nfs_nfsdport.c revision 280258
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 */
33
34#include <sys/cdefs.h>
35__FBSDID("$FreeBSD: stable/10/sys/fs/nfsserver/nfs_nfsdport.c 280258 2015-03-19 13:37:36Z rwatson $");
36
37#include <sys/capsicum.h>
38
39/*
40 * Functions that perform the vfs operations required by the routines in
41 * nfsd_serv.c. It is hoped that this change will make the server more
42 * portable.
43 */
44
45#include <fs/nfs/nfsport.h>
46#include <sys/hash.h>
47#include <sys/sysctl.h>
48#include <nlm/nlm_prot.h>
49#include <nlm/nlm.h>
50
51FEATURE(nfsd, "NFSv4 server");
52
53extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
54extern int nfsrv_useacl;
55extern int newnfs_numnfsd;
56extern struct mount nfsv4root_mnt;
57extern struct nfsrv_stablefirst nfsrv_stablefirst;
58extern void (*nfsd_call_servertimer)(void);
59extern SVCPOOL	*nfsrvd_pool;
60extern struct nfsv4lock nfsd_suspend_lock;
61extern struct nfssessionhash nfssessionhash[NFSSESSIONHASHSIZE];
62struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
63NFSDLOCKMUTEX;
64struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
65struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
66struct mtx nfsrc_udpmtx;
67struct mtx nfs_v4root_mutex;
68struct nfsrvfh nfs_rootfh, nfs_pubfh;
69int nfs_pubfhset = 0, nfs_rootfhset = 0;
70struct proc *nfsd_master_proc = NULL;
71int nfsd_debuglevel = 0;
72static pid_t nfsd_master_pid = (pid_t)-1;
73static char nfsd_master_comm[MAXCOMLEN + 1];
74static struct timeval nfsd_master_start;
75static uint32_t nfsv4_sysid = 0;
76
77static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
78    struct ucred *);
79
80int nfsrv_enable_crossmntpt = 1;
81static int nfs_commit_blks;
82static int nfs_commit_miss;
83extern int nfsrv_issuedelegs;
84extern int nfsrv_dolocallocks;
85extern int nfsd_enable_stringtouid;
86
87SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server");
88SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
89    &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
90SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
91    0, "");
92SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
93    0, "");
94SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
95    &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
96SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
97    &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
98SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
99    0, "Debug level for new nfs server");
100SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
101    &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
102
103#define	MAX_REORDERED_RPC	16
104#define	NUM_HEURISTIC		1031
105#define	NHUSE_INIT		64
106#define	NHUSE_INC		16
107#define	NHUSE_MAX		2048
108
109static struct nfsheur {
110	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
111	off_t nh_nextoff;	/* next offset for sequential detection */
112	int nh_use;		/* use count for selection */
113	int nh_seqcount;	/* heuristic */
114} nfsheur[NUM_HEURISTIC];
115
116
117/*
118 * Heuristic to detect sequential operation.
119 */
120static struct nfsheur *
121nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
122{
123	struct nfsheur *nh;
124	int hi, try;
125
126	/* Locate best candidate. */
127	try = 32;
128	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
129	nh = &nfsheur[hi];
130	while (try--) {
131		if (nfsheur[hi].nh_vp == vp) {
132			nh = &nfsheur[hi];
133			break;
134		}
135		if (nfsheur[hi].nh_use > 0)
136			--nfsheur[hi].nh_use;
137		hi = (hi + 1) % NUM_HEURISTIC;
138		if (nfsheur[hi].nh_use < nh->nh_use)
139			nh = &nfsheur[hi];
140	}
141
142	/* Initialize hint if this is a new file. */
143	if (nh->nh_vp != vp) {
144		nh->nh_vp = vp;
145		nh->nh_nextoff = uio->uio_offset;
146		nh->nh_use = NHUSE_INIT;
147		if (uio->uio_offset == 0)
148			nh->nh_seqcount = 4;
149		else
150			nh->nh_seqcount = 1;
151	}
152
153	/* Calculate heuristic. */
154	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
155	    uio->uio_offset == nh->nh_nextoff) {
156		/* See comments in vfs_vnops.c:sequential_heuristic(). */
157		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
158		if (nh->nh_seqcount > IO_SEQMAX)
159			nh->nh_seqcount = IO_SEQMAX;
160	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
161	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
162		/* Probably a reordered RPC, leave seqcount alone. */
163	} else if (nh->nh_seqcount > 1) {
164		nh->nh_seqcount /= 2;
165	} else {
166		nh->nh_seqcount = 0;
167	}
168	nh->nh_use += NHUSE_INC;
169	if (nh->nh_use > NHUSE_MAX)
170		nh->nh_use = NHUSE_MAX;
171	return (nh);
172}
173
174/*
175 * Get attributes into nfsvattr structure.
176 */
177int
178nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
179    struct thread *p, int vpislocked)
180{
181	int error, lockedit = 0;
182
183	if (vpislocked == 0) {
184		/*
185		 * When vpislocked == 0, the vnode is either exclusively
186		 * locked by this thread or not locked by this thread.
187		 * As such, shared lock it, if not exclusively locked.
188		 */
189		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
190			lockedit = 1;
191			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
192		}
193	}
194	error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
195	if (lockedit != 0)
196		NFSVOPUNLOCK(vp, 0);
197
198	NFSEXITCODE(error);
199	return (error);
200}
201
202/*
203 * Get a file handle for a vnode.
204 */
205int
206nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
207{
208	int error;
209
210	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
211	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
212	error = VOP_VPTOFH(vp, &fhp->fh_fid);
213
214	NFSEXITCODE(error);
215	return (error);
216}
217
218/*
219 * Perform access checking for vnodes obtained from file handles that would
220 * refer to files already opened by a Unix client. You cannot just use
221 * vn_writechk() and VOP_ACCESSX() for two reasons.
222 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
223 *     case.
224 * 2 - The owner is to be given access irrespective of mode bits for some
225 *     operations, so that processes that chmod after opening a file don't
226 *     break.
227 */
228int
229nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
230    struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
231    u_int32_t *supportedtypep)
232{
233	struct vattr vattr;
234	int error = 0, getret = 0;
235
236	if (vpislocked == 0) {
237		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
238			error = EPERM;
239			goto out;
240		}
241	}
242	if (accmode & VWRITE) {
243		/* Just vn_writechk() changed to check rdonly */
244		/*
245		 * Disallow write attempts on read-only file systems;
246		 * unless the file is a socket or a block or character
247		 * device resident on the file system.
248		 */
249		if (NFSVNO_EXRDONLY(exp) ||
250		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
251			switch (vp->v_type) {
252			case VREG:
253			case VDIR:
254			case VLNK:
255				error = EROFS;
256			default:
257				break;
258			}
259		}
260		/*
261		 * If there's shared text associated with
262		 * the inode, try to free it up once.  If
263		 * we fail, we can't allow writing.
264		 */
265		if (VOP_IS_TEXT(vp) && error == 0)
266			error = ETXTBSY;
267	}
268	if (error != 0) {
269		if (vpislocked == 0)
270			NFSVOPUNLOCK(vp, 0);
271		goto out;
272	}
273
274	/*
275	 * Should the override still be applied when ACLs are enabled?
276	 */
277	error = VOP_ACCESSX(vp, accmode, cred, p);
278	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
279		/*
280		 * Try again with VEXPLICIT_DENY, to see if the test for
281		 * deletion is supported.
282		 */
283		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
284		if (error == 0) {
285			if (vp->v_type == VDIR) {
286				accmode &= ~(VDELETE | VDELETE_CHILD);
287				accmode |= VWRITE;
288				error = VOP_ACCESSX(vp, accmode, cred, p);
289			} else if (supportedtypep != NULL) {
290				*supportedtypep &= ~NFSACCESS_DELETE;
291			}
292		}
293	}
294
295	/*
296	 * Allow certain operations for the owner (reads and writes
297	 * on files that are already open).
298	 */
299	if (override != NFSACCCHK_NOOVERRIDE &&
300	    (error == EPERM || error == EACCES)) {
301		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
302			error = 0;
303		else if (override & NFSACCCHK_ALLOWOWNER) {
304			getret = VOP_GETATTR(vp, &vattr, cred);
305			if (getret == 0 && cred->cr_uid == vattr.va_uid)
306				error = 0;
307		}
308	}
309	if (vpislocked == 0)
310		NFSVOPUNLOCK(vp, 0);
311
312out:
313	NFSEXITCODE(error);
314	return (error);
315}
316
317/*
318 * Set attribute(s) vnop.
319 */
320int
321nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
322    struct thread *p, struct nfsexstuff *exp)
323{
324	int error;
325
326	error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
327	NFSEXITCODE(error);
328	return (error);
329}
330
331/*
332 * Set up nameidata for a lookup() call and do it.
333 */
334int
335nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
336    struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
337    struct vnode **retdirp)
338{
339	struct componentname *cnp = &ndp->ni_cnd;
340	int i;
341	struct iovec aiov;
342	struct uio auio;
343	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
344	int error = 0, crossmnt;
345	char *cp;
346
347	*retdirp = NULL;
348	cnp->cn_nameptr = cnp->cn_pnbuf;
349	ndp->ni_strictrelative = 0;
350	/*
351	 * Extract and set starting directory.
352	 */
353	if (dp->v_type != VDIR) {
354		if (islocked)
355			vput(dp);
356		else
357			vrele(dp);
358		nfsvno_relpathbuf(ndp);
359		error = ENOTDIR;
360		goto out1;
361	}
362	if (islocked)
363		NFSVOPUNLOCK(dp, 0);
364	VREF(dp);
365	*retdirp = dp;
366	if (NFSVNO_EXRDONLY(exp))
367		cnp->cn_flags |= RDONLY;
368	ndp->ni_segflg = UIO_SYSSPACE;
369	crossmnt = 1;
370
371	if (nd->nd_flag & ND_PUBLOOKUP) {
372		ndp->ni_loopcnt = 0;
373		if (cnp->cn_pnbuf[0] == '/') {
374			vrele(dp);
375			/*
376			 * Check for degenerate pathnames here, since lookup()
377			 * panics on them.
378			 */
379			for (i = 1; i < ndp->ni_pathlen; i++)
380				if (cnp->cn_pnbuf[i] != '/')
381					break;
382			if (i == ndp->ni_pathlen) {
383				error = NFSERR_ACCES;
384				goto out;
385			}
386			dp = rootvnode;
387			VREF(dp);
388		}
389	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
390	    (nd->nd_flag & ND_NFSV4) == 0) {
391		/*
392		 * Only cross mount points for NFSv4 when doing a
393		 * mount while traversing the file system above
394		 * the mount point, unless nfsrv_enable_crossmntpt is set.
395		 */
396		cnp->cn_flags |= NOCROSSMOUNT;
397		crossmnt = 0;
398	}
399
400	/*
401	 * Initialize for scan, set ni_startdir and bump ref on dp again
402	 * because lookup() will dereference ni_startdir.
403	 */
404
405	cnp->cn_thread = p;
406	ndp->ni_startdir = dp;
407	ndp->ni_rootdir = rootvnode;
408	ndp->ni_topdir = NULL;
409
410	if (!lockleaf)
411		cnp->cn_flags |= LOCKLEAF;
412	for (;;) {
413		cnp->cn_nameptr = cnp->cn_pnbuf;
414		/*
415		 * Call lookup() to do the real work.  If an error occurs,
416		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
417		 * we do not have to dereference anything before returning.
418		 * In either case ni_startdir will be dereferenced and NULLed
419		 * out.
420		 */
421		error = lookup(ndp);
422		if (error)
423			break;
424
425		/*
426		 * Check for encountering a symbolic link.  Trivial
427		 * termination occurs if no symlink encountered.
428		 */
429		if ((cnp->cn_flags & ISSYMLINK) == 0) {
430			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
431				nfsvno_relpathbuf(ndp);
432			if (ndp->ni_vp && !lockleaf)
433				NFSVOPUNLOCK(ndp->ni_vp, 0);
434			break;
435		}
436
437		/*
438		 * Validate symlink
439		 */
440		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
441			NFSVOPUNLOCK(ndp->ni_dvp, 0);
442		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
443			error = EINVAL;
444			goto badlink2;
445		}
446
447		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
448			error = ELOOP;
449			goto badlink2;
450		}
451		if (ndp->ni_pathlen > 1)
452			cp = uma_zalloc(namei_zone, M_WAITOK);
453		else
454			cp = cnp->cn_pnbuf;
455		aiov.iov_base = cp;
456		aiov.iov_len = MAXPATHLEN;
457		auio.uio_iov = &aiov;
458		auio.uio_iovcnt = 1;
459		auio.uio_offset = 0;
460		auio.uio_rw = UIO_READ;
461		auio.uio_segflg = UIO_SYSSPACE;
462		auio.uio_td = NULL;
463		auio.uio_resid = MAXPATHLEN;
464		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
465		if (error) {
466		badlink1:
467			if (ndp->ni_pathlen > 1)
468				uma_zfree(namei_zone, cp);
469		badlink2:
470			vrele(ndp->ni_dvp);
471			vput(ndp->ni_vp);
472			break;
473		}
474		linklen = MAXPATHLEN - auio.uio_resid;
475		if (linklen == 0) {
476			error = ENOENT;
477			goto badlink1;
478		}
479		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
480			error = ENAMETOOLONG;
481			goto badlink1;
482		}
483
484		/*
485		 * Adjust or replace path
486		 */
487		if (ndp->ni_pathlen > 1) {
488			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
489			uma_zfree(namei_zone, cnp->cn_pnbuf);
490			cnp->cn_pnbuf = cp;
491		} else
492			cnp->cn_pnbuf[linklen] = '\0';
493		ndp->ni_pathlen += linklen;
494
495		/*
496		 * Cleanup refs for next loop and check if root directory
497		 * should replace current directory.  Normally ni_dvp
498		 * becomes the new base directory and is cleaned up when
499		 * we loop.  Explicitly null pointers after invalidation
500		 * to clarify operation.
501		 */
502		vput(ndp->ni_vp);
503		ndp->ni_vp = NULL;
504
505		if (cnp->cn_pnbuf[0] == '/') {
506			vrele(ndp->ni_dvp);
507			ndp->ni_dvp = ndp->ni_rootdir;
508			VREF(ndp->ni_dvp);
509		}
510		ndp->ni_startdir = ndp->ni_dvp;
511		ndp->ni_dvp = NULL;
512	}
513	if (!lockleaf)
514		cnp->cn_flags &= ~LOCKLEAF;
515
516out:
517	if (error) {
518		nfsvno_relpathbuf(ndp);
519		ndp->ni_vp = NULL;
520		ndp->ni_dvp = NULL;
521		ndp->ni_startdir = NULL;
522	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
523		ndp->ni_dvp = NULL;
524	}
525
526out1:
527	NFSEXITCODE2(error, nd);
528	return (error);
529}
530
531/*
532 * Set up a pathname buffer and return a pointer to it and, optionally
533 * set a hash pointer.
534 */
535void
536nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
537{
538	struct componentname *cnp = &ndp->ni_cnd;
539
540	cnp->cn_flags |= (NOMACCHECK | HASBUF);
541	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
542	if (hashpp != NULL)
543		*hashpp = NULL;
544	*bufpp = cnp->cn_pnbuf;
545}
546
547/*
548 * Release the above path buffer, if not released by nfsvno_namei().
549 */
550void
551nfsvno_relpathbuf(struct nameidata *ndp)
552{
553
554	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
555		panic("nfsrelpath");
556	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
557	ndp->ni_cnd.cn_flags &= ~HASBUF;
558}
559
560/*
561 * Readlink vnode op into an mbuf list.
562 */
563int
564nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
565    struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
566{
567	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
568	struct iovec *ivp = iv;
569	struct uio io, *uiop = &io;
570	struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
571	int i, len, tlen, error = 0;
572
573	len = 0;
574	i = 0;
575	while (len < NFS_MAXPATHLEN) {
576		NFSMGET(mp);
577		MCLGET(mp, M_WAITOK);
578		mp->m_len = NFSMSIZ(mp);
579		if (len == 0) {
580			mp3 = mp2 = mp;
581		} else {
582			mp2->m_next = mp;
583			mp2 = mp;
584		}
585		if ((len + mp->m_len) > NFS_MAXPATHLEN) {
586			mp->m_len = NFS_MAXPATHLEN - len;
587			len = NFS_MAXPATHLEN;
588		} else {
589			len += mp->m_len;
590		}
591		ivp->iov_base = mtod(mp, caddr_t);
592		ivp->iov_len = mp->m_len;
593		i++;
594		ivp++;
595	}
596	uiop->uio_iov = iv;
597	uiop->uio_iovcnt = i;
598	uiop->uio_offset = 0;
599	uiop->uio_resid = len;
600	uiop->uio_rw = UIO_READ;
601	uiop->uio_segflg = UIO_SYSSPACE;
602	uiop->uio_td = NULL;
603	error = VOP_READLINK(vp, uiop, cred);
604	if (error) {
605		m_freem(mp3);
606		*lenp = 0;
607		goto out;
608	}
609	if (uiop->uio_resid > 0) {
610		len -= uiop->uio_resid;
611		tlen = NFSM_RNDUP(len);
612		nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
613	}
614	*lenp = len;
615	*mpp = mp3;
616	*mpendp = mp;
617
618out:
619	NFSEXITCODE(error);
620	return (error);
621}
622
623/*
624 * Read vnode op call into mbuf list.
625 */
626int
627nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
628    struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
629{
630	struct mbuf *m;
631	int i;
632	struct iovec *iv;
633	struct iovec *iv2;
634	int error = 0, len, left, siz, tlen, ioflag = 0;
635	struct mbuf *m2 = NULL, *m3;
636	struct uio io, *uiop = &io;
637	struct nfsheur *nh;
638
639	len = left = NFSM_RNDUP(cnt);
640	m3 = NULL;
641	/*
642	 * Generate the mbuf list with the uio_iov ref. to it.
643	 */
644	i = 0;
645	while (left > 0) {
646		NFSMGET(m);
647		MCLGET(m, M_WAITOK);
648		m->m_len = 0;
649		siz = min(M_TRAILINGSPACE(m), left);
650		left -= siz;
651		i++;
652		if (m3)
653			m2->m_next = m;
654		else
655			m3 = m;
656		m2 = m;
657	}
658	MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
659	    M_TEMP, M_WAITOK);
660	uiop->uio_iov = iv2 = iv;
661	m = m3;
662	left = len;
663	i = 0;
664	while (left > 0) {
665		if (m == NULL)
666			panic("nfsvno_read iov");
667		siz = min(M_TRAILINGSPACE(m), left);
668		if (siz > 0) {
669			iv->iov_base = mtod(m, caddr_t) + m->m_len;
670			iv->iov_len = siz;
671			m->m_len += siz;
672			left -= siz;
673			iv++;
674			i++;
675		}
676		m = m->m_next;
677	}
678	uiop->uio_iovcnt = i;
679	uiop->uio_offset = off;
680	uiop->uio_resid = len;
681	uiop->uio_rw = UIO_READ;
682	uiop->uio_segflg = UIO_SYSSPACE;
683	uiop->uio_td = NULL;
684	nh = nfsrv_sequential_heuristic(uiop, vp);
685	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
686	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
687	FREE((caddr_t)iv2, M_TEMP);
688	if (error) {
689		m_freem(m3);
690		*mpp = NULL;
691		goto out;
692	}
693	nh->nh_nextoff = uiop->uio_offset;
694	tlen = len - uiop->uio_resid;
695	cnt = cnt < tlen ? cnt : tlen;
696	tlen = NFSM_RNDUP(cnt);
697	if (tlen == 0) {
698		m_freem(m3);
699		m3 = NULL;
700	} else if (len != tlen || tlen != cnt)
701		nfsrv_adj(m3, len - tlen, tlen - cnt);
702	*mpp = m3;
703	*mpendp = m2;
704
705out:
706	NFSEXITCODE(error);
707	return (error);
708}
709
710/*
711 * Write vnode op from an mbuf list.
712 */
713int
714nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
715    struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
716{
717	struct iovec *ivp;
718	int i, len;
719	struct iovec *iv;
720	int ioflags, error;
721	struct uio io, *uiop = &io;
722	struct nfsheur *nh;
723
724	MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
725	    M_WAITOK);
726	uiop->uio_iov = iv = ivp;
727	uiop->uio_iovcnt = cnt;
728	i = mtod(mp, caddr_t) + mp->m_len - cp;
729	len = retlen;
730	while (len > 0) {
731		if (mp == NULL)
732			panic("nfsvno_write");
733		if (i > 0) {
734			i = min(i, len);
735			ivp->iov_base = cp;
736			ivp->iov_len = i;
737			ivp++;
738			len -= i;
739		}
740		mp = mp->m_next;
741		if (mp) {
742			i = mp->m_len;
743			cp = mtod(mp, caddr_t);
744		}
745	}
746
747	if (stable == NFSWRITE_UNSTABLE)
748		ioflags = IO_NODELOCKED;
749	else
750		ioflags = (IO_SYNC | IO_NODELOCKED);
751	uiop->uio_resid = retlen;
752	uiop->uio_rw = UIO_WRITE;
753	uiop->uio_segflg = UIO_SYSSPACE;
754	NFSUIOPROC(uiop, p);
755	uiop->uio_offset = off;
756	nh = nfsrv_sequential_heuristic(uiop, vp);
757	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
758	error = VOP_WRITE(vp, uiop, ioflags, cred);
759	if (error == 0)
760		nh->nh_nextoff = uiop->uio_offset;
761	FREE((caddr_t)iv, M_TEMP);
762
763	NFSEXITCODE(error);
764	return (error);
765}
766
767/*
768 * Common code for creating a regular file (plus special files for V2).
769 */
770int
771nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
772    struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
773    int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
774{
775	u_quad_t tempsize;
776	int error;
777
778	error = nd->nd_repstat;
779	if (!error && ndp->ni_vp == NULL) {
780		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
781			vrele(ndp->ni_startdir);
782			error = VOP_CREATE(ndp->ni_dvp,
783			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
784			vput(ndp->ni_dvp);
785			nfsvno_relpathbuf(ndp);
786			if (!error) {
787				if (*exclusive_flagp) {
788					*exclusive_flagp = 0;
789					NFSVNO_ATTRINIT(nvap);
790					nvap->na_atime.tv_sec = cverf[0];
791					nvap->na_atime.tv_nsec = cverf[1];
792					error = VOP_SETATTR(ndp->ni_vp,
793					    &nvap->na_vattr, nd->nd_cred);
794				}
795			}
796		/*
797		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
798		 * (This implies, just get out on an error.)
799		 */
800		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
801			nvap->na_type == VFIFO) {
802			if (nvap->na_type == VCHR && rdev == 0xffffffff)
803				nvap->na_type = VFIFO;
804                        if (nvap->na_type != VFIFO &&
805			    (error = priv_check_cred(nd->nd_cred,
806			     PRIV_VFS_MKNOD_DEV, 0))) {
807				vrele(ndp->ni_startdir);
808				nfsvno_relpathbuf(ndp);
809				vput(ndp->ni_dvp);
810				goto out;
811			}
812			nvap->na_rdev = rdev;
813			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
814			    &ndp->ni_cnd, &nvap->na_vattr);
815			vput(ndp->ni_dvp);
816			nfsvno_relpathbuf(ndp);
817			vrele(ndp->ni_startdir);
818			if (error)
819				goto out;
820		} else {
821			vrele(ndp->ni_startdir);
822			nfsvno_relpathbuf(ndp);
823			vput(ndp->ni_dvp);
824			error = ENXIO;
825			goto out;
826		}
827		*vpp = ndp->ni_vp;
828	} else {
829		/*
830		 * Handle cases where error is already set and/or
831		 * the file exists.
832		 * 1 - clean up the lookup
833		 * 2 - iff !error and na_size set, truncate it
834		 */
835		vrele(ndp->ni_startdir);
836		nfsvno_relpathbuf(ndp);
837		*vpp = ndp->ni_vp;
838		if (ndp->ni_dvp == *vpp)
839			vrele(ndp->ni_dvp);
840		else
841			vput(ndp->ni_dvp);
842		if (!error && nvap->na_size != VNOVAL) {
843			error = nfsvno_accchk(*vpp, VWRITE,
844			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
845			    NFSACCCHK_VPISLOCKED, NULL);
846			if (!error) {
847				tempsize = nvap->na_size;
848				NFSVNO_ATTRINIT(nvap);
849				nvap->na_size = tempsize;
850				error = VOP_SETATTR(*vpp,
851				    &nvap->na_vattr, nd->nd_cred);
852			}
853		}
854		if (error)
855			vput(*vpp);
856	}
857
858out:
859	NFSEXITCODE(error);
860	return (error);
861}
862
863/*
864 * Do a mknod vnode op.
865 */
866int
867nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
868    struct thread *p)
869{
870	int error = 0;
871	enum vtype vtyp;
872
873	vtyp = nvap->na_type;
874	/*
875	 * Iff doesn't exist, create it.
876	 */
877	if (ndp->ni_vp) {
878		vrele(ndp->ni_startdir);
879		nfsvno_relpathbuf(ndp);
880		vput(ndp->ni_dvp);
881		vrele(ndp->ni_vp);
882		error = EEXIST;
883		goto out;
884	}
885	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
886		vrele(ndp->ni_startdir);
887		nfsvno_relpathbuf(ndp);
888		vput(ndp->ni_dvp);
889		error = NFSERR_BADTYPE;
890		goto out;
891	}
892	if (vtyp == VSOCK) {
893		vrele(ndp->ni_startdir);
894		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
895		    &ndp->ni_cnd, &nvap->na_vattr);
896		vput(ndp->ni_dvp);
897		nfsvno_relpathbuf(ndp);
898	} else {
899		if (nvap->na_type != VFIFO &&
900		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
901			vrele(ndp->ni_startdir);
902			nfsvno_relpathbuf(ndp);
903			vput(ndp->ni_dvp);
904			goto out;
905		}
906		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
907		    &ndp->ni_cnd, &nvap->na_vattr);
908		vput(ndp->ni_dvp);
909		nfsvno_relpathbuf(ndp);
910		vrele(ndp->ni_startdir);
911		/*
912		 * Since VOP_MKNOD returns the ni_vp, I can't
913		 * see any reason to do the lookup.
914		 */
915	}
916
917out:
918	NFSEXITCODE(error);
919	return (error);
920}
921
922/*
923 * Mkdir vnode op.
924 */
925int
926nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
927    struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
928{
929	int error = 0;
930
931	if (ndp->ni_vp != NULL) {
932		if (ndp->ni_dvp == ndp->ni_vp)
933			vrele(ndp->ni_dvp);
934		else
935			vput(ndp->ni_dvp);
936		vrele(ndp->ni_vp);
937		nfsvno_relpathbuf(ndp);
938		error = EEXIST;
939		goto out;
940	}
941	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
942	    &nvap->na_vattr);
943	vput(ndp->ni_dvp);
944	nfsvno_relpathbuf(ndp);
945
946out:
947	NFSEXITCODE(error);
948	return (error);
949}
950
951/*
952 * symlink vnode op.
953 */
954int
955nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
956    int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
957    struct nfsexstuff *exp)
958{
959	int error = 0;
960
961	if (ndp->ni_vp) {
962		vrele(ndp->ni_startdir);
963		nfsvno_relpathbuf(ndp);
964		if (ndp->ni_dvp == ndp->ni_vp)
965			vrele(ndp->ni_dvp);
966		else
967			vput(ndp->ni_dvp);
968		vrele(ndp->ni_vp);
969		error = EEXIST;
970		goto out;
971	}
972
973	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
974	    &nvap->na_vattr, pathcp);
975	vput(ndp->ni_dvp);
976	vrele(ndp->ni_startdir);
977	nfsvno_relpathbuf(ndp);
978	/*
979	 * Although FreeBSD still had the lookup code in
980	 * it for 7/current, there doesn't seem to be any
981	 * point, since VOP_SYMLINK() returns the ni_vp.
982	 * Just vput it for v2.
983	 */
984	if (!not_v2 && !error)
985		vput(ndp->ni_vp);
986
987out:
988	NFSEXITCODE(error);
989	return (error);
990}
991
992/*
993 * Parse symbolic link arguments.
994 * This function has an ugly side effect. It will MALLOC() an area for
995 * the symlink and set iov_base to point to it, only if it succeeds.
996 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
997 * be FREE'd later.
998 */
999int
1000nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1001    struct thread *p, char **pathcpp, int *lenp)
1002{
1003	u_int32_t *tl;
1004	char *pathcp = NULL;
1005	int error = 0, len;
1006	struct nfsv2_sattr *sp;
1007
1008	*pathcpp = NULL;
1009	*lenp = 0;
1010	if ((nd->nd_flag & ND_NFSV3) &&
1011	    (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
1012		goto nfsmout;
1013	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1014	len = fxdr_unsigned(int, *tl);
1015	if (len > NFS_MAXPATHLEN || len <= 0) {
1016		error = EBADRPC;
1017		goto nfsmout;
1018	}
1019	MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
1020	error = nfsrv_mtostr(nd, pathcp, len);
1021	if (error)
1022		goto nfsmout;
1023	if (nd->nd_flag & ND_NFSV2) {
1024		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1025		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1026	}
1027	*pathcpp = pathcp;
1028	*lenp = len;
1029	NFSEXITCODE2(0, nd);
1030	return (0);
1031nfsmout:
1032	if (pathcp)
1033		free(pathcp, M_TEMP);
1034	NFSEXITCODE2(error, nd);
1035	return (error);
1036}
1037
1038/*
1039 * Remove a non-directory object.
1040 */
1041int
1042nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1043    struct thread *p, struct nfsexstuff *exp)
1044{
1045	struct vnode *vp;
1046	int error = 0;
1047
1048	vp = ndp->ni_vp;
1049	if (vp->v_type == VDIR)
1050		error = NFSERR_ISDIR;
1051	else if (is_v4)
1052		error = nfsrv_checkremove(vp, 1, p);
1053	if (!error)
1054		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1055	if (ndp->ni_dvp == vp)
1056		vrele(ndp->ni_dvp);
1057	else
1058		vput(ndp->ni_dvp);
1059	vput(vp);
1060	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1061		nfsvno_relpathbuf(ndp);
1062	NFSEXITCODE(error);
1063	return (error);
1064}
1065
1066/*
1067 * Remove a directory.
1068 */
1069int
1070nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1071    struct thread *p, struct nfsexstuff *exp)
1072{
1073	struct vnode *vp;
1074	int error = 0;
1075
1076	vp = ndp->ni_vp;
1077	if (vp->v_type != VDIR) {
1078		error = ENOTDIR;
1079		goto out;
1080	}
1081	/*
1082	 * No rmdir "." please.
1083	 */
1084	if (ndp->ni_dvp == vp) {
1085		error = EINVAL;
1086		goto out;
1087	}
1088	/*
1089	 * The root of a mounted filesystem cannot be deleted.
1090	 */
1091	if (vp->v_vflag & VV_ROOT)
1092		error = EBUSY;
1093out:
1094	if (!error)
1095		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1096	if (ndp->ni_dvp == vp)
1097		vrele(ndp->ni_dvp);
1098	else
1099		vput(ndp->ni_dvp);
1100	vput(vp);
1101	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1102		nfsvno_relpathbuf(ndp);
1103	NFSEXITCODE(error);
1104	return (error);
1105}
1106
1107/*
1108 * Rename vnode op.
1109 */
1110int
1111nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1112    u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1113{
1114	struct vnode *fvp, *tvp, *tdvp;
1115	int error = 0;
1116
1117	fvp = fromndp->ni_vp;
1118	if (ndstat) {
1119		vrele(fromndp->ni_dvp);
1120		vrele(fvp);
1121		error = ndstat;
1122		goto out1;
1123	}
1124	tdvp = tondp->ni_dvp;
1125	tvp = tondp->ni_vp;
1126	if (tvp != NULL) {
1127		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1128			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1129			goto out;
1130		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1131			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1132			goto out;
1133		}
1134		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1135			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1136			goto out;
1137		}
1138
1139		/*
1140		 * A rename to '.' or '..' results in a prematurely
1141		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1142		 * here.
1143		 */
1144		if ((tondp->ni_cnd.cn_namelen == 1 &&
1145		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1146		    (tondp->ni_cnd.cn_namelen == 2 &&
1147		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1148		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1149			error = EINVAL;
1150			goto out;
1151		}
1152	}
1153	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1154		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1155		goto out;
1156	}
1157	if (fvp->v_mount != tdvp->v_mount) {
1158		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1159		goto out;
1160	}
1161	if (fvp == tdvp) {
1162		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1163		goto out;
1164	}
1165	if (fvp == tvp) {
1166		/*
1167		 * If source and destination are the same, there is nothing to
1168		 * do. Set error to -1 to indicate this.
1169		 */
1170		error = -1;
1171		goto out;
1172	}
1173	if (ndflag & ND_NFSV4) {
1174		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1175			error = nfsrv_checkremove(fvp, 0, p);
1176			NFSVOPUNLOCK(fvp, 0);
1177		} else
1178			error = EPERM;
1179		if (tvp && !error)
1180			error = nfsrv_checkremove(tvp, 1, p);
1181	} else {
1182		/*
1183		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1184		 * that the NFSv4 client won't be confused by the rename.
1185		 * Since nfsd_recalldelegation() can only be called on an
1186		 * unlocked vnode at this point and fvp is the file that will
1187		 * still exist after the rename, just do fvp.
1188		 */
1189		nfsd_recalldelegation(fvp, p);
1190	}
1191out:
1192	if (!error) {
1193		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1194		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1195		    &tondp->ni_cnd);
1196	} else {
1197		if (tdvp == tvp)
1198			vrele(tdvp);
1199		else
1200			vput(tdvp);
1201		if (tvp)
1202			vput(tvp);
1203		vrele(fromndp->ni_dvp);
1204		vrele(fvp);
1205		if (error == -1)
1206			error = 0;
1207	}
1208	vrele(tondp->ni_startdir);
1209	nfsvno_relpathbuf(tondp);
1210out1:
1211	vrele(fromndp->ni_startdir);
1212	nfsvno_relpathbuf(fromndp);
1213	NFSEXITCODE(error);
1214	return (error);
1215}
1216
1217/*
1218 * Link vnode op.
1219 */
1220int
1221nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1222    struct thread *p, struct nfsexstuff *exp)
1223{
1224	struct vnode *xp;
1225	int error = 0;
1226
1227	xp = ndp->ni_vp;
1228	if (xp != NULL) {
1229		error = EEXIST;
1230	} else {
1231		xp = ndp->ni_dvp;
1232		if (vp->v_mount != xp->v_mount)
1233			error = EXDEV;
1234	}
1235	if (!error) {
1236		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1237		if ((vp->v_iflag & VI_DOOMED) == 0)
1238			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1239		else
1240			error = EPERM;
1241		if (ndp->ni_dvp == vp)
1242			vrele(ndp->ni_dvp);
1243		else
1244			vput(ndp->ni_dvp);
1245		NFSVOPUNLOCK(vp, 0);
1246	} else {
1247		if (ndp->ni_dvp == ndp->ni_vp)
1248			vrele(ndp->ni_dvp);
1249		else
1250			vput(ndp->ni_dvp);
1251		if (ndp->ni_vp)
1252			vrele(ndp->ni_vp);
1253	}
1254	nfsvno_relpathbuf(ndp);
1255	NFSEXITCODE(error);
1256	return (error);
1257}
1258
1259/*
1260 * Do the fsync() appropriate for the commit.
1261 */
1262int
1263nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1264    struct thread *td)
1265{
1266	int error = 0;
1267
1268	/*
1269	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1270	 * file is done.  At this time VOP_FSYNC does not accept offset and
1271	 * byte count parameters so call VOP_FSYNC the whole file for now.
1272	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1273	 */
1274	if (cnt == 0 || cnt > MAX_COMMIT_COUNT) {
1275		/*
1276		 * Give up and do the whole thing
1277		 */
1278		if (vp->v_object &&
1279		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1280			VM_OBJECT_WLOCK(vp->v_object);
1281			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1282			VM_OBJECT_WUNLOCK(vp->v_object);
1283		}
1284		error = VOP_FSYNC(vp, MNT_WAIT, td);
1285	} else {
1286		/*
1287		 * Locate and synchronously write any buffers that fall
1288		 * into the requested range.  Note:  we are assuming that
1289		 * f_iosize is a power of 2.
1290		 */
1291		int iosize = vp->v_mount->mnt_stat.f_iosize;
1292		int iomask = iosize - 1;
1293		struct bufobj *bo;
1294		daddr_t lblkno;
1295
1296		/*
1297		 * Align to iosize boundry, super-align to page boundry.
1298		 */
1299		if (off & iomask) {
1300			cnt += off & iomask;
1301			off &= ~(u_quad_t)iomask;
1302		}
1303		if (off & PAGE_MASK) {
1304			cnt += off & PAGE_MASK;
1305			off &= ~(u_quad_t)PAGE_MASK;
1306		}
1307		lblkno = off / iosize;
1308
1309		if (vp->v_object &&
1310		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1311			VM_OBJECT_WLOCK(vp->v_object);
1312			vm_object_page_clean(vp->v_object, off, off + cnt,
1313			    OBJPC_SYNC);
1314			VM_OBJECT_WUNLOCK(vp->v_object);
1315		}
1316
1317		bo = &vp->v_bufobj;
1318		BO_LOCK(bo);
1319		while (cnt > 0) {
1320			struct buf *bp;
1321
1322			/*
1323			 * If we have a buffer and it is marked B_DELWRI we
1324			 * have to lock and write it.  Otherwise the prior
1325			 * write is assumed to have already been committed.
1326			 *
1327			 * gbincore() can return invalid buffers now so we
1328			 * have to check that bit as well (though B_DELWRI
1329			 * should not be set if B_INVAL is set there could be
1330			 * a race here since we haven't locked the buffer).
1331			 */
1332			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1333				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1334				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1335					BO_LOCK(bo);
1336					continue; /* retry */
1337				}
1338			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1339				    B_DELWRI) {
1340					bremfree(bp);
1341					bp->b_flags &= ~B_ASYNC;
1342					bwrite(bp);
1343					++nfs_commit_miss;
1344				} else
1345					BUF_UNLOCK(bp);
1346				BO_LOCK(bo);
1347			}
1348			++nfs_commit_blks;
1349			if (cnt < iosize)
1350				break;
1351			cnt -= iosize;
1352			++lblkno;
1353		}
1354		BO_UNLOCK(bo);
1355	}
1356	NFSEXITCODE(error);
1357	return (error);
1358}
1359
1360/*
1361 * Statfs vnode op.
1362 */
1363int
1364nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1365{
1366	int error;
1367
1368	error = VFS_STATFS(vp->v_mount, sf);
1369	if (error == 0) {
1370		/*
1371		 * Since NFS handles these values as unsigned on the
1372		 * wire, there is no way to represent negative values,
1373		 * so set them to 0. Without this, they will appear
1374		 * to be very large positive values for clients like
1375		 * Solaris10.
1376		 */
1377		if (sf->f_bavail < 0)
1378			sf->f_bavail = 0;
1379		if (sf->f_ffree < 0)
1380			sf->f_ffree = 0;
1381	}
1382	NFSEXITCODE(error);
1383	return (error);
1384}
1385
1386/*
1387 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1388 * must handle nfsrv_opencheck() calls after any other access checks.
1389 */
1390void
1391nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1392    nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1393    int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1394    NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1395    struct nfsexstuff *exp, struct vnode **vpp)
1396{
1397	struct vnode *vp = NULL;
1398	u_quad_t tempsize;
1399	struct nfsexstuff nes;
1400
1401	if (ndp->ni_vp == NULL)
1402		nd->nd_repstat = nfsrv_opencheck(clientid,
1403		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1404	if (!nd->nd_repstat) {
1405		if (ndp->ni_vp == NULL) {
1406			vrele(ndp->ni_startdir);
1407			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1408			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1409			vput(ndp->ni_dvp);
1410			nfsvno_relpathbuf(ndp);
1411			if (!nd->nd_repstat) {
1412				if (*exclusive_flagp) {
1413					*exclusive_flagp = 0;
1414					NFSVNO_ATTRINIT(nvap);
1415					nvap->na_atime.tv_sec = cverf[0];
1416					nvap->na_atime.tv_nsec = cverf[1];
1417					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1418					    &nvap->na_vattr, cred);
1419				} else {
1420					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1421					    aclp, p, attrbitp, exp);
1422				}
1423			}
1424			vp = ndp->ni_vp;
1425		} else {
1426			if (ndp->ni_startdir)
1427				vrele(ndp->ni_startdir);
1428			nfsvno_relpathbuf(ndp);
1429			vp = ndp->ni_vp;
1430			if (create == NFSV4OPEN_CREATE) {
1431				if (ndp->ni_dvp == vp)
1432					vrele(ndp->ni_dvp);
1433				else
1434					vput(ndp->ni_dvp);
1435			}
1436			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1437				if (ndp->ni_cnd.cn_flags & RDONLY)
1438					NFSVNO_SETEXRDONLY(&nes);
1439				else
1440					NFSVNO_EXINIT(&nes);
1441				nd->nd_repstat = nfsvno_accchk(vp,
1442				    VWRITE, cred, &nes, p,
1443				    NFSACCCHK_NOOVERRIDE,
1444				    NFSACCCHK_VPISLOCKED, NULL);
1445				nd->nd_repstat = nfsrv_opencheck(clientid,
1446				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1447				if (!nd->nd_repstat) {
1448					tempsize = nvap->na_size;
1449					NFSVNO_ATTRINIT(nvap);
1450					nvap->na_size = tempsize;
1451					nd->nd_repstat = VOP_SETATTR(vp,
1452					    &nvap->na_vattr, cred);
1453				}
1454			} else if (vp->v_type == VREG) {
1455				nd->nd_repstat = nfsrv_opencheck(clientid,
1456				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1457			}
1458		}
1459	} else {
1460		if (ndp->ni_cnd.cn_flags & HASBUF)
1461			nfsvno_relpathbuf(ndp);
1462		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1463			vrele(ndp->ni_startdir);
1464			if (ndp->ni_dvp == ndp->ni_vp)
1465				vrele(ndp->ni_dvp);
1466			else
1467				vput(ndp->ni_dvp);
1468			if (ndp->ni_vp)
1469				vput(ndp->ni_vp);
1470		}
1471	}
1472	*vpp = vp;
1473
1474	NFSEXITCODE2(0, nd);
1475}
1476
1477/*
1478 * Updates the file rev and sets the mtime and ctime
1479 * to the current clock time, returning the va_filerev and va_Xtime
1480 * values.
1481 * Return ESTALE to indicate the vnode is VI_DOOMED.
1482 */
1483int
1484nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1485    struct ucred *cred, struct thread *p)
1486{
1487	struct vattr va;
1488
1489	VATTR_NULL(&va);
1490	vfs_timestamp(&va.va_mtime);
1491	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1492		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1493		if ((vp->v_iflag & VI_DOOMED) != 0)
1494			return (ESTALE);
1495	}
1496	(void) VOP_SETATTR(vp, &va, cred);
1497	(void) nfsvno_getattr(vp, nvap, cred, p, 1);
1498	return (0);
1499}
1500
1501/*
1502 * Glue routine to nfsv4_fillattr().
1503 */
1504int
1505nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1506    struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1507    struct ucred *cred, struct thread *p, int isdgram, int reterr,
1508    int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1509{
1510	int error;
1511
1512	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1513	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1514	    mounted_on_fileno);
1515	NFSEXITCODE2(0, nd);
1516	return (error);
1517}
1518
1519/* Since the Readdir vnode ops vary, put the entire functions in here. */
1520/*
1521 * nfs readdir service
1522 * - mallocs what it thinks is enough to read
1523 *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1524 * - calls VOP_READDIR()
1525 * - loops around building the reply
1526 *	if the output generated exceeds count break out of loop
1527 *	The NFSM_CLGET macro is used here so that the reply will be packed
1528 *	tightly in mbuf clusters.
1529 * - it trims out records with d_fileno == 0
1530 *	this doesn't matter for Unix clients, but they might confuse clients
1531 *	for other os'.
1532 * - it trims out records with d_type == DT_WHT
1533 *	these cannot be seen through NFS (unless we extend the protocol)
1534 *     The alternate call nfsrvd_readdirplus() does lookups as well.
1535 * PS: The NFS protocol spec. does not clarify what the "count" byte
1536 *	argument is a count of.. just name strings and file id's or the
1537 *	entire reply rpc or ...
1538 *	I tried just file name and id sizes and it confused the Sun client,
1539 *	so I am using the full rpc size now. The "paranoia.." comment refers
1540 *	to including the status longwords that are not a part of the dir.
1541 *	"entry" structures, but are in the rpc.
1542 */
1543int
1544nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1545    struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1546{
1547	struct dirent *dp;
1548	u_int32_t *tl;
1549	int dirlen;
1550	char *cpos, *cend, *rbuf;
1551	struct nfsvattr at;
1552	int nlen, error = 0, getret = 1;
1553	int siz, cnt, fullsiz, eofflag, ncookies;
1554	u_int64_t off, toff, verf;
1555	u_long *cookies = NULL, *cookiep;
1556	struct uio io;
1557	struct iovec iv;
1558	int is_ufs;
1559
1560	if (nd->nd_repstat) {
1561		nfsrv_postopattr(nd, getret, &at);
1562		goto out;
1563	}
1564	if (nd->nd_flag & ND_NFSV2) {
1565		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1566		off = fxdr_unsigned(u_quad_t, *tl++);
1567	} else {
1568		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1569		off = fxdr_hyper(tl);
1570		tl += 2;
1571		verf = fxdr_hyper(tl);
1572		tl += 2;
1573	}
1574	toff = off;
1575	cnt = fxdr_unsigned(int, *tl);
1576	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1577		cnt = NFS_SRVMAXDATA(nd);
1578	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1579	fullsiz = siz;
1580	if (nd->nd_flag & ND_NFSV3) {
1581		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1582		    p, 1);
1583#if 0
1584		/*
1585		 * va_filerev is not sufficient as a cookie verifier,
1586		 * since it is not supposed to change when entries are
1587		 * removed/added unless that offset cookies returned to
1588		 * the client are no longer valid.
1589		 */
1590		if (!nd->nd_repstat && toff && verf != at.na_filerev)
1591			nd->nd_repstat = NFSERR_BAD_COOKIE;
1592#endif
1593	}
1594	if (!nd->nd_repstat && vp->v_type != VDIR)
1595		nd->nd_repstat = NFSERR_NOTDIR;
1596	if (nd->nd_repstat == 0 && cnt == 0) {
1597		if (nd->nd_flag & ND_NFSV2)
1598			/* NFSv2 does not have NFSERR_TOOSMALL */
1599			nd->nd_repstat = EPERM;
1600		else
1601			nd->nd_repstat = NFSERR_TOOSMALL;
1602	}
1603	if (!nd->nd_repstat)
1604		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1605		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1606		    NFSACCCHK_VPISLOCKED, NULL);
1607	if (nd->nd_repstat) {
1608		vput(vp);
1609		if (nd->nd_flag & ND_NFSV3)
1610			nfsrv_postopattr(nd, getret, &at);
1611		goto out;
1612	}
1613	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1614	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1615again:
1616	eofflag = 0;
1617	if (cookies) {
1618		free((caddr_t)cookies, M_TEMP);
1619		cookies = NULL;
1620	}
1621
1622	iv.iov_base = rbuf;
1623	iv.iov_len = siz;
1624	io.uio_iov = &iv;
1625	io.uio_iovcnt = 1;
1626	io.uio_offset = (off_t)off;
1627	io.uio_resid = siz;
1628	io.uio_segflg = UIO_SYSSPACE;
1629	io.uio_rw = UIO_READ;
1630	io.uio_td = NULL;
1631	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1632	    &cookies);
1633	off = (u_int64_t)io.uio_offset;
1634	if (io.uio_resid)
1635		siz -= io.uio_resid;
1636
1637	if (!cookies && !nd->nd_repstat)
1638		nd->nd_repstat = NFSERR_PERM;
1639	if (nd->nd_flag & ND_NFSV3) {
1640		getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1641		if (!nd->nd_repstat)
1642			nd->nd_repstat = getret;
1643	}
1644
1645	/*
1646	 * Handles the failed cases. nd->nd_repstat == 0 past here.
1647	 */
1648	if (nd->nd_repstat) {
1649		vput(vp);
1650		free((caddr_t)rbuf, M_TEMP);
1651		if (cookies)
1652			free((caddr_t)cookies, M_TEMP);
1653		if (nd->nd_flag & ND_NFSV3)
1654			nfsrv_postopattr(nd, getret, &at);
1655		goto out;
1656	}
1657	/*
1658	 * If nothing read, return eof
1659	 * rpc reply
1660	 */
1661	if (siz == 0) {
1662		vput(vp);
1663		if (nd->nd_flag & ND_NFSV2) {
1664			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1665		} else {
1666			nfsrv_postopattr(nd, getret, &at);
1667			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1668			txdr_hyper(at.na_filerev, tl);
1669			tl += 2;
1670		}
1671		*tl++ = newnfs_false;
1672		*tl = newnfs_true;
1673		FREE((caddr_t)rbuf, M_TEMP);
1674		FREE((caddr_t)cookies, M_TEMP);
1675		goto out;
1676	}
1677
1678	/*
1679	 * Check for degenerate cases of nothing useful read.
1680	 * If so go try again
1681	 */
1682	cpos = rbuf;
1683	cend = rbuf + siz;
1684	dp = (struct dirent *)cpos;
1685	cookiep = cookies;
1686
1687	/*
1688	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1689	 * directory offset up to a block boundary, so it is necessary to
1690	 * skip over the records that precede the requested offset. This
1691	 * requires the assumption that file offset cookies monotonically
1692	 * increase.
1693	 */
1694	while (cpos < cend && ncookies > 0 &&
1695	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1696	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
1697		cpos += dp->d_reclen;
1698		dp = (struct dirent *)cpos;
1699		cookiep++;
1700		ncookies--;
1701	}
1702	if (cpos >= cend || ncookies == 0) {
1703		siz = fullsiz;
1704		toff = off;
1705		goto again;
1706	}
1707	vput(vp);
1708
1709	/*
1710	 * dirlen is the size of the reply, including all XDR and must
1711	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1712	 * if the XDR should be included in "count", but to be safe, we do.
1713	 * (Include the two booleans at the end of the reply in dirlen now.)
1714	 */
1715	if (nd->nd_flag & ND_NFSV3) {
1716		nfsrv_postopattr(nd, getret, &at);
1717		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1718		txdr_hyper(at.na_filerev, tl);
1719		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1720	} else {
1721		dirlen = 2 * NFSX_UNSIGNED;
1722	}
1723
1724	/* Loop through the records and build reply */
1725	while (cpos < cend && ncookies > 0) {
1726		nlen = dp->d_namlen;
1727		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1728			nlen <= NFS_MAXNAMLEN) {
1729			if (nd->nd_flag & ND_NFSV3)
1730				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1731			else
1732				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1733			if (dirlen > cnt) {
1734				eofflag = 0;
1735				break;
1736			}
1737
1738			/*
1739			 * Build the directory record xdr from
1740			 * the dirent entry.
1741			 */
1742			if (nd->nd_flag & ND_NFSV3) {
1743				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1744				*tl++ = newnfs_true;
1745				*tl++ = 0;
1746			} else {
1747				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1748				*tl++ = newnfs_true;
1749			}
1750			*tl = txdr_unsigned(dp->d_fileno);
1751			(void) nfsm_strtom(nd, dp->d_name, nlen);
1752			if (nd->nd_flag & ND_NFSV3) {
1753				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1754				*tl++ = 0;
1755			} else
1756				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1757			*tl = txdr_unsigned(*cookiep);
1758		}
1759		cpos += dp->d_reclen;
1760		dp = (struct dirent *)cpos;
1761		cookiep++;
1762		ncookies--;
1763	}
1764	if (cpos < cend)
1765		eofflag = 0;
1766	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1767	*tl++ = newnfs_false;
1768	if (eofflag)
1769		*tl = newnfs_true;
1770	else
1771		*tl = newnfs_false;
1772	FREE((caddr_t)rbuf, M_TEMP);
1773	FREE((caddr_t)cookies, M_TEMP);
1774
1775out:
1776	NFSEXITCODE2(0, nd);
1777	return (0);
1778nfsmout:
1779	vput(vp);
1780	NFSEXITCODE2(error, nd);
1781	return (error);
1782}
1783
1784/*
1785 * Readdirplus for V3 and Readdir for V4.
1786 */
1787int
1788nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1789    struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1790{
1791	struct dirent *dp;
1792	u_int32_t *tl;
1793	int dirlen;
1794	char *cpos, *cend, *rbuf;
1795	struct vnode *nvp;
1796	fhandle_t nfh;
1797	struct nfsvattr nva, at, *nvap = &nva;
1798	struct mbuf *mb0, *mb1;
1799	struct nfsreferral *refp;
1800	int nlen, r, error = 0, getret = 1, usevget = 1;
1801	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1802	caddr_t bpos0, bpos1;
1803	u_int64_t off, toff, verf;
1804	u_long *cookies = NULL, *cookiep;
1805	nfsattrbit_t attrbits, rderrbits, savbits;
1806	struct uio io;
1807	struct iovec iv;
1808	struct componentname cn;
1809	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
1810	struct mount *mp, *new_mp;
1811	uint64_t mounted_on_fileno;
1812
1813	if (nd->nd_repstat) {
1814		nfsrv_postopattr(nd, getret, &at);
1815		goto out;
1816	}
1817	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1818	off = fxdr_hyper(tl);
1819	toff = off;
1820	tl += 2;
1821	verf = fxdr_hyper(tl);
1822	tl += 2;
1823	siz = fxdr_unsigned(int, *tl++);
1824	cnt = fxdr_unsigned(int, *tl);
1825
1826	/*
1827	 * Use the server's maximum data transfer size as the upper bound
1828	 * on reply datalen.
1829	 */
1830	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1831		cnt = NFS_SRVMAXDATA(nd);
1832
1833	/*
1834	 * siz is a "hint" of how much directory information (name, fileid,
1835	 * cookie) should be in the reply. At least one client "hints" 0,
1836	 * so I set it to cnt for that case. I also round it up to the
1837	 * next multiple of DIRBLKSIZ.
1838	 */
1839	if (siz <= 0)
1840		siz = cnt;
1841	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1842
1843	if (nd->nd_flag & ND_NFSV4) {
1844		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1845		if (error)
1846			goto nfsmout;
1847		NFSSET_ATTRBIT(&savbits, &attrbits);
1848		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1849		NFSZERO_ATTRBIT(&rderrbits);
1850		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1851	} else {
1852		NFSZERO_ATTRBIT(&attrbits);
1853	}
1854	fullsiz = siz;
1855	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1856	if (!nd->nd_repstat) {
1857	    if (off && verf != at.na_filerev) {
1858		/*
1859		 * va_filerev is not sufficient as a cookie verifier,
1860		 * since it is not supposed to change when entries are
1861		 * removed/added unless that offset cookies returned to
1862		 * the client are no longer valid.
1863		 */
1864#if 0
1865		if (nd->nd_flag & ND_NFSV4) {
1866			nd->nd_repstat = NFSERR_NOTSAME;
1867		} else {
1868			nd->nd_repstat = NFSERR_BAD_COOKIE;
1869		}
1870#endif
1871	    } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1872		nd->nd_repstat = NFSERR_BAD_COOKIE;
1873	    }
1874	}
1875	if (!nd->nd_repstat && vp->v_type != VDIR)
1876		nd->nd_repstat = NFSERR_NOTDIR;
1877	if (!nd->nd_repstat && cnt == 0)
1878		nd->nd_repstat = NFSERR_TOOSMALL;
1879	if (!nd->nd_repstat)
1880		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1881		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1882		    NFSACCCHK_VPISLOCKED, NULL);
1883	if (nd->nd_repstat) {
1884		vput(vp);
1885		if (nd->nd_flag & ND_NFSV3)
1886			nfsrv_postopattr(nd, getret, &at);
1887		goto out;
1888	}
1889	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1890	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
1891
1892	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1893again:
1894	eofflag = 0;
1895	if (cookies) {
1896		free((caddr_t)cookies, M_TEMP);
1897		cookies = NULL;
1898	}
1899
1900	iv.iov_base = rbuf;
1901	iv.iov_len = siz;
1902	io.uio_iov = &iv;
1903	io.uio_iovcnt = 1;
1904	io.uio_offset = (off_t)off;
1905	io.uio_resid = siz;
1906	io.uio_segflg = UIO_SYSSPACE;
1907	io.uio_rw = UIO_READ;
1908	io.uio_td = NULL;
1909	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1910	    &cookies);
1911	off = (u_int64_t)io.uio_offset;
1912	if (io.uio_resid)
1913		siz -= io.uio_resid;
1914
1915	getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1916
1917	if (!cookies && !nd->nd_repstat)
1918		nd->nd_repstat = NFSERR_PERM;
1919	if (!nd->nd_repstat)
1920		nd->nd_repstat = getret;
1921	if (nd->nd_repstat) {
1922		vput(vp);
1923		if (cookies)
1924			free((caddr_t)cookies, M_TEMP);
1925		free((caddr_t)rbuf, M_TEMP);
1926		if (nd->nd_flag & ND_NFSV3)
1927			nfsrv_postopattr(nd, getret, &at);
1928		goto out;
1929	}
1930	/*
1931	 * If nothing read, return eof
1932	 * rpc reply
1933	 */
1934	if (siz == 0) {
1935		vput(vp);
1936		if (nd->nd_flag & ND_NFSV3)
1937			nfsrv_postopattr(nd, getret, &at);
1938		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1939		txdr_hyper(at.na_filerev, tl);
1940		tl += 2;
1941		*tl++ = newnfs_false;
1942		*tl = newnfs_true;
1943		free((caddr_t)cookies, M_TEMP);
1944		free((caddr_t)rbuf, M_TEMP);
1945		goto out;
1946	}
1947
1948	/*
1949	 * Check for degenerate cases of nothing useful read.
1950	 * If so go try again
1951	 */
1952	cpos = rbuf;
1953	cend = rbuf + siz;
1954	dp = (struct dirent *)cpos;
1955	cookiep = cookies;
1956
1957	/*
1958	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1959	 * directory offset up to a block boundary, so it is necessary to
1960	 * skip over the records that precede the requested offset. This
1961	 * requires the assumption that file offset cookies monotonically
1962	 * increase.
1963	 */
1964	while (cpos < cend && ncookies > 0 &&
1965	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1966	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
1967	   ((nd->nd_flag & ND_NFSV4) &&
1968	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1969	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1970		cpos += dp->d_reclen;
1971		dp = (struct dirent *)cpos;
1972		cookiep++;
1973		ncookies--;
1974	}
1975	if (cpos >= cend || ncookies == 0) {
1976		siz = fullsiz;
1977		toff = off;
1978		goto again;
1979	}
1980
1981	/*
1982	 * Busy the file system so that the mount point won't go away
1983	 * and, as such, VFS_VGET() can be used safely.
1984	 */
1985	mp = vp->v_mount;
1986	vfs_ref(mp);
1987	NFSVOPUNLOCK(vp, 0);
1988	nd->nd_repstat = vfs_busy(mp, 0);
1989	vfs_rel(mp);
1990	if (nd->nd_repstat != 0) {
1991		vrele(vp);
1992		free(cookies, M_TEMP);
1993		free(rbuf, M_TEMP);
1994		if (nd->nd_flag & ND_NFSV3)
1995			nfsrv_postopattr(nd, getret, &at);
1996		goto out;
1997	}
1998
1999	/*
2000	 * Check to see if entries in this directory can be safely acquired
2001	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2002	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2003	 * automount of the snapshot directory that is required will
2004	 * be done.
2005	 * This needs to be done here for NFSv4, since NFSv4 never does
2006	 * a VFS_VGET() for "." or "..".
2007	 */
2008	if (is_zfs == 1) {
2009		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2010		if (r == EOPNOTSUPP) {
2011			usevget = 0;
2012			cn.cn_nameiop = LOOKUP;
2013			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2014			cn.cn_cred = nd->nd_cred;
2015			cn.cn_thread = p;
2016		} else if (r == 0)
2017			vput(nvp);
2018	}
2019
2020	/*
2021	 * Save this position, in case there is an error before one entry
2022	 * is created.
2023	 */
2024	mb0 = nd->nd_mb;
2025	bpos0 = nd->nd_bpos;
2026
2027	/*
2028	 * Fill in the first part of the reply.
2029	 * dirlen is the reply length in bytes and cannot exceed cnt.
2030	 * (Include the two booleans at the end of the reply in dirlen now,
2031	 *  so we recognize when we have exceeded cnt.)
2032	 */
2033	if (nd->nd_flag & ND_NFSV3) {
2034		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2035		nfsrv_postopattr(nd, getret, &at);
2036	} else {
2037		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2038	}
2039	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2040	txdr_hyper(at.na_filerev, tl);
2041
2042	/*
2043	 * Save this position, in case there is an empty reply needed.
2044	 */
2045	mb1 = nd->nd_mb;
2046	bpos1 = nd->nd_bpos;
2047
2048	/* Loop through the records and build reply */
2049	entrycnt = 0;
2050	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2051		nlen = dp->d_namlen;
2052		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2053		    nlen <= NFS_MAXNAMLEN &&
2054		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2055		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2056		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2057			/*
2058			 * Save the current position in the reply, in case
2059			 * this entry exceeds cnt.
2060			 */
2061			mb1 = nd->nd_mb;
2062			bpos1 = nd->nd_bpos;
2063
2064			/*
2065			 * For readdir_and_lookup get the vnode using
2066			 * the file number.
2067			 */
2068			nvp = NULL;
2069			refp = NULL;
2070			r = 0;
2071			at_root = 0;
2072			needs_unbusy = 0;
2073			new_mp = mp;
2074			mounted_on_fileno = (uint64_t)dp->d_fileno;
2075			if ((nd->nd_flag & ND_NFSV3) ||
2076			    NFSNONZERO_ATTRBIT(&savbits)) {
2077				if (nd->nd_flag & ND_NFSV4)
2078					refp = nfsv4root_getreferral(NULL,
2079					    vp, dp->d_fileno);
2080				if (refp == NULL) {
2081					if (usevget)
2082						r = VFS_VGET(mp, dp->d_fileno,
2083						    LK_SHARED, &nvp);
2084					else
2085						r = EOPNOTSUPP;
2086					if (r == EOPNOTSUPP) {
2087						if (usevget) {
2088							usevget = 0;
2089							cn.cn_nameiop = LOOKUP;
2090							cn.cn_lkflags =
2091							    LK_SHARED |
2092							    LK_RETRY;
2093							cn.cn_cred =
2094							    nd->nd_cred;
2095							cn.cn_thread = p;
2096						}
2097						cn.cn_nameptr = dp->d_name;
2098						cn.cn_namelen = nlen;
2099						cn.cn_flags = ISLASTCN |
2100						    NOFOLLOW | LOCKLEAF;
2101						if (nlen == 2 &&
2102						    dp->d_name[0] == '.' &&
2103						    dp->d_name[1] == '.')
2104							cn.cn_flags |=
2105							    ISDOTDOT;
2106						if (NFSVOPLOCK(vp, LK_SHARED)
2107						    != 0) {
2108							nd->nd_repstat = EPERM;
2109							break;
2110						}
2111						if ((vp->v_vflag & VV_ROOT) != 0
2112						    && (cn.cn_flags & ISDOTDOT)
2113						    != 0) {
2114							vref(vp);
2115							nvp = vp;
2116							r = 0;
2117						} else {
2118							r = VOP_LOOKUP(vp, &nvp,
2119							    &cn);
2120							if (vp != nvp)
2121								NFSVOPUNLOCK(vp,
2122								    0);
2123						}
2124					}
2125
2126					/*
2127					 * For NFSv4, check to see if nvp is
2128					 * a mount point and get the mount
2129					 * point vnode, as required.
2130					 */
2131					if (r == 0 &&
2132					    nfsrv_enable_crossmntpt != 0 &&
2133					    (nd->nd_flag & ND_NFSV4) != 0 &&
2134					    nvp->v_type == VDIR &&
2135					    nvp->v_mountedhere != NULL) {
2136						new_mp = nvp->v_mountedhere;
2137						r = vfs_busy(new_mp, 0);
2138						vput(nvp);
2139						nvp = NULL;
2140						if (r == 0) {
2141							r = VFS_ROOT(new_mp,
2142							    LK_SHARED, &nvp);
2143							needs_unbusy = 1;
2144							if (r == 0)
2145								at_root = 1;
2146						}
2147					}
2148				}
2149				if (!r) {
2150				    if (refp == NULL &&
2151					((nd->nd_flag & ND_NFSV3) ||
2152					 NFSNONZERO_ATTRBIT(&attrbits))) {
2153					r = nfsvno_getfh(nvp, &nfh, p);
2154					if (!r)
2155					    r = nfsvno_getattr(nvp, nvap,
2156						nd->nd_cred, p, 1);
2157					if (r == 0 && is_zfs == 1 &&
2158					    nfsrv_enable_crossmntpt != 0 &&
2159					    (nd->nd_flag & ND_NFSV4) != 0 &&
2160					    nvp->v_type == VDIR &&
2161					    vp->v_mount != nvp->v_mount) {
2162					    /*
2163					     * For a ZFS snapshot, there is a
2164					     * pseudo mount that does not set
2165					     * v_mountedhere, so it needs to
2166					     * be detected via a different
2167					     * mount structure.
2168					     */
2169					    at_root = 1;
2170					    if (new_mp == mp)
2171						new_mp = nvp->v_mount;
2172					}
2173				    }
2174				} else {
2175				    nvp = NULL;
2176				}
2177				if (r) {
2178					if (!NFSISSET_ATTRBIT(&attrbits,
2179					    NFSATTRBIT_RDATTRERROR)) {
2180						if (nvp != NULL)
2181							vput(nvp);
2182						if (needs_unbusy != 0)
2183							vfs_unbusy(new_mp);
2184						nd->nd_repstat = r;
2185						break;
2186					}
2187				}
2188			}
2189
2190			/*
2191			 * Build the directory record xdr
2192			 */
2193			if (nd->nd_flag & ND_NFSV3) {
2194				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2195				*tl++ = newnfs_true;
2196				*tl++ = 0;
2197				*tl = txdr_unsigned(dp->d_fileno);
2198				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2199				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2200				*tl++ = 0;
2201				*tl = txdr_unsigned(*cookiep);
2202				nfsrv_postopattr(nd, 0, nvap);
2203				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2204				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2205				if (nvp != NULL)
2206					vput(nvp);
2207			} else {
2208				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2209				*tl++ = newnfs_true;
2210				*tl++ = 0;
2211				*tl = txdr_unsigned(*cookiep);
2212				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2213				if (nvp != NULL) {
2214					supports_nfsv4acls =
2215					    nfs_supportsnfsv4acls(nvp);
2216					NFSVOPUNLOCK(nvp, 0);
2217				} else
2218					supports_nfsv4acls = 0;
2219				if (refp != NULL) {
2220					dirlen += nfsrv_putreferralattr(nd,
2221					    &savbits, refp, 0,
2222					    &nd->nd_repstat);
2223					if (nd->nd_repstat) {
2224						if (nvp != NULL)
2225							vrele(nvp);
2226						if (needs_unbusy != 0)
2227							vfs_unbusy(new_mp);
2228						break;
2229					}
2230				} else if (r) {
2231					dirlen += nfsvno_fillattr(nd, new_mp,
2232					    nvp, nvap, &nfh, r, &rderrbits,
2233					    nd->nd_cred, p, isdgram, 0,
2234					    supports_nfsv4acls, at_root,
2235					    mounted_on_fileno);
2236				} else {
2237					dirlen += nfsvno_fillattr(nd, new_mp,
2238					    nvp, nvap, &nfh, r, &attrbits,
2239					    nd->nd_cred, p, isdgram, 0,
2240					    supports_nfsv4acls, at_root,
2241					    mounted_on_fileno);
2242				}
2243				if (nvp != NULL)
2244					vrele(nvp);
2245				dirlen += (3 * NFSX_UNSIGNED);
2246			}
2247			if (needs_unbusy != 0)
2248				vfs_unbusy(new_mp);
2249			if (dirlen <= cnt)
2250				entrycnt++;
2251		}
2252		cpos += dp->d_reclen;
2253		dp = (struct dirent *)cpos;
2254		cookiep++;
2255		ncookies--;
2256	}
2257	vrele(vp);
2258	vfs_unbusy(mp);
2259
2260	/*
2261	 * If dirlen > cnt, we must strip off the last entry. If that
2262	 * results in an empty reply, report NFSERR_TOOSMALL.
2263	 */
2264	if (dirlen > cnt || nd->nd_repstat) {
2265		if (!nd->nd_repstat && entrycnt == 0)
2266			nd->nd_repstat = NFSERR_TOOSMALL;
2267		if (nd->nd_repstat) {
2268			newnfs_trimtrailing(nd, mb0, bpos0);
2269			if (nd->nd_flag & ND_NFSV3)
2270				nfsrv_postopattr(nd, getret, &at);
2271		} else
2272			newnfs_trimtrailing(nd, mb1, bpos1);
2273		eofflag = 0;
2274	} else if (cpos < cend)
2275		eofflag = 0;
2276	if (!nd->nd_repstat) {
2277		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2278		*tl++ = newnfs_false;
2279		if (eofflag)
2280			*tl = newnfs_true;
2281		else
2282			*tl = newnfs_false;
2283	}
2284	FREE((caddr_t)cookies, M_TEMP);
2285	FREE((caddr_t)rbuf, M_TEMP);
2286
2287out:
2288	NFSEXITCODE2(0, nd);
2289	return (0);
2290nfsmout:
2291	vput(vp);
2292	NFSEXITCODE2(error, nd);
2293	return (error);
2294}
2295
2296/*
2297 * Get the settable attributes out of the mbuf list.
2298 * (Return 0 or EBADRPC)
2299 */
2300int
2301nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2302    nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2303{
2304	u_int32_t *tl;
2305	struct nfsv2_sattr *sp;
2306	int error = 0, toclient = 0;
2307
2308	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2309	case ND_NFSV2:
2310		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2311		/*
2312		 * Some old clients didn't fill in the high order 16bits.
2313		 * --> check the low order 2 bytes for 0xffff
2314		 */
2315		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2316			nvap->na_mode = nfstov_mode(sp->sa_mode);
2317		if (sp->sa_uid != newnfs_xdrneg1)
2318			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2319		if (sp->sa_gid != newnfs_xdrneg1)
2320			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2321		if (sp->sa_size != newnfs_xdrneg1)
2322			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2323		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2324#ifdef notyet
2325			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2326#else
2327			nvap->na_atime.tv_sec =
2328				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2329			nvap->na_atime.tv_nsec = 0;
2330#endif
2331		}
2332		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2333			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2334		break;
2335	case ND_NFSV3:
2336		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2337		if (*tl == newnfs_true) {
2338			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2339			nvap->na_mode = nfstov_mode(*tl);
2340		}
2341		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2342		if (*tl == newnfs_true) {
2343			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2344			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2345		}
2346		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2347		if (*tl == newnfs_true) {
2348			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2349			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2350		}
2351		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2352		if (*tl == newnfs_true) {
2353			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2354			nvap->na_size = fxdr_hyper(tl);
2355		}
2356		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2357		switch (fxdr_unsigned(int, *tl)) {
2358		case NFSV3SATTRTIME_TOCLIENT:
2359			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2360			fxdr_nfsv3time(tl, &nvap->na_atime);
2361			toclient = 1;
2362			break;
2363		case NFSV3SATTRTIME_TOSERVER:
2364			vfs_timestamp(&nvap->na_atime);
2365			nvap->na_vaflags |= VA_UTIMES_NULL;
2366			break;
2367		};
2368		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2369		switch (fxdr_unsigned(int, *tl)) {
2370		case NFSV3SATTRTIME_TOCLIENT:
2371			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2372			fxdr_nfsv3time(tl, &nvap->na_mtime);
2373			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2374			break;
2375		case NFSV3SATTRTIME_TOSERVER:
2376			vfs_timestamp(&nvap->na_mtime);
2377			if (!toclient)
2378				nvap->na_vaflags |= VA_UTIMES_NULL;
2379			break;
2380		};
2381		break;
2382	case ND_NFSV4:
2383		error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2384	};
2385nfsmout:
2386	NFSEXITCODE2(error, nd);
2387	return (error);
2388}
2389
2390/*
2391 * Handle the setable attributes for V4.
2392 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2393 */
2394int
2395nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2396    nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2397{
2398	u_int32_t *tl;
2399	int attrsum = 0;
2400	int i, j;
2401	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2402	int toclient = 0;
2403	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2404	uid_t uid;
2405	gid_t gid;
2406
2407	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2408	if (error)
2409		goto nfsmout;
2410	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2411	attrsize = fxdr_unsigned(int, *tl);
2412
2413	/*
2414	 * Loop around getting the setable attributes. If an unsupported
2415	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2416	 */
2417	if (retnotsup) {
2418		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2419		bitpos = NFSATTRBIT_MAX;
2420	} else {
2421		bitpos = 0;
2422	}
2423	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2424	    if (attrsum > attrsize) {
2425		error = NFSERR_BADXDR;
2426		goto nfsmout;
2427	    }
2428	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2429		switch (bitpos) {
2430		case NFSATTRBIT_SIZE:
2431			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2432			nvap->na_size = fxdr_hyper(tl);
2433			attrsum += NFSX_HYPER;
2434			break;
2435		case NFSATTRBIT_ACL:
2436			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2437			    p);
2438			if (error)
2439				goto nfsmout;
2440			if (aceerr && !nd->nd_repstat)
2441				nd->nd_repstat = aceerr;
2442			attrsum += aclsize;
2443			break;
2444		case NFSATTRBIT_ARCHIVE:
2445			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2446			if (!nd->nd_repstat)
2447				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2448			attrsum += NFSX_UNSIGNED;
2449			break;
2450		case NFSATTRBIT_HIDDEN:
2451			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2452			if (!nd->nd_repstat)
2453				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2454			attrsum += NFSX_UNSIGNED;
2455			break;
2456		case NFSATTRBIT_MIMETYPE:
2457			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2458			i = fxdr_unsigned(int, *tl);
2459			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2460			if (error)
2461				goto nfsmout;
2462			if (!nd->nd_repstat)
2463				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2464			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2465			break;
2466		case NFSATTRBIT_MODE:
2467			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2468			nvap->na_mode = nfstov_mode(*tl);
2469			attrsum += NFSX_UNSIGNED;
2470			break;
2471		case NFSATTRBIT_OWNER:
2472			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2473			j = fxdr_unsigned(int, *tl);
2474			if (j < 0) {
2475				error = NFSERR_BADXDR;
2476				goto nfsmout;
2477			}
2478			if (j > NFSV4_SMALLSTR)
2479				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2480			else
2481				cp = namestr;
2482			error = nfsrv_mtostr(nd, cp, j);
2483			if (error) {
2484				if (j > NFSV4_SMALLSTR)
2485					free(cp, M_NFSSTRING);
2486				goto nfsmout;
2487			}
2488			if (!nd->nd_repstat) {
2489				nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2490				    p);
2491				if (!nd->nd_repstat)
2492					nvap->na_uid = uid;
2493			}
2494			if (j > NFSV4_SMALLSTR)
2495				free(cp, M_NFSSTRING);
2496			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2497			break;
2498		case NFSATTRBIT_OWNERGROUP:
2499			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2500			j = fxdr_unsigned(int, *tl);
2501			if (j < 0) {
2502				error = NFSERR_BADXDR;
2503				goto nfsmout;
2504			}
2505			if (j > NFSV4_SMALLSTR)
2506				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2507			else
2508				cp = namestr;
2509			error = nfsrv_mtostr(nd, cp, j);
2510			if (error) {
2511				if (j > NFSV4_SMALLSTR)
2512					free(cp, M_NFSSTRING);
2513				goto nfsmout;
2514			}
2515			if (!nd->nd_repstat) {
2516				nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2517				    p);
2518				if (!nd->nd_repstat)
2519					nvap->na_gid = gid;
2520			}
2521			if (j > NFSV4_SMALLSTR)
2522				free(cp, M_NFSSTRING);
2523			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2524			break;
2525		case NFSATTRBIT_SYSTEM:
2526			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2527			if (!nd->nd_repstat)
2528				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2529			attrsum += NFSX_UNSIGNED;
2530			break;
2531		case NFSATTRBIT_TIMEACCESSSET:
2532			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2533			attrsum += NFSX_UNSIGNED;
2534			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2535			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2536			    fxdr_nfsv4time(tl, &nvap->na_atime);
2537			    toclient = 1;
2538			    attrsum += NFSX_V4TIME;
2539			} else {
2540			    vfs_timestamp(&nvap->na_atime);
2541			    nvap->na_vaflags |= VA_UTIMES_NULL;
2542			}
2543			break;
2544		case NFSATTRBIT_TIMEBACKUP:
2545			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2546			if (!nd->nd_repstat)
2547				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2548			attrsum += NFSX_V4TIME;
2549			break;
2550		case NFSATTRBIT_TIMECREATE:
2551			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2552			if (!nd->nd_repstat)
2553				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2554			attrsum += NFSX_V4TIME;
2555			break;
2556		case NFSATTRBIT_TIMEMODIFYSET:
2557			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2558			attrsum += NFSX_UNSIGNED;
2559			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2560			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2561			    fxdr_nfsv4time(tl, &nvap->na_mtime);
2562			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
2563			    attrsum += NFSX_V4TIME;
2564			} else {
2565			    vfs_timestamp(&nvap->na_mtime);
2566			    if (!toclient)
2567				nvap->na_vaflags |= VA_UTIMES_NULL;
2568			}
2569			break;
2570		default:
2571			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2572			/*
2573			 * set bitpos so we drop out of the loop.
2574			 */
2575			bitpos = NFSATTRBIT_MAX;
2576			break;
2577		};
2578	}
2579
2580	/*
2581	 * some clients pad the attrlist, so we need to skip over the
2582	 * padding.
2583	 */
2584	if (attrsum > attrsize) {
2585		error = NFSERR_BADXDR;
2586	} else {
2587		attrsize = NFSM_RNDUP(attrsize);
2588		if (attrsum < attrsize)
2589			error = nfsm_advance(nd, attrsize - attrsum, -1);
2590	}
2591nfsmout:
2592	NFSEXITCODE2(error, nd);
2593	return (error);
2594}
2595
2596/*
2597 * Check/setup export credentials.
2598 */
2599int
2600nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2601    struct ucred *credanon)
2602{
2603	int error = 0;
2604
2605	/*
2606	 * Check/setup credentials.
2607	 */
2608	if (nd->nd_flag & ND_GSS)
2609		exp->nes_exflag &= ~MNT_EXPORTANON;
2610
2611	/*
2612	 * Check to see if the operation is allowed for this security flavor.
2613	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2614	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2615	 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2616	 */
2617	if (nfsvno_testexp(nd, exp) &&
2618	    nd->nd_procnum != NFSV4OP_SECINFO &&
2619	    nd->nd_procnum != NFSPROC_FSINFO) {
2620		if (nd->nd_flag & ND_NFSV4)
2621			error = NFSERR_WRONGSEC;
2622		else
2623			error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2624		goto out;
2625	}
2626
2627	/*
2628	 * Check to see if the file system is exported V4 only.
2629	 */
2630	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2631		error = NFSERR_PROGNOTV4;
2632		goto out;
2633	}
2634
2635	/*
2636	 * Now, map the user credentials.
2637	 * (Note that ND_AUTHNONE will only be set for an NFSv3
2638	 *  Fsinfo RPC. If set for anything else, this code might need
2639	 *  to change.)
2640	 */
2641	if (NFSVNO_EXPORTED(exp) &&
2642	    ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2643	     NFSVNO_EXPORTANON(exp) ||
2644	     (nd->nd_flag & ND_AUTHNONE))) {
2645		nd->nd_cred->cr_uid = credanon->cr_uid;
2646		nd->nd_cred->cr_gid = credanon->cr_gid;
2647		crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2648		    credanon->cr_groups);
2649	}
2650
2651out:
2652	NFSEXITCODE2(error, nd);
2653	return (error);
2654}
2655
2656/*
2657 * Check exports.
2658 */
2659int
2660nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2661    struct ucred **credp)
2662{
2663	int i, error, *secflavors;
2664
2665	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2666	    &exp->nes_numsecflavor, &secflavors);
2667	if (error) {
2668		if (nfs_rootfhset) {
2669			exp->nes_exflag = 0;
2670			exp->nes_numsecflavor = 0;
2671			error = 0;
2672		}
2673	} else {
2674		/* Copy the security flavors. */
2675		for (i = 0; i < exp->nes_numsecflavor; i++)
2676			exp->nes_secflavors[i] = secflavors[i];
2677	}
2678	NFSEXITCODE(error);
2679	return (error);
2680}
2681
2682/*
2683 * Get a vnode for a file handle and export stuff.
2684 */
2685int
2686nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2687    int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2688    struct ucred **credp)
2689{
2690	int i, error, *secflavors;
2691
2692	*credp = NULL;
2693	exp->nes_numsecflavor = 0;
2694	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2695	if (error != 0)
2696		/* Make sure the server replies ESTALE to the client. */
2697		error = ESTALE;
2698	if (nam && !error) {
2699		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2700		    &exp->nes_numsecflavor, &secflavors);
2701		if (error) {
2702			if (nfs_rootfhset) {
2703				exp->nes_exflag = 0;
2704				exp->nes_numsecflavor = 0;
2705				error = 0;
2706			} else {
2707				vput(*vpp);
2708			}
2709		} else {
2710			/* Copy the security flavors. */
2711			for (i = 0; i < exp->nes_numsecflavor; i++)
2712				exp->nes_secflavors[i] = secflavors[i];
2713		}
2714	}
2715	NFSEXITCODE(error);
2716	return (error);
2717}
2718
2719/*
2720 * nfsd_fhtovp() - convert a fh to a vnode ptr
2721 * 	- look up fsid in mount list (if not found ret error)
2722 *	- get vp and export rights by calling nfsvno_fhtovp()
2723 *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2724 *	  for AUTH_SYS
2725 *	- if mpp != NULL, return the mount point so that it can
2726 *	  be used for vn_finished_write() by the caller
2727 */
2728void
2729nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2730    struct vnode **vpp, struct nfsexstuff *exp,
2731    struct mount **mpp, int startwrite, struct thread *p)
2732{
2733	struct mount *mp;
2734	struct ucred *credanon;
2735	fhandle_t *fhp;
2736
2737	fhp = (fhandle_t *)nfp->nfsrvfh_data;
2738	/*
2739	 * Check for the special case of the nfsv4root_fh.
2740	 */
2741	mp = vfs_busyfs(&fhp->fh_fsid);
2742	if (mpp != NULL)
2743		*mpp = mp;
2744	if (mp == NULL) {
2745		*vpp = NULL;
2746		nd->nd_repstat = ESTALE;
2747		goto out;
2748	}
2749
2750	if (startwrite) {
2751		vn_start_write(NULL, mpp, V_WAIT);
2752		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
2753			lktype = LK_EXCLUSIVE;
2754	}
2755	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2756	    &credanon);
2757	vfs_unbusy(mp);
2758
2759	/*
2760	 * For NFSv4 without a pseudo root fs, unexported file handles
2761	 * can be returned, so that Lookup works everywhere.
2762	 */
2763	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2764	    !(nd->nd_flag & ND_NFSV4)) {
2765		vput(*vpp);
2766		nd->nd_repstat = EACCES;
2767	}
2768
2769	/*
2770	 * Personally, I've never seen any point in requiring a
2771	 * reserved port#, since only in the rare case where the
2772	 * clients are all boxes with secure system priviledges,
2773	 * does it provide any enhanced security, but... some people
2774	 * believe it to be useful and keep putting this code back in.
2775	 * (There is also some "security checker" out there that
2776	 *  complains if the nfs server doesn't enforce this.)
2777	 * However, note the following:
2778	 * RFC3530 (NFSv4) specifies that a reserved port# not be
2779	 *	required.
2780	 * RFC2623 recommends that, if a reserved port# is checked for,
2781	 *	that there be a way to turn that off--> ifdef'd.
2782	 */
2783#ifdef NFS_REQRSVPORT
2784	if (!nd->nd_repstat) {
2785		struct sockaddr_in *saddr;
2786		struct sockaddr_in6 *saddr6;
2787
2788		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2789		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2790		if (!(nd->nd_flag & ND_NFSV4) &&
2791		    ((saddr->sin_family == AF_INET &&
2792		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2793		     (saddr6->sin6_family == AF_INET6 &&
2794		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2795			vput(*vpp);
2796			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2797		}
2798	}
2799#endif	/* NFS_REQRSVPORT */
2800
2801	/*
2802	 * Check/setup credentials.
2803	 */
2804	if (!nd->nd_repstat) {
2805		nd->nd_saveduid = nd->nd_cred->cr_uid;
2806		nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2807		if (nd->nd_repstat)
2808			vput(*vpp);
2809	}
2810	if (credanon != NULL)
2811		crfree(credanon);
2812	if (nd->nd_repstat) {
2813		if (startwrite)
2814			vn_finished_write(mp);
2815		*vpp = NULL;
2816		if (mpp != NULL)
2817			*mpp = NULL;
2818	}
2819
2820out:
2821	NFSEXITCODE2(0, nd);
2822}
2823
2824/*
2825 * glue for fp.
2826 */
2827static int
2828fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2829{
2830	struct filedesc *fdp;
2831	struct file *fp;
2832	int error = 0;
2833
2834	fdp = p->td_proc->p_fd;
2835	if (fd < 0 || fd >= fdp->fd_nfiles ||
2836	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
2837		error = EBADF;
2838		goto out;
2839	}
2840	*fpp = fp;
2841
2842out:
2843	NFSEXITCODE(error);
2844	return (error);
2845}
2846
2847/*
2848 * Called from nfssvc() to update the exports list. Just call
2849 * vfs_export(). This has to be done, since the v4 root fake fs isn't
2850 * in the mount list.
2851 */
2852int
2853nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2854{
2855	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2856	int error = 0;
2857	struct nameidata nd;
2858	fhandle_t fh;
2859
2860	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2861	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2862		nfs_rootfhset = 0;
2863	else if (error == 0) {
2864		if (nfsexargp->fspec == NULL) {
2865			error = EPERM;
2866			goto out;
2867		}
2868		/*
2869		 * If fspec != NULL, this is the v4root path.
2870		 */
2871		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
2872		    nfsexargp->fspec, p);
2873		if ((error = namei(&nd)) != 0)
2874			goto out;
2875		error = nfsvno_getfh(nd.ni_vp, &fh, p);
2876		vrele(nd.ni_vp);
2877		if (!error) {
2878			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2879			NFSBCOPY((caddr_t)&fh,
2880			    nfs_rootfh.nfsrvfh_data,
2881			    sizeof (fhandle_t));
2882			nfs_rootfhset = 1;
2883		}
2884	}
2885
2886out:
2887	NFSEXITCODE(error);
2888	return (error);
2889}
2890
2891/*
2892 * This function needs to test to see if the system is near its limit
2893 * for memory allocation via malloc() or mget() and return True iff
2894 * either of these resources are near their limit.
2895 * XXX (For now, this is just a stub.)
2896 */
2897int nfsrv_testmalloclimit = 0;
2898int
2899nfsrv_mallocmget_limit(void)
2900{
2901	static int printmesg = 0;
2902	static int testval = 1;
2903
2904	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2905		if ((printmesg++ % 100) == 0)
2906			printf("nfsd: malloc/mget near limit\n");
2907		return (1);
2908	}
2909	return (0);
2910}
2911
2912/*
2913 * BSD specific initialization of a mount point.
2914 */
2915void
2916nfsd_mntinit(void)
2917{
2918	static int inited = 0;
2919
2920	if (inited)
2921		return;
2922	inited = 1;
2923	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2924	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2925	TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2926	nfsv4root_mnt.mnt_export = NULL;
2927	TAILQ_INIT(&nfsv4root_opt);
2928	TAILQ_INIT(&nfsv4root_newopt);
2929	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2930	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2931	nfsv4root_mnt.mnt_nvnodelistsize = 0;
2932	nfsv4root_mnt.mnt_activevnodelistsize = 0;
2933}
2934
2935/*
2936 * Get a vnode for a file handle, without checking exports, etc.
2937 */
2938struct vnode *
2939nfsvno_getvp(fhandle_t *fhp)
2940{
2941	struct mount *mp;
2942	struct vnode *vp;
2943	int error;
2944
2945	mp = vfs_busyfs(&fhp->fh_fsid);
2946	if (mp == NULL)
2947		return (NULL);
2948	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2949	vfs_unbusy(mp);
2950	if (error)
2951		return (NULL);
2952	return (vp);
2953}
2954
2955/*
2956 * Do a local VOP_ADVLOCK().
2957 */
2958int
2959nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2960    u_int64_t end, struct thread *td)
2961{
2962	int error = 0;
2963	struct flock fl;
2964	u_int64_t tlen;
2965
2966	if (nfsrv_dolocallocks == 0)
2967		goto out;
2968	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
2969
2970	fl.l_whence = SEEK_SET;
2971	fl.l_type = ftype;
2972	fl.l_start = (off_t)first;
2973	if (end == NFS64BITSSET) {
2974		fl.l_len = 0;
2975	} else {
2976		tlen = end - first;
2977		fl.l_len = (off_t)tlen;
2978	}
2979	/*
2980	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2981	 * values for all calls, so that all locks will be held by the
2982	 * nfsd server. (The nfsd server handles conflicts between the
2983	 * various clients.)
2984	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2985	 * bytes, so it can't be put in l_sysid.
2986	 */
2987	if (nfsv4_sysid == 0)
2988		nfsv4_sysid = nlm_acquire_next_sysid();
2989	fl.l_pid = (pid_t)0;
2990	fl.l_sysid = (int)nfsv4_sysid;
2991
2992	if (ftype == F_UNLCK)
2993		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
2994		    (F_POSIX | F_REMOTE));
2995	else
2996		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
2997		    (F_POSIX | F_REMOTE));
2998
2999out:
3000	NFSEXITCODE(error);
3001	return (error);
3002}
3003
3004/*
3005 * Check the nfsv4 root exports.
3006 */
3007int
3008nfsvno_v4rootexport(struct nfsrv_descript *nd)
3009{
3010	struct ucred *credanon;
3011	int exflags, error = 0, numsecflavor, *secflavors, i;
3012
3013	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3014	    &credanon, &numsecflavor, &secflavors);
3015	if (error) {
3016		error = NFSERR_PROGUNAVAIL;
3017		goto out;
3018	}
3019	if (credanon != NULL)
3020		crfree(credanon);
3021	for (i = 0; i < numsecflavor; i++) {
3022		if (secflavors[i] == AUTH_SYS)
3023			nd->nd_flag |= ND_EXAUTHSYS;
3024		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3025			nd->nd_flag |= ND_EXGSS;
3026		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3027			nd->nd_flag |= ND_EXGSSINTEGRITY;
3028		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3029			nd->nd_flag |= ND_EXGSSPRIVACY;
3030	}
3031
3032out:
3033	NFSEXITCODE(error);
3034	return (error);
3035}
3036
3037/*
3038 * Nfs server psuedo system call for the nfsd's
3039 */
3040/*
3041 * MPSAFE
3042 */
3043static int
3044nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3045{
3046	struct file *fp;
3047	struct nfsd_addsock_args sockarg;
3048	struct nfsd_nfsd_args nfsdarg;
3049	cap_rights_t rights;
3050	int error;
3051
3052	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3053		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3054		if (error)
3055			goto out;
3056		/*
3057		 * Since we don't know what rights might be required,
3058		 * pretend that we need them all. It is better to be too
3059		 * careful than too reckless.
3060		 */
3061		error = fget(td, sockarg.sock,
3062		    cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3063		if (error != 0)
3064			goto out;
3065		if (fp->f_type != DTYPE_SOCKET) {
3066			fdrop(fp, td);
3067			error = EPERM;
3068			goto out;
3069		}
3070		error = nfsrvd_addsock(fp);
3071		fdrop(fp, td);
3072	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3073		if (uap->argp == NULL) {
3074			error = EINVAL;
3075			goto out;
3076		}
3077		error = copyin(uap->argp, (caddr_t)&nfsdarg,
3078		    sizeof (nfsdarg));
3079		if (error)
3080			goto out;
3081		error = nfsrvd_nfsd(td, &nfsdarg);
3082	} else {
3083		error = nfssvc_srvcall(td, uap, td->td_ucred);
3084	}
3085
3086out:
3087	NFSEXITCODE(error);
3088	return (error);
3089}
3090
3091static int
3092nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3093{
3094	struct nfsex_args export;
3095	struct file *fp = NULL;
3096	int stablefd, len;
3097	struct nfsd_clid adminrevoke;
3098	struct nfsd_dumplist dumplist;
3099	struct nfsd_dumpclients *dumpclients;
3100	struct nfsd_dumplocklist dumplocklist;
3101	struct nfsd_dumplocks *dumplocks;
3102	struct nameidata nd;
3103	vnode_t vp;
3104	int error = EINVAL, igotlock;
3105	struct proc *procp;
3106	static int suspend_nfsd = 0;
3107
3108	if (uap->flag & NFSSVC_PUBLICFH) {
3109		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3110		    sizeof (fhandle_t));
3111		error = copyin(uap->argp,
3112		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3113		if (!error)
3114			nfs_pubfhset = 1;
3115	} else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3116		error = copyin(uap->argp,(caddr_t)&export,
3117		    sizeof (struct nfsex_args));
3118		if (!error)
3119			error = nfsrv_v4rootexport(&export, cred, p);
3120	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3121		nfs_pubfhset = 0;
3122		error = 0;
3123	} else if (uap->flag & NFSSVC_STABLERESTART) {
3124		error = copyin(uap->argp, (caddr_t)&stablefd,
3125		    sizeof (int));
3126		if (!error)
3127			error = fp_getfvp(p, stablefd, &fp, &vp);
3128		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3129			error = EBADF;
3130		if (!error && newnfs_numnfsd != 0)
3131			error = EPERM;
3132		if (!error) {
3133			nfsrv_stablefirst.nsf_fp = fp;
3134			nfsrv_setupstable(p);
3135		}
3136	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3137		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3138		    sizeof (struct nfsd_clid));
3139		if (!error)
3140			error = nfsrv_adminrevoke(&adminrevoke, p);
3141	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3142		error = copyin(uap->argp, (caddr_t)&dumplist,
3143		    sizeof (struct nfsd_dumplist));
3144		if (!error && (dumplist.ndl_size < 1 ||
3145			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3146			error = EPERM;
3147		if (!error) {
3148		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3149		    dumpclients = (struct nfsd_dumpclients *)malloc(len,
3150			M_TEMP, M_WAITOK);
3151		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3152		    error = copyout(dumpclients,
3153			CAST_USER_ADDR_T(dumplist.ndl_list), len);
3154		    free((caddr_t)dumpclients, M_TEMP);
3155		}
3156	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3157		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3158		    sizeof (struct nfsd_dumplocklist));
3159		if (!error && (dumplocklist.ndllck_size < 1 ||
3160			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3161			error = EPERM;
3162		if (!error)
3163			error = nfsrv_lookupfilename(&nd,
3164				dumplocklist.ndllck_fname, p);
3165		if (!error) {
3166			len = sizeof (struct nfsd_dumplocks) *
3167				dumplocklist.ndllck_size;
3168			dumplocks = (struct nfsd_dumplocks *)malloc(len,
3169				M_TEMP, M_WAITOK);
3170			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3171			    dumplocklist.ndllck_size, p);
3172			vput(nd.ni_vp);
3173			error = copyout(dumplocks,
3174			    CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3175			free((caddr_t)dumplocks, M_TEMP);
3176		}
3177	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3178		procp = p->td_proc;
3179		PROC_LOCK(procp);
3180		nfsd_master_pid = procp->p_pid;
3181		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3182		nfsd_master_start = procp->p_stats->p_start;
3183		nfsd_master_proc = procp;
3184		PROC_UNLOCK(procp);
3185	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3186		NFSLOCKV4ROOTMUTEX();
3187		if (suspend_nfsd == 0) {
3188			/* Lock out all nfsd threads */
3189			do {
3190				igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3191				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3192			} while (igotlock == 0 && suspend_nfsd == 0);
3193			suspend_nfsd = 1;
3194		}
3195		NFSUNLOCKV4ROOTMUTEX();
3196		error = 0;
3197	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3198		NFSLOCKV4ROOTMUTEX();
3199		if (suspend_nfsd != 0) {
3200			nfsv4_unlock(&nfsd_suspend_lock, 0);
3201			suspend_nfsd = 0;
3202		}
3203		NFSUNLOCKV4ROOTMUTEX();
3204		error = 0;
3205	}
3206
3207	NFSEXITCODE(error);
3208	return (error);
3209}
3210
3211/*
3212 * Check exports.
3213 * Returns 0 if ok, 1 otherwise.
3214 */
3215int
3216nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3217{
3218	int i;
3219
3220	/*
3221	 * This seems odd, but allow the case where the security flavor
3222	 * list is empty. This happens when NFSv4 is traversing non-exported
3223	 * file systems. Exported file systems should always have a non-empty
3224	 * security flavor list.
3225	 */
3226	if (exp->nes_numsecflavor == 0)
3227		return (0);
3228
3229	for (i = 0; i < exp->nes_numsecflavor; i++) {
3230		/*
3231		 * The tests for privacy and integrity must be first,
3232		 * since ND_GSS is set for everything but AUTH_SYS.
3233		 */
3234		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3235		    (nd->nd_flag & ND_GSSPRIVACY))
3236			return (0);
3237		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3238		    (nd->nd_flag & ND_GSSINTEGRITY))
3239			return (0);
3240		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3241		    (nd->nd_flag & ND_GSS))
3242			return (0);
3243		if (exp->nes_secflavors[i] == AUTH_SYS &&
3244		    (nd->nd_flag & ND_GSS) == 0)
3245			return (0);
3246	}
3247	return (1);
3248}
3249
3250/*
3251 * Calculate a hash value for the fid in a file handle.
3252 */
3253uint32_t
3254nfsrv_hashfh(fhandle_t *fhp)
3255{
3256	uint32_t hashval;
3257
3258	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3259	return (hashval);
3260}
3261
3262/*
3263 * Calculate a hash value for the sessionid.
3264 */
3265uint32_t
3266nfsrv_hashsessionid(uint8_t *sessionid)
3267{
3268	uint32_t hashval;
3269
3270	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
3271	return (hashval);
3272}
3273
3274/*
3275 * Signal the userland master nfsd to backup the stable restart file.
3276 */
3277void
3278nfsrv_backupstable(void)
3279{
3280	struct proc *procp;
3281
3282	if (nfsd_master_proc != NULL) {
3283		procp = pfind(nfsd_master_pid);
3284		/* Try to make sure it is the correct process. */
3285		if (procp == nfsd_master_proc &&
3286		    procp->p_stats->p_start.tv_sec ==
3287		    nfsd_master_start.tv_sec &&
3288		    procp->p_stats->p_start.tv_usec ==
3289		    nfsd_master_start.tv_usec &&
3290		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
3291			kern_psignal(procp, SIGUSR2);
3292		else
3293			nfsd_master_proc = NULL;
3294
3295		if (procp != NULL)
3296			PROC_UNLOCK(procp);
3297	}
3298}
3299
3300extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3301
3302/*
3303 * Called once to initialize data structures...
3304 */
3305static int
3306nfsd_modevent(module_t mod, int type, void *data)
3307{
3308	int error = 0, i;
3309	static int loaded = 0;
3310
3311	switch (type) {
3312	case MOD_LOAD:
3313		if (loaded)
3314			goto out;
3315		newnfs_portinit();
3316		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3317			mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
3318			    MTX_DEF);
3319			mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
3320			    MTX_DEF);
3321		}
3322		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
3323		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
3324		mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
3325		for (i = 0; i < NFSSESSIONHASHSIZE; i++)
3326			mtx_init(&nfssessionhash[i].mtx, "nfssm",
3327			    NULL, MTX_DEF);
3328		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3329		nfsrvd_initcache();
3330		nfsd_init();
3331		NFSD_LOCK();
3332		nfsrvd_init(0);
3333		NFSD_UNLOCK();
3334		nfsd_mntinit();
3335#ifdef VV_DISABLEDELEG
3336		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3337		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3338#endif
3339		nfsd_call_servertimer = nfsrv_servertimer;
3340		nfsd_call_nfsd = nfssvc_nfsd;
3341		loaded = 1;
3342		break;
3343
3344	case MOD_UNLOAD:
3345		if (newnfs_numnfsd != 0) {
3346			error = EBUSY;
3347			break;
3348		}
3349
3350#ifdef VV_DISABLEDELEG
3351		vn_deleg_ops.vndeleg_recall = NULL;
3352		vn_deleg_ops.vndeleg_disable = NULL;
3353#endif
3354		nfsd_call_servertimer = NULL;
3355		nfsd_call_nfsd = NULL;
3356
3357		/* Clean out all NFSv4 state. */
3358		nfsrv_throwawayallstate(curthread);
3359
3360		/* Clean the NFS server reply cache */
3361		nfsrvd_cleancache();
3362
3363		/* Free up the krpc server pool. */
3364		if (nfsrvd_pool != NULL)
3365			svcpool_destroy(nfsrvd_pool);
3366
3367		/* and get rid of the locks */
3368		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3369			mtx_destroy(&nfsrchash_table[i].mtx);
3370			mtx_destroy(&nfsrcahash_table[i].mtx);
3371		}
3372		mtx_destroy(&nfsrc_udpmtx);
3373		mtx_destroy(&nfs_v4root_mutex);
3374		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3375		for (i = 0; i < NFSSESSIONHASHSIZE; i++)
3376			mtx_destroy(&nfssessionhash[i].mtx);
3377		lockdestroy(&nfsv4root_mnt.mnt_explock);
3378		loaded = 0;
3379		break;
3380	default:
3381		error = EOPNOTSUPP;
3382		break;
3383	}
3384
3385out:
3386	NFSEXITCODE(error);
3387	return (error);
3388}
3389static moduledata_t nfsd_mod = {
3390	"nfsd",
3391	nfsd_modevent,
3392	NULL,
3393};
3394DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3395
3396/* So that loader and kldload(2) can find us, wherever we are.. */
3397MODULE_VERSION(nfsd, 1);
3398MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3399MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3400MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3401MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3402MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3403
3404