xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision 97cb52fa9aefd90fad38790fded50905aeeb9b9e)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/capsicum.h>
40 
41 /*
42  * Functions that perform the vfs operations required by the routines in
43  * nfsd_serv.c. It is hoped that this change will make the server more
44  * portable.
45  */
46 
47 #include <fs/nfs/nfsport.h>
48 #include <sys/hash.h>
49 #include <sys/sysctl.h>
50 #include <nlm/nlm_prot.h>
51 #include <nlm/nlm.h>
52 
53 FEATURE(nfsd, "NFSv4 server");
54 
55 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
56 extern int nfsrv_useacl;
57 extern int newnfs_numnfsd;
58 extern struct mount nfsv4root_mnt;
59 extern struct nfsrv_stablefirst nfsrv_stablefirst;
60 extern void (*nfsd_call_servertimer)(void);
61 extern SVCPOOL	*nfsrvd_pool;
62 extern struct nfsv4lock nfsd_suspend_lock;
63 extern struct nfsclienthashhead *nfsclienthash;
64 extern struct nfslockhashhead *nfslockhash;
65 extern struct nfssessionhash *nfssessionhash;
66 extern int nfsrv_sessionhashsize;
67 extern struct nfsstatsv1 nfsstatsv1;
68 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
69 NFSDLOCKMUTEX;
70 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
71 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
72 struct mtx nfsrc_udpmtx;
73 struct mtx nfs_v4root_mutex;
74 struct nfsrvfh nfs_rootfh, nfs_pubfh;
75 int nfs_pubfhset = 0, nfs_rootfhset = 0;
76 struct proc *nfsd_master_proc = NULL;
77 int nfsd_debuglevel = 0;
78 static pid_t nfsd_master_pid = (pid_t)-1;
79 static char nfsd_master_comm[MAXCOMLEN + 1];
80 static struct timeval nfsd_master_start;
81 static uint32_t nfsv4_sysid = 0;
82 
83 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
84     struct ucred *);
85 
86 int nfsrv_enable_crossmntpt = 1;
87 static int nfs_commit_blks;
88 static int nfs_commit_miss;
89 extern int nfsrv_issuedelegs;
90 extern int nfsrv_dolocallocks;
91 extern int nfsd_enable_stringtouid;
92 
93 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "NFS server");
94 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
95     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
96 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
97     0, "");
98 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
99     0, "");
100 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
101     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
102 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
103     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
104 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
105     0, "Debug level for NFS server");
106 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
107     &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
108 
109 #define	MAX_REORDERED_RPC	16
110 #define	NUM_HEURISTIC		1031
111 #define	NHUSE_INIT		64
112 #define	NHUSE_INC		16
113 #define	NHUSE_MAX		2048
114 
115 static struct nfsheur {
116 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
117 	off_t nh_nextoff;	/* next offset for sequential detection */
118 	int nh_use;		/* use count for selection */
119 	int nh_seqcount;	/* heuristic */
120 } nfsheur[NUM_HEURISTIC];
121 
122 
123 /*
124  * Heuristic to detect sequential operation.
125  */
126 static struct nfsheur *
127 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
128 {
129 	struct nfsheur *nh;
130 	int hi, try;
131 
132 	/* Locate best candidate. */
133 	try = 32;
134 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
135 	nh = &nfsheur[hi];
136 	while (try--) {
137 		if (nfsheur[hi].nh_vp == vp) {
138 			nh = &nfsheur[hi];
139 			break;
140 		}
141 		if (nfsheur[hi].nh_use > 0)
142 			--nfsheur[hi].nh_use;
143 		hi = (hi + 1) % NUM_HEURISTIC;
144 		if (nfsheur[hi].nh_use < nh->nh_use)
145 			nh = &nfsheur[hi];
146 	}
147 
148 	/* Initialize hint if this is a new file. */
149 	if (nh->nh_vp != vp) {
150 		nh->nh_vp = vp;
151 		nh->nh_nextoff = uio->uio_offset;
152 		nh->nh_use = NHUSE_INIT;
153 		if (uio->uio_offset == 0)
154 			nh->nh_seqcount = 4;
155 		else
156 			nh->nh_seqcount = 1;
157 	}
158 
159 	/* Calculate heuristic. */
160 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
161 	    uio->uio_offset == nh->nh_nextoff) {
162 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
163 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
164 		if (nh->nh_seqcount > IO_SEQMAX)
165 			nh->nh_seqcount = IO_SEQMAX;
166 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
167 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
168 		/* Probably a reordered RPC, leave seqcount alone. */
169 	} else if (nh->nh_seqcount > 1) {
170 		nh->nh_seqcount /= 2;
171 	} else {
172 		nh->nh_seqcount = 0;
173 	}
174 	nh->nh_use += NHUSE_INC;
175 	if (nh->nh_use > NHUSE_MAX)
176 		nh->nh_use = NHUSE_MAX;
177 	return (nh);
178 }
179 
180 /*
181  * Get attributes into nfsvattr structure.
182  */
183 int
184 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
185     struct thread *p, int vpislocked)
186 {
187 	int error, lockedit = 0;
188 
189 	if (vpislocked == 0) {
190 		/*
191 		 * When vpislocked == 0, the vnode is either exclusively
192 		 * locked by this thread or not locked by this thread.
193 		 * As such, shared lock it, if not exclusively locked.
194 		 */
195 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
196 			lockedit = 1;
197 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
198 		}
199 	}
200 	error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
201 	if (lockedit != 0)
202 		NFSVOPUNLOCK(vp, 0);
203 
204 	NFSEXITCODE(error);
205 	return (error);
206 }
207 
208 /*
209  * Get a file handle for a vnode.
210  */
211 int
212 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
213 {
214 	int error;
215 
216 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
217 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
218 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
219 
220 	NFSEXITCODE(error);
221 	return (error);
222 }
223 
224 /*
225  * Perform access checking for vnodes obtained from file handles that would
226  * refer to files already opened by a Unix client. You cannot just use
227  * vn_writechk() and VOP_ACCESSX() for two reasons.
228  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
229  *     case.
230  * 2 - The owner is to be given access irrespective of mode bits for some
231  *     operations, so that processes that chmod after opening a file don't
232  *     break.
233  */
234 int
235 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
236     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
237     u_int32_t *supportedtypep)
238 {
239 	struct vattr vattr;
240 	int error = 0, getret = 0;
241 
242 	if (vpislocked == 0) {
243 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
244 			error = EPERM;
245 			goto out;
246 		}
247 	}
248 	if (accmode & VWRITE) {
249 		/* Just vn_writechk() changed to check rdonly */
250 		/*
251 		 * Disallow write attempts on read-only file systems;
252 		 * unless the file is a socket or a block or character
253 		 * device resident on the file system.
254 		 */
255 		if (NFSVNO_EXRDONLY(exp) ||
256 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
257 			switch (vp->v_type) {
258 			case VREG:
259 			case VDIR:
260 			case VLNK:
261 				error = EROFS;
262 			default:
263 				break;
264 			}
265 		}
266 		/*
267 		 * If there's shared text associated with
268 		 * the inode, try to free it up once.  If
269 		 * we fail, we can't allow writing.
270 		 */
271 		if (VOP_IS_TEXT(vp) && error == 0)
272 			error = ETXTBSY;
273 	}
274 	if (error != 0) {
275 		if (vpislocked == 0)
276 			NFSVOPUNLOCK(vp, 0);
277 		goto out;
278 	}
279 
280 	/*
281 	 * Should the override still be applied when ACLs are enabled?
282 	 */
283 	error = VOP_ACCESSX(vp, accmode, cred, p);
284 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
285 		/*
286 		 * Try again with VEXPLICIT_DENY, to see if the test for
287 		 * deletion is supported.
288 		 */
289 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
290 		if (error == 0) {
291 			if (vp->v_type == VDIR) {
292 				accmode &= ~(VDELETE | VDELETE_CHILD);
293 				accmode |= VWRITE;
294 				error = VOP_ACCESSX(vp, accmode, cred, p);
295 			} else if (supportedtypep != NULL) {
296 				*supportedtypep &= ~NFSACCESS_DELETE;
297 			}
298 		}
299 	}
300 
301 	/*
302 	 * Allow certain operations for the owner (reads and writes
303 	 * on files that are already open).
304 	 */
305 	if (override != NFSACCCHK_NOOVERRIDE &&
306 	    (error == EPERM || error == EACCES)) {
307 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
308 			error = 0;
309 		else if (override & NFSACCCHK_ALLOWOWNER) {
310 			getret = VOP_GETATTR(vp, &vattr, cred);
311 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
312 				error = 0;
313 		}
314 	}
315 	if (vpislocked == 0)
316 		NFSVOPUNLOCK(vp, 0);
317 
318 out:
319 	NFSEXITCODE(error);
320 	return (error);
321 }
322 
323 /*
324  * Set attribute(s) vnop.
325  */
326 int
327 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
328     struct thread *p, struct nfsexstuff *exp)
329 {
330 	int error;
331 
332 	error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
333 	NFSEXITCODE(error);
334 	return (error);
335 }
336 
337 /*
338  * Set up nameidata for a lookup() call and do it.
339  */
340 int
341 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
342     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
343     struct vnode **retdirp)
344 {
345 	struct componentname *cnp = &ndp->ni_cnd;
346 	int i;
347 	struct iovec aiov;
348 	struct uio auio;
349 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
350 	int error = 0, crossmnt;
351 	char *cp;
352 
353 	*retdirp = NULL;
354 	cnp->cn_nameptr = cnp->cn_pnbuf;
355 	ndp->ni_lcf = 0;
356 	/*
357 	 * Extract and set starting directory.
358 	 */
359 	if (dp->v_type != VDIR) {
360 		if (islocked)
361 			vput(dp);
362 		else
363 			vrele(dp);
364 		nfsvno_relpathbuf(ndp);
365 		error = ENOTDIR;
366 		goto out1;
367 	}
368 	if (islocked)
369 		NFSVOPUNLOCK(dp, 0);
370 	VREF(dp);
371 	*retdirp = dp;
372 	if (NFSVNO_EXRDONLY(exp))
373 		cnp->cn_flags |= RDONLY;
374 	ndp->ni_segflg = UIO_SYSSPACE;
375 	crossmnt = 1;
376 
377 	if (nd->nd_flag & ND_PUBLOOKUP) {
378 		ndp->ni_loopcnt = 0;
379 		if (cnp->cn_pnbuf[0] == '/') {
380 			vrele(dp);
381 			/*
382 			 * Check for degenerate pathnames here, since lookup()
383 			 * panics on them.
384 			 */
385 			for (i = 1; i < ndp->ni_pathlen; i++)
386 				if (cnp->cn_pnbuf[i] != '/')
387 					break;
388 			if (i == ndp->ni_pathlen) {
389 				error = NFSERR_ACCES;
390 				goto out;
391 			}
392 			dp = rootvnode;
393 			VREF(dp);
394 		}
395 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
396 	    (nd->nd_flag & ND_NFSV4) == 0) {
397 		/*
398 		 * Only cross mount points for NFSv4 when doing a
399 		 * mount while traversing the file system above
400 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
401 		 */
402 		cnp->cn_flags |= NOCROSSMOUNT;
403 		crossmnt = 0;
404 	}
405 
406 	/*
407 	 * Initialize for scan, set ni_startdir and bump ref on dp again
408 	 * because lookup() will dereference ni_startdir.
409 	 */
410 
411 	cnp->cn_thread = p;
412 	ndp->ni_startdir = dp;
413 	ndp->ni_rootdir = rootvnode;
414 	ndp->ni_topdir = NULL;
415 
416 	if (!lockleaf)
417 		cnp->cn_flags |= LOCKLEAF;
418 	for (;;) {
419 		cnp->cn_nameptr = cnp->cn_pnbuf;
420 		/*
421 		 * Call lookup() to do the real work.  If an error occurs,
422 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
423 		 * we do not have to dereference anything before returning.
424 		 * In either case ni_startdir will be dereferenced and NULLed
425 		 * out.
426 		 */
427 		error = lookup(ndp);
428 		if (error)
429 			break;
430 
431 		/*
432 		 * Check for encountering a symbolic link.  Trivial
433 		 * termination occurs if no symlink encountered.
434 		 */
435 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
436 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
437 				nfsvno_relpathbuf(ndp);
438 			if (ndp->ni_vp && !lockleaf)
439 				NFSVOPUNLOCK(ndp->ni_vp, 0);
440 			break;
441 		}
442 
443 		/*
444 		 * Validate symlink
445 		 */
446 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
447 			NFSVOPUNLOCK(ndp->ni_dvp, 0);
448 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
449 			error = EINVAL;
450 			goto badlink2;
451 		}
452 
453 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
454 			error = ELOOP;
455 			goto badlink2;
456 		}
457 		if (ndp->ni_pathlen > 1)
458 			cp = uma_zalloc(namei_zone, M_WAITOK);
459 		else
460 			cp = cnp->cn_pnbuf;
461 		aiov.iov_base = cp;
462 		aiov.iov_len = MAXPATHLEN;
463 		auio.uio_iov = &aiov;
464 		auio.uio_iovcnt = 1;
465 		auio.uio_offset = 0;
466 		auio.uio_rw = UIO_READ;
467 		auio.uio_segflg = UIO_SYSSPACE;
468 		auio.uio_td = NULL;
469 		auio.uio_resid = MAXPATHLEN;
470 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
471 		if (error) {
472 		badlink1:
473 			if (ndp->ni_pathlen > 1)
474 				uma_zfree(namei_zone, cp);
475 		badlink2:
476 			vrele(ndp->ni_dvp);
477 			vput(ndp->ni_vp);
478 			break;
479 		}
480 		linklen = MAXPATHLEN - auio.uio_resid;
481 		if (linklen == 0) {
482 			error = ENOENT;
483 			goto badlink1;
484 		}
485 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
486 			error = ENAMETOOLONG;
487 			goto badlink1;
488 		}
489 
490 		/*
491 		 * Adjust or replace path
492 		 */
493 		if (ndp->ni_pathlen > 1) {
494 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
495 			uma_zfree(namei_zone, cnp->cn_pnbuf);
496 			cnp->cn_pnbuf = cp;
497 		} else
498 			cnp->cn_pnbuf[linklen] = '\0';
499 		ndp->ni_pathlen += linklen;
500 
501 		/*
502 		 * Cleanup refs for next loop and check if root directory
503 		 * should replace current directory.  Normally ni_dvp
504 		 * becomes the new base directory and is cleaned up when
505 		 * we loop.  Explicitly null pointers after invalidation
506 		 * to clarify operation.
507 		 */
508 		vput(ndp->ni_vp);
509 		ndp->ni_vp = NULL;
510 
511 		if (cnp->cn_pnbuf[0] == '/') {
512 			vrele(ndp->ni_dvp);
513 			ndp->ni_dvp = ndp->ni_rootdir;
514 			VREF(ndp->ni_dvp);
515 		}
516 		ndp->ni_startdir = ndp->ni_dvp;
517 		ndp->ni_dvp = NULL;
518 	}
519 	if (!lockleaf)
520 		cnp->cn_flags &= ~LOCKLEAF;
521 
522 out:
523 	if (error) {
524 		nfsvno_relpathbuf(ndp);
525 		ndp->ni_vp = NULL;
526 		ndp->ni_dvp = NULL;
527 		ndp->ni_startdir = NULL;
528 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
529 		ndp->ni_dvp = NULL;
530 	}
531 
532 out1:
533 	NFSEXITCODE2(error, nd);
534 	return (error);
535 }
536 
537 /*
538  * Set up a pathname buffer and return a pointer to it and, optionally
539  * set a hash pointer.
540  */
541 void
542 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
543 {
544 	struct componentname *cnp = &ndp->ni_cnd;
545 
546 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
547 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
548 	if (hashpp != NULL)
549 		*hashpp = NULL;
550 	*bufpp = cnp->cn_pnbuf;
551 }
552 
553 /*
554  * Release the above path buffer, if not released by nfsvno_namei().
555  */
556 void
557 nfsvno_relpathbuf(struct nameidata *ndp)
558 {
559 
560 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
561 		panic("nfsrelpath");
562 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
563 	ndp->ni_cnd.cn_flags &= ~HASBUF;
564 }
565 
566 /*
567  * Readlink vnode op into an mbuf list.
568  */
569 int
570 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
571     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
572 {
573 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
574 	struct iovec *ivp = iv;
575 	struct uio io, *uiop = &io;
576 	struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
577 	int i, len, tlen, error = 0;
578 
579 	len = 0;
580 	i = 0;
581 	while (len < NFS_MAXPATHLEN) {
582 		NFSMGET(mp);
583 		MCLGET(mp, M_WAITOK);
584 		mp->m_len = M_SIZE(mp);
585 		if (len == 0) {
586 			mp3 = mp2 = mp;
587 		} else {
588 			mp2->m_next = mp;
589 			mp2 = mp;
590 		}
591 		if ((len + mp->m_len) > NFS_MAXPATHLEN) {
592 			mp->m_len = NFS_MAXPATHLEN - len;
593 			len = NFS_MAXPATHLEN;
594 		} else {
595 			len += mp->m_len;
596 		}
597 		ivp->iov_base = mtod(mp, caddr_t);
598 		ivp->iov_len = mp->m_len;
599 		i++;
600 		ivp++;
601 	}
602 	uiop->uio_iov = iv;
603 	uiop->uio_iovcnt = i;
604 	uiop->uio_offset = 0;
605 	uiop->uio_resid = len;
606 	uiop->uio_rw = UIO_READ;
607 	uiop->uio_segflg = UIO_SYSSPACE;
608 	uiop->uio_td = NULL;
609 	error = VOP_READLINK(vp, uiop, cred);
610 	if (error) {
611 		m_freem(mp3);
612 		*lenp = 0;
613 		goto out;
614 	}
615 	if (uiop->uio_resid > 0) {
616 		len -= uiop->uio_resid;
617 		tlen = NFSM_RNDUP(len);
618 		nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
619 	}
620 	*lenp = len;
621 	*mpp = mp3;
622 	*mpendp = mp;
623 
624 out:
625 	NFSEXITCODE(error);
626 	return (error);
627 }
628 
629 /*
630  * Read vnode op call into mbuf list.
631  */
632 int
633 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
634     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
635 {
636 	struct mbuf *m;
637 	int i;
638 	struct iovec *iv;
639 	struct iovec *iv2;
640 	int error = 0, len, left, siz, tlen, ioflag = 0;
641 	struct mbuf *m2 = NULL, *m3;
642 	struct uio io, *uiop = &io;
643 	struct nfsheur *nh;
644 
645 	len = left = NFSM_RNDUP(cnt);
646 	m3 = NULL;
647 	/*
648 	 * Generate the mbuf list with the uio_iov ref. to it.
649 	 */
650 	i = 0;
651 	while (left > 0) {
652 		NFSMGET(m);
653 		MCLGET(m, M_WAITOK);
654 		m->m_len = 0;
655 		siz = min(M_TRAILINGSPACE(m), left);
656 		left -= siz;
657 		i++;
658 		if (m3)
659 			m2->m_next = m;
660 		else
661 			m3 = m;
662 		m2 = m;
663 	}
664 	MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
665 	    M_TEMP, M_WAITOK);
666 	uiop->uio_iov = iv2 = iv;
667 	m = m3;
668 	left = len;
669 	i = 0;
670 	while (left > 0) {
671 		if (m == NULL)
672 			panic("nfsvno_read iov");
673 		siz = min(M_TRAILINGSPACE(m), left);
674 		if (siz > 0) {
675 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
676 			iv->iov_len = siz;
677 			m->m_len += siz;
678 			left -= siz;
679 			iv++;
680 			i++;
681 		}
682 		m = m->m_next;
683 	}
684 	uiop->uio_iovcnt = i;
685 	uiop->uio_offset = off;
686 	uiop->uio_resid = len;
687 	uiop->uio_rw = UIO_READ;
688 	uiop->uio_segflg = UIO_SYSSPACE;
689 	uiop->uio_td = NULL;
690 	nh = nfsrv_sequential_heuristic(uiop, vp);
691 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
692 	/* XXX KDM make this more systematic? */
693 	nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid;
694 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
695 	FREE((caddr_t)iv2, M_TEMP);
696 	if (error) {
697 		m_freem(m3);
698 		*mpp = NULL;
699 		goto out;
700 	}
701 	nh->nh_nextoff = uiop->uio_offset;
702 	tlen = len - uiop->uio_resid;
703 	cnt = cnt < tlen ? cnt : tlen;
704 	tlen = NFSM_RNDUP(cnt);
705 	if (tlen == 0) {
706 		m_freem(m3);
707 		m3 = NULL;
708 	} else if (len != tlen || tlen != cnt)
709 		nfsrv_adj(m3, len - tlen, tlen - cnt);
710 	*mpp = m3;
711 	*mpendp = m2;
712 
713 out:
714 	NFSEXITCODE(error);
715 	return (error);
716 }
717 
718 /*
719  * Write vnode op from an mbuf list.
720  */
721 int
722 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
723     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
724 {
725 	struct iovec *ivp;
726 	int i, len;
727 	struct iovec *iv;
728 	int ioflags, error;
729 	struct uio io, *uiop = &io;
730 	struct nfsheur *nh;
731 
732 	MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
733 	    M_WAITOK);
734 	uiop->uio_iov = iv = ivp;
735 	uiop->uio_iovcnt = cnt;
736 	i = mtod(mp, caddr_t) + mp->m_len - cp;
737 	len = retlen;
738 	while (len > 0) {
739 		if (mp == NULL)
740 			panic("nfsvno_write");
741 		if (i > 0) {
742 			i = min(i, len);
743 			ivp->iov_base = cp;
744 			ivp->iov_len = i;
745 			ivp++;
746 			len -= i;
747 		}
748 		mp = mp->m_next;
749 		if (mp) {
750 			i = mp->m_len;
751 			cp = mtod(mp, caddr_t);
752 		}
753 	}
754 
755 	if (stable == NFSWRITE_UNSTABLE)
756 		ioflags = IO_NODELOCKED;
757 	else
758 		ioflags = (IO_SYNC | IO_NODELOCKED);
759 	uiop->uio_resid = retlen;
760 	uiop->uio_rw = UIO_WRITE;
761 	uiop->uio_segflg = UIO_SYSSPACE;
762 	NFSUIOPROC(uiop, p);
763 	uiop->uio_offset = off;
764 	nh = nfsrv_sequential_heuristic(uiop, vp);
765 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
766 	/* XXX KDM make this more systematic? */
767 	nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
768 	error = VOP_WRITE(vp, uiop, ioflags, cred);
769 	if (error == 0)
770 		nh->nh_nextoff = uiop->uio_offset;
771 	FREE((caddr_t)iv, M_TEMP);
772 
773 	NFSEXITCODE(error);
774 	return (error);
775 }
776 
777 /*
778  * Common code for creating a regular file (plus special files for V2).
779  */
780 int
781 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
782     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
783     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
784 {
785 	u_quad_t tempsize;
786 	int error;
787 
788 	error = nd->nd_repstat;
789 	if (!error && ndp->ni_vp == NULL) {
790 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
791 			vrele(ndp->ni_startdir);
792 			error = VOP_CREATE(ndp->ni_dvp,
793 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
794 			vput(ndp->ni_dvp);
795 			nfsvno_relpathbuf(ndp);
796 			if (!error) {
797 				if (*exclusive_flagp) {
798 					*exclusive_flagp = 0;
799 					NFSVNO_ATTRINIT(nvap);
800 					nvap->na_atime.tv_sec = cverf[0];
801 					nvap->na_atime.tv_nsec = cverf[1];
802 					error = VOP_SETATTR(ndp->ni_vp,
803 					    &nvap->na_vattr, nd->nd_cred);
804 					if (error != 0) {
805 						vput(ndp->ni_vp);
806 						ndp->ni_vp = NULL;
807 						error = NFSERR_NOTSUPP;
808 					}
809 				}
810 			}
811 		/*
812 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
813 		 * (This implies, just get out on an error.)
814 		 */
815 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
816 			nvap->na_type == VFIFO) {
817 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
818 				nvap->na_type = VFIFO;
819                         if (nvap->na_type != VFIFO &&
820 			    (error = priv_check_cred(nd->nd_cred,
821 			     PRIV_VFS_MKNOD_DEV, 0))) {
822 				vrele(ndp->ni_startdir);
823 				nfsvno_relpathbuf(ndp);
824 				vput(ndp->ni_dvp);
825 				goto out;
826 			}
827 			nvap->na_rdev = rdev;
828 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
829 			    &ndp->ni_cnd, &nvap->na_vattr);
830 			vput(ndp->ni_dvp);
831 			nfsvno_relpathbuf(ndp);
832 			vrele(ndp->ni_startdir);
833 			if (error)
834 				goto out;
835 		} else {
836 			vrele(ndp->ni_startdir);
837 			nfsvno_relpathbuf(ndp);
838 			vput(ndp->ni_dvp);
839 			error = ENXIO;
840 			goto out;
841 		}
842 		*vpp = ndp->ni_vp;
843 	} else {
844 		/*
845 		 * Handle cases where error is already set and/or
846 		 * the file exists.
847 		 * 1 - clean up the lookup
848 		 * 2 - iff !error and na_size set, truncate it
849 		 */
850 		vrele(ndp->ni_startdir);
851 		nfsvno_relpathbuf(ndp);
852 		*vpp = ndp->ni_vp;
853 		if (ndp->ni_dvp == *vpp)
854 			vrele(ndp->ni_dvp);
855 		else
856 			vput(ndp->ni_dvp);
857 		if (!error && nvap->na_size != VNOVAL) {
858 			error = nfsvno_accchk(*vpp, VWRITE,
859 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
860 			    NFSACCCHK_VPISLOCKED, NULL);
861 			if (!error) {
862 				tempsize = nvap->na_size;
863 				NFSVNO_ATTRINIT(nvap);
864 				nvap->na_size = tempsize;
865 				error = VOP_SETATTR(*vpp,
866 				    &nvap->na_vattr, nd->nd_cred);
867 			}
868 		}
869 		if (error)
870 			vput(*vpp);
871 	}
872 
873 out:
874 	NFSEXITCODE(error);
875 	return (error);
876 }
877 
878 /*
879  * Do a mknod vnode op.
880  */
881 int
882 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
883     struct thread *p)
884 {
885 	int error = 0;
886 	enum vtype vtyp;
887 
888 	vtyp = nvap->na_type;
889 	/*
890 	 * Iff doesn't exist, create it.
891 	 */
892 	if (ndp->ni_vp) {
893 		vrele(ndp->ni_startdir);
894 		nfsvno_relpathbuf(ndp);
895 		vput(ndp->ni_dvp);
896 		vrele(ndp->ni_vp);
897 		error = EEXIST;
898 		goto out;
899 	}
900 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
901 		vrele(ndp->ni_startdir);
902 		nfsvno_relpathbuf(ndp);
903 		vput(ndp->ni_dvp);
904 		error = NFSERR_BADTYPE;
905 		goto out;
906 	}
907 	if (vtyp == VSOCK) {
908 		vrele(ndp->ni_startdir);
909 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
910 		    &ndp->ni_cnd, &nvap->na_vattr);
911 		vput(ndp->ni_dvp);
912 		nfsvno_relpathbuf(ndp);
913 	} else {
914 		if (nvap->na_type != VFIFO &&
915 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
916 			vrele(ndp->ni_startdir);
917 			nfsvno_relpathbuf(ndp);
918 			vput(ndp->ni_dvp);
919 			goto out;
920 		}
921 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
922 		    &ndp->ni_cnd, &nvap->na_vattr);
923 		vput(ndp->ni_dvp);
924 		nfsvno_relpathbuf(ndp);
925 		vrele(ndp->ni_startdir);
926 		/*
927 		 * Since VOP_MKNOD returns the ni_vp, I can't
928 		 * see any reason to do the lookup.
929 		 */
930 	}
931 
932 out:
933 	NFSEXITCODE(error);
934 	return (error);
935 }
936 
937 /*
938  * Mkdir vnode op.
939  */
940 int
941 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
942     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
943 {
944 	int error = 0;
945 
946 	if (ndp->ni_vp != NULL) {
947 		if (ndp->ni_dvp == ndp->ni_vp)
948 			vrele(ndp->ni_dvp);
949 		else
950 			vput(ndp->ni_dvp);
951 		vrele(ndp->ni_vp);
952 		nfsvno_relpathbuf(ndp);
953 		error = EEXIST;
954 		goto out;
955 	}
956 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
957 	    &nvap->na_vattr);
958 	vput(ndp->ni_dvp);
959 	nfsvno_relpathbuf(ndp);
960 
961 out:
962 	NFSEXITCODE(error);
963 	return (error);
964 }
965 
966 /*
967  * symlink vnode op.
968  */
969 int
970 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
971     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
972     struct nfsexstuff *exp)
973 {
974 	int error = 0;
975 
976 	if (ndp->ni_vp) {
977 		vrele(ndp->ni_startdir);
978 		nfsvno_relpathbuf(ndp);
979 		if (ndp->ni_dvp == ndp->ni_vp)
980 			vrele(ndp->ni_dvp);
981 		else
982 			vput(ndp->ni_dvp);
983 		vrele(ndp->ni_vp);
984 		error = EEXIST;
985 		goto out;
986 	}
987 
988 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
989 	    &nvap->na_vattr, pathcp);
990 	vput(ndp->ni_dvp);
991 	vrele(ndp->ni_startdir);
992 	nfsvno_relpathbuf(ndp);
993 	/*
994 	 * Although FreeBSD still had the lookup code in
995 	 * it for 7/current, there doesn't seem to be any
996 	 * point, since VOP_SYMLINK() returns the ni_vp.
997 	 * Just vput it for v2.
998 	 */
999 	if (!not_v2 && !error)
1000 		vput(ndp->ni_vp);
1001 
1002 out:
1003 	NFSEXITCODE(error);
1004 	return (error);
1005 }
1006 
1007 /*
1008  * Parse symbolic link arguments.
1009  * This function has an ugly side effect. It will MALLOC() an area for
1010  * the symlink and set iov_base to point to it, only if it succeeds.
1011  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1012  * be FREE'd later.
1013  */
1014 int
1015 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1016     struct thread *p, char **pathcpp, int *lenp)
1017 {
1018 	u_int32_t *tl;
1019 	char *pathcp = NULL;
1020 	int error = 0, len;
1021 	struct nfsv2_sattr *sp;
1022 
1023 	*pathcpp = NULL;
1024 	*lenp = 0;
1025 	if ((nd->nd_flag & ND_NFSV3) &&
1026 	    (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1027 		goto nfsmout;
1028 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1029 	len = fxdr_unsigned(int, *tl);
1030 	if (len > NFS_MAXPATHLEN || len <= 0) {
1031 		error = EBADRPC;
1032 		goto nfsmout;
1033 	}
1034 	MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
1035 	error = nfsrv_mtostr(nd, pathcp, len);
1036 	if (error)
1037 		goto nfsmout;
1038 	if (nd->nd_flag & ND_NFSV2) {
1039 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1040 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1041 	}
1042 	*pathcpp = pathcp;
1043 	*lenp = len;
1044 	NFSEXITCODE2(0, nd);
1045 	return (0);
1046 nfsmout:
1047 	if (pathcp)
1048 		free(pathcp, M_TEMP);
1049 	NFSEXITCODE2(error, nd);
1050 	return (error);
1051 }
1052 
1053 /*
1054  * Remove a non-directory object.
1055  */
1056 int
1057 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1058     struct thread *p, struct nfsexstuff *exp)
1059 {
1060 	struct vnode *vp;
1061 	int error = 0;
1062 
1063 	vp = ndp->ni_vp;
1064 	if (vp->v_type == VDIR)
1065 		error = NFSERR_ISDIR;
1066 	else if (is_v4)
1067 		error = nfsrv_checkremove(vp, 1, p);
1068 	if (!error)
1069 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1070 	if (ndp->ni_dvp == vp)
1071 		vrele(ndp->ni_dvp);
1072 	else
1073 		vput(ndp->ni_dvp);
1074 	vput(vp);
1075 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1076 		nfsvno_relpathbuf(ndp);
1077 	NFSEXITCODE(error);
1078 	return (error);
1079 }
1080 
1081 /*
1082  * Remove a directory.
1083  */
1084 int
1085 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1086     struct thread *p, struct nfsexstuff *exp)
1087 {
1088 	struct vnode *vp;
1089 	int error = 0;
1090 
1091 	vp = ndp->ni_vp;
1092 	if (vp->v_type != VDIR) {
1093 		error = ENOTDIR;
1094 		goto out;
1095 	}
1096 	/*
1097 	 * No rmdir "." please.
1098 	 */
1099 	if (ndp->ni_dvp == vp) {
1100 		error = EINVAL;
1101 		goto out;
1102 	}
1103 	/*
1104 	 * The root of a mounted filesystem cannot be deleted.
1105 	 */
1106 	if (vp->v_vflag & VV_ROOT)
1107 		error = EBUSY;
1108 out:
1109 	if (!error)
1110 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1111 	if (ndp->ni_dvp == vp)
1112 		vrele(ndp->ni_dvp);
1113 	else
1114 		vput(ndp->ni_dvp);
1115 	vput(vp);
1116 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1117 		nfsvno_relpathbuf(ndp);
1118 	NFSEXITCODE(error);
1119 	return (error);
1120 }
1121 
1122 /*
1123  * Rename vnode op.
1124  */
1125 int
1126 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1127     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1128 {
1129 	struct vnode *fvp, *tvp, *tdvp;
1130 	int error = 0;
1131 
1132 	fvp = fromndp->ni_vp;
1133 	if (ndstat) {
1134 		vrele(fromndp->ni_dvp);
1135 		vrele(fvp);
1136 		error = ndstat;
1137 		goto out1;
1138 	}
1139 	tdvp = tondp->ni_dvp;
1140 	tvp = tondp->ni_vp;
1141 	if (tvp != NULL) {
1142 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1143 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1144 			goto out;
1145 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1146 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1147 			goto out;
1148 		}
1149 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1150 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1151 			goto out;
1152 		}
1153 
1154 		/*
1155 		 * A rename to '.' or '..' results in a prematurely
1156 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1157 		 * here.
1158 		 */
1159 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1160 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1161 		    (tondp->ni_cnd.cn_namelen == 2 &&
1162 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1163 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1164 			error = EINVAL;
1165 			goto out;
1166 		}
1167 	}
1168 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1169 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1170 		goto out;
1171 	}
1172 	if (fvp->v_mount != tdvp->v_mount) {
1173 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1174 		goto out;
1175 	}
1176 	if (fvp == tdvp) {
1177 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1178 		goto out;
1179 	}
1180 	if (fvp == tvp) {
1181 		/*
1182 		 * If source and destination are the same, there is nothing to
1183 		 * do. Set error to -1 to indicate this.
1184 		 */
1185 		error = -1;
1186 		goto out;
1187 	}
1188 	if (ndflag & ND_NFSV4) {
1189 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1190 			error = nfsrv_checkremove(fvp, 0, p);
1191 			NFSVOPUNLOCK(fvp, 0);
1192 		} else
1193 			error = EPERM;
1194 		if (tvp && !error)
1195 			error = nfsrv_checkremove(tvp, 1, p);
1196 	} else {
1197 		/*
1198 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1199 		 * that the NFSv4 client won't be confused by the rename.
1200 		 * Since nfsd_recalldelegation() can only be called on an
1201 		 * unlocked vnode at this point and fvp is the file that will
1202 		 * still exist after the rename, just do fvp.
1203 		 */
1204 		nfsd_recalldelegation(fvp, p);
1205 	}
1206 out:
1207 	if (!error) {
1208 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1209 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1210 		    &tondp->ni_cnd);
1211 	} else {
1212 		if (tdvp == tvp)
1213 			vrele(tdvp);
1214 		else
1215 			vput(tdvp);
1216 		if (tvp)
1217 			vput(tvp);
1218 		vrele(fromndp->ni_dvp);
1219 		vrele(fvp);
1220 		if (error == -1)
1221 			error = 0;
1222 	}
1223 	vrele(tondp->ni_startdir);
1224 	nfsvno_relpathbuf(tondp);
1225 out1:
1226 	vrele(fromndp->ni_startdir);
1227 	nfsvno_relpathbuf(fromndp);
1228 	NFSEXITCODE(error);
1229 	return (error);
1230 }
1231 
1232 /*
1233  * Link vnode op.
1234  */
1235 int
1236 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1237     struct thread *p, struct nfsexstuff *exp)
1238 {
1239 	struct vnode *xp;
1240 	int error = 0;
1241 
1242 	xp = ndp->ni_vp;
1243 	if (xp != NULL) {
1244 		error = EEXIST;
1245 	} else {
1246 		xp = ndp->ni_dvp;
1247 		if (vp->v_mount != xp->v_mount)
1248 			error = EXDEV;
1249 	}
1250 	if (!error) {
1251 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1252 		if ((vp->v_iflag & VI_DOOMED) == 0)
1253 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1254 		else
1255 			error = EPERM;
1256 		if (ndp->ni_dvp == vp)
1257 			vrele(ndp->ni_dvp);
1258 		else
1259 			vput(ndp->ni_dvp);
1260 		NFSVOPUNLOCK(vp, 0);
1261 	} else {
1262 		if (ndp->ni_dvp == ndp->ni_vp)
1263 			vrele(ndp->ni_dvp);
1264 		else
1265 			vput(ndp->ni_dvp);
1266 		if (ndp->ni_vp)
1267 			vrele(ndp->ni_vp);
1268 	}
1269 	nfsvno_relpathbuf(ndp);
1270 	NFSEXITCODE(error);
1271 	return (error);
1272 }
1273 
1274 /*
1275  * Do the fsync() appropriate for the commit.
1276  */
1277 int
1278 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1279     struct thread *td)
1280 {
1281 	int error = 0;
1282 
1283 	/*
1284 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1285 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1286 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1287 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1288 	 * File systems that do not use the buffer cache (as indicated
1289 	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1290 	 */
1291 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1292 	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1293 		/*
1294 		 * Give up and do the whole thing
1295 		 */
1296 		if (vp->v_object &&
1297 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1298 			VM_OBJECT_WLOCK(vp->v_object);
1299 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1300 			VM_OBJECT_WUNLOCK(vp->v_object);
1301 		}
1302 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1303 	} else {
1304 		/*
1305 		 * Locate and synchronously write any buffers that fall
1306 		 * into the requested range.  Note:  we are assuming that
1307 		 * f_iosize is a power of 2.
1308 		 */
1309 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1310 		int iomask = iosize - 1;
1311 		struct bufobj *bo;
1312 		daddr_t lblkno;
1313 
1314 		/*
1315 		 * Align to iosize boundary, super-align to page boundary.
1316 		 */
1317 		if (off & iomask) {
1318 			cnt += off & iomask;
1319 			off &= ~(u_quad_t)iomask;
1320 		}
1321 		if (off & PAGE_MASK) {
1322 			cnt += off & PAGE_MASK;
1323 			off &= ~(u_quad_t)PAGE_MASK;
1324 		}
1325 		lblkno = off / iosize;
1326 
1327 		if (vp->v_object &&
1328 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1329 			VM_OBJECT_WLOCK(vp->v_object);
1330 			vm_object_page_clean(vp->v_object, off, off + cnt,
1331 			    OBJPC_SYNC);
1332 			VM_OBJECT_WUNLOCK(vp->v_object);
1333 		}
1334 
1335 		bo = &vp->v_bufobj;
1336 		BO_LOCK(bo);
1337 		while (cnt > 0) {
1338 			struct buf *bp;
1339 
1340 			/*
1341 			 * If we have a buffer and it is marked B_DELWRI we
1342 			 * have to lock and write it.  Otherwise the prior
1343 			 * write is assumed to have already been committed.
1344 			 *
1345 			 * gbincore() can return invalid buffers now so we
1346 			 * have to check that bit as well (though B_DELWRI
1347 			 * should not be set if B_INVAL is set there could be
1348 			 * a race here since we haven't locked the buffer).
1349 			 */
1350 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1351 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1352 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1353 					BO_LOCK(bo);
1354 					continue; /* retry */
1355 				}
1356 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1357 				    B_DELWRI) {
1358 					bremfree(bp);
1359 					bp->b_flags &= ~B_ASYNC;
1360 					bwrite(bp);
1361 					++nfs_commit_miss;
1362 				} else
1363 					BUF_UNLOCK(bp);
1364 				BO_LOCK(bo);
1365 			}
1366 			++nfs_commit_blks;
1367 			if (cnt < iosize)
1368 				break;
1369 			cnt -= iosize;
1370 			++lblkno;
1371 		}
1372 		BO_UNLOCK(bo);
1373 	}
1374 	NFSEXITCODE(error);
1375 	return (error);
1376 }
1377 
1378 /*
1379  * Statfs vnode op.
1380  */
1381 int
1382 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1383 {
1384 	int error;
1385 
1386 	error = VFS_STATFS(vp->v_mount, sf);
1387 	if (error == 0) {
1388 		/*
1389 		 * Since NFS handles these values as unsigned on the
1390 		 * wire, there is no way to represent negative values,
1391 		 * so set them to 0. Without this, they will appear
1392 		 * to be very large positive values for clients like
1393 		 * Solaris10.
1394 		 */
1395 		if (sf->f_bavail < 0)
1396 			sf->f_bavail = 0;
1397 		if (sf->f_ffree < 0)
1398 			sf->f_ffree = 0;
1399 	}
1400 	NFSEXITCODE(error);
1401 	return (error);
1402 }
1403 
1404 /*
1405  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1406  * must handle nfsrv_opencheck() calls after any other access checks.
1407  */
1408 void
1409 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1410     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1411     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1412     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1413     struct nfsexstuff *exp, struct vnode **vpp)
1414 {
1415 	struct vnode *vp = NULL;
1416 	u_quad_t tempsize;
1417 	struct nfsexstuff nes;
1418 
1419 	if (ndp->ni_vp == NULL)
1420 		nd->nd_repstat = nfsrv_opencheck(clientid,
1421 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1422 	if (!nd->nd_repstat) {
1423 		if (ndp->ni_vp == NULL) {
1424 			vrele(ndp->ni_startdir);
1425 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1426 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1427 			vput(ndp->ni_dvp);
1428 			nfsvno_relpathbuf(ndp);
1429 			if (!nd->nd_repstat) {
1430 				if (*exclusive_flagp) {
1431 					*exclusive_flagp = 0;
1432 					NFSVNO_ATTRINIT(nvap);
1433 					nvap->na_atime.tv_sec = cverf[0];
1434 					nvap->na_atime.tv_nsec = cverf[1];
1435 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1436 					    &nvap->na_vattr, cred);
1437 					if (nd->nd_repstat != 0) {
1438 						vput(ndp->ni_vp);
1439 						ndp->ni_vp = NULL;
1440 						nd->nd_repstat = NFSERR_NOTSUPP;
1441 					} else
1442 						NFSSETBIT_ATTRBIT(attrbitp,
1443 						    NFSATTRBIT_TIMEACCESS);
1444 				} else {
1445 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1446 					    aclp, p, attrbitp, exp);
1447 				}
1448 			}
1449 			vp = ndp->ni_vp;
1450 		} else {
1451 			if (ndp->ni_startdir)
1452 				vrele(ndp->ni_startdir);
1453 			nfsvno_relpathbuf(ndp);
1454 			vp = ndp->ni_vp;
1455 			if (create == NFSV4OPEN_CREATE) {
1456 				if (ndp->ni_dvp == vp)
1457 					vrele(ndp->ni_dvp);
1458 				else
1459 					vput(ndp->ni_dvp);
1460 			}
1461 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1462 				if (ndp->ni_cnd.cn_flags & RDONLY)
1463 					NFSVNO_SETEXRDONLY(&nes);
1464 				else
1465 					NFSVNO_EXINIT(&nes);
1466 				nd->nd_repstat = nfsvno_accchk(vp,
1467 				    VWRITE, cred, &nes, p,
1468 				    NFSACCCHK_NOOVERRIDE,
1469 				    NFSACCCHK_VPISLOCKED, NULL);
1470 				nd->nd_repstat = nfsrv_opencheck(clientid,
1471 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1472 				if (!nd->nd_repstat) {
1473 					tempsize = nvap->na_size;
1474 					NFSVNO_ATTRINIT(nvap);
1475 					nvap->na_size = tempsize;
1476 					nd->nd_repstat = VOP_SETATTR(vp,
1477 					    &nvap->na_vattr, cred);
1478 				}
1479 			} else if (vp->v_type == VREG) {
1480 				nd->nd_repstat = nfsrv_opencheck(clientid,
1481 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1482 			}
1483 		}
1484 	} else {
1485 		if (ndp->ni_cnd.cn_flags & HASBUF)
1486 			nfsvno_relpathbuf(ndp);
1487 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1488 			vrele(ndp->ni_startdir);
1489 			if (ndp->ni_dvp == ndp->ni_vp)
1490 				vrele(ndp->ni_dvp);
1491 			else
1492 				vput(ndp->ni_dvp);
1493 			if (ndp->ni_vp)
1494 				vput(ndp->ni_vp);
1495 		}
1496 	}
1497 	*vpp = vp;
1498 
1499 	NFSEXITCODE2(0, nd);
1500 }
1501 
1502 /*
1503  * Updates the file rev and sets the mtime and ctime
1504  * to the current clock time, returning the va_filerev and va_Xtime
1505  * values.
1506  * Return ESTALE to indicate the vnode is VI_DOOMED.
1507  */
1508 int
1509 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1510     struct ucred *cred, struct thread *p)
1511 {
1512 	struct vattr va;
1513 
1514 	VATTR_NULL(&va);
1515 	vfs_timestamp(&va.va_mtime);
1516 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1517 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1518 		if ((vp->v_iflag & VI_DOOMED) != 0)
1519 			return (ESTALE);
1520 	}
1521 	(void) VOP_SETATTR(vp, &va, cred);
1522 	(void) nfsvno_getattr(vp, nvap, cred, p, 1);
1523 	return (0);
1524 }
1525 
1526 /*
1527  * Glue routine to nfsv4_fillattr().
1528  */
1529 int
1530 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1531     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1532     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1533     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1534 {
1535 	int error;
1536 
1537 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1538 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1539 	    mounted_on_fileno);
1540 	NFSEXITCODE2(0, nd);
1541 	return (error);
1542 }
1543 
1544 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1545 /*
1546  * nfs readdir service
1547  * - mallocs what it thinks is enough to read
1548  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1549  * - calls VOP_READDIR()
1550  * - loops around building the reply
1551  *	if the output generated exceeds count break out of loop
1552  *	The NFSM_CLGET macro is used here so that the reply will be packed
1553  *	tightly in mbuf clusters.
1554  * - it trims out records with d_fileno == 0
1555  *	this doesn't matter for Unix clients, but they might confuse clients
1556  *	for other os'.
1557  * - it trims out records with d_type == DT_WHT
1558  *	these cannot be seen through NFS (unless we extend the protocol)
1559  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1560  * PS: The NFS protocol spec. does not clarify what the "count" byte
1561  *	argument is a count of.. just name strings and file id's or the
1562  *	entire reply rpc or ...
1563  *	I tried just file name and id sizes and it confused the Sun client,
1564  *	so I am using the full rpc size now. The "paranoia.." comment refers
1565  *	to including the status longwords that are not a part of the dir.
1566  *	"entry" structures, but are in the rpc.
1567  */
1568 int
1569 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1570     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1571 {
1572 	struct dirent *dp;
1573 	u_int32_t *tl;
1574 	int dirlen;
1575 	char *cpos, *cend, *rbuf;
1576 	struct nfsvattr at;
1577 	int nlen, error = 0, getret = 1;
1578 	int siz, cnt, fullsiz, eofflag, ncookies;
1579 	u_int64_t off, toff, verf;
1580 	u_long *cookies = NULL, *cookiep;
1581 	struct uio io;
1582 	struct iovec iv;
1583 	int is_ufs;
1584 
1585 	if (nd->nd_repstat) {
1586 		nfsrv_postopattr(nd, getret, &at);
1587 		goto out;
1588 	}
1589 	if (nd->nd_flag & ND_NFSV2) {
1590 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1591 		off = fxdr_unsigned(u_quad_t, *tl++);
1592 	} else {
1593 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1594 		off = fxdr_hyper(tl);
1595 		tl += 2;
1596 		verf = fxdr_hyper(tl);
1597 		tl += 2;
1598 	}
1599 	toff = off;
1600 	cnt = fxdr_unsigned(int, *tl);
1601 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1602 		cnt = NFS_SRVMAXDATA(nd);
1603 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1604 	fullsiz = siz;
1605 	if (nd->nd_flag & ND_NFSV3) {
1606 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1607 		    p, 1);
1608 #if 0
1609 		/*
1610 		 * va_filerev is not sufficient as a cookie verifier,
1611 		 * since it is not supposed to change when entries are
1612 		 * removed/added unless that offset cookies returned to
1613 		 * the client are no longer valid.
1614 		 */
1615 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
1616 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1617 #endif
1618 	}
1619 	if (!nd->nd_repstat && vp->v_type != VDIR)
1620 		nd->nd_repstat = NFSERR_NOTDIR;
1621 	if (nd->nd_repstat == 0 && cnt == 0) {
1622 		if (nd->nd_flag & ND_NFSV2)
1623 			/* NFSv2 does not have NFSERR_TOOSMALL */
1624 			nd->nd_repstat = EPERM;
1625 		else
1626 			nd->nd_repstat = NFSERR_TOOSMALL;
1627 	}
1628 	if (!nd->nd_repstat)
1629 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1630 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1631 		    NFSACCCHK_VPISLOCKED, NULL);
1632 	if (nd->nd_repstat) {
1633 		vput(vp);
1634 		if (nd->nd_flag & ND_NFSV3)
1635 			nfsrv_postopattr(nd, getret, &at);
1636 		goto out;
1637 	}
1638 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1639 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1640 again:
1641 	eofflag = 0;
1642 	if (cookies) {
1643 		free((caddr_t)cookies, M_TEMP);
1644 		cookies = NULL;
1645 	}
1646 
1647 	iv.iov_base = rbuf;
1648 	iv.iov_len = siz;
1649 	io.uio_iov = &iv;
1650 	io.uio_iovcnt = 1;
1651 	io.uio_offset = (off_t)off;
1652 	io.uio_resid = siz;
1653 	io.uio_segflg = UIO_SYSSPACE;
1654 	io.uio_rw = UIO_READ;
1655 	io.uio_td = NULL;
1656 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1657 	    &cookies);
1658 	off = (u_int64_t)io.uio_offset;
1659 	if (io.uio_resid)
1660 		siz -= io.uio_resid;
1661 
1662 	if (!cookies && !nd->nd_repstat)
1663 		nd->nd_repstat = NFSERR_PERM;
1664 	if (nd->nd_flag & ND_NFSV3) {
1665 		getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1666 		if (!nd->nd_repstat)
1667 			nd->nd_repstat = getret;
1668 	}
1669 
1670 	/*
1671 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
1672 	 */
1673 	if (nd->nd_repstat) {
1674 		vput(vp);
1675 		free((caddr_t)rbuf, M_TEMP);
1676 		if (cookies)
1677 			free((caddr_t)cookies, M_TEMP);
1678 		if (nd->nd_flag & ND_NFSV3)
1679 			nfsrv_postopattr(nd, getret, &at);
1680 		goto out;
1681 	}
1682 	/*
1683 	 * If nothing read, return eof
1684 	 * rpc reply
1685 	 */
1686 	if (siz == 0) {
1687 		vput(vp);
1688 		if (nd->nd_flag & ND_NFSV2) {
1689 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1690 		} else {
1691 			nfsrv_postopattr(nd, getret, &at);
1692 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1693 			txdr_hyper(at.na_filerev, tl);
1694 			tl += 2;
1695 		}
1696 		*tl++ = newnfs_false;
1697 		*tl = newnfs_true;
1698 		FREE((caddr_t)rbuf, M_TEMP);
1699 		FREE((caddr_t)cookies, M_TEMP);
1700 		goto out;
1701 	}
1702 
1703 	/*
1704 	 * Check for degenerate cases of nothing useful read.
1705 	 * If so go try again
1706 	 */
1707 	cpos = rbuf;
1708 	cend = rbuf + siz;
1709 	dp = (struct dirent *)cpos;
1710 	cookiep = cookies;
1711 
1712 	/*
1713 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1714 	 * directory offset up to a block boundary, so it is necessary to
1715 	 * skip over the records that precede the requested offset. This
1716 	 * requires the assumption that file offset cookies monotonically
1717 	 * increase.
1718 	 */
1719 	while (cpos < cend && ncookies > 0 &&
1720 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1721 	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
1722 		cpos += dp->d_reclen;
1723 		dp = (struct dirent *)cpos;
1724 		cookiep++;
1725 		ncookies--;
1726 	}
1727 	if (cpos >= cend || ncookies == 0) {
1728 		siz = fullsiz;
1729 		toff = off;
1730 		goto again;
1731 	}
1732 	vput(vp);
1733 
1734 	/*
1735 	 * dirlen is the size of the reply, including all XDR and must
1736 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1737 	 * if the XDR should be included in "count", but to be safe, we do.
1738 	 * (Include the two booleans at the end of the reply in dirlen now.)
1739 	 */
1740 	if (nd->nd_flag & ND_NFSV3) {
1741 		nfsrv_postopattr(nd, getret, &at);
1742 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1743 		txdr_hyper(at.na_filerev, tl);
1744 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1745 	} else {
1746 		dirlen = 2 * NFSX_UNSIGNED;
1747 	}
1748 
1749 	/* Loop through the records and build reply */
1750 	while (cpos < cend && ncookies > 0) {
1751 		nlen = dp->d_namlen;
1752 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1753 			nlen <= NFS_MAXNAMLEN) {
1754 			if (nd->nd_flag & ND_NFSV3)
1755 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1756 			else
1757 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1758 			if (dirlen > cnt) {
1759 				eofflag = 0;
1760 				break;
1761 			}
1762 
1763 			/*
1764 			 * Build the directory record xdr from
1765 			 * the dirent entry.
1766 			 */
1767 			if (nd->nd_flag & ND_NFSV3) {
1768 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1769 				*tl++ = newnfs_true;
1770 				*tl++ = 0;
1771 			} else {
1772 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1773 				*tl++ = newnfs_true;
1774 			}
1775 			*tl = txdr_unsigned(dp->d_fileno);
1776 			(void) nfsm_strtom(nd, dp->d_name, nlen);
1777 			if (nd->nd_flag & ND_NFSV3) {
1778 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1779 				*tl++ = 0;
1780 			} else
1781 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1782 			*tl = txdr_unsigned(*cookiep);
1783 		}
1784 		cpos += dp->d_reclen;
1785 		dp = (struct dirent *)cpos;
1786 		cookiep++;
1787 		ncookies--;
1788 	}
1789 	if (cpos < cend)
1790 		eofflag = 0;
1791 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1792 	*tl++ = newnfs_false;
1793 	if (eofflag)
1794 		*tl = newnfs_true;
1795 	else
1796 		*tl = newnfs_false;
1797 	FREE((caddr_t)rbuf, M_TEMP);
1798 	FREE((caddr_t)cookies, M_TEMP);
1799 
1800 out:
1801 	NFSEXITCODE2(0, nd);
1802 	return (0);
1803 nfsmout:
1804 	vput(vp);
1805 	NFSEXITCODE2(error, nd);
1806 	return (error);
1807 }
1808 
1809 /*
1810  * Readdirplus for V3 and Readdir for V4.
1811  */
1812 int
1813 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1814     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1815 {
1816 	struct dirent *dp;
1817 	u_int32_t *tl;
1818 	int dirlen;
1819 	char *cpos, *cend, *rbuf;
1820 	struct vnode *nvp;
1821 	fhandle_t nfh;
1822 	struct nfsvattr nva, at, *nvap = &nva;
1823 	struct mbuf *mb0, *mb1;
1824 	struct nfsreferral *refp;
1825 	int nlen, r, error = 0, getret = 1, usevget = 1;
1826 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1827 	caddr_t bpos0, bpos1;
1828 	u_int64_t off, toff, verf;
1829 	u_long *cookies = NULL, *cookiep;
1830 	nfsattrbit_t attrbits, rderrbits, savbits;
1831 	struct uio io;
1832 	struct iovec iv;
1833 	struct componentname cn;
1834 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
1835 	struct mount *mp, *new_mp;
1836 	uint64_t mounted_on_fileno;
1837 
1838 	if (nd->nd_repstat) {
1839 		nfsrv_postopattr(nd, getret, &at);
1840 		goto out;
1841 	}
1842 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1843 	off = fxdr_hyper(tl);
1844 	toff = off;
1845 	tl += 2;
1846 	verf = fxdr_hyper(tl);
1847 	tl += 2;
1848 	siz = fxdr_unsigned(int, *tl++);
1849 	cnt = fxdr_unsigned(int, *tl);
1850 
1851 	/*
1852 	 * Use the server's maximum data transfer size as the upper bound
1853 	 * on reply datalen.
1854 	 */
1855 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1856 		cnt = NFS_SRVMAXDATA(nd);
1857 
1858 	/*
1859 	 * siz is a "hint" of how much directory information (name, fileid,
1860 	 * cookie) should be in the reply. At least one client "hints" 0,
1861 	 * so I set it to cnt for that case. I also round it up to the
1862 	 * next multiple of DIRBLKSIZ.
1863 	 */
1864 	if (siz <= 0)
1865 		siz = cnt;
1866 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1867 
1868 	if (nd->nd_flag & ND_NFSV4) {
1869 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1870 		if (error)
1871 			goto nfsmout;
1872 		NFSSET_ATTRBIT(&savbits, &attrbits);
1873 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1874 		NFSZERO_ATTRBIT(&rderrbits);
1875 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1876 	} else {
1877 		NFSZERO_ATTRBIT(&attrbits);
1878 	}
1879 	fullsiz = siz;
1880 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1881 	if (!nd->nd_repstat) {
1882 	    if (off && verf != at.na_filerev) {
1883 		/*
1884 		 * va_filerev is not sufficient as a cookie verifier,
1885 		 * since it is not supposed to change when entries are
1886 		 * removed/added unless that offset cookies returned to
1887 		 * the client are no longer valid.
1888 		 */
1889 #if 0
1890 		if (nd->nd_flag & ND_NFSV4) {
1891 			nd->nd_repstat = NFSERR_NOTSAME;
1892 		} else {
1893 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1894 		}
1895 #endif
1896 	    } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1897 		nd->nd_repstat = NFSERR_BAD_COOKIE;
1898 	    }
1899 	}
1900 	if (!nd->nd_repstat && vp->v_type != VDIR)
1901 		nd->nd_repstat = NFSERR_NOTDIR;
1902 	if (!nd->nd_repstat && cnt == 0)
1903 		nd->nd_repstat = NFSERR_TOOSMALL;
1904 	if (!nd->nd_repstat)
1905 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1906 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1907 		    NFSACCCHK_VPISLOCKED, NULL);
1908 	if (nd->nd_repstat) {
1909 		vput(vp);
1910 		if (nd->nd_flag & ND_NFSV3)
1911 			nfsrv_postopattr(nd, getret, &at);
1912 		goto out;
1913 	}
1914 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1915 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
1916 
1917 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1918 again:
1919 	eofflag = 0;
1920 	if (cookies) {
1921 		free((caddr_t)cookies, M_TEMP);
1922 		cookies = NULL;
1923 	}
1924 
1925 	iv.iov_base = rbuf;
1926 	iv.iov_len = siz;
1927 	io.uio_iov = &iv;
1928 	io.uio_iovcnt = 1;
1929 	io.uio_offset = (off_t)off;
1930 	io.uio_resid = siz;
1931 	io.uio_segflg = UIO_SYSSPACE;
1932 	io.uio_rw = UIO_READ;
1933 	io.uio_td = NULL;
1934 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1935 	    &cookies);
1936 	off = (u_int64_t)io.uio_offset;
1937 	if (io.uio_resid)
1938 		siz -= io.uio_resid;
1939 
1940 	getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1941 
1942 	if (!cookies && !nd->nd_repstat)
1943 		nd->nd_repstat = NFSERR_PERM;
1944 	if (!nd->nd_repstat)
1945 		nd->nd_repstat = getret;
1946 	if (nd->nd_repstat) {
1947 		vput(vp);
1948 		if (cookies)
1949 			free((caddr_t)cookies, M_TEMP);
1950 		free((caddr_t)rbuf, M_TEMP);
1951 		if (nd->nd_flag & ND_NFSV3)
1952 			nfsrv_postopattr(nd, getret, &at);
1953 		goto out;
1954 	}
1955 	/*
1956 	 * If nothing read, return eof
1957 	 * rpc reply
1958 	 */
1959 	if (siz == 0) {
1960 		vput(vp);
1961 		if (nd->nd_flag & ND_NFSV3)
1962 			nfsrv_postopattr(nd, getret, &at);
1963 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1964 		txdr_hyper(at.na_filerev, tl);
1965 		tl += 2;
1966 		*tl++ = newnfs_false;
1967 		*tl = newnfs_true;
1968 		free((caddr_t)cookies, M_TEMP);
1969 		free((caddr_t)rbuf, M_TEMP);
1970 		goto out;
1971 	}
1972 
1973 	/*
1974 	 * Check for degenerate cases of nothing useful read.
1975 	 * If so go try again
1976 	 */
1977 	cpos = rbuf;
1978 	cend = rbuf + siz;
1979 	dp = (struct dirent *)cpos;
1980 	cookiep = cookies;
1981 
1982 	/*
1983 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1984 	 * directory offset up to a block boundary, so it is necessary to
1985 	 * skip over the records that precede the requested offset. This
1986 	 * requires the assumption that file offset cookies monotonically
1987 	 * increase.
1988 	 */
1989 	while (cpos < cend && ncookies > 0 &&
1990 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1991 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
1992 	   ((nd->nd_flag & ND_NFSV4) &&
1993 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1994 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1995 		cpos += dp->d_reclen;
1996 		dp = (struct dirent *)cpos;
1997 		cookiep++;
1998 		ncookies--;
1999 	}
2000 	if (cpos >= cend || ncookies == 0) {
2001 		siz = fullsiz;
2002 		toff = off;
2003 		goto again;
2004 	}
2005 
2006 	/*
2007 	 * Busy the file system so that the mount point won't go away
2008 	 * and, as such, VFS_VGET() can be used safely.
2009 	 */
2010 	mp = vp->v_mount;
2011 	vfs_ref(mp);
2012 	NFSVOPUNLOCK(vp, 0);
2013 	nd->nd_repstat = vfs_busy(mp, 0);
2014 	vfs_rel(mp);
2015 	if (nd->nd_repstat != 0) {
2016 		vrele(vp);
2017 		free(cookies, M_TEMP);
2018 		free(rbuf, M_TEMP);
2019 		if (nd->nd_flag & ND_NFSV3)
2020 			nfsrv_postopattr(nd, getret, &at);
2021 		goto out;
2022 	}
2023 
2024 	/*
2025 	 * Check to see if entries in this directory can be safely acquired
2026 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2027 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2028 	 * automount of the snapshot directory that is required will
2029 	 * be done.
2030 	 * This needs to be done here for NFSv4, since NFSv4 never does
2031 	 * a VFS_VGET() for "." or "..".
2032 	 */
2033 	if (is_zfs == 1) {
2034 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2035 		if (r == EOPNOTSUPP) {
2036 			usevget = 0;
2037 			cn.cn_nameiop = LOOKUP;
2038 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2039 			cn.cn_cred = nd->nd_cred;
2040 			cn.cn_thread = p;
2041 		} else if (r == 0)
2042 			vput(nvp);
2043 	}
2044 
2045 	/*
2046 	 * Save this position, in case there is an error before one entry
2047 	 * is created.
2048 	 */
2049 	mb0 = nd->nd_mb;
2050 	bpos0 = nd->nd_bpos;
2051 
2052 	/*
2053 	 * Fill in the first part of the reply.
2054 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2055 	 * (Include the two booleans at the end of the reply in dirlen now,
2056 	 *  so we recognize when we have exceeded cnt.)
2057 	 */
2058 	if (nd->nd_flag & ND_NFSV3) {
2059 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2060 		nfsrv_postopattr(nd, getret, &at);
2061 	} else {
2062 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2063 	}
2064 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2065 	txdr_hyper(at.na_filerev, tl);
2066 
2067 	/*
2068 	 * Save this position, in case there is an empty reply needed.
2069 	 */
2070 	mb1 = nd->nd_mb;
2071 	bpos1 = nd->nd_bpos;
2072 
2073 	/* Loop through the records and build reply */
2074 	entrycnt = 0;
2075 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2076 		nlen = dp->d_namlen;
2077 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2078 		    nlen <= NFS_MAXNAMLEN &&
2079 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2080 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2081 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2082 			/*
2083 			 * Save the current position in the reply, in case
2084 			 * this entry exceeds cnt.
2085 			 */
2086 			mb1 = nd->nd_mb;
2087 			bpos1 = nd->nd_bpos;
2088 
2089 			/*
2090 			 * For readdir_and_lookup get the vnode using
2091 			 * the file number.
2092 			 */
2093 			nvp = NULL;
2094 			refp = NULL;
2095 			r = 0;
2096 			at_root = 0;
2097 			needs_unbusy = 0;
2098 			new_mp = mp;
2099 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2100 			if ((nd->nd_flag & ND_NFSV3) ||
2101 			    NFSNONZERO_ATTRBIT(&savbits)) {
2102 				if (nd->nd_flag & ND_NFSV4)
2103 					refp = nfsv4root_getreferral(NULL,
2104 					    vp, dp->d_fileno);
2105 				if (refp == NULL) {
2106 					if (usevget)
2107 						r = VFS_VGET(mp, dp->d_fileno,
2108 						    LK_SHARED, &nvp);
2109 					else
2110 						r = EOPNOTSUPP;
2111 					if (r == EOPNOTSUPP) {
2112 						if (usevget) {
2113 							usevget = 0;
2114 							cn.cn_nameiop = LOOKUP;
2115 							cn.cn_lkflags =
2116 							    LK_SHARED |
2117 							    LK_RETRY;
2118 							cn.cn_cred =
2119 							    nd->nd_cred;
2120 							cn.cn_thread = p;
2121 						}
2122 						cn.cn_nameptr = dp->d_name;
2123 						cn.cn_namelen = nlen;
2124 						cn.cn_flags = ISLASTCN |
2125 						    NOFOLLOW | LOCKLEAF;
2126 						if (nlen == 2 &&
2127 						    dp->d_name[0] == '.' &&
2128 						    dp->d_name[1] == '.')
2129 							cn.cn_flags |=
2130 							    ISDOTDOT;
2131 						if (NFSVOPLOCK(vp, LK_SHARED)
2132 						    != 0) {
2133 							nd->nd_repstat = EPERM;
2134 							break;
2135 						}
2136 						if ((vp->v_vflag & VV_ROOT) != 0
2137 						    && (cn.cn_flags & ISDOTDOT)
2138 						    != 0) {
2139 							vref(vp);
2140 							nvp = vp;
2141 							r = 0;
2142 						} else {
2143 							r = VOP_LOOKUP(vp, &nvp,
2144 							    &cn);
2145 							if (vp != nvp)
2146 								NFSVOPUNLOCK(vp,
2147 								    0);
2148 						}
2149 					}
2150 
2151 					/*
2152 					 * For NFSv4, check to see if nvp is
2153 					 * a mount point and get the mount
2154 					 * point vnode, as required.
2155 					 */
2156 					if (r == 0 &&
2157 					    nfsrv_enable_crossmntpt != 0 &&
2158 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2159 					    nvp->v_type == VDIR &&
2160 					    nvp->v_mountedhere != NULL) {
2161 						new_mp = nvp->v_mountedhere;
2162 						r = vfs_busy(new_mp, 0);
2163 						vput(nvp);
2164 						nvp = NULL;
2165 						if (r == 0) {
2166 							r = VFS_ROOT(new_mp,
2167 							    LK_SHARED, &nvp);
2168 							needs_unbusy = 1;
2169 							if (r == 0)
2170 								at_root = 1;
2171 						}
2172 					}
2173 				}
2174 				if (!r) {
2175 				    if (refp == NULL &&
2176 					((nd->nd_flag & ND_NFSV3) ||
2177 					 NFSNONZERO_ATTRBIT(&attrbits))) {
2178 					r = nfsvno_getfh(nvp, &nfh, p);
2179 					if (!r)
2180 					    r = nfsvno_getattr(nvp, nvap,
2181 						nd->nd_cred, p, 1);
2182 					if (r == 0 && is_zfs == 1 &&
2183 					    nfsrv_enable_crossmntpt != 0 &&
2184 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2185 					    nvp->v_type == VDIR &&
2186 					    vp->v_mount != nvp->v_mount) {
2187 					    /*
2188 					     * For a ZFS snapshot, there is a
2189 					     * pseudo mount that does not set
2190 					     * v_mountedhere, so it needs to
2191 					     * be detected via a different
2192 					     * mount structure.
2193 					     */
2194 					    at_root = 1;
2195 					    if (new_mp == mp)
2196 						new_mp = nvp->v_mount;
2197 					}
2198 				    }
2199 				} else {
2200 				    nvp = NULL;
2201 				}
2202 				if (r) {
2203 					if (!NFSISSET_ATTRBIT(&attrbits,
2204 					    NFSATTRBIT_RDATTRERROR)) {
2205 						if (nvp != NULL)
2206 							vput(nvp);
2207 						if (needs_unbusy != 0)
2208 							vfs_unbusy(new_mp);
2209 						nd->nd_repstat = r;
2210 						break;
2211 					}
2212 				}
2213 			}
2214 
2215 			/*
2216 			 * Build the directory record xdr
2217 			 */
2218 			if (nd->nd_flag & ND_NFSV3) {
2219 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2220 				*tl++ = newnfs_true;
2221 				*tl++ = 0;
2222 				*tl = txdr_unsigned(dp->d_fileno);
2223 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2224 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2225 				*tl++ = 0;
2226 				*tl = txdr_unsigned(*cookiep);
2227 				nfsrv_postopattr(nd, 0, nvap);
2228 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2229 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2230 				if (nvp != NULL)
2231 					vput(nvp);
2232 			} else {
2233 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2234 				*tl++ = newnfs_true;
2235 				*tl++ = 0;
2236 				*tl = txdr_unsigned(*cookiep);
2237 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2238 				if (nvp != NULL) {
2239 					supports_nfsv4acls =
2240 					    nfs_supportsnfsv4acls(nvp);
2241 					NFSVOPUNLOCK(nvp, 0);
2242 				} else
2243 					supports_nfsv4acls = 0;
2244 				if (refp != NULL) {
2245 					dirlen += nfsrv_putreferralattr(nd,
2246 					    &savbits, refp, 0,
2247 					    &nd->nd_repstat);
2248 					if (nd->nd_repstat) {
2249 						if (nvp != NULL)
2250 							vrele(nvp);
2251 						if (needs_unbusy != 0)
2252 							vfs_unbusy(new_mp);
2253 						break;
2254 					}
2255 				} else if (r) {
2256 					dirlen += nfsvno_fillattr(nd, new_mp,
2257 					    nvp, nvap, &nfh, r, &rderrbits,
2258 					    nd->nd_cred, p, isdgram, 0,
2259 					    supports_nfsv4acls, at_root,
2260 					    mounted_on_fileno);
2261 				} else {
2262 					dirlen += nfsvno_fillattr(nd, new_mp,
2263 					    nvp, nvap, &nfh, r, &attrbits,
2264 					    nd->nd_cred, p, isdgram, 0,
2265 					    supports_nfsv4acls, at_root,
2266 					    mounted_on_fileno);
2267 				}
2268 				if (nvp != NULL)
2269 					vrele(nvp);
2270 				dirlen += (3 * NFSX_UNSIGNED);
2271 			}
2272 			if (needs_unbusy != 0)
2273 				vfs_unbusy(new_mp);
2274 			if (dirlen <= cnt)
2275 				entrycnt++;
2276 		}
2277 		cpos += dp->d_reclen;
2278 		dp = (struct dirent *)cpos;
2279 		cookiep++;
2280 		ncookies--;
2281 	}
2282 	vrele(vp);
2283 	vfs_unbusy(mp);
2284 
2285 	/*
2286 	 * If dirlen > cnt, we must strip off the last entry. If that
2287 	 * results in an empty reply, report NFSERR_TOOSMALL.
2288 	 */
2289 	if (dirlen > cnt || nd->nd_repstat) {
2290 		if (!nd->nd_repstat && entrycnt == 0)
2291 			nd->nd_repstat = NFSERR_TOOSMALL;
2292 		if (nd->nd_repstat) {
2293 			newnfs_trimtrailing(nd, mb0, bpos0);
2294 			if (nd->nd_flag & ND_NFSV3)
2295 				nfsrv_postopattr(nd, getret, &at);
2296 		} else
2297 			newnfs_trimtrailing(nd, mb1, bpos1);
2298 		eofflag = 0;
2299 	} else if (cpos < cend)
2300 		eofflag = 0;
2301 	if (!nd->nd_repstat) {
2302 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2303 		*tl++ = newnfs_false;
2304 		if (eofflag)
2305 			*tl = newnfs_true;
2306 		else
2307 			*tl = newnfs_false;
2308 	}
2309 	FREE((caddr_t)cookies, M_TEMP);
2310 	FREE((caddr_t)rbuf, M_TEMP);
2311 
2312 out:
2313 	NFSEXITCODE2(0, nd);
2314 	return (0);
2315 nfsmout:
2316 	vput(vp);
2317 	NFSEXITCODE2(error, nd);
2318 	return (error);
2319 }
2320 
2321 /*
2322  * Get the settable attributes out of the mbuf list.
2323  * (Return 0 or EBADRPC)
2324  */
2325 int
2326 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2327     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2328 {
2329 	u_int32_t *tl;
2330 	struct nfsv2_sattr *sp;
2331 	int error = 0, toclient = 0;
2332 
2333 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2334 	case ND_NFSV2:
2335 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2336 		/*
2337 		 * Some old clients didn't fill in the high order 16bits.
2338 		 * --> check the low order 2 bytes for 0xffff
2339 		 */
2340 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2341 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2342 		if (sp->sa_uid != newnfs_xdrneg1)
2343 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2344 		if (sp->sa_gid != newnfs_xdrneg1)
2345 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2346 		if (sp->sa_size != newnfs_xdrneg1)
2347 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2348 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2349 #ifdef notyet
2350 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2351 #else
2352 			nvap->na_atime.tv_sec =
2353 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2354 			nvap->na_atime.tv_nsec = 0;
2355 #endif
2356 		}
2357 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2358 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2359 		break;
2360 	case ND_NFSV3:
2361 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2362 		if (*tl == newnfs_true) {
2363 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2364 			nvap->na_mode = nfstov_mode(*tl);
2365 		}
2366 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2367 		if (*tl == newnfs_true) {
2368 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2369 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2370 		}
2371 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2372 		if (*tl == newnfs_true) {
2373 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2374 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2375 		}
2376 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2377 		if (*tl == newnfs_true) {
2378 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2379 			nvap->na_size = fxdr_hyper(tl);
2380 		}
2381 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2382 		switch (fxdr_unsigned(int, *tl)) {
2383 		case NFSV3SATTRTIME_TOCLIENT:
2384 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2385 			fxdr_nfsv3time(tl, &nvap->na_atime);
2386 			toclient = 1;
2387 			break;
2388 		case NFSV3SATTRTIME_TOSERVER:
2389 			vfs_timestamp(&nvap->na_atime);
2390 			nvap->na_vaflags |= VA_UTIMES_NULL;
2391 			break;
2392 		}
2393 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2394 		switch (fxdr_unsigned(int, *tl)) {
2395 		case NFSV3SATTRTIME_TOCLIENT:
2396 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2397 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2398 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2399 			break;
2400 		case NFSV3SATTRTIME_TOSERVER:
2401 			vfs_timestamp(&nvap->na_mtime);
2402 			if (!toclient)
2403 				nvap->na_vaflags |= VA_UTIMES_NULL;
2404 			break;
2405 		}
2406 		break;
2407 	case ND_NFSV4:
2408 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2409 	}
2410 nfsmout:
2411 	NFSEXITCODE2(error, nd);
2412 	return (error);
2413 }
2414 
2415 /*
2416  * Handle the setable attributes for V4.
2417  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2418  */
2419 int
2420 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2421     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2422 {
2423 	u_int32_t *tl;
2424 	int attrsum = 0;
2425 	int i, j;
2426 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2427 	int toclient = 0;
2428 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2429 	uid_t uid;
2430 	gid_t gid;
2431 
2432 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2433 	if (error)
2434 		goto nfsmout;
2435 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2436 	attrsize = fxdr_unsigned(int, *tl);
2437 
2438 	/*
2439 	 * Loop around getting the setable attributes. If an unsupported
2440 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2441 	 */
2442 	if (retnotsup) {
2443 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2444 		bitpos = NFSATTRBIT_MAX;
2445 	} else {
2446 		bitpos = 0;
2447 	}
2448 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2449 	    if (attrsum > attrsize) {
2450 		error = NFSERR_BADXDR;
2451 		goto nfsmout;
2452 	    }
2453 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2454 		switch (bitpos) {
2455 		case NFSATTRBIT_SIZE:
2456 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2457                      if (vp != NULL && vp->v_type != VREG) {
2458                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2459                                 NFSERR_INVAL;
2460                             goto nfsmout;
2461 			}
2462 			nvap->na_size = fxdr_hyper(tl);
2463 			attrsum += NFSX_HYPER;
2464 			break;
2465 		case NFSATTRBIT_ACL:
2466 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2467 			    p);
2468 			if (error)
2469 				goto nfsmout;
2470 			if (aceerr && !nd->nd_repstat)
2471 				nd->nd_repstat = aceerr;
2472 			attrsum += aclsize;
2473 			break;
2474 		case NFSATTRBIT_ARCHIVE:
2475 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2476 			if (!nd->nd_repstat)
2477 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2478 			attrsum += NFSX_UNSIGNED;
2479 			break;
2480 		case NFSATTRBIT_HIDDEN:
2481 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2482 			if (!nd->nd_repstat)
2483 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2484 			attrsum += NFSX_UNSIGNED;
2485 			break;
2486 		case NFSATTRBIT_MIMETYPE:
2487 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2488 			i = fxdr_unsigned(int, *tl);
2489 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2490 			if (error)
2491 				goto nfsmout;
2492 			if (!nd->nd_repstat)
2493 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2494 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2495 			break;
2496 		case NFSATTRBIT_MODE:
2497 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2498 			nvap->na_mode = nfstov_mode(*tl);
2499 			attrsum += NFSX_UNSIGNED;
2500 			break;
2501 		case NFSATTRBIT_OWNER:
2502 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2503 			j = fxdr_unsigned(int, *tl);
2504 			if (j < 0) {
2505 				error = NFSERR_BADXDR;
2506 				goto nfsmout;
2507 			}
2508 			if (j > NFSV4_SMALLSTR)
2509 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2510 			else
2511 				cp = namestr;
2512 			error = nfsrv_mtostr(nd, cp, j);
2513 			if (error) {
2514 				if (j > NFSV4_SMALLSTR)
2515 					free(cp, M_NFSSTRING);
2516 				goto nfsmout;
2517 			}
2518 			if (!nd->nd_repstat) {
2519 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2520 				    p);
2521 				if (!nd->nd_repstat)
2522 					nvap->na_uid = uid;
2523 			}
2524 			if (j > NFSV4_SMALLSTR)
2525 				free(cp, M_NFSSTRING);
2526 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2527 			break;
2528 		case NFSATTRBIT_OWNERGROUP:
2529 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2530 			j = fxdr_unsigned(int, *tl);
2531 			if (j < 0) {
2532 				error = NFSERR_BADXDR;
2533 				goto nfsmout;
2534 			}
2535 			if (j > NFSV4_SMALLSTR)
2536 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2537 			else
2538 				cp = namestr;
2539 			error = nfsrv_mtostr(nd, cp, j);
2540 			if (error) {
2541 				if (j > NFSV4_SMALLSTR)
2542 					free(cp, M_NFSSTRING);
2543 				goto nfsmout;
2544 			}
2545 			if (!nd->nd_repstat) {
2546 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2547 				    p);
2548 				if (!nd->nd_repstat)
2549 					nvap->na_gid = gid;
2550 			}
2551 			if (j > NFSV4_SMALLSTR)
2552 				free(cp, M_NFSSTRING);
2553 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2554 			break;
2555 		case NFSATTRBIT_SYSTEM:
2556 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2557 			if (!nd->nd_repstat)
2558 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2559 			attrsum += NFSX_UNSIGNED;
2560 			break;
2561 		case NFSATTRBIT_TIMEACCESSSET:
2562 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2563 			attrsum += NFSX_UNSIGNED;
2564 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2565 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2566 			    fxdr_nfsv4time(tl, &nvap->na_atime);
2567 			    toclient = 1;
2568 			    attrsum += NFSX_V4TIME;
2569 			} else {
2570 			    vfs_timestamp(&nvap->na_atime);
2571 			    nvap->na_vaflags |= VA_UTIMES_NULL;
2572 			}
2573 			break;
2574 		case NFSATTRBIT_TIMEBACKUP:
2575 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2576 			if (!nd->nd_repstat)
2577 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2578 			attrsum += NFSX_V4TIME;
2579 			break;
2580 		case NFSATTRBIT_TIMECREATE:
2581 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2582 			if (!nd->nd_repstat)
2583 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2584 			attrsum += NFSX_V4TIME;
2585 			break;
2586 		case NFSATTRBIT_TIMEMODIFYSET:
2587 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2588 			attrsum += NFSX_UNSIGNED;
2589 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2590 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2591 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
2592 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
2593 			    attrsum += NFSX_V4TIME;
2594 			} else {
2595 			    vfs_timestamp(&nvap->na_mtime);
2596 			    if (!toclient)
2597 				nvap->na_vaflags |= VA_UTIMES_NULL;
2598 			}
2599 			break;
2600 		default:
2601 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2602 			/*
2603 			 * set bitpos so we drop out of the loop.
2604 			 */
2605 			bitpos = NFSATTRBIT_MAX;
2606 			break;
2607 		}
2608 	}
2609 
2610 	/*
2611 	 * some clients pad the attrlist, so we need to skip over the
2612 	 * padding.
2613 	 */
2614 	if (attrsum > attrsize) {
2615 		error = NFSERR_BADXDR;
2616 	} else {
2617 		attrsize = NFSM_RNDUP(attrsize);
2618 		if (attrsum < attrsize)
2619 			error = nfsm_advance(nd, attrsize - attrsum, -1);
2620 	}
2621 nfsmout:
2622 	NFSEXITCODE2(error, nd);
2623 	return (error);
2624 }
2625 
2626 /*
2627  * Check/setup export credentials.
2628  */
2629 int
2630 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2631     struct ucred *credanon)
2632 {
2633 	int error = 0;
2634 
2635 	/*
2636 	 * Check/setup credentials.
2637 	 */
2638 	if (nd->nd_flag & ND_GSS)
2639 		exp->nes_exflag &= ~MNT_EXPORTANON;
2640 
2641 	/*
2642 	 * Check to see if the operation is allowed for this security flavor.
2643 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2644 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2645 	 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2646 	 */
2647 	if (nfsvno_testexp(nd, exp) &&
2648 	    nd->nd_procnum != NFSV4OP_SECINFO &&
2649 	    nd->nd_procnum != NFSPROC_FSINFO) {
2650 		if (nd->nd_flag & ND_NFSV4)
2651 			error = NFSERR_WRONGSEC;
2652 		else
2653 			error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2654 		goto out;
2655 	}
2656 
2657 	/*
2658 	 * Check to see if the file system is exported V4 only.
2659 	 */
2660 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2661 		error = NFSERR_PROGNOTV4;
2662 		goto out;
2663 	}
2664 
2665 	/*
2666 	 * Now, map the user credentials.
2667 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
2668 	 *  Fsinfo RPC. If set for anything else, this code might need
2669 	 *  to change.)
2670 	 */
2671 	if (NFSVNO_EXPORTED(exp)) {
2672 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
2673 		     NFSVNO_EXPORTANON(exp) ||
2674 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
2675 			nd->nd_cred->cr_uid = credanon->cr_uid;
2676 			nd->nd_cred->cr_gid = credanon->cr_gid;
2677 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2678 			    credanon->cr_groups);
2679 		} else if ((nd->nd_flag & ND_GSS) == 0) {
2680 			/*
2681 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
2682 			 * if there is a replacement credential with a group
2683 			 * list set up by "nfsuserd -manage-gids".
2684 			 * If there is no replacement, nfsrv_getgrpscred()
2685 			 * simply returns its argument.
2686 			 */
2687 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
2688 		}
2689 	}
2690 
2691 out:
2692 	NFSEXITCODE2(error, nd);
2693 	return (error);
2694 }
2695 
2696 /*
2697  * Check exports.
2698  */
2699 int
2700 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2701     struct ucred **credp)
2702 {
2703 	int i, error, *secflavors;
2704 
2705 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2706 	    &exp->nes_numsecflavor, &secflavors);
2707 	if (error) {
2708 		if (nfs_rootfhset) {
2709 			exp->nes_exflag = 0;
2710 			exp->nes_numsecflavor = 0;
2711 			error = 0;
2712 		}
2713 	} else {
2714 		/* Copy the security flavors. */
2715 		for (i = 0; i < exp->nes_numsecflavor; i++)
2716 			exp->nes_secflavors[i] = secflavors[i];
2717 	}
2718 	NFSEXITCODE(error);
2719 	return (error);
2720 }
2721 
2722 /*
2723  * Get a vnode for a file handle and export stuff.
2724  */
2725 int
2726 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2727     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2728     struct ucred **credp)
2729 {
2730 	int i, error, *secflavors;
2731 
2732 	*credp = NULL;
2733 	exp->nes_numsecflavor = 0;
2734 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2735 	if (error != 0)
2736 		/* Make sure the server replies ESTALE to the client. */
2737 		error = ESTALE;
2738 	if (nam && !error) {
2739 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2740 		    &exp->nes_numsecflavor, &secflavors);
2741 		if (error) {
2742 			if (nfs_rootfhset) {
2743 				exp->nes_exflag = 0;
2744 				exp->nes_numsecflavor = 0;
2745 				error = 0;
2746 			} else {
2747 				vput(*vpp);
2748 			}
2749 		} else {
2750 			/* Copy the security flavors. */
2751 			for (i = 0; i < exp->nes_numsecflavor; i++)
2752 				exp->nes_secflavors[i] = secflavors[i];
2753 		}
2754 	}
2755 	NFSEXITCODE(error);
2756 	return (error);
2757 }
2758 
2759 /*
2760  * nfsd_fhtovp() - convert a fh to a vnode ptr
2761  * 	- look up fsid in mount list (if not found ret error)
2762  *	- get vp and export rights by calling nfsvno_fhtovp()
2763  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2764  *	  for AUTH_SYS
2765  *	- if mpp != NULL, return the mount point so that it can
2766  *	  be used for vn_finished_write() by the caller
2767  */
2768 void
2769 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2770     struct vnode **vpp, struct nfsexstuff *exp,
2771     struct mount **mpp, int startwrite, struct thread *p)
2772 {
2773 	struct mount *mp;
2774 	struct ucred *credanon;
2775 	fhandle_t *fhp;
2776 
2777 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
2778 	/*
2779 	 * Check for the special case of the nfsv4root_fh.
2780 	 */
2781 	mp = vfs_busyfs(&fhp->fh_fsid);
2782 	if (mpp != NULL)
2783 		*mpp = mp;
2784 	if (mp == NULL) {
2785 		*vpp = NULL;
2786 		nd->nd_repstat = ESTALE;
2787 		goto out;
2788 	}
2789 
2790 	if (startwrite) {
2791 		vn_start_write(NULL, mpp, V_WAIT);
2792 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
2793 			lktype = LK_EXCLUSIVE;
2794 	}
2795 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2796 	    &credanon);
2797 	vfs_unbusy(mp);
2798 
2799 	/*
2800 	 * For NFSv4 without a pseudo root fs, unexported file handles
2801 	 * can be returned, so that Lookup works everywhere.
2802 	 */
2803 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2804 	    !(nd->nd_flag & ND_NFSV4)) {
2805 		vput(*vpp);
2806 		nd->nd_repstat = EACCES;
2807 	}
2808 
2809 	/*
2810 	 * Personally, I've never seen any point in requiring a
2811 	 * reserved port#, since only in the rare case where the
2812 	 * clients are all boxes with secure system privileges,
2813 	 * does it provide any enhanced security, but... some people
2814 	 * believe it to be useful and keep putting this code back in.
2815 	 * (There is also some "security checker" out there that
2816 	 *  complains if the nfs server doesn't enforce this.)
2817 	 * However, note the following:
2818 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
2819 	 *	required.
2820 	 * RFC2623 recommends that, if a reserved port# is checked for,
2821 	 *	that there be a way to turn that off--> ifdef'd.
2822 	 */
2823 #ifdef NFS_REQRSVPORT
2824 	if (!nd->nd_repstat) {
2825 		struct sockaddr_in *saddr;
2826 		struct sockaddr_in6 *saddr6;
2827 
2828 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2829 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2830 		if (!(nd->nd_flag & ND_NFSV4) &&
2831 		    ((saddr->sin_family == AF_INET &&
2832 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2833 		     (saddr6->sin6_family == AF_INET6 &&
2834 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2835 			vput(*vpp);
2836 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2837 		}
2838 	}
2839 #endif	/* NFS_REQRSVPORT */
2840 
2841 	/*
2842 	 * Check/setup credentials.
2843 	 */
2844 	if (!nd->nd_repstat) {
2845 		nd->nd_saveduid = nd->nd_cred->cr_uid;
2846 		nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2847 		if (nd->nd_repstat)
2848 			vput(*vpp);
2849 	}
2850 	if (credanon != NULL)
2851 		crfree(credanon);
2852 	if (nd->nd_repstat) {
2853 		if (startwrite)
2854 			vn_finished_write(mp);
2855 		*vpp = NULL;
2856 		if (mpp != NULL)
2857 			*mpp = NULL;
2858 	}
2859 
2860 out:
2861 	NFSEXITCODE2(0, nd);
2862 }
2863 
2864 /*
2865  * glue for fp.
2866  */
2867 static int
2868 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2869 {
2870 	struct filedesc *fdp;
2871 	struct file *fp;
2872 	int error = 0;
2873 
2874 	fdp = p->td_proc->p_fd;
2875 	if (fd < 0 || fd >= fdp->fd_nfiles ||
2876 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
2877 		error = EBADF;
2878 		goto out;
2879 	}
2880 	*fpp = fp;
2881 
2882 out:
2883 	NFSEXITCODE(error);
2884 	return (error);
2885 }
2886 
2887 /*
2888  * Called from nfssvc() to update the exports list. Just call
2889  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2890  * in the mount list.
2891  */
2892 int
2893 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2894 {
2895 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2896 	int error = 0;
2897 	struct nameidata nd;
2898 	fhandle_t fh;
2899 
2900 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2901 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2902 		nfs_rootfhset = 0;
2903 	else if (error == 0) {
2904 		if (nfsexargp->fspec == NULL) {
2905 			error = EPERM;
2906 			goto out;
2907 		}
2908 		/*
2909 		 * If fspec != NULL, this is the v4root path.
2910 		 */
2911 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
2912 		    nfsexargp->fspec, p);
2913 		if ((error = namei(&nd)) != 0)
2914 			goto out;
2915 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
2916 		vrele(nd.ni_vp);
2917 		if (!error) {
2918 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2919 			NFSBCOPY((caddr_t)&fh,
2920 			    nfs_rootfh.nfsrvfh_data,
2921 			    sizeof (fhandle_t));
2922 			nfs_rootfhset = 1;
2923 		}
2924 	}
2925 
2926 out:
2927 	NFSEXITCODE(error);
2928 	return (error);
2929 }
2930 
2931 /*
2932  * This function needs to test to see if the system is near its limit
2933  * for memory allocation via malloc() or mget() and return True iff
2934  * either of these resources are near their limit.
2935  * XXX (For now, this is just a stub.)
2936  */
2937 int nfsrv_testmalloclimit = 0;
2938 int
2939 nfsrv_mallocmget_limit(void)
2940 {
2941 	static int printmesg = 0;
2942 	static int testval = 1;
2943 
2944 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2945 		if ((printmesg++ % 100) == 0)
2946 			printf("nfsd: malloc/mget near limit\n");
2947 		return (1);
2948 	}
2949 	return (0);
2950 }
2951 
2952 /*
2953  * BSD specific initialization of a mount point.
2954  */
2955 void
2956 nfsd_mntinit(void)
2957 {
2958 	static int inited = 0;
2959 
2960 	if (inited)
2961 		return;
2962 	inited = 1;
2963 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2964 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2965 	TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2966 	nfsv4root_mnt.mnt_export = NULL;
2967 	TAILQ_INIT(&nfsv4root_opt);
2968 	TAILQ_INIT(&nfsv4root_newopt);
2969 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2970 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2971 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
2972 	nfsv4root_mnt.mnt_activevnodelistsize = 0;
2973 }
2974 
2975 /*
2976  * Get a vnode for a file handle, without checking exports, etc.
2977  */
2978 struct vnode *
2979 nfsvno_getvp(fhandle_t *fhp)
2980 {
2981 	struct mount *mp;
2982 	struct vnode *vp;
2983 	int error;
2984 
2985 	mp = vfs_busyfs(&fhp->fh_fsid);
2986 	if (mp == NULL)
2987 		return (NULL);
2988 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2989 	vfs_unbusy(mp);
2990 	if (error)
2991 		return (NULL);
2992 	return (vp);
2993 }
2994 
2995 /*
2996  * Do a local VOP_ADVLOCK().
2997  */
2998 int
2999 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3000     u_int64_t end, struct thread *td)
3001 {
3002 	int error = 0;
3003 	struct flock fl;
3004 	u_int64_t tlen;
3005 
3006 	if (nfsrv_dolocallocks == 0)
3007 		goto out;
3008 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3009 
3010 	fl.l_whence = SEEK_SET;
3011 	fl.l_type = ftype;
3012 	fl.l_start = (off_t)first;
3013 	if (end == NFS64BITSSET) {
3014 		fl.l_len = 0;
3015 	} else {
3016 		tlen = end - first;
3017 		fl.l_len = (off_t)tlen;
3018 	}
3019 	/*
3020 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3021 	 * values for all calls, so that all locks will be held by the
3022 	 * nfsd server. (The nfsd server handles conflicts between the
3023 	 * various clients.)
3024 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3025 	 * bytes, so it can't be put in l_sysid.
3026 	 */
3027 	if (nfsv4_sysid == 0)
3028 		nfsv4_sysid = nlm_acquire_next_sysid();
3029 	fl.l_pid = (pid_t)0;
3030 	fl.l_sysid = (int)nfsv4_sysid;
3031 
3032 	if (ftype == F_UNLCK)
3033 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3034 		    (F_POSIX | F_REMOTE));
3035 	else
3036 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3037 		    (F_POSIX | F_REMOTE));
3038 
3039 out:
3040 	NFSEXITCODE(error);
3041 	return (error);
3042 }
3043 
3044 /*
3045  * Check the nfsv4 root exports.
3046  */
3047 int
3048 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3049 {
3050 	struct ucred *credanon;
3051 	int exflags, error = 0, numsecflavor, *secflavors, i;
3052 
3053 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3054 	    &credanon, &numsecflavor, &secflavors);
3055 	if (error) {
3056 		error = NFSERR_PROGUNAVAIL;
3057 		goto out;
3058 	}
3059 	if (credanon != NULL)
3060 		crfree(credanon);
3061 	for (i = 0; i < numsecflavor; i++) {
3062 		if (secflavors[i] == AUTH_SYS)
3063 			nd->nd_flag |= ND_EXAUTHSYS;
3064 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3065 			nd->nd_flag |= ND_EXGSS;
3066 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3067 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3068 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3069 			nd->nd_flag |= ND_EXGSSPRIVACY;
3070 	}
3071 
3072 out:
3073 	NFSEXITCODE(error);
3074 	return (error);
3075 }
3076 
3077 /*
3078  * Nfs server pseudo system call for the nfsd's
3079  */
3080 /*
3081  * MPSAFE
3082  */
3083 static int
3084 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3085 {
3086 	struct file *fp;
3087 	struct nfsd_addsock_args sockarg;
3088 	struct nfsd_nfsd_args nfsdarg;
3089 	cap_rights_t rights;
3090 	int error;
3091 
3092 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3093 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3094 		if (error)
3095 			goto out;
3096 		/*
3097 		 * Since we don't know what rights might be required,
3098 		 * pretend that we need them all. It is better to be too
3099 		 * careful than too reckless.
3100 		 */
3101 		error = fget(td, sockarg.sock,
3102 		    cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3103 		if (error != 0)
3104 			goto out;
3105 		if (fp->f_type != DTYPE_SOCKET) {
3106 			fdrop(fp, td);
3107 			error = EPERM;
3108 			goto out;
3109 		}
3110 		error = nfsrvd_addsock(fp);
3111 		fdrop(fp, td);
3112 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3113 		if (uap->argp == NULL) {
3114 			error = EINVAL;
3115 			goto out;
3116 		}
3117 		error = copyin(uap->argp, (caddr_t)&nfsdarg,
3118 		    sizeof (nfsdarg));
3119 		if (error)
3120 			goto out;
3121 		error = nfsrvd_nfsd(td, &nfsdarg);
3122 	} else {
3123 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3124 	}
3125 
3126 out:
3127 	NFSEXITCODE(error);
3128 	return (error);
3129 }
3130 
3131 static int
3132 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3133 {
3134 	struct nfsex_args export;
3135 	struct file *fp = NULL;
3136 	int stablefd, len;
3137 	struct nfsd_clid adminrevoke;
3138 	struct nfsd_dumplist dumplist;
3139 	struct nfsd_dumpclients *dumpclients;
3140 	struct nfsd_dumplocklist dumplocklist;
3141 	struct nfsd_dumplocks *dumplocks;
3142 	struct nameidata nd;
3143 	vnode_t vp;
3144 	int error = EINVAL, igotlock;
3145 	struct proc *procp;
3146 	static int suspend_nfsd = 0;
3147 
3148 	if (uap->flag & NFSSVC_PUBLICFH) {
3149 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3150 		    sizeof (fhandle_t));
3151 		error = copyin(uap->argp,
3152 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3153 		if (!error)
3154 			nfs_pubfhset = 1;
3155 	} else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3156 		error = copyin(uap->argp,(caddr_t)&export,
3157 		    sizeof (struct nfsex_args));
3158 		if (!error)
3159 			error = nfsrv_v4rootexport(&export, cred, p);
3160 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3161 		nfs_pubfhset = 0;
3162 		error = 0;
3163 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3164 		error = copyin(uap->argp, (caddr_t)&stablefd,
3165 		    sizeof (int));
3166 		if (!error)
3167 			error = fp_getfvp(p, stablefd, &fp, &vp);
3168 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3169 			error = EBADF;
3170 		if (!error && newnfs_numnfsd != 0)
3171 			error = EPERM;
3172 		if (!error) {
3173 			nfsrv_stablefirst.nsf_fp = fp;
3174 			nfsrv_setupstable(p);
3175 		}
3176 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3177 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3178 		    sizeof (struct nfsd_clid));
3179 		if (!error)
3180 			error = nfsrv_adminrevoke(&adminrevoke, p);
3181 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3182 		error = copyin(uap->argp, (caddr_t)&dumplist,
3183 		    sizeof (struct nfsd_dumplist));
3184 		if (!error && (dumplist.ndl_size < 1 ||
3185 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3186 			error = EPERM;
3187 		if (!error) {
3188 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3189 		    dumpclients = (struct nfsd_dumpclients *)malloc(len,
3190 			M_TEMP, M_WAITOK);
3191 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3192 		    error = copyout(dumpclients,
3193 			CAST_USER_ADDR_T(dumplist.ndl_list), len);
3194 		    free((caddr_t)dumpclients, M_TEMP);
3195 		}
3196 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3197 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3198 		    sizeof (struct nfsd_dumplocklist));
3199 		if (!error && (dumplocklist.ndllck_size < 1 ||
3200 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3201 			error = EPERM;
3202 		if (!error)
3203 			error = nfsrv_lookupfilename(&nd,
3204 				dumplocklist.ndllck_fname, p);
3205 		if (!error) {
3206 			len = sizeof (struct nfsd_dumplocks) *
3207 				dumplocklist.ndllck_size;
3208 			dumplocks = (struct nfsd_dumplocks *)malloc(len,
3209 				M_TEMP, M_WAITOK);
3210 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3211 			    dumplocklist.ndllck_size, p);
3212 			vput(nd.ni_vp);
3213 			error = copyout(dumplocks,
3214 			    CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3215 			free((caddr_t)dumplocks, M_TEMP);
3216 		}
3217 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3218 		procp = p->td_proc;
3219 		PROC_LOCK(procp);
3220 		nfsd_master_pid = procp->p_pid;
3221 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3222 		nfsd_master_start = procp->p_stats->p_start;
3223 		nfsd_master_proc = procp;
3224 		PROC_UNLOCK(procp);
3225 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3226 		NFSLOCKV4ROOTMUTEX();
3227 		if (suspend_nfsd == 0) {
3228 			/* Lock out all nfsd threads */
3229 			do {
3230 				igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3231 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3232 			} while (igotlock == 0 && suspend_nfsd == 0);
3233 			suspend_nfsd = 1;
3234 		}
3235 		NFSUNLOCKV4ROOTMUTEX();
3236 		error = 0;
3237 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3238 		NFSLOCKV4ROOTMUTEX();
3239 		if (suspend_nfsd != 0) {
3240 			nfsv4_unlock(&nfsd_suspend_lock, 0);
3241 			suspend_nfsd = 0;
3242 		}
3243 		NFSUNLOCKV4ROOTMUTEX();
3244 		error = 0;
3245 	}
3246 
3247 	NFSEXITCODE(error);
3248 	return (error);
3249 }
3250 
3251 /*
3252  * Check exports.
3253  * Returns 0 if ok, 1 otherwise.
3254  */
3255 int
3256 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3257 {
3258 	int i;
3259 
3260 	/*
3261 	 * This seems odd, but allow the case where the security flavor
3262 	 * list is empty. This happens when NFSv4 is traversing non-exported
3263 	 * file systems. Exported file systems should always have a non-empty
3264 	 * security flavor list.
3265 	 */
3266 	if (exp->nes_numsecflavor == 0)
3267 		return (0);
3268 
3269 	for (i = 0; i < exp->nes_numsecflavor; i++) {
3270 		/*
3271 		 * The tests for privacy and integrity must be first,
3272 		 * since ND_GSS is set for everything but AUTH_SYS.
3273 		 */
3274 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3275 		    (nd->nd_flag & ND_GSSPRIVACY))
3276 			return (0);
3277 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3278 		    (nd->nd_flag & ND_GSSINTEGRITY))
3279 			return (0);
3280 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3281 		    (nd->nd_flag & ND_GSS))
3282 			return (0);
3283 		if (exp->nes_secflavors[i] == AUTH_SYS &&
3284 		    (nd->nd_flag & ND_GSS) == 0)
3285 			return (0);
3286 	}
3287 	return (1);
3288 }
3289 
3290 /*
3291  * Calculate a hash value for the fid in a file handle.
3292  */
3293 uint32_t
3294 nfsrv_hashfh(fhandle_t *fhp)
3295 {
3296 	uint32_t hashval;
3297 
3298 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3299 	return (hashval);
3300 }
3301 
3302 /*
3303  * Calculate a hash value for the sessionid.
3304  */
3305 uint32_t
3306 nfsrv_hashsessionid(uint8_t *sessionid)
3307 {
3308 	uint32_t hashval;
3309 
3310 	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
3311 	return (hashval);
3312 }
3313 
3314 /*
3315  * Signal the userland master nfsd to backup the stable restart file.
3316  */
3317 void
3318 nfsrv_backupstable(void)
3319 {
3320 	struct proc *procp;
3321 
3322 	if (nfsd_master_proc != NULL) {
3323 		procp = pfind(nfsd_master_pid);
3324 		/* Try to make sure it is the correct process. */
3325 		if (procp == nfsd_master_proc &&
3326 		    procp->p_stats->p_start.tv_sec ==
3327 		    nfsd_master_start.tv_sec &&
3328 		    procp->p_stats->p_start.tv_usec ==
3329 		    nfsd_master_start.tv_usec &&
3330 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
3331 			kern_psignal(procp, SIGUSR2);
3332 		else
3333 			nfsd_master_proc = NULL;
3334 
3335 		if (procp != NULL)
3336 			PROC_UNLOCK(procp);
3337 	}
3338 }
3339 
3340 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3341 
3342 /*
3343  * Called once to initialize data structures...
3344  */
3345 static int
3346 nfsd_modevent(module_t mod, int type, void *data)
3347 {
3348 	int error = 0, i;
3349 	static int loaded = 0;
3350 
3351 	switch (type) {
3352 	case MOD_LOAD:
3353 		if (loaded)
3354 			goto out;
3355 		newnfs_portinit();
3356 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3357 			mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
3358 			    MTX_DEF);
3359 			mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
3360 			    MTX_DEF);
3361 		}
3362 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
3363 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
3364 		mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
3365 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3366 		nfsrvd_initcache();
3367 		nfsd_init();
3368 		NFSD_LOCK();
3369 		nfsrvd_init(0);
3370 		NFSD_UNLOCK();
3371 		nfsd_mntinit();
3372 #ifdef VV_DISABLEDELEG
3373 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3374 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3375 #endif
3376 		nfsd_call_servertimer = nfsrv_servertimer;
3377 		nfsd_call_nfsd = nfssvc_nfsd;
3378 		loaded = 1;
3379 		break;
3380 
3381 	case MOD_UNLOAD:
3382 		if (newnfs_numnfsd != 0) {
3383 			error = EBUSY;
3384 			break;
3385 		}
3386 
3387 #ifdef VV_DISABLEDELEG
3388 		vn_deleg_ops.vndeleg_recall = NULL;
3389 		vn_deleg_ops.vndeleg_disable = NULL;
3390 #endif
3391 		nfsd_call_servertimer = NULL;
3392 		nfsd_call_nfsd = NULL;
3393 
3394 		/* Clean out all NFSv4 state. */
3395 		nfsrv_throwawayallstate(curthread);
3396 
3397 		/* Clean the NFS server reply cache */
3398 		nfsrvd_cleancache();
3399 
3400 		/* Free up the krpc server pool. */
3401 		if (nfsrvd_pool != NULL)
3402 			svcpool_destroy(nfsrvd_pool);
3403 
3404 		/* and get rid of the locks */
3405 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3406 			mtx_destroy(&nfsrchash_table[i].mtx);
3407 			mtx_destroy(&nfsrcahash_table[i].mtx);
3408 		}
3409 		mtx_destroy(&nfsrc_udpmtx);
3410 		mtx_destroy(&nfs_v4root_mutex);
3411 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3412 		for (i = 0; i < nfsrv_sessionhashsize; i++)
3413 			mtx_destroy(&nfssessionhash[i].mtx);
3414 		lockdestroy(&nfsv4root_mnt.mnt_explock);
3415 		free(nfsclienthash, M_NFSDCLIENT);
3416 		free(nfslockhash, M_NFSDLOCKFILE);
3417 		free(nfssessionhash, M_NFSDSESSION);
3418 		loaded = 0;
3419 		break;
3420 	default:
3421 		error = EOPNOTSUPP;
3422 		break;
3423 	}
3424 
3425 out:
3426 	NFSEXITCODE(error);
3427 	return (error);
3428 }
3429 static moduledata_t nfsd_mod = {
3430 	"nfsd",
3431 	nfsd_modevent,
3432 	NULL,
3433 };
3434 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3435 
3436 /* So that loader and kldload(2) can find us, wherever we are.. */
3437 MODULE_VERSION(nfsd, 1);
3438 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3439 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3440 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3441 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3442 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3443 
3444