nfs_serv.c revision 268961
1/*-
2 * Copyright (c) 1989, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 *    may be used to endorse or promote products derived from this software
18 *    without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 *	@(#)nfs_serv.c  8.8 (Berkeley) 7/31/95
33 */
34
35#include <sys/cdefs.h>
36__FBSDID("$FreeBSD: stable/10/sys/nfsserver/nfs_serv.c 268961 2014-07-21 22:21:09Z bdrewery $");
37
38/*
39 * nfs version 2 and 3 server calls to vnode ops
40 * - these routines generally have 3 phases
41 *   1 - break down and validate rpc request in mbuf list
42 *   2 - do the vnode ops for the request
43 *       (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
44 *   3 - build the rpc reply in an mbuf list
45 *   nb:
46 *	- do not mix the phases, since the nfsm_?? macros can return failures
47 *	  on a bad rpc or similar and do not do any vrele() or vput()'s
48 *
49 *      - the nfsm_reply() macro generates an nfs rpc reply with the nfs
50 *	error number iff error != 0 whereas
51 *	returning an error from the server function implies a fatal error
52 *	such as a badly constructed rpc request that should be dropped without
53 *	a reply.
54 *	For nfsm_reply(), the case where error == EBADRPC is treated
55 *	specially; after constructing a reply, it does an immediate
56 *	`goto nfsmout' to avoid getting any V3 post-op status appended.
57 *
58 * Other notes:
59 *	Warning: always pay careful attention to resource cleanup on return
60 *	and note that nfsm_*() macros can terminate a procedure on certain
61 *	errors.
62 *
63 *	lookup() and namei()
64 *	may return garbage in various structural fields/return elements
65 *	if an error is returned, and may garbage up nd.ni_dvp even if no
66 *	error is returned and you did not request LOCKPARENT or WANTPARENT.
67 *
68 *	We use the ni_cnd.cn_flags 'HASBUF' flag to track whether the name
69 *	buffer has been freed or not.
70 */
71
72#include <sys/param.h>
73#include <sys/systm.h>
74#include <sys/proc.h>
75#include <sys/namei.h>
76#include <sys/unistd.h>
77#include <sys/vnode.h>
78#include <sys/mount.h>
79#include <sys/socket.h>
80#include <sys/socketvar.h>
81#include <sys/malloc.h>
82#include <sys/mbuf.h>
83#include <sys/priv.h>
84#include <sys/dirent.h>
85#include <sys/stat.h>
86#include <sys/kernel.h>
87#include <sys/sysctl.h>
88#include <sys/bio.h>
89#include <sys/buf.h>
90#include <sys/rwlock.h>
91
92#include <vm/vm.h>
93#include <vm/vm_extern.h>
94#include <vm/vm_object.h>
95
96#include <nfs/nfsproto.h>
97#include <nfsserver/nfs.h>
98#include <nfs/xdr_subs.h>
99#include <nfsserver/nfsm_subs.h>
100
101FEATURE(nfsserver, "NFS server");
102
103#ifdef NFSRV_DEBUG
104#define nfsdbprintf(info)	printf info
105#else
106#define nfsdbprintf(info)
107#endif
108
109#define MAX_COMMIT_COUNT	(1024 * 1024)
110
111#define	MAX_REORDERED_RPC	16
112#define NUM_HEURISTIC		1031
113#define NHUSE_INIT		64
114#define NHUSE_INC		16
115#define NHUSE_MAX		2048
116
117static struct nfsheur {
118	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
119	off_t nh_nextoff;	/* next offset for sequential detection */
120	int nh_use;		/* use count for selection */
121	int nh_seqcount;	/* heuristic */
122} nfsheur[NUM_HEURISTIC];
123
124/* Global vars */
125
126int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
127int nfsrvw_procrastinate_v3 = 0;
128
129static struct timeval	nfsver = { 0 };
130
131SYSCTL_NODE(_vfs, OID_AUTO, nfsrv, CTLFLAG_RW, 0, "NFS server");
132
133static int nfs_async;
134static int nfs_commit_blks;
135static int nfs_commit_miss;
136SYSCTL_INT(_vfs_nfsrv, OID_AUTO, async, CTLFLAG_RW, &nfs_async, 0,
137    "Tell client that writes were synced even though they were not");
138SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0,
139    "Number of completed commits");
140SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, "");
141
142struct nfsrvstats nfsrvstats;
143SYSCTL_STRUCT(_vfs_nfsrv, NFS_NFSRVSTATS, nfsrvstats, CTLFLAG_RW,
144	&nfsrvstats, nfsrvstats, "S,nfsrvstats");
145
146static int	nfsrv_access(struct vnode *, accmode_t, struct ucred *,
147		    int, int);
148
149/*
150 * Clear nameidata fields that are tested in nsfmout cleanup code prior
151 * to using first nfsm macro (that might jump to the cleanup code).
152 */
153
154static __inline void
155ndclear(struct nameidata *nd)
156{
157
158	nd->ni_cnd.cn_flags = 0;
159	nd->ni_vp = NULL;
160	nd->ni_dvp = NULL;
161	nd->ni_startdir = NULL;
162	nd->ni_strictrelative = 0;
163}
164
165/*
166 * Heuristic to detect sequential operation.
167 */
168static struct nfsheur *
169nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
170{
171	struct nfsheur *nh;
172	int hi, try;
173
174	/* Locate best candidate. */
175	try = 32;
176	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
177	nh = &nfsheur[hi];
178	while (try--) {
179		if (nfsheur[hi].nh_vp == vp) {
180			nh = &nfsheur[hi];
181			break;
182		}
183		if (nfsheur[hi].nh_use > 0)
184			--nfsheur[hi].nh_use;
185		hi = (hi + 1) % NUM_HEURISTIC;
186		if (nfsheur[hi].nh_use < nh->nh_use)
187			nh = &nfsheur[hi];
188	}
189
190	/* Initialize hint if this is a new file. */
191	if (nh->nh_vp != vp) {
192		nh->nh_vp = vp;
193		nh->nh_nextoff = uio->uio_offset;
194		nh->nh_use = NHUSE_INIT;
195		if (uio->uio_offset == 0)
196			nh->nh_seqcount = 4;
197		else
198			nh->nh_seqcount = 1;
199	}
200
201	/* Calculate heuristic. */
202	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
203	    uio->uio_offset == nh->nh_nextoff) {
204		/* See comments in vfs_vnops.c:sequential_heuristic(). */
205		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
206		if (nh->nh_seqcount > IO_SEQMAX)
207			nh->nh_seqcount = IO_SEQMAX;
208	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
209	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
210		/* Probably a reordered RPC, leave seqcount alone. */
211	} else if (nh->nh_seqcount > 1) {
212		nh->nh_seqcount /= 2;
213	} else {
214		nh->nh_seqcount = 0;
215	}
216	nh->nh_use += NHUSE_INC;
217	if (nh->nh_use > NHUSE_MAX)
218		nh->nh_use = NHUSE_MAX;
219	return (nh);
220}
221
222/*
223 * nfs v3 access service
224 */
225int
226nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
227    struct mbuf **mrq)
228{
229	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
230	struct sockaddr *nam = nfsd->nd_nam;
231	caddr_t dpos = nfsd->nd_dpos;
232	struct ucred *cred = nfsd->nd_cr;
233	struct vnode *vp = NULL;
234	nfsfh_t nfh;
235	fhandle_t *fhp;
236	u_int32_t *tl;
237	caddr_t bpos;
238	int error = 0, rdonly, getret;
239	struct mbuf *mb, *mreq;
240	struct vattr vattr, *vap = &vattr;
241	u_long testmode, nfsmode;
242	int v3 = (nfsd->nd_flag & ND_NFSV3);
243
244	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
245	if (!v3)
246		panic("nfsrv3_access: v3 proc called on a v2 connection");
247	fhp = &nfh.fh_generic;
248	nfsm_srvmtofh(fhp);
249	tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
250	error = nfsrv_fhtovp(fhp, 0, &vp, nfsd, slp, nam, &rdonly);
251	if (error) {
252		nfsm_reply(NFSX_UNSIGNED);
253		nfsm_srvpostop_attr(1, NULL);
254		error = 0;
255		goto nfsmout;
256	}
257	nfsmode = fxdr_unsigned(u_int32_t, *tl);
258	if ((nfsmode & NFSV3ACCESS_READ) &&
259		nfsrv_access(vp, VREAD, cred, rdonly, 0))
260		nfsmode &= ~NFSV3ACCESS_READ;
261	if (vp->v_type == VDIR)
262		testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
263			NFSV3ACCESS_DELETE);
264	else
265		testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
266	if ((nfsmode & testmode) &&
267		nfsrv_access(vp, VWRITE, cred, rdonly, 0))
268		nfsmode &= ~testmode;
269	if (vp->v_type == VDIR)
270		testmode = NFSV3ACCESS_LOOKUP;
271	else
272		testmode = NFSV3ACCESS_EXECUTE;
273	if ((nfsmode & testmode) &&
274		nfsrv_access(vp, VEXEC, cred, rdonly, 0))
275		nfsmode &= ~testmode;
276	getret = VOP_GETATTR(vp, vap, cred);
277	vput(vp);
278	vp = NULL;
279	nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
280	nfsm_srvpostop_attr(getret, vap);
281	tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
282	*tl = txdr_unsigned(nfsmode);
283nfsmout:
284	if (vp)
285		vput(vp);
286	return(error);
287}
288
289/*
290 * nfs getattr service
291 */
292int
293nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
294    struct mbuf **mrq)
295{
296	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
297	struct sockaddr *nam = nfsd->nd_nam;
298	caddr_t dpos = nfsd->nd_dpos;
299	struct ucred *cred = nfsd->nd_cr;
300	struct nfs_fattr *fp;
301	struct vattr va;
302	struct vattr *vap = &va;
303	struct vnode *vp = NULL;
304	nfsfh_t nfh;
305	fhandle_t *fhp;
306	caddr_t bpos;
307	int error = 0, rdonly;
308	struct mbuf *mb, *mreq;
309
310	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
311	fhp = &nfh.fh_generic;
312	nfsm_srvmtofh(fhp);
313	error = nfsrv_fhtovp(fhp, 0, &vp, nfsd, slp, nam, &rdonly);
314	if (error) {
315		nfsm_reply(0);
316		error = 0;
317		goto nfsmout;
318	}
319	error = VOP_GETATTR(vp, vap, cred);
320	vput(vp);
321	vp = NULL;
322	nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
323	if (error) {
324		error = 0;
325		goto nfsmout;
326	}
327	fp = nfsm_build(struct nfs_fattr *,
328	    NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
329	nfsm_srvfillattr(vap, fp);
330	/* fall through */
331
332nfsmout:
333	if (vp)
334		vput(vp);
335	return(error);
336}
337
338/*
339 * nfs setattr service
340 */
341int
342nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
343    struct mbuf **mrq)
344{
345	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
346	struct sockaddr *nam = nfsd->nd_nam;
347	caddr_t dpos = nfsd->nd_dpos;
348	struct ucred *cred = nfsd->nd_cr;
349	struct vattr va, preat;
350	struct vattr *vap = &va;
351	struct nfsv2_sattr *sp;
352	struct nfs_fattr *fp;
353	struct vnode *vp = NULL;
354	nfsfh_t nfh;
355	fhandle_t *fhp;
356	u_int32_t *tl;
357	caddr_t bpos;
358	int error = 0, rdonly, preat_ret = 1, postat_ret = 1;
359	int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
360	struct mbuf *mb, *mreq;
361	struct timespec guard = { 0, 0 };
362	struct mount *mp = NULL;
363
364	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
365	fhp = &nfh.fh_generic;
366	nfsm_srvmtofh(fhp);
367	if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
368		error = ESTALE;
369		goto out;
370	}
371	(void) vn_start_write(NULL, &mp, V_WAIT);
372	vfs_rel(mp);		/* The write holds a ref. */
373	VATTR_NULL(vap);
374	if (v3) {
375		nfsm_srvsattr(vap);
376		tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
377		gcheck = fxdr_unsigned(int, *tl);
378		if (gcheck) {
379			tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
380			fxdr_nfsv3time(tl, &guard);
381		}
382	} else {
383		sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
384		/*
385		 * Nah nah nah nah na nah
386		 * There is a bug in the Sun client that puts 0xffff in the mode
387		 * field of sattr when it should put in 0xffffffff. The u_short
388		 * doesn't sign extend.
389		 * --> check the low order 2 bytes for 0xffff
390		 */
391		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
392			vap->va_mode = nfstov_mode(sp->sa_mode);
393		if (sp->sa_uid != nfsrv_nfs_xdrneg1)
394			vap->va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
395		if (sp->sa_gid != nfsrv_nfs_xdrneg1)
396			vap->va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
397		if (sp->sa_size != nfsrv_nfs_xdrneg1)
398			vap->va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
399		if (sp->sa_atime.nfsv2_sec != nfsrv_nfs_xdrneg1) {
400#ifdef notyet
401			fxdr_nfsv2time(&sp->sa_atime, &vap->va_atime);
402#else
403			vap->va_atime.tv_sec =
404				fxdr_unsigned(int32_t, sp->sa_atime.nfsv2_sec);
405			vap->va_atime.tv_nsec = 0;
406#endif
407		}
408		if (sp->sa_mtime.nfsv2_sec != nfsrv_nfs_xdrneg1)
409			fxdr_nfsv2time(&sp->sa_mtime, &vap->va_mtime);
410
411	}
412
413	/*
414	 * Now that we have all the fields, lets do it.
415	 */
416	error = nfsrv_fhtovp(fhp, 0, &vp, nfsd, slp, nam, &rdonly);
417	if (error) {
418		nfsm_reply(2 * NFSX_UNSIGNED);
419		if (v3)
420			nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
421		error = 0;
422		goto nfsmout;
423	}
424
425	/*
426	 * vp now an active resource, pay careful attention to cleanup
427	 */
428	if (v3) {
429		error = preat_ret = VOP_GETATTR(vp, &preat, cred);
430		if (!error && gcheck &&
431			(preat.va_ctime.tv_sec != guard.tv_sec ||
432			 preat.va_ctime.tv_nsec != guard.tv_nsec))
433			error = NFSERR_NOT_SYNC;
434		if (error) {
435			vput(vp);
436			vp = NULL;
437			nfsm_reply(NFSX_WCCDATA(v3));
438			if (v3)
439				nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
440			error = 0;
441			goto nfsmout;
442		}
443	}
444
445	/*
446	 * If the size is being changed write acces is required, otherwise
447	 * just check for a read only filesystem.
448	 */
449	if (vap->va_size == ((u_quad_t)((quad_t) -1))) {
450		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
451			error = EROFS;
452			goto out;
453		}
454	} else {
455		if (vp->v_type == VDIR) {
456			error = EISDIR;
457			goto out;
458		} else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
459		    0)) != 0)
460			goto out;
461	}
462	error = VOP_SETATTR(vp, vap, cred);
463	postat_ret = VOP_GETATTR(vp, vap, cred);
464	if (!error)
465		error = postat_ret;
466out:
467	if (vp != NULL)
468		vput(vp);
469
470	vp = NULL;
471	nfsm_reply(NFSX_WCCORFATTR(v3));
472	if (v3) {
473		nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
474	} else if (!error) {
475		/* v2 non-error case. */
476		fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
477		nfsm_srvfillattr(vap, fp);
478	}
479	error = 0;
480	/* fall through */
481
482nfsmout:
483	if (vp)
484		vput(vp);
485	vn_finished_write(mp);
486	return(error);
487}
488
489/*
490 * nfs lookup rpc
491 */
492int
493nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
494    struct mbuf **mrq)
495{
496	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
497	struct sockaddr *nam = nfsd->nd_nam;
498	caddr_t dpos = nfsd->nd_dpos;
499	struct ucred *cred = nfsd->nd_cr;
500	struct nfs_fattr *fp;
501	struct nameidata nd, ind, *ndp = &nd;
502	struct vnode *vp, *dirp = NULL;
503	nfsfh_t nfh;
504	fhandle_t *fhp;
505	caddr_t bpos;
506	int error = 0, len, dirattr_ret = 1;
507	int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
508	struct mbuf *mb, *mreq;
509	struct vattr va, dirattr, *vap = &va;
510
511	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
512	ndclear(&nd);
513
514	fhp = &nfh.fh_generic;
515	nfsm_srvmtofh(fhp);
516	nfsm_srvnamesiz(len);
517
518	pubflag = nfs_ispublicfh(fhp);
519
520	nd.ni_cnd.cn_cred = cred;
521	nd.ni_cnd.cn_nameiop = LOOKUP;
522	nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART;
523	error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
524		&dirp, v3, &dirattr, &dirattr_ret, pubflag);
525
526	/*
527	 * namei failure, only dirp to cleanup.  Clear out garbarge from
528	 * structure in case macros jump to nfsmout.
529	 */
530
531	if (error) {
532		if (dirp) {
533			vrele(dirp);
534			dirp = NULL;
535		}
536		nfsm_reply(NFSX_POSTOPATTR(v3));
537		if (v3)
538			nfsm_srvpostop_attr(dirattr_ret, &dirattr);
539		error = 0;
540		goto nfsmout;
541	}
542
543	/*
544	 * Locate index file for public filehandle
545	 *
546	 * error is 0 on entry and 0 on exit from this block.
547	 */
548
549	if (pubflag) {
550		if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
551			/*
552			 * Setup call to lookup() to see if we can find
553			 * the index file. Arguably, this doesn't belong
554			 * in a kernel.. Ugh.  If an error occurs, do not
555			 * try to install an index file and then clear the
556			 * error.
557			 *
558			 * When we replace nd with ind and redirect ndp,
559			 * maintenance of ni_startdir and ni_vp shift to
560			 * ind and we have to clean them up in the old nd.
561			 * However, the cnd resource continues to be maintained
562			 * via the original nd.  Confused?  You aren't alone!
563			 */
564			ind = nd;
565			VOP_UNLOCK(nd.ni_vp, 0);
566			ind.ni_pathlen = strlen(nfs_pub.np_index);
567			ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
568			    nfs_pub.np_index;
569			ind.ni_startdir = nd.ni_vp;
570			VREF(ind.ni_startdir);
571			error = lookup(&ind);
572			ind.ni_dvp = NULL;
573
574			if (error == 0) {
575				/*
576				 * Found an index file. Get rid of
577				 * the old references.  transfer nd.ni_vp'
578				 */
579				if (dirp)
580					vrele(dirp);
581				dirp = nd.ni_vp;
582				nd.ni_vp = NULL;
583				vrele(nd.ni_startdir);
584				nd.ni_startdir = NULL;
585				ndp = &ind;
586			}
587			error = 0;
588		}
589		/*
590		 * If the public filehandle was used, check that this lookup
591		 * didn't result in a filehandle outside the publicly exported
592		 * filesystem.  We clear the poor vp here to avoid lockups due
593		 * to NFS I/O.
594		 */
595
596		if (ndp->ni_vp->v_mount != nfs_pub.np_mount) {
597			vput(nd.ni_vp);
598			nd.ni_vp = NULL;
599			error = EPERM;
600		}
601	}
602
603	/*
604	 * Resources at this point:
605	 *	ndp->ni_vp	may not be NULL
606	 */
607
608	if (error) {
609		nfsm_reply(NFSX_POSTOPATTR(v3));
610		if (v3)
611			nfsm_srvpostop_attr(dirattr_ret, &dirattr);
612		error = 0;
613		goto nfsmout;
614	}
615
616	/*
617	 * Get underlying attribute, then release remaining resources ( for
618	 * the same potential blocking reason ) and reply.
619	 */
620	vp = ndp->ni_vp;
621	bzero((caddr_t)fhp, sizeof(nfh));
622	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
623	error = VOP_VPTOFH(vp, &fhp->fh_fid);
624	if (!error)
625		error = VOP_GETATTR(vp, vap, cred);
626
627	vput(vp);
628	vrele(ndp->ni_startdir);
629	vrele(dirp);
630	ndp->ni_vp = NULL;
631	ndp->ni_startdir = NULL;
632	dirp = NULL;
633	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
634	if (error) {
635		if (v3)
636			nfsm_srvpostop_attr(dirattr_ret, &dirattr);
637		error = 0;
638		goto nfsmout;
639	}
640	nfsm_srvfhtom(fhp, v3);
641	if (v3) {
642		nfsm_srvpostop_attr(0, vap);
643		nfsm_srvpostop_attr(dirattr_ret, &dirattr);
644	} else {
645		fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
646		nfsm_srvfillattr(vap, fp);
647	}
648
649nfsmout:
650	if (ndp->ni_vp || dirp || ndp->ni_startdir) {
651		if (ndp->ni_vp)
652			vput(ndp->ni_vp);
653		if (dirp)
654			vrele(dirp);
655		if (ndp->ni_startdir)
656			vrele(ndp->ni_startdir);
657	}
658	NDFREE(&nd, NDF_ONLY_PNBUF);
659	return (error);
660}
661
662/*
663 * nfs readlink service
664 */
665int
666nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
667    struct mbuf **mrq)
668{
669	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
670	struct sockaddr *nam = nfsd->nd_nam;
671	caddr_t dpos = nfsd->nd_dpos;
672	struct ucred *cred = nfsd->nd_cr;
673	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
674	struct iovec *ivp = iv;
675	struct mbuf *mp;
676	u_int32_t *tl;
677	caddr_t bpos;
678	int error = 0, rdonly, i, tlen, len, getret;
679	int v3 = (nfsd->nd_flag & ND_NFSV3);
680	struct mbuf *mb, *mp3, *nmp, *mreq;
681	struct vnode *vp = NULL;
682	struct vattr attr;
683	nfsfh_t nfh;
684	fhandle_t *fhp;
685	struct uio io, *uiop = &io;
686
687	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
688#ifndef nolint
689	mp = NULL;
690#endif
691	mp3 = NULL;
692	fhp = &nfh.fh_generic;
693	nfsm_srvmtofh(fhp);
694	len = 0;
695	i = 0;
696	while (len < NFS_MAXPATHLEN) {
697		MGET(nmp, M_WAITOK, MT_DATA);
698		MCLGET(nmp, M_WAITOK);
699		nmp->m_len = NFSMSIZ(nmp);
700		if (len == 0)
701			mp3 = mp = nmp;
702		else {
703			mp->m_next = nmp;
704			mp = nmp;
705		}
706		if ((len + mp->m_len) > NFS_MAXPATHLEN) {
707			mp->m_len = NFS_MAXPATHLEN - len;
708			len = NFS_MAXPATHLEN;
709		} else
710			len += mp->m_len;
711		ivp->iov_base = mtod(mp, caddr_t);
712		ivp->iov_len = mp->m_len;
713		i++;
714		ivp++;
715	}
716	uiop->uio_iov = iv;
717	uiop->uio_iovcnt = i;
718	uiop->uio_offset = 0;
719	uiop->uio_resid = len;
720	uiop->uio_rw = UIO_READ;
721	uiop->uio_segflg = UIO_SYSSPACE;
722	uiop->uio_td = NULL;
723	error = nfsrv_fhtovp(fhp, 0, &vp, nfsd, slp, nam, &rdonly);
724	if (error) {
725		nfsm_reply(2 * NFSX_UNSIGNED);
726		if (v3)
727			nfsm_srvpostop_attr(1, NULL);
728		error = 0;
729		goto nfsmout;
730	}
731	if (vp->v_type != VLNK) {
732		if (v3)
733			error = EINVAL;
734		else
735			error = ENXIO;
736	} else
737		error = VOP_READLINK(vp, uiop, cred);
738	getret = VOP_GETATTR(vp, &attr, cred);
739	vput(vp);
740	vp = NULL;
741	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
742	if (v3)
743		nfsm_srvpostop_attr(getret, &attr);
744	if (error) {
745		error = 0;
746		goto nfsmout;
747	}
748	if (uiop->uio_resid > 0) {
749		len -= uiop->uio_resid;
750		tlen = nfsm_rndup(len);
751		nfsm_adj(mp3, NFS_MAXPATHLEN-tlen, tlen-len);
752	}
753	tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
754	*tl = txdr_unsigned(len);
755	mb->m_next = mp3;
756	mp3 = NULL;
757nfsmout:
758	if (mp3)
759		m_freem(mp3);
760	if (vp)
761		vput(vp);
762	return(error);
763}
764
765/*
766 * nfs read service
767 */
768int
769nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
770    struct mbuf **mrq)
771{
772	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
773	struct sockaddr *nam = nfsd->nd_nam;
774	caddr_t dpos = nfsd->nd_dpos;
775	struct ucred *cred = nfsd->nd_cr;
776	struct iovec *iv;
777	struct iovec *iv2;
778	struct mbuf *m;
779	struct nfs_fattr *fp;
780	u_int32_t *tl;
781	int i;
782	caddr_t bpos;
783	int error = 0, rdonly, cnt, len, left, siz, tlen, getret;
784	int v3 = (nfsd->nd_flag & ND_NFSV3), reqlen;
785	struct mbuf *mb, *mreq;
786	struct mbuf *m2;
787	struct vnode *vp = NULL;
788	nfsfh_t nfh;
789	fhandle_t *fhp;
790	struct uio io, *uiop = &io;
791	struct vattr va, *vap = &va;
792	struct nfsheur *nh;
793	off_t off;
794	int ioflag = 0;
795
796	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
797	fhp = &nfh.fh_generic;
798	nfsm_srvmtofh(fhp);
799	if (v3) {
800		tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
801		off = fxdr_hyper(tl);
802	} else {
803		tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
804		off = (off_t)fxdr_unsigned(u_int32_t, *tl);
805	}
806	nfsm_srvstrsiz(reqlen, NFS_SRVMAXDATA(nfsd));
807
808	/*
809	 * Reference vp.  If an error occurs, vp will be invalid, but we
810	 * have to NULL it just in case.  The macros might goto nfsmout
811	 * as well.
812	 */
813
814	error = nfsrv_fhtovp(fhp, 0, &vp, nfsd, slp, nam, &rdonly);
815	if (error) {
816		vp = NULL;
817		nfsm_reply(2 * NFSX_UNSIGNED);
818		if (v3)
819			nfsm_srvpostop_attr(1, NULL);
820		error = 0;
821		goto nfsmout;
822	}
823
824	if (vp->v_type != VREG) {
825		if (v3)
826			error = EINVAL;
827		else
828			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
829	}
830	if (!error) {
831		if ((error = nfsrv_access(vp, VREAD, cred, rdonly, 1)) != 0)
832			error = nfsrv_access(vp, VEXEC, cred, rdonly, 1);
833	}
834	getret = VOP_GETATTR(vp, vap, cred);
835	if (!error)
836		error = getret;
837	if (error) {
838		vput(vp);
839		vp = NULL;
840		nfsm_reply(NFSX_POSTOPATTR(v3));
841		if (v3)
842			nfsm_srvpostop_attr(getret, vap);
843		error = 0;
844		goto nfsmout;
845	}
846
847	/*
848	 * Calculate byte count to read
849	 */
850	if (off >= vap->va_size)
851		cnt = 0;
852	else if ((off + reqlen) > vap->va_size)
853		cnt = vap->va_size - off;
854	else
855		cnt = reqlen;
856
857	nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
858	if (v3) {
859		tl = nfsm_build(u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
860		*tl++ = nfsrv_nfs_true;
861		fp = (struct nfs_fattr *)tl;
862		tl += (NFSX_V3FATTR / sizeof (u_int32_t));
863	} else {
864		tl = nfsm_build(u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
865		fp = (struct nfs_fattr *)tl;
866		tl += (NFSX_V2FATTR / sizeof (u_int32_t));
867	}
868	len = left = nfsm_rndup(cnt);
869	if (cnt > 0) {
870		/*
871		 * Generate the mbuf list with the uio_iov ref. to it.
872		 */
873		i = 0;
874		m = m2 = mb;
875		while (left > 0) {
876			siz = min(M_TRAILINGSPACE(m), left);
877			if (siz > 0) {
878				left -= siz;
879				i++;
880			}
881			if (left > 0) {
882				MGET(m, M_WAITOK, MT_DATA);
883				MCLGET(m, M_WAITOK);
884				m->m_len = 0;
885				m2->m_next = m;
886				m2 = m;
887			}
888		}
889		iv = malloc(i * sizeof (struct iovec),
890		       M_TEMP, M_WAITOK);
891		uiop->uio_iov = iv2 = iv;
892		m = mb;
893		left = len;
894		i = 0;
895		while (left > 0) {
896			if (m == NULL)
897				panic("nfsrv_read iov");
898			siz = min(M_TRAILINGSPACE(m), left);
899			if (siz > 0) {
900				iv->iov_base = mtod(m, caddr_t) + m->m_len;
901				iv->iov_len = siz;
902				m->m_len += siz;
903				left -= siz;
904				iv++;
905				i++;
906			}
907			m = m->m_next;
908		}
909		uiop->uio_iovcnt = i;
910		uiop->uio_offset = off;
911		uiop->uio_resid = len;
912		uiop->uio_rw = UIO_READ;
913		uiop->uio_segflg = UIO_SYSSPACE;
914		uiop->uio_td = NULL;
915		nh = nfsrv_sequential_heuristic(uiop, vp);
916		ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
917		error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
918		if (error == 0)
919			nh->nh_nextoff = uiop->uio_offset;
920		free((caddr_t)iv2, M_TEMP);
921		if (error || (getret = VOP_GETATTR(vp, vap, cred))) {
922			if (!error)
923				error = getret;
924			m_freem(mreq);
925			vput(vp);
926			vp = NULL;
927			nfsm_reply(NFSX_POSTOPATTR(v3));
928			if (v3)
929				nfsm_srvpostop_attr(getret, vap);
930			error = 0;
931			goto nfsmout;
932		}
933	} else
934		uiop->uio_resid = 0;
935	vput(vp);
936	vp = NULL;
937	nfsm_srvfillattr(vap, fp);
938	tlen = len - uiop->uio_resid;
939	cnt = cnt < tlen ? cnt : tlen;
940	tlen = nfsm_rndup(cnt);
941	if (len != tlen || tlen != cnt)
942		nfsm_adj(mb, len - tlen, tlen - cnt);
943	if (v3) {
944		*tl++ = txdr_unsigned(cnt);
945		if (cnt < reqlen)
946			*tl++ = nfsrv_nfs_true;
947		else
948			*tl++ = nfsrv_nfs_false;
949	}
950	*tl = txdr_unsigned(cnt);
951nfsmout:
952	if (vp)
953		vput(vp);
954	return(error);
955}
956
957/*
958 * nfs write service
959 */
960int
961nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
962    struct mbuf **mrq)
963{
964	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
965	struct sockaddr *nam = nfsd->nd_nam;
966	caddr_t dpos = nfsd->nd_dpos;
967	struct ucred *cred = nfsd->nd_cr;
968	struct iovec *ivp;
969	int i, cnt;
970	struct mbuf *mp;
971	struct nfs_fattr *fp;
972	struct iovec *iv;
973	struct vattr va, forat;
974	struct vattr *vap = &va;
975	u_int32_t *tl;
976	caddr_t bpos;
977	int error = 0, rdonly, len, forat_ret = 1;
978	int ioflags, aftat_ret = 1, retlen = 0, zeroing, adjust;
979	int stable = NFSV3WRITE_FILESYNC;
980	int v3 = (nfsd->nd_flag & ND_NFSV3);
981	struct mbuf *mb, *mreq;
982	struct vnode *vp = NULL;
983	struct nfsheur *nh;
984	nfsfh_t nfh;
985	fhandle_t *fhp;
986	struct uio io, *uiop = &io;
987	off_t off;
988	struct mount *mntp = NULL;
989
990	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
991	if (mrep == NULL) {
992		*mrq = NULL;
993		error = 0;
994		goto nfsmout;
995	}
996	fhp = &nfh.fh_generic;
997	nfsm_srvmtofh(fhp);
998	if ((mntp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
999		error = ESTALE;
1000		goto ereply;
1001	}
1002	(void) vn_start_write(NULL, &mntp, V_WAIT);
1003	vfs_rel(mntp);		/* The write holds a ref. */
1004	if (v3) {
1005		tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
1006		off = fxdr_hyper(tl);
1007		tl += 3;
1008		stable = fxdr_unsigned(int, *tl++);
1009	} else {
1010		tl = nfsm_dissect_nonblock(u_int32_t *, 4 * NFSX_UNSIGNED);
1011		off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1012		tl += 2;
1013		if (nfs_async)
1014	    		stable = NFSV3WRITE_UNSTABLE;
1015	}
1016	retlen = len = fxdr_unsigned(int32_t, *tl);
1017	cnt = i = 0;
1018
1019	/*
1020	 * For NFS Version 2, it is not obvious what a write of zero length
1021	 * should do, but I might as well be consistent with Version 3,
1022	 * which is to return ok so long as there are no permission problems.
1023	 */
1024	if (len > 0) {
1025	    zeroing = 1;
1026	    mp = mrep;
1027	    while (mp) {
1028		if (mp == md) {
1029			zeroing = 0;
1030			adjust = dpos - mtod(mp, caddr_t);
1031			mp->m_len -= adjust;
1032			if (mp->m_len > 0 && adjust > 0)
1033				mp->m_data += adjust;
1034		}
1035		if (zeroing)
1036			mp->m_len = 0;
1037		else if (mp->m_len > 0) {
1038			i += mp->m_len;
1039			if (i > len) {
1040				mp->m_len -= (i - len);
1041				zeroing	= 1;
1042			}
1043			if (mp->m_len > 0)
1044				cnt++;
1045		}
1046		mp = mp->m_next;
1047	    }
1048	}
1049	if (len > NFS_MAXDATA || len < 0 || i < len) {
1050		error = EIO;
1051		nfsm_reply(2 * NFSX_UNSIGNED);
1052		if (v3)
1053			nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1054		error = 0;
1055		goto nfsmout;
1056	}
1057	error = nfsrv_fhtovp(fhp, 0, &vp, nfsd, slp, nam, &rdonly);
1058	if (error) {
1059		vp = NULL;
1060		nfsm_reply(2 * NFSX_UNSIGNED);
1061		if (v3)
1062			nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1063		error = 0;
1064		goto nfsmout;
1065	}
1066	if (v3)
1067		forat_ret = VOP_GETATTR(vp, &forat, cred);
1068	if (vp->v_type != VREG) {
1069		if (v3)
1070			error = EINVAL;
1071		else
1072			error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1073	}
1074	if (!error)
1075		error = nfsrv_access(vp, VWRITE, cred, rdonly, 1);
1076	if (error) {
1077		vput(vp);
1078		vp = NULL;
1079		nfsm_reply(NFSX_WCCDATA(v3));
1080		if (v3)
1081			nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1082		error = 0;
1083		goto nfsmout;
1084	}
1085
1086	if (len > 0) {
1087	    ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
1088		M_WAITOK);
1089	    uiop->uio_iov = iv = ivp;
1090	    uiop->uio_iovcnt = cnt;
1091	    mp = mrep;
1092	    while (mp) {
1093		if (mp->m_len > 0) {
1094			ivp->iov_base = mtod(mp, caddr_t);
1095			ivp->iov_len = mp->m_len;
1096			ivp++;
1097		}
1098		mp = mp->m_next;
1099	    }
1100
1101	    /*
1102	     * XXX
1103	     * The IO_METASYNC flag indicates that all metadata (and not just
1104	     * enough to ensure data integrity) mus be written to stable storage
1105	     * synchronously.
1106	     * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
1107	     */
1108	    if (stable == NFSV3WRITE_UNSTABLE)
1109		ioflags = IO_NODELOCKED;
1110	    else if (stable == NFSV3WRITE_DATASYNC)
1111		ioflags = (IO_SYNC | IO_NODELOCKED);
1112	    else
1113		ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1114	    uiop->uio_resid = len;
1115	    uiop->uio_rw = UIO_WRITE;
1116	    uiop->uio_segflg = UIO_SYSSPACE;
1117	    uiop->uio_td = NULL;
1118	    uiop->uio_offset = off;
1119	    nh = nfsrv_sequential_heuristic(uiop, vp);
1120	    ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
1121	    error = VOP_WRITE(vp, uiop, ioflags, cred);
1122	    if (error == 0)
1123		    nh->nh_nextoff = uiop->uio_offset;
1124	    /* Unlocked write. */
1125	    nfsrvstats.srvvop_writes++;
1126	    free((caddr_t)iv, M_TEMP);
1127	}
1128	aftat_ret = VOP_GETATTR(vp, vap, cred);
1129	vput(vp);
1130	vp = NULL;
1131	if (!error)
1132		error = aftat_ret;
1133ereply:
1134	nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
1135		2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
1136	if (v3) {
1137		nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1138		if (error) {
1139			error = 0;
1140			goto nfsmout;
1141		}
1142		tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1143		*tl++ = txdr_unsigned(retlen);
1144		/*
1145		 * If nfs_async is set, then pretend the write was FILESYNC.
1146		 */
1147		if (stable == NFSV3WRITE_UNSTABLE && !nfs_async)
1148			*tl++ = txdr_unsigned(stable);
1149		else
1150			*tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1151		/*
1152		 * Actually, there is no need to txdr these fields,
1153		 * but it may make the values more human readable,
1154		 * for debugging purposes.
1155		 */
1156		if (nfsver.tv_sec == 0)
1157			nfsver = boottime;
1158		*tl++ = txdr_unsigned(nfsver.tv_sec);
1159		*tl = txdr_unsigned(nfsver.tv_usec);
1160	} else if (!error) {
1161		/* v2 non-error case. */
1162		fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1163		nfsm_srvfillattr(vap, fp);
1164	}
1165	error = 0;
1166nfsmout:
1167	if (vp)
1168		vput(vp);
1169	vn_finished_write(mntp);
1170	return(error);
1171}
1172
1173/*
1174 * nfs create service
1175 * now does a truncate to 0 length via. setattr if it already exists
1176 */
1177int
1178nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1179    struct mbuf **mrq)
1180{
1181	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1182	struct sockaddr *nam = nfsd->nd_nam;
1183	caddr_t dpos = nfsd->nd_dpos;
1184	struct ucred *cred = nfsd->nd_cr;
1185	struct nfs_fattr *fp;
1186	struct vattr va, dirfor, diraft;
1187	struct vattr *vap = &va;
1188	struct nfsv2_sattr *sp;
1189	u_int32_t *tl;
1190	struct nameidata nd;
1191	caddr_t bpos;
1192	int error = 0, rdev, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1193	int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1194	struct mbuf *mb, *mreq;
1195	struct vnode *dirp = NULL;
1196	nfsfh_t nfh;
1197	fhandle_t *fhp;
1198	u_quad_t tempsize;
1199	struct timespec cverf;
1200	struct mount *mp = NULL;
1201
1202	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1203#ifndef nolint
1204	rdev = 0;
1205#endif
1206	ndclear(&nd);
1207
1208	fhp = &nfh.fh_generic;
1209	nfsm_srvmtofh(fhp);
1210	if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1211		error = ESTALE;
1212		goto ereply;
1213	}
1214	(void) vn_start_write(NULL, &mp, V_WAIT);
1215	vfs_rel(mp);		/* The write holds a ref. */
1216	nfsm_srvnamesiz(len);
1217
1218	nd.ni_cnd.cn_cred = cred;
1219	nd.ni_cnd.cn_nameiop = CREATE;
1220	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART;
1221
1222	/*
1223	 * Call namei and do initial cleanup to get a few things
1224	 * out of the way.  If we get an initial error we cleanup
1225	 * and return here to avoid special-casing the invalid nd
1226	 * structure through the rest of the case.  dirp may be
1227	 * set even if an error occurs, but the nd structure will not
1228	 * be valid at all if an error occurs so we have to invalidate it
1229	 * prior to calling nfsm_reply ( which might goto nfsmout ).
1230	 */
1231	error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
1232		&dirp, v3, &dirfor, &dirfor_ret, FALSE);
1233	if (dirp && !v3) {
1234		vrele(dirp);
1235		dirp = NULL;
1236	}
1237	if (error) {
1238		nfsm_reply(NFSX_WCCDATA(v3));
1239		if (v3)
1240			nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1241		error = 0;
1242		goto nfsmout;
1243	}
1244
1245	/*
1246	 * No error.  Continue.  State:
1247	 *
1248	 *	startdir	is valid ( we release this immediately )
1249	 *	dirp 		may be valid
1250	 *	nd.ni_vp	may be valid
1251	 *	nd.ni_dvp	is valid
1252	 *
1253	 * The error state is set through the code and we may also do some
1254	 * opportunistic releasing of vnodes to avoid holding locks through
1255	 * NFS I/O.  The cleanup at the end is a catch-all
1256	 */
1257
1258	VATTR_NULL(vap);
1259	if (v3) {
1260		tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
1261		how = fxdr_unsigned(int, *tl);
1262		switch (how) {
1263		case NFSV3CREATE_GUARDED:
1264			if (nd.ni_vp) {
1265				error = EEXIST;
1266				break;
1267			}
1268			/* fall through */
1269		case NFSV3CREATE_UNCHECKED:
1270			nfsm_srvsattr(vap);
1271			break;
1272		case NFSV3CREATE_EXCLUSIVE:
1273			tl = nfsm_dissect_nonblock(u_int32_t *,
1274			    NFSX_V3CREATEVERF);
1275			/* Unique bytes, endianness is not important. */
1276			cverf.tv_sec  = (int32_t)tl[0];
1277			cverf.tv_nsec = tl[1];
1278			exclusive_flag = 1;
1279			break;
1280		};
1281		vap->va_type = VREG;
1282	} else {
1283		sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
1284		vap->va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1285		if (vap->va_type == VNON)
1286			vap->va_type = VREG;
1287		vap->va_mode = nfstov_mode(sp->sa_mode);
1288		switch (vap->va_type) {
1289		case VREG:
1290			tsize = fxdr_unsigned(int32_t, sp->sa_size);
1291			if (tsize != -1)
1292				vap->va_size = (u_quad_t)tsize;
1293			break;
1294		case VCHR:
1295		case VBLK:
1296		case VFIFO:
1297			rdev = fxdr_unsigned(long, sp->sa_size);
1298			break;
1299		default:
1300			break;
1301		};
1302	}
1303
1304	/*
1305	 * Iff doesn't exist, create it
1306	 * otherwise just truncate to 0 length
1307	 *   should I set the mode too ?
1308	 *
1309	 * The only possible error we can have at this point is EEXIST.
1310	 * nd.ni_vp will also be non-NULL in that case.
1311	 */
1312	if (nd.ni_vp == NULL) {
1313		if (vap->va_mode == (mode_t)VNOVAL)
1314			vap->va_mode = 0;
1315		if (vap->va_type == VREG || vap->va_type == VSOCK) {
1316			error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1317			if (error)
1318				NDFREE(&nd, NDF_ONLY_PNBUF);
1319			else {
1320				if (exclusive_flag) {
1321					exclusive_flag = 0;
1322					VATTR_NULL(vap);
1323					vap->va_atime = cverf;
1324					error = VOP_SETATTR(nd.ni_vp, vap,
1325					    cred);
1326				}
1327			}
1328		} else if (vap->va_type == VCHR || vap->va_type == VBLK ||
1329		    vap->va_type == VFIFO) {
1330			/*
1331			 * NFSv2-specific code for creating device nodes
1332			 * and fifos.
1333			 *
1334			 * Handle SysV FIFO node special cases.  All other
1335			 * devices require super user to access.
1336			 */
1337			if (vap->va_type == VCHR && rdev == 0xffffffff)
1338				vap->va_type = VFIFO;
1339                        if (vap->va_type != VFIFO &&
1340			    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV,
1341			    0))) {
1342				goto ereply;
1343                        }
1344			vap->va_rdev = rdev;
1345			error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1346			if (error) {
1347				NDFREE(&nd, NDF_ONLY_PNBUF);
1348				goto ereply;
1349			}
1350			vput(nd.ni_vp);
1351			nd.ni_vp = NULL;
1352
1353			/*
1354			 * release dvp prior to lookup
1355			 */
1356			vput(nd.ni_dvp);
1357			nd.ni_dvp = NULL;
1358			/*
1359			 * Setup for lookup.
1360			 *
1361			 * Even though LOCKPARENT was cleared, ni_dvp may
1362			 * be garbage.
1363			 */
1364			nd.ni_cnd.cn_nameiop = LOOKUP;
1365			nd.ni_cnd.cn_flags &= ~(LOCKPARENT);
1366			nd.ni_cnd.cn_thread = curthread;
1367			nd.ni_cnd.cn_cred = cred;
1368			error = lookup(&nd);
1369			nd.ni_dvp = NULL;
1370			if (error)
1371				goto ereply;
1372
1373			if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1374				error = EINVAL;
1375				goto ereply;
1376			}
1377		} else {
1378			error = ENXIO;
1379		}
1380	} else {
1381		if (vap->va_size != -1) {
1382			error = nfsrv_access(nd.ni_vp, VWRITE,
1383			    cred, (nd.ni_cnd.cn_flags & RDONLY), 0);
1384			if (!error) {
1385				tempsize = vap->va_size;
1386				VATTR_NULL(vap);
1387				vap->va_size = tempsize;
1388				error = VOP_SETATTR(nd.ni_vp, vap, cred);
1389			}
1390		}
1391	}
1392
1393	if (!error) {
1394		bzero((caddr_t)fhp, sizeof(nfh));
1395		fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
1396		error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
1397		if (!error)
1398			error = VOP_GETATTR(nd.ni_vp, vap, cred);
1399	}
1400	if (v3) {
1401		if (exclusive_flag && !error &&
1402		    bcmp(&cverf, &vap->va_atime, sizeof (cverf)))
1403			error = EEXIST;
1404		if (dirp == nd.ni_dvp)
1405			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1406		else {
1407			/* Drop the other locks to avoid deadlock. */
1408			if (nd.ni_dvp) {
1409				if (nd.ni_dvp == nd.ni_vp)
1410					vrele(nd.ni_dvp);
1411				else
1412					vput(nd.ni_dvp);
1413			}
1414			if (nd.ni_vp)
1415				vput(nd.ni_vp);
1416			nd.ni_dvp = NULL;
1417			nd.ni_vp = NULL;
1418
1419			vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
1420			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1421			VOP_UNLOCK(dirp, 0);
1422		}
1423	}
1424ereply:
1425	nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1426	if (v3) {
1427		if (!error) {
1428			nfsm_srvpostop_fh(fhp);
1429			nfsm_srvpostop_attr(0, vap);
1430		}
1431		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1432	} else if (!error) {
1433		/* v2 non-error case. */
1434		nfsm_srvfhtom(fhp, v3);
1435		fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1436		nfsm_srvfillattr(vap, fp);
1437	}
1438	error = 0;
1439
1440nfsmout:
1441	if (nd.ni_dvp) {
1442		if (nd.ni_dvp == nd.ni_vp)
1443			vrele(nd.ni_dvp);
1444		else
1445			vput(nd.ni_dvp);
1446	}
1447	if (nd.ni_vp)
1448		vput(nd.ni_vp);
1449	if (nd.ni_startdir) {
1450		vrele(nd.ni_startdir);
1451		nd.ni_startdir = NULL;
1452	}
1453	if (dirp)
1454		vrele(dirp);
1455	NDFREE(&nd, NDF_ONLY_PNBUF);
1456	vn_finished_write(mp);
1457	return (error);
1458}
1459
1460/*
1461 * nfs v3 mknod service
1462 */
1463int
1464nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1465    struct mbuf **mrq)
1466{
1467	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1468	struct sockaddr *nam = nfsd->nd_nam;
1469	caddr_t dpos = nfsd->nd_dpos;
1470	struct ucred *cred = nfsd->nd_cr;
1471	struct vattr va, dirfor, diraft;
1472	struct vattr *vap = &va;
1473	struct thread *td = curthread;
1474	u_int32_t *tl;
1475	struct nameidata nd;
1476	caddr_t bpos;
1477	int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1478	u_int32_t major, minor;
1479	enum vtype vtyp;
1480	struct mbuf *mb, *mreq;
1481	struct vnode *vp, *dirp = NULL;
1482	nfsfh_t nfh;
1483	fhandle_t *fhp;
1484	struct mount *mp = NULL;
1485	int v3 = (nfsd->nd_flag & ND_NFSV3);
1486
1487	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1488	if (!v3)
1489		panic("nfsrv_mknod: v3 proc called on a v2 connection");
1490	ndclear(&nd);
1491
1492	fhp = &nfh.fh_generic;
1493	nfsm_srvmtofh(fhp);
1494	if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1495		error = ESTALE;
1496		goto ereply;
1497	}
1498	(void) vn_start_write(NULL, &mp, V_WAIT);
1499	vfs_rel(mp);		/* The write holds a ref. */
1500	nfsm_srvnamesiz(len);
1501
1502	nd.ni_cnd.cn_cred = cred;
1503	nd.ni_cnd.cn_nameiop = CREATE;
1504	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART;
1505
1506	/*
1507	 * Handle nfs_namei() call.  If an error occurs, the nd structure
1508	 * is not valid.  However, nfsm_*() routines may still jump to
1509	 * nfsmout.
1510	 */
1511
1512	error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
1513		&dirp, v3, &dirfor, &dirfor_ret, FALSE);
1514	if (error) {
1515		nfsm_reply(NFSX_WCCDATA(1));
1516		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1517		error = 0;
1518		goto nfsmout;
1519	}
1520	tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
1521	vtyp = nfsv3tov_type(*tl);
1522	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1523		error = NFSERR_BADTYPE;
1524		goto out;
1525	}
1526	VATTR_NULL(vap);
1527	nfsm_srvsattr(vap);
1528	if (vtyp == VCHR || vtyp == VBLK) {
1529		tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
1530		major = fxdr_unsigned(u_int32_t, *tl++);
1531		minor = fxdr_unsigned(u_int32_t, *tl);
1532		vap->va_rdev = makedev(major, minor);
1533	}
1534
1535	/*
1536	 * Iff doesn't exist, create it.
1537	 */
1538	if (nd.ni_vp) {
1539		error = EEXIST;
1540		goto out;
1541	}
1542	vap->va_type = vtyp;
1543	if (vap->va_mode == (mode_t)VNOVAL)
1544		vap->va_mode = 0;
1545	if (vtyp == VSOCK) {
1546		vrele(nd.ni_startdir);
1547		nd.ni_startdir = NULL;
1548		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1549		if (error)
1550			NDFREE(&nd, NDF_ONLY_PNBUF);
1551	} else {
1552		if (vtyp != VFIFO && (error = priv_check_cred(cred,
1553		    PRIV_VFS_MKNOD_DEV, 0)))
1554			goto out;
1555		error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1556		if (error) {
1557			NDFREE(&nd, NDF_ONLY_PNBUF);
1558			goto out;
1559		}
1560		vput(nd.ni_vp);
1561		nd.ni_vp = NULL;
1562
1563		/*
1564		 * Release dvp prior to lookup
1565		 */
1566		vput(nd.ni_dvp);
1567		nd.ni_dvp = NULL;
1568
1569		nd.ni_cnd.cn_nameiop = LOOKUP;
1570		nd.ni_cnd.cn_flags &= ~(LOCKPARENT);
1571		nd.ni_cnd.cn_thread = td;
1572		nd.ni_cnd.cn_cred = td->td_ucred;
1573		error = lookup(&nd);
1574		nd.ni_dvp = NULL;
1575
1576		if (error)
1577			goto out;
1578		if (nd.ni_cnd.cn_flags & ISSYMLINK)
1579			error = EINVAL;
1580	}
1581
1582	/*
1583	 * send response, cleanup, return.
1584	 */
1585out:
1586	vp = nd.ni_vp;
1587	if (!error) {
1588		bzero((caddr_t)fhp, sizeof(nfh));
1589		fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1590		error = VOP_VPTOFH(vp, &fhp->fh_fid);
1591		if (!error)
1592			error = VOP_GETATTR(vp, vap, cred);
1593	}
1594	if (nd.ni_dvp) {
1595		if (nd.ni_dvp == nd.ni_vp)
1596			vrele(nd.ni_dvp);
1597		else
1598			vput(nd.ni_dvp);
1599		nd.ni_dvp = NULL;
1600	}
1601	if (vp) {
1602		vput(vp);
1603		vp = NULL;
1604		nd.ni_vp = NULL;
1605	}
1606	if (nd.ni_startdir) {
1607		vrele(nd.ni_startdir);
1608		nd.ni_startdir = NULL;
1609	}
1610	NDFREE(&nd, NDF_ONLY_PNBUF);
1611	if (dirp) {
1612		vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
1613		diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1614		vput(dirp);
1615	}
1616ereply:
1617	nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
1618	if (v3) {
1619		if (!error) {
1620			nfsm_srvpostop_fh(fhp);
1621			nfsm_srvpostop_attr(0, vap);
1622		}
1623		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1624	}
1625	vn_finished_write(mp);
1626	return (0);
1627nfsmout:
1628	if (nd.ni_dvp) {
1629		if (nd.ni_dvp == nd.ni_vp)
1630			vrele(nd.ni_dvp);
1631		else
1632			vput(nd.ni_dvp);
1633	}
1634	if (nd.ni_vp)
1635		vput(nd.ni_vp);
1636	if (dirp)
1637		vrele(dirp);
1638	if (nd.ni_startdir)
1639		vrele(nd.ni_startdir);
1640	NDFREE(&nd, NDF_ONLY_PNBUF);
1641	vn_finished_write(mp);
1642	return (error);
1643}
1644
1645/*
1646 * nfs remove service
1647 */
1648int
1649nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1650    struct mbuf **mrq)
1651{
1652	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1653	struct sockaddr *nam = nfsd->nd_nam;
1654	caddr_t dpos = nfsd->nd_dpos;
1655	struct ucred *cred = nfsd->nd_cr;
1656	struct nameidata nd;
1657	caddr_t bpos;
1658	int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1659	int v3 = (nfsd->nd_flag & ND_NFSV3);
1660	struct mbuf *mb, *mreq;
1661	struct vnode *dirp;
1662	struct vattr dirfor, diraft;
1663	nfsfh_t nfh;
1664	fhandle_t *fhp;
1665	struct mount *mp = NULL;
1666
1667	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1668	ndclear(&nd);
1669
1670	fhp = &nfh.fh_generic;
1671	nfsm_srvmtofh(fhp);
1672	if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1673		error = ESTALE;
1674		goto ereply;
1675	}
1676	(void) vn_start_write(NULL, &mp, V_WAIT);
1677	vfs_rel(mp);		/* The write holds a ref. */
1678	nfsm_srvnamesiz(len);
1679
1680	nd.ni_cnd.cn_cred = cred;
1681	nd.ni_cnd.cn_nameiop = DELETE;
1682	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1683	error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
1684		&dirp, v3,  &dirfor, &dirfor_ret, FALSE);
1685	if (dirp && !v3) {
1686		vrele(dirp);
1687		dirp = NULL;
1688	}
1689	if (error == 0) {
1690		if (nd.ni_vp->v_type == VDIR) {
1691			error = EPERM;		/* POSIX */
1692			goto out;
1693		}
1694		/*
1695		 * The root of a mounted filesystem cannot be deleted.
1696		 */
1697		if (nd.ni_vp->v_vflag & VV_ROOT) {
1698			error = EBUSY;
1699			goto out;
1700		}
1701out:
1702		if (!error) {
1703			error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
1704			NDFREE(&nd, NDF_ONLY_PNBUF);
1705		}
1706	}
1707	if (dirp && v3) {
1708		if (dirp == nd.ni_dvp)
1709			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1710		else {
1711			/* Drop the other locks to avoid deadlock. */
1712			if (nd.ni_dvp) {
1713				if (nd.ni_dvp == nd.ni_vp)
1714					vrele(nd.ni_dvp);
1715				else
1716					vput(nd.ni_dvp);
1717			}
1718			if (nd.ni_vp)
1719				vput(nd.ni_vp);
1720			nd.ni_dvp = NULL;
1721			nd.ni_vp = NULL;
1722
1723			vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
1724			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1725			VOP_UNLOCK(dirp, 0);
1726		}
1727		vrele(dirp);
1728		dirp = NULL;
1729	}
1730ereply:
1731	nfsm_reply(NFSX_WCCDATA(v3));
1732	if (v3)
1733		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1734	error = 0;
1735nfsmout:
1736	NDFREE(&nd, NDF_ONLY_PNBUF);
1737	if (nd.ni_dvp) {
1738		if (nd.ni_dvp == nd.ni_vp)
1739			vrele(nd.ni_dvp);
1740		else
1741			vput(nd.ni_dvp);
1742	}
1743	if (nd.ni_vp)
1744		vput(nd.ni_vp);
1745	vn_finished_write(mp);
1746	return(error);
1747}
1748
1749/*
1750 * nfs rename service
1751 */
1752int
1753nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1754    struct mbuf **mrq)
1755{
1756	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1757	struct sockaddr *nam = nfsd->nd_nam;
1758	caddr_t dpos = nfsd->nd_dpos;
1759	struct ucred *cred = nfsd->nd_cr;
1760	caddr_t bpos;
1761	int error = 0, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
1762	int tdirfor_ret = 1, tdiraft_ret = 1;
1763	int v3 = (nfsd->nd_flag & ND_NFSV3);
1764	struct mbuf *mb, *mreq;
1765	struct nameidata fromnd, tond;
1766	struct vnode *fvp, *tvp, *tdvp, *fdirp = NULL;
1767	struct vnode *tdirp = NULL;
1768	struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
1769	nfsfh_t fnfh, tnfh;
1770	fhandle_t *ffhp, *tfhp;
1771	uid_t saved_uid;
1772	struct mount *mp = NULL;
1773
1774	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1775#ifndef nolint
1776	fvp = NULL;
1777#endif
1778	ffhp = &fnfh.fh_generic;
1779	tfhp = &tnfh.fh_generic;
1780
1781	/*
1782	 * Clear fields incase goto nfsmout occurs from macro.
1783	 */
1784
1785	ndclear(&fromnd);
1786	ndclear(&tond);
1787
1788	nfsm_srvmtofh(ffhp);
1789	if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL) {
1790		error = ESTALE;
1791		goto out1;
1792	}
1793	(void) vn_start_write(NULL, &mp, V_WAIT);
1794	vfs_rel(mp);		/* The write holds a ref. */
1795	nfsm_srvnamesiz(len);
1796	/*
1797	 * Remember our original uid so that we can reset cr_uid before
1798	 * the second nfs_namei() call, in case it is remapped.
1799	 */
1800	saved_uid = cred->cr_uid;
1801	fromnd.ni_cnd.cn_cred = cred;
1802	fromnd.ni_cnd.cn_nameiop = DELETE;
1803	fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART;
1804	error = nfs_namei(&fromnd, nfsd, ffhp, len, slp, nam, &md,
1805		&dpos, &fdirp, v3, &fdirfor, &fdirfor_ret, FALSE);
1806	if (fdirp && !v3) {
1807		vrele(fdirp);
1808		fdirp = NULL;
1809	}
1810	if (error) {
1811		nfsm_reply(2 * NFSX_WCCDATA(v3));
1812		if (v3) {
1813			nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1814			nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1815		}
1816		error = 0;
1817		goto nfsmout;
1818	}
1819	fvp = fromnd.ni_vp;
1820	nfsm_srvmtofh(tfhp);
1821	nfsm_srvnamesiz(len2);
1822	cred->cr_uid = saved_uid;
1823	tond.ni_cnd.cn_cred = cred;
1824	tond.ni_cnd.cn_nameiop = RENAME;
1825	tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART;
1826	error = nfs_namei(&tond, nfsd, tfhp, len2, slp, nam, &md,
1827		&dpos, &tdirp, v3, &tdirfor, &tdirfor_ret, FALSE);
1828	if (tdirp && !v3) {
1829		vrele(tdirp);
1830		tdirp = NULL;
1831	}
1832	if (error)
1833		goto out1;
1834
1835	tdvp = tond.ni_dvp;
1836	tvp = tond.ni_vp;
1837	if (tvp != NULL) {
1838		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1839			if (v3)
1840				error = EEXIST;
1841			else
1842				error = EISDIR;
1843			goto out;
1844		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1845			if (v3)
1846				error = EEXIST;
1847			else
1848				error = ENOTDIR;
1849			goto out;
1850		}
1851		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1852			if (v3)
1853				error = EXDEV;
1854			else
1855				error = ENOTEMPTY;
1856			goto out;
1857		}
1858	}
1859	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1860		if (v3)
1861			error = EXDEV;
1862		else
1863			error = ENOTEMPTY;
1864		goto out;
1865	}
1866	if (fvp->v_mount != tdvp->v_mount) {
1867		if (v3)
1868			error = EXDEV;
1869		else
1870			error = ENOTEMPTY;
1871		goto out;
1872	}
1873	if (fvp == tdvp) {
1874		if (v3)
1875			error = EINVAL;
1876		else
1877			error = ENOTEMPTY;
1878	}
1879	/*
1880	 * If source is the same as the destination (that is the
1881	 * same vnode with the same name in the same directory),
1882	 * then there is nothing to do.
1883	 */
1884	if (fvp == tvp && fromnd.ni_dvp == tdvp &&
1885	    fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
1886	    !bcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
1887	      fromnd.ni_cnd.cn_namelen))
1888		error = -1;
1889out:
1890	if (!error) {
1891		/*
1892		 * The VOP_RENAME function releases all vnode references &
1893		 * locks prior to returning so we need to clear the pointers
1894		 * to bypass cleanup code later on.
1895		 */
1896		error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
1897				   tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
1898		fromnd.ni_dvp = NULL;
1899		fromnd.ni_vp = NULL;
1900		tond.ni_dvp = NULL;
1901		tond.ni_vp = NULL;
1902		if (error) {
1903			NDFREE(&fromnd, NDF_ONLY_PNBUF);
1904			NDFREE(&tond, NDF_ONLY_PNBUF);
1905		}
1906	} else {
1907		if (error == -1)
1908			error = 0;
1909	}
1910	/* fall through */
1911out1:
1912	nfsm_reply(2 * NFSX_WCCDATA(v3));
1913	if (v3) {
1914		/* Release existing locks to prevent deadlock. */
1915		if (tond.ni_dvp) {
1916			if (tond.ni_dvp == tond.ni_vp)
1917				vrele(tond.ni_dvp);
1918			else
1919				vput(tond.ni_dvp);
1920		}
1921		if (tond.ni_vp)
1922			vput(tond.ni_vp);
1923		tond.ni_dvp = NULL;
1924		tond.ni_vp = NULL;
1925
1926		if (fdirp) {
1927			vn_lock(fdirp, LK_EXCLUSIVE | LK_RETRY);
1928			fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred);
1929			VOP_UNLOCK(fdirp, 0);
1930		}
1931		if (tdirp) {
1932			vn_lock(tdirp, LK_EXCLUSIVE | LK_RETRY);
1933			tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred);
1934			VOP_UNLOCK(tdirp, 0);
1935		}
1936		nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1937		nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1938	}
1939	error = 0;
1940	/* fall through */
1941
1942nfsmout:
1943	/*
1944	 * Clear out tond related fields
1945	 */
1946	if (tond.ni_dvp) {
1947		if (tond.ni_dvp == tond.ni_vp)
1948			vrele(tond.ni_dvp);
1949		else
1950			vput(tond.ni_dvp);
1951	}
1952	if (tond.ni_vp)
1953		vput(tond.ni_vp);
1954	if (tdirp)
1955		vrele(tdirp);
1956	if (tond.ni_startdir)
1957		vrele(tond.ni_startdir);
1958	NDFREE(&tond, NDF_ONLY_PNBUF);
1959	/*
1960	 * Clear out fromnd related fields
1961	 */
1962	if (fdirp)
1963		vrele(fdirp);
1964	if (fromnd.ni_startdir)
1965		vrele(fromnd.ni_startdir);
1966	NDFREE(&fromnd, NDF_ONLY_PNBUF);
1967	if (fromnd.ni_dvp)
1968		vrele(fromnd.ni_dvp);
1969	if (fromnd.ni_vp)
1970		vrele(fromnd.ni_vp);
1971
1972	vn_finished_write(mp);
1973	return (error);
1974}
1975
1976/*
1977 * nfs link service
1978 */
1979int
1980nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1981    struct mbuf **mrq)
1982{
1983	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1984	struct sockaddr *nam = nfsd->nd_nam;
1985	caddr_t dpos = nfsd->nd_dpos;
1986	struct ucred *cred = nfsd->nd_cr;
1987	struct nameidata nd;
1988	caddr_t bpos;
1989	int error = 0, rdonly, len, dirfor_ret = 1, diraft_ret = 1;
1990	int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
1991	struct mbuf *mb, *mreq;
1992	struct vnode *vp = NULL, *xp, *dirp = NULL;
1993	struct vattr dirfor, diraft, at;
1994	nfsfh_t nfh, dnfh;
1995	fhandle_t *fhp, *dfhp;
1996	struct mount *mp = NULL;
1997
1998	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1999	ndclear(&nd);
2000
2001	fhp = &nfh.fh_generic;
2002	dfhp = &dnfh.fh_generic;
2003	nfsm_srvmtofh(fhp);
2004	if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2005		error = ESTALE;
2006		goto ereply;
2007	}
2008	(void) vn_start_write(NULL, &mp, V_WAIT);
2009	vfs_rel(mp);		/* The write holds a ref. */
2010	nfsm_srvmtofh(dfhp);
2011	nfsm_srvnamesiz(len);
2012
2013	error = nfsrv_fhtovp(fhp, 0, &vp, nfsd, slp, nam, &rdonly);
2014	if (error) {
2015		nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2016		if (v3) {
2017			nfsm_srvpostop_attr(getret, &at);
2018			nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2019		}
2020		vp = NULL;
2021		error = 0;
2022		goto nfsmout;
2023	}
2024	if (v3)
2025		getret = VOP_GETATTR(vp, &at, cred);
2026	if (vp->v_type == VDIR) {
2027		error = EPERM;		/* POSIX */
2028		goto out1;
2029	}
2030	VOP_UNLOCK(vp, 0);
2031	nd.ni_cnd.cn_cred = cred;
2032	nd.ni_cnd.cn_nameiop = CREATE;
2033	nd.ni_cnd.cn_flags = LOCKPARENT;
2034	error = nfs_namei(&nd, nfsd, dfhp, len, slp, nam, &md, &dpos,
2035		&dirp, v3, &dirfor, &dirfor_ret, FALSE);
2036	if (dirp && !v3) {
2037		vrele(dirp);
2038		dirp = NULL;
2039	}
2040	if (error) {
2041		vrele(vp);
2042		vp = NULL;
2043		goto out2;
2044	}
2045	xp = nd.ni_vp;
2046	if (xp != NULL) {
2047		error = EEXIST;
2048		vrele(vp);
2049		vp = NULL;
2050		goto out2;
2051	}
2052	xp = nd.ni_dvp;
2053	if (vp->v_mount != xp->v_mount) {
2054		error = EXDEV;
2055		vrele(vp);
2056		vp = NULL;
2057		goto out2;
2058	}
2059	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2060	error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2061	NDFREE(&nd, NDF_ONLY_PNBUF);
2062	/* fall through */
2063
2064out1:
2065	if (v3)
2066		getret = VOP_GETATTR(vp, &at, cred);
2067out2:
2068	if (dirp) {
2069		if (dirp == nd.ni_dvp)
2070			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2071		else {
2072			/* Release existing locks to prevent deadlock. */
2073			if (nd.ni_dvp) {
2074				if (nd.ni_dvp == nd.ni_vp)
2075					vrele(nd.ni_dvp);
2076				else
2077					vput(nd.ni_dvp);
2078			}
2079			if (nd.ni_vp)
2080				vrele(nd.ni_vp);
2081			nd.ni_dvp = NULL;
2082			nd.ni_vp = NULL;
2083
2084			vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2085			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2086			VOP_UNLOCK(dirp, 0);
2087		}
2088	}
2089ereply:
2090	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2091	if (v3) {
2092		nfsm_srvpostop_attr(getret, &at);
2093		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2094	}
2095	error = 0;
2096	/* fall through */
2097
2098nfsmout:
2099	NDFREE(&nd, NDF_ONLY_PNBUF);
2100	if (vp)
2101		vput(vp);
2102	if (nd.ni_dvp) {
2103		if (nd.ni_dvp == nd.ni_vp)
2104			vrele(nd.ni_dvp);
2105		else
2106			vput(nd.ni_dvp);
2107	}
2108	if (dirp)
2109		vrele(dirp);
2110	if (nd.ni_vp)
2111		vrele(nd.ni_vp);
2112	vn_finished_write(mp);
2113	return(error);
2114}
2115
2116/*
2117 * nfs symbolic link service
2118 */
2119int
2120nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2121    struct mbuf **mrq)
2122{
2123	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2124	struct sockaddr *nam = nfsd->nd_nam;
2125	caddr_t dpos = nfsd->nd_dpos;
2126	struct ucred *cred = nfsd->nd_cr;
2127	struct vattr va, dirfor, diraft;
2128	struct nameidata nd;
2129	struct vattr *vap = &va;
2130	struct nfsv2_sattr *sp;
2131	char *bpos, *pathcp = NULL;
2132	struct uio io;
2133	struct iovec iv;
2134	int error = 0, len, len2, dirfor_ret = 1, diraft_ret = 1;
2135	int v3 = (nfsd->nd_flag & ND_NFSV3);
2136	struct mbuf *mb, *mreq;
2137	struct vnode *dirp = NULL;
2138	nfsfh_t nfh;
2139	fhandle_t *fhp;
2140	struct mount *mp = NULL;
2141
2142	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2143	ndclear(&nd);
2144
2145	fhp = &nfh.fh_generic;
2146	nfsm_srvmtofh(fhp);
2147	if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2148		error = ESTALE;
2149		goto out;
2150	}
2151	(void) vn_start_write(NULL, &mp, V_WAIT);
2152	vfs_rel(mp);		/* The write holds a ref. */
2153	nfsm_srvnamesiz(len);
2154	nd.ni_cnd.cn_cred = cred;
2155	nd.ni_cnd.cn_nameiop = CREATE;
2156	nd.ni_cnd.cn_flags = LOCKPARENT | SAVESTART;
2157	error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
2158		&dirp, v3, &dirfor, &dirfor_ret, FALSE);
2159	if (error == 0) {
2160		VATTR_NULL(vap);
2161		if (v3)
2162			nfsm_srvsattr(vap);
2163		nfsm_srvpathsiz(len2);
2164	}
2165	if (dirp && !v3) {
2166		vrele(dirp);
2167		dirp = NULL;
2168	}
2169	if (error)
2170		goto out;
2171	pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK);
2172	iv.iov_base = pathcp;
2173	iv.iov_len = len2;
2174	io.uio_resid = len2;
2175	io.uio_offset = 0;
2176	io.uio_iov = &iv;
2177	io.uio_iovcnt = 1;
2178	io.uio_segflg = UIO_SYSSPACE;
2179	io.uio_rw = UIO_READ;
2180	io.uio_td = NULL;
2181	nfsm_mtouio(&io, len2);
2182	if (!v3) {
2183		sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
2184		vap->va_mode = nfstov_mode(sp->sa_mode);
2185	}
2186	*(pathcp + len2) = '\0';
2187	if (nd.ni_vp) {
2188		error = EEXIST;
2189		goto out;
2190	}
2191
2192	/*
2193	 * issue symlink op.  SAVESTART is set so the underlying path component
2194	 * is only freed by the VOP if an error occurs.
2195	 */
2196	if (vap->va_mode == (mode_t)VNOVAL)
2197		vap->va_mode = 0;
2198	error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap, pathcp);
2199	if (error)
2200		NDFREE(&nd, NDF_ONLY_PNBUF);
2201	else
2202		vput(nd.ni_vp);
2203	nd.ni_vp = NULL;
2204	/*
2205	 * releases directory prior to potential lookup op.
2206	 */
2207	vput(nd.ni_dvp);
2208	nd.ni_dvp = NULL;
2209
2210	if (error == 0) {
2211	    if (v3) {
2212		/*
2213		 * Issue lookup.  Leave SAVESTART set so we can easily free
2214		 * the name buffer later on.
2215		 *
2216		 * since LOCKPARENT is not set, ni_dvp will be garbage on
2217		 * return whether an error occurs or not.
2218		 */
2219		nd.ni_cnd.cn_nameiop = LOOKUP;
2220		nd.ni_cnd.cn_flags &= ~(LOCKPARENT | FOLLOW);
2221		nd.ni_cnd.cn_flags |= (NOFOLLOW | LOCKLEAF);
2222		nd.ni_cnd.cn_thread = curthread;
2223		nd.ni_cnd.cn_cred = cred;
2224		error = lookup(&nd);
2225		nd.ni_dvp = NULL;
2226
2227		if (error == 0) {
2228			bzero((caddr_t)fhp, sizeof(nfh));
2229			fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2230			error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2231			if (!error)
2232				error = VOP_GETATTR(nd.ni_vp, vap, cred);
2233			vput(nd.ni_vp);
2234			nd.ni_vp = NULL;
2235		}
2236	    }
2237	}
2238out:
2239	/*
2240	 * These releases aren't strictly required, does even doing them
2241	 * make any sense? XXX can nfsm_reply() block?
2242	 */
2243	if (pathcp) {
2244		free(pathcp, M_TEMP);
2245		pathcp = NULL;
2246	}
2247	if (dirp) {
2248		vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2249		diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2250		VOP_UNLOCK(dirp, 0);
2251	}
2252	if (nd.ni_startdir) {
2253		vrele(nd.ni_startdir);
2254		nd.ni_startdir = NULL;
2255	}
2256	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2257	if (v3) {
2258		if (!error) {
2259			nfsm_srvpostop_fh(fhp);
2260			nfsm_srvpostop_attr(0, vap);
2261		}
2262		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2263	}
2264	error = 0;
2265	/* fall through */
2266
2267nfsmout:
2268	NDFREE(&nd, NDF_ONLY_PNBUF);
2269	if (nd.ni_dvp) {
2270		if (nd.ni_dvp == nd.ni_vp)
2271			vrele(nd.ni_dvp);
2272		else
2273			vput(nd.ni_dvp);
2274	}
2275	if (nd.ni_vp)
2276		vrele(nd.ni_vp);
2277	if (nd.ni_startdir)
2278		vrele(nd.ni_startdir);
2279	if (dirp)
2280		vrele(dirp);
2281	if (pathcp)
2282		free(pathcp, M_TEMP);
2283
2284	vn_finished_write(mp);
2285	return (error);
2286}
2287
2288/*
2289 * nfs mkdir service
2290 */
2291int
2292nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2293    struct mbuf **mrq)
2294{
2295	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2296	struct sockaddr *nam = nfsd->nd_nam;
2297	caddr_t dpos = nfsd->nd_dpos;
2298	struct ucred *cred = nfsd->nd_cr;
2299	struct vattr va, dirfor, diraft;
2300	struct vattr *vap = &va;
2301	struct nfs_fattr *fp;
2302	struct nameidata nd;
2303	u_int32_t *tl;
2304	caddr_t bpos;
2305	int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2306	int v3 = (nfsd->nd_flag & ND_NFSV3);
2307	struct mbuf *mb, *mreq;
2308	struct vnode *dirp = NULL;
2309	int vpexcl = 0;
2310	nfsfh_t nfh;
2311	fhandle_t *fhp;
2312	struct mount *mp = NULL;
2313
2314	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2315	ndclear(&nd);
2316
2317	fhp = &nfh.fh_generic;
2318	nfsm_srvmtofh(fhp);
2319	if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2320		error = ESTALE;
2321		goto out;
2322	}
2323	(void) vn_start_write(NULL, &mp, V_WAIT);
2324	vfs_rel(mp);		/* The write holds a ref. */
2325	nfsm_srvnamesiz(len);
2326	nd.ni_cnd.cn_cred = cred;
2327	nd.ni_cnd.cn_nameiop = CREATE;
2328	nd.ni_cnd.cn_flags = LOCKPARENT;
2329
2330	error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
2331		&dirp, v3, &dirfor, &dirfor_ret, FALSE);
2332	if (dirp && !v3) {
2333		vrele(dirp);
2334		dirp = NULL;
2335	}
2336	if (error) {
2337		nfsm_reply(NFSX_WCCDATA(v3));
2338		if (v3)
2339			nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2340		error = 0;
2341		goto nfsmout;
2342	}
2343	VATTR_NULL(vap);
2344	if (v3) {
2345		nfsm_srvsattr(vap);
2346	} else {
2347		tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
2348		vap->va_mode = nfstov_mode(*tl++);
2349	}
2350
2351	/*
2352	 * At this point nd.ni_dvp is referenced and exclusively locked and
2353	 * nd.ni_vp, if it exists, is referenced but not locked.
2354	 */
2355
2356	vap->va_type = VDIR;
2357	if (nd.ni_vp != NULL) {
2358		NDFREE(&nd, NDF_ONLY_PNBUF);
2359		error = EEXIST;
2360		goto out;
2361	}
2362
2363	/*
2364	 * Issue mkdir op.  Since SAVESTART is not set, the pathname
2365	 * component is freed by the VOP call.  This will fill-in
2366	 * nd.ni_vp, reference, and exclusively lock it.
2367	 */
2368	if (vap->va_mode == (mode_t)VNOVAL)
2369		vap->va_mode = 0;
2370	error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2371	NDFREE(&nd, NDF_ONLY_PNBUF);
2372	vpexcl = 1;
2373
2374	vput(nd.ni_dvp);
2375	nd.ni_dvp = NULL;
2376
2377	if (!error) {
2378		bzero((caddr_t)fhp, sizeof(nfh));
2379		fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2380		error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2381		if (!error)
2382			error = VOP_GETATTR(nd.ni_vp, vap, cred);
2383	}
2384out:
2385	if (dirp) {
2386		if (dirp == nd.ni_dvp) {
2387			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2388		} else {
2389			/* Release existing locks to prevent deadlock. */
2390			if (nd.ni_dvp) {
2391				NDFREE(&nd, NDF_ONLY_PNBUF);
2392				if (nd.ni_dvp == nd.ni_vp && vpexcl)
2393					vrele(nd.ni_dvp);
2394				else
2395					vput(nd.ni_dvp);
2396			}
2397			if (nd.ni_vp) {
2398				if (vpexcl)
2399					vput(nd.ni_vp);
2400				else
2401					vrele(nd.ni_vp);
2402			}
2403			nd.ni_dvp = NULL;
2404			nd.ni_vp = NULL;
2405			vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2406			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2407			VOP_UNLOCK(dirp, 0);
2408		}
2409	}
2410	nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2411	if (v3) {
2412		if (!error) {
2413			nfsm_srvpostop_fh(fhp);
2414			nfsm_srvpostop_attr(0, vap);
2415		}
2416		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2417	} else if (!error) {
2418		/* v2 non-error case. */
2419		nfsm_srvfhtom(fhp, v3);
2420		fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
2421		nfsm_srvfillattr(vap, fp);
2422	}
2423	error = 0;
2424	/* fall through */
2425
2426nfsmout:
2427	if (nd.ni_dvp) {
2428		NDFREE(&nd, NDF_ONLY_PNBUF);
2429		if (nd.ni_dvp == nd.ni_vp && vpexcl)
2430			vrele(nd.ni_dvp);
2431		else
2432			vput(nd.ni_dvp);
2433	}
2434	if (nd.ni_vp) {
2435		if (vpexcl)
2436			vput(nd.ni_vp);
2437		else
2438			vrele(nd.ni_vp);
2439	}
2440	if (dirp)
2441		vrele(dirp);
2442	vn_finished_write(mp);
2443	return (error);
2444}
2445
2446/*
2447 * nfs rmdir service
2448 */
2449int
2450nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2451    struct mbuf **mrq)
2452{
2453	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2454	struct sockaddr *nam = nfsd->nd_nam;
2455	caddr_t dpos = nfsd->nd_dpos;
2456	struct ucred *cred = nfsd->nd_cr;
2457	caddr_t bpos;
2458	int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2459	int v3 = (nfsd->nd_flag & ND_NFSV3);
2460	struct mbuf *mb, *mreq;
2461	struct vnode *vp, *dirp = NULL;
2462	struct vattr dirfor, diraft;
2463	nfsfh_t nfh;
2464	fhandle_t *fhp;
2465	struct nameidata nd;
2466	struct mount *mp = NULL;
2467
2468	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2469	ndclear(&nd);
2470
2471	fhp = &nfh.fh_generic;
2472	nfsm_srvmtofh(fhp);
2473	if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2474		error = ESTALE;
2475		goto out;
2476	}
2477	(void) vn_start_write(NULL, &mp, V_WAIT);
2478	vfs_rel(mp);		/* The write holds a ref. */
2479	nfsm_srvnamesiz(len);
2480	nd.ni_cnd.cn_cred = cred;
2481	nd.ni_cnd.cn_nameiop = DELETE;
2482	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
2483	error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
2484		&dirp, v3, &dirfor, &dirfor_ret, FALSE);
2485	if (dirp && !v3) {
2486		vrele(dirp);
2487		dirp = NULL;
2488	}
2489	if (error) {
2490		nfsm_reply(NFSX_WCCDATA(v3));
2491		if (v3)
2492			nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2493		error = 0;
2494		goto nfsmout;
2495	}
2496	vp = nd.ni_vp;
2497	if (vp->v_type != VDIR) {
2498		error = ENOTDIR;
2499		goto out;
2500	}
2501	/*
2502	 * No rmdir "." please.
2503	 */
2504	if (nd.ni_dvp == vp) {
2505		error = EINVAL;
2506		goto out;
2507	}
2508	/*
2509	 * The root of a mounted filesystem cannot be deleted.
2510	 */
2511	if (vp->v_vflag & VV_ROOT)
2512		error = EBUSY;
2513out:
2514	/*
2515	 * Issue or abort op.  Since SAVESTART is not set, path name
2516	 * component is freed by the VOP after either.
2517	 */
2518	if (!error)
2519		error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2520	NDFREE(&nd, NDF_ONLY_PNBUF);
2521
2522	if (dirp) {
2523		if (dirp == nd.ni_dvp)
2524			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2525		else {
2526			/* Release existing locks to prevent deadlock. */
2527			if (nd.ni_dvp) {
2528				if (nd.ni_dvp == nd.ni_vp)
2529					vrele(nd.ni_dvp);
2530				else
2531					vput(nd.ni_dvp);
2532			}
2533			if (nd.ni_vp)
2534				vput(nd.ni_vp);
2535			nd.ni_dvp = NULL;
2536			nd.ni_vp = NULL;
2537			vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2538			diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2539			VOP_UNLOCK(dirp, 0);
2540		}
2541	}
2542	nfsm_reply(NFSX_WCCDATA(v3));
2543	error = 0;
2544	if (v3)
2545		nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2546	/* fall through */
2547
2548nfsmout:
2549	NDFREE(&nd, NDF_ONLY_PNBUF);
2550	if (nd.ni_dvp) {
2551		if (nd.ni_dvp == nd.ni_vp)
2552			vrele(nd.ni_dvp);
2553		else
2554			vput(nd.ni_dvp);
2555	}
2556	if (nd.ni_vp)
2557		vput(nd.ni_vp);
2558	if (dirp)
2559		vrele(dirp);
2560
2561	vn_finished_write(mp);
2562	return(error);
2563}
2564
2565/*
2566 * nfs readdir service
2567 * - mallocs what it thinks is enough to read
2568 *	count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
2569 * - calls VOP_READDIR()
2570 * - loops around building the reply
2571 *	if the output generated exceeds count break out of loop
2572 *	The nfsm_clget macro is used here so that the reply will be packed
2573 *	tightly in mbuf clusters.
2574 * - it only knows that it has encountered eof when the VOP_READDIR()
2575 *	reads nothing
2576 * - as such one readdir rpc will return eof false although you are there
2577 *	and then the next will return eof
2578 * - it trims out records with d_fileno == 0
2579 *	this doesn't matter for Unix clients, but they might confuse clients
2580 *	for other os'.
2581 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2582 *	than requested, but this may not apply to all filesystems. For
2583 *	example, client NFS does not { although it is never remote mounted
2584 *	anyhow }
2585 *     The alternate call nfsrv_readdirplus() does lookups as well.
2586 * PS: The NFS protocol spec. does not clarify what the "count" byte
2587 *	argument is a count of.. just name strings and file id's or the
2588 *	entire reply rpc or ...
2589 *	I tried just file name and id sizes and it confused the Sun client,
2590 *	so I am using the full rpc size now. The "paranoia.." comment refers
2591 *	to including the status longwords that are not a part of the dir.
2592 *	"entry" structures, but are in the rpc.
2593 */
2594struct flrep {
2595	nfsuint64	fl_off;
2596	u_int32_t	fl_postopok;
2597	u_int32_t	fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2598	u_int32_t	fl_fhok;
2599	u_int32_t	fl_fhsize;
2600	u_int32_t	fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
2601};
2602
2603int
2604nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2605    struct mbuf **mrq)
2606{
2607	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2608	struct sockaddr *nam = nfsd->nd_nam;
2609	caddr_t dpos = nfsd->nd_dpos;
2610	struct ucred *cred = nfsd->nd_cr;
2611	char *bp, *be;
2612	struct mbuf *mp;
2613	struct dirent *dp;
2614	caddr_t cp;
2615	u_int32_t *tl;
2616	caddr_t bpos;
2617	struct mbuf *mb, *mreq;
2618	char *cpos, *cend, *rbuf;
2619	struct vnode *vp = NULL;
2620	struct vattr at;
2621	nfsfh_t nfh;
2622	fhandle_t *fhp;
2623	struct uio io;
2624	struct iovec iv;
2625	int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2626	int siz, cnt, fullsiz, eofflag, rdonly, ncookies;
2627	int v3 = (nfsd->nd_flag & ND_NFSV3);
2628	u_quad_t off, toff, verf;
2629	u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
2630	int is_ufs;
2631
2632	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2633	fhp = &nfh.fh_generic;
2634	nfsm_srvmtofh(fhp);
2635	if (v3) {
2636		tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
2637		toff = fxdr_hyper(tl);
2638		tl += 2;
2639		verf = fxdr_hyper(tl);
2640		tl += 2;
2641	} else {
2642		tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
2643		toff = fxdr_unsigned(u_quad_t, *tl++);
2644		verf = 0;	/* shut up gcc */
2645	}
2646	off = toff;
2647	cnt = fxdr_unsigned(int, *tl);
2648	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2649	xfer = NFS_SRVMAXDATA(nfsd);
2650	if (cnt > xfer)
2651		cnt = xfer;
2652	if (siz > xfer)
2653		siz = xfer;
2654	fullsiz = siz;
2655	error = nfsrv_fhtovp(fhp, 0, &vp, nfsd, slp, nam, &rdonly);
2656	if (!error && vp->v_type != VDIR) {
2657		error = ENOTDIR;
2658		vput(vp);
2659		vp = NULL;
2660	}
2661	if (error) {
2662		nfsm_reply(NFSX_UNSIGNED);
2663		if (v3)
2664			nfsm_srvpostop_attr(getret, &at);
2665		error = 0;
2666		goto nfsmout;
2667	}
2668
2669	/*
2670	 * Obtain lock on vnode for this section of the code
2671	 */
2672	if (v3) {
2673		error = getret = VOP_GETATTR(vp, &at, cred);
2674#if 0
2675		/*
2676		 * XXX This check may be too strict for Solaris 2.5 clients.
2677		 */
2678		if (!error && toff && verf && verf != at.va_filerev)
2679			error = NFSERR_BAD_COOKIE;
2680#endif
2681	}
2682	if (!error)
2683		error = nfsrv_access(vp, VEXEC, cred, rdonly, 0);
2684	if (error) {
2685		vput(vp);
2686		vp = NULL;
2687		nfsm_reply(NFSX_POSTOPATTR(v3));
2688		if (v3)
2689			nfsm_srvpostop_attr(getret, &at);
2690		error = 0;
2691		goto nfsmout;
2692	}
2693	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2694	VOP_UNLOCK(vp, 0);
2695
2696	/*
2697	 * end section.  Allocate rbuf and continue
2698	 */
2699	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2700again:
2701	iv.iov_base = rbuf;
2702	iv.iov_len = fullsiz;
2703	io.uio_iov = &iv;
2704	io.uio_iovcnt = 1;
2705	io.uio_offset = (off_t)off;
2706	io.uio_resid = fullsiz;
2707	io.uio_segflg = UIO_SYSSPACE;
2708	io.uio_rw = UIO_READ;
2709	io.uio_td = NULL;
2710	eofflag = 0;
2711	if (cookies) {
2712		free((caddr_t)cookies, M_TEMP);
2713		cookies = NULL;
2714	}
2715	vn_lock(vp, LK_SHARED | LK_RETRY);
2716	error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
2717	off = (off_t)io.uio_offset;
2718	if (!cookies && !error)
2719		error = NFSERR_PERM;
2720	if (v3) {
2721		getret = VOP_GETATTR(vp, &at, cred);
2722		if (!error)
2723			error = getret;
2724	}
2725	VOP_UNLOCK(vp, 0);
2726	if (error) {
2727		vrele(vp);
2728		vp = NULL;
2729		free((caddr_t)rbuf, M_TEMP);
2730		if (cookies)
2731			free((caddr_t)cookies, M_TEMP);
2732		nfsm_reply(NFSX_POSTOPATTR(v3));
2733		if (v3)
2734			nfsm_srvpostop_attr(getret, &at);
2735		error = 0;
2736		goto nfsmout;
2737	}
2738	if (io.uio_resid) {
2739		siz -= io.uio_resid;
2740
2741		/*
2742		 * If nothing read, return eof
2743		 * rpc reply
2744		 */
2745		if (siz == 0) {
2746			vrele(vp);
2747			vp = NULL;
2748			nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2749				2 * NFSX_UNSIGNED);
2750			if (v3) {
2751				nfsm_srvpostop_attr(getret, &at);
2752				tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
2753				txdr_hyper(at.va_filerev, tl);
2754				tl += 2;
2755			} else
2756				tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2757			*tl++ = nfsrv_nfs_false;
2758			*tl = nfsrv_nfs_true;
2759			free((caddr_t)rbuf, M_TEMP);
2760			free((caddr_t)cookies, M_TEMP);
2761			error = 0;
2762			goto nfsmout;
2763		}
2764	}
2765
2766	/*
2767	 * Check for degenerate cases of nothing useful read.
2768	 * If so go try again
2769	 */
2770	cpos = rbuf;
2771	cend = rbuf + siz;
2772	dp = (struct dirent *)cpos;
2773	cookiep = cookies;
2774	/*
2775	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2776	 * directory offset up to a block boundary, so it is necessary to
2777	 * skip over the records that precede the requested offset. This
2778	 * requires the assumption that file offset cookies monotonically
2779	 * increase.
2780	 */
2781	while (cpos < cend && ncookies > 0 &&
2782		(dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2783		 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
2784		cpos += dp->d_reclen;
2785		dp = (struct dirent *)cpos;
2786		cookiep++;
2787		ncookies--;
2788	}
2789	if (cpos >= cend || ncookies == 0) {
2790		toff = off;
2791		siz = fullsiz;
2792		goto again;
2793	}
2794
2795	len = 3 * NFSX_UNSIGNED;	/* paranoia, probably can be 0 */
2796	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2797	if (v3) {
2798		nfsm_srvpostop_attr(getret, &at);
2799		tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2800		txdr_hyper(at.va_filerev, tl);
2801	}
2802	mp = mb;
2803	bp = bpos;
2804	be = bp + M_TRAILINGSPACE(mp);
2805
2806	/* Loop through the records and build reply */
2807	while (cpos < cend && ncookies > 0) {
2808		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2809			nlen = dp->d_namlen;
2810			rem = nfsm_rndup(nlen) - nlen;
2811			len += (4 * NFSX_UNSIGNED + nlen + rem);
2812			if (v3)
2813				len += 2 * NFSX_UNSIGNED;
2814			if (len > cnt) {
2815				eofflag = 0;
2816				break;
2817			}
2818			/*
2819			 * Build the directory record xdr from
2820			 * the dirent entry.
2821			 */
2822			nfsm_clget;
2823			*tl = nfsrv_nfs_true;
2824			bp += NFSX_UNSIGNED;
2825			if (v3) {
2826				nfsm_clget;
2827				*tl = 0;
2828				bp += NFSX_UNSIGNED;
2829			}
2830			nfsm_clget;
2831			*tl = txdr_unsigned(dp->d_fileno);
2832			bp += NFSX_UNSIGNED;
2833			nfsm_clget;
2834			*tl = txdr_unsigned(nlen);
2835			bp += NFSX_UNSIGNED;
2836
2837			/* And loop around copying the name */
2838			xfer = nlen;
2839			cp = dp->d_name;
2840			while (xfer > 0) {
2841				nfsm_clget;
2842				if ((bp+xfer) > be)
2843					tsiz = be-bp;
2844				else
2845					tsiz = xfer;
2846				bcopy(cp, bp, tsiz);
2847				bp += tsiz;
2848				xfer -= tsiz;
2849				if (xfer > 0)
2850					cp += tsiz;
2851			}
2852			/* And null pad to an int32_t boundary. */
2853			for (i = 0; i < rem; i++)
2854				*bp++ = '\0';
2855			nfsm_clget;
2856
2857			/* Finish off the record */
2858			if (v3) {
2859				*tl = 0;
2860				bp += NFSX_UNSIGNED;
2861				nfsm_clget;
2862			}
2863			*tl = txdr_unsigned(*cookiep);
2864			bp += NFSX_UNSIGNED;
2865		}
2866		cpos += dp->d_reclen;
2867		dp = (struct dirent *)cpos;
2868		cookiep++;
2869		ncookies--;
2870	}
2871	vrele(vp);
2872	vp = NULL;
2873	nfsm_clget;
2874	*tl = nfsrv_nfs_false;
2875	bp += NFSX_UNSIGNED;
2876	nfsm_clget;
2877	if (eofflag)
2878		*tl = nfsrv_nfs_true;
2879	else
2880		*tl = nfsrv_nfs_false;
2881	bp += NFSX_UNSIGNED;
2882	if (mp != mb) {
2883		if (bp < be)
2884			mp->m_len = bp - mtod(mp, caddr_t);
2885	} else
2886		mp->m_len += bp - bpos;
2887	free((caddr_t)rbuf, M_TEMP);
2888	free((caddr_t)cookies, M_TEMP);
2889
2890nfsmout:
2891	if (vp)
2892		vrele(vp);
2893	return(error);
2894}
2895
2896int
2897nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2898    struct mbuf **mrq)
2899{
2900	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2901	struct sockaddr *nam = nfsd->nd_nam;
2902	caddr_t dpos = nfsd->nd_dpos;
2903	struct ucred *cred = nfsd->nd_cr;
2904	char *bp, *be;
2905	struct mbuf *mp;
2906	struct dirent *dp;
2907	caddr_t cp;
2908	u_int32_t *tl;
2909	caddr_t bpos;
2910	struct mbuf *mb, *mreq;
2911	char *cpos, *cend, *rbuf;
2912	struct vnode *vp = NULL, *nvp;
2913	struct flrep fl;
2914	nfsfh_t nfh;
2915	fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
2916	struct uio io;
2917	struct iovec iv;
2918	struct vattr va, at, *vap = &va;
2919	struct nfs_fattr *fp;
2920	int len, nlen, rem, xfer, tsiz, i, error = 0, error1, getret = 1;
2921	int vp_locked;
2922	int siz, cnt, fullsiz, eofflag, rdonly, dirlen, ncookies;
2923	u_quad_t off, toff, verf;
2924	u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
2925	int v3 = (nfsd->nd_flag & ND_NFSV3);
2926	int usevget = 1;
2927	struct componentname cn;
2928	struct mount *mntp = NULL;
2929	int is_ufs;
2930
2931	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2932	vp_locked = 0;
2933	if (!v3)
2934		panic("nfsrv_readdirplus: v3 proc called on a v2 connection");
2935	fhp = &nfh.fh_generic;
2936	nfsm_srvmtofh(fhp);
2937	tl = nfsm_dissect_nonblock(u_int32_t *, 6 * NFSX_UNSIGNED);
2938	toff = fxdr_hyper(tl);
2939	tl += 2;
2940	verf = fxdr_hyper(tl);
2941	tl += 2;
2942	siz = fxdr_unsigned(int, *tl++);
2943	cnt = fxdr_unsigned(int, *tl);
2944	off = toff;
2945	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2946	xfer = NFS_SRVMAXDATA(nfsd);
2947	if (cnt > xfer)
2948		cnt = xfer;
2949	if (siz > xfer)
2950		siz = xfer;
2951	fullsiz = siz;
2952	error = nfsrv_fhtovp(fhp, NFSRV_FLAG_BUSY, &vp, nfsd, slp,
2953	    nam, &rdonly);
2954	if (!error) {
2955		vp_locked = 1;
2956		mntp = vp->v_mount;
2957		if (vp->v_type != VDIR) {
2958			error = ENOTDIR;
2959			vput(vp);
2960			vp = NULL;
2961			vp_locked = 0;
2962		}
2963	}
2964	if (error) {
2965		nfsm_reply(NFSX_UNSIGNED);
2966		nfsm_srvpostop_attr(getret, &at);
2967		error = 0;
2968		goto nfsmout;
2969	}
2970	error = getret = VOP_GETATTR(vp, &at, cred);
2971#if 0
2972	/*
2973	 * XXX This check may be too strict for Solaris 2.5 clients.
2974	 */
2975	if (!error && toff && verf && verf != at.va_filerev)
2976		error = NFSERR_BAD_COOKIE;
2977#endif
2978	if (!error)
2979		error = nfsrv_access(vp, VEXEC, cred, rdonly, 0);
2980	if (error) {
2981		vput(vp);
2982		vp_locked = 0;
2983		vp = NULL;
2984		nfsm_reply(NFSX_V3POSTOPATTR);
2985		nfsm_srvpostop_attr(getret, &at);
2986		error = 0;
2987		goto nfsmout;
2988	}
2989	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2990	VOP_UNLOCK(vp, 0);
2991	vp_locked = 0;
2992	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2993again:
2994	iv.iov_base = rbuf;
2995	iv.iov_len = fullsiz;
2996	io.uio_iov = &iv;
2997	io.uio_iovcnt = 1;
2998	io.uio_offset = (off_t)off;
2999	io.uio_resid = fullsiz;
3000	io.uio_segflg = UIO_SYSSPACE;
3001	io.uio_rw = UIO_READ;
3002	io.uio_td = NULL;
3003	eofflag = 0;
3004	vp_locked = 1;
3005	if (cookies) {
3006		free((caddr_t)cookies, M_TEMP);
3007		cookies = NULL;
3008	}
3009	vn_lock(vp, LK_SHARED | LK_RETRY);
3010	error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3011	off = (u_quad_t)io.uio_offset;
3012	getret = VOP_GETATTR(vp, &at, cred);
3013	VOP_UNLOCK(vp, 0);
3014	vp_locked = 0;
3015	if (!cookies && !error)
3016		error = NFSERR_PERM;
3017	if (!error)
3018		error = getret;
3019	if (error) {
3020		vrele(vp);
3021		vp = NULL;
3022		if (cookies)
3023			free((caddr_t)cookies, M_TEMP);
3024		free((caddr_t)rbuf, M_TEMP);
3025		nfsm_reply(NFSX_V3POSTOPATTR);
3026		nfsm_srvpostop_attr(getret, &at);
3027		error = 0;
3028		goto nfsmout;
3029	}
3030	if (io.uio_resid) {
3031		siz -= io.uio_resid;
3032
3033		/*
3034		 * If nothing read, return eof
3035		 * rpc reply
3036		 */
3037		if (siz == 0) {
3038			vrele(vp);
3039			vp = NULL;
3040			nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3041				2 * NFSX_UNSIGNED);
3042			nfsm_srvpostop_attr(getret, &at);
3043			tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
3044			txdr_hyper(at.va_filerev, tl);
3045			tl += 2;
3046			*tl++ = nfsrv_nfs_false;
3047			*tl = nfsrv_nfs_true;
3048			free((caddr_t)cookies, M_TEMP);
3049			free((caddr_t)rbuf, M_TEMP);
3050			error = 0;
3051			goto nfsmout;
3052		}
3053	}
3054
3055	/*
3056	 * Check for degenerate cases of nothing useful read.
3057	 * If so go try again
3058	 */
3059	cpos = rbuf;
3060	cend = rbuf + siz;
3061	dp = (struct dirent *)cpos;
3062	cookiep = cookies;
3063	/*
3064	 * For some reason FreeBSD's ufs_readdir() chooses to back the
3065	 * directory offset up to a block boundary, so it is necessary to
3066	 * skip over the records that precede the requested offset. This
3067	 * requires the assumption that file offset cookies monotonically
3068	 * increase.
3069	 */
3070	while (cpos < cend && ncookies > 0 &&
3071		(dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3072		 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
3073		cpos += dp->d_reclen;
3074		dp = (struct dirent *)cpos;
3075		cookiep++;
3076		ncookies--;
3077	}
3078	if (cpos >= cend || ncookies == 0) {
3079		toff = off;
3080		siz = fullsiz;
3081		goto again;
3082	}
3083
3084	dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3085	    2 * NFSX_UNSIGNED;
3086	nfsm_reply(cnt);
3087	nfsm_srvpostop_attr(getret, &at);
3088	tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3089	txdr_hyper(at.va_filerev, tl);
3090	mp = mb;
3091	bp = bpos;
3092	be = bp + M_TRAILINGSPACE(mp);
3093
3094	/* Loop through the records and build reply */
3095	while (cpos < cend && ncookies > 0) {
3096		if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3097			nlen = dp->d_namlen;
3098			rem = nfsm_rndup(nlen)-nlen;
3099
3100			if (usevget) {
3101				/*
3102				 * For readdir_and_lookup get the vnode using
3103				 * the file number.
3104				 */
3105				error = VFS_VGET(mntp, dp->d_fileno, LK_SHARED,
3106				    &nvp);
3107				if (error != 0 && error != EOPNOTSUPP) {
3108					error = 0;
3109					goto invalid;
3110				} else if (error == EOPNOTSUPP) {
3111					/*
3112					 * VFS_VGET() not supported?
3113					 * Let's switch to VOP_LOOKUP().
3114					 */
3115					error = 0;
3116					usevget = 0;
3117					cn.cn_nameiop = LOOKUP;
3118					cn.cn_flags = ISLASTCN | NOFOLLOW | \
3119					    LOCKSHARED | LOCKLEAF;
3120					cn.cn_lkflags = LK_SHARED | LK_RETRY;
3121					cn.cn_cred = cred;
3122					cn.cn_thread = curthread;
3123				}
3124			}
3125			if (!usevget) {
3126				cn.cn_nameptr = dp->d_name;
3127				cn.cn_namelen = dp->d_namlen;
3128				if (dp->d_namlen == 2 &&
3129				    dp->d_name[0] == '.' &&
3130				    dp->d_name[1] == '.') {
3131					cn.cn_flags |= ISDOTDOT;
3132				} else {
3133					cn.cn_flags &= ~ISDOTDOT;
3134				}
3135				if (!vp_locked) {
3136					vn_lock(vp, LK_SHARED | LK_RETRY);
3137					vp_locked = 1;
3138				}
3139				if ((vp->v_vflag & VV_ROOT) != 0 &&
3140				    (cn.cn_flags & ISDOTDOT) != 0) {
3141					vref(vp);
3142					nvp = vp;
3143				} else if (VOP_LOOKUP(vp, &nvp, &cn) != 0)
3144					goto invalid;
3145			}
3146
3147			bzero((caddr_t)nfhp, NFSX_V3FH);
3148			nfhp->fh_fsid = nvp->v_mount->mnt_stat.f_fsid;
3149			if ((error1 = VOP_VPTOFH(nvp, &nfhp->fh_fid)) == 0)
3150				error1 = VOP_GETATTR(nvp, vap, cred);
3151			if (!usevget && vp == nvp)
3152				vunref(nvp);
3153			else
3154				vput(nvp);
3155			nvp = NULL;
3156			if (error1 != 0)
3157				goto invalid;
3158
3159			/*
3160			 * If either the dircount or maxcount will be
3161			 * exceeded, get out now. Both of these lengths
3162			 * are calculated conservatively, including all
3163			 * XDR overheads.
3164			 */
3165			len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3166				NFSX_V3POSTOPATTR);
3167			dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3168			if (len > cnt || dirlen > fullsiz) {
3169				eofflag = 0;
3170				break;
3171			}
3172
3173			/*
3174			 * Build the directory record xdr from
3175			 * the dirent entry.
3176			 */
3177			fp = (struct nfs_fattr *)&fl.fl_fattr;
3178			nfsm_srvfillattr(vap, fp);
3179			fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3180			fl.fl_fhok = nfsrv_nfs_true;
3181			fl.fl_postopok = nfsrv_nfs_true;
3182			fl.fl_off.nfsuquad[0] = 0;
3183			fl.fl_off.nfsuquad[1] = txdr_unsigned(*cookiep);
3184
3185			nfsm_clget;
3186			*tl = nfsrv_nfs_true;
3187			bp += NFSX_UNSIGNED;
3188			nfsm_clget;
3189			*tl = 0;
3190			bp += NFSX_UNSIGNED;
3191			nfsm_clget;
3192			*tl = txdr_unsigned(dp->d_fileno);
3193			bp += NFSX_UNSIGNED;
3194			nfsm_clget;
3195			*tl = txdr_unsigned(nlen);
3196			bp += NFSX_UNSIGNED;
3197
3198			/* And loop around copying the name */
3199			xfer = nlen;
3200			cp = dp->d_name;
3201			while (xfer > 0) {
3202				nfsm_clget;
3203				if ((bp + xfer) > be)
3204					tsiz = be - bp;
3205				else
3206					tsiz = xfer;
3207				bcopy(cp, bp, tsiz);
3208				bp += tsiz;
3209				xfer -= tsiz;
3210				if (xfer > 0)
3211					cp += tsiz;
3212			}
3213			/* And null pad to an int32_t boundary. */
3214			for (i = 0; i < rem; i++)
3215				*bp++ = '\0';
3216
3217			/*
3218			 * Now copy the flrep structure out.
3219			 */
3220			xfer = sizeof (struct flrep);
3221			cp = (caddr_t)&fl;
3222			while (xfer > 0) {
3223				nfsm_clget;
3224				if ((bp + xfer) > be)
3225					tsiz = be - bp;
3226				else
3227					tsiz = xfer;
3228				bcopy(cp, bp, tsiz);
3229				bp += tsiz;
3230				xfer -= tsiz;
3231				if (xfer > 0)
3232					cp += tsiz;
3233			}
3234		}
3235invalid:
3236		cpos += dp->d_reclen;
3237		dp = (struct dirent *)cpos;
3238		cookiep++;
3239		ncookies--;
3240	}
3241	if (!usevget && vp_locked)
3242		vput(vp);
3243	else
3244		vrele(vp);
3245	vp = NULL;
3246	nfsm_clget;
3247	*tl = nfsrv_nfs_false;
3248	bp += NFSX_UNSIGNED;
3249	nfsm_clget;
3250	if (eofflag)
3251		*tl = nfsrv_nfs_true;
3252	else
3253		*tl = nfsrv_nfs_false;
3254	bp += NFSX_UNSIGNED;
3255	if (mp != mb) {
3256		if (bp < be)
3257			mp->m_len = bp - mtod(mp, caddr_t);
3258	} else
3259		mp->m_len += bp - bpos;
3260	free((caddr_t)cookies, M_TEMP);
3261	free((caddr_t)rbuf, M_TEMP);
3262nfsmout:
3263	if (vp)
3264		vrele(vp);
3265	if (mntp)
3266		vfs_unbusy(mntp);
3267	return(error);
3268}
3269
3270/*
3271 * nfs commit service
3272 */
3273int
3274nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3275    struct mbuf **mrq)
3276{
3277	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3278	struct sockaddr *nam = nfsd->nd_nam;
3279	caddr_t dpos = nfsd->nd_dpos;
3280	struct ucred *cred = nfsd->nd_cr;
3281	struct vattr bfor, aft;
3282	struct vnode *vp = NULL;
3283	nfsfh_t nfh;
3284	fhandle_t *fhp;
3285	u_int32_t *tl;
3286	caddr_t bpos;
3287	int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt;
3288	struct mbuf *mb, *mreq;
3289	u_quad_t off;
3290	struct mount *mp = NULL;
3291	int v3 = (nfsd->nd_flag & ND_NFSV3);
3292
3293	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3294	if (!v3)
3295		panic("nfsrv_commit: v3 proc called on a v2 connection");
3296	fhp = &nfh.fh_generic;
3297	nfsm_srvmtofh(fhp);
3298	if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
3299		error = ESTALE;
3300		goto ereply;
3301	}
3302	(void) vn_start_write(NULL, &mp, V_WAIT);
3303	vfs_rel(mp);		/* The write holds a ref. */
3304	tl = nfsm_dissect_nonblock(u_int32_t *, 3 * NFSX_UNSIGNED);
3305
3306	/*
3307	 * XXX At this time VOP_FSYNC() does not accept offset and byte
3308	 * count parameters, so these arguments are useless (someday maybe).
3309	 */
3310	off = fxdr_hyper(tl);
3311	tl += 2;
3312	cnt = fxdr_unsigned(int, *tl);
3313	error = nfsrv_fhtovp(fhp, 0, &vp, nfsd, slp, nam, &rdonly);
3314	if (error) {
3315		nfsm_reply(2 * NFSX_UNSIGNED);
3316		nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3317		error = 0;
3318		goto nfsmout;
3319	}
3320	for_ret = VOP_GETATTR(vp, &bfor, cred);
3321
3322	/*
3323	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of file
3324	 * is done.  At this time VOP_FSYNC does not accept offset and byte count
3325	 * parameters so call VOP_FSYNC the whole file for now.
3326	 */
3327	if (cnt == 0 || cnt > MAX_COMMIT_COUNT) {
3328		/*
3329		 * Give up and do the whole thing
3330		 */
3331		if (vp->v_object &&
3332		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3333			VM_OBJECT_WLOCK(vp->v_object);
3334			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3335			VM_OBJECT_WUNLOCK(vp->v_object);
3336		}
3337		error = VOP_FSYNC(vp, MNT_WAIT, curthread);
3338	} else {
3339		/*
3340		 * Locate and synchronously write any buffers that fall
3341		 * into the requested range.  Note:  we are assuming that
3342		 * f_iosize is a power of 2.
3343		 */
3344		int iosize = vp->v_mount->mnt_stat.f_iosize;
3345		int iomask = iosize - 1;
3346		struct bufobj *bo;
3347		daddr_t lblkno;
3348
3349		/*
3350		 * Align to iosize boundry, super-align to page boundry.
3351		 */
3352		if (off & iomask) {
3353			cnt += off & iomask;
3354			off &= ~(u_quad_t)iomask;
3355		}
3356		if (off & PAGE_MASK) {
3357			cnt += off & PAGE_MASK;
3358			off &= ~(u_quad_t)PAGE_MASK;
3359		}
3360		lblkno = off / iosize;
3361
3362		if (vp->v_object &&
3363		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3364			VM_OBJECT_WLOCK(vp->v_object);
3365			vm_object_page_clean(vp->v_object, off, off + cnt,
3366			    OBJPC_SYNC);
3367			VM_OBJECT_WUNLOCK(vp->v_object);
3368		}
3369
3370		bo = &vp->v_bufobj;
3371		BO_LOCK(bo);
3372		while (cnt > 0) {
3373			struct buf *bp;
3374
3375			/*
3376			 * If we have a buffer and it is marked B_DELWRI we
3377			 * have to lock and write it.  Otherwise the prior
3378			 * write is assumed to have already been committed.
3379			 *
3380			 * gbincore() can return invalid buffers now so we
3381			 * have to check that bit as well (though B_DELWRI
3382			 * should not be set if B_INVAL is set there could be
3383			 * a race here since we haven't locked the buffer).
3384			 */
3385			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
3386				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
3387				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
3388					BO_LOCK(bo);
3389					continue; /* retry */
3390				}
3391			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
3392				    B_DELWRI) {
3393					bremfree(bp);
3394					bp->b_flags &= ~B_ASYNC;
3395					bwrite(bp);
3396					++nfs_commit_miss;
3397				} else
3398					BUF_UNLOCK(bp);
3399				BO_LOCK(bo);
3400			}
3401			++nfs_commit_blks;
3402			if (cnt < iosize)
3403				break;
3404			cnt -= iosize;
3405			++lblkno;
3406		}
3407		BO_UNLOCK(bo);
3408	}
3409
3410	aft_ret = VOP_GETATTR(vp, &aft, cred);
3411	vput(vp);
3412	vp = NULL;
3413ereply:
3414	nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3415	nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3416	if (!error) {
3417		tl = nfsm_build(u_int32_t *, NFSX_V3WRITEVERF);
3418		if (nfsver.tv_sec == 0)
3419			nfsver = boottime;
3420		*tl++ = txdr_unsigned(nfsver.tv_sec);
3421		*tl = txdr_unsigned(nfsver.tv_usec);
3422	} else {
3423		error = 0;
3424	}
3425nfsmout:
3426	if (vp)
3427		vput(vp);
3428	vn_finished_write(mp);
3429	return(error);
3430}
3431
3432/*
3433 * nfs statfs service
3434 */
3435int
3436nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3437    struct mbuf **mrq)
3438{
3439	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3440	struct sockaddr *nam = nfsd->nd_nam;
3441	caddr_t dpos = nfsd->nd_dpos;
3442	struct ucred *cred = nfsd->nd_cr;
3443	struct statfs *sf;
3444	struct nfs_statfs *sfp;
3445	caddr_t bpos;
3446	int error = 0, rdonly, getret = 1;
3447	int v3 = (nfsd->nd_flag & ND_NFSV3);
3448	struct mbuf *mb, *mreq;
3449	struct vnode *vp = NULL;
3450	struct vattr at;
3451	nfsfh_t nfh;
3452	fhandle_t *fhp;
3453	struct statfs statfs;
3454	u_quad_t tval;
3455
3456	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3457	fhp = &nfh.fh_generic;
3458	nfsm_srvmtofh(fhp);
3459	error = nfsrv_fhtovp(fhp, 0, &vp, nfsd, slp, nam, &rdonly);
3460	if (error) {
3461		nfsm_reply(NFSX_UNSIGNED);
3462		if (v3)
3463			nfsm_srvpostop_attr(getret, &at);
3464		error = 0;
3465		goto nfsmout;
3466	}
3467	sf = &statfs;
3468	error = VFS_STATFS(vp->v_mount, sf);
3469	getret = VOP_GETATTR(vp, &at, cred);
3470	vput(vp);
3471	vp = NULL;
3472	nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3473	if (v3)
3474		nfsm_srvpostop_attr(getret, &at);
3475	if (error) {
3476		error = 0;
3477		goto nfsmout;
3478	}
3479	sfp = nfsm_build(struct nfs_statfs *, NFSX_STATFS(v3));
3480	if (v3) {
3481		tval = (u_quad_t)sf->f_blocks;
3482		tval *= (u_quad_t)sf->f_bsize;
3483		txdr_hyper(tval, &sfp->sf_tbytes);
3484		tval = (u_quad_t)sf->f_bfree;
3485		tval *= (u_quad_t)sf->f_bsize;
3486		txdr_hyper(tval, &sfp->sf_fbytes);
3487		/*
3488		 * Don't send negative values for available space,
3489		 * since this field is unsigned in the NFS protocol.
3490		 * Otherwise, the client would see absurdly high
3491		 * numbers for free space.
3492		 */
3493		if (sf->f_bavail < 0)
3494			tval = 0;
3495		else
3496			tval = (u_quad_t)sf->f_bavail;
3497		tval *= (u_quad_t)sf->f_bsize;
3498		txdr_hyper(tval, &sfp->sf_abytes);
3499		sfp->sf_tfiles.nfsuquad[0] = 0;
3500		sfp->sf_tfiles.nfsuquad[1] = txdr_unsigned(sf->f_files);
3501		sfp->sf_ffiles.nfsuquad[0] = 0;
3502		sfp->sf_ffiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3503		sfp->sf_afiles.nfsuquad[0] = 0;
3504		sfp->sf_afiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3505		sfp->sf_invarsec = 0;
3506	} else {
3507		sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3508		sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
3509		sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3510		sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3511		if (sf->f_bavail < 0)
3512			sfp->sf_bavail = 0;
3513		else
3514			sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3515	}
3516nfsmout:
3517	if (vp)
3518		vput(vp);
3519	return(error);
3520}
3521
3522/*
3523 * nfs fsinfo service
3524 */
3525int
3526nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3527    struct mbuf **mrq)
3528{
3529	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3530	struct sockaddr *nam = nfsd->nd_nam;
3531	caddr_t dpos = nfsd->nd_dpos;
3532	struct ucred *cred = nfsd->nd_cr;
3533	struct nfsv3_fsinfo *sip;
3534	caddr_t bpos;
3535	int error = 0, rdonly, getret = 1, pref;
3536	struct mbuf *mb, *mreq;
3537	struct vnode *vp = NULL;
3538	struct vattr at;
3539	nfsfh_t nfh;
3540	fhandle_t *fhp;
3541	u_quad_t maxfsize;
3542	struct statfs sb;
3543	int v3 = (nfsd->nd_flag & ND_NFSV3);
3544
3545	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3546	if (!v3)
3547		panic("nfsrv_fsinfo: v3 proc called on a v2 connection");
3548	fhp = &nfh.fh_generic;
3549	nfsm_srvmtofh(fhp);
3550	error = nfsrv_fhtovp(fhp, 0, &vp, nfsd, slp, nam, &rdonly);
3551	if (error) {
3552		nfsm_reply(NFSX_UNSIGNED);
3553		nfsm_srvpostop_attr(getret, &at);
3554		error = 0;
3555		goto nfsmout;
3556	}
3557
3558	/* XXX Try to make a guess on the max file size. */
3559	VFS_STATFS(vp->v_mount, &sb);
3560	maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
3561
3562	getret = VOP_GETATTR(vp, &at, cred);
3563	vput(vp);
3564	vp = NULL;
3565	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3566	nfsm_srvpostop_attr(getret, &at);
3567	sip = nfsm_build(struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3568
3569	/*
3570	 * XXX
3571	 * There should be filesystem VFS OP(s) to get this information.
3572	 * For now, assume ufs.
3573	 */
3574	pref = NFS_SRVMAXDATA(nfsd);
3575	sip->fs_rtmax = txdr_unsigned(pref);
3576	sip->fs_rtpref = txdr_unsigned(pref);
3577	sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3578	sip->fs_wtmax = txdr_unsigned(pref);
3579	sip->fs_wtpref = txdr_unsigned(pref);
3580	sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3581	sip->fs_dtpref = txdr_unsigned(pref);
3582	txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3583	sip->fs_timedelta.nfsv3_sec = 0;
3584	sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3585	sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3586		NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3587		NFSV3FSINFO_CANSETTIME);
3588nfsmout:
3589	if (vp)
3590		vput(vp);
3591	return(error);
3592}
3593
3594/*
3595 * nfs pathconf service
3596 */
3597int
3598nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3599    struct mbuf **mrq)
3600{
3601	struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3602	struct sockaddr *nam = nfsd->nd_nam;
3603	caddr_t dpos = nfsd->nd_dpos;
3604	struct ucred *cred = nfsd->nd_cr;
3605	struct nfsv3_pathconf *pc;
3606	caddr_t bpos;
3607	int error = 0, rdonly, getret = 1;
3608	register_t linkmax, namemax, chownres, notrunc;
3609	struct mbuf *mb, *mreq;
3610	struct vnode *vp = NULL;
3611	struct vattr at;
3612	nfsfh_t nfh;
3613	fhandle_t *fhp;
3614	int v3 = (nfsd->nd_flag & ND_NFSV3);
3615
3616	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3617	if (!v3)
3618		panic("nfsrv_pathconf: v3 proc called on a v2 connection");
3619	fhp = &nfh.fh_generic;
3620	nfsm_srvmtofh(fhp);
3621	error = nfsrv_fhtovp(fhp, 0, &vp, nfsd, slp, nam, &rdonly);
3622	if (error) {
3623		nfsm_reply(NFSX_UNSIGNED);
3624		nfsm_srvpostop_attr(getret, &at);
3625		error = 0;
3626		goto nfsmout;
3627	}
3628	error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3629	if (!error)
3630		error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3631	if (!error)
3632		error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3633	if (!error)
3634		error = VOP_PATHCONF(vp, _PC_NO_TRUNC, &notrunc);
3635	getret = VOP_GETATTR(vp, &at, cred);
3636	vput(vp);
3637	vp = NULL;
3638	nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3639	nfsm_srvpostop_attr(getret, &at);
3640	if (error) {
3641		error = 0;
3642		goto nfsmout;
3643	}
3644	pc = nfsm_build(struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3645
3646	pc->pc_linkmax = txdr_unsigned(linkmax);
3647	pc->pc_namemax = txdr_unsigned(namemax);
3648	pc->pc_notrunc = txdr_unsigned(notrunc);
3649	pc->pc_chownrestricted = txdr_unsigned(chownres);
3650
3651	/*
3652	 * These should probably be supported by VOP_PATHCONF(), but
3653	 * until msdosfs is exportable (why would you want to?), the
3654	 * Unix defaults should be ok.
3655	 */
3656	pc->pc_caseinsensitive = nfsrv_nfs_false;
3657	pc->pc_casepreserving = nfsrv_nfs_true;
3658nfsmout:
3659	if (vp)
3660		vput(vp);
3661	return(error);
3662}
3663
3664/*
3665 * Null operation, used by clients to ping server
3666 */
3667/* ARGSUSED */
3668int
3669nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3670    struct mbuf **mrq)
3671{
3672	struct mbuf *mrep = nfsd->nd_mrep;
3673	caddr_t bpos;
3674	int error = NFSERR_RETVOID;
3675	struct mbuf *mb, *mreq;
3676
3677	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3678	nfsm_reply(0);
3679nfsmout:
3680	return (error);
3681}
3682
3683/*
3684 * No operation, used for obsolete procedures
3685 */
3686/* ARGSUSED */
3687int
3688nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3689    struct mbuf **mrq)
3690{
3691	struct mbuf *mrep = nfsd->nd_mrep;
3692	caddr_t bpos;
3693	int error;
3694	struct mbuf *mb, *mreq;
3695
3696	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3697	if (nfsd->nd_repstat)
3698		error = nfsd->nd_repstat;
3699	else
3700		error = EPROCUNAVAIL;
3701	nfsm_reply(0);
3702	error = 0;
3703nfsmout:
3704	return (error);
3705}
3706
3707/*
3708 * Perform access checking for vnodes obtained from file handles that would
3709 * refer to files already opened by a Unix client. You cannot just use
3710 * vn_writechk() and VOP_ACCESS() for two reasons.
3711 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
3712 *     case.
3713 * 2 - The owner is to be given access irrespective of mode bits for some
3714 *     operations, so that processes that chmod after opening a file don't
3715 *     break. I don't like this because it opens a security hole, but since
3716 *     the nfs server opens a security hole the size of a barn door anyhow,
3717 *     what the heck.
3718 *
3719 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3720 * will return EPERM instead of EACCES. EPERM is always an error.
3721 */
3722static int
3723nfsrv_access(struct vnode *vp, accmode_t accmode, struct ucred *cred,
3724    int rdonly, int override)
3725{
3726	struct vattr vattr;
3727	int error;
3728
3729	nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3730
3731	if (accmode & VWRITE) {
3732		/* Just vn_writechk() changed to check rdonly */
3733		/*
3734		 * Disallow write attempts on read-only filesystems;
3735		 * unless the file is a socket or a block or character
3736		 * device resident on the filesystem.
3737		 */
3738		if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3739			switch (vp->v_type) {
3740			case VREG:
3741			case VDIR:
3742			case VLNK:
3743				return (EROFS);
3744			default:
3745				break;
3746			}
3747		}
3748		/*
3749		 * If there's shared text associated with
3750		 * the inode, we can't allow writing.
3751		 */
3752		if (VOP_IS_TEXT(vp))
3753			return (ETXTBSY);
3754	}
3755
3756	error = VOP_GETATTR(vp, &vattr, cred);
3757	if (error)
3758		return (error);
3759	error = VOP_ACCESS(vp, accmode, cred, curthread);
3760	/*
3761	 * Allow certain operations for the owner (reads and writes
3762	 * on files that are already open).
3763	 */
3764	if (override && error == EACCES && cred->cr_uid == vattr.va_uid)
3765		error = 0;
3766	return (error);
3767}
3768