nfs_nfsdport.c revision 282270
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 282270 2015-04-30 12:39:24Z rmacklem $");
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	 * File systems that do not use the buffer cache (as indicated
1274	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1275	 */
1276	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1277	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1278		/*
1279		 * Give up and do the whole thing
1280		 */
1281		if (vp->v_object &&
1282		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1283			VM_OBJECT_WLOCK(vp->v_object);
1284			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1285			VM_OBJECT_WUNLOCK(vp->v_object);
1286		}
1287		error = VOP_FSYNC(vp, MNT_WAIT, td);
1288	} else {
1289		/*
1290		 * Locate and synchronously write any buffers that fall
1291		 * into the requested range.  Note:  we are assuming that
1292		 * f_iosize is a power of 2.
1293		 */
1294		int iosize = vp->v_mount->mnt_stat.f_iosize;
1295		int iomask = iosize - 1;
1296		struct bufobj *bo;
1297		daddr_t lblkno;
1298
1299		/*
1300		 * Align to iosize boundry, super-align to page boundry.
1301		 */
1302		if (off & iomask) {
1303			cnt += off & iomask;
1304			off &= ~(u_quad_t)iomask;
1305		}
1306		if (off & PAGE_MASK) {
1307			cnt += off & PAGE_MASK;
1308			off &= ~(u_quad_t)PAGE_MASK;
1309		}
1310		lblkno = off / iosize;
1311
1312		if (vp->v_object &&
1313		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1314			VM_OBJECT_WLOCK(vp->v_object);
1315			vm_object_page_clean(vp->v_object, off, off + cnt,
1316			    OBJPC_SYNC);
1317			VM_OBJECT_WUNLOCK(vp->v_object);
1318		}
1319
1320		bo = &vp->v_bufobj;
1321		BO_LOCK(bo);
1322		while (cnt > 0) {
1323			struct buf *bp;
1324
1325			/*
1326			 * If we have a buffer and it is marked B_DELWRI we
1327			 * have to lock and write it.  Otherwise the prior
1328			 * write is assumed to have already been committed.
1329			 *
1330			 * gbincore() can return invalid buffers now so we
1331			 * have to check that bit as well (though B_DELWRI
1332			 * should not be set if B_INVAL is set there could be
1333			 * a race here since we haven't locked the buffer).
1334			 */
1335			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1336				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1337				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1338					BO_LOCK(bo);
1339					continue; /* retry */
1340				}
1341			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1342				    B_DELWRI) {
1343					bremfree(bp);
1344					bp->b_flags &= ~B_ASYNC;
1345					bwrite(bp);
1346					++nfs_commit_miss;
1347				} else
1348					BUF_UNLOCK(bp);
1349				BO_LOCK(bo);
1350			}
1351			++nfs_commit_blks;
1352			if (cnt < iosize)
1353				break;
1354			cnt -= iosize;
1355			++lblkno;
1356		}
1357		BO_UNLOCK(bo);
1358	}
1359	NFSEXITCODE(error);
1360	return (error);
1361}
1362
1363/*
1364 * Statfs vnode op.
1365 */
1366int
1367nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1368{
1369	int error;
1370
1371	error = VFS_STATFS(vp->v_mount, sf);
1372	if (error == 0) {
1373		/*
1374		 * Since NFS handles these values as unsigned on the
1375		 * wire, there is no way to represent negative values,
1376		 * so set them to 0. Without this, they will appear
1377		 * to be very large positive values for clients like
1378		 * Solaris10.
1379		 */
1380		if (sf->f_bavail < 0)
1381			sf->f_bavail = 0;
1382		if (sf->f_ffree < 0)
1383			sf->f_ffree = 0;
1384	}
1385	NFSEXITCODE(error);
1386	return (error);
1387}
1388
1389/*
1390 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1391 * must handle nfsrv_opencheck() calls after any other access checks.
1392 */
1393void
1394nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1395    nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1396    int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1397    NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1398    struct nfsexstuff *exp, struct vnode **vpp)
1399{
1400	struct vnode *vp = NULL;
1401	u_quad_t tempsize;
1402	struct nfsexstuff nes;
1403
1404	if (ndp->ni_vp == NULL)
1405		nd->nd_repstat = nfsrv_opencheck(clientid,
1406		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1407	if (!nd->nd_repstat) {
1408		if (ndp->ni_vp == NULL) {
1409			vrele(ndp->ni_startdir);
1410			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1411			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1412			vput(ndp->ni_dvp);
1413			nfsvno_relpathbuf(ndp);
1414			if (!nd->nd_repstat) {
1415				if (*exclusive_flagp) {
1416					*exclusive_flagp = 0;
1417					NFSVNO_ATTRINIT(nvap);
1418					nvap->na_atime.tv_sec = cverf[0];
1419					nvap->na_atime.tv_nsec = cverf[1];
1420					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1421					    &nvap->na_vattr, cred);
1422				} else {
1423					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1424					    aclp, p, attrbitp, exp);
1425				}
1426			}
1427			vp = ndp->ni_vp;
1428		} else {
1429			if (ndp->ni_startdir)
1430				vrele(ndp->ni_startdir);
1431			nfsvno_relpathbuf(ndp);
1432			vp = ndp->ni_vp;
1433			if (create == NFSV4OPEN_CREATE) {
1434				if (ndp->ni_dvp == vp)
1435					vrele(ndp->ni_dvp);
1436				else
1437					vput(ndp->ni_dvp);
1438			}
1439			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1440				if (ndp->ni_cnd.cn_flags & RDONLY)
1441					NFSVNO_SETEXRDONLY(&nes);
1442				else
1443					NFSVNO_EXINIT(&nes);
1444				nd->nd_repstat = nfsvno_accchk(vp,
1445				    VWRITE, cred, &nes, p,
1446				    NFSACCCHK_NOOVERRIDE,
1447				    NFSACCCHK_VPISLOCKED, NULL);
1448				nd->nd_repstat = nfsrv_opencheck(clientid,
1449				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1450				if (!nd->nd_repstat) {
1451					tempsize = nvap->na_size;
1452					NFSVNO_ATTRINIT(nvap);
1453					nvap->na_size = tempsize;
1454					nd->nd_repstat = VOP_SETATTR(vp,
1455					    &nvap->na_vattr, cred);
1456				}
1457			} else if (vp->v_type == VREG) {
1458				nd->nd_repstat = nfsrv_opencheck(clientid,
1459				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1460			}
1461		}
1462	} else {
1463		if (ndp->ni_cnd.cn_flags & HASBUF)
1464			nfsvno_relpathbuf(ndp);
1465		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1466			vrele(ndp->ni_startdir);
1467			if (ndp->ni_dvp == ndp->ni_vp)
1468				vrele(ndp->ni_dvp);
1469			else
1470				vput(ndp->ni_dvp);
1471			if (ndp->ni_vp)
1472				vput(ndp->ni_vp);
1473		}
1474	}
1475	*vpp = vp;
1476
1477	NFSEXITCODE2(0, nd);
1478}
1479
1480/*
1481 * Updates the file rev and sets the mtime and ctime
1482 * to the current clock time, returning the va_filerev and va_Xtime
1483 * values.
1484 * Return ESTALE to indicate the vnode is VI_DOOMED.
1485 */
1486int
1487nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1488    struct ucred *cred, struct thread *p)
1489{
1490	struct vattr va;
1491
1492	VATTR_NULL(&va);
1493	vfs_timestamp(&va.va_mtime);
1494	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1495		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1496		if ((vp->v_iflag & VI_DOOMED) != 0)
1497			return (ESTALE);
1498	}
1499	(void) VOP_SETATTR(vp, &va, cred);
1500	(void) nfsvno_getattr(vp, nvap, cred, p, 1);
1501	return (0);
1502}
1503
1504/*
1505 * Glue routine to nfsv4_fillattr().
1506 */
1507int
1508nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1509    struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1510    struct ucred *cred, struct thread *p, int isdgram, int reterr,
1511    int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1512{
1513	int error;
1514
1515	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1516	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1517	    mounted_on_fileno);
1518	NFSEXITCODE2(0, nd);
1519	return (error);
1520}
1521
1522/* Since the Readdir vnode ops vary, put the entire functions in here. */
1523/*
1524 * nfs readdir service
1525 * - mallocs what it thinks is enough to read
1526 *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1527 * - calls VOP_READDIR()
1528 * - loops around building the reply
1529 *	if the output generated exceeds count break out of loop
1530 *	The NFSM_CLGET macro is used here so that the reply will be packed
1531 *	tightly in mbuf clusters.
1532 * - it trims out records with d_fileno == 0
1533 *	this doesn't matter for Unix clients, but they might confuse clients
1534 *	for other os'.
1535 * - it trims out records with d_type == DT_WHT
1536 *	these cannot be seen through NFS (unless we extend the protocol)
1537 *     The alternate call nfsrvd_readdirplus() does lookups as well.
1538 * PS: The NFS protocol spec. does not clarify what the "count" byte
1539 *	argument is a count of.. just name strings and file id's or the
1540 *	entire reply rpc or ...
1541 *	I tried just file name and id sizes and it confused the Sun client,
1542 *	so I am using the full rpc size now. The "paranoia.." comment refers
1543 *	to including the status longwords that are not a part of the dir.
1544 *	"entry" structures, but are in the rpc.
1545 */
1546int
1547nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1548    struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1549{
1550	struct dirent *dp;
1551	u_int32_t *tl;
1552	int dirlen;
1553	char *cpos, *cend, *rbuf;
1554	struct nfsvattr at;
1555	int nlen, error = 0, getret = 1;
1556	int siz, cnt, fullsiz, eofflag, ncookies;
1557	u_int64_t off, toff, verf;
1558	u_long *cookies = NULL, *cookiep;
1559	struct uio io;
1560	struct iovec iv;
1561	int is_ufs;
1562
1563	if (nd->nd_repstat) {
1564		nfsrv_postopattr(nd, getret, &at);
1565		goto out;
1566	}
1567	if (nd->nd_flag & ND_NFSV2) {
1568		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1569		off = fxdr_unsigned(u_quad_t, *tl++);
1570	} else {
1571		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1572		off = fxdr_hyper(tl);
1573		tl += 2;
1574		verf = fxdr_hyper(tl);
1575		tl += 2;
1576	}
1577	toff = off;
1578	cnt = fxdr_unsigned(int, *tl);
1579	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1580		cnt = NFS_SRVMAXDATA(nd);
1581	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1582	fullsiz = siz;
1583	if (nd->nd_flag & ND_NFSV3) {
1584		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1585		    p, 1);
1586#if 0
1587		/*
1588		 * va_filerev is not sufficient as a cookie verifier,
1589		 * since it is not supposed to change when entries are
1590		 * removed/added unless that offset cookies returned to
1591		 * the client are no longer valid.
1592		 */
1593		if (!nd->nd_repstat && toff && verf != at.na_filerev)
1594			nd->nd_repstat = NFSERR_BAD_COOKIE;
1595#endif
1596	}
1597	if (!nd->nd_repstat && vp->v_type != VDIR)
1598		nd->nd_repstat = NFSERR_NOTDIR;
1599	if (nd->nd_repstat == 0 && cnt == 0) {
1600		if (nd->nd_flag & ND_NFSV2)
1601			/* NFSv2 does not have NFSERR_TOOSMALL */
1602			nd->nd_repstat = EPERM;
1603		else
1604			nd->nd_repstat = NFSERR_TOOSMALL;
1605	}
1606	if (!nd->nd_repstat)
1607		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1608		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1609		    NFSACCCHK_VPISLOCKED, NULL);
1610	if (nd->nd_repstat) {
1611		vput(vp);
1612		if (nd->nd_flag & ND_NFSV3)
1613			nfsrv_postopattr(nd, getret, &at);
1614		goto out;
1615	}
1616	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1617	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1618again:
1619	eofflag = 0;
1620	if (cookies) {
1621		free((caddr_t)cookies, M_TEMP);
1622		cookies = NULL;
1623	}
1624
1625	iv.iov_base = rbuf;
1626	iv.iov_len = siz;
1627	io.uio_iov = &iv;
1628	io.uio_iovcnt = 1;
1629	io.uio_offset = (off_t)off;
1630	io.uio_resid = siz;
1631	io.uio_segflg = UIO_SYSSPACE;
1632	io.uio_rw = UIO_READ;
1633	io.uio_td = NULL;
1634	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1635	    &cookies);
1636	off = (u_int64_t)io.uio_offset;
1637	if (io.uio_resid)
1638		siz -= io.uio_resid;
1639
1640	if (!cookies && !nd->nd_repstat)
1641		nd->nd_repstat = NFSERR_PERM;
1642	if (nd->nd_flag & ND_NFSV3) {
1643		getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1644		if (!nd->nd_repstat)
1645			nd->nd_repstat = getret;
1646	}
1647
1648	/*
1649	 * Handles the failed cases. nd->nd_repstat == 0 past here.
1650	 */
1651	if (nd->nd_repstat) {
1652		vput(vp);
1653		free((caddr_t)rbuf, M_TEMP);
1654		if (cookies)
1655			free((caddr_t)cookies, M_TEMP);
1656		if (nd->nd_flag & ND_NFSV3)
1657			nfsrv_postopattr(nd, getret, &at);
1658		goto out;
1659	}
1660	/*
1661	 * If nothing read, return eof
1662	 * rpc reply
1663	 */
1664	if (siz == 0) {
1665		vput(vp);
1666		if (nd->nd_flag & ND_NFSV2) {
1667			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1668		} else {
1669			nfsrv_postopattr(nd, getret, &at);
1670			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1671			txdr_hyper(at.na_filerev, tl);
1672			tl += 2;
1673		}
1674		*tl++ = newnfs_false;
1675		*tl = newnfs_true;
1676		FREE((caddr_t)rbuf, M_TEMP);
1677		FREE((caddr_t)cookies, M_TEMP);
1678		goto out;
1679	}
1680
1681	/*
1682	 * Check for degenerate cases of nothing useful read.
1683	 * If so go try again
1684	 */
1685	cpos = rbuf;
1686	cend = rbuf + siz;
1687	dp = (struct dirent *)cpos;
1688	cookiep = cookies;
1689
1690	/*
1691	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1692	 * directory offset up to a block boundary, so it is necessary to
1693	 * skip over the records that precede the requested offset. This
1694	 * requires the assumption that file offset cookies monotonically
1695	 * increase.
1696	 */
1697	while (cpos < cend && ncookies > 0 &&
1698	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1699	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
1700		cpos += dp->d_reclen;
1701		dp = (struct dirent *)cpos;
1702		cookiep++;
1703		ncookies--;
1704	}
1705	if (cpos >= cend || ncookies == 0) {
1706		siz = fullsiz;
1707		toff = off;
1708		goto again;
1709	}
1710	vput(vp);
1711
1712	/*
1713	 * dirlen is the size of the reply, including all XDR and must
1714	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1715	 * if the XDR should be included in "count", but to be safe, we do.
1716	 * (Include the two booleans at the end of the reply in dirlen now.)
1717	 */
1718	if (nd->nd_flag & ND_NFSV3) {
1719		nfsrv_postopattr(nd, getret, &at);
1720		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1721		txdr_hyper(at.na_filerev, tl);
1722		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1723	} else {
1724		dirlen = 2 * NFSX_UNSIGNED;
1725	}
1726
1727	/* Loop through the records and build reply */
1728	while (cpos < cend && ncookies > 0) {
1729		nlen = dp->d_namlen;
1730		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1731			nlen <= NFS_MAXNAMLEN) {
1732			if (nd->nd_flag & ND_NFSV3)
1733				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1734			else
1735				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1736			if (dirlen > cnt) {
1737				eofflag = 0;
1738				break;
1739			}
1740
1741			/*
1742			 * Build the directory record xdr from
1743			 * the dirent entry.
1744			 */
1745			if (nd->nd_flag & ND_NFSV3) {
1746				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1747				*tl++ = newnfs_true;
1748				*tl++ = 0;
1749			} else {
1750				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1751				*tl++ = newnfs_true;
1752			}
1753			*tl = txdr_unsigned(dp->d_fileno);
1754			(void) nfsm_strtom(nd, dp->d_name, nlen);
1755			if (nd->nd_flag & ND_NFSV3) {
1756				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1757				*tl++ = 0;
1758			} else
1759				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1760			*tl = txdr_unsigned(*cookiep);
1761		}
1762		cpos += dp->d_reclen;
1763		dp = (struct dirent *)cpos;
1764		cookiep++;
1765		ncookies--;
1766	}
1767	if (cpos < cend)
1768		eofflag = 0;
1769	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1770	*tl++ = newnfs_false;
1771	if (eofflag)
1772		*tl = newnfs_true;
1773	else
1774		*tl = newnfs_false;
1775	FREE((caddr_t)rbuf, M_TEMP);
1776	FREE((caddr_t)cookies, M_TEMP);
1777
1778out:
1779	NFSEXITCODE2(0, nd);
1780	return (0);
1781nfsmout:
1782	vput(vp);
1783	NFSEXITCODE2(error, nd);
1784	return (error);
1785}
1786
1787/*
1788 * Readdirplus for V3 and Readdir for V4.
1789 */
1790int
1791nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1792    struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1793{
1794	struct dirent *dp;
1795	u_int32_t *tl;
1796	int dirlen;
1797	char *cpos, *cend, *rbuf;
1798	struct vnode *nvp;
1799	fhandle_t nfh;
1800	struct nfsvattr nva, at, *nvap = &nva;
1801	struct mbuf *mb0, *mb1;
1802	struct nfsreferral *refp;
1803	int nlen, r, error = 0, getret = 1, usevget = 1;
1804	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1805	caddr_t bpos0, bpos1;
1806	u_int64_t off, toff, verf;
1807	u_long *cookies = NULL, *cookiep;
1808	nfsattrbit_t attrbits, rderrbits, savbits;
1809	struct uio io;
1810	struct iovec iv;
1811	struct componentname cn;
1812	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
1813	struct mount *mp, *new_mp;
1814	uint64_t mounted_on_fileno;
1815
1816	if (nd->nd_repstat) {
1817		nfsrv_postopattr(nd, getret, &at);
1818		goto out;
1819	}
1820	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1821	off = fxdr_hyper(tl);
1822	toff = off;
1823	tl += 2;
1824	verf = fxdr_hyper(tl);
1825	tl += 2;
1826	siz = fxdr_unsigned(int, *tl++);
1827	cnt = fxdr_unsigned(int, *tl);
1828
1829	/*
1830	 * Use the server's maximum data transfer size as the upper bound
1831	 * on reply datalen.
1832	 */
1833	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1834		cnt = NFS_SRVMAXDATA(nd);
1835
1836	/*
1837	 * siz is a "hint" of how much directory information (name, fileid,
1838	 * cookie) should be in the reply. At least one client "hints" 0,
1839	 * so I set it to cnt for that case. I also round it up to the
1840	 * next multiple of DIRBLKSIZ.
1841	 */
1842	if (siz <= 0)
1843		siz = cnt;
1844	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1845
1846	if (nd->nd_flag & ND_NFSV4) {
1847		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1848		if (error)
1849			goto nfsmout;
1850		NFSSET_ATTRBIT(&savbits, &attrbits);
1851		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1852		NFSZERO_ATTRBIT(&rderrbits);
1853		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1854	} else {
1855		NFSZERO_ATTRBIT(&attrbits);
1856	}
1857	fullsiz = siz;
1858	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1859	if (!nd->nd_repstat) {
1860	    if (off && verf != at.na_filerev) {
1861		/*
1862		 * va_filerev is not sufficient as a cookie verifier,
1863		 * since it is not supposed to change when entries are
1864		 * removed/added unless that offset cookies returned to
1865		 * the client are no longer valid.
1866		 */
1867#if 0
1868		if (nd->nd_flag & ND_NFSV4) {
1869			nd->nd_repstat = NFSERR_NOTSAME;
1870		} else {
1871			nd->nd_repstat = NFSERR_BAD_COOKIE;
1872		}
1873#endif
1874	    } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1875		nd->nd_repstat = NFSERR_BAD_COOKIE;
1876	    }
1877	}
1878	if (!nd->nd_repstat && vp->v_type != VDIR)
1879		nd->nd_repstat = NFSERR_NOTDIR;
1880	if (!nd->nd_repstat && cnt == 0)
1881		nd->nd_repstat = NFSERR_TOOSMALL;
1882	if (!nd->nd_repstat)
1883		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1884		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1885		    NFSACCCHK_VPISLOCKED, NULL);
1886	if (nd->nd_repstat) {
1887		vput(vp);
1888		if (nd->nd_flag & ND_NFSV3)
1889			nfsrv_postopattr(nd, getret, &at);
1890		goto out;
1891	}
1892	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1893	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
1894
1895	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1896again:
1897	eofflag = 0;
1898	if (cookies) {
1899		free((caddr_t)cookies, M_TEMP);
1900		cookies = NULL;
1901	}
1902
1903	iv.iov_base = rbuf;
1904	iv.iov_len = siz;
1905	io.uio_iov = &iv;
1906	io.uio_iovcnt = 1;
1907	io.uio_offset = (off_t)off;
1908	io.uio_resid = siz;
1909	io.uio_segflg = UIO_SYSSPACE;
1910	io.uio_rw = UIO_READ;
1911	io.uio_td = NULL;
1912	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1913	    &cookies);
1914	off = (u_int64_t)io.uio_offset;
1915	if (io.uio_resid)
1916		siz -= io.uio_resid;
1917
1918	getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1919
1920	if (!cookies && !nd->nd_repstat)
1921		nd->nd_repstat = NFSERR_PERM;
1922	if (!nd->nd_repstat)
1923		nd->nd_repstat = getret;
1924	if (nd->nd_repstat) {
1925		vput(vp);
1926		if (cookies)
1927			free((caddr_t)cookies, M_TEMP);
1928		free((caddr_t)rbuf, M_TEMP);
1929		if (nd->nd_flag & ND_NFSV3)
1930			nfsrv_postopattr(nd, getret, &at);
1931		goto out;
1932	}
1933	/*
1934	 * If nothing read, return eof
1935	 * rpc reply
1936	 */
1937	if (siz == 0) {
1938		vput(vp);
1939		if (nd->nd_flag & ND_NFSV3)
1940			nfsrv_postopattr(nd, getret, &at);
1941		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1942		txdr_hyper(at.na_filerev, tl);
1943		tl += 2;
1944		*tl++ = newnfs_false;
1945		*tl = newnfs_true;
1946		free((caddr_t)cookies, M_TEMP);
1947		free((caddr_t)rbuf, M_TEMP);
1948		goto out;
1949	}
1950
1951	/*
1952	 * Check for degenerate cases of nothing useful read.
1953	 * If so go try again
1954	 */
1955	cpos = rbuf;
1956	cend = rbuf + siz;
1957	dp = (struct dirent *)cpos;
1958	cookiep = cookies;
1959
1960	/*
1961	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1962	 * directory offset up to a block boundary, so it is necessary to
1963	 * skip over the records that precede the requested offset. This
1964	 * requires the assumption that file offset cookies monotonically
1965	 * increase.
1966	 */
1967	while (cpos < cend && ncookies > 0 &&
1968	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1969	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
1970	   ((nd->nd_flag & ND_NFSV4) &&
1971	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1972	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1973		cpos += dp->d_reclen;
1974		dp = (struct dirent *)cpos;
1975		cookiep++;
1976		ncookies--;
1977	}
1978	if (cpos >= cend || ncookies == 0) {
1979		siz = fullsiz;
1980		toff = off;
1981		goto again;
1982	}
1983
1984	/*
1985	 * Busy the file system so that the mount point won't go away
1986	 * and, as such, VFS_VGET() can be used safely.
1987	 */
1988	mp = vp->v_mount;
1989	vfs_ref(mp);
1990	NFSVOPUNLOCK(vp, 0);
1991	nd->nd_repstat = vfs_busy(mp, 0);
1992	vfs_rel(mp);
1993	if (nd->nd_repstat != 0) {
1994		vrele(vp);
1995		free(cookies, M_TEMP);
1996		free(rbuf, M_TEMP);
1997		if (nd->nd_flag & ND_NFSV3)
1998			nfsrv_postopattr(nd, getret, &at);
1999		goto out;
2000	}
2001
2002	/*
2003	 * Check to see if entries in this directory can be safely acquired
2004	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2005	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2006	 * automount of the snapshot directory that is required will
2007	 * be done.
2008	 * This needs to be done here for NFSv4, since NFSv4 never does
2009	 * a VFS_VGET() for "." or "..".
2010	 */
2011	if (is_zfs == 1) {
2012		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2013		if (r == EOPNOTSUPP) {
2014			usevget = 0;
2015			cn.cn_nameiop = LOOKUP;
2016			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2017			cn.cn_cred = nd->nd_cred;
2018			cn.cn_thread = p;
2019		} else if (r == 0)
2020			vput(nvp);
2021	}
2022
2023	/*
2024	 * Save this position, in case there is an error before one entry
2025	 * is created.
2026	 */
2027	mb0 = nd->nd_mb;
2028	bpos0 = nd->nd_bpos;
2029
2030	/*
2031	 * Fill in the first part of the reply.
2032	 * dirlen is the reply length in bytes and cannot exceed cnt.
2033	 * (Include the two booleans at the end of the reply in dirlen now,
2034	 *  so we recognize when we have exceeded cnt.)
2035	 */
2036	if (nd->nd_flag & ND_NFSV3) {
2037		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2038		nfsrv_postopattr(nd, getret, &at);
2039	} else {
2040		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2041	}
2042	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2043	txdr_hyper(at.na_filerev, tl);
2044
2045	/*
2046	 * Save this position, in case there is an empty reply needed.
2047	 */
2048	mb1 = nd->nd_mb;
2049	bpos1 = nd->nd_bpos;
2050
2051	/* Loop through the records and build reply */
2052	entrycnt = 0;
2053	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2054		nlen = dp->d_namlen;
2055		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2056		    nlen <= NFS_MAXNAMLEN &&
2057		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2058		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2059		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2060			/*
2061			 * Save the current position in the reply, in case
2062			 * this entry exceeds cnt.
2063			 */
2064			mb1 = nd->nd_mb;
2065			bpos1 = nd->nd_bpos;
2066
2067			/*
2068			 * For readdir_and_lookup get the vnode using
2069			 * the file number.
2070			 */
2071			nvp = NULL;
2072			refp = NULL;
2073			r = 0;
2074			at_root = 0;
2075			needs_unbusy = 0;
2076			new_mp = mp;
2077			mounted_on_fileno = (uint64_t)dp->d_fileno;
2078			if ((nd->nd_flag & ND_NFSV3) ||
2079			    NFSNONZERO_ATTRBIT(&savbits)) {
2080				if (nd->nd_flag & ND_NFSV4)
2081					refp = nfsv4root_getreferral(NULL,
2082					    vp, dp->d_fileno);
2083				if (refp == NULL) {
2084					if (usevget)
2085						r = VFS_VGET(mp, dp->d_fileno,
2086						    LK_SHARED, &nvp);
2087					else
2088						r = EOPNOTSUPP;
2089					if (r == EOPNOTSUPP) {
2090						if (usevget) {
2091							usevget = 0;
2092							cn.cn_nameiop = LOOKUP;
2093							cn.cn_lkflags =
2094							    LK_SHARED |
2095							    LK_RETRY;
2096							cn.cn_cred =
2097							    nd->nd_cred;
2098							cn.cn_thread = p;
2099						}
2100						cn.cn_nameptr = dp->d_name;
2101						cn.cn_namelen = nlen;
2102						cn.cn_flags = ISLASTCN |
2103						    NOFOLLOW | LOCKLEAF;
2104						if (nlen == 2 &&
2105						    dp->d_name[0] == '.' &&
2106						    dp->d_name[1] == '.')
2107							cn.cn_flags |=
2108							    ISDOTDOT;
2109						if (NFSVOPLOCK(vp, LK_SHARED)
2110						    != 0) {
2111							nd->nd_repstat = EPERM;
2112							break;
2113						}
2114						if ((vp->v_vflag & VV_ROOT) != 0
2115						    && (cn.cn_flags & ISDOTDOT)
2116						    != 0) {
2117							vref(vp);
2118							nvp = vp;
2119							r = 0;
2120						} else {
2121							r = VOP_LOOKUP(vp, &nvp,
2122							    &cn);
2123							if (vp != nvp)
2124								NFSVOPUNLOCK(vp,
2125								    0);
2126						}
2127					}
2128
2129					/*
2130					 * For NFSv4, check to see if nvp is
2131					 * a mount point and get the mount
2132					 * point vnode, as required.
2133					 */
2134					if (r == 0 &&
2135					    nfsrv_enable_crossmntpt != 0 &&
2136					    (nd->nd_flag & ND_NFSV4) != 0 &&
2137					    nvp->v_type == VDIR &&
2138					    nvp->v_mountedhere != NULL) {
2139						new_mp = nvp->v_mountedhere;
2140						r = vfs_busy(new_mp, 0);
2141						vput(nvp);
2142						nvp = NULL;
2143						if (r == 0) {
2144							r = VFS_ROOT(new_mp,
2145							    LK_SHARED, &nvp);
2146							needs_unbusy = 1;
2147							if (r == 0)
2148								at_root = 1;
2149						}
2150					}
2151				}
2152				if (!r) {
2153				    if (refp == NULL &&
2154					((nd->nd_flag & ND_NFSV3) ||
2155					 NFSNONZERO_ATTRBIT(&attrbits))) {
2156					r = nfsvno_getfh(nvp, &nfh, p);
2157					if (!r)
2158					    r = nfsvno_getattr(nvp, nvap,
2159						nd->nd_cred, p, 1);
2160					if (r == 0 && is_zfs == 1 &&
2161					    nfsrv_enable_crossmntpt != 0 &&
2162					    (nd->nd_flag & ND_NFSV4) != 0 &&
2163					    nvp->v_type == VDIR &&
2164					    vp->v_mount != nvp->v_mount) {
2165					    /*
2166					     * For a ZFS snapshot, there is a
2167					     * pseudo mount that does not set
2168					     * v_mountedhere, so it needs to
2169					     * be detected via a different
2170					     * mount structure.
2171					     */
2172					    at_root = 1;
2173					    if (new_mp == mp)
2174						new_mp = nvp->v_mount;
2175					}
2176				    }
2177				} else {
2178				    nvp = NULL;
2179				}
2180				if (r) {
2181					if (!NFSISSET_ATTRBIT(&attrbits,
2182					    NFSATTRBIT_RDATTRERROR)) {
2183						if (nvp != NULL)
2184							vput(nvp);
2185						if (needs_unbusy != 0)
2186							vfs_unbusy(new_mp);
2187						nd->nd_repstat = r;
2188						break;
2189					}
2190				}
2191			}
2192
2193			/*
2194			 * Build the directory record xdr
2195			 */
2196			if (nd->nd_flag & ND_NFSV3) {
2197				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2198				*tl++ = newnfs_true;
2199				*tl++ = 0;
2200				*tl = txdr_unsigned(dp->d_fileno);
2201				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2202				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2203				*tl++ = 0;
2204				*tl = txdr_unsigned(*cookiep);
2205				nfsrv_postopattr(nd, 0, nvap);
2206				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2207				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2208				if (nvp != NULL)
2209					vput(nvp);
2210			} else {
2211				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2212				*tl++ = newnfs_true;
2213				*tl++ = 0;
2214				*tl = txdr_unsigned(*cookiep);
2215				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2216				if (nvp != NULL) {
2217					supports_nfsv4acls =
2218					    nfs_supportsnfsv4acls(nvp);
2219					NFSVOPUNLOCK(nvp, 0);
2220				} else
2221					supports_nfsv4acls = 0;
2222				if (refp != NULL) {
2223					dirlen += nfsrv_putreferralattr(nd,
2224					    &savbits, refp, 0,
2225					    &nd->nd_repstat);
2226					if (nd->nd_repstat) {
2227						if (nvp != NULL)
2228							vrele(nvp);
2229						if (needs_unbusy != 0)
2230							vfs_unbusy(new_mp);
2231						break;
2232					}
2233				} else if (r) {
2234					dirlen += nfsvno_fillattr(nd, new_mp,
2235					    nvp, nvap, &nfh, r, &rderrbits,
2236					    nd->nd_cred, p, isdgram, 0,
2237					    supports_nfsv4acls, at_root,
2238					    mounted_on_fileno);
2239				} else {
2240					dirlen += nfsvno_fillattr(nd, new_mp,
2241					    nvp, nvap, &nfh, r, &attrbits,
2242					    nd->nd_cred, p, isdgram, 0,
2243					    supports_nfsv4acls, at_root,
2244					    mounted_on_fileno);
2245				}
2246				if (nvp != NULL)
2247					vrele(nvp);
2248				dirlen += (3 * NFSX_UNSIGNED);
2249			}
2250			if (needs_unbusy != 0)
2251				vfs_unbusy(new_mp);
2252			if (dirlen <= cnt)
2253				entrycnt++;
2254		}
2255		cpos += dp->d_reclen;
2256		dp = (struct dirent *)cpos;
2257		cookiep++;
2258		ncookies--;
2259	}
2260	vrele(vp);
2261	vfs_unbusy(mp);
2262
2263	/*
2264	 * If dirlen > cnt, we must strip off the last entry. If that
2265	 * results in an empty reply, report NFSERR_TOOSMALL.
2266	 */
2267	if (dirlen > cnt || nd->nd_repstat) {
2268		if (!nd->nd_repstat && entrycnt == 0)
2269			nd->nd_repstat = NFSERR_TOOSMALL;
2270		if (nd->nd_repstat) {
2271			newnfs_trimtrailing(nd, mb0, bpos0);
2272			if (nd->nd_flag & ND_NFSV3)
2273				nfsrv_postopattr(nd, getret, &at);
2274		} else
2275			newnfs_trimtrailing(nd, mb1, bpos1);
2276		eofflag = 0;
2277	} else if (cpos < cend)
2278		eofflag = 0;
2279	if (!nd->nd_repstat) {
2280		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2281		*tl++ = newnfs_false;
2282		if (eofflag)
2283			*tl = newnfs_true;
2284		else
2285			*tl = newnfs_false;
2286	}
2287	FREE((caddr_t)cookies, M_TEMP);
2288	FREE((caddr_t)rbuf, M_TEMP);
2289
2290out:
2291	NFSEXITCODE2(0, nd);
2292	return (0);
2293nfsmout:
2294	vput(vp);
2295	NFSEXITCODE2(error, nd);
2296	return (error);
2297}
2298
2299/*
2300 * Get the settable attributes out of the mbuf list.
2301 * (Return 0 or EBADRPC)
2302 */
2303int
2304nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2305    nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2306{
2307	u_int32_t *tl;
2308	struct nfsv2_sattr *sp;
2309	int error = 0, toclient = 0;
2310
2311	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2312	case ND_NFSV2:
2313		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2314		/*
2315		 * Some old clients didn't fill in the high order 16bits.
2316		 * --> check the low order 2 bytes for 0xffff
2317		 */
2318		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2319			nvap->na_mode = nfstov_mode(sp->sa_mode);
2320		if (sp->sa_uid != newnfs_xdrneg1)
2321			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2322		if (sp->sa_gid != newnfs_xdrneg1)
2323			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2324		if (sp->sa_size != newnfs_xdrneg1)
2325			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2326		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2327#ifdef notyet
2328			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2329#else
2330			nvap->na_atime.tv_sec =
2331				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2332			nvap->na_atime.tv_nsec = 0;
2333#endif
2334		}
2335		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2336			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2337		break;
2338	case ND_NFSV3:
2339		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2340		if (*tl == newnfs_true) {
2341			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2342			nvap->na_mode = nfstov_mode(*tl);
2343		}
2344		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2345		if (*tl == newnfs_true) {
2346			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2347			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2348		}
2349		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2350		if (*tl == newnfs_true) {
2351			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2352			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2353		}
2354		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2355		if (*tl == newnfs_true) {
2356			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2357			nvap->na_size = fxdr_hyper(tl);
2358		}
2359		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2360		switch (fxdr_unsigned(int, *tl)) {
2361		case NFSV3SATTRTIME_TOCLIENT:
2362			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2363			fxdr_nfsv3time(tl, &nvap->na_atime);
2364			toclient = 1;
2365			break;
2366		case NFSV3SATTRTIME_TOSERVER:
2367			vfs_timestamp(&nvap->na_atime);
2368			nvap->na_vaflags |= VA_UTIMES_NULL;
2369			break;
2370		};
2371		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2372		switch (fxdr_unsigned(int, *tl)) {
2373		case NFSV3SATTRTIME_TOCLIENT:
2374			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2375			fxdr_nfsv3time(tl, &nvap->na_mtime);
2376			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2377			break;
2378		case NFSV3SATTRTIME_TOSERVER:
2379			vfs_timestamp(&nvap->na_mtime);
2380			if (!toclient)
2381				nvap->na_vaflags |= VA_UTIMES_NULL;
2382			break;
2383		};
2384		break;
2385	case ND_NFSV4:
2386		error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2387	};
2388nfsmout:
2389	NFSEXITCODE2(error, nd);
2390	return (error);
2391}
2392
2393/*
2394 * Handle the setable attributes for V4.
2395 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2396 */
2397int
2398nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2399    nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2400{
2401	u_int32_t *tl;
2402	int attrsum = 0;
2403	int i, j;
2404	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2405	int toclient = 0;
2406	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2407	uid_t uid;
2408	gid_t gid;
2409
2410	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2411	if (error)
2412		goto nfsmout;
2413	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2414	attrsize = fxdr_unsigned(int, *tl);
2415
2416	/*
2417	 * Loop around getting the setable attributes. If an unsupported
2418	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2419	 */
2420	if (retnotsup) {
2421		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2422		bitpos = NFSATTRBIT_MAX;
2423	} else {
2424		bitpos = 0;
2425	}
2426	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2427	    if (attrsum > attrsize) {
2428		error = NFSERR_BADXDR;
2429		goto nfsmout;
2430	    }
2431	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2432		switch (bitpos) {
2433		case NFSATTRBIT_SIZE:
2434			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2435			nvap->na_size = fxdr_hyper(tl);
2436			attrsum += NFSX_HYPER;
2437			break;
2438		case NFSATTRBIT_ACL:
2439			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2440			    p);
2441			if (error)
2442				goto nfsmout;
2443			if (aceerr && !nd->nd_repstat)
2444				nd->nd_repstat = aceerr;
2445			attrsum += aclsize;
2446			break;
2447		case NFSATTRBIT_ARCHIVE:
2448			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2449			if (!nd->nd_repstat)
2450				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2451			attrsum += NFSX_UNSIGNED;
2452			break;
2453		case NFSATTRBIT_HIDDEN:
2454			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2455			if (!nd->nd_repstat)
2456				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2457			attrsum += NFSX_UNSIGNED;
2458			break;
2459		case NFSATTRBIT_MIMETYPE:
2460			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2461			i = fxdr_unsigned(int, *tl);
2462			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2463			if (error)
2464				goto nfsmout;
2465			if (!nd->nd_repstat)
2466				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2467			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2468			break;
2469		case NFSATTRBIT_MODE:
2470			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2471			nvap->na_mode = nfstov_mode(*tl);
2472			attrsum += NFSX_UNSIGNED;
2473			break;
2474		case NFSATTRBIT_OWNER:
2475			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2476			j = fxdr_unsigned(int, *tl);
2477			if (j < 0) {
2478				error = NFSERR_BADXDR;
2479				goto nfsmout;
2480			}
2481			if (j > NFSV4_SMALLSTR)
2482				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2483			else
2484				cp = namestr;
2485			error = nfsrv_mtostr(nd, cp, j);
2486			if (error) {
2487				if (j > NFSV4_SMALLSTR)
2488					free(cp, M_NFSSTRING);
2489				goto nfsmout;
2490			}
2491			if (!nd->nd_repstat) {
2492				nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2493				    p);
2494				if (!nd->nd_repstat)
2495					nvap->na_uid = uid;
2496			}
2497			if (j > NFSV4_SMALLSTR)
2498				free(cp, M_NFSSTRING);
2499			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2500			break;
2501		case NFSATTRBIT_OWNERGROUP:
2502			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2503			j = fxdr_unsigned(int, *tl);
2504			if (j < 0) {
2505				error = NFSERR_BADXDR;
2506				goto nfsmout;
2507			}
2508			if (j > NFSV4_SMALLSTR)
2509				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2510			else
2511				cp = namestr;
2512			error = nfsrv_mtostr(nd, cp, j);
2513			if (error) {
2514				if (j > NFSV4_SMALLSTR)
2515					free(cp, M_NFSSTRING);
2516				goto nfsmout;
2517			}
2518			if (!nd->nd_repstat) {
2519				nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2520				    p);
2521				if (!nd->nd_repstat)
2522					nvap->na_gid = gid;
2523			}
2524			if (j > NFSV4_SMALLSTR)
2525				free(cp, M_NFSSTRING);
2526			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2527			break;
2528		case NFSATTRBIT_SYSTEM:
2529			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2530			if (!nd->nd_repstat)
2531				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2532			attrsum += NFSX_UNSIGNED;
2533			break;
2534		case NFSATTRBIT_TIMEACCESSSET:
2535			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2536			attrsum += NFSX_UNSIGNED;
2537			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2538			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2539			    fxdr_nfsv4time(tl, &nvap->na_atime);
2540			    toclient = 1;
2541			    attrsum += NFSX_V4TIME;
2542			} else {
2543			    vfs_timestamp(&nvap->na_atime);
2544			    nvap->na_vaflags |= VA_UTIMES_NULL;
2545			}
2546			break;
2547		case NFSATTRBIT_TIMEBACKUP:
2548			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2549			if (!nd->nd_repstat)
2550				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2551			attrsum += NFSX_V4TIME;
2552			break;
2553		case NFSATTRBIT_TIMECREATE:
2554			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2555			if (!nd->nd_repstat)
2556				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2557			attrsum += NFSX_V4TIME;
2558			break;
2559		case NFSATTRBIT_TIMEMODIFYSET:
2560			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2561			attrsum += NFSX_UNSIGNED;
2562			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2563			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2564			    fxdr_nfsv4time(tl, &nvap->na_mtime);
2565			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
2566			    attrsum += NFSX_V4TIME;
2567			} else {
2568			    vfs_timestamp(&nvap->na_mtime);
2569			    if (!toclient)
2570				nvap->na_vaflags |= VA_UTIMES_NULL;
2571			}
2572			break;
2573		default:
2574			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2575			/*
2576			 * set bitpos so we drop out of the loop.
2577			 */
2578			bitpos = NFSATTRBIT_MAX;
2579			break;
2580		};
2581	}
2582
2583	/*
2584	 * some clients pad the attrlist, so we need to skip over the
2585	 * padding.
2586	 */
2587	if (attrsum > attrsize) {
2588		error = NFSERR_BADXDR;
2589	} else {
2590		attrsize = NFSM_RNDUP(attrsize);
2591		if (attrsum < attrsize)
2592			error = nfsm_advance(nd, attrsize - attrsum, -1);
2593	}
2594nfsmout:
2595	NFSEXITCODE2(error, nd);
2596	return (error);
2597}
2598
2599/*
2600 * Check/setup export credentials.
2601 */
2602int
2603nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2604    struct ucred *credanon)
2605{
2606	int error = 0;
2607
2608	/*
2609	 * Check/setup credentials.
2610	 */
2611	if (nd->nd_flag & ND_GSS)
2612		exp->nes_exflag &= ~MNT_EXPORTANON;
2613
2614	/*
2615	 * Check to see if the operation is allowed for this security flavor.
2616	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2617	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2618	 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2619	 */
2620	if (nfsvno_testexp(nd, exp) &&
2621	    nd->nd_procnum != NFSV4OP_SECINFO &&
2622	    nd->nd_procnum != NFSPROC_FSINFO) {
2623		if (nd->nd_flag & ND_NFSV4)
2624			error = NFSERR_WRONGSEC;
2625		else
2626			error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2627		goto out;
2628	}
2629
2630	/*
2631	 * Check to see if the file system is exported V4 only.
2632	 */
2633	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2634		error = NFSERR_PROGNOTV4;
2635		goto out;
2636	}
2637
2638	/*
2639	 * Now, map the user credentials.
2640	 * (Note that ND_AUTHNONE will only be set for an NFSv3
2641	 *  Fsinfo RPC. If set for anything else, this code might need
2642	 *  to change.)
2643	 */
2644	if (NFSVNO_EXPORTED(exp) &&
2645	    ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2646	     NFSVNO_EXPORTANON(exp) ||
2647	     (nd->nd_flag & ND_AUTHNONE))) {
2648		nd->nd_cred->cr_uid = credanon->cr_uid;
2649		nd->nd_cred->cr_gid = credanon->cr_gid;
2650		crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2651		    credanon->cr_groups);
2652	}
2653
2654out:
2655	NFSEXITCODE2(error, nd);
2656	return (error);
2657}
2658
2659/*
2660 * Check exports.
2661 */
2662int
2663nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2664    struct ucred **credp)
2665{
2666	int i, error, *secflavors;
2667
2668	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2669	    &exp->nes_numsecflavor, &secflavors);
2670	if (error) {
2671		if (nfs_rootfhset) {
2672			exp->nes_exflag = 0;
2673			exp->nes_numsecflavor = 0;
2674			error = 0;
2675		}
2676	} else {
2677		/* Copy the security flavors. */
2678		for (i = 0; i < exp->nes_numsecflavor; i++)
2679			exp->nes_secflavors[i] = secflavors[i];
2680	}
2681	NFSEXITCODE(error);
2682	return (error);
2683}
2684
2685/*
2686 * Get a vnode for a file handle and export stuff.
2687 */
2688int
2689nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2690    int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2691    struct ucred **credp)
2692{
2693	int i, error, *secflavors;
2694
2695	*credp = NULL;
2696	exp->nes_numsecflavor = 0;
2697	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2698	if (error != 0)
2699		/* Make sure the server replies ESTALE to the client. */
2700		error = ESTALE;
2701	if (nam && !error) {
2702		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2703		    &exp->nes_numsecflavor, &secflavors);
2704		if (error) {
2705			if (nfs_rootfhset) {
2706				exp->nes_exflag = 0;
2707				exp->nes_numsecflavor = 0;
2708				error = 0;
2709			} else {
2710				vput(*vpp);
2711			}
2712		} else {
2713			/* Copy the security flavors. */
2714			for (i = 0; i < exp->nes_numsecflavor; i++)
2715				exp->nes_secflavors[i] = secflavors[i];
2716		}
2717	}
2718	NFSEXITCODE(error);
2719	return (error);
2720}
2721
2722/*
2723 * nfsd_fhtovp() - convert a fh to a vnode ptr
2724 * 	- look up fsid in mount list (if not found ret error)
2725 *	- get vp and export rights by calling nfsvno_fhtovp()
2726 *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2727 *	  for AUTH_SYS
2728 *	- if mpp != NULL, return the mount point so that it can
2729 *	  be used for vn_finished_write() by the caller
2730 */
2731void
2732nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2733    struct vnode **vpp, struct nfsexstuff *exp,
2734    struct mount **mpp, int startwrite, struct thread *p)
2735{
2736	struct mount *mp;
2737	struct ucred *credanon;
2738	fhandle_t *fhp;
2739
2740	fhp = (fhandle_t *)nfp->nfsrvfh_data;
2741	/*
2742	 * Check for the special case of the nfsv4root_fh.
2743	 */
2744	mp = vfs_busyfs(&fhp->fh_fsid);
2745	if (mpp != NULL)
2746		*mpp = mp;
2747	if (mp == NULL) {
2748		*vpp = NULL;
2749		nd->nd_repstat = ESTALE;
2750		goto out;
2751	}
2752
2753	if (startwrite) {
2754		vn_start_write(NULL, mpp, V_WAIT);
2755		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
2756			lktype = LK_EXCLUSIVE;
2757	}
2758	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2759	    &credanon);
2760	vfs_unbusy(mp);
2761
2762	/*
2763	 * For NFSv4 without a pseudo root fs, unexported file handles
2764	 * can be returned, so that Lookup works everywhere.
2765	 */
2766	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2767	    !(nd->nd_flag & ND_NFSV4)) {
2768		vput(*vpp);
2769		nd->nd_repstat = EACCES;
2770	}
2771
2772	/*
2773	 * Personally, I've never seen any point in requiring a
2774	 * reserved port#, since only in the rare case where the
2775	 * clients are all boxes with secure system priviledges,
2776	 * does it provide any enhanced security, but... some people
2777	 * believe it to be useful and keep putting this code back in.
2778	 * (There is also some "security checker" out there that
2779	 *  complains if the nfs server doesn't enforce this.)
2780	 * However, note the following:
2781	 * RFC3530 (NFSv4) specifies that a reserved port# not be
2782	 *	required.
2783	 * RFC2623 recommends that, if a reserved port# is checked for,
2784	 *	that there be a way to turn that off--> ifdef'd.
2785	 */
2786#ifdef NFS_REQRSVPORT
2787	if (!nd->nd_repstat) {
2788		struct sockaddr_in *saddr;
2789		struct sockaddr_in6 *saddr6;
2790
2791		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2792		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2793		if (!(nd->nd_flag & ND_NFSV4) &&
2794		    ((saddr->sin_family == AF_INET &&
2795		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2796		     (saddr6->sin6_family == AF_INET6 &&
2797		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2798			vput(*vpp);
2799			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2800		}
2801	}
2802#endif	/* NFS_REQRSVPORT */
2803
2804	/*
2805	 * Check/setup credentials.
2806	 */
2807	if (!nd->nd_repstat) {
2808		nd->nd_saveduid = nd->nd_cred->cr_uid;
2809		nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2810		if (nd->nd_repstat)
2811			vput(*vpp);
2812	}
2813	if (credanon != NULL)
2814		crfree(credanon);
2815	if (nd->nd_repstat) {
2816		if (startwrite)
2817			vn_finished_write(mp);
2818		*vpp = NULL;
2819		if (mpp != NULL)
2820			*mpp = NULL;
2821	}
2822
2823out:
2824	NFSEXITCODE2(0, nd);
2825}
2826
2827/*
2828 * glue for fp.
2829 */
2830static int
2831fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2832{
2833	struct filedesc *fdp;
2834	struct file *fp;
2835	int error = 0;
2836
2837	fdp = p->td_proc->p_fd;
2838	if (fd < 0 || fd >= fdp->fd_nfiles ||
2839	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
2840		error = EBADF;
2841		goto out;
2842	}
2843	*fpp = fp;
2844
2845out:
2846	NFSEXITCODE(error);
2847	return (error);
2848}
2849
2850/*
2851 * Called from nfssvc() to update the exports list. Just call
2852 * vfs_export(). This has to be done, since the v4 root fake fs isn't
2853 * in the mount list.
2854 */
2855int
2856nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2857{
2858	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2859	int error = 0;
2860	struct nameidata nd;
2861	fhandle_t fh;
2862
2863	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2864	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2865		nfs_rootfhset = 0;
2866	else if (error == 0) {
2867		if (nfsexargp->fspec == NULL) {
2868			error = EPERM;
2869			goto out;
2870		}
2871		/*
2872		 * If fspec != NULL, this is the v4root path.
2873		 */
2874		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
2875		    nfsexargp->fspec, p);
2876		if ((error = namei(&nd)) != 0)
2877			goto out;
2878		error = nfsvno_getfh(nd.ni_vp, &fh, p);
2879		vrele(nd.ni_vp);
2880		if (!error) {
2881			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2882			NFSBCOPY((caddr_t)&fh,
2883			    nfs_rootfh.nfsrvfh_data,
2884			    sizeof (fhandle_t));
2885			nfs_rootfhset = 1;
2886		}
2887	}
2888
2889out:
2890	NFSEXITCODE(error);
2891	return (error);
2892}
2893
2894/*
2895 * This function needs to test to see if the system is near its limit
2896 * for memory allocation via malloc() or mget() and return True iff
2897 * either of these resources are near their limit.
2898 * XXX (For now, this is just a stub.)
2899 */
2900int nfsrv_testmalloclimit = 0;
2901int
2902nfsrv_mallocmget_limit(void)
2903{
2904	static int printmesg = 0;
2905	static int testval = 1;
2906
2907	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2908		if ((printmesg++ % 100) == 0)
2909			printf("nfsd: malloc/mget near limit\n");
2910		return (1);
2911	}
2912	return (0);
2913}
2914
2915/*
2916 * BSD specific initialization of a mount point.
2917 */
2918void
2919nfsd_mntinit(void)
2920{
2921	static int inited = 0;
2922
2923	if (inited)
2924		return;
2925	inited = 1;
2926	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2927	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2928	TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2929	nfsv4root_mnt.mnt_export = NULL;
2930	TAILQ_INIT(&nfsv4root_opt);
2931	TAILQ_INIT(&nfsv4root_newopt);
2932	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2933	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2934	nfsv4root_mnt.mnt_nvnodelistsize = 0;
2935	nfsv4root_mnt.mnt_activevnodelistsize = 0;
2936}
2937
2938/*
2939 * Get a vnode for a file handle, without checking exports, etc.
2940 */
2941struct vnode *
2942nfsvno_getvp(fhandle_t *fhp)
2943{
2944	struct mount *mp;
2945	struct vnode *vp;
2946	int error;
2947
2948	mp = vfs_busyfs(&fhp->fh_fsid);
2949	if (mp == NULL)
2950		return (NULL);
2951	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2952	vfs_unbusy(mp);
2953	if (error)
2954		return (NULL);
2955	return (vp);
2956}
2957
2958/*
2959 * Do a local VOP_ADVLOCK().
2960 */
2961int
2962nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2963    u_int64_t end, struct thread *td)
2964{
2965	int error = 0;
2966	struct flock fl;
2967	u_int64_t tlen;
2968
2969	if (nfsrv_dolocallocks == 0)
2970		goto out;
2971	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
2972
2973	fl.l_whence = SEEK_SET;
2974	fl.l_type = ftype;
2975	fl.l_start = (off_t)first;
2976	if (end == NFS64BITSSET) {
2977		fl.l_len = 0;
2978	} else {
2979		tlen = end - first;
2980		fl.l_len = (off_t)tlen;
2981	}
2982	/*
2983	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2984	 * values for all calls, so that all locks will be held by the
2985	 * nfsd server. (The nfsd server handles conflicts between the
2986	 * various clients.)
2987	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2988	 * bytes, so it can't be put in l_sysid.
2989	 */
2990	if (nfsv4_sysid == 0)
2991		nfsv4_sysid = nlm_acquire_next_sysid();
2992	fl.l_pid = (pid_t)0;
2993	fl.l_sysid = (int)nfsv4_sysid;
2994
2995	if (ftype == F_UNLCK)
2996		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
2997		    (F_POSIX | F_REMOTE));
2998	else
2999		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3000		    (F_POSIX | F_REMOTE));
3001
3002out:
3003	NFSEXITCODE(error);
3004	return (error);
3005}
3006
3007/*
3008 * Check the nfsv4 root exports.
3009 */
3010int
3011nfsvno_v4rootexport(struct nfsrv_descript *nd)
3012{
3013	struct ucred *credanon;
3014	int exflags, error = 0, numsecflavor, *secflavors, i;
3015
3016	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3017	    &credanon, &numsecflavor, &secflavors);
3018	if (error) {
3019		error = NFSERR_PROGUNAVAIL;
3020		goto out;
3021	}
3022	if (credanon != NULL)
3023		crfree(credanon);
3024	for (i = 0; i < numsecflavor; i++) {
3025		if (secflavors[i] == AUTH_SYS)
3026			nd->nd_flag |= ND_EXAUTHSYS;
3027		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3028			nd->nd_flag |= ND_EXGSS;
3029		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3030			nd->nd_flag |= ND_EXGSSINTEGRITY;
3031		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3032			nd->nd_flag |= ND_EXGSSPRIVACY;
3033	}
3034
3035out:
3036	NFSEXITCODE(error);
3037	return (error);
3038}
3039
3040/*
3041 * Nfs server psuedo system call for the nfsd's
3042 */
3043/*
3044 * MPSAFE
3045 */
3046static int
3047nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3048{
3049	struct file *fp;
3050	struct nfsd_addsock_args sockarg;
3051	struct nfsd_nfsd_args nfsdarg;
3052	cap_rights_t rights;
3053	int error;
3054
3055	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3056		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3057		if (error)
3058			goto out;
3059		/*
3060		 * Since we don't know what rights might be required,
3061		 * pretend that we need them all. It is better to be too
3062		 * careful than too reckless.
3063		 */
3064		error = fget(td, sockarg.sock,
3065		    cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3066		if (error != 0)
3067			goto out;
3068		if (fp->f_type != DTYPE_SOCKET) {
3069			fdrop(fp, td);
3070			error = EPERM;
3071			goto out;
3072		}
3073		error = nfsrvd_addsock(fp);
3074		fdrop(fp, td);
3075	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3076		if (uap->argp == NULL) {
3077			error = EINVAL;
3078			goto out;
3079		}
3080		error = copyin(uap->argp, (caddr_t)&nfsdarg,
3081		    sizeof (nfsdarg));
3082		if (error)
3083			goto out;
3084		error = nfsrvd_nfsd(td, &nfsdarg);
3085	} else {
3086		error = nfssvc_srvcall(td, uap, td->td_ucred);
3087	}
3088
3089out:
3090	NFSEXITCODE(error);
3091	return (error);
3092}
3093
3094static int
3095nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3096{
3097	struct nfsex_args export;
3098	struct file *fp = NULL;
3099	int stablefd, len;
3100	struct nfsd_clid adminrevoke;
3101	struct nfsd_dumplist dumplist;
3102	struct nfsd_dumpclients *dumpclients;
3103	struct nfsd_dumplocklist dumplocklist;
3104	struct nfsd_dumplocks *dumplocks;
3105	struct nameidata nd;
3106	vnode_t vp;
3107	int error = EINVAL, igotlock;
3108	struct proc *procp;
3109	static int suspend_nfsd = 0;
3110
3111	if (uap->flag & NFSSVC_PUBLICFH) {
3112		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3113		    sizeof (fhandle_t));
3114		error = copyin(uap->argp,
3115		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3116		if (!error)
3117			nfs_pubfhset = 1;
3118	} else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3119		error = copyin(uap->argp,(caddr_t)&export,
3120		    sizeof (struct nfsex_args));
3121		if (!error)
3122			error = nfsrv_v4rootexport(&export, cred, p);
3123	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3124		nfs_pubfhset = 0;
3125		error = 0;
3126	} else if (uap->flag & NFSSVC_STABLERESTART) {
3127		error = copyin(uap->argp, (caddr_t)&stablefd,
3128		    sizeof (int));
3129		if (!error)
3130			error = fp_getfvp(p, stablefd, &fp, &vp);
3131		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3132			error = EBADF;
3133		if (!error && newnfs_numnfsd != 0)
3134			error = EPERM;
3135		if (!error) {
3136			nfsrv_stablefirst.nsf_fp = fp;
3137			nfsrv_setupstable(p);
3138		}
3139	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3140		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3141		    sizeof (struct nfsd_clid));
3142		if (!error)
3143			error = nfsrv_adminrevoke(&adminrevoke, p);
3144	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3145		error = copyin(uap->argp, (caddr_t)&dumplist,
3146		    sizeof (struct nfsd_dumplist));
3147		if (!error && (dumplist.ndl_size < 1 ||
3148			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3149			error = EPERM;
3150		if (!error) {
3151		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3152		    dumpclients = (struct nfsd_dumpclients *)malloc(len,
3153			M_TEMP, M_WAITOK);
3154		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3155		    error = copyout(dumpclients,
3156			CAST_USER_ADDR_T(dumplist.ndl_list), len);
3157		    free((caddr_t)dumpclients, M_TEMP);
3158		}
3159	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3160		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3161		    sizeof (struct nfsd_dumplocklist));
3162		if (!error && (dumplocklist.ndllck_size < 1 ||
3163			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3164			error = EPERM;
3165		if (!error)
3166			error = nfsrv_lookupfilename(&nd,
3167				dumplocklist.ndllck_fname, p);
3168		if (!error) {
3169			len = sizeof (struct nfsd_dumplocks) *
3170				dumplocklist.ndllck_size;
3171			dumplocks = (struct nfsd_dumplocks *)malloc(len,
3172				M_TEMP, M_WAITOK);
3173			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3174			    dumplocklist.ndllck_size, p);
3175			vput(nd.ni_vp);
3176			error = copyout(dumplocks,
3177			    CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3178			free((caddr_t)dumplocks, M_TEMP);
3179		}
3180	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3181		procp = p->td_proc;
3182		PROC_LOCK(procp);
3183		nfsd_master_pid = procp->p_pid;
3184		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3185		nfsd_master_start = procp->p_stats->p_start;
3186		nfsd_master_proc = procp;
3187		PROC_UNLOCK(procp);
3188	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3189		NFSLOCKV4ROOTMUTEX();
3190		if (suspend_nfsd == 0) {
3191			/* Lock out all nfsd threads */
3192			do {
3193				igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3194				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3195			} while (igotlock == 0 && suspend_nfsd == 0);
3196			suspend_nfsd = 1;
3197		}
3198		NFSUNLOCKV4ROOTMUTEX();
3199		error = 0;
3200	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3201		NFSLOCKV4ROOTMUTEX();
3202		if (suspend_nfsd != 0) {
3203			nfsv4_unlock(&nfsd_suspend_lock, 0);
3204			suspend_nfsd = 0;
3205		}
3206		NFSUNLOCKV4ROOTMUTEX();
3207		error = 0;
3208	}
3209
3210	NFSEXITCODE(error);
3211	return (error);
3212}
3213
3214/*
3215 * Check exports.
3216 * Returns 0 if ok, 1 otherwise.
3217 */
3218int
3219nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3220{
3221	int i;
3222
3223	/*
3224	 * This seems odd, but allow the case where the security flavor
3225	 * list is empty. This happens when NFSv4 is traversing non-exported
3226	 * file systems. Exported file systems should always have a non-empty
3227	 * security flavor list.
3228	 */
3229	if (exp->nes_numsecflavor == 0)
3230		return (0);
3231
3232	for (i = 0; i < exp->nes_numsecflavor; i++) {
3233		/*
3234		 * The tests for privacy and integrity must be first,
3235		 * since ND_GSS is set for everything but AUTH_SYS.
3236		 */
3237		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3238		    (nd->nd_flag & ND_GSSPRIVACY))
3239			return (0);
3240		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3241		    (nd->nd_flag & ND_GSSINTEGRITY))
3242			return (0);
3243		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3244		    (nd->nd_flag & ND_GSS))
3245			return (0);
3246		if (exp->nes_secflavors[i] == AUTH_SYS &&
3247		    (nd->nd_flag & ND_GSS) == 0)
3248			return (0);
3249	}
3250	return (1);
3251}
3252
3253/*
3254 * Calculate a hash value for the fid in a file handle.
3255 */
3256uint32_t
3257nfsrv_hashfh(fhandle_t *fhp)
3258{
3259	uint32_t hashval;
3260
3261	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3262	return (hashval);
3263}
3264
3265/*
3266 * Calculate a hash value for the sessionid.
3267 */
3268uint32_t
3269nfsrv_hashsessionid(uint8_t *sessionid)
3270{
3271	uint32_t hashval;
3272
3273	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
3274	return (hashval);
3275}
3276
3277/*
3278 * Signal the userland master nfsd to backup the stable restart file.
3279 */
3280void
3281nfsrv_backupstable(void)
3282{
3283	struct proc *procp;
3284
3285	if (nfsd_master_proc != NULL) {
3286		procp = pfind(nfsd_master_pid);
3287		/* Try to make sure it is the correct process. */
3288		if (procp == nfsd_master_proc &&
3289		    procp->p_stats->p_start.tv_sec ==
3290		    nfsd_master_start.tv_sec &&
3291		    procp->p_stats->p_start.tv_usec ==
3292		    nfsd_master_start.tv_usec &&
3293		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
3294			kern_psignal(procp, SIGUSR2);
3295		else
3296			nfsd_master_proc = NULL;
3297
3298		if (procp != NULL)
3299			PROC_UNLOCK(procp);
3300	}
3301}
3302
3303extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3304
3305/*
3306 * Called once to initialize data structures...
3307 */
3308static int
3309nfsd_modevent(module_t mod, int type, void *data)
3310{
3311	int error = 0, i;
3312	static int loaded = 0;
3313
3314	switch (type) {
3315	case MOD_LOAD:
3316		if (loaded)
3317			goto out;
3318		newnfs_portinit();
3319		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3320			mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
3321			    MTX_DEF);
3322			mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
3323			    MTX_DEF);
3324		}
3325		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
3326		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
3327		mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
3328		for (i = 0; i < NFSSESSIONHASHSIZE; i++)
3329			mtx_init(&nfssessionhash[i].mtx, "nfssm",
3330			    NULL, MTX_DEF);
3331		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3332		nfsrvd_initcache();
3333		nfsd_init();
3334		NFSD_LOCK();
3335		nfsrvd_init(0);
3336		NFSD_UNLOCK();
3337		nfsd_mntinit();
3338#ifdef VV_DISABLEDELEG
3339		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3340		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3341#endif
3342		nfsd_call_servertimer = nfsrv_servertimer;
3343		nfsd_call_nfsd = nfssvc_nfsd;
3344		loaded = 1;
3345		break;
3346
3347	case MOD_UNLOAD:
3348		if (newnfs_numnfsd != 0) {
3349			error = EBUSY;
3350			break;
3351		}
3352
3353#ifdef VV_DISABLEDELEG
3354		vn_deleg_ops.vndeleg_recall = NULL;
3355		vn_deleg_ops.vndeleg_disable = NULL;
3356#endif
3357		nfsd_call_servertimer = NULL;
3358		nfsd_call_nfsd = NULL;
3359
3360		/* Clean out all NFSv4 state. */
3361		nfsrv_throwawayallstate(curthread);
3362
3363		/* Clean the NFS server reply cache */
3364		nfsrvd_cleancache();
3365
3366		/* Free up the krpc server pool. */
3367		if (nfsrvd_pool != NULL)
3368			svcpool_destroy(nfsrvd_pool);
3369
3370		/* and get rid of the locks */
3371		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3372			mtx_destroy(&nfsrchash_table[i].mtx);
3373			mtx_destroy(&nfsrcahash_table[i].mtx);
3374		}
3375		mtx_destroy(&nfsrc_udpmtx);
3376		mtx_destroy(&nfs_v4root_mutex);
3377		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3378		for (i = 0; i < NFSSESSIONHASHSIZE; i++)
3379			mtx_destroy(&nfssessionhash[i].mtx);
3380		lockdestroy(&nfsv4root_mnt.mnt_explock);
3381		loaded = 0;
3382		break;
3383	default:
3384		error = EOPNOTSUPP;
3385		break;
3386	}
3387
3388out:
3389	NFSEXITCODE(error);
3390	return (error);
3391}
3392static moduledata_t nfsd_mod = {
3393	"nfsd",
3394	nfsd_modevent,
3395	NULL,
3396};
3397DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3398
3399/* So that loader and kldload(2) can find us, wherever we are.. */
3400MODULE_VERSION(nfsd, 1);
3401MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3402MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3403MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3404MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3405MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3406
3407