xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision e52d92164754cbfff84767d4c6eb3cc93e8c21ae)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/capsicum.h>
40 
41 /*
42  * Functions that perform the vfs operations required by the routines in
43  * nfsd_serv.c. It is hoped that this change will make the server more
44  * portable.
45  */
46 
47 #include <fs/nfs/nfsport.h>
48 #include <sys/hash.h>
49 #include <sys/sysctl.h>
50 #include <nlm/nlm_prot.h>
51 #include <nlm/nlm.h>
52 
53 FEATURE(nfsd, "NFSv4 server");
54 
55 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
56 extern int nfsrv_useacl;
57 extern int newnfs_numnfsd;
58 extern struct mount nfsv4root_mnt;
59 extern struct nfsrv_stablefirst nfsrv_stablefirst;
60 extern void (*nfsd_call_servertimer)(void);
61 extern SVCPOOL	*nfsrvd_pool;
62 extern struct nfsv4lock nfsd_suspend_lock;
63 extern struct nfsclienthashhead *nfsclienthash;
64 extern struct nfslockhashhead *nfslockhash;
65 extern struct nfssessionhash *nfssessionhash;
66 extern int nfsrv_sessionhashsize;
67 extern struct nfsstatsv1 nfsstatsv1;
68 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
69 NFSDLOCKMUTEX;
70 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
71 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
72 struct mtx nfsrc_udpmtx;
73 struct mtx nfs_v4root_mutex;
74 struct nfsrvfh nfs_rootfh, nfs_pubfh;
75 int nfs_pubfhset = 0, nfs_rootfhset = 0;
76 struct proc *nfsd_master_proc = NULL;
77 int nfsd_debuglevel = 0;
78 static pid_t nfsd_master_pid = (pid_t)-1;
79 static char nfsd_master_comm[MAXCOMLEN + 1];
80 static struct timeval nfsd_master_start;
81 static uint32_t nfsv4_sysid = 0;
82 
83 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
84     struct ucred *);
85 
86 int nfsrv_enable_crossmntpt = 1;
87 static int nfs_commit_blks;
88 static int nfs_commit_miss;
89 extern int nfsrv_issuedelegs;
90 extern int nfsrv_dolocallocks;
91 extern int nfsd_enable_stringtouid;
92 
93 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "NFS server");
94 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
95     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
96 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
97     0, "");
98 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
99     0, "");
100 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
101     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
102 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
103     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
104 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
105     0, "Debug level for NFS server");
106 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
107     &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
108 
109 #define	MAX_REORDERED_RPC	16
110 #define	NUM_HEURISTIC		1031
111 #define	NHUSE_INIT		64
112 #define	NHUSE_INC		16
113 #define	NHUSE_MAX		2048
114 
115 static struct nfsheur {
116 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
117 	off_t nh_nextoff;	/* next offset for sequential detection */
118 	int nh_use;		/* use count for selection */
119 	int nh_seqcount;	/* heuristic */
120 } nfsheur[NUM_HEURISTIC];
121 
122 
123 /*
124  * Heuristic to detect sequential operation.
125  */
126 static struct nfsheur *
127 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
128 {
129 	struct nfsheur *nh;
130 	int hi, try;
131 
132 	/* Locate best candidate. */
133 	try = 32;
134 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
135 	nh = &nfsheur[hi];
136 	while (try--) {
137 		if (nfsheur[hi].nh_vp == vp) {
138 			nh = &nfsheur[hi];
139 			break;
140 		}
141 		if (nfsheur[hi].nh_use > 0)
142 			--nfsheur[hi].nh_use;
143 		hi = (hi + 1) % NUM_HEURISTIC;
144 		if (nfsheur[hi].nh_use < nh->nh_use)
145 			nh = &nfsheur[hi];
146 	}
147 
148 	/* Initialize hint if this is a new file. */
149 	if (nh->nh_vp != vp) {
150 		nh->nh_vp = vp;
151 		nh->nh_nextoff = uio->uio_offset;
152 		nh->nh_use = NHUSE_INIT;
153 		if (uio->uio_offset == 0)
154 			nh->nh_seqcount = 4;
155 		else
156 			nh->nh_seqcount = 1;
157 	}
158 
159 	/* Calculate heuristic. */
160 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
161 	    uio->uio_offset == nh->nh_nextoff) {
162 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
163 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
164 		if (nh->nh_seqcount > IO_SEQMAX)
165 			nh->nh_seqcount = IO_SEQMAX;
166 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
167 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
168 		/* Probably a reordered RPC, leave seqcount alone. */
169 	} else if (nh->nh_seqcount > 1) {
170 		nh->nh_seqcount /= 2;
171 	} else {
172 		nh->nh_seqcount = 0;
173 	}
174 	nh->nh_use += NHUSE_INC;
175 	if (nh->nh_use > NHUSE_MAX)
176 		nh->nh_use = NHUSE_MAX;
177 	return (nh);
178 }
179 
180 /*
181  * Get attributes into nfsvattr structure.
182  */
183 int
184 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
185     struct thread *p, int vpislocked)
186 {
187 	int error, lockedit = 0;
188 
189 	if (vpislocked == 0) {
190 		/*
191 		 * When vpislocked == 0, the vnode is either exclusively
192 		 * locked by this thread or not locked by this thread.
193 		 * As such, shared lock it, if not exclusively locked.
194 		 */
195 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
196 			lockedit = 1;
197 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
198 		}
199 	}
200 	error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
201 	if (lockedit != 0)
202 		NFSVOPUNLOCK(vp, 0);
203 
204 	NFSEXITCODE(error);
205 	return (error);
206 }
207 
208 /*
209  * Get a file handle for a vnode.
210  */
211 int
212 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
213 {
214 	int error;
215 
216 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
217 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
218 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
219 
220 	NFSEXITCODE(error);
221 	return (error);
222 }
223 
224 /*
225  * Perform access checking for vnodes obtained from file handles that would
226  * refer to files already opened by a Unix client. You cannot just use
227  * vn_writechk() and VOP_ACCESSX() for two reasons.
228  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
229  *     case.
230  * 2 - The owner is to be given access irrespective of mode bits for some
231  *     operations, so that processes that chmod after opening a file don't
232  *     break.
233  */
234 int
235 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
236     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
237     u_int32_t *supportedtypep)
238 {
239 	struct vattr vattr;
240 	int error = 0, getret = 0;
241 
242 	if (vpislocked == 0) {
243 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
244 			error = EPERM;
245 			goto out;
246 		}
247 	}
248 	if (accmode & VWRITE) {
249 		/* Just vn_writechk() changed to check rdonly */
250 		/*
251 		 * Disallow write attempts on read-only file systems;
252 		 * unless the file is a socket or a block or character
253 		 * device resident on the file system.
254 		 */
255 		if (NFSVNO_EXRDONLY(exp) ||
256 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
257 			switch (vp->v_type) {
258 			case VREG:
259 			case VDIR:
260 			case VLNK:
261 				error = EROFS;
262 			default:
263 				break;
264 			}
265 		}
266 		/*
267 		 * If there's shared text associated with
268 		 * the inode, try to free it up once.  If
269 		 * we fail, we can't allow writing.
270 		 */
271 		if (VOP_IS_TEXT(vp) && error == 0)
272 			error = ETXTBSY;
273 	}
274 	if (error != 0) {
275 		if (vpislocked == 0)
276 			NFSVOPUNLOCK(vp, 0);
277 		goto out;
278 	}
279 
280 	/*
281 	 * Should the override still be applied when ACLs are enabled?
282 	 */
283 	error = VOP_ACCESSX(vp, accmode, cred, p);
284 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
285 		/*
286 		 * Try again with VEXPLICIT_DENY, to see if the test for
287 		 * deletion is supported.
288 		 */
289 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
290 		if (error == 0) {
291 			if (vp->v_type == VDIR) {
292 				accmode &= ~(VDELETE | VDELETE_CHILD);
293 				accmode |= VWRITE;
294 				error = VOP_ACCESSX(vp, accmode, cred, p);
295 			} else if (supportedtypep != NULL) {
296 				*supportedtypep &= ~NFSACCESS_DELETE;
297 			}
298 		}
299 	}
300 
301 	/*
302 	 * Allow certain operations for the owner (reads and writes
303 	 * on files that are already open).
304 	 */
305 	if (override != NFSACCCHK_NOOVERRIDE &&
306 	    (error == EPERM || error == EACCES)) {
307 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
308 			error = 0;
309 		else if (override & NFSACCCHK_ALLOWOWNER) {
310 			getret = VOP_GETATTR(vp, &vattr, cred);
311 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
312 				error = 0;
313 		}
314 	}
315 	if (vpislocked == 0)
316 		NFSVOPUNLOCK(vp, 0);
317 
318 out:
319 	NFSEXITCODE(error);
320 	return (error);
321 }
322 
323 /*
324  * Set attribute(s) vnop.
325  */
326 int
327 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
328     struct thread *p, struct nfsexstuff *exp)
329 {
330 	int error;
331 
332 	error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
333 	NFSEXITCODE(error);
334 	return (error);
335 }
336 
337 /*
338  * Set up nameidata for a lookup() call and do it.
339  */
340 int
341 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
342     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
343     struct vnode **retdirp)
344 {
345 	struct componentname *cnp = &ndp->ni_cnd;
346 	int i;
347 	struct iovec aiov;
348 	struct uio auio;
349 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
350 	int error = 0;
351 	char *cp;
352 
353 	*retdirp = NULL;
354 	cnp->cn_nameptr = cnp->cn_pnbuf;
355 	ndp->ni_lcf = 0;
356 	/*
357 	 * Extract and set starting directory.
358 	 */
359 	if (dp->v_type != VDIR) {
360 		if (islocked)
361 			vput(dp);
362 		else
363 			vrele(dp);
364 		nfsvno_relpathbuf(ndp);
365 		error = ENOTDIR;
366 		goto out1;
367 	}
368 	if (islocked)
369 		NFSVOPUNLOCK(dp, 0);
370 	VREF(dp);
371 	*retdirp = dp;
372 	if (NFSVNO_EXRDONLY(exp))
373 		cnp->cn_flags |= RDONLY;
374 	ndp->ni_segflg = UIO_SYSSPACE;
375 
376 	if (nd->nd_flag & ND_PUBLOOKUP) {
377 		ndp->ni_loopcnt = 0;
378 		if (cnp->cn_pnbuf[0] == '/') {
379 			vrele(dp);
380 			/*
381 			 * Check for degenerate pathnames here, since lookup()
382 			 * panics on them.
383 			 */
384 			for (i = 1; i < ndp->ni_pathlen; i++)
385 				if (cnp->cn_pnbuf[i] != '/')
386 					break;
387 			if (i == ndp->ni_pathlen) {
388 				error = NFSERR_ACCES;
389 				goto out;
390 			}
391 			dp = rootvnode;
392 			VREF(dp);
393 		}
394 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
395 	    (nd->nd_flag & ND_NFSV4) == 0) {
396 		/*
397 		 * Only cross mount points for NFSv4 when doing a
398 		 * mount while traversing the file system above
399 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
400 		 */
401 		cnp->cn_flags |= NOCROSSMOUNT;
402 	}
403 
404 	/*
405 	 * Initialize for scan, set ni_startdir and bump ref on dp again
406 	 * because lookup() will dereference ni_startdir.
407 	 */
408 
409 	cnp->cn_thread = p;
410 	ndp->ni_startdir = dp;
411 	ndp->ni_rootdir = rootvnode;
412 	ndp->ni_topdir = NULL;
413 
414 	if (!lockleaf)
415 		cnp->cn_flags |= LOCKLEAF;
416 	for (;;) {
417 		cnp->cn_nameptr = cnp->cn_pnbuf;
418 		/*
419 		 * Call lookup() to do the real work.  If an error occurs,
420 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
421 		 * we do not have to dereference anything before returning.
422 		 * In either case ni_startdir will be dereferenced and NULLed
423 		 * out.
424 		 */
425 		error = lookup(ndp);
426 		if (error)
427 			break;
428 
429 		/*
430 		 * Check for encountering a symbolic link.  Trivial
431 		 * termination occurs if no symlink encountered.
432 		 */
433 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
434 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
435 				nfsvno_relpathbuf(ndp);
436 			if (ndp->ni_vp && !lockleaf)
437 				NFSVOPUNLOCK(ndp->ni_vp, 0);
438 			break;
439 		}
440 
441 		/*
442 		 * Validate symlink
443 		 */
444 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
445 			NFSVOPUNLOCK(ndp->ni_dvp, 0);
446 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
447 			error = EINVAL;
448 			goto badlink2;
449 		}
450 
451 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
452 			error = ELOOP;
453 			goto badlink2;
454 		}
455 		if (ndp->ni_pathlen > 1)
456 			cp = uma_zalloc(namei_zone, M_WAITOK);
457 		else
458 			cp = cnp->cn_pnbuf;
459 		aiov.iov_base = cp;
460 		aiov.iov_len = MAXPATHLEN;
461 		auio.uio_iov = &aiov;
462 		auio.uio_iovcnt = 1;
463 		auio.uio_offset = 0;
464 		auio.uio_rw = UIO_READ;
465 		auio.uio_segflg = UIO_SYSSPACE;
466 		auio.uio_td = NULL;
467 		auio.uio_resid = MAXPATHLEN;
468 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
469 		if (error) {
470 		badlink1:
471 			if (ndp->ni_pathlen > 1)
472 				uma_zfree(namei_zone, cp);
473 		badlink2:
474 			vrele(ndp->ni_dvp);
475 			vput(ndp->ni_vp);
476 			break;
477 		}
478 		linklen = MAXPATHLEN - auio.uio_resid;
479 		if (linklen == 0) {
480 			error = ENOENT;
481 			goto badlink1;
482 		}
483 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
484 			error = ENAMETOOLONG;
485 			goto badlink1;
486 		}
487 
488 		/*
489 		 * Adjust or replace path
490 		 */
491 		if (ndp->ni_pathlen > 1) {
492 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
493 			uma_zfree(namei_zone, cnp->cn_pnbuf);
494 			cnp->cn_pnbuf = cp;
495 		} else
496 			cnp->cn_pnbuf[linklen] = '\0';
497 		ndp->ni_pathlen += linklen;
498 
499 		/*
500 		 * Cleanup refs for next loop and check if root directory
501 		 * should replace current directory.  Normally ni_dvp
502 		 * becomes the new base directory and is cleaned up when
503 		 * we loop.  Explicitly null pointers after invalidation
504 		 * to clarify operation.
505 		 */
506 		vput(ndp->ni_vp);
507 		ndp->ni_vp = NULL;
508 
509 		if (cnp->cn_pnbuf[0] == '/') {
510 			vrele(ndp->ni_dvp);
511 			ndp->ni_dvp = ndp->ni_rootdir;
512 			VREF(ndp->ni_dvp);
513 		}
514 		ndp->ni_startdir = ndp->ni_dvp;
515 		ndp->ni_dvp = NULL;
516 	}
517 	if (!lockleaf)
518 		cnp->cn_flags &= ~LOCKLEAF;
519 
520 out:
521 	if (error) {
522 		nfsvno_relpathbuf(ndp);
523 		ndp->ni_vp = NULL;
524 		ndp->ni_dvp = NULL;
525 		ndp->ni_startdir = NULL;
526 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
527 		ndp->ni_dvp = NULL;
528 	}
529 
530 out1:
531 	NFSEXITCODE2(error, nd);
532 	return (error);
533 }
534 
535 /*
536  * Set up a pathname buffer and return a pointer to it and, optionally
537  * set a hash pointer.
538  */
539 void
540 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
541 {
542 	struct componentname *cnp = &ndp->ni_cnd;
543 
544 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
545 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
546 	if (hashpp != NULL)
547 		*hashpp = NULL;
548 	*bufpp = cnp->cn_pnbuf;
549 }
550 
551 /*
552  * Release the above path buffer, if not released by nfsvno_namei().
553  */
554 void
555 nfsvno_relpathbuf(struct nameidata *ndp)
556 {
557 
558 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
559 		panic("nfsrelpath");
560 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
561 	ndp->ni_cnd.cn_flags &= ~HASBUF;
562 }
563 
564 /*
565  * Readlink vnode op into an mbuf list.
566  */
567 int
568 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
569     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
570 {
571 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
572 	struct iovec *ivp = iv;
573 	struct uio io, *uiop = &io;
574 	struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
575 	int i, len, tlen, error = 0;
576 
577 	len = 0;
578 	i = 0;
579 	while (len < NFS_MAXPATHLEN) {
580 		NFSMGET(mp);
581 		MCLGET(mp, M_WAITOK);
582 		mp->m_len = M_SIZE(mp);
583 		if (len == 0) {
584 			mp3 = mp2 = mp;
585 		} else {
586 			mp2->m_next = mp;
587 			mp2 = mp;
588 		}
589 		if ((len + mp->m_len) > NFS_MAXPATHLEN) {
590 			mp->m_len = NFS_MAXPATHLEN - len;
591 			len = NFS_MAXPATHLEN;
592 		} else {
593 			len += mp->m_len;
594 		}
595 		ivp->iov_base = mtod(mp, caddr_t);
596 		ivp->iov_len = mp->m_len;
597 		i++;
598 		ivp++;
599 	}
600 	uiop->uio_iov = iv;
601 	uiop->uio_iovcnt = i;
602 	uiop->uio_offset = 0;
603 	uiop->uio_resid = len;
604 	uiop->uio_rw = UIO_READ;
605 	uiop->uio_segflg = UIO_SYSSPACE;
606 	uiop->uio_td = NULL;
607 	error = VOP_READLINK(vp, uiop, cred);
608 	if (error) {
609 		m_freem(mp3);
610 		*lenp = 0;
611 		goto out;
612 	}
613 	if (uiop->uio_resid > 0) {
614 		len -= uiop->uio_resid;
615 		tlen = NFSM_RNDUP(len);
616 		nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
617 	}
618 	*lenp = len;
619 	*mpp = mp3;
620 	*mpendp = mp;
621 
622 out:
623 	NFSEXITCODE(error);
624 	return (error);
625 }
626 
627 /*
628  * Read vnode op call into mbuf list.
629  */
630 int
631 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
632     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
633 {
634 	struct mbuf *m;
635 	int i;
636 	struct iovec *iv;
637 	struct iovec *iv2;
638 	int error = 0, len, left, siz, tlen, ioflag = 0;
639 	struct mbuf *m2 = NULL, *m3;
640 	struct uio io, *uiop = &io;
641 	struct nfsheur *nh;
642 
643 	len = left = NFSM_RNDUP(cnt);
644 	m3 = NULL;
645 	/*
646 	 * Generate the mbuf list with the uio_iov ref. to it.
647 	 */
648 	i = 0;
649 	while (left > 0) {
650 		NFSMGET(m);
651 		MCLGET(m, M_WAITOK);
652 		m->m_len = 0;
653 		siz = min(M_TRAILINGSPACE(m), left);
654 		left -= siz;
655 		i++;
656 		if (m3)
657 			m2->m_next = m;
658 		else
659 			m3 = m;
660 		m2 = m;
661 	}
662 	iv = malloc(i * sizeof (struct iovec),
663 	    M_TEMP, M_WAITOK);
664 	uiop->uio_iov = iv2 = iv;
665 	m = m3;
666 	left = len;
667 	i = 0;
668 	while (left > 0) {
669 		if (m == NULL)
670 			panic("nfsvno_read iov");
671 		siz = min(M_TRAILINGSPACE(m), left);
672 		if (siz > 0) {
673 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
674 			iv->iov_len = siz;
675 			m->m_len += siz;
676 			left -= siz;
677 			iv++;
678 			i++;
679 		}
680 		m = m->m_next;
681 	}
682 	uiop->uio_iovcnt = i;
683 	uiop->uio_offset = off;
684 	uiop->uio_resid = len;
685 	uiop->uio_rw = UIO_READ;
686 	uiop->uio_segflg = UIO_SYSSPACE;
687 	uiop->uio_td = NULL;
688 	nh = nfsrv_sequential_heuristic(uiop, vp);
689 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
690 	/* XXX KDM make this more systematic? */
691 	nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid;
692 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
693 	free(iv2, M_TEMP);
694 	if (error) {
695 		m_freem(m3);
696 		*mpp = NULL;
697 		goto out;
698 	}
699 	nh->nh_nextoff = uiop->uio_offset;
700 	tlen = len - uiop->uio_resid;
701 	cnt = cnt < tlen ? cnt : tlen;
702 	tlen = NFSM_RNDUP(cnt);
703 	if (tlen == 0) {
704 		m_freem(m3);
705 		m3 = NULL;
706 	} else if (len != tlen || tlen != cnt)
707 		nfsrv_adj(m3, len - tlen, tlen - cnt);
708 	*mpp = m3;
709 	*mpendp = m2;
710 
711 out:
712 	NFSEXITCODE(error);
713 	return (error);
714 }
715 
716 /*
717  * Write vnode op from an mbuf list.
718  */
719 int
720 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
721     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
722 {
723 	struct iovec *ivp;
724 	int i, len;
725 	struct iovec *iv;
726 	int ioflags, error;
727 	struct uio io, *uiop = &io;
728 	struct nfsheur *nh;
729 
730 	ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
731 	    M_WAITOK);
732 	uiop->uio_iov = iv = ivp;
733 	uiop->uio_iovcnt = cnt;
734 	i = mtod(mp, caddr_t) + mp->m_len - cp;
735 	len = retlen;
736 	while (len > 0) {
737 		if (mp == NULL)
738 			panic("nfsvno_write");
739 		if (i > 0) {
740 			i = min(i, len);
741 			ivp->iov_base = cp;
742 			ivp->iov_len = i;
743 			ivp++;
744 			len -= i;
745 		}
746 		mp = mp->m_next;
747 		if (mp) {
748 			i = mp->m_len;
749 			cp = mtod(mp, caddr_t);
750 		}
751 	}
752 
753 	if (stable == NFSWRITE_UNSTABLE)
754 		ioflags = IO_NODELOCKED;
755 	else
756 		ioflags = (IO_SYNC | IO_NODELOCKED);
757 	uiop->uio_resid = retlen;
758 	uiop->uio_rw = UIO_WRITE;
759 	uiop->uio_segflg = UIO_SYSSPACE;
760 	NFSUIOPROC(uiop, p);
761 	uiop->uio_offset = off;
762 	nh = nfsrv_sequential_heuristic(uiop, vp);
763 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
764 	/* XXX KDM make this more systematic? */
765 	nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
766 	error = VOP_WRITE(vp, uiop, ioflags, cred);
767 	if (error == 0)
768 		nh->nh_nextoff = uiop->uio_offset;
769 	free(iv, M_TEMP);
770 
771 	NFSEXITCODE(error);
772 	return (error);
773 }
774 
775 /*
776  * Common code for creating a regular file (plus special files for V2).
777  */
778 int
779 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
780     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
781     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
782 {
783 	u_quad_t tempsize;
784 	int error;
785 
786 	error = nd->nd_repstat;
787 	if (!error && ndp->ni_vp == NULL) {
788 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
789 			vrele(ndp->ni_startdir);
790 			error = VOP_CREATE(ndp->ni_dvp,
791 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
792 			vput(ndp->ni_dvp);
793 			nfsvno_relpathbuf(ndp);
794 			if (!error) {
795 				if (*exclusive_flagp) {
796 					*exclusive_flagp = 0;
797 					NFSVNO_ATTRINIT(nvap);
798 					nvap->na_atime.tv_sec = cverf[0];
799 					nvap->na_atime.tv_nsec = cverf[1];
800 					error = VOP_SETATTR(ndp->ni_vp,
801 					    &nvap->na_vattr, nd->nd_cred);
802 					if (error != 0) {
803 						vput(ndp->ni_vp);
804 						ndp->ni_vp = NULL;
805 						error = NFSERR_NOTSUPP;
806 					}
807 				}
808 			}
809 		/*
810 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
811 		 * (This implies, just get out on an error.)
812 		 */
813 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
814 			nvap->na_type == VFIFO) {
815 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
816 				nvap->na_type = VFIFO;
817                         if (nvap->na_type != VFIFO &&
818 			    (error = priv_check_cred(nd->nd_cred,
819 			     PRIV_VFS_MKNOD_DEV, 0))) {
820 				vrele(ndp->ni_startdir);
821 				nfsvno_relpathbuf(ndp);
822 				vput(ndp->ni_dvp);
823 				goto out;
824 			}
825 			nvap->na_rdev = rdev;
826 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
827 			    &ndp->ni_cnd, &nvap->na_vattr);
828 			vput(ndp->ni_dvp);
829 			nfsvno_relpathbuf(ndp);
830 			vrele(ndp->ni_startdir);
831 			if (error)
832 				goto out;
833 		} else {
834 			vrele(ndp->ni_startdir);
835 			nfsvno_relpathbuf(ndp);
836 			vput(ndp->ni_dvp);
837 			error = ENXIO;
838 			goto out;
839 		}
840 		*vpp = ndp->ni_vp;
841 	} else {
842 		/*
843 		 * Handle cases where error is already set and/or
844 		 * the file exists.
845 		 * 1 - clean up the lookup
846 		 * 2 - iff !error and na_size set, truncate it
847 		 */
848 		vrele(ndp->ni_startdir);
849 		nfsvno_relpathbuf(ndp);
850 		*vpp = ndp->ni_vp;
851 		if (ndp->ni_dvp == *vpp)
852 			vrele(ndp->ni_dvp);
853 		else
854 			vput(ndp->ni_dvp);
855 		if (!error && nvap->na_size != VNOVAL) {
856 			error = nfsvno_accchk(*vpp, VWRITE,
857 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
858 			    NFSACCCHK_VPISLOCKED, NULL);
859 			if (!error) {
860 				tempsize = nvap->na_size;
861 				NFSVNO_ATTRINIT(nvap);
862 				nvap->na_size = tempsize;
863 				error = VOP_SETATTR(*vpp,
864 				    &nvap->na_vattr, nd->nd_cred);
865 			}
866 		}
867 		if (error)
868 			vput(*vpp);
869 	}
870 
871 out:
872 	NFSEXITCODE(error);
873 	return (error);
874 }
875 
876 /*
877  * Do a mknod vnode op.
878  */
879 int
880 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
881     struct thread *p)
882 {
883 	int error = 0;
884 	enum vtype vtyp;
885 
886 	vtyp = nvap->na_type;
887 	/*
888 	 * Iff doesn't exist, create it.
889 	 */
890 	if (ndp->ni_vp) {
891 		vrele(ndp->ni_startdir);
892 		nfsvno_relpathbuf(ndp);
893 		vput(ndp->ni_dvp);
894 		vrele(ndp->ni_vp);
895 		error = EEXIST;
896 		goto out;
897 	}
898 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
899 		vrele(ndp->ni_startdir);
900 		nfsvno_relpathbuf(ndp);
901 		vput(ndp->ni_dvp);
902 		error = NFSERR_BADTYPE;
903 		goto out;
904 	}
905 	if (vtyp == VSOCK) {
906 		vrele(ndp->ni_startdir);
907 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
908 		    &ndp->ni_cnd, &nvap->na_vattr);
909 		vput(ndp->ni_dvp);
910 		nfsvno_relpathbuf(ndp);
911 	} else {
912 		if (nvap->na_type != VFIFO &&
913 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
914 			vrele(ndp->ni_startdir);
915 			nfsvno_relpathbuf(ndp);
916 			vput(ndp->ni_dvp);
917 			goto out;
918 		}
919 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
920 		    &ndp->ni_cnd, &nvap->na_vattr);
921 		vput(ndp->ni_dvp);
922 		nfsvno_relpathbuf(ndp);
923 		vrele(ndp->ni_startdir);
924 		/*
925 		 * Since VOP_MKNOD returns the ni_vp, I can't
926 		 * see any reason to do the lookup.
927 		 */
928 	}
929 
930 out:
931 	NFSEXITCODE(error);
932 	return (error);
933 }
934 
935 /*
936  * Mkdir vnode op.
937  */
938 int
939 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
940     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
941 {
942 	int error = 0;
943 
944 	if (ndp->ni_vp != NULL) {
945 		if (ndp->ni_dvp == ndp->ni_vp)
946 			vrele(ndp->ni_dvp);
947 		else
948 			vput(ndp->ni_dvp);
949 		vrele(ndp->ni_vp);
950 		nfsvno_relpathbuf(ndp);
951 		error = EEXIST;
952 		goto out;
953 	}
954 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
955 	    &nvap->na_vattr);
956 	vput(ndp->ni_dvp);
957 	nfsvno_relpathbuf(ndp);
958 
959 out:
960 	NFSEXITCODE(error);
961 	return (error);
962 }
963 
964 /*
965  * symlink vnode op.
966  */
967 int
968 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
969     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
970     struct nfsexstuff *exp)
971 {
972 	int error = 0;
973 
974 	if (ndp->ni_vp) {
975 		vrele(ndp->ni_startdir);
976 		nfsvno_relpathbuf(ndp);
977 		if (ndp->ni_dvp == ndp->ni_vp)
978 			vrele(ndp->ni_dvp);
979 		else
980 			vput(ndp->ni_dvp);
981 		vrele(ndp->ni_vp);
982 		error = EEXIST;
983 		goto out;
984 	}
985 
986 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
987 	    &nvap->na_vattr, pathcp);
988 	vput(ndp->ni_dvp);
989 	vrele(ndp->ni_startdir);
990 	nfsvno_relpathbuf(ndp);
991 	/*
992 	 * Although FreeBSD still had the lookup code in
993 	 * it for 7/current, there doesn't seem to be any
994 	 * point, since VOP_SYMLINK() returns the ni_vp.
995 	 * Just vput it for v2.
996 	 */
997 	if (!not_v2 && !error)
998 		vput(ndp->ni_vp);
999 
1000 out:
1001 	NFSEXITCODE(error);
1002 	return (error);
1003 }
1004 
1005 /*
1006  * Parse symbolic link arguments.
1007  * This function has an ugly side effect. It will malloc() an area for
1008  * the symlink and set iov_base to point to it, only if it succeeds.
1009  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1010  * be FREE'd later.
1011  */
1012 int
1013 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1014     struct thread *p, char **pathcpp, int *lenp)
1015 {
1016 	u_int32_t *tl;
1017 	char *pathcp = NULL;
1018 	int error = 0, len;
1019 	struct nfsv2_sattr *sp;
1020 
1021 	*pathcpp = NULL;
1022 	*lenp = 0;
1023 	if ((nd->nd_flag & ND_NFSV3) &&
1024 	    (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1025 		goto nfsmout;
1026 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1027 	len = fxdr_unsigned(int, *tl);
1028 	if (len > NFS_MAXPATHLEN || len <= 0) {
1029 		error = EBADRPC;
1030 		goto nfsmout;
1031 	}
1032 	pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1033 	error = nfsrv_mtostr(nd, pathcp, len);
1034 	if (error)
1035 		goto nfsmout;
1036 	if (nd->nd_flag & ND_NFSV2) {
1037 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1038 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1039 	}
1040 	*pathcpp = pathcp;
1041 	*lenp = len;
1042 	NFSEXITCODE2(0, nd);
1043 	return (0);
1044 nfsmout:
1045 	if (pathcp)
1046 		free(pathcp, M_TEMP);
1047 	NFSEXITCODE2(error, nd);
1048 	return (error);
1049 }
1050 
1051 /*
1052  * Remove a non-directory object.
1053  */
1054 int
1055 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1056     struct thread *p, struct nfsexstuff *exp)
1057 {
1058 	struct vnode *vp;
1059 	int error = 0;
1060 
1061 	vp = ndp->ni_vp;
1062 	if (vp->v_type == VDIR)
1063 		error = NFSERR_ISDIR;
1064 	else if (is_v4)
1065 		error = nfsrv_checkremove(vp, 1, p);
1066 	if (!error)
1067 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1068 	if (ndp->ni_dvp == vp)
1069 		vrele(ndp->ni_dvp);
1070 	else
1071 		vput(ndp->ni_dvp);
1072 	vput(vp);
1073 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1074 		nfsvno_relpathbuf(ndp);
1075 	NFSEXITCODE(error);
1076 	return (error);
1077 }
1078 
1079 /*
1080  * Remove a directory.
1081  */
1082 int
1083 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1084     struct thread *p, struct nfsexstuff *exp)
1085 {
1086 	struct vnode *vp;
1087 	int error = 0;
1088 
1089 	vp = ndp->ni_vp;
1090 	if (vp->v_type != VDIR) {
1091 		error = ENOTDIR;
1092 		goto out;
1093 	}
1094 	/*
1095 	 * No rmdir "." please.
1096 	 */
1097 	if (ndp->ni_dvp == vp) {
1098 		error = EINVAL;
1099 		goto out;
1100 	}
1101 	/*
1102 	 * The root of a mounted filesystem cannot be deleted.
1103 	 */
1104 	if (vp->v_vflag & VV_ROOT)
1105 		error = EBUSY;
1106 out:
1107 	if (!error)
1108 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1109 	if (ndp->ni_dvp == vp)
1110 		vrele(ndp->ni_dvp);
1111 	else
1112 		vput(ndp->ni_dvp);
1113 	vput(vp);
1114 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1115 		nfsvno_relpathbuf(ndp);
1116 	NFSEXITCODE(error);
1117 	return (error);
1118 }
1119 
1120 /*
1121  * Rename vnode op.
1122  */
1123 int
1124 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1125     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1126 {
1127 	struct vnode *fvp, *tvp, *tdvp;
1128 	int error = 0;
1129 
1130 	fvp = fromndp->ni_vp;
1131 	if (ndstat) {
1132 		vrele(fromndp->ni_dvp);
1133 		vrele(fvp);
1134 		error = ndstat;
1135 		goto out1;
1136 	}
1137 	tdvp = tondp->ni_dvp;
1138 	tvp = tondp->ni_vp;
1139 	if (tvp != NULL) {
1140 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1141 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1142 			goto out;
1143 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1144 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1145 			goto out;
1146 		}
1147 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1148 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1149 			goto out;
1150 		}
1151 
1152 		/*
1153 		 * A rename to '.' or '..' results in a prematurely
1154 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1155 		 * here.
1156 		 */
1157 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1158 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1159 		    (tondp->ni_cnd.cn_namelen == 2 &&
1160 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1161 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1162 			error = EINVAL;
1163 			goto out;
1164 		}
1165 	}
1166 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1167 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1168 		goto out;
1169 	}
1170 	if (fvp->v_mount != tdvp->v_mount) {
1171 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1172 		goto out;
1173 	}
1174 	if (fvp == tdvp) {
1175 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1176 		goto out;
1177 	}
1178 	if (fvp == tvp) {
1179 		/*
1180 		 * If source and destination are the same, there is nothing to
1181 		 * do. Set error to -1 to indicate this.
1182 		 */
1183 		error = -1;
1184 		goto out;
1185 	}
1186 	if (ndflag & ND_NFSV4) {
1187 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1188 			error = nfsrv_checkremove(fvp, 0, p);
1189 			NFSVOPUNLOCK(fvp, 0);
1190 		} else
1191 			error = EPERM;
1192 		if (tvp && !error)
1193 			error = nfsrv_checkremove(tvp, 1, p);
1194 	} else {
1195 		/*
1196 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1197 		 * that the NFSv4 client won't be confused by the rename.
1198 		 * Since nfsd_recalldelegation() can only be called on an
1199 		 * unlocked vnode at this point and fvp is the file that will
1200 		 * still exist after the rename, just do fvp.
1201 		 */
1202 		nfsd_recalldelegation(fvp, p);
1203 	}
1204 out:
1205 	if (!error) {
1206 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1207 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1208 		    &tondp->ni_cnd);
1209 	} else {
1210 		if (tdvp == tvp)
1211 			vrele(tdvp);
1212 		else
1213 			vput(tdvp);
1214 		if (tvp)
1215 			vput(tvp);
1216 		vrele(fromndp->ni_dvp);
1217 		vrele(fvp);
1218 		if (error == -1)
1219 			error = 0;
1220 	}
1221 	vrele(tondp->ni_startdir);
1222 	nfsvno_relpathbuf(tondp);
1223 out1:
1224 	vrele(fromndp->ni_startdir);
1225 	nfsvno_relpathbuf(fromndp);
1226 	NFSEXITCODE(error);
1227 	return (error);
1228 }
1229 
1230 /*
1231  * Link vnode op.
1232  */
1233 int
1234 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1235     struct thread *p, struct nfsexstuff *exp)
1236 {
1237 	struct vnode *xp;
1238 	int error = 0;
1239 
1240 	xp = ndp->ni_vp;
1241 	if (xp != NULL) {
1242 		error = EEXIST;
1243 	} else {
1244 		xp = ndp->ni_dvp;
1245 		if (vp->v_mount != xp->v_mount)
1246 			error = EXDEV;
1247 	}
1248 	if (!error) {
1249 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1250 		if ((vp->v_iflag & VI_DOOMED) == 0)
1251 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1252 		else
1253 			error = EPERM;
1254 		if (ndp->ni_dvp == vp)
1255 			vrele(ndp->ni_dvp);
1256 		else
1257 			vput(ndp->ni_dvp);
1258 		NFSVOPUNLOCK(vp, 0);
1259 	} else {
1260 		if (ndp->ni_dvp == ndp->ni_vp)
1261 			vrele(ndp->ni_dvp);
1262 		else
1263 			vput(ndp->ni_dvp);
1264 		if (ndp->ni_vp)
1265 			vrele(ndp->ni_vp);
1266 	}
1267 	nfsvno_relpathbuf(ndp);
1268 	NFSEXITCODE(error);
1269 	return (error);
1270 }
1271 
1272 /*
1273  * Do the fsync() appropriate for the commit.
1274  */
1275 int
1276 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1277     struct thread *td)
1278 {
1279 	int error = 0;
1280 
1281 	/*
1282 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1283 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1284 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1285 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1286 	 * File systems that do not use the buffer cache (as indicated
1287 	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1288 	 */
1289 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1290 	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1291 		/*
1292 		 * Give up and do the whole thing
1293 		 */
1294 		if (vp->v_object &&
1295 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1296 			VM_OBJECT_WLOCK(vp->v_object);
1297 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1298 			VM_OBJECT_WUNLOCK(vp->v_object);
1299 		}
1300 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1301 	} else {
1302 		/*
1303 		 * Locate and synchronously write any buffers that fall
1304 		 * into the requested range.  Note:  we are assuming that
1305 		 * f_iosize is a power of 2.
1306 		 */
1307 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1308 		int iomask = iosize - 1;
1309 		struct bufobj *bo;
1310 		daddr_t lblkno;
1311 
1312 		/*
1313 		 * Align to iosize boundary, super-align to page boundary.
1314 		 */
1315 		if (off & iomask) {
1316 			cnt += off & iomask;
1317 			off &= ~(u_quad_t)iomask;
1318 		}
1319 		if (off & PAGE_MASK) {
1320 			cnt += off & PAGE_MASK;
1321 			off &= ~(u_quad_t)PAGE_MASK;
1322 		}
1323 		lblkno = off / iosize;
1324 
1325 		if (vp->v_object &&
1326 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1327 			VM_OBJECT_WLOCK(vp->v_object);
1328 			vm_object_page_clean(vp->v_object, off, off + cnt,
1329 			    OBJPC_SYNC);
1330 			VM_OBJECT_WUNLOCK(vp->v_object);
1331 		}
1332 
1333 		bo = &vp->v_bufobj;
1334 		BO_LOCK(bo);
1335 		while (cnt > 0) {
1336 			struct buf *bp;
1337 
1338 			/*
1339 			 * If we have a buffer and it is marked B_DELWRI we
1340 			 * have to lock and write it.  Otherwise the prior
1341 			 * write is assumed to have already been committed.
1342 			 *
1343 			 * gbincore() can return invalid buffers now so we
1344 			 * have to check that bit as well (though B_DELWRI
1345 			 * should not be set if B_INVAL is set there could be
1346 			 * a race here since we haven't locked the buffer).
1347 			 */
1348 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1349 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1350 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1351 					BO_LOCK(bo);
1352 					continue; /* retry */
1353 				}
1354 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1355 				    B_DELWRI) {
1356 					bremfree(bp);
1357 					bp->b_flags &= ~B_ASYNC;
1358 					bwrite(bp);
1359 					++nfs_commit_miss;
1360 				} else
1361 					BUF_UNLOCK(bp);
1362 				BO_LOCK(bo);
1363 			}
1364 			++nfs_commit_blks;
1365 			if (cnt < iosize)
1366 				break;
1367 			cnt -= iosize;
1368 			++lblkno;
1369 		}
1370 		BO_UNLOCK(bo);
1371 	}
1372 	NFSEXITCODE(error);
1373 	return (error);
1374 }
1375 
1376 /*
1377  * Statfs vnode op.
1378  */
1379 int
1380 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1381 {
1382 	int error;
1383 
1384 	error = VFS_STATFS(vp->v_mount, sf);
1385 	if (error == 0) {
1386 		/*
1387 		 * Since NFS handles these values as unsigned on the
1388 		 * wire, there is no way to represent negative values,
1389 		 * so set them to 0. Without this, they will appear
1390 		 * to be very large positive values for clients like
1391 		 * Solaris10.
1392 		 */
1393 		if (sf->f_bavail < 0)
1394 			sf->f_bavail = 0;
1395 		if (sf->f_ffree < 0)
1396 			sf->f_ffree = 0;
1397 	}
1398 	NFSEXITCODE(error);
1399 	return (error);
1400 }
1401 
1402 /*
1403  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1404  * must handle nfsrv_opencheck() calls after any other access checks.
1405  */
1406 void
1407 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1408     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1409     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1410     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1411     struct nfsexstuff *exp, struct vnode **vpp)
1412 {
1413 	struct vnode *vp = NULL;
1414 	u_quad_t tempsize;
1415 	struct nfsexstuff nes;
1416 
1417 	if (ndp->ni_vp == NULL)
1418 		nd->nd_repstat = nfsrv_opencheck(clientid,
1419 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1420 	if (!nd->nd_repstat) {
1421 		if (ndp->ni_vp == NULL) {
1422 			vrele(ndp->ni_startdir);
1423 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1424 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1425 			vput(ndp->ni_dvp);
1426 			nfsvno_relpathbuf(ndp);
1427 			if (!nd->nd_repstat) {
1428 				if (*exclusive_flagp) {
1429 					*exclusive_flagp = 0;
1430 					NFSVNO_ATTRINIT(nvap);
1431 					nvap->na_atime.tv_sec = cverf[0];
1432 					nvap->na_atime.tv_nsec = cverf[1];
1433 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1434 					    &nvap->na_vattr, cred);
1435 					if (nd->nd_repstat != 0) {
1436 						vput(ndp->ni_vp);
1437 						ndp->ni_vp = NULL;
1438 						nd->nd_repstat = NFSERR_NOTSUPP;
1439 					} else
1440 						NFSSETBIT_ATTRBIT(attrbitp,
1441 						    NFSATTRBIT_TIMEACCESS);
1442 				} else {
1443 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1444 					    aclp, p, attrbitp, exp);
1445 				}
1446 			}
1447 			vp = ndp->ni_vp;
1448 		} else {
1449 			if (ndp->ni_startdir)
1450 				vrele(ndp->ni_startdir);
1451 			nfsvno_relpathbuf(ndp);
1452 			vp = ndp->ni_vp;
1453 			if (create == NFSV4OPEN_CREATE) {
1454 				if (ndp->ni_dvp == vp)
1455 					vrele(ndp->ni_dvp);
1456 				else
1457 					vput(ndp->ni_dvp);
1458 			}
1459 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1460 				if (ndp->ni_cnd.cn_flags & RDONLY)
1461 					NFSVNO_SETEXRDONLY(&nes);
1462 				else
1463 					NFSVNO_EXINIT(&nes);
1464 				nd->nd_repstat = nfsvno_accchk(vp,
1465 				    VWRITE, cred, &nes, p,
1466 				    NFSACCCHK_NOOVERRIDE,
1467 				    NFSACCCHK_VPISLOCKED, NULL);
1468 				nd->nd_repstat = nfsrv_opencheck(clientid,
1469 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1470 				if (!nd->nd_repstat) {
1471 					tempsize = nvap->na_size;
1472 					NFSVNO_ATTRINIT(nvap);
1473 					nvap->na_size = tempsize;
1474 					nd->nd_repstat = VOP_SETATTR(vp,
1475 					    &nvap->na_vattr, cred);
1476 				}
1477 			} else if (vp->v_type == VREG) {
1478 				nd->nd_repstat = nfsrv_opencheck(clientid,
1479 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1480 			}
1481 		}
1482 	} else {
1483 		if (ndp->ni_cnd.cn_flags & HASBUF)
1484 			nfsvno_relpathbuf(ndp);
1485 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1486 			vrele(ndp->ni_startdir);
1487 			if (ndp->ni_dvp == ndp->ni_vp)
1488 				vrele(ndp->ni_dvp);
1489 			else
1490 				vput(ndp->ni_dvp);
1491 			if (ndp->ni_vp)
1492 				vput(ndp->ni_vp);
1493 		}
1494 	}
1495 	*vpp = vp;
1496 
1497 	NFSEXITCODE2(0, nd);
1498 }
1499 
1500 /*
1501  * Updates the file rev and sets the mtime and ctime
1502  * to the current clock time, returning the va_filerev and va_Xtime
1503  * values.
1504  * Return ESTALE to indicate the vnode is VI_DOOMED.
1505  */
1506 int
1507 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1508     struct ucred *cred, struct thread *p)
1509 {
1510 	struct vattr va;
1511 
1512 	VATTR_NULL(&va);
1513 	vfs_timestamp(&va.va_mtime);
1514 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1515 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1516 		if ((vp->v_iflag & VI_DOOMED) != 0)
1517 			return (ESTALE);
1518 	}
1519 	(void) VOP_SETATTR(vp, &va, cred);
1520 	(void) nfsvno_getattr(vp, nvap, cred, p, 1);
1521 	return (0);
1522 }
1523 
1524 /*
1525  * Glue routine to nfsv4_fillattr().
1526  */
1527 int
1528 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1529     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1530     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1531     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1532 {
1533 	int error;
1534 
1535 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1536 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1537 	    mounted_on_fileno);
1538 	NFSEXITCODE2(0, nd);
1539 	return (error);
1540 }
1541 
1542 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1543 /*
1544  * nfs readdir service
1545  * - mallocs what it thinks is enough to read
1546  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1547  * - calls VOP_READDIR()
1548  * - loops around building the reply
1549  *	if the output generated exceeds count break out of loop
1550  *	The NFSM_CLGET macro is used here so that the reply will be packed
1551  *	tightly in mbuf clusters.
1552  * - it trims out records with d_fileno == 0
1553  *	this doesn't matter for Unix clients, but they might confuse clients
1554  *	for other os'.
1555  * - it trims out records with d_type == DT_WHT
1556  *	these cannot be seen through NFS (unless we extend the protocol)
1557  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1558  * PS: The NFS protocol spec. does not clarify what the "count" byte
1559  *	argument is a count of.. just name strings and file id's or the
1560  *	entire reply rpc or ...
1561  *	I tried just file name and id sizes and it confused the Sun client,
1562  *	so I am using the full rpc size now. The "paranoia.." comment refers
1563  *	to including the status longwords that are not a part of the dir.
1564  *	"entry" structures, but are in the rpc.
1565  */
1566 int
1567 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1568     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1569 {
1570 	struct dirent *dp;
1571 	u_int32_t *tl;
1572 	int dirlen;
1573 	char *cpos, *cend, *rbuf;
1574 	struct nfsvattr at;
1575 	int nlen, error = 0, getret = 1;
1576 	int siz, cnt, fullsiz, eofflag, ncookies;
1577 	u_int64_t off, toff;
1578 #if 0
1579 	u_int64_t verf;
1580 #endif
1581 	u_long *cookies = NULL, *cookiep;
1582 	struct uio io;
1583 	struct iovec iv;
1584 	int is_ufs;
1585 
1586 	if (nd->nd_repstat) {
1587 		nfsrv_postopattr(nd, getret, &at);
1588 		goto out;
1589 	}
1590 	if (nd->nd_flag & ND_NFSV2) {
1591 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1592 		off = fxdr_unsigned(u_quad_t, *tl++);
1593 	} else {
1594 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1595 		off = fxdr_hyper(tl);
1596 		tl += 2;
1597 #if 0
1598 		verf = fxdr_hyper(tl);
1599 #endif
1600 		tl += 2;
1601 	}
1602 	toff = off;
1603 	cnt = fxdr_unsigned(int, *tl);
1604 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1605 		cnt = NFS_SRVMAXDATA(nd);
1606 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1607 	fullsiz = siz;
1608 	if (nd->nd_flag & ND_NFSV3) {
1609 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1610 		    p, 1);
1611 #if 0
1612 		/*
1613 		 * va_filerev is not sufficient as a cookie verifier,
1614 		 * since it is not supposed to change when entries are
1615 		 * removed/added unless that offset cookies returned to
1616 		 * the client are no longer valid.
1617 		 */
1618 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
1619 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1620 #endif
1621 	}
1622 	if (!nd->nd_repstat && vp->v_type != VDIR)
1623 		nd->nd_repstat = NFSERR_NOTDIR;
1624 	if (nd->nd_repstat == 0 && cnt == 0) {
1625 		if (nd->nd_flag & ND_NFSV2)
1626 			/* NFSv2 does not have NFSERR_TOOSMALL */
1627 			nd->nd_repstat = EPERM;
1628 		else
1629 			nd->nd_repstat = NFSERR_TOOSMALL;
1630 	}
1631 	if (!nd->nd_repstat)
1632 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1633 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1634 		    NFSACCCHK_VPISLOCKED, NULL);
1635 	if (nd->nd_repstat) {
1636 		vput(vp);
1637 		if (nd->nd_flag & ND_NFSV3)
1638 			nfsrv_postopattr(nd, getret, &at);
1639 		goto out;
1640 	}
1641 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1642 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
1643 again:
1644 	eofflag = 0;
1645 	if (cookies) {
1646 		free(cookies, M_TEMP);
1647 		cookies = NULL;
1648 	}
1649 
1650 	iv.iov_base = rbuf;
1651 	iv.iov_len = siz;
1652 	io.uio_iov = &iv;
1653 	io.uio_iovcnt = 1;
1654 	io.uio_offset = (off_t)off;
1655 	io.uio_resid = siz;
1656 	io.uio_segflg = UIO_SYSSPACE;
1657 	io.uio_rw = UIO_READ;
1658 	io.uio_td = NULL;
1659 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1660 	    &cookies);
1661 	off = (u_int64_t)io.uio_offset;
1662 	if (io.uio_resid)
1663 		siz -= io.uio_resid;
1664 
1665 	if (!cookies && !nd->nd_repstat)
1666 		nd->nd_repstat = NFSERR_PERM;
1667 	if (nd->nd_flag & ND_NFSV3) {
1668 		getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1669 		if (!nd->nd_repstat)
1670 			nd->nd_repstat = getret;
1671 	}
1672 
1673 	/*
1674 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
1675 	 */
1676 	if (nd->nd_repstat) {
1677 		vput(vp);
1678 		free(rbuf, M_TEMP);
1679 		if (cookies)
1680 			free(cookies, M_TEMP);
1681 		if (nd->nd_flag & ND_NFSV3)
1682 			nfsrv_postopattr(nd, getret, &at);
1683 		goto out;
1684 	}
1685 	/*
1686 	 * If nothing read, return eof
1687 	 * rpc reply
1688 	 */
1689 	if (siz == 0) {
1690 		vput(vp);
1691 		if (nd->nd_flag & ND_NFSV2) {
1692 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1693 		} else {
1694 			nfsrv_postopattr(nd, getret, &at);
1695 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1696 			txdr_hyper(at.na_filerev, tl);
1697 			tl += 2;
1698 		}
1699 		*tl++ = newnfs_false;
1700 		*tl = newnfs_true;
1701 		free(rbuf, M_TEMP);
1702 		free(cookies, M_TEMP);
1703 		goto out;
1704 	}
1705 
1706 	/*
1707 	 * Check for degenerate cases of nothing useful read.
1708 	 * If so go try again
1709 	 */
1710 	cpos = rbuf;
1711 	cend = rbuf + siz;
1712 	dp = (struct dirent *)cpos;
1713 	cookiep = cookies;
1714 
1715 	/*
1716 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1717 	 * directory offset up to a block boundary, so it is necessary to
1718 	 * skip over the records that precede the requested offset. This
1719 	 * requires the assumption that file offset cookies monotonically
1720 	 * increase.
1721 	 */
1722 	while (cpos < cend && ncookies > 0 &&
1723 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1724 	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
1725 		cpos += dp->d_reclen;
1726 		dp = (struct dirent *)cpos;
1727 		cookiep++;
1728 		ncookies--;
1729 	}
1730 	if (cpos >= cend || ncookies == 0) {
1731 		siz = fullsiz;
1732 		toff = off;
1733 		goto again;
1734 	}
1735 	vput(vp);
1736 
1737 	/*
1738 	 * dirlen is the size of the reply, including all XDR and must
1739 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1740 	 * if the XDR should be included in "count", but to be safe, we do.
1741 	 * (Include the two booleans at the end of the reply in dirlen now.)
1742 	 */
1743 	if (nd->nd_flag & ND_NFSV3) {
1744 		nfsrv_postopattr(nd, getret, &at);
1745 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1746 		txdr_hyper(at.na_filerev, tl);
1747 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1748 	} else {
1749 		dirlen = 2 * NFSX_UNSIGNED;
1750 	}
1751 
1752 	/* Loop through the records and build reply */
1753 	while (cpos < cend && ncookies > 0) {
1754 		nlen = dp->d_namlen;
1755 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1756 			nlen <= NFS_MAXNAMLEN) {
1757 			if (nd->nd_flag & ND_NFSV3)
1758 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1759 			else
1760 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1761 			if (dirlen > cnt) {
1762 				eofflag = 0;
1763 				break;
1764 			}
1765 
1766 			/*
1767 			 * Build the directory record xdr from
1768 			 * the dirent entry.
1769 			 */
1770 			if (nd->nd_flag & ND_NFSV3) {
1771 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1772 				*tl++ = newnfs_true;
1773 				*tl++ = 0;
1774 			} else {
1775 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1776 				*tl++ = newnfs_true;
1777 			}
1778 			*tl = txdr_unsigned(dp->d_fileno);
1779 			(void) nfsm_strtom(nd, dp->d_name, nlen);
1780 			if (nd->nd_flag & ND_NFSV3) {
1781 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1782 				*tl++ = 0;
1783 			} else
1784 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1785 			*tl = txdr_unsigned(*cookiep);
1786 		}
1787 		cpos += dp->d_reclen;
1788 		dp = (struct dirent *)cpos;
1789 		cookiep++;
1790 		ncookies--;
1791 	}
1792 	if (cpos < cend)
1793 		eofflag = 0;
1794 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1795 	*tl++ = newnfs_false;
1796 	if (eofflag)
1797 		*tl = newnfs_true;
1798 	else
1799 		*tl = newnfs_false;
1800 	free(rbuf, M_TEMP);
1801 	free(cookies, M_TEMP);
1802 
1803 out:
1804 	NFSEXITCODE2(0, nd);
1805 	return (0);
1806 nfsmout:
1807 	vput(vp);
1808 	NFSEXITCODE2(error, nd);
1809 	return (error);
1810 }
1811 
1812 /*
1813  * Readdirplus for V3 and Readdir for V4.
1814  */
1815 int
1816 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1817     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1818 {
1819 	struct dirent *dp;
1820 	u_int32_t *tl;
1821 	int dirlen;
1822 	char *cpos, *cend, *rbuf;
1823 	struct vnode *nvp;
1824 	fhandle_t nfh;
1825 	struct nfsvattr nva, at, *nvap = &nva;
1826 	struct mbuf *mb0, *mb1;
1827 	struct nfsreferral *refp;
1828 	int nlen, r, error = 0, getret = 1, usevget = 1;
1829 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1830 	caddr_t bpos0, bpos1;
1831 	u_int64_t off, toff, verf;
1832 	u_long *cookies = NULL, *cookiep;
1833 	nfsattrbit_t attrbits, rderrbits, savbits;
1834 	struct uio io;
1835 	struct iovec iv;
1836 	struct componentname cn;
1837 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
1838 	struct mount *mp, *new_mp;
1839 	uint64_t mounted_on_fileno;
1840 
1841 	if (nd->nd_repstat) {
1842 		nfsrv_postopattr(nd, getret, &at);
1843 		goto out;
1844 	}
1845 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1846 	off = fxdr_hyper(tl);
1847 	toff = off;
1848 	tl += 2;
1849 	verf = fxdr_hyper(tl);
1850 	tl += 2;
1851 	siz = fxdr_unsigned(int, *tl++);
1852 	cnt = fxdr_unsigned(int, *tl);
1853 
1854 	/*
1855 	 * Use the server's maximum data transfer size as the upper bound
1856 	 * on reply datalen.
1857 	 */
1858 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1859 		cnt = NFS_SRVMAXDATA(nd);
1860 
1861 	/*
1862 	 * siz is a "hint" of how much directory information (name, fileid,
1863 	 * cookie) should be in the reply. At least one client "hints" 0,
1864 	 * so I set it to cnt for that case. I also round it up to the
1865 	 * next multiple of DIRBLKSIZ.
1866 	 */
1867 	if (siz <= 0)
1868 		siz = cnt;
1869 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1870 
1871 	if (nd->nd_flag & ND_NFSV4) {
1872 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1873 		if (error)
1874 			goto nfsmout;
1875 		NFSSET_ATTRBIT(&savbits, &attrbits);
1876 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1877 		NFSZERO_ATTRBIT(&rderrbits);
1878 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1879 	} else {
1880 		NFSZERO_ATTRBIT(&attrbits);
1881 	}
1882 	fullsiz = siz;
1883 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1884 	if (!nd->nd_repstat) {
1885 	    if (off && verf != at.na_filerev) {
1886 		/*
1887 		 * va_filerev is not sufficient as a cookie verifier,
1888 		 * since it is not supposed to change when entries are
1889 		 * removed/added unless that offset cookies returned to
1890 		 * the client are no longer valid.
1891 		 */
1892 #if 0
1893 		if (nd->nd_flag & ND_NFSV4) {
1894 			nd->nd_repstat = NFSERR_NOTSAME;
1895 		} else {
1896 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1897 		}
1898 #endif
1899 	    } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1900 		nd->nd_repstat = NFSERR_BAD_COOKIE;
1901 	    }
1902 	}
1903 	if (!nd->nd_repstat && vp->v_type != VDIR)
1904 		nd->nd_repstat = NFSERR_NOTDIR;
1905 	if (!nd->nd_repstat && cnt == 0)
1906 		nd->nd_repstat = NFSERR_TOOSMALL;
1907 	if (!nd->nd_repstat)
1908 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1909 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1910 		    NFSACCCHK_VPISLOCKED, NULL);
1911 	if (nd->nd_repstat) {
1912 		vput(vp);
1913 		if (nd->nd_flag & ND_NFSV3)
1914 			nfsrv_postopattr(nd, getret, &at);
1915 		goto out;
1916 	}
1917 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1918 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
1919 
1920 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
1921 again:
1922 	eofflag = 0;
1923 	if (cookies) {
1924 		free(cookies, M_TEMP);
1925 		cookies = NULL;
1926 	}
1927 
1928 	iv.iov_base = rbuf;
1929 	iv.iov_len = siz;
1930 	io.uio_iov = &iv;
1931 	io.uio_iovcnt = 1;
1932 	io.uio_offset = (off_t)off;
1933 	io.uio_resid = siz;
1934 	io.uio_segflg = UIO_SYSSPACE;
1935 	io.uio_rw = UIO_READ;
1936 	io.uio_td = NULL;
1937 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1938 	    &cookies);
1939 	off = (u_int64_t)io.uio_offset;
1940 	if (io.uio_resid)
1941 		siz -= io.uio_resid;
1942 
1943 	getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1944 
1945 	if (!cookies && !nd->nd_repstat)
1946 		nd->nd_repstat = NFSERR_PERM;
1947 	if (!nd->nd_repstat)
1948 		nd->nd_repstat = getret;
1949 	if (nd->nd_repstat) {
1950 		vput(vp);
1951 		if (cookies)
1952 			free(cookies, M_TEMP);
1953 		free(rbuf, M_TEMP);
1954 		if (nd->nd_flag & ND_NFSV3)
1955 			nfsrv_postopattr(nd, getret, &at);
1956 		goto out;
1957 	}
1958 	/*
1959 	 * If nothing read, return eof
1960 	 * rpc reply
1961 	 */
1962 	if (siz == 0) {
1963 		vput(vp);
1964 		if (nd->nd_flag & ND_NFSV3)
1965 			nfsrv_postopattr(nd, getret, &at);
1966 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1967 		txdr_hyper(at.na_filerev, tl);
1968 		tl += 2;
1969 		*tl++ = newnfs_false;
1970 		*tl = newnfs_true;
1971 		free(cookies, M_TEMP);
1972 		free(rbuf, M_TEMP);
1973 		goto out;
1974 	}
1975 
1976 	/*
1977 	 * Check for degenerate cases of nothing useful read.
1978 	 * If so go try again
1979 	 */
1980 	cpos = rbuf;
1981 	cend = rbuf + siz;
1982 	dp = (struct dirent *)cpos;
1983 	cookiep = cookies;
1984 
1985 	/*
1986 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1987 	 * directory offset up to a block boundary, so it is necessary to
1988 	 * skip over the records that precede the requested offset. This
1989 	 * requires the assumption that file offset cookies monotonically
1990 	 * increase.
1991 	 */
1992 	while (cpos < cend && ncookies > 0 &&
1993 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1994 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
1995 	   ((nd->nd_flag & ND_NFSV4) &&
1996 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1997 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1998 		cpos += dp->d_reclen;
1999 		dp = (struct dirent *)cpos;
2000 		cookiep++;
2001 		ncookies--;
2002 	}
2003 	if (cpos >= cend || ncookies == 0) {
2004 		siz = fullsiz;
2005 		toff = off;
2006 		goto again;
2007 	}
2008 
2009 	/*
2010 	 * Busy the file system so that the mount point won't go away
2011 	 * and, as such, VFS_VGET() can be used safely.
2012 	 */
2013 	mp = vp->v_mount;
2014 	vfs_ref(mp);
2015 	NFSVOPUNLOCK(vp, 0);
2016 	nd->nd_repstat = vfs_busy(mp, 0);
2017 	vfs_rel(mp);
2018 	if (nd->nd_repstat != 0) {
2019 		vrele(vp);
2020 		free(cookies, M_TEMP);
2021 		free(rbuf, M_TEMP);
2022 		if (nd->nd_flag & ND_NFSV3)
2023 			nfsrv_postopattr(nd, getret, &at);
2024 		goto out;
2025 	}
2026 
2027 	/*
2028 	 * Check to see if entries in this directory can be safely acquired
2029 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2030 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2031 	 * automount of the snapshot directory that is required will
2032 	 * be done.
2033 	 * This needs to be done here for NFSv4, since NFSv4 never does
2034 	 * a VFS_VGET() for "." or "..".
2035 	 */
2036 	if (is_zfs == 1) {
2037 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2038 		if (r == EOPNOTSUPP) {
2039 			usevget = 0;
2040 			cn.cn_nameiop = LOOKUP;
2041 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2042 			cn.cn_cred = nd->nd_cred;
2043 			cn.cn_thread = p;
2044 		} else if (r == 0)
2045 			vput(nvp);
2046 	}
2047 
2048 	/*
2049 	 * Save this position, in case there is an error before one entry
2050 	 * is created.
2051 	 */
2052 	mb0 = nd->nd_mb;
2053 	bpos0 = nd->nd_bpos;
2054 
2055 	/*
2056 	 * Fill in the first part of the reply.
2057 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2058 	 * (Include the two booleans at the end of the reply in dirlen now,
2059 	 *  so we recognize when we have exceeded cnt.)
2060 	 */
2061 	if (nd->nd_flag & ND_NFSV3) {
2062 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2063 		nfsrv_postopattr(nd, getret, &at);
2064 	} else {
2065 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2066 	}
2067 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2068 	txdr_hyper(at.na_filerev, tl);
2069 
2070 	/*
2071 	 * Save this position, in case there is an empty reply needed.
2072 	 */
2073 	mb1 = nd->nd_mb;
2074 	bpos1 = nd->nd_bpos;
2075 
2076 	/* Loop through the records and build reply */
2077 	entrycnt = 0;
2078 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2079 		nlen = dp->d_namlen;
2080 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2081 		    nlen <= NFS_MAXNAMLEN &&
2082 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2083 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2084 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2085 			/*
2086 			 * Save the current position in the reply, in case
2087 			 * this entry exceeds cnt.
2088 			 */
2089 			mb1 = nd->nd_mb;
2090 			bpos1 = nd->nd_bpos;
2091 
2092 			/*
2093 			 * For readdir_and_lookup get the vnode using
2094 			 * the file number.
2095 			 */
2096 			nvp = NULL;
2097 			refp = NULL;
2098 			r = 0;
2099 			at_root = 0;
2100 			needs_unbusy = 0;
2101 			new_mp = mp;
2102 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2103 			if ((nd->nd_flag & ND_NFSV3) ||
2104 			    NFSNONZERO_ATTRBIT(&savbits)) {
2105 				if (nd->nd_flag & ND_NFSV4)
2106 					refp = nfsv4root_getreferral(NULL,
2107 					    vp, dp->d_fileno);
2108 				if (refp == NULL) {
2109 					if (usevget)
2110 						r = VFS_VGET(mp, dp->d_fileno,
2111 						    LK_SHARED, &nvp);
2112 					else
2113 						r = EOPNOTSUPP;
2114 					if (r == EOPNOTSUPP) {
2115 						if (usevget) {
2116 							usevget = 0;
2117 							cn.cn_nameiop = LOOKUP;
2118 							cn.cn_lkflags =
2119 							    LK_SHARED |
2120 							    LK_RETRY;
2121 							cn.cn_cred =
2122 							    nd->nd_cred;
2123 							cn.cn_thread = p;
2124 						}
2125 						cn.cn_nameptr = dp->d_name;
2126 						cn.cn_namelen = nlen;
2127 						cn.cn_flags = ISLASTCN |
2128 						    NOFOLLOW | LOCKLEAF;
2129 						if (nlen == 2 &&
2130 						    dp->d_name[0] == '.' &&
2131 						    dp->d_name[1] == '.')
2132 							cn.cn_flags |=
2133 							    ISDOTDOT;
2134 						if (NFSVOPLOCK(vp, LK_SHARED)
2135 						    != 0) {
2136 							nd->nd_repstat = EPERM;
2137 							break;
2138 						}
2139 						if ((vp->v_vflag & VV_ROOT) != 0
2140 						    && (cn.cn_flags & ISDOTDOT)
2141 						    != 0) {
2142 							vref(vp);
2143 							nvp = vp;
2144 							r = 0;
2145 						} else {
2146 							r = VOP_LOOKUP(vp, &nvp,
2147 							    &cn);
2148 							if (vp != nvp)
2149 								NFSVOPUNLOCK(vp,
2150 								    0);
2151 						}
2152 					}
2153 
2154 					/*
2155 					 * For NFSv4, check to see if nvp is
2156 					 * a mount point and get the mount
2157 					 * point vnode, as required.
2158 					 */
2159 					if (r == 0 &&
2160 					    nfsrv_enable_crossmntpt != 0 &&
2161 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2162 					    nvp->v_type == VDIR &&
2163 					    nvp->v_mountedhere != NULL) {
2164 						new_mp = nvp->v_mountedhere;
2165 						r = vfs_busy(new_mp, 0);
2166 						vput(nvp);
2167 						nvp = NULL;
2168 						if (r == 0) {
2169 							r = VFS_ROOT(new_mp,
2170 							    LK_SHARED, &nvp);
2171 							needs_unbusy = 1;
2172 							if (r == 0)
2173 								at_root = 1;
2174 						}
2175 					}
2176 				}
2177 				if (!r) {
2178 				    if (refp == NULL &&
2179 					((nd->nd_flag & ND_NFSV3) ||
2180 					 NFSNONZERO_ATTRBIT(&attrbits))) {
2181 					r = nfsvno_getfh(nvp, &nfh, p);
2182 					if (!r)
2183 					    r = nfsvno_getattr(nvp, nvap,
2184 						nd->nd_cred, p, 1);
2185 					if (r == 0 && is_zfs == 1 &&
2186 					    nfsrv_enable_crossmntpt != 0 &&
2187 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2188 					    nvp->v_type == VDIR &&
2189 					    vp->v_mount != nvp->v_mount) {
2190 					    /*
2191 					     * For a ZFS snapshot, there is a
2192 					     * pseudo mount that does not set
2193 					     * v_mountedhere, so it needs to
2194 					     * be detected via a different
2195 					     * mount structure.
2196 					     */
2197 					    at_root = 1;
2198 					    if (new_mp == mp)
2199 						new_mp = nvp->v_mount;
2200 					}
2201 				    }
2202 				} else {
2203 				    nvp = NULL;
2204 				}
2205 				if (r) {
2206 					if (!NFSISSET_ATTRBIT(&attrbits,
2207 					    NFSATTRBIT_RDATTRERROR)) {
2208 						if (nvp != NULL)
2209 							vput(nvp);
2210 						if (needs_unbusy != 0)
2211 							vfs_unbusy(new_mp);
2212 						nd->nd_repstat = r;
2213 						break;
2214 					}
2215 				}
2216 			}
2217 
2218 			/*
2219 			 * Build the directory record xdr
2220 			 */
2221 			if (nd->nd_flag & ND_NFSV3) {
2222 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2223 				*tl++ = newnfs_true;
2224 				*tl++ = 0;
2225 				*tl = txdr_unsigned(dp->d_fileno);
2226 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2227 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2228 				*tl++ = 0;
2229 				*tl = txdr_unsigned(*cookiep);
2230 				nfsrv_postopattr(nd, 0, nvap);
2231 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2232 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2233 				if (nvp != NULL)
2234 					vput(nvp);
2235 			} else {
2236 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2237 				*tl++ = newnfs_true;
2238 				*tl++ = 0;
2239 				*tl = txdr_unsigned(*cookiep);
2240 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2241 				if (nvp != NULL) {
2242 					supports_nfsv4acls =
2243 					    nfs_supportsnfsv4acls(nvp);
2244 					NFSVOPUNLOCK(nvp, 0);
2245 				} else
2246 					supports_nfsv4acls = 0;
2247 				if (refp != NULL) {
2248 					dirlen += nfsrv_putreferralattr(nd,
2249 					    &savbits, refp, 0,
2250 					    &nd->nd_repstat);
2251 					if (nd->nd_repstat) {
2252 						if (nvp != NULL)
2253 							vrele(nvp);
2254 						if (needs_unbusy != 0)
2255 							vfs_unbusy(new_mp);
2256 						break;
2257 					}
2258 				} else if (r) {
2259 					dirlen += nfsvno_fillattr(nd, new_mp,
2260 					    nvp, nvap, &nfh, r, &rderrbits,
2261 					    nd->nd_cred, p, isdgram, 0,
2262 					    supports_nfsv4acls, at_root,
2263 					    mounted_on_fileno);
2264 				} else {
2265 					dirlen += nfsvno_fillattr(nd, new_mp,
2266 					    nvp, nvap, &nfh, r, &attrbits,
2267 					    nd->nd_cred, p, isdgram, 0,
2268 					    supports_nfsv4acls, at_root,
2269 					    mounted_on_fileno);
2270 				}
2271 				if (nvp != NULL)
2272 					vrele(nvp);
2273 				dirlen += (3 * NFSX_UNSIGNED);
2274 			}
2275 			if (needs_unbusy != 0)
2276 				vfs_unbusy(new_mp);
2277 			if (dirlen <= cnt)
2278 				entrycnt++;
2279 		}
2280 		cpos += dp->d_reclen;
2281 		dp = (struct dirent *)cpos;
2282 		cookiep++;
2283 		ncookies--;
2284 	}
2285 	vrele(vp);
2286 	vfs_unbusy(mp);
2287 
2288 	/*
2289 	 * If dirlen > cnt, we must strip off the last entry. If that
2290 	 * results in an empty reply, report NFSERR_TOOSMALL.
2291 	 */
2292 	if (dirlen > cnt || nd->nd_repstat) {
2293 		if (!nd->nd_repstat && entrycnt == 0)
2294 			nd->nd_repstat = NFSERR_TOOSMALL;
2295 		if (nd->nd_repstat) {
2296 			newnfs_trimtrailing(nd, mb0, bpos0);
2297 			if (nd->nd_flag & ND_NFSV3)
2298 				nfsrv_postopattr(nd, getret, &at);
2299 		} else
2300 			newnfs_trimtrailing(nd, mb1, bpos1);
2301 		eofflag = 0;
2302 	} else if (cpos < cend)
2303 		eofflag = 0;
2304 	if (!nd->nd_repstat) {
2305 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2306 		*tl++ = newnfs_false;
2307 		if (eofflag)
2308 			*tl = newnfs_true;
2309 		else
2310 			*tl = newnfs_false;
2311 	}
2312 	free(cookies, M_TEMP);
2313 	free(rbuf, M_TEMP);
2314 
2315 out:
2316 	NFSEXITCODE2(0, nd);
2317 	return (0);
2318 nfsmout:
2319 	vput(vp);
2320 	NFSEXITCODE2(error, nd);
2321 	return (error);
2322 }
2323 
2324 /*
2325  * Get the settable attributes out of the mbuf list.
2326  * (Return 0 or EBADRPC)
2327  */
2328 int
2329 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2330     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2331 {
2332 	u_int32_t *tl;
2333 	struct nfsv2_sattr *sp;
2334 	int error = 0, toclient = 0;
2335 
2336 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2337 	case ND_NFSV2:
2338 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2339 		/*
2340 		 * Some old clients didn't fill in the high order 16bits.
2341 		 * --> check the low order 2 bytes for 0xffff
2342 		 */
2343 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2344 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2345 		if (sp->sa_uid != newnfs_xdrneg1)
2346 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2347 		if (sp->sa_gid != newnfs_xdrneg1)
2348 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2349 		if (sp->sa_size != newnfs_xdrneg1)
2350 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2351 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2352 #ifdef notyet
2353 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2354 #else
2355 			nvap->na_atime.tv_sec =
2356 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2357 			nvap->na_atime.tv_nsec = 0;
2358 #endif
2359 		}
2360 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2361 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2362 		break;
2363 	case ND_NFSV3:
2364 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2365 		if (*tl == newnfs_true) {
2366 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2367 			nvap->na_mode = nfstov_mode(*tl);
2368 		}
2369 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2370 		if (*tl == newnfs_true) {
2371 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2372 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2373 		}
2374 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2375 		if (*tl == newnfs_true) {
2376 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2377 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2378 		}
2379 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2380 		if (*tl == newnfs_true) {
2381 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2382 			nvap->na_size = fxdr_hyper(tl);
2383 		}
2384 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2385 		switch (fxdr_unsigned(int, *tl)) {
2386 		case NFSV3SATTRTIME_TOCLIENT:
2387 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2388 			fxdr_nfsv3time(tl, &nvap->na_atime);
2389 			toclient = 1;
2390 			break;
2391 		case NFSV3SATTRTIME_TOSERVER:
2392 			vfs_timestamp(&nvap->na_atime);
2393 			nvap->na_vaflags |= VA_UTIMES_NULL;
2394 			break;
2395 		}
2396 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2397 		switch (fxdr_unsigned(int, *tl)) {
2398 		case NFSV3SATTRTIME_TOCLIENT:
2399 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2400 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2401 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2402 			break;
2403 		case NFSV3SATTRTIME_TOSERVER:
2404 			vfs_timestamp(&nvap->na_mtime);
2405 			if (!toclient)
2406 				nvap->na_vaflags |= VA_UTIMES_NULL;
2407 			break;
2408 		}
2409 		break;
2410 	case ND_NFSV4:
2411 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2412 	}
2413 nfsmout:
2414 	NFSEXITCODE2(error, nd);
2415 	return (error);
2416 }
2417 
2418 /*
2419  * Handle the setable attributes for V4.
2420  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2421  */
2422 int
2423 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2424     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2425 {
2426 	u_int32_t *tl;
2427 	int attrsum = 0;
2428 	int i, j;
2429 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2430 	int toclient = 0;
2431 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2432 	uid_t uid;
2433 	gid_t gid;
2434 
2435 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2436 	if (error)
2437 		goto nfsmout;
2438 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2439 	attrsize = fxdr_unsigned(int, *tl);
2440 
2441 	/*
2442 	 * Loop around getting the setable attributes. If an unsupported
2443 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2444 	 */
2445 	if (retnotsup) {
2446 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2447 		bitpos = NFSATTRBIT_MAX;
2448 	} else {
2449 		bitpos = 0;
2450 	}
2451 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2452 	    if (attrsum > attrsize) {
2453 		error = NFSERR_BADXDR;
2454 		goto nfsmout;
2455 	    }
2456 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2457 		switch (bitpos) {
2458 		case NFSATTRBIT_SIZE:
2459 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2460                      if (vp != NULL && vp->v_type != VREG) {
2461                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2462                                 NFSERR_INVAL;
2463                             goto nfsmout;
2464 			}
2465 			nvap->na_size = fxdr_hyper(tl);
2466 			attrsum += NFSX_HYPER;
2467 			break;
2468 		case NFSATTRBIT_ACL:
2469 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2470 			    p);
2471 			if (error)
2472 				goto nfsmout;
2473 			if (aceerr && !nd->nd_repstat)
2474 				nd->nd_repstat = aceerr;
2475 			attrsum += aclsize;
2476 			break;
2477 		case NFSATTRBIT_ARCHIVE:
2478 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2479 			if (!nd->nd_repstat)
2480 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2481 			attrsum += NFSX_UNSIGNED;
2482 			break;
2483 		case NFSATTRBIT_HIDDEN:
2484 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2485 			if (!nd->nd_repstat)
2486 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2487 			attrsum += NFSX_UNSIGNED;
2488 			break;
2489 		case NFSATTRBIT_MIMETYPE:
2490 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2491 			i = fxdr_unsigned(int, *tl);
2492 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2493 			if (error)
2494 				goto nfsmout;
2495 			if (!nd->nd_repstat)
2496 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2497 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2498 			break;
2499 		case NFSATTRBIT_MODE:
2500 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2501 			nvap->na_mode = nfstov_mode(*tl);
2502 			attrsum += NFSX_UNSIGNED;
2503 			break;
2504 		case NFSATTRBIT_OWNER:
2505 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2506 			j = fxdr_unsigned(int, *tl);
2507 			if (j < 0) {
2508 				error = NFSERR_BADXDR;
2509 				goto nfsmout;
2510 			}
2511 			if (j > NFSV4_SMALLSTR)
2512 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2513 			else
2514 				cp = namestr;
2515 			error = nfsrv_mtostr(nd, cp, j);
2516 			if (error) {
2517 				if (j > NFSV4_SMALLSTR)
2518 					free(cp, M_NFSSTRING);
2519 				goto nfsmout;
2520 			}
2521 			if (!nd->nd_repstat) {
2522 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2523 				    p);
2524 				if (!nd->nd_repstat)
2525 					nvap->na_uid = uid;
2526 			}
2527 			if (j > NFSV4_SMALLSTR)
2528 				free(cp, M_NFSSTRING);
2529 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2530 			break;
2531 		case NFSATTRBIT_OWNERGROUP:
2532 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2533 			j = fxdr_unsigned(int, *tl);
2534 			if (j < 0) {
2535 				error = NFSERR_BADXDR;
2536 				goto nfsmout;
2537 			}
2538 			if (j > NFSV4_SMALLSTR)
2539 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2540 			else
2541 				cp = namestr;
2542 			error = nfsrv_mtostr(nd, cp, j);
2543 			if (error) {
2544 				if (j > NFSV4_SMALLSTR)
2545 					free(cp, M_NFSSTRING);
2546 				goto nfsmout;
2547 			}
2548 			if (!nd->nd_repstat) {
2549 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2550 				    p);
2551 				if (!nd->nd_repstat)
2552 					nvap->na_gid = gid;
2553 			}
2554 			if (j > NFSV4_SMALLSTR)
2555 				free(cp, M_NFSSTRING);
2556 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2557 			break;
2558 		case NFSATTRBIT_SYSTEM:
2559 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2560 			if (!nd->nd_repstat)
2561 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2562 			attrsum += NFSX_UNSIGNED;
2563 			break;
2564 		case NFSATTRBIT_TIMEACCESSSET:
2565 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2566 			attrsum += NFSX_UNSIGNED;
2567 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2568 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2569 			    fxdr_nfsv4time(tl, &nvap->na_atime);
2570 			    toclient = 1;
2571 			    attrsum += NFSX_V4TIME;
2572 			} else {
2573 			    vfs_timestamp(&nvap->na_atime);
2574 			    nvap->na_vaflags |= VA_UTIMES_NULL;
2575 			}
2576 			break;
2577 		case NFSATTRBIT_TIMEBACKUP:
2578 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2579 			if (!nd->nd_repstat)
2580 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2581 			attrsum += NFSX_V4TIME;
2582 			break;
2583 		case NFSATTRBIT_TIMECREATE:
2584 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2585 			if (!nd->nd_repstat)
2586 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2587 			attrsum += NFSX_V4TIME;
2588 			break;
2589 		case NFSATTRBIT_TIMEMODIFYSET:
2590 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2591 			attrsum += NFSX_UNSIGNED;
2592 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2593 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2594 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
2595 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
2596 			    attrsum += NFSX_V4TIME;
2597 			} else {
2598 			    vfs_timestamp(&nvap->na_mtime);
2599 			    if (!toclient)
2600 				nvap->na_vaflags |= VA_UTIMES_NULL;
2601 			}
2602 			break;
2603 		default:
2604 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2605 			/*
2606 			 * set bitpos so we drop out of the loop.
2607 			 */
2608 			bitpos = NFSATTRBIT_MAX;
2609 			break;
2610 		}
2611 	}
2612 
2613 	/*
2614 	 * some clients pad the attrlist, so we need to skip over the
2615 	 * padding.
2616 	 */
2617 	if (attrsum > attrsize) {
2618 		error = NFSERR_BADXDR;
2619 	} else {
2620 		attrsize = NFSM_RNDUP(attrsize);
2621 		if (attrsum < attrsize)
2622 			error = nfsm_advance(nd, attrsize - attrsum, -1);
2623 	}
2624 nfsmout:
2625 	NFSEXITCODE2(error, nd);
2626 	return (error);
2627 }
2628 
2629 /*
2630  * Check/setup export credentials.
2631  */
2632 int
2633 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2634     struct ucred *credanon)
2635 {
2636 	int error = 0;
2637 
2638 	/*
2639 	 * Check/setup credentials.
2640 	 */
2641 	if (nd->nd_flag & ND_GSS)
2642 		exp->nes_exflag &= ~MNT_EXPORTANON;
2643 
2644 	/*
2645 	 * Check to see if the operation is allowed for this security flavor.
2646 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2647 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2648 	 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2649 	 */
2650 	if (nfsvno_testexp(nd, exp) &&
2651 	    nd->nd_procnum != NFSV4OP_SECINFO &&
2652 	    nd->nd_procnum != NFSPROC_FSINFO) {
2653 		if (nd->nd_flag & ND_NFSV4)
2654 			error = NFSERR_WRONGSEC;
2655 		else
2656 			error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2657 		goto out;
2658 	}
2659 
2660 	/*
2661 	 * Check to see if the file system is exported V4 only.
2662 	 */
2663 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2664 		error = NFSERR_PROGNOTV4;
2665 		goto out;
2666 	}
2667 
2668 	/*
2669 	 * Now, map the user credentials.
2670 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
2671 	 *  Fsinfo RPC. If set for anything else, this code might need
2672 	 *  to change.)
2673 	 */
2674 	if (NFSVNO_EXPORTED(exp)) {
2675 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
2676 		     NFSVNO_EXPORTANON(exp) ||
2677 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
2678 			nd->nd_cred->cr_uid = credanon->cr_uid;
2679 			nd->nd_cred->cr_gid = credanon->cr_gid;
2680 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2681 			    credanon->cr_groups);
2682 		} else if ((nd->nd_flag & ND_GSS) == 0) {
2683 			/*
2684 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
2685 			 * if there is a replacement credential with a group
2686 			 * list set up by "nfsuserd -manage-gids".
2687 			 * If there is no replacement, nfsrv_getgrpscred()
2688 			 * simply returns its argument.
2689 			 */
2690 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
2691 		}
2692 	}
2693 
2694 out:
2695 	NFSEXITCODE2(error, nd);
2696 	return (error);
2697 }
2698 
2699 /*
2700  * Check exports.
2701  */
2702 int
2703 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2704     struct ucred **credp)
2705 {
2706 	int i, error, *secflavors;
2707 
2708 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2709 	    &exp->nes_numsecflavor, &secflavors);
2710 	if (error) {
2711 		if (nfs_rootfhset) {
2712 			exp->nes_exflag = 0;
2713 			exp->nes_numsecflavor = 0;
2714 			error = 0;
2715 		}
2716 	} else {
2717 		/* Copy the security flavors. */
2718 		for (i = 0; i < exp->nes_numsecflavor; i++)
2719 			exp->nes_secflavors[i] = secflavors[i];
2720 	}
2721 	NFSEXITCODE(error);
2722 	return (error);
2723 }
2724 
2725 /*
2726  * Get a vnode for a file handle and export stuff.
2727  */
2728 int
2729 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2730     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2731     struct ucred **credp)
2732 {
2733 	int i, error, *secflavors;
2734 
2735 	*credp = NULL;
2736 	exp->nes_numsecflavor = 0;
2737 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2738 	if (error != 0)
2739 		/* Make sure the server replies ESTALE to the client. */
2740 		error = ESTALE;
2741 	if (nam && !error) {
2742 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2743 		    &exp->nes_numsecflavor, &secflavors);
2744 		if (error) {
2745 			if (nfs_rootfhset) {
2746 				exp->nes_exflag = 0;
2747 				exp->nes_numsecflavor = 0;
2748 				error = 0;
2749 			} else {
2750 				vput(*vpp);
2751 			}
2752 		} else {
2753 			/* Copy the security flavors. */
2754 			for (i = 0; i < exp->nes_numsecflavor; i++)
2755 				exp->nes_secflavors[i] = secflavors[i];
2756 		}
2757 	}
2758 	NFSEXITCODE(error);
2759 	return (error);
2760 }
2761 
2762 /*
2763  * nfsd_fhtovp() - convert a fh to a vnode ptr
2764  * 	- look up fsid in mount list (if not found ret error)
2765  *	- get vp and export rights by calling nfsvno_fhtovp()
2766  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2767  *	  for AUTH_SYS
2768  *	- if mpp != NULL, return the mount point so that it can
2769  *	  be used for vn_finished_write() by the caller
2770  */
2771 void
2772 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2773     struct vnode **vpp, struct nfsexstuff *exp,
2774     struct mount **mpp, int startwrite, struct thread *p)
2775 {
2776 	struct mount *mp;
2777 	struct ucred *credanon;
2778 	fhandle_t *fhp;
2779 
2780 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
2781 	/*
2782 	 * Check for the special case of the nfsv4root_fh.
2783 	 */
2784 	mp = vfs_busyfs(&fhp->fh_fsid);
2785 	if (mpp != NULL)
2786 		*mpp = mp;
2787 	if (mp == NULL) {
2788 		*vpp = NULL;
2789 		nd->nd_repstat = ESTALE;
2790 		goto out;
2791 	}
2792 
2793 	if (startwrite) {
2794 		vn_start_write(NULL, mpp, V_WAIT);
2795 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
2796 			lktype = LK_EXCLUSIVE;
2797 	}
2798 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2799 	    &credanon);
2800 	vfs_unbusy(mp);
2801 
2802 	/*
2803 	 * For NFSv4 without a pseudo root fs, unexported file handles
2804 	 * can be returned, so that Lookup works everywhere.
2805 	 */
2806 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2807 	    !(nd->nd_flag & ND_NFSV4)) {
2808 		vput(*vpp);
2809 		nd->nd_repstat = EACCES;
2810 	}
2811 
2812 	/*
2813 	 * Personally, I've never seen any point in requiring a
2814 	 * reserved port#, since only in the rare case where the
2815 	 * clients are all boxes with secure system privileges,
2816 	 * does it provide any enhanced security, but... some people
2817 	 * believe it to be useful and keep putting this code back in.
2818 	 * (There is also some "security checker" out there that
2819 	 *  complains if the nfs server doesn't enforce this.)
2820 	 * However, note the following:
2821 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
2822 	 *	required.
2823 	 * RFC2623 recommends that, if a reserved port# is checked for,
2824 	 *	that there be a way to turn that off--> ifdef'd.
2825 	 */
2826 #ifdef NFS_REQRSVPORT
2827 	if (!nd->nd_repstat) {
2828 		struct sockaddr_in *saddr;
2829 		struct sockaddr_in6 *saddr6;
2830 
2831 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2832 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2833 		if (!(nd->nd_flag & ND_NFSV4) &&
2834 		    ((saddr->sin_family == AF_INET &&
2835 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2836 		     (saddr6->sin6_family == AF_INET6 &&
2837 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2838 			vput(*vpp);
2839 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2840 		}
2841 	}
2842 #endif	/* NFS_REQRSVPORT */
2843 
2844 	/*
2845 	 * Check/setup credentials.
2846 	 */
2847 	if (!nd->nd_repstat) {
2848 		nd->nd_saveduid = nd->nd_cred->cr_uid;
2849 		nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2850 		if (nd->nd_repstat)
2851 			vput(*vpp);
2852 	}
2853 	if (credanon != NULL)
2854 		crfree(credanon);
2855 	if (nd->nd_repstat) {
2856 		if (startwrite)
2857 			vn_finished_write(mp);
2858 		*vpp = NULL;
2859 		if (mpp != NULL)
2860 			*mpp = NULL;
2861 	}
2862 
2863 out:
2864 	NFSEXITCODE2(0, nd);
2865 }
2866 
2867 /*
2868  * glue for fp.
2869  */
2870 static int
2871 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2872 {
2873 	struct filedesc *fdp;
2874 	struct file *fp;
2875 	int error = 0;
2876 
2877 	fdp = p->td_proc->p_fd;
2878 	if (fd < 0 || fd >= fdp->fd_nfiles ||
2879 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
2880 		error = EBADF;
2881 		goto out;
2882 	}
2883 	*fpp = fp;
2884 
2885 out:
2886 	NFSEXITCODE(error);
2887 	return (error);
2888 }
2889 
2890 /*
2891  * Called from nfssvc() to update the exports list. Just call
2892  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2893  * in the mount list.
2894  */
2895 int
2896 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2897 {
2898 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2899 	int error = 0;
2900 	struct nameidata nd;
2901 	fhandle_t fh;
2902 
2903 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2904 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2905 		nfs_rootfhset = 0;
2906 	else if (error == 0) {
2907 		if (nfsexargp->fspec == NULL) {
2908 			error = EPERM;
2909 			goto out;
2910 		}
2911 		/*
2912 		 * If fspec != NULL, this is the v4root path.
2913 		 */
2914 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
2915 		    nfsexargp->fspec, p);
2916 		if ((error = namei(&nd)) != 0)
2917 			goto out;
2918 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
2919 		vrele(nd.ni_vp);
2920 		if (!error) {
2921 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2922 			NFSBCOPY((caddr_t)&fh,
2923 			    nfs_rootfh.nfsrvfh_data,
2924 			    sizeof (fhandle_t));
2925 			nfs_rootfhset = 1;
2926 		}
2927 	}
2928 
2929 out:
2930 	NFSEXITCODE(error);
2931 	return (error);
2932 }
2933 
2934 /*
2935  * This function needs to test to see if the system is near its limit
2936  * for memory allocation via malloc() or mget() and return True iff
2937  * either of these resources are near their limit.
2938  * XXX (For now, this is just a stub.)
2939  */
2940 int nfsrv_testmalloclimit = 0;
2941 int
2942 nfsrv_mallocmget_limit(void)
2943 {
2944 	static int printmesg = 0;
2945 	static int testval = 1;
2946 
2947 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2948 		if ((printmesg++ % 100) == 0)
2949 			printf("nfsd: malloc/mget near limit\n");
2950 		return (1);
2951 	}
2952 	return (0);
2953 }
2954 
2955 /*
2956  * BSD specific initialization of a mount point.
2957  */
2958 void
2959 nfsd_mntinit(void)
2960 {
2961 	static int inited = 0;
2962 
2963 	if (inited)
2964 		return;
2965 	inited = 1;
2966 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2967 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2968 	TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2969 	nfsv4root_mnt.mnt_export = NULL;
2970 	TAILQ_INIT(&nfsv4root_opt);
2971 	TAILQ_INIT(&nfsv4root_newopt);
2972 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2973 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2974 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
2975 	nfsv4root_mnt.mnt_activevnodelistsize = 0;
2976 }
2977 
2978 /*
2979  * Get a vnode for a file handle, without checking exports, etc.
2980  */
2981 struct vnode *
2982 nfsvno_getvp(fhandle_t *fhp)
2983 {
2984 	struct mount *mp;
2985 	struct vnode *vp;
2986 	int error;
2987 
2988 	mp = vfs_busyfs(&fhp->fh_fsid);
2989 	if (mp == NULL)
2990 		return (NULL);
2991 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2992 	vfs_unbusy(mp);
2993 	if (error)
2994 		return (NULL);
2995 	return (vp);
2996 }
2997 
2998 /*
2999  * Do a local VOP_ADVLOCK().
3000  */
3001 int
3002 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3003     u_int64_t end, struct thread *td)
3004 {
3005 	int error = 0;
3006 	struct flock fl;
3007 	u_int64_t tlen;
3008 
3009 	if (nfsrv_dolocallocks == 0)
3010 		goto out;
3011 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3012 
3013 	fl.l_whence = SEEK_SET;
3014 	fl.l_type = ftype;
3015 	fl.l_start = (off_t)first;
3016 	if (end == NFS64BITSSET) {
3017 		fl.l_len = 0;
3018 	} else {
3019 		tlen = end - first;
3020 		fl.l_len = (off_t)tlen;
3021 	}
3022 	/*
3023 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3024 	 * values for all calls, so that all locks will be held by the
3025 	 * nfsd server. (The nfsd server handles conflicts between the
3026 	 * various clients.)
3027 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3028 	 * bytes, so it can't be put in l_sysid.
3029 	 */
3030 	if (nfsv4_sysid == 0)
3031 		nfsv4_sysid = nlm_acquire_next_sysid();
3032 	fl.l_pid = (pid_t)0;
3033 	fl.l_sysid = (int)nfsv4_sysid;
3034 
3035 	if (ftype == F_UNLCK)
3036 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3037 		    (F_POSIX | F_REMOTE));
3038 	else
3039 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3040 		    (F_POSIX | F_REMOTE));
3041 
3042 out:
3043 	NFSEXITCODE(error);
3044 	return (error);
3045 }
3046 
3047 /*
3048  * Check the nfsv4 root exports.
3049  */
3050 int
3051 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3052 {
3053 	struct ucred *credanon;
3054 	int exflags, error = 0, numsecflavor, *secflavors, i;
3055 
3056 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3057 	    &credanon, &numsecflavor, &secflavors);
3058 	if (error) {
3059 		error = NFSERR_PROGUNAVAIL;
3060 		goto out;
3061 	}
3062 	if (credanon != NULL)
3063 		crfree(credanon);
3064 	for (i = 0; i < numsecflavor; i++) {
3065 		if (secflavors[i] == AUTH_SYS)
3066 			nd->nd_flag |= ND_EXAUTHSYS;
3067 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3068 			nd->nd_flag |= ND_EXGSS;
3069 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3070 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3071 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3072 			nd->nd_flag |= ND_EXGSSPRIVACY;
3073 	}
3074 
3075 out:
3076 	NFSEXITCODE(error);
3077 	return (error);
3078 }
3079 
3080 /*
3081  * Nfs server pseudo system call for the nfsd's
3082  */
3083 /*
3084  * MPSAFE
3085  */
3086 static int
3087 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3088 {
3089 	struct file *fp;
3090 	struct nfsd_addsock_args sockarg;
3091 	struct nfsd_nfsd_args nfsdarg;
3092 	cap_rights_t rights;
3093 	int error;
3094 
3095 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3096 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3097 		if (error)
3098 			goto out;
3099 		/*
3100 		 * Since we don't know what rights might be required,
3101 		 * pretend that we need them all. It is better to be too
3102 		 * careful than too reckless.
3103 		 */
3104 		error = fget(td, sockarg.sock,
3105 		    cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3106 		if (error != 0)
3107 			goto out;
3108 		if (fp->f_type != DTYPE_SOCKET) {
3109 			fdrop(fp, td);
3110 			error = EPERM;
3111 			goto out;
3112 		}
3113 		error = nfsrvd_addsock(fp);
3114 		fdrop(fp, td);
3115 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3116 		if (uap->argp == NULL) {
3117 			error = EINVAL;
3118 			goto out;
3119 		}
3120 		error = copyin(uap->argp, (caddr_t)&nfsdarg,
3121 		    sizeof (nfsdarg));
3122 		if (error)
3123 			goto out;
3124 		error = nfsrvd_nfsd(td, &nfsdarg);
3125 	} else {
3126 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3127 	}
3128 
3129 out:
3130 	NFSEXITCODE(error);
3131 	return (error);
3132 }
3133 
3134 static int
3135 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3136 {
3137 	struct nfsex_args export;
3138 	struct file *fp = NULL;
3139 	int stablefd, len;
3140 	struct nfsd_clid adminrevoke;
3141 	struct nfsd_dumplist dumplist;
3142 	struct nfsd_dumpclients *dumpclients;
3143 	struct nfsd_dumplocklist dumplocklist;
3144 	struct nfsd_dumplocks *dumplocks;
3145 	struct nameidata nd;
3146 	vnode_t vp;
3147 	int error = EINVAL, igotlock;
3148 	struct proc *procp;
3149 	static int suspend_nfsd = 0;
3150 
3151 	if (uap->flag & NFSSVC_PUBLICFH) {
3152 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3153 		    sizeof (fhandle_t));
3154 		error = copyin(uap->argp,
3155 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3156 		if (!error)
3157 			nfs_pubfhset = 1;
3158 	} else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3159 		error = copyin(uap->argp,(caddr_t)&export,
3160 		    sizeof (struct nfsex_args));
3161 		if (!error)
3162 			error = nfsrv_v4rootexport(&export, cred, p);
3163 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3164 		nfs_pubfhset = 0;
3165 		error = 0;
3166 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3167 		error = copyin(uap->argp, (caddr_t)&stablefd,
3168 		    sizeof (int));
3169 		if (!error)
3170 			error = fp_getfvp(p, stablefd, &fp, &vp);
3171 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3172 			error = EBADF;
3173 		if (!error && newnfs_numnfsd != 0)
3174 			error = EPERM;
3175 		if (!error) {
3176 			nfsrv_stablefirst.nsf_fp = fp;
3177 			nfsrv_setupstable(p);
3178 		}
3179 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3180 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3181 		    sizeof (struct nfsd_clid));
3182 		if (!error)
3183 			error = nfsrv_adminrevoke(&adminrevoke, p);
3184 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3185 		error = copyin(uap->argp, (caddr_t)&dumplist,
3186 		    sizeof (struct nfsd_dumplist));
3187 		if (!error && (dumplist.ndl_size < 1 ||
3188 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3189 			error = EPERM;
3190 		if (!error) {
3191 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3192 		    dumpclients = (struct nfsd_dumpclients *)malloc(len,
3193 			M_TEMP, M_WAITOK);
3194 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3195 		    error = copyout(dumpclients,
3196 			CAST_USER_ADDR_T(dumplist.ndl_list), len);
3197 		    free(dumpclients, M_TEMP);
3198 		}
3199 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3200 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3201 		    sizeof (struct nfsd_dumplocklist));
3202 		if (!error && (dumplocklist.ndllck_size < 1 ||
3203 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3204 			error = EPERM;
3205 		if (!error)
3206 			error = nfsrv_lookupfilename(&nd,
3207 				dumplocklist.ndllck_fname, p);
3208 		if (!error) {
3209 			len = sizeof (struct nfsd_dumplocks) *
3210 				dumplocklist.ndllck_size;
3211 			dumplocks = (struct nfsd_dumplocks *)malloc(len,
3212 				M_TEMP, M_WAITOK);
3213 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3214 			    dumplocklist.ndllck_size, p);
3215 			vput(nd.ni_vp);
3216 			error = copyout(dumplocks,
3217 			    CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3218 			free(dumplocks, M_TEMP);
3219 		}
3220 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3221 		procp = p->td_proc;
3222 		PROC_LOCK(procp);
3223 		nfsd_master_pid = procp->p_pid;
3224 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3225 		nfsd_master_start = procp->p_stats->p_start;
3226 		nfsd_master_proc = procp;
3227 		PROC_UNLOCK(procp);
3228 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3229 		NFSLOCKV4ROOTMUTEX();
3230 		if (suspend_nfsd == 0) {
3231 			/* Lock out all nfsd threads */
3232 			do {
3233 				igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3234 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3235 			} while (igotlock == 0 && suspend_nfsd == 0);
3236 			suspend_nfsd = 1;
3237 		}
3238 		NFSUNLOCKV4ROOTMUTEX();
3239 		error = 0;
3240 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3241 		NFSLOCKV4ROOTMUTEX();
3242 		if (suspend_nfsd != 0) {
3243 			nfsv4_unlock(&nfsd_suspend_lock, 0);
3244 			suspend_nfsd = 0;
3245 		}
3246 		NFSUNLOCKV4ROOTMUTEX();
3247 		error = 0;
3248 	}
3249 
3250 	NFSEXITCODE(error);
3251 	return (error);
3252 }
3253 
3254 /*
3255  * Check exports.
3256  * Returns 0 if ok, 1 otherwise.
3257  */
3258 int
3259 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3260 {
3261 	int i;
3262 
3263 	/*
3264 	 * This seems odd, but allow the case where the security flavor
3265 	 * list is empty. This happens when NFSv4 is traversing non-exported
3266 	 * file systems. Exported file systems should always have a non-empty
3267 	 * security flavor list.
3268 	 */
3269 	if (exp->nes_numsecflavor == 0)
3270 		return (0);
3271 
3272 	for (i = 0; i < exp->nes_numsecflavor; i++) {
3273 		/*
3274 		 * The tests for privacy and integrity must be first,
3275 		 * since ND_GSS is set for everything but AUTH_SYS.
3276 		 */
3277 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3278 		    (nd->nd_flag & ND_GSSPRIVACY))
3279 			return (0);
3280 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3281 		    (nd->nd_flag & ND_GSSINTEGRITY))
3282 			return (0);
3283 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3284 		    (nd->nd_flag & ND_GSS))
3285 			return (0);
3286 		if (exp->nes_secflavors[i] == AUTH_SYS &&
3287 		    (nd->nd_flag & ND_GSS) == 0)
3288 			return (0);
3289 	}
3290 	return (1);
3291 }
3292 
3293 /*
3294  * Calculate a hash value for the fid in a file handle.
3295  */
3296 uint32_t
3297 nfsrv_hashfh(fhandle_t *fhp)
3298 {
3299 	uint32_t hashval;
3300 
3301 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3302 	return (hashval);
3303 }
3304 
3305 /*
3306  * Calculate a hash value for the sessionid.
3307  */
3308 uint32_t
3309 nfsrv_hashsessionid(uint8_t *sessionid)
3310 {
3311 	uint32_t hashval;
3312 
3313 	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
3314 	return (hashval);
3315 }
3316 
3317 /*
3318  * Signal the userland master nfsd to backup the stable restart file.
3319  */
3320 void
3321 nfsrv_backupstable(void)
3322 {
3323 	struct proc *procp;
3324 
3325 	if (nfsd_master_proc != NULL) {
3326 		procp = pfind(nfsd_master_pid);
3327 		/* Try to make sure it is the correct process. */
3328 		if (procp == nfsd_master_proc &&
3329 		    procp->p_stats->p_start.tv_sec ==
3330 		    nfsd_master_start.tv_sec &&
3331 		    procp->p_stats->p_start.tv_usec ==
3332 		    nfsd_master_start.tv_usec &&
3333 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
3334 			kern_psignal(procp, SIGUSR2);
3335 		else
3336 			nfsd_master_proc = NULL;
3337 
3338 		if (procp != NULL)
3339 			PROC_UNLOCK(procp);
3340 	}
3341 }
3342 
3343 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3344 
3345 /*
3346  * Called once to initialize data structures...
3347  */
3348 static int
3349 nfsd_modevent(module_t mod, int type, void *data)
3350 {
3351 	int error = 0, i;
3352 	static int loaded = 0;
3353 
3354 	switch (type) {
3355 	case MOD_LOAD:
3356 		if (loaded)
3357 			goto out;
3358 		newnfs_portinit();
3359 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3360 			mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
3361 			    MTX_DEF);
3362 			mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
3363 			    MTX_DEF);
3364 		}
3365 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
3366 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
3367 		mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
3368 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3369 		nfsrvd_initcache();
3370 		nfsd_init();
3371 		NFSD_LOCK();
3372 		nfsrvd_init(0);
3373 		NFSD_UNLOCK();
3374 		nfsd_mntinit();
3375 #ifdef VV_DISABLEDELEG
3376 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3377 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3378 #endif
3379 		nfsd_call_servertimer = nfsrv_servertimer;
3380 		nfsd_call_nfsd = nfssvc_nfsd;
3381 		loaded = 1;
3382 		break;
3383 
3384 	case MOD_UNLOAD:
3385 		if (newnfs_numnfsd != 0) {
3386 			error = EBUSY;
3387 			break;
3388 		}
3389 
3390 #ifdef VV_DISABLEDELEG
3391 		vn_deleg_ops.vndeleg_recall = NULL;
3392 		vn_deleg_ops.vndeleg_disable = NULL;
3393 #endif
3394 		nfsd_call_servertimer = NULL;
3395 		nfsd_call_nfsd = NULL;
3396 
3397 		/* Clean out all NFSv4 state. */
3398 		nfsrv_throwawayallstate(curthread);
3399 
3400 		/* Clean the NFS server reply cache */
3401 		nfsrvd_cleancache();
3402 
3403 		/* Free up the krpc server pool. */
3404 		if (nfsrvd_pool != NULL)
3405 			svcpool_destroy(nfsrvd_pool);
3406 
3407 		/* and get rid of the locks */
3408 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3409 			mtx_destroy(&nfsrchash_table[i].mtx);
3410 			mtx_destroy(&nfsrcahash_table[i].mtx);
3411 		}
3412 		mtx_destroy(&nfsrc_udpmtx);
3413 		mtx_destroy(&nfs_v4root_mutex);
3414 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3415 		for (i = 0; i < nfsrv_sessionhashsize; i++)
3416 			mtx_destroy(&nfssessionhash[i].mtx);
3417 		lockdestroy(&nfsv4root_mnt.mnt_explock);
3418 		free(nfsclienthash, M_NFSDCLIENT);
3419 		free(nfslockhash, M_NFSDLOCKFILE);
3420 		free(nfssessionhash, M_NFSDSESSION);
3421 		loaded = 0;
3422 		break;
3423 	default:
3424 		error = EOPNOTSUPP;
3425 		break;
3426 	}
3427 
3428 out:
3429 	NFSEXITCODE(error);
3430 	return (error);
3431 }
3432 static moduledata_t nfsd_mod = {
3433 	"nfsd",
3434 	nfsd_modevent,
3435 	NULL,
3436 };
3437 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3438 
3439 /* So that loader and kldload(2) can find us, wherever we are.. */
3440 MODULE_VERSION(nfsd, 1);
3441 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3442 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3443 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3444 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3445 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3446 
3447