xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision 3ef51c5fb9163f2aafb1c14729e06a8bf0c4d113)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include <sys/capability.h>
38 
39 /*
40  * Functions that perform the vfs operations required by the routines in
41  * nfsd_serv.c. It is hoped that this change will make the server more
42  * portable.
43  */
44 
45 #include <fs/nfs/nfsport.h>
46 #include <sys/hash.h>
47 #include <sys/sysctl.h>
48 #include <nlm/nlm_prot.h>
49 #include <nlm/nlm.h>
50 
51 FEATURE(nfsd, "NFSv4 server");
52 
53 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
54 extern int nfsrv_useacl;
55 extern int newnfs_numnfsd;
56 extern struct mount nfsv4root_mnt;
57 extern struct nfsrv_stablefirst nfsrv_stablefirst;
58 extern void (*nfsd_call_servertimer)(void);
59 extern SVCPOOL	*nfsrvd_pool;
60 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
61 NFSDLOCKMUTEX;
62 struct mtx nfs_cache_mutex;
63 struct mtx nfs_v4root_mutex;
64 struct nfsrvfh nfs_rootfh, nfs_pubfh;
65 int nfs_pubfhset = 0, nfs_rootfhset = 0;
66 struct proc *nfsd_master_proc = NULL;
67 static pid_t nfsd_master_pid = (pid_t)-1;
68 static char nfsd_master_comm[MAXCOMLEN + 1];
69 static struct timeval nfsd_master_start;
70 static uint32_t nfsv4_sysid = 0;
71 
72 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
73     struct ucred *);
74 
75 int nfsrv_enable_crossmntpt = 1;
76 static int nfs_commit_blks;
77 static int nfs_commit_miss;
78 extern int nfsrv_issuedelegs;
79 extern int nfsrv_dolocallocks;
80 
81 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server");
82 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
83     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
84 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
85     0, "");
86 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
87     0, "");
88 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
89     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
90 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
91     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
92 
93 #define	MAX_REORDERED_RPC	16
94 #define	NUM_HEURISTIC		1031
95 #define	NHUSE_INIT		64
96 #define	NHUSE_INC		16
97 #define	NHUSE_MAX		2048
98 
99 static struct nfsheur {
100 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
101 	off_t nh_nextoff;	/* next offset for sequential detection */
102 	int nh_use;		/* use count for selection */
103 	int nh_seqcount;	/* heuristic */
104 } nfsheur[NUM_HEURISTIC];
105 
106 
107 /*
108  * Heuristic to detect sequential operation.
109  */
110 static struct nfsheur *
111 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
112 {
113 	struct nfsheur *nh;
114 	int hi, try;
115 
116 	/* Locate best candidate. */
117 	try = 32;
118 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
119 	nh = &nfsheur[hi];
120 	while (try--) {
121 		if (nfsheur[hi].nh_vp == vp) {
122 			nh = &nfsheur[hi];
123 			break;
124 		}
125 		if (nfsheur[hi].nh_use > 0)
126 			--nfsheur[hi].nh_use;
127 		hi = (hi + 1) % NUM_HEURISTIC;
128 		if (nfsheur[hi].nh_use < nh->nh_use)
129 			nh = &nfsheur[hi];
130 	}
131 
132 	/* Initialize hint if this is a new file. */
133 	if (nh->nh_vp != vp) {
134 		nh->nh_vp = vp;
135 		nh->nh_nextoff = uio->uio_offset;
136 		nh->nh_use = NHUSE_INIT;
137 		if (uio->uio_offset == 0)
138 			nh->nh_seqcount = 4;
139 		else
140 			nh->nh_seqcount = 1;
141 	}
142 
143 	/* Calculate heuristic. */
144 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
145 	    uio->uio_offset == nh->nh_nextoff) {
146 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
147 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
148 		if (nh->nh_seqcount > IO_SEQMAX)
149 			nh->nh_seqcount = IO_SEQMAX;
150 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
151 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
152 		/* Probably a reordered RPC, leave seqcount alone. */
153 	} else if (nh->nh_seqcount > 1) {
154 		nh->nh_seqcount /= 2;
155 	} else {
156 		nh->nh_seqcount = 0;
157 	}
158 	nh->nh_use += NHUSE_INC;
159 	if (nh->nh_use > NHUSE_MAX)
160 		nh->nh_use = NHUSE_MAX;
161 	return (nh);
162 }
163 
164 /*
165  * Get attributes into nfsvattr structure.
166  */
167 int
168 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
169     struct thread *p, int vpislocked)
170 {
171 	int error, lockedit = 0;
172 
173 	if (vpislocked == 0) {
174 		/*
175 		 * When vpislocked == 0, the vnode is either exclusively
176 		 * locked by this thread or not locked by this thread.
177 		 * As such, shared lock it, if not exclusively locked.
178 		 */
179 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
180 			lockedit = 1;
181 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
182 		}
183 	}
184 	error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
185 	if (lockedit != 0)
186 		NFSVOPUNLOCK(vp, 0);
187 
188 	NFSEXITCODE(error);
189 	return (error);
190 }
191 
192 /*
193  * Get a file handle for a vnode.
194  */
195 int
196 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
197 {
198 	int error;
199 
200 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
201 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
202 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
203 
204 	NFSEXITCODE(error);
205 	return (error);
206 }
207 
208 /*
209  * Perform access checking for vnodes obtained from file handles that would
210  * refer to files already opened by a Unix client. You cannot just use
211  * vn_writechk() and VOP_ACCESSX() for two reasons.
212  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
213  *     case.
214  * 2 - The owner is to be given access irrespective of mode bits for some
215  *     operations, so that processes that chmod after opening a file don't
216  *     break.
217  */
218 int
219 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
220     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
221     u_int32_t *supportedtypep)
222 {
223 	struct vattr vattr;
224 	int error = 0, getret = 0;
225 
226 	if (vpislocked == 0) {
227 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
228 			error = EPERM;
229 			goto out;
230 		}
231 	}
232 	if (accmode & VWRITE) {
233 		/* Just vn_writechk() changed to check rdonly */
234 		/*
235 		 * Disallow write attempts on read-only file systems;
236 		 * unless the file is a socket or a block or character
237 		 * device resident on the file system.
238 		 */
239 		if (NFSVNO_EXRDONLY(exp) ||
240 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
241 			switch (vp->v_type) {
242 			case VREG:
243 			case VDIR:
244 			case VLNK:
245 				error = EROFS;
246 			default:
247 				break;
248 			}
249 		}
250 		/*
251 		 * If there's shared text associated with
252 		 * the inode, try to free it up once.  If
253 		 * we fail, we can't allow writing.
254 		 */
255 		if ((vp->v_vflag & VV_TEXT) != 0 && error == 0)
256 			error = ETXTBSY;
257 	}
258 	if (error != 0) {
259 		if (vpislocked == 0)
260 			NFSVOPUNLOCK(vp, 0);
261 		goto out;
262 	}
263 
264 	/*
265 	 * Should the override still be applied when ACLs are enabled?
266 	 */
267 	error = VOP_ACCESSX(vp, accmode, cred, p);
268 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
269 		/*
270 		 * Try again with VEXPLICIT_DENY, to see if the test for
271 		 * deletion is supported.
272 		 */
273 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
274 		if (error == 0) {
275 			if (vp->v_type == VDIR) {
276 				accmode &= ~(VDELETE | VDELETE_CHILD);
277 				accmode |= VWRITE;
278 				error = VOP_ACCESSX(vp, accmode, cred, p);
279 			} else if (supportedtypep != NULL) {
280 				*supportedtypep &= ~NFSACCESS_DELETE;
281 			}
282 		}
283 	}
284 
285 	/*
286 	 * Allow certain operations for the owner (reads and writes
287 	 * on files that are already open).
288 	 */
289 	if (override != NFSACCCHK_NOOVERRIDE &&
290 	    (error == EPERM || error == EACCES)) {
291 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
292 			error = 0;
293 		else if (override & NFSACCCHK_ALLOWOWNER) {
294 			getret = VOP_GETATTR(vp, &vattr, cred);
295 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
296 				error = 0;
297 		}
298 	}
299 	if (vpislocked == 0)
300 		NFSVOPUNLOCK(vp, 0);
301 
302 out:
303 	NFSEXITCODE(error);
304 	return (error);
305 }
306 
307 /*
308  * Set attribute(s) vnop.
309  */
310 int
311 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
312     struct thread *p, struct nfsexstuff *exp)
313 {
314 	int error;
315 
316 	error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
317 	NFSEXITCODE(error);
318 	return (error);
319 }
320 
321 /*
322  * Set up nameidata for a lookup() call and do it.
323  */
324 int
325 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
326     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
327     struct vnode **retdirp)
328 {
329 	struct componentname *cnp = &ndp->ni_cnd;
330 	int i;
331 	struct iovec aiov;
332 	struct uio auio;
333 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
334 	int error = 0, crossmnt;
335 	char *cp;
336 
337 	*retdirp = NULL;
338 	cnp->cn_nameptr = cnp->cn_pnbuf;
339 	ndp->ni_strictrelative = 0;
340 	/*
341 	 * Extract and set starting directory.
342 	 */
343 	if (dp->v_type != VDIR) {
344 		if (islocked)
345 			vput(dp);
346 		else
347 			vrele(dp);
348 		nfsvno_relpathbuf(ndp);
349 		error = ENOTDIR;
350 		goto out1;
351 	}
352 	if (islocked)
353 		NFSVOPUNLOCK(dp, 0);
354 	VREF(dp);
355 	*retdirp = dp;
356 	if (NFSVNO_EXRDONLY(exp))
357 		cnp->cn_flags |= RDONLY;
358 	ndp->ni_segflg = UIO_SYSSPACE;
359 	crossmnt = 1;
360 
361 	if (nd->nd_flag & ND_PUBLOOKUP) {
362 		ndp->ni_loopcnt = 0;
363 		if (cnp->cn_pnbuf[0] == '/') {
364 			vrele(dp);
365 			/*
366 			 * Check for degenerate pathnames here, since lookup()
367 			 * panics on them.
368 			 */
369 			for (i = 1; i < ndp->ni_pathlen; i++)
370 				if (cnp->cn_pnbuf[i] != '/')
371 					break;
372 			if (i == ndp->ni_pathlen) {
373 				error = NFSERR_ACCES;
374 				goto out;
375 			}
376 			dp = rootvnode;
377 			VREF(dp);
378 		}
379 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
380 	    (nd->nd_flag & ND_NFSV4) == 0) {
381 		/*
382 		 * Only cross mount points for NFSv4 when doing a
383 		 * mount while traversing the file system above
384 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
385 		 */
386 		cnp->cn_flags |= NOCROSSMOUNT;
387 		crossmnt = 0;
388 	}
389 
390 	/*
391 	 * Initialize for scan, set ni_startdir and bump ref on dp again
392 	 * becuase lookup() will dereference ni_startdir.
393 	 */
394 
395 	cnp->cn_thread = p;
396 	ndp->ni_startdir = dp;
397 	ndp->ni_rootdir = rootvnode;
398 	ndp->ni_topdir = NULL;
399 
400 	if (!lockleaf)
401 		cnp->cn_flags |= LOCKLEAF;
402 	for (;;) {
403 		cnp->cn_nameptr = cnp->cn_pnbuf;
404 		/*
405 		 * Call lookup() to do the real work.  If an error occurs,
406 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
407 		 * we do not have to dereference anything before returning.
408 		 * In either case ni_startdir will be dereferenced and NULLed
409 		 * out.
410 		 */
411 		error = lookup(ndp);
412 		if (error)
413 			break;
414 
415 		/*
416 		 * Check for encountering a symbolic link.  Trivial
417 		 * termination occurs if no symlink encountered.
418 		 */
419 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
420 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
421 				nfsvno_relpathbuf(ndp);
422 			if (ndp->ni_vp && !lockleaf)
423 				NFSVOPUNLOCK(ndp->ni_vp, 0);
424 			break;
425 		}
426 
427 		/*
428 		 * Validate symlink
429 		 */
430 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
431 			NFSVOPUNLOCK(ndp->ni_dvp, 0);
432 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
433 			error = EINVAL;
434 			goto badlink2;
435 		}
436 
437 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
438 			error = ELOOP;
439 			goto badlink2;
440 		}
441 		if (ndp->ni_pathlen > 1)
442 			cp = uma_zalloc(namei_zone, M_WAITOK);
443 		else
444 			cp = cnp->cn_pnbuf;
445 		aiov.iov_base = cp;
446 		aiov.iov_len = MAXPATHLEN;
447 		auio.uio_iov = &aiov;
448 		auio.uio_iovcnt = 1;
449 		auio.uio_offset = 0;
450 		auio.uio_rw = UIO_READ;
451 		auio.uio_segflg = UIO_SYSSPACE;
452 		auio.uio_td = NULL;
453 		auio.uio_resid = MAXPATHLEN;
454 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
455 		if (error) {
456 		badlink1:
457 			if (ndp->ni_pathlen > 1)
458 				uma_zfree(namei_zone, cp);
459 		badlink2:
460 			vrele(ndp->ni_dvp);
461 			vput(ndp->ni_vp);
462 			break;
463 		}
464 		linklen = MAXPATHLEN - auio.uio_resid;
465 		if (linklen == 0) {
466 			error = ENOENT;
467 			goto badlink1;
468 		}
469 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
470 			error = ENAMETOOLONG;
471 			goto badlink1;
472 		}
473 
474 		/*
475 		 * Adjust or replace path
476 		 */
477 		if (ndp->ni_pathlen > 1) {
478 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
479 			uma_zfree(namei_zone, cnp->cn_pnbuf);
480 			cnp->cn_pnbuf = cp;
481 		} else
482 			cnp->cn_pnbuf[linklen] = '\0';
483 		ndp->ni_pathlen += linklen;
484 
485 		/*
486 		 * Cleanup refs for next loop and check if root directory
487 		 * should replace current directory.  Normally ni_dvp
488 		 * becomes the new base directory and is cleaned up when
489 		 * we loop.  Explicitly null pointers after invalidation
490 		 * to clarify operation.
491 		 */
492 		vput(ndp->ni_vp);
493 		ndp->ni_vp = NULL;
494 
495 		if (cnp->cn_pnbuf[0] == '/') {
496 			vrele(ndp->ni_dvp);
497 			ndp->ni_dvp = ndp->ni_rootdir;
498 			VREF(ndp->ni_dvp);
499 		}
500 		ndp->ni_startdir = ndp->ni_dvp;
501 		ndp->ni_dvp = NULL;
502 	}
503 	if (!lockleaf)
504 		cnp->cn_flags &= ~LOCKLEAF;
505 
506 out:
507 	if (error) {
508 		uma_zfree(namei_zone, cnp->cn_pnbuf);
509 		ndp->ni_vp = NULL;
510 		ndp->ni_dvp = NULL;
511 		ndp->ni_startdir = NULL;
512 		cnp->cn_flags &= ~HASBUF;
513 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
514 		ndp->ni_dvp = NULL;
515 	}
516 
517 out1:
518 	NFSEXITCODE2(error, nd);
519 	return (error);
520 }
521 
522 /*
523  * Set up a pathname buffer and return a pointer to it and, optionally
524  * set a hash pointer.
525  */
526 void
527 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
528 {
529 	struct componentname *cnp = &ndp->ni_cnd;
530 
531 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
532 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
533 	if (hashpp != NULL)
534 		*hashpp = NULL;
535 	*bufpp = cnp->cn_pnbuf;
536 }
537 
538 /*
539  * Release the above path buffer, if not released by nfsvno_namei().
540  */
541 void
542 nfsvno_relpathbuf(struct nameidata *ndp)
543 {
544 
545 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
546 		panic("nfsrelpath");
547 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
548 	ndp->ni_cnd.cn_flags &= ~HASBUF;
549 }
550 
551 /*
552  * Readlink vnode op into an mbuf list.
553  */
554 int
555 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
556     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
557 {
558 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
559 	struct iovec *ivp = iv;
560 	struct uio io, *uiop = &io;
561 	struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
562 	int i, len, tlen, error = 0;
563 
564 	len = 0;
565 	i = 0;
566 	while (len < NFS_MAXPATHLEN) {
567 		NFSMGET(mp);
568 		MCLGET(mp, M_WAIT);
569 		mp->m_len = NFSMSIZ(mp);
570 		if (len == 0) {
571 			mp3 = mp2 = mp;
572 		} else {
573 			mp2->m_next = mp;
574 			mp2 = mp;
575 		}
576 		if ((len + mp->m_len) > NFS_MAXPATHLEN) {
577 			mp->m_len = NFS_MAXPATHLEN - len;
578 			len = NFS_MAXPATHLEN;
579 		} else {
580 			len += mp->m_len;
581 		}
582 		ivp->iov_base = mtod(mp, caddr_t);
583 		ivp->iov_len = mp->m_len;
584 		i++;
585 		ivp++;
586 	}
587 	uiop->uio_iov = iv;
588 	uiop->uio_iovcnt = i;
589 	uiop->uio_offset = 0;
590 	uiop->uio_resid = len;
591 	uiop->uio_rw = UIO_READ;
592 	uiop->uio_segflg = UIO_SYSSPACE;
593 	uiop->uio_td = NULL;
594 	error = VOP_READLINK(vp, uiop, cred);
595 	if (error) {
596 		m_freem(mp3);
597 		*lenp = 0;
598 		goto out;
599 	}
600 	if (uiop->uio_resid > 0) {
601 		len -= uiop->uio_resid;
602 		tlen = NFSM_RNDUP(len);
603 		nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
604 	}
605 	*lenp = len;
606 	*mpp = mp3;
607 	*mpendp = mp;
608 
609 out:
610 	NFSEXITCODE(error);
611 	return (error);
612 }
613 
614 /*
615  * Read vnode op call into mbuf list.
616  */
617 int
618 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
619     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
620 {
621 	struct mbuf *m;
622 	int i;
623 	struct iovec *iv;
624 	struct iovec *iv2;
625 	int error = 0, len, left, siz, tlen, ioflag = 0;
626 	struct mbuf *m2 = NULL, *m3;
627 	struct uio io, *uiop = &io;
628 	struct nfsheur *nh;
629 
630 	len = left = NFSM_RNDUP(cnt);
631 	m3 = NULL;
632 	/*
633 	 * Generate the mbuf list with the uio_iov ref. to it.
634 	 */
635 	i = 0;
636 	while (left > 0) {
637 		NFSMGET(m);
638 		MCLGET(m, M_WAIT);
639 		m->m_len = 0;
640 		siz = min(M_TRAILINGSPACE(m), left);
641 		left -= siz;
642 		i++;
643 		if (m3)
644 			m2->m_next = m;
645 		else
646 			m3 = m;
647 		m2 = m;
648 	}
649 	MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
650 	    M_TEMP, M_WAITOK);
651 	uiop->uio_iov = iv2 = iv;
652 	m = m3;
653 	left = len;
654 	i = 0;
655 	while (left > 0) {
656 		if (m == NULL)
657 			panic("nfsvno_read iov");
658 		siz = min(M_TRAILINGSPACE(m), left);
659 		if (siz > 0) {
660 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
661 			iv->iov_len = siz;
662 			m->m_len += siz;
663 			left -= siz;
664 			iv++;
665 			i++;
666 		}
667 		m = m->m_next;
668 	}
669 	uiop->uio_iovcnt = i;
670 	uiop->uio_offset = off;
671 	uiop->uio_resid = len;
672 	uiop->uio_rw = UIO_READ;
673 	uiop->uio_segflg = UIO_SYSSPACE;
674 	nh = nfsrv_sequential_heuristic(uiop, vp);
675 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
676 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
677 	FREE((caddr_t)iv2, M_TEMP);
678 	if (error) {
679 		m_freem(m3);
680 		*mpp = NULL;
681 		goto out;
682 	}
683 	nh->nh_nextoff = uiop->uio_offset;
684 	tlen = len - uiop->uio_resid;
685 	cnt = cnt < tlen ? cnt : tlen;
686 	tlen = NFSM_RNDUP(cnt);
687 	if (tlen == 0) {
688 		m_freem(m3);
689 		m3 = NULL;
690 	} else if (len != tlen || tlen != cnt)
691 		nfsrv_adj(m3, len - tlen, tlen - cnt);
692 	*mpp = m3;
693 	*mpendp = m2;
694 
695 out:
696 	NFSEXITCODE(error);
697 	return (error);
698 }
699 
700 /*
701  * Write vnode op from an mbuf list.
702  */
703 int
704 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
705     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
706 {
707 	struct iovec *ivp;
708 	int i, len;
709 	struct iovec *iv;
710 	int ioflags, error;
711 	struct uio io, *uiop = &io;
712 	struct nfsheur *nh;
713 
714 	MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
715 	    M_WAITOK);
716 	uiop->uio_iov = iv = ivp;
717 	uiop->uio_iovcnt = cnt;
718 	i = mtod(mp, caddr_t) + mp->m_len - cp;
719 	len = retlen;
720 	while (len > 0) {
721 		if (mp == NULL)
722 			panic("nfsvno_write");
723 		if (i > 0) {
724 			i = min(i, len);
725 			ivp->iov_base = cp;
726 			ivp->iov_len = i;
727 			ivp++;
728 			len -= i;
729 		}
730 		mp = mp->m_next;
731 		if (mp) {
732 			i = mp->m_len;
733 			cp = mtod(mp, caddr_t);
734 		}
735 	}
736 
737 	if (stable == NFSWRITE_UNSTABLE)
738 		ioflags = IO_NODELOCKED;
739 	else
740 		ioflags = (IO_SYNC | IO_NODELOCKED);
741 	uiop->uio_resid = retlen;
742 	uiop->uio_rw = UIO_WRITE;
743 	uiop->uio_segflg = UIO_SYSSPACE;
744 	NFSUIOPROC(uiop, p);
745 	uiop->uio_offset = off;
746 	nh = nfsrv_sequential_heuristic(uiop, vp);
747 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
748 	error = VOP_WRITE(vp, uiop, ioflags, cred);
749 	if (error == 0)
750 		nh->nh_nextoff = uiop->uio_offset;
751 	FREE((caddr_t)iv, M_TEMP);
752 
753 	NFSEXITCODE(error);
754 	return (error);
755 }
756 
757 /*
758  * Common code for creating a regular file (plus special files for V2).
759  */
760 int
761 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
762     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
763     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
764 {
765 	u_quad_t tempsize;
766 	int error;
767 
768 	error = nd->nd_repstat;
769 	if (!error && ndp->ni_vp == NULL) {
770 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
771 			vrele(ndp->ni_startdir);
772 			error = VOP_CREATE(ndp->ni_dvp,
773 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
774 			vput(ndp->ni_dvp);
775 			nfsvno_relpathbuf(ndp);
776 			if (!error) {
777 				if (*exclusive_flagp) {
778 					*exclusive_flagp = 0;
779 					NFSVNO_ATTRINIT(nvap);
780 					nvap->na_atime.tv_sec = cverf[0];
781 					nvap->na_atime.tv_nsec = cverf[1];
782 					error = VOP_SETATTR(ndp->ni_vp,
783 					    &nvap->na_vattr, nd->nd_cred);
784 				}
785 			}
786 		/*
787 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
788 		 * (This implies, just get out on an error.)
789 		 */
790 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
791 			nvap->na_type == VFIFO) {
792 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
793 				nvap->na_type = VFIFO;
794                         if (nvap->na_type != VFIFO &&
795 			    (error = priv_check_cred(nd->nd_cred,
796 			     PRIV_VFS_MKNOD_DEV, 0))) {
797 				vrele(ndp->ni_startdir);
798 				nfsvno_relpathbuf(ndp);
799 				vput(ndp->ni_dvp);
800 				goto out;
801 			}
802 			nvap->na_rdev = rdev;
803 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
804 			    &ndp->ni_cnd, &nvap->na_vattr);
805 			vput(ndp->ni_dvp);
806 			nfsvno_relpathbuf(ndp);
807 			vrele(ndp->ni_startdir);
808 			if (error)
809 				goto out;
810 		} else {
811 			vrele(ndp->ni_startdir);
812 			nfsvno_relpathbuf(ndp);
813 			vput(ndp->ni_dvp);
814 			error = ENXIO;
815 			goto out;
816 		}
817 		*vpp = ndp->ni_vp;
818 	} else {
819 		/*
820 		 * Handle cases where error is already set and/or
821 		 * the file exists.
822 		 * 1 - clean up the lookup
823 		 * 2 - iff !error and na_size set, truncate it
824 		 */
825 		vrele(ndp->ni_startdir);
826 		nfsvno_relpathbuf(ndp);
827 		*vpp = ndp->ni_vp;
828 		if (ndp->ni_dvp == *vpp)
829 			vrele(ndp->ni_dvp);
830 		else
831 			vput(ndp->ni_dvp);
832 		if (!error && nvap->na_size != VNOVAL) {
833 			error = nfsvno_accchk(*vpp, VWRITE,
834 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
835 			    NFSACCCHK_VPISLOCKED, NULL);
836 			if (!error) {
837 				tempsize = nvap->na_size;
838 				NFSVNO_ATTRINIT(nvap);
839 				nvap->na_size = tempsize;
840 				error = VOP_SETATTR(*vpp,
841 				    &nvap->na_vattr, nd->nd_cred);
842 			}
843 		}
844 		if (error)
845 			vput(*vpp);
846 	}
847 
848 out:
849 	NFSEXITCODE(error);
850 	return (error);
851 }
852 
853 /*
854  * Do a mknod vnode op.
855  */
856 int
857 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
858     struct thread *p)
859 {
860 	int error = 0;
861 	enum vtype vtyp;
862 
863 	vtyp = nvap->na_type;
864 	/*
865 	 * Iff doesn't exist, create it.
866 	 */
867 	if (ndp->ni_vp) {
868 		vrele(ndp->ni_startdir);
869 		nfsvno_relpathbuf(ndp);
870 		vput(ndp->ni_dvp);
871 		vrele(ndp->ni_vp);
872 		error = EEXIST;
873 		goto out;
874 	}
875 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
876 		vrele(ndp->ni_startdir);
877 		nfsvno_relpathbuf(ndp);
878 		vput(ndp->ni_dvp);
879 		error = NFSERR_BADTYPE;
880 		goto out;
881 	}
882 	if (vtyp == VSOCK) {
883 		vrele(ndp->ni_startdir);
884 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
885 		    &ndp->ni_cnd, &nvap->na_vattr);
886 		vput(ndp->ni_dvp);
887 		nfsvno_relpathbuf(ndp);
888 	} else {
889 		if (nvap->na_type != VFIFO &&
890 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
891 			vrele(ndp->ni_startdir);
892 			nfsvno_relpathbuf(ndp);
893 			vput(ndp->ni_dvp);
894 			goto out;
895 		}
896 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
897 		    &ndp->ni_cnd, &nvap->na_vattr);
898 		vput(ndp->ni_dvp);
899 		nfsvno_relpathbuf(ndp);
900 		vrele(ndp->ni_startdir);
901 		/*
902 		 * Since VOP_MKNOD returns the ni_vp, I can't
903 		 * see any reason to do the lookup.
904 		 */
905 	}
906 
907 out:
908 	NFSEXITCODE(error);
909 	return (error);
910 }
911 
912 /*
913  * Mkdir vnode op.
914  */
915 int
916 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
917     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
918 {
919 	int error = 0;
920 
921 	if (ndp->ni_vp != NULL) {
922 		if (ndp->ni_dvp == ndp->ni_vp)
923 			vrele(ndp->ni_dvp);
924 		else
925 			vput(ndp->ni_dvp);
926 		vrele(ndp->ni_vp);
927 		nfsvno_relpathbuf(ndp);
928 		error = EEXIST;
929 		goto out;
930 	}
931 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
932 	    &nvap->na_vattr);
933 	vput(ndp->ni_dvp);
934 	nfsvno_relpathbuf(ndp);
935 
936 out:
937 	NFSEXITCODE(error);
938 	return (error);
939 }
940 
941 /*
942  * symlink vnode op.
943  */
944 int
945 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
946     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
947     struct nfsexstuff *exp)
948 {
949 	int error = 0;
950 
951 	if (ndp->ni_vp) {
952 		vrele(ndp->ni_startdir);
953 		nfsvno_relpathbuf(ndp);
954 		if (ndp->ni_dvp == ndp->ni_vp)
955 			vrele(ndp->ni_dvp);
956 		else
957 			vput(ndp->ni_dvp);
958 		vrele(ndp->ni_vp);
959 		error = EEXIST;
960 		goto out;
961 	}
962 
963 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
964 	    &nvap->na_vattr, pathcp);
965 	vput(ndp->ni_dvp);
966 	vrele(ndp->ni_startdir);
967 	nfsvno_relpathbuf(ndp);
968 	/*
969 	 * Although FreeBSD still had the lookup code in
970 	 * it for 7/current, there doesn't seem to be any
971 	 * point, since VOP_SYMLINK() returns the ni_vp.
972 	 * Just vput it for v2.
973 	 */
974 	if (!not_v2 && !error)
975 		vput(ndp->ni_vp);
976 
977 out:
978 	NFSEXITCODE(error);
979 	return (error);
980 }
981 
982 /*
983  * Parse symbolic link arguments.
984  * This function has an ugly side effect. It will MALLOC() an area for
985  * the symlink and set iov_base to point to it, only if it succeeds.
986  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
987  * be FREE'd later.
988  */
989 int
990 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
991     struct thread *p, char **pathcpp, int *lenp)
992 {
993 	u_int32_t *tl;
994 	char *pathcp = NULL;
995 	int error = 0, len;
996 	struct nfsv2_sattr *sp;
997 
998 	*pathcpp = NULL;
999 	*lenp = 0;
1000 	if ((nd->nd_flag & ND_NFSV3) &&
1001 	    (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
1002 		goto nfsmout;
1003 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1004 	len = fxdr_unsigned(int, *tl);
1005 	if (len > NFS_MAXPATHLEN || len <= 0) {
1006 		error = EBADRPC;
1007 		goto nfsmout;
1008 	}
1009 	MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
1010 	error = nfsrv_mtostr(nd, pathcp, len);
1011 	if (error)
1012 		goto nfsmout;
1013 	if (nd->nd_flag & ND_NFSV2) {
1014 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1015 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1016 	}
1017 	*pathcpp = pathcp;
1018 	*lenp = len;
1019 	NFSEXITCODE2(0, nd);
1020 	return (0);
1021 nfsmout:
1022 	if (pathcp)
1023 		free(pathcp, M_TEMP);
1024 	NFSEXITCODE2(error, nd);
1025 	return (error);
1026 }
1027 
1028 /*
1029  * Remove a non-directory object.
1030  */
1031 int
1032 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1033     struct thread *p, struct nfsexstuff *exp)
1034 {
1035 	struct vnode *vp;
1036 	int error = 0;
1037 
1038 	vp = ndp->ni_vp;
1039 	if (vp->v_type == VDIR)
1040 		error = NFSERR_ISDIR;
1041 	else if (is_v4)
1042 		error = nfsrv_checkremove(vp, 1, p);
1043 	if (!error)
1044 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1045 	if (ndp->ni_dvp == vp)
1046 		vrele(ndp->ni_dvp);
1047 	else
1048 		vput(ndp->ni_dvp);
1049 	vput(vp);
1050 	NFSEXITCODE(error);
1051 	return (error);
1052 }
1053 
1054 /*
1055  * Remove a directory.
1056  */
1057 int
1058 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1059     struct thread *p, struct nfsexstuff *exp)
1060 {
1061 	struct vnode *vp;
1062 	int error = 0;
1063 
1064 	vp = ndp->ni_vp;
1065 	if (vp->v_type != VDIR) {
1066 		error = ENOTDIR;
1067 		goto out;
1068 	}
1069 	/*
1070 	 * No rmdir "." please.
1071 	 */
1072 	if (ndp->ni_dvp == vp) {
1073 		error = EINVAL;
1074 		goto out;
1075 	}
1076 	/*
1077 	 * The root of a mounted filesystem cannot be deleted.
1078 	 */
1079 	if (vp->v_vflag & VV_ROOT)
1080 		error = EBUSY;
1081 out:
1082 	if (!error)
1083 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1084 	if (ndp->ni_dvp == vp)
1085 		vrele(ndp->ni_dvp);
1086 	else
1087 		vput(ndp->ni_dvp);
1088 	vput(vp);
1089 	NFSEXITCODE(error);
1090 	return (error);
1091 }
1092 
1093 /*
1094  * Rename vnode op.
1095  */
1096 int
1097 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1098     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1099 {
1100 	struct vnode *fvp, *tvp, *tdvp;
1101 	int error = 0;
1102 
1103 	fvp = fromndp->ni_vp;
1104 	if (ndstat) {
1105 		vrele(fromndp->ni_dvp);
1106 		vrele(fvp);
1107 		error = ndstat;
1108 		goto out1;
1109 	}
1110 	tdvp = tondp->ni_dvp;
1111 	tvp = tondp->ni_vp;
1112 	if (tvp != NULL) {
1113 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1114 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1115 			goto out;
1116 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1117 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1118 			goto out;
1119 		}
1120 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1121 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1122 			goto out;
1123 		}
1124 
1125 		/*
1126 		 * A rename to '.' or '..' results in a prematurely
1127 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1128 		 * here.
1129 		 */
1130 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1131 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1132 		    (tondp->ni_cnd.cn_namelen == 2 &&
1133 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1134 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1135 			error = EINVAL;
1136 			goto out;
1137 		}
1138 	}
1139 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1140 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1141 		goto out;
1142 	}
1143 	if (fvp->v_mount != tdvp->v_mount) {
1144 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1145 		goto out;
1146 	}
1147 	if (fvp == tdvp) {
1148 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1149 		goto out;
1150 	}
1151 	if (fvp == tvp) {
1152 		/*
1153 		 * If source and destination are the same, there is nothing to
1154 		 * do. Set error to -1 to indicate this.
1155 		 */
1156 		error = -1;
1157 		goto out;
1158 	}
1159 	if (ndflag & ND_NFSV4) {
1160 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1161 			error = nfsrv_checkremove(fvp, 0, p);
1162 			NFSVOPUNLOCK(fvp, 0);
1163 		} else
1164 			error = EPERM;
1165 		if (tvp && !error)
1166 			error = nfsrv_checkremove(tvp, 1, p);
1167 	} else {
1168 		/*
1169 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1170 		 * that the NFSv4 client won't be confused by the rename.
1171 		 * Since nfsd_recalldelegation() can only be called on an
1172 		 * unlocked vnode at this point and fvp is the file that will
1173 		 * still exist after the rename, just do fvp.
1174 		 */
1175 		nfsd_recalldelegation(fvp, p);
1176 	}
1177 out:
1178 	if (!error) {
1179 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1180 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1181 		    &tondp->ni_cnd);
1182 	} else {
1183 		if (tdvp == tvp)
1184 			vrele(tdvp);
1185 		else
1186 			vput(tdvp);
1187 		if (tvp)
1188 			vput(tvp);
1189 		vrele(fromndp->ni_dvp);
1190 		vrele(fvp);
1191 		if (error == -1)
1192 			error = 0;
1193 	}
1194 	vrele(tondp->ni_startdir);
1195 	nfsvno_relpathbuf(tondp);
1196 out1:
1197 	vrele(fromndp->ni_startdir);
1198 	nfsvno_relpathbuf(fromndp);
1199 	NFSEXITCODE(error);
1200 	return (error);
1201 }
1202 
1203 /*
1204  * Link vnode op.
1205  */
1206 int
1207 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1208     struct thread *p, struct nfsexstuff *exp)
1209 {
1210 	struct vnode *xp;
1211 	int error = 0;
1212 
1213 	xp = ndp->ni_vp;
1214 	if (xp != NULL) {
1215 		error = EEXIST;
1216 	} else {
1217 		xp = ndp->ni_dvp;
1218 		if (vp->v_mount != xp->v_mount)
1219 			error = EXDEV;
1220 	}
1221 	if (!error) {
1222 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1223 		if ((vp->v_iflag & VI_DOOMED) == 0)
1224 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1225 		else
1226 			error = EPERM;
1227 		if (ndp->ni_dvp == vp)
1228 			vrele(ndp->ni_dvp);
1229 		else
1230 			vput(ndp->ni_dvp);
1231 		NFSVOPUNLOCK(vp, 0);
1232 	} else {
1233 		if (ndp->ni_dvp == ndp->ni_vp)
1234 			vrele(ndp->ni_dvp);
1235 		else
1236 			vput(ndp->ni_dvp);
1237 		if (ndp->ni_vp)
1238 			vrele(ndp->ni_vp);
1239 	}
1240 	nfsvno_relpathbuf(ndp);
1241 	NFSEXITCODE(error);
1242 	return (error);
1243 }
1244 
1245 /*
1246  * Do the fsync() appropriate for the commit.
1247  */
1248 int
1249 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1250     struct thread *td)
1251 {
1252 	int error = 0;
1253 
1254 	/*
1255 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1256 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1257 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1258 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1259 	 */
1260 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT) {
1261 		/*
1262 		 * Give up and do the whole thing
1263 		 */
1264 		if (vp->v_object &&
1265 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1266 			VM_OBJECT_LOCK(vp->v_object);
1267 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1268 			VM_OBJECT_UNLOCK(vp->v_object);
1269 		}
1270 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1271 	} else {
1272 		/*
1273 		 * Locate and synchronously write any buffers that fall
1274 		 * into the requested range.  Note:  we are assuming that
1275 		 * f_iosize is a power of 2.
1276 		 */
1277 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1278 		int iomask = iosize - 1;
1279 		struct bufobj *bo;
1280 		daddr_t lblkno;
1281 
1282 		/*
1283 		 * Align to iosize boundry, super-align to page boundry.
1284 		 */
1285 		if (off & iomask) {
1286 			cnt += off & iomask;
1287 			off &= ~(u_quad_t)iomask;
1288 		}
1289 		if (off & PAGE_MASK) {
1290 			cnt += off & PAGE_MASK;
1291 			off &= ~(u_quad_t)PAGE_MASK;
1292 		}
1293 		lblkno = off / iosize;
1294 
1295 		if (vp->v_object &&
1296 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1297 			VM_OBJECT_LOCK(vp->v_object);
1298 			vm_object_page_clean(vp->v_object, off, off + cnt,
1299 			    OBJPC_SYNC);
1300 			VM_OBJECT_UNLOCK(vp->v_object);
1301 		}
1302 
1303 		bo = &vp->v_bufobj;
1304 		BO_LOCK(bo);
1305 		while (cnt > 0) {
1306 			struct buf *bp;
1307 
1308 			/*
1309 			 * If we have a buffer and it is marked B_DELWRI we
1310 			 * have to lock and write it.  Otherwise the prior
1311 			 * write is assumed to have already been committed.
1312 			 *
1313 			 * gbincore() can return invalid buffers now so we
1314 			 * have to check that bit as well (though B_DELWRI
1315 			 * should not be set if B_INVAL is set there could be
1316 			 * a race here since we haven't locked the buffer).
1317 			 */
1318 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1319 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1320 				    LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
1321 					BO_LOCK(bo);
1322 					continue; /* retry */
1323 				}
1324 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1325 				    B_DELWRI) {
1326 					bremfree(bp);
1327 					bp->b_flags &= ~B_ASYNC;
1328 					bwrite(bp);
1329 					++nfs_commit_miss;
1330 				} else
1331 					BUF_UNLOCK(bp);
1332 				BO_LOCK(bo);
1333 			}
1334 			++nfs_commit_blks;
1335 			if (cnt < iosize)
1336 				break;
1337 			cnt -= iosize;
1338 			++lblkno;
1339 		}
1340 		BO_UNLOCK(bo);
1341 	}
1342 	NFSEXITCODE(error);
1343 	return (error);
1344 }
1345 
1346 /*
1347  * Statfs vnode op.
1348  */
1349 int
1350 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1351 {
1352 	int error;
1353 
1354 	error = VFS_STATFS(vp->v_mount, sf);
1355 	if (error == 0) {
1356 		/*
1357 		 * Since NFS handles these values as unsigned on the
1358 		 * wire, there is no way to represent negative values,
1359 		 * so set them to 0. Without this, they will appear
1360 		 * to be very large positive values for clients like
1361 		 * Solaris10.
1362 		 */
1363 		if (sf->f_bavail < 0)
1364 			sf->f_bavail = 0;
1365 		if (sf->f_ffree < 0)
1366 			sf->f_ffree = 0;
1367 	}
1368 	NFSEXITCODE(error);
1369 	return (error);
1370 }
1371 
1372 /*
1373  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1374  * must handle nfsrv_opencheck() calls after any other access checks.
1375  */
1376 void
1377 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1378     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1379     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1380     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1381     struct nfsexstuff *exp, struct vnode **vpp)
1382 {
1383 	struct vnode *vp = NULL;
1384 	u_quad_t tempsize;
1385 	struct nfsexstuff nes;
1386 
1387 	if (ndp->ni_vp == NULL)
1388 		nd->nd_repstat = nfsrv_opencheck(clientid,
1389 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1390 	if (!nd->nd_repstat) {
1391 		if (ndp->ni_vp == NULL) {
1392 			vrele(ndp->ni_startdir);
1393 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1394 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1395 			vput(ndp->ni_dvp);
1396 			nfsvno_relpathbuf(ndp);
1397 			if (!nd->nd_repstat) {
1398 				if (*exclusive_flagp) {
1399 					*exclusive_flagp = 0;
1400 					NFSVNO_ATTRINIT(nvap);
1401 					nvap->na_atime.tv_sec = cverf[0];
1402 					nvap->na_atime.tv_nsec = cverf[1];
1403 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1404 					    &nvap->na_vattr, cred);
1405 				} else {
1406 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1407 					    aclp, p, attrbitp, exp);
1408 				}
1409 			}
1410 			vp = ndp->ni_vp;
1411 		} else {
1412 			if (ndp->ni_startdir)
1413 				vrele(ndp->ni_startdir);
1414 			nfsvno_relpathbuf(ndp);
1415 			vp = ndp->ni_vp;
1416 			if (create == NFSV4OPEN_CREATE) {
1417 				if (ndp->ni_dvp == vp)
1418 					vrele(ndp->ni_dvp);
1419 				else
1420 					vput(ndp->ni_dvp);
1421 			}
1422 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1423 				if (ndp->ni_cnd.cn_flags & RDONLY)
1424 					NFSVNO_SETEXRDONLY(&nes);
1425 				else
1426 					NFSVNO_EXINIT(&nes);
1427 				nd->nd_repstat = nfsvno_accchk(vp,
1428 				    VWRITE, cred, &nes, p,
1429 				    NFSACCCHK_NOOVERRIDE,
1430 				    NFSACCCHK_VPISLOCKED, NULL);
1431 				nd->nd_repstat = nfsrv_opencheck(clientid,
1432 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1433 				if (!nd->nd_repstat) {
1434 					tempsize = nvap->na_size;
1435 					NFSVNO_ATTRINIT(nvap);
1436 					nvap->na_size = tempsize;
1437 					nd->nd_repstat = VOP_SETATTR(vp,
1438 					    &nvap->na_vattr, cred);
1439 				}
1440 			} else if (vp->v_type == VREG) {
1441 				nd->nd_repstat = nfsrv_opencheck(clientid,
1442 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1443 			}
1444 		}
1445 	} else {
1446 		if (ndp->ni_cnd.cn_flags & HASBUF)
1447 			nfsvno_relpathbuf(ndp);
1448 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1449 			vrele(ndp->ni_startdir);
1450 			if (ndp->ni_dvp == ndp->ni_vp)
1451 				vrele(ndp->ni_dvp);
1452 			else
1453 				vput(ndp->ni_dvp);
1454 			if (ndp->ni_vp)
1455 				vput(ndp->ni_vp);
1456 		}
1457 	}
1458 	*vpp = vp;
1459 
1460 	NFSEXITCODE2(0, nd);
1461 }
1462 
1463 /*
1464  * Updates the file rev and sets the mtime and ctime
1465  * to the current clock time, returning the va_filerev and va_Xtime
1466  * values.
1467  */
1468 void
1469 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1470     struct ucred *cred, struct thread *p)
1471 {
1472 	struct vattr va;
1473 
1474 	VATTR_NULL(&va);
1475 	getnanotime(&va.va_mtime);
1476 	(void) VOP_SETATTR(vp, &va, cred);
1477 	(void) nfsvno_getattr(vp, nvap, cred, p, 1);
1478 }
1479 
1480 /*
1481  * Glue routine to nfsv4_fillattr().
1482  */
1483 int
1484 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1485     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1486     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1487     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1488 {
1489 	int error;
1490 
1491 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1492 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1493 	    mounted_on_fileno);
1494 	NFSEXITCODE2(0, nd);
1495 	return (error);
1496 }
1497 
1498 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1499 /*
1500  * nfs readdir service
1501  * - mallocs what it thinks is enough to read
1502  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1503  * - calls VOP_READDIR()
1504  * - loops around building the reply
1505  *	if the output generated exceeds count break out of loop
1506  *	The NFSM_CLGET macro is used here so that the reply will be packed
1507  *	tightly in mbuf clusters.
1508  * - it trims out records with d_fileno == 0
1509  *	this doesn't matter for Unix clients, but they might confuse clients
1510  *	for other os'.
1511  * - it trims out records with d_type == DT_WHT
1512  *	these cannot be seen through NFS (unless we extend the protocol)
1513  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1514  * PS: The NFS protocol spec. does not clarify what the "count" byte
1515  *	argument is a count of.. just name strings and file id's or the
1516  *	entire reply rpc or ...
1517  *	I tried just file name and id sizes and it confused the Sun client,
1518  *	so I am using the full rpc size now. The "paranoia.." comment refers
1519  *	to including the status longwords that are not a part of the dir.
1520  *	"entry" structures, but are in the rpc.
1521  */
1522 int
1523 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1524     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1525 {
1526 	struct dirent *dp;
1527 	u_int32_t *tl;
1528 	int dirlen;
1529 	char *cpos, *cend, *rbuf;
1530 	struct nfsvattr at;
1531 	int nlen, error = 0, getret = 1;
1532 	int siz, cnt, fullsiz, eofflag, ncookies;
1533 	u_int64_t off, toff, verf;
1534 	u_long *cookies = NULL, *cookiep;
1535 	struct uio io;
1536 	struct iovec iv;
1537 	int not_zfs;
1538 
1539 	if (nd->nd_repstat) {
1540 		nfsrv_postopattr(nd, getret, &at);
1541 		goto out;
1542 	}
1543 	if (nd->nd_flag & ND_NFSV2) {
1544 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1545 		off = fxdr_unsigned(u_quad_t, *tl++);
1546 	} else {
1547 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1548 		off = fxdr_hyper(tl);
1549 		tl += 2;
1550 		verf = fxdr_hyper(tl);
1551 		tl += 2;
1552 	}
1553 	toff = off;
1554 	cnt = fxdr_unsigned(int, *tl);
1555 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1556 		cnt = NFS_SRVMAXDATA(nd);
1557 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1558 	fullsiz = siz;
1559 	if (nd->nd_flag & ND_NFSV3) {
1560 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1561 		    p, 1);
1562 #if 0
1563 		/*
1564 		 * va_filerev is not sufficient as a cookie verifier,
1565 		 * since it is not supposed to change when entries are
1566 		 * removed/added unless that offset cookies returned to
1567 		 * the client are no longer valid.
1568 		 */
1569 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
1570 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1571 #endif
1572 	}
1573 	if (nd->nd_repstat == 0 && cnt == 0) {
1574 		if (nd->nd_flag & ND_NFSV2)
1575 			/* NFSv2 does not have NFSERR_TOOSMALL */
1576 			nd->nd_repstat = EPERM;
1577 		else
1578 			nd->nd_repstat = NFSERR_TOOSMALL;
1579 	}
1580 	if (!nd->nd_repstat)
1581 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1582 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1583 		    NFSACCCHK_VPISLOCKED, NULL);
1584 	if (nd->nd_repstat) {
1585 		vput(vp);
1586 		if (nd->nd_flag & ND_NFSV3)
1587 			nfsrv_postopattr(nd, getret, &at);
1588 		goto out;
1589 	}
1590 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1591 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1592 again:
1593 	eofflag = 0;
1594 	if (cookies) {
1595 		free((caddr_t)cookies, M_TEMP);
1596 		cookies = NULL;
1597 	}
1598 
1599 	iv.iov_base = rbuf;
1600 	iv.iov_len = siz;
1601 	io.uio_iov = &iv;
1602 	io.uio_iovcnt = 1;
1603 	io.uio_offset = (off_t)off;
1604 	io.uio_resid = siz;
1605 	io.uio_segflg = UIO_SYSSPACE;
1606 	io.uio_rw = UIO_READ;
1607 	io.uio_td = NULL;
1608 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1609 	    &cookies);
1610 	off = (u_int64_t)io.uio_offset;
1611 	if (io.uio_resid)
1612 		siz -= io.uio_resid;
1613 
1614 	if (!cookies && !nd->nd_repstat)
1615 		nd->nd_repstat = NFSERR_PERM;
1616 	if (nd->nd_flag & ND_NFSV3) {
1617 		getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1618 		if (!nd->nd_repstat)
1619 			nd->nd_repstat = getret;
1620 	}
1621 
1622 	/*
1623 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
1624 	 */
1625 	if (nd->nd_repstat) {
1626 		vput(vp);
1627 		free((caddr_t)rbuf, M_TEMP);
1628 		if (cookies)
1629 			free((caddr_t)cookies, M_TEMP);
1630 		if (nd->nd_flag & ND_NFSV3)
1631 			nfsrv_postopattr(nd, getret, &at);
1632 		goto out;
1633 	}
1634 	/*
1635 	 * If nothing read, return eof
1636 	 * rpc reply
1637 	 */
1638 	if (siz == 0) {
1639 		vput(vp);
1640 		if (nd->nd_flag & ND_NFSV2) {
1641 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1642 		} else {
1643 			nfsrv_postopattr(nd, getret, &at);
1644 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1645 			txdr_hyper(at.na_filerev, tl);
1646 			tl += 2;
1647 		}
1648 		*tl++ = newnfs_false;
1649 		*tl = newnfs_true;
1650 		FREE((caddr_t)rbuf, M_TEMP);
1651 		FREE((caddr_t)cookies, M_TEMP);
1652 		goto out;
1653 	}
1654 
1655 	/*
1656 	 * Check for degenerate cases of nothing useful read.
1657 	 * If so go try again
1658 	 */
1659 	cpos = rbuf;
1660 	cend = rbuf + siz;
1661 	dp = (struct dirent *)cpos;
1662 	cookiep = cookies;
1663 
1664 	/*
1665 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1666 	 * directory offset up to a block boundary, so it is necessary to
1667 	 * skip over the records that precede the requested offset. This
1668 	 * requires the assumption that file offset cookies monotonically
1669 	 * increase.
1670 	 * Since the offset cookies don't monotonically increase for ZFS,
1671 	 * this is not done when ZFS is the file system.
1672 	 */
1673 	while (cpos < cend && ncookies > 0 &&
1674 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1675 	     (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
1676 		cpos += dp->d_reclen;
1677 		dp = (struct dirent *)cpos;
1678 		cookiep++;
1679 		ncookies--;
1680 	}
1681 	if (cpos >= cend || ncookies == 0) {
1682 		siz = fullsiz;
1683 		toff = off;
1684 		goto again;
1685 	}
1686 	vput(vp);
1687 
1688 	/*
1689 	 * dirlen is the size of the reply, including all XDR and must
1690 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1691 	 * if the XDR should be included in "count", but to be safe, we do.
1692 	 * (Include the two booleans at the end of the reply in dirlen now.)
1693 	 */
1694 	if (nd->nd_flag & ND_NFSV3) {
1695 		nfsrv_postopattr(nd, getret, &at);
1696 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1697 		txdr_hyper(at.na_filerev, tl);
1698 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1699 	} else {
1700 		dirlen = 2 * NFSX_UNSIGNED;
1701 	}
1702 
1703 	/* Loop through the records and build reply */
1704 	while (cpos < cend && ncookies > 0) {
1705 		nlen = dp->d_namlen;
1706 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1707 			nlen <= NFS_MAXNAMLEN) {
1708 			if (nd->nd_flag & ND_NFSV3)
1709 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1710 			else
1711 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1712 			if (dirlen > cnt) {
1713 				eofflag = 0;
1714 				break;
1715 			}
1716 
1717 			/*
1718 			 * Build the directory record xdr from
1719 			 * the dirent entry.
1720 			 */
1721 			if (nd->nd_flag & ND_NFSV3) {
1722 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1723 				*tl++ = newnfs_true;
1724 				*tl++ = 0;
1725 			} else {
1726 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1727 				*tl++ = newnfs_true;
1728 			}
1729 			*tl = txdr_unsigned(dp->d_fileno);
1730 			(void) nfsm_strtom(nd, dp->d_name, nlen);
1731 			if (nd->nd_flag & ND_NFSV3) {
1732 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1733 				*tl++ = 0;
1734 			} else
1735 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1736 			*tl = txdr_unsigned(*cookiep);
1737 		}
1738 		cpos += dp->d_reclen;
1739 		dp = (struct dirent *)cpos;
1740 		cookiep++;
1741 		ncookies--;
1742 	}
1743 	if (cpos < cend)
1744 		eofflag = 0;
1745 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1746 	*tl++ = newnfs_false;
1747 	if (eofflag)
1748 		*tl = newnfs_true;
1749 	else
1750 		*tl = newnfs_false;
1751 	FREE((caddr_t)rbuf, M_TEMP);
1752 	FREE((caddr_t)cookies, M_TEMP);
1753 
1754 out:
1755 	NFSEXITCODE2(0, nd);
1756 	return (0);
1757 nfsmout:
1758 	vput(vp);
1759 	NFSEXITCODE2(error, nd);
1760 	return (error);
1761 }
1762 
1763 /*
1764  * Readdirplus for V3 and Readdir for V4.
1765  */
1766 int
1767 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1768     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1769 {
1770 	struct dirent *dp;
1771 	u_int32_t *tl;
1772 	int dirlen;
1773 	char *cpos, *cend, *rbuf;
1774 	struct vnode *nvp;
1775 	fhandle_t nfh;
1776 	struct nfsvattr nva, at, *nvap = &nva;
1777 	struct mbuf *mb0, *mb1;
1778 	struct nfsreferral *refp;
1779 	int nlen, r, error = 0, getret = 1, usevget = 1;
1780 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1781 	caddr_t bpos0, bpos1;
1782 	u_int64_t off, toff, verf;
1783 	u_long *cookies = NULL, *cookiep;
1784 	nfsattrbit_t attrbits, rderrbits, savbits;
1785 	struct uio io;
1786 	struct iovec iv;
1787 	struct componentname cn;
1788 	int at_root, needs_unbusy, not_zfs, supports_nfsv4acls;
1789 	struct mount *mp, *new_mp;
1790 	uint64_t mounted_on_fileno;
1791 
1792 	if (nd->nd_repstat) {
1793 		nfsrv_postopattr(nd, getret, &at);
1794 		goto out;
1795 	}
1796 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1797 	off = fxdr_hyper(tl);
1798 	toff = off;
1799 	tl += 2;
1800 	verf = fxdr_hyper(tl);
1801 	tl += 2;
1802 	siz = fxdr_unsigned(int, *tl++);
1803 	cnt = fxdr_unsigned(int, *tl);
1804 
1805 	/*
1806 	 * Use the server's maximum data transfer size as the upper bound
1807 	 * on reply datalen.
1808 	 */
1809 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1810 		cnt = NFS_SRVMAXDATA(nd);
1811 
1812 	/*
1813 	 * siz is a "hint" of how much directory information (name, fileid,
1814 	 * cookie) should be in the reply. At least one client "hints" 0,
1815 	 * so I set it to cnt for that case. I also round it up to the
1816 	 * next multiple of DIRBLKSIZ.
1817 	 */
1818 	if (siz <= 0)
1819 		siz = cnt;
1820 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1821 
1822 	if (nd->nd_flag & ND_NFSV4) {
1823 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1824 		if (error)
1825 			goto nfsmout;
1826 		NFSSET_ATTRBIT(&savbits, &attrbits);
1827 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1828 		NFSZERO_ATTRBIT(&rderrbits);
1829 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1830 	} else {
1831 		NFSZERO_ATTRBIT(&attrbits);
1832 	}
1833 	fullsiz = siz;
1834 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1835 	if (!nd->nd_repstat) {
1836 	    if (off && verf != at.na_filerev) {
1837 		/*
1838 		 * va_filerev is not sufficient as a cookie verifier,
1839 		 * since it is not supposed to change when entries are
1840 		 * removed/added unless that offset cookies returned to
1841 		 * the client are no longer valid.
1842 		 */
1843 #if 0
1844 		if (nd->nd_flag & ND_NFSV4) {
1845 			nd->nd_repstat = NFSERR_NOTSAME;
1846 		} else {
1847 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1848 		}
1849 #endif
1850 	    } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1851 		nd->nd_repstat = NFSERR_BAD_COOKIE;
1852 	    }
1853 	}
1854 	if (!nd->nd_repstat && vp->v_type != VDIR)
1855 		nd->nd_repstat = NFSERR_NOTDIR;
1856 	if (!nd->nd_repstat && cnt == 0)
1857 		nd->nd_repstat = NFSERR_TOOSMALL;
1858 	if (!nd->nd_repstat)
1859 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1860 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1861 		    NFSACCCHK_VPISLOCKED, NULL);
1862 	if (nd->nd_repstat) {
1863 		vput(vp);
1864 		if (nd->nd_flag & ND_NFSV3)
1865 			nfsrv_postopattr(nd, getret, &at);
1866 		goto out;
1867 	}
1868 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1869 
1870 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1871 again:
1872 	eofflag = 0;
1873 	if (cookies) {
1874 		free((caddr_t)cookies, M_TEMP);
1875 		cookies = NULL;
1876 	}
1877 
1878 	iv.iov_base = rbuf;
1879 	iv.iov_len = siz;
1880 	io.uio_iov = &iv;
1881 	io.uio_iovcnt = 1;
1882 	io.uio_offset = (off_t)off;
1883 	io.uio_resid = siz;
1884 	io.uio_segflg = UIO_SYSSPACE;
1885 	io.uio_rw = UIO_READ;
1886 	io.uio_td = NULL;
1887 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1888 	    &cookies);
1889 	off = (u_int64_t)io.uio_offset;
1890 	if (io.uio_resid)
1891 		siz -= io.uio_resid;
1892 
1893 	getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1894 
1895 	if (!cookies && !nd->nd_repstat)
1896 		nd->nd_repstat = NFSERR_PERM;
1897 	if (!nd->nd_repstat)
1898 		nd->nd_repstat = getret;
1899 	if (nd->nd_repstat) {
1900 		vput(vp);
1901 		if (cookies)
1902 			free((caddr_t)cookies, M_TEMP);
1903 		free((caddr_t)rbuf, M_TEMP);
1904 		if (nd->nd_flag & ND_NFSV3)
1905 			nfsrv_postopattr(nd, getret, &at);
1906 		goto out;
1907 	}
1908 	/*
1909 	 * If nothing read, return eof
1910 	 * rpc reply
1911 	 */
1912 	if (siz == 0) {
1913 		vput(vp);
1914 		if (nd->nd_flag & ND_NFSV3)
1915 			nfsrv_postopattr(nd, getret, &at);
1916 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1917 		txdr_hyper(at.na_filerev, tl);
1918 		tl += 2;
1919 		*tl++ = newnfs_false;
1920 		*tl = newnfs_true;
1921 		free((caddr_t)cookies, M_TEMP);
1922 		free((caddr_t)rbuf, M_TEMP);
1923 		goto out;
1924 	}
1925 
1926 	/*
1927 	 * Check for degenerate cases of nothing useful read.
1928 	 * If so go try again
1929 	 */
1930 	cpos = rbuf;
1931 	cend = rbuf + siz;
1932 	dp = (struct dirent *)cpos;
1933 	cookiep = cookies;
1934 
1935 	/*
1936 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1937 	 * directory offset up to a block boundary, so it is necessary to
1938 	 * skip over the records that precede the requested offset. This
1939 	 * requires the assumption that file offset cookies monotonically
1940 	 * increase.
1941 	 * Since the offset cookies don't monotonically increase for ZFS,
1942 	 * this is not done when ZFS is the file system.
1943 	 */
1944 	while (cpos < cend && ncookies > 0 &&
1945 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1946 	   (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff) ||
1947 	   ((nd->nd_flag & ND_NFSV4) &&
1948 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1949 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1950 		cpos += dp->d_reclen;
1951 		dp = (struct dirent *)cpos;
1952 		cookiep++;
1953 		ncookies--;
1954 	}
1955 	if (cpos >= cend || ncookies == 0) {
1956 		siz = fullsiz;
1957 		toff = off;
1958 		goto again;
1959 	}
1960 
1961 	/*
1962 	 * Busy the file system so that the mount point won't go away
1963 	 * and, as such, VFS_VGET() can be used safely.
1964 	 */
1965 	mp = vp->v_mount;
1966 	vfs_ref(mp);
1967 	NFSVOPUNLOCK(vp, 0);
1968 	nd->nd_repstat = vfs_busy(mp, 0);
1969 	vfs_rel(mp);
1970 	if (nd->nd_repstat != 0) {
1971 		vrele(vp);
1972 		free(cookies, M_TEMP);
1973 		free(rbuf, M_TEMP);
1974 		if (nd->nd_flag & ND_NFSV3)
1975 			nfsrv_postopattr(nd, getret, &at);
1976 		goto out;
1977 	}
1978 
1979 	/*
1980 	 * Save this position, in case there is an error before one entry
1981 	 * is created.
1982 	 */
1983 	mb0 = nd->nd_mb;
1984 	bpos0 = nd->nd_bpos;
1985 
1986 	/*
1987 	 * Fill in the first part of the reply.
1988 	 * dirlen is the reply length in bytes and cannot exceed cnt.
1989 	 * (Include the two booleans at the end of the reply in dirlen now,
1990 	 *  so we recognize when we have exceeded cnt.)
1991 	 */
1992 	if (nd->nd_flag & ND_NFSV3) {
1993 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1994 		nfsrv_postopattr(nd, getret, &at);
1995 	} else {
1996 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
1997 	}
1998 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
1999 	txdr_hyper(at.na_filerev, tl);
2000 
2001 	/*
2002 	 * Save this position, in case there is an empty reply needed.
2003 	 */
2004 	mb1 = nd->nd_mb;
2005 	bpos1 = nd->nd_bpos;
2006 
2007 	/* Loop through the records and build reply */
2008 	entrycnt = 0;
2009 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2010 		nlen = dp->d_namlen;
2011 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2012 		    nlen <= NFS_MAXNAMLEN &&
2013 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2014 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2015 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2016 			/*
2017 			 * Save the current position in the reply, in case
2018 			 * this entry exceeds cnt.
2019 			 */
2020 			mb1 = nd->nd_mb;
2021 			bpos1 = nd->nd_bpos;
2022 
2023 			/*
2024 			 * For readdir_and_lookup get the vnode using
2025 			 * the file number.
2026 			 */
2027 			nvp = NULL;
2028 			refp = NULL;
2029 			r = 0;
2030 			at_root = 0;
2031 			needs_unbusy = 0;
2032 			new_mp = mp;
2033 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2034 			if ((nd->nd_flag & ND_NFSV3) ||
2035 			    NFSNONZERO_ATTRBIT(&savbits)) {
2036 				if (nd->nd_flag & ND_NFSV4)
2037 					refp = nfsv4root_getreferral(NULL,
2038 					    vp, dp->d_fileno);
2039 				if (refp == NULL) {
2040 					if (usevget)
2041 						r = VFS_VGET(mp, dp->d_fileno,
2042 						    LK_SHARED, &nvp);
2043 					else
2044 						r = EOPNOTSUPP;
2045 					if (r == EOPNOTSUPP) {
2046 						if (usevget) {
2047 							usevget = 0;
2048 							cn.cn_nameiop = LOOKUP;
2049 							cn.cn_lkflags =
2050 							    LK_SHARED |
2051 							    LK_RETRY;
2052 							cn.cn_cred =
2053 							    nd->nd_cred;
2054 							cn.cn_thread = p;
2055 						}
2056 						cn.cn_nameptr = dp->d_name;
2057 						cn.cn_namelen = nlen;
2058 						cn.cn_flags = ISLASTCN |
2059 						    NOFOLLOW | LOCKLEAF |
2060 						    MPSAFE;
2061 						if (nlen == 2 &&
2062 						    dp->d_name[0] == '.' &&
2063 						    dp->d_name[1] == '.')
2064 							cn.cn_flags |=
2065 							    ISDOTDOT;
2066 						if (NFSVOPLOCK(vp, LK_SHARED)
2067 						    != 0) {
2068 							nd->nd_repstat = EPERM;
2069 							break;
2070 						}
2071 						if ((vp->v_vflag & VV_ROOT) != 0
2072 						    && (cn.cn_flags & ISDOTDOT)
2073 						    != 0) {
2074 							vref(vp);
2075 							nvp = vp;
2076 							r = 0;
2077 						} else {
2078 							r = VOP_LOOKUP(vp, &nvp,
2079 							    &cn);
2080 							if (vp != nvp)
2081 								NFSVOPUNLOCK(vp,
2082 								    0);
2083 						}
2084 					}
2085 
2086 					/*
2087 					 * For NFSv4, check to see if nvp is
2088 					 * a mount point and get the mount
2089 					 * point vnode, as required.
2090 					 */
2091 					if (r == 0 &&
2092 					    nfsrv_enable_crossmntpt != 0 &&
2093 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2094 					    nvp->v_type == VDIR &&
2095 					    nvp->v_mountedhere != NULL) {
2096 						new_mp = nvp->v_mountedhere;
2097 						r = vfs_busy(new_mp, 0);
2098 						vput(nvp);
2099 						nvp = NULL;
2100 						if (r == 0) {
2101 							r = VFS_ROOT(new_mp,
2102 							    LK_SHARED, &nvp);
2103 							needs_unbusy = 1;
2104 							if (r == 0)
2105 								at_root = 1;
2106 						}
2107 					}
2108 				}
2109 				if (!r) {
2110 				    if (refp == NULL &&
2111 					((nd->nd_flag & ND_NFSV3) ||
2112 					 NFSNONZERO_ATTRBIT(&attrbits))) {
2113 					r = nfsvno_getfh(nvp, &nfh, p);
2114 					if (!r)
2115 					    r = nfsvno_getattr(nvp, nvap,
2116 						nd->nd_cred, p, 1);
2117 				    }
2118 				} else {
2119 				    nvp = NULL;
2120 				}
2121 				if (r) {
2122 					if (!NFSISSET_ATTRBIT(&attrbits,
2123 					    NFSATTRBIT_RDATTRERROR)) {
2124 						if (nvp != NULL)
2125 							vput(nvp);
2126 						if (needs_unbusy != 0)
2127 							vfs_unbusy(new_mp);
2128 						nd->nd_repstat = r;
2129 						break;
2130 					}
2131 				}
2132 			}
2133 
2134 			/*
2135 			 * Build the directory record xdr
2136 			 */
2137 			if (nd->nd_flag & ND_NFSV3) {
2138 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2139 				*tl++ = newnfs_true;
2140 				*tl++ = 0;
2141 				*tl = txdr_unsigned(dp->d_fileno);
2142 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2143 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2144 				*tl++ = 0;
2145 				*tl = txdr_unsigned(*cookiep);
2146 				nfsrv_postopattr(nd, 0, nvap);
2147 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2148 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2149 				if (nvp != NULL)
2150 					vput(nvp);
2151 			} else {
2152 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2153 				*tl++ = newnfs_true;
2154 				*tl++ = 0;
2155 				*tl = txdr_unsigned(*cookiep);
2156 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2157 				if (nvp != NULL) {
2158 					supports_nfsv4acls =
2159 					    nfs_supportsnfsv4acls(nvp);
2160 					NFSVOPUNLOCK(nvp, 0);
2161 				} else
2162 					supports_nfsv4acls = 0;
2163 				if (refp != NULL) {
2164 					dirlen += nfsrv_putreferralattr(nd,
2165 					    &savbits, refp, 0,
2166 					    &nd->nd_repstat);
2167 					if (nd->nd_repstat) {
2168 						if (nvp != NULL)
2169 							vrele(nvp);
2170 						if (needs_unbusy != 0)
2171 							vfs_unbusy(new_mp);
2172 						break;
2173 					}
2174 				} else if (r) {
2175 					dirlen += nfsvno_fillattr(nd, new_mp,
2176 					    nvp, nvap, &nfh, r, &rderrbits,
2177 					    nd->nd_cred, p, isdgram, 0,
2178 					    supports_nfsv4acls, at_root,
2179 					    mounted_on_fileno);
2180 				} else {
2181 					dirlen += nfsvno_fillattr(nd, new_mp,
2182 					    nvp, nvap, &nfh, r, &attrbits,
2183 					    nd->nd_cred, p, isdgram, 0,
2184 					    supports_nfsv4acls, at_root,
2185 					    mounted_on_fileno);
2186 				}
2187 				if (nvp != NULL)
2188 					vrele(nvp);
2189 				dirlen += (3 * NFSX_UNSIGNED);
2190 			}
2191 			if (needs_unbusy != 0)
2192 				vfs_unbusy(new_mp);
2193 			if (dirlen <= cnt)
2194 				entrycnt++;
2195 		}
2196 		cpos += dp->d_reclen;
2197 		dp = (struct dirent *)cpos;
2198 		cookiep++;
2199 		ncookies--;
2200 	}
2201 	vrele(vp);
2202 	vfs_unbusy(mp);
2203 
2204 	/*
2205 	 * If dirlen > cnt, we must strip off the last entry. If that
2206 	 * results in an empty reply, report NFSERR_TOOSMALL.
2207 	 */
2208 	if (dirlen > cnt || nd->nd_repstat) {
2209 		if (!nd->nd_repstat && entrycnt == 0)
2210 			nd->nd_repstat = NFSERR_TOOSMALL;
2211 		if (nd->nd_repstat)
2212 			newnfs_trimtrailing(nd, mb0, bpos0);
2213 		else
2214 			newnfs_trimtrailing(nd, mb1, bpos1);
2215 		eofflag = 0;
2216 	} else if (cpos < cend)
2217 		eofflag = 0;
2218 	if (!nd->nd_repstat) {
2219 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2220 		*tl++ = newnfs_false;
2221 		if (eofflag)
2222 			*tl = newnfs_true;
2223 		else
2224 			*tl = newnfs_false;
2225 	}
2226 	FREE((caddr_t)cookies, M_TEMP);
2227 	FREE((caddr_t)rbuf, M_TEMP);
2228 
2229 out:
2230 	NFSEXITCODE2(0, nd);
2231 	return (0);
2232 nfsmout:
2233 	vput(vp);
2234 	NFSEXITCODE2(error, nd);
2235 	return (error);
2236 }
2237 
2238 /*
2239  * Get the settable attributes out of the mbuf list.
2240  * (Return 0 or EBADRPC)
2241  */
2242 int
2243 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2244     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2245 {
2246 	u_int32_t *tl;
2247 	struct nfsv2_sattr *sp;
2248 	struct timeval curtime;
2249 	int error = 0, toclient = 0;
2250 
2251 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2252 	case ND_NFSV2:
2253 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2254 		/*
2255 		 * Some old clients didn't fill in the high order 16bits.
2256 		 * --> check the low order 2 bytes for 0xffff
2257 		 */
2258 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2259 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2260 		if (sp->sa_uid != newnfs_xdrneg1)
2261 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2262 		if (sp->sa_gid != newnfs_xdrneg1)
2263 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2264 		if (sp->sa_size != newnfs_xdrneg1)
2265 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2266 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2267 #ifdef notyet
2268 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2269 #else
2270 			nvap->na_atime.tv_sec =
2271 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2272 			nvap->na_atime.tv_nsec = 0;
2273 #endif
2274 		}
2275 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2276 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2277 		break;
2278 	case ND_NFSV3:
2279 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2280 		if (*tl == newnfs_true) {
2281 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2282 			nvap->na_mode = nfstov_mode(*tl);
2283 		}
2284 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2285 		if (*tl == newnfs_true) {
2286 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2287 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2288 		}
2289 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2290 		if (*tl == newnfs_true) {
2291 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2292 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2293 		}
2294 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2295 		if (*tl == newnfs_true) {
2296 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2297 			nvap->na_size = fxdr_hyper(tl);
2298 		}
2299 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2300 		switch (fxdr_unsigned(int, *tl)) {
2301 		case NFSV3SATTRTIME_TOCLIENT:
2302 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2303 			fxdr_nfsv3time(tl, &nvap->na_atime);
2304 			toclient = 1;
2305 			break;
2306 		case NFSV3SATTRTIME_TOSERVER:
2307 			NFSGETTIME(&curtime);
2308 			nvap->na_atime.tv_sec = curtime.tv_sec;
2309 			nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2310 			nvap->na_vaflags |= VA_UTIMES_NULL;
2311 			break;
2312 		};
2313 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2314 		switch (fxdr_unsigned(int, *tl)) {
2315 		case NFSV3SATTRTIME_TOCLIENT:
2316 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2317 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2318 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2319 			break;
2320 		case NFSV3SATTRTIME_TOSERVER:
2321 			NFSGETTIME(&curtime);
2322 			nvap->na_mtime.tv_sec = curtime.tv_sec;
2323 			nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2324 			if (!toclient)
2325 				nvap->na_vaflags |= VA_UTIMES_NULL;
2326 			break;
2327 		};
2328 		break;
2329 	case ND_NFSV4:
2330 		error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2331 	};
2332 nfsmout:
2333 	NFSEXITCODE2(error, nd);
2334 	return (error);
2335 }
2336 
2337 /*
2338  * Handle the setable attributes for V4.
2339  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2340  */
2341 int
2342 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2343     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2344 {
2345 	u_int32_t *tl;
2346 	int attrsum = 0;
2347 	int i, j;
2348 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2349 	int toclient = 0;
2350 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2351 	uid_t uid;
2352 	gid_t gid;
2353 	struct timeval curtime;
2354 
2355 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2356 	if (error)
2357 		goto nfsmout;
2358 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2359 	attrsize = fxdr_unsigned(int, *tl);
2360 
2361 	/*
2362 	 * Loop around getting the setable attributes. If an unsupported
2363 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2364 	 */
2365 	if (retnotsup) {
2366 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2367 		bitpos = NFSATTRBIT_MAX;
2368 	} else {
2369 		bitpos = 0;
2370 	}
2371 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2372 	    if (attrsum > attrsize) {
2373 		error = NFSERR_BADXDR;
2374 		goto nfsmout;
2375 	    }
2376 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2377 		switch (bitpos) {
2378 		case NFSATTRBIT_SIZE:
2379 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2380 			nvap->na_size = fxdr_hyper(tl);
2381 			attrsum += NFSX_HYPER;
2382 			break;
2383 		case NFSATTRBIT_ACL:
2384 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2385 			    p);
2386 			if (error)
2387 				goto nfsmout;
2388 			if (aceerr && !nd->nd_repstat)
2389 				nd->nd_repstat = aceerr;
2390 			attrsum += aclsize;
2391 			break;
2392 		case NFSATTRBIT_ARCHIVE:
2393 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2394 			if (!nd->nd_repstat)
2395 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2396 			attrsum += NFSX_UNSIGNED;
2397 			break;
2398 		case NFSATTRBIT_HIDDEN:
2399 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2400 			if (!nd->nd_repstat)
2401 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2402 			attrsum += NFSX_UNSIGNED;
2403 			break;
2404 		case NFSATTRBIT_MIMETYPE:
2405 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2406 			i = fxdr_unsigned(int, *tl);
2407 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2408 			if (error)
2409 				goto nfsmout;
2410 			if (!nd->nd_repstat)
2411 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2412 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2413 			break;
2414 		case NFSATTRBIT_MODE:
2415 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2416 			nvap->na_mode = nfstov_mode(*tl);
2417 			attrsum += NFSX_UNSIGNED;
2418 			break;
2419 		case NFSATTRBIT_OWNER:
2420 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2421 			j = fxdr_unsigned(int, *tl);
2422 			if (j < 0) {
2423 				error = NFSERR_BADXDR;
2424 				goto nfsmout;
2425 			}
2426 			if (j > NFSV4_SMALLSTR)
2427 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2428 			else
2429 				cp = namestr;
2430 			error = nfsrv_mtostr(nd, cp, j);
2431 			if (error) {
2432 				if (j > NFSV4_SMALLSTR)
2433 					free(cp, M_NFSSTRING);
2434 				goto nfsmout;
2435 			}
2436 			if (!nd->nd_repstat) {
2437 				nd->nd_repstat = nfsv4_strtouid(cp,j,&uid,p);
2438 				if (!nd->nd_repstat)
2439 					nvap->na_uid = uid;
2440 			}
2441 			if (j > NFSV4_SMALLSTR)
2442 				free(cp, M_NFSSTRING);
2443 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2444 			break;
2445 		case NFSATTRBIT_OWNERGROUP:
2446 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2447 			j = fxdr_unsigned(int, *tl);
2448 			if (j < 0) {
2449 				error = NFSERR_BADXDR;
2450 				goto nfsmout;
2451 			}
2452 			if (j > NFSV4_SMALLSTR)
2453 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2454 			else
2455 				cp = namestr;
2456 			error = nfsrv_mtostr(nd, cp, j);
2457 			if (error) {
2458 				if (j > NFSV4_SMALLSTR)
2459 					free(cp, M_NFSSTRING);
2460 				goto nfsmout;
2461 			}
2462 			if (!nd->nd_repstat) {
2463 				nd->nd_repstat = nfsv4_strtogid(cp,j,&gid,p);
2464 				if (!nd->nd_repstat)
2465 					nvap->na_gid = gid;
2466 			}
2467 			if (j > NFSV4_SMALLSTR)
2468 				free(cp, M_NFSSTRING);
2469 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2470 			break;
2471 		case NFSATTRBIT_SYSTEM:
2472 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2473 			if (!nd->nd_repstat)
2474 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2475 			attrsum += NFSX_UNSIGNED;
2476 			break;
2477 		case NFSATTRBIT_TIMEACCESSSET:
2478 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2479 			attrsum += NFSX_UNSIGNED;
2480 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2481 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2482 			    fxdr_nfsv4time(tl, &nvap->na_atime);
2483 			    toclient = 1;
2484 			    attrsum += NFSX_V4TIME;
2485 			} else {
2486 			    NFSGETTIME(&curtime);
2487 			    nvap->na_atime.tv_sec = curtime.tv_sec;
2488 			    nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2489 			    nvap->na_vaflags |= VA_UTIMES_NULL;
2490 			}
2491 			break;
2492 		case NFSATTRBIT_TIMEBACKUP:
2493 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2494 			if (!nd->nd_repstat)
2495 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2496 			attrsum += NFSX_V4TIME;
2497 			break;
2498 		case NFSATTRBIT_TIMECREATE:
2499 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2500 			if (!nd->nd_repstat)
2501 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2502 			attrsum += NFSX_V4TIME;
2503 			break;
2504 		case NFSATTRBIT_TIMEMODIFYSET:
2505 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2506 			attrsum += NFSX_UNSIGNED;
2507 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2508 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2509 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
2510 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
2511 			    attrsum += NFSX_V4TIME;
2512 			} else {
2513 			    NFSGETTIME(&curtime);
2514 			    nvap->na_mtime.tv_sec = curtime.tv_sec;
2515 			    nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2516 			    if (!toclient)
2517 				nvap->na_vaflags |= VA_UTIMES_NULL;
2518 			}
2519 			break;
2520 		default:
2521 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2522 			/*
2523 			 * set bitpos so we drop out of the loop.
2524 			 */
2525 			bitpos = NFSATTRBIT_MAX;
2526 			break;
2527 		};
2528 	}
2529 
2530 	/*
2531 	 * some clients pad the attrlist, so we need to skip over the
2532 	 * padding.
2533 	 */
2534 	if (attrsum > attrsize) {
2535 		error = NFSERR_BADXDR;
2536 	} else {
2537 		attrsize = NFSM_RNDUP(attrsize);
2538 		if (attrsum < attrsize)
2539 			error = nfsm_advance(nd, attrsize - attrsum, -1);
2540 	}
2541 nfsmout:
2542 	NFSEXITCODE2(error, nd);
2543 	return (error);
2544 }
2545 
2546 /*
2547  * Check/setup export credentials.
2548  */
2549 int
2550 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2551     struct ucred *credanon)
2552 {
2553 	int error = 0;
2554 
2555 	/*
2556 	 * Check/setup credentials.
2557 	 */
2558 	if (nd->nd_flag & ND_GSS)
2559 		exp->nes_exflag &= ~MNT_EXPORTANON;
2560 
2561 	/*
2562 	 * Check to see if the operation is allowed for this security flavor.
2563 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2564 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2565 	 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2566 	 */
2567 	if (nfsvno_testexp(nd, exp) &&
2568 	    nd->nd_procnum != NFSV4OP_SECINFO &&
2569 	    nd->nd_procnum != NFSPROC_FSINFO) {
2570 		if (nd->nd_flag & ND_NFSV4)
2571 			error = NFSERR_WRONGSEC;
2572 		else
2573 			error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2574 		goto out;
2575 	}
2576 
2577 	/*
2578 	 * Check to see if the file system is exported V4 only.
2579 	 */
2580 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2581 		error = NFSERR_PROGNOTV4;
2582 		goto out;
2583 	}
2584 
2585 	/*
2586 	 * Now, map the user credentials.
2587 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
2588 	 *  Fsinfo RPC. If set for anything else, this code might need
2589 	 *  to change.)
2590 	 */
2591 	if (NFSVNO_EXPORTED(exp) &&
2592 	    ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2593 	     NFSVNO_EXPORTANON(exp) ||
2594 	     (nd->nd_flag & ND_AUTHNONE))) {
2595 		nd->nd_cred->cr_uid = credanon->cr_uid;
2596 		nd->nd_cred->cr_gid = credanon->cr_gid;
2597 		crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2598 		    credanon->cr_groups);
2599 	}
2600 
2601 out:
2602 	NFSEXITCODE2(error, nd);
2603 	return (error);
2604 }
2605 
2606 /*
2607  * Check exports.
2608  */
2609 int
2610 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2611     struct ucred **credp)
2612 {
2613 	int i, error, *secflavors;
2614 
2615 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2616 	    &exp->nes_numsecflavor, &secflavors);
2617 	if (error) {
2618 		if (nfs_rootfhset) {
2619 			exp->nes_exflag = 0;
2620 			exp->nes_numsecflavor = 0;
2621 			error = 0;
2622 		}
2623 	} else {
2624 		/* Copy the security flavors. */
2625 		for (i = 0; i < exp->nes_numsecflavor; i++)
2626 			exp->nes_secflavors[i] = secflavors[i];
2627 	}
2628 	NFSEXITCODE(error);
2629 	return (error);
2630 }
2631 
2632 /*
2633  * Get a vnode for a file handle and export stuff.
2634  */
2635 int
2636 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2637     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2638     struct ucred **credp)
2639 {
2640 	int i, error, *secflavors;
2641 
2642 	*credp = NULL;
2643 	exp->nes_numsecflavor = 0;
2644 	if (VFS_NEEDSGIANT(mp))
2645 		error = ESTALE;
2646 	else
2647 		error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2648 	if (error != 0)
2649 		/* Make sure the server replies ESTALE to the client. */
2650 		error = ESTALE;
2651 	if (nam && !error) {
2652 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2653 		    &exp->nes_numsecflavor, &secflavors);
2654 		if (error) {
2655 			if (nfs_rootfhset) {
2656 				exp->nes_exflag = 0;
2657 				exp->nes_numsecflavor = 0;
2658 				error = 0;
2659 			} else {
2660 				vput(*vpp);
2661 			}
2662 		} else {
2663 			/* Copy the security flavors. */
2664 			for (i = 0; i < exp->nes_numsecflavor; i++)
2665 				exp->nes_secflavors[i] = secflavors[i];
2666 		}
2667 	}
2668 	NFSEXITCODE(error);
2669 	return (error);
2670 }
2671 
2672 /*
2673  * nfsd_fhtovp() - convert a fh to a vnode ptr
2674  * 	- look up fsid in mount list (if not found ret error)
2675  *	- get vp and export rights by calling nfsvno_fhtovp()
2676  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2677  *	  for AUTH_SYS
2678  *	- if mpp != NULL, return the mount point so that it can
2679  *	  be used for vn_finished_write() by the caller
2680  */
2681 void
2682 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2683     struct vnode **vpp, struct nfsexstuff *exp,
2684     struct mount **mpp, int startwrite, struct thread *p)
2685 {
2686 	struct mount *mp;
2687 	struct ucred *credanon;
2688 	fhandle_t *fhp;
2689 
2690 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
2691 	/*
2692 	 * Check for the special case of the nfsv4root_fh.
2693 	 */
2694 	mp = vfs_busyfs(&fhp->fh_fsid);
2695 	if (mpp != NULL)
2696 		*mpp = mp;
2697 	if (mp == NULL) {
2698 		*vpp = NULL;
2699 		nd->nd_repstat = ESTALE;
2700 		goto out;
2701 	}
2702 
2703 	if (startwrite)
2704 		vn_start_write(NULL, mpp, V_WAIT);
2705 
2706 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2707 	    &credanon);
2708 	vfs_unbusy(mp);
2709 
2710 	/*
2711 	 * For NFSv4 without a pseudo root fs, unexported file handles
2712 	 * can be returned, so that Lookup works everywhere.
2713 	 */
2714 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2715 	    !(nd->nd_flag & ND_NFSV4)) {
2716 		vput(*vpp);
2717 		nd->nd_repstat = EACCES;
2718 	}
2719 
2720 	/*
2721 	 * Personally, I've never seen any point in requiring a
2722 	 * reserved port#, since only in the rare case where the
2723 	 * clients are all boxes with secure system priviledges,
2724 	 * does it provide any enhanced security, but... some people
2725 	 * believe it to be useful and keep putting this code back in.
2726 	 * (There is also some "security checker" out there that
2727 	 *  complains if the nfs server doesn't enforce this.)
2728 	 * However, note the following:
2729 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
2730 	 *	required.
2731 	 * RFC2623 recommends that, if a reserved port# is checked for,
2732 	 *	that there be a way to turn that off--> ifdef'd.
2733 	 */
2734 #ifdef NFS_REQRSVPORT
2735 	if (!nd->nd_repstat) {
2736 		struct sockaddr_in *saddr;
2737 		struct sockaddr_in6 *saddr6;
2738 
2739 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2740 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2741 		if (!(nd->nd_flag & ND_NFSV4) &&
2742 		    ((saddr->sin_family == AF_INET &&
2743 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2744 		     (saddr6->sin6_family == AF_INET6 &&
2745 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2746 			vput(*vpp);
2747 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2748 		}
2749 	}
2750 #endif	/* NFS_REQRSVPORT */
2751 
2752 	/*
2753 	 * Check/setup credentials.
2754 	 */
2755 	if (!nd->nd_repstat) {
2756 		nd->nd_saveduid = nd->nd_cred->cr_uid;
2757 		nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2758 		if (nd->nd_repstat)
2759 			vput(*vpp);
2760 	}
2761 	if (credanon != NULL)
2762 		crfree(credanon);
2763 	if (nd->nd_repstat) {
2764 		if (startwrite)
2765 			vn_finished_write(mp);
2766 		*vpp = NULL;
2767 		if (mpp != NULL)
2768 			*mpp = NULL;
2769 	}
2770 
2771 out:
2772 	NFSEXITCODE2(0, nd);
2773 }
2774 
2775 /*
2776  * glue for fp.
2777  */
2778 int
2779 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2780 {
2781 	struct filedesc *fdp;
2782 	struct file *fp;
2783 	int error = 0;
2784 
2785 	fdp = p->td_proc->p_fd;
2786 	if (fd >= fdp->fd_nfiles ||
2787 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
2788 		error = EBADF;
2789 		goto out;
2790 	}
2791 	*fpp = fp;
2792 
2793 out:
2794 	NFSEXITCODE(error);
2795 	return (error);
2796 }
2797 
2798 /*
2799  * Called from nfssvc() to update the exports list. Just call
2800  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2801  * in the mount list.
2802  */
2803 int
2804 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2805 {
2806 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2807 	int error = 0;
2808 	struct nameidata nd;
2809 	fhandle_t fh;
2810 
2811 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2812 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2813 		nfs_rootfhset = 0;
2814 	else if (error == 0) {
2815 		if (nfsexargp->fspec == NULL) {
2816 			error = EPERM;
2817 			goto out;
2818 		}
2819 		/*
2820 		 * If fspec != NULL, this is the v4root path.
2821 		 */
2822 		NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_USERSPACE,
2823 		    nfsexargp->fspec, p);
2824 		if ((error = namei(&nd)) != 0)
2825 			goto out;
2826 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
2827 		vrele(nd.ni_vp);
2828 		if (!error) {
2829 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2830 			NFSBCOPY((caddr_t)&fh,
2831 			    nfs_rootfh.nfsrvfh_data,
2832 			    sizeof (fhandle_t));
2833 			nfs_rootfhset = 1;
2834 		}
2835 	}
2836 
2837 out:
2838 	NFSEXITCODE(error);
2839 	return (error);
2840 }
2841 
2842 /*
2843  * Get the tcp socket sequence numbers we need.
2844  * (Maybe this should be moved to the tcp sources?)
2845  */
2846 int
2847 nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap)
2848 {
2849 	struct inpcb *inp;
2850 	struct tcpcb *tp;
2851 	int error = 0;
2852 
2853 	inp = sotoinpcb(so);
2854 	KASSERT(inp != NULL, ("nfsrv_getsocksndseq: inp == NULL"));
2855 	INP_RLOCK(inp);
2856 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2857 		INP_RUNLOCK(inp);
2858 		error = EPIPE;
2859 		goto out;
2860 	}
2861 	tp = intotcpcb(inp);
2862 	if (tp->t_state != TCPS_ESTABLISHED) {
2863 		INP_RUNLOCK(inp);
2864 		error = EPIPE;
2865 		goto out;
2866 	}
2867 	*maxp = tp->snd_max;
2868 	*unap = tp->snd_una;
2869 	INP_RUNLOCK(inp);
2870 
2871 out:
2872 	NFSEXITCODE(error);
2873 	return (error);
2874 }
2875 
2876 /*
2877  * This function needs to test to see if the system is near its limit
2878  * for memory allocation via malloc() or mget() and return True iff
2879  * either of these resources are near their limit.
2880  * XXX (For now, this is just a stub.)
2881  */
2882 int nfsrv_testmalloclimit = 0;
2883 int
2884 nfsrv_mallocmget_limit(void)
2885 {
2886 	static int printmesg = 0;
2887 	static int testval = 1;
2888 
2889 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2890 		if ((printmesg++ % 100) == 0)
2891 			printf("nfsd: malloc/mget near limit\n");
2892 		return (1);
2893 	}
2894 	return (0);
2895 }
2896 
2897 /*
2898  * BSD specific initialization of a mount point.
2899  */
2900 void
2901 nfsd_mntinit(void)
2902 {
2903 	static int inited = 0;
2904 
2905 	if (inited)
2906 		return;
2907 	inited = 1;
2908 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2909 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2910 	nfsv4root_mnt.mnt_export = NULL;
2911 	TAILQ_INIT(&nfsv4root_opt);
2912 	TAILQ_INIT(&nfsv4root_newopt);
2913 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2914 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2915 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
2916 }
2917 
2918 /*
2919  * Get a vnode for a file handle, without checking exports, etc.
2920  */
2921 struct vnode *
2922 nfsvno_getvp(fhandle_t *fhp)
2923 {
2924 	struct mount *mp;
2925 	struct vnode *vp;
2926 	int error;
2927 
2928 	mp = vfs_busyfs(&fhp->fh_fsid);
2929 	if (mp == NULL)
2930 		return (NULL);
2931 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2932 	vfs_unbusy(mp);
2933 	if (error)
2934 		return (NULL);
2935 	return (vp);
2936 }
2937 
2938 /*
2939  * Do a local VOP_ADVLOCK().
2940  */
2941 int
2942 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2943     u_int64_t end, struct thread *td)
2944 {
2945 	int error = 0;
2946 	struct flock fl;
2947 	u_int64_t tlen;
2948 
2949 	if (nfsrv_dolocallocks == 0)
2950 		goto out;
2951 
2952 	/* Check for VI_DOOMED here, so that VOP_ADVLOCK() isn't performed. */
2953 	if ((vp->v_iflag & VI_DOOMED) != 0) {
2954 		error = EPERM;
2955 		goto out;
2956 	}
2957 
2958 	fl.l_whence = SEEK_SET;
2959 	fl.l_type = ftype;
2960 	fl.l_start = (off_t)first;
2961 	if (end == NFS64BITSSET) {
2962 		fl.l_len = 0;
2963 	} else {
2964 		tlen = end - first;
2965 		fl.l_len = (off_t)tlen;
2966 	}
2967 	/*
2968 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2969 	 * values for all calls, so that all locks will be held by the
2970 	 * nfsd server. (The nfsd server handles conflicts between the
2971 	 * various clients.)
2972 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2973 	 * bytes, so it can't be put in l_sysid.
2974 	 */
2975 	if (nfsv4_sysid == 0)
2976 		nfsv4_sysid = nlm_acquire_next_sysid();
2977 	fl.l_pid = (pid_t)0;
2978 	fl.l_sysid = (int)nfsv4_sysid;
2979 
2980 	NFSVOPUNLOCK(vp, 0);
2981 	if (ftype == F_UNLCK)
2982 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
2983 		    (F_POSIX | F_REMOTE));
2984 	else
2985 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
2986 		    (F_POSIX | F_REMOTE));
2987 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2988 
2989 out:
2990 	NFSEXITCODE(error);
2991 	return (error);
2992 }
2993 
2994 /*
2995  * Check the nfsv4 root exports.
2996  */
2997 int
2998 nfsvno_v4rootexport(struct nfsrv_descript *nd)
2999 {
3000 	struct ucred *credanon;
3001 	int exflags, error = 0, numsecflavor, *secflavors, i;
3002 
3003 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3004 	    &credanon, &numsecflavor, &secflavors);
3005 	if (error) {
3006 		error = NFSERR_PROGUNAVAIL;
3007 		goto out;
3008 	}
3009 	if (credanon != NULL)
3010 		crfree(credanon);
3011 	for (i = 0; i < numsecflavor; i++) {
3012 		if (secflavors[i] == AUTH_SYS)
3013 			nd->nd_flag |= ND_EXAUTHSYS;
3014 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3015 			nd->nd_flag |= ND_EXGSS;
3016 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3017 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3018 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3019 			nd->nd_flag |= ND_EXGSSPRIVACY;
3020 	}
3021 
3022 out:
3023 	NFSEXITCODE(error);
3024 	return (error);
3025 }
3026 
3027 /*
3028  * Nfs server psuedo system call for the nfsd's
3029  */
3030 /*
3031  * MPSAFE
3032  */
3033 static int
3034 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3035 {
3036 	struct file *fp;
3037 	struct nfsd_addsock_args sockarg;
3038 	struct nfsd_nfsd_args nfsdarg;
3039 	int error;
3040 
3041 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3042 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3043 		if (error)
3044 			goto out;
3045 		/*
3046 		 * Since we don't know what rights might be required,
3047 		 * pretend that we need them all. It is better to be too
3048 		 * careful than too reckless.
3049 		 */
3050 		if ((error = fget(td, sockarg.sock, CAP_SOCK_ALL, &fp)) != 0)
3051 			goto out;
3052 		if (fp->f_type != DTYPE_SOCKET) {
3053 			fdrop(fp, td);
3054 			error = EPERM;
3055 			goto out;
3056 		}
3057 		error = nfsrvd_addsock(fp);
3058 		fdrop(fp, td);
3059 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3060 		if (uap->argp == NULL) {
3061 			error = EINVAL;
3062 			goto out;
3063 		}
3064 		error = copyin(uap->argp, (caddr_t)&nfsdarg,
3065 		    sizeof (nfsdarg));
3066 		if (error)
3067 			goto out;
3068 		error = nfsrvd_nfsd(td, &nfsdarg);
3069 	} else {
3070 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3071 	}
3072 
3073 out:
3074 	NFSEXITCODE(error);
3075 	return (error);
3076 }
3077 
3078 static int
3079 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3080 {
3081 	struct nfsex_args export;
3082 	struct file *fp = NULL;
3083 	int stablefd, len;
3084 	struct nfsd_clid adminrevoke;
3085 	struct nfsd_dumplist dumplist;
3086 	struct nfsd_dumpclients *dumpclients;
3087 	struct nfsd_dumplocklist dumplocklist;
3088 	struct nfsd_dumplocks *dumplocks;
3089 	struct nameidata nd;
3090 	vnode_t vp;
3091 	int error = EINVAL;
3092 	struct proc *procp;
3093 
3094 	if (uap->flag & NFSSVC_PUBLICFH) {
3095 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3096 		    sizeof (fhandle_t));
3097 		error = copyin(uap->argp,
3098 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3099 		if (!error)
3100 			nfs_pubfhset = 1;
3101 	} else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3102 		error = copyin(uap->argp,(caddr_t)&export,
3103 		    sizeof (struct nfsex_args));
3104 		if (!error)
3105 			error = nfsrv_v4rootexport(&export, cred, p);
3106 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3107 		nfs_pubfhset = 0;
3108 		error = 0;
3109 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3110 		error = copyin(uap->argp, (caddr_t)&stablefd,
3111 		    sizeof (int));
3112 		if (!error)
3113 			error = fp_getfvp(p, stablefd, &fp, &vp);
3114 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3115 			error = EBADF;
3116 		if (!error && newnfs_numnfsd != 0)
3117 			error = EPERM;
3118 		if (!error) {
3119 			nfsrv_stablefirst.nsf_fp = fp;
3120 			nfsrv_setupstable(p);
3121 		}
3122 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3123 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3124 		    sizeof (struct nfsd_clid));
3125 		if (!error)
3126 			error = nfsrv_adminrevoke(&adminrevoke, p);
3127 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3128 		error = copyin(uap->argp, (caddr_t)&dumplist,
3129 		    sizeof (struct nfsd_dumplist));
3130 		if (!error && (dumplist.ndl_size < 1 ||
3131 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3132 			error = EPERM;
3133 		if (!error) {
3134 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3135 		    dumpclients = (struct nfsd_dumpclients *)malloc(len,
3136 			M_TEMP, M_WAITOK);
3137 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3138 		    error = copyout(dumpclients,
3139 			CAST_USER_ADDR_T(dumplist.ndl_list), len);
3140 		    free((caddr_t)dumpclients, M_TEMP);
3141 		}
3142 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3143 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3144 		    sizeof (struct nfsd_dumplocklist));
3145 		if (!error && (dumplocklist.ndllck_size < 1 ||
3146 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3147 			error = EPERM;
3148 		if (!error)
3149 			error = nfsrv_lookupfilename(&nd,
3150 				dumplocklist.ndllck_fname, p);
3151 		if (!error) {
3152 			len = sizeof (struct nfsd_dumplocks) *
3153 				dumplocklist.ndllck_size;
3154 			dumplocks = (struct nfsd_dumplocks *)malloc(len,
3155 				M_TEMP, M_WAITOK);
3156 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3157 			    dumplocklist.ndllck_size, p);
3158 			vput(nd.ni_vp);
3159 			error = copyout(dumplocks,
3160 			    CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3161 			free((caddr_t)dumplocks, M_TEMP);
3162 		}
3163 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3164 		procp = p->td_proc;
3165 		PROC_LOCK(procp);
3166 		nfsd_master_pid = procp->p_pid;
3167 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3168 		nfsd_master_start = procp->p_stats->p_start;
3169 		nfsd_master_proc = procp;
3170 		PROC_UNLOCK(procp);
3171 	}
3172 
3173 	NFSEXITCODE(error);
3174 	return (error);
3175 }
3176 
3177 /*
3178  * Check exports.
3179  * Returns 0 if ok, 1 otherwise.
3180  */
3181 int
3182 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3183 {
3184 	int i;
3185 
3186 	/*
3187 	 * This seems odd, but allow the case where the security flavor
3188 	 * list is empty. This happens when NFSv4 is traversing non-exported
3189 	 * file systems. Exported file systems should always have a non-empty
3190 	 * security flavor list.
3191 	 */
3192 	if (exp->nes_numsecflavor == 0)
3193 		return (0);
3194 
3195 	for (i = 0; i < exp->nes_numsecflavor; i++) {
3196 		/*
3197 		 * The tests for privacy and integrity must be first,
3198 		 * since ND_GSS is set for everything but AUTH_SYS.
3199 		 */
3200 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3201 		    (nd->nd_flag & ND_GSSPRIVACY))
3202 			return (0);
3203 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3204 		    (nd->nd_flag & ND_GSSINTEGRITY))
3205 			return (0);
3206 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3207 		    (nd->nd_flag & ND_GSS))
3208 			return (0);
3209 		if (exp->nes_secflavors[i] == AUTH_SYS &&
3210 		    (nd->nd_flag & ND_GSS) == 0)
3211 			return (0);
3212 	}
3213 	return (1);
3214 }
3215 
3216 /*
3217  * Calculate a hash value for the fid in a file handle.
3218  */
3219 uint32_t
3220 nfsrv_hashfh(fhandle_t *fhp)
3221 {
3222 	uint32_t hashval;
3223 
3224 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3225 	return (hashval);
3226 }
3227 
3228 /*
3229  * Signal the userland master nfsd to backup the stable restart file.
3230  */
3231 void
3232 nfsrv_backupstable(void)
3233 {
3234 	struct proc *procp;
3235 
3236 	if (nfsd_master_proc != NULL) {
3237 		procp = pfind(nfsd_master_pid);
3238 		/* Try to make sure it is the correct process. */
3239 		if (procp == nfsd_master_proc &&
3240 		    procp->p_stats->p_start.tv_sec ==
3241 		    nfsd_master_start.tv_sec &&
3242 		    procp->p_stats->p_start.tv_usec ==
3243 		    nfsd_master_start.tv_usec &&
3244 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
3245 			kern_psignal(procp, SIGUSR2);
3246 		else
3247 			nfsd_master_proc = NULL;
3248 
3249 		if (procp != NULL)
3250 			PROC_UNLOCK(procp);
3251 	}
3252 }
3253 
3254 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3255 
3256 /*
3257  * Called once to initialize data structures...
3258  */
3259 static int
3260 nfsd_modevent(module_t mod, int type, void *data)
3261 {
3262 	int error = 0;
3263 	static int loaded = 0;
3264 
3265 	switch (type) {
3266 	case MOD_LOAD:
3267 		if (loaded)
3268 			goto out;
3269 		newnfs_portinit();
3270 		mtx_init(&nfs_cache_mutex, "nfs_cache_mutex", NULL, MTX_DEF);
3271 		mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3272 		mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3273 		    MTX_DEF);
3274 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3275 		nfsrvd_initcache();
3276 		nfsd_init();
3277 		NFSD_LOCK();
3278 		nfsrvd_init(0);
3279 		NFSD_UNLOCK();
3280 		nfsd_mntinit();
3281 #ifdef VV_DISABLEDELEG
3282 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3283 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3284 #endif
3285 		nfsd_call_servertimer = nfsrv_servertimer;
3286 		nfsd_call_nfsd = nfssvc_nfsd;
3287 		loaded = 1;
3288 		break;
3289 
3290 	case MOD_UNLOAD:
3291 		if (newnfs_numnfsd != 0) {
3292 			error = EBUSY;
3293 			break;
3294 		}
3295 
3296 #ifdef VV_DISABLEDELEG
3297 		vn_deleg_ops.vndeleg_recall = NULL;
3298 		vn_deleg_ops.vndeleg_disable = NULL;
3299 #endif
3300 		nfsd_call_servertimer = NULL;
3301 		nfsd_call_nfsd = NULL;
3302 
3303 		/* Clean out all NFSv4 state. */
3304 		nfsrv_throwawayallstate(curthread);
3305 
3306 		/* Clean the NFS server reply cache */
3307 		nfsrvd_cleancache();
3308 
3309 		/* Free up the krpc server pool. */
3310 		if (nfsrvd_pool != NULL)
3311 			svcpool_destroy(nfsrvd_pool);
3312 
3313 		/* and get rid of the locks */
3314 		mtx_destroy(&nfs_cache_mutex);
3315 		mtx_destroy(&nfs_v4root_mutex);
3316 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3317 		lockdestroy(&nfsv4root_mnt.mnt_explock);
3318 		loaded = 0;
3319 		break;
3320 	default:
3321 		error = EOPNOTSUPP;
3322 		break;
3323 	}
3324 
3325 out:
3326 	NFSEXITCODE(error);
3327 	return (error);
3328 }
3329 static moduledata_t nfsd_mod = {
3330 	"nfsd",
3331 	nfsd_modevent,
3332 	NULL,
3333 };
3334 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3335 
3336 /* So that loader and kldload(2) can find us, wherever we are.. */
3337 MODULE_VERSION(nfsd, 1);
3338 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3339 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3340 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3341 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3342 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3343 
3344