And null pad to an int32_t boundary */
2919 			for (i = 0; i < rem; i++)
2920 				*bp++ = '\0';
2921 
2922 			/*
2923 			 * Now copy the flrep structure out.
2924 			 */
2925 			xfer = sizeof (struct flrep);
2926 			cp = (caddr_t)&fl;
2927 			while (xfer > 0) {
2928 				nfsm_clget;
2929 				if ((bp + xfer) > be)
2930 					tsiz = be - bp;
2931 				else
2932 					tsiz = xfer;
2933 				memcpy(bp, cp, tsiz);
2934 				bp += tsiz;
2935 				xfer -= tsiz;
2936 				if (xfer > 0)
2937 					cp += tsiz;
2938 			}
2939 		}
2940 invalid:
2941 		cpos += dp->d_reclen;
2942 		dp = (struct dirent *)cpos;
2943 		cookiep++;
2944 		ncookies--;
2945 	}
2946 	vrele(vp);
2947 	nfsm_clget;
2948 	*tl = nfs_false;
2949 	bp += NFSX_UNSIGNED;
2950 	nfsm_clget;
2951 	if (eofflag)
2952 		*tl = nfs_true;
2953 	else
2954 		*tl = nfs_false;
2955 	bp += NFSX_UNSIGNED;
2956 	if (mp != mb) {
2957 		if (bp < be)
2958 			mp->m_len = bp - mtod(mp, caddr_t);
2959 	} else
2960 		mp->m_len += bp - bpos;
2961 	free((caddr_t)cookies, M_TEMP);
2962 	free((caddr_t)rbuf, M_TEMP);
2963 	nfsm_srvdone;
2964 }
2965 
2966 /*
2967  * nfs commit service
2968  */
2969 int
2970 nfsrv_commit(nfsd, slp, procp, mrq)
2971 	struct nfsrv_descript *nfsd;
2972 	struct nfssvc_sock *slp;
2973 	struct proc *procp;
2974 	struct mbuf **mrq;
2975 {
2976 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2977 	struct mbuf *nam = nfsd->nd_nam;
2978 	caddr_t dpos = nfsd->nd_dpos;
2979 	struct ucred *cred = &nfsd->nd_cr;
2980 	struct vattr bfor, aft;
2981 	struct vnode *vp;
2982 	nfsfh_t nfh;
2983 	fhandle_t *fhp;
2984 	u_int32_t *tl;
2985 	int32_t t1;
2986 	caddr_t bpos;
2987 	int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt, cache;
2988 	char *cp2;
2989 	struct mbuf *mb, *mb2, *mreq;
2990 	u_quad_t frev, off;
2991 
2992 #ifndef nolint
2993 	cache = 0;
2994 #endif
2995 	fhp = &nfh.fh_generic;
2996 	nfsm_srvmtofh(fhp);
2997 	nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2998 
2999 	off = fxdr_hyper(tl);
3000 	tl += 2;
3001 	cnt = fxdr_unsigned(int, *tl);
3002 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3003 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3004 	if (error) {
3005 		nfsm_reply(2 * NFSX_UNSIGNED);
3006 		nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3007 		return (0);
3008 	}
3009 	for_ret = VOP_GETATTR(vp, &bfor, cred, procp);
3010 	error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, off + cnt, procp);
3011 	aft_ret = VOP_GETATTR(vp, &aft, cred, procp);
3012 	vput(vp);
3013 	nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3014 	nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3015 	if (!error) {
3016 		nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3017 		*tl++ = txdr_unsigned(boottime.tv_sec);
3018 		*tl = txdr_unsigned(boottime.tv_usec);
3019 	} else
3020 		return (0);
3021 	nfsm_srvdone;
3022 }
3023 
3024 /*
3025  * nfs statfs service
3026  */
3027 int
3028 nfsrv_statfs(nfsd, slp, procp, mrq)
3029 	struct nfsrv_descript *nfsd;
3030 	struct nfssvc_sock *slp;
3031 	struct proc *procp;
3032 	struct mbuf **mrq;
3033 {
3034 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3035 	struct mbuf *nam = nfsd->nd_nam;
3036 	caddr_t dpos = nfsd->nd_dpos;
3037 	struct ucred *cred = &nfsd->nd_cr;
3038 	struct statfs *sf;
3039 	struct nfs_statfs *sfp;
3040 	u_int32_t *tl;
3041 	int32_t t1;
3042 	caddr_t bpos;
3043 	int error = 0, rdonly, cache, getret = 1;
3044 	int v3 = (nfsd->nd_flag & ND_NFSV3);
3045 	char *cp2;
3046 	struct mbuf *mb, *mb2, *mreq;
3047 	struct vnode *vp;
3048 	struct vattr at;
3049 	nfsfh_t nfh;
3050 	fhandle_t *fhp;
3051 	struct statfs statfs;
3052 	u_quad_t frev, tval;
3053 
3054 #ifndef nolint
3055 	cache = 0;
3056 #endif
3057 	fhp = &nfh.fh_generic;
3058 	nfsm_srvmtofh(fhp);
3059 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3060 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3061 	if (error) {
3062 		nfsm_reply(NFSX_UNSIGNED);
3063 		nfsm_srvpostop_attr(getret, &at);
3064 		return (0);
3065 	}
3066 	sf = &statfs;
3067 	error = VFS_STATFS(vp->v_mount, sf, procp);
3068 	getret = VOP_GETATTR(vp, &at, cred, procp);
3069 	vput(vp);
3070 	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3071 	if (v3)
3072 		nfsm_srvpostop_attr(getret, &at);
3073 	if (error)
3074 		return (0);
3075 	nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3076 	if (v3) {
3077 		tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_bsize);
3078 		txdr_hyper(tval, &sfp->sf_tbytes);
3079 		tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_bsize);
3080 		txdr_hyper(tval, &sfp->sf_fbytes);
3081 		tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_bsize);
3082 		txdr_hyper(tval, &sfp->sf_abytes);
3083 		tval = (u_quad_t)sf->f_files;
3084 		txdr_hyper(tval, &sfp->sf_tfiles);
3085 		tval = (u_quad_t)sf->f_ffree;
3086 		txdr_hyper(tval, &sfp->sf_ffiles);
3087 		txdr_hyper(tval, &sfp->sf_afiles);
3088 		sfp->sf_invarsec = 0;
3089 	} else {
3090 		sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3091 		sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
3092 		sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3093 		sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3094 		sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3095 	}
3096 	nfsm_srvdone;
3097 }
3098 
3099 /*
3100  * nfs fsinfo service
3101  */
3102 int
3103 nfsrv_fsinfo(nfsd, slp, procp, mrq)
3104 	struct nfsrv_descript *nfsd;
3105 	struct nfssvc_sock *slp;
3106 	struct proc *procp;
3107 	struct mbuf **mrq;
3108 {
3109 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3110 	struct mbuf *nam = nfsd->nd_nam;
3111 	caddr_t dpos = nfsd->nd_dpos;
3112 	struct ucred *cred = &nfsd->nd_cr;
3113 	u_int32_t *tl;
3114 	struct nfsv3_fsinfo *sip;
3115 	int32_t t1;
3116 	caddr_t bpos;
3117 	int error = 0, rdonly, cache, getret = 1, pref;
3118 	char *cp2;
3119 	struct mbuf *mb, *mb2, *mreq;
3120 	struct vnode *vp;
3121 	struct vattr at;
3122 	nfsfh_t nfh;
3123 	fhandle_t *fhp;
3124 	u_quad_t frev, maxfsize;
3125 	struct statfs sb;
3126 
3127 #ifndef nolint
3128 	cache = 0;
3129 #endif
3130 	fhp = &nfh.fh_generic;
3131 	nfsm_srvmtofh(fhp);
3132 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3133 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3134 	if (error) {
3135 		nfsm_reply(NFSX_UNSIGNED);
3136 		nfsm_srvpostop_attr(getret, &at);
3137 		return (0);
3138 	}
3139 
3140 	/* XXX Try to make a guess on the max file size. */
3141 	VFS_STATFS(vp->v_mount, &sb, (struct proc *)0);
3142 	maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
3143 
3144 	getret = VOP_GETATTR(vp, &at, cred, procp);
3145 	vput(vp);
3146 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3147 	nfsm_srvpostop_attr(getret, &at);
3148 	nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3149 
3150 	/*
3151 	 * XXX
3152 	 * There should be file system VFS OP(s) to get this information.
3153 	 * For now, assume ufs.
3154 	 */
3155 	if (slp->ns_so->so_type == SOCK_DGRAM)
3156 		pref = NFS_MAXDGRAMDATA;
3157 	else
3158 		pref = NFS_MAXDATA;
3159 	sip->fs_rtmax = txdr_unsigned(NFS_MAXDATA);
3160 	sip->fs_rtpref = txdr_unsigned(pref);
3161 	sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3162 	sip->fs_wtmax = txdr_unsigned(NFS_MAXDATA);
3163 	sip->fs_wtpref = txdr_unsigned(pref);
3164 	sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3165 	sip->fs_dtpref = txdr_unsigned(pref);
3166 	txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3167 	sip->fs_timedelta.nfsv3_sec = 0;
3168 	sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3169 	sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3170 		NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3171 		NFSV3FSINFO_CANSETTIME);
3172 	nfsm_srvdone;
3173 }
3174 
3175 /*
3176  * nfs pathconf service
3177  */
3178 int
3179 nfsrv_pathconf(nfsd, slp, procp, mrq)
3180 	struct nfsrv_descript *nfsd;
3181 	struct nfssvc_sock *slp;
3182 	struct proc *procp;
3183 	struct mbuf **mrq;
3184 {
3185 	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3186 	struct mbuf *nam = nfsd->nd_nam;
3187 	caddr_t dpos = nfsd->nd_dpos;
3188 	struct ucred *cred = &nfsd->nd_cr;
3189 	u_int32_t *tl;
3190 	struct nfsv3_pathconf *pc;
3191 	int32_t t1;
3192 	caddr_t bpos;
3193 	int error = 0, rdonly, cache, getret = 1;
3194 	register_t linkmax, namemax, chownres, notrunc;
3195 	char *cp2;
3196 	struct mbuf *mb, *mb2, *mreq;
3197 	struct vnode *vp;
3198 	struct vattr at;
3199 	nfsfh_t nfh;
3200 	fhandle_t *fhp;
3201 	u_quad_t frev;
3202 
3203 #ifndef nolint
3204 	cache = 0;
3205 #endif
3206 	fhp = &nfh.fh_generic;
3207 	nfsm_srvmtofh(fhp);
3208 	error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3209 		 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3210 	if (error) {
3211 		nfsm_reply(NFSX_UNSIGNED);
3212 		nfsm_srvpostop_attr(getret, &at);
3213 		return (0);
3214 	}
3215 	error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3216 	if (!error)
3217 		error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3218 	if (!error)
3219 		error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3220 	if (!error)
3221 		error = VOP_PATHCONF(vp, _PC_NO_TRUNC, &notrunc);
3222 	getret = VOP_GETATTR(vp, &at, cred, procp);
3223 	vput(vp);
3224 	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3225 	nfsm_srvpostop_attr(getret, &at);
3226 	if (error)
3227 		return (0);
3228 	nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3229 
3230 	pc->pc_linkmax = txdr_unsigned(linkmax);
3231 	pc->pc_namemax = txdr_unsigned(namemax);
3232 	pc->pc_notrunc = txdr_unsigned(notrunc);
3233 	pc->pc_chownrestricted = txdr_unsigned(chownres);
3234 
3235 	/*
3236 	 * These should probably be supported by VOP_PATHCONF(), but
3237 	 * until msdosfs is exportable (why would you want to?), the
3238 	 * Unix defaults should be ok.
3239 	 */
3240 	pc->pc_caseinsensitive = nfs_false;
3241 	pc->pc_casepreserving = nfs_true;
3242 	nfsm_srvdone;
3243 }
3244 
3245 /*
3246  * Null operation, used by clients to ping server
3247  */
3248 /* ARGSUSED */
3249 int
3250 nfsrv_null(nfsd, slp, procp, mrq)
3251 	struct nfsrv_descript *nfsd;
3252 	struct nfssvc_sock *slp;
3253 	struct proc *procp;
3254 	struct mbuf **mrq;
3255 {
3256 	struct mbuf *mrep = nfsd->nd_mrep;
3257 	caddr_t bpos;
3258 	int error = NFSERR_RETVOID, cache = 0;
3259 	struct mbuf *mb, *mreq;
3260 	u_quad_t frev;
3261 
3262 	nfsm_reply(0);
3263 	return (0);
3264 }
3265 
3266 /*
3267  * No operation, used for obsolete procedures
3268  */
3269 /* ARGSUSED */
3270 int
3271 nfsrv_noop(nfsd, slp, procp, mrq)
3272 	struct nfsrv_descript *nfsd;
3273 	struct nfssvc_sock *slp;
3274 	struct proc *procp;
3275 	struct mbuf **mrq;
3276 {
3277 	struct mbuf *mrep = nfsd->nd_mrep;
3278 	caddr_t bpos;
3279 	int error, cache = 0;
3280 	struct mbuf *mb, *mreq;
3281 	u_quad_t frev;
3282 
3283 	if (nfsd->nd_repstat)
3284 		error = nfsd->nd_repstat;
3285 	else
3286 		error = EPROCUNAVAIL;
3287 	nfsm_reply(0);
3288 	return (0);
3289 }
3290 
3291 /*
3292  * Perform access checking for vnodes obtained from file handles that would
3293  * refer to files already opened by a Unix client. You cannot just use
3294  * vn_writechk() and VOP_ACCESS() for two reasons.
3295  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3296  * 2 - The owner is to be given access irrespective of mode bits for some
3297  *     operations, so that processes that chmod after opening a file don't
3298  *     break. I don't like this because it opens a security hole, but since
3299  *     the nfs server opens a security hole the size of a barn door anyhow,
3300  *     what the heck.
3301  *
3302  * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3303  * will return EPERM instead of EACCESS. EPERM is always an error.
3304  */
3305 int
3306 nfsrv_access(vp, flags, cred, rdonly, p, override)
3307 	struct vnode *vp;
3308 	int flags;
3309 	struct ucred *cred;
3310 	int rdonly;
3311 	struct proc *p;
3312 {
3313 	struct vattr vattr;
3314 	int error;
3315 	if (flags & VWRITE) {
3316 		/* Just vn_writechk() changed to check rdonly */
3317 		/*
3318 		 * Disallow write attempts on read-only file systems;
3319 		 * unless the file is a socket or a block or character
3320 		 * device resident on the file system.
3321 		 */
3322 		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3323 			switch (vp->v_type) {
3324 			case VREG:
3325 			case VDIR:
3326 			case VLNK:
3327 				return (EROFS);
3328 			default:
3329 				break;
3330 			}
3331 		}
3332 
3333 		/*
3334 		 * If the vnode is in use as a process's text,
3335 		 * we can't allow writing.
3336 		 */
3337 		if (vp->v_flag & VTEXT)
3338 			return (ETXTBSY);
3339 	}
3340 	error = VOP_GETATTR(vp, &vattr, cred, p);
3341 	if (error)
3342 		return (error);
3343 	error = VOP_ACCESS(vp, flags, cred, p);
3344 	/*
3345 	 * Allow certain operations for the owner (reads and writes
3346 	 * on files that are already open).
3347 	 */
3348 	if (override && error == EACCES && cred->cr_uid == vattr.va_uid)
3349 		error = 0;
3350 	return error;
3351 }
3352