xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 /*
38  * Functions that perform the vfs operations required by the routines in
39  * nfsd_serv.c. It is hoped that this change will make the server more
40  * portable.
41  */
42 
43 #include <fs/nfs/nfsport.h>
44 #include <sys/hash.h>
45 #include <sys/sysctl.h>
46 #include <nlm/nlm_prot.h>
47 #include <nlm/nlm.h>
48 
49 FEATURE(nfsd, "NFSv4 server");
50 
51 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
52 extern int nfsrv_useacl;
53 extern int newnfs_numnfsd;
54 extern struct mount nfsv4root_mnt;
55 extern struct nfsrv_stablefirst nfsrv_stablefirst;
56 extern void (*nfsd_call_servertimer)(void);
57 extern SVCPOOL	*nfsrvd_pool;
58 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
59 NFSDLOCKMUTEX;
60 struct mtx nfs_cache_mutex;
61 struct mtx nfs_v4root_mutex;
62 struct nfsrvfh nfs_rootfh, nfs_pubfh;
63 int nfs_pubfhset = 0, nfs_rootfhset = 0;
64 struct proc *nfsd_master_proc = NULL;
65 static pid_t nfsd_master_pid = (pid_t)-1;
66 static char nfsd_master_comm[MAXCOMLEN + 1];
67 static struct timeval nfsd_master_start;
68 static uint32_t nfsv4_sysid = 0;
69 
70 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
71     struct ucred *);
72 
73 int nfsrv_enable_crossmntpt = 1;
74 static int nfs_commit_blks;
75 static int nfs_commit_miss;
76 extern int nfsrv_issuedelegs;
77 extern int nfsrv_dolocallocks;
78 
79 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server");
80 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
81     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
82 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
83     0, "");
84 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
85     0, "");
86 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
87     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
88 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
89     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
90 
91 #define	NUM_HEURISTIC		1017
92 #define	NHUSE_INIT		64
93 #define	NHUSE_INC		16
94 #define	NHUSE_MAX		2048
95 
96 static struct nfsheur {
97 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
98 	off_t nh_nextr;		/* next offset for sequential detection */
99 	int nh_use;		/* use count for selection */
100 	int nh_seqcount;	/* heuristic */
101 } nfsheur[NUM_HEURISTIC];
102 
103 
104 /*
105  * Get attributes into nfsvattr structure.
106  */
107 int
108 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
109     struct thread *p, int vpislocked)
110 {
111 	int error, lockedit = 0;
112 
113 	if (vpislocked == 0) {
114 		/*
115 		 * When vpislocked == 0, the vnode is either exclusively
116 		 * locked by this thread or not locked by this thread.
117 		 * As such, shared lock it, if not exclusively locked.
118 		 */
119 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
120 			lockedit = 1;
121 			vn_lock(vp, LK_SHARED | LK_RETRY);
122 		}
123 	}
124 	error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
125 	if (lockedit != 0)
126 		VOP_UNLOCK(vp, 0);
127 	return (error);
128 }
129 
130 /*
131  * Get a file handle for a vnode.
132  */
133 int
134 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
135 {
136 	int error;
137 
138 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
139 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
140 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
141 	return (error);
142 }
143 
144 /*
145  * Perform access checking for vnodes obtained from file handles that would
146  * refer to files already opened by a Unix client. You cannot just use
147  * vn_writechk() and VOP_ACCESSX() for two reasons.
148  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
149  *     case.
150  * 2 - The owner is to be given access irrespective of mode bits for some
151  *     operations, so that processes that chmod after opening a file don't
152  *     break.
153  */
154 int
155 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
156     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
157     u_int32_t *supportedtypep)
158 {
159 	struct vattr vattr;
160 	int error = 0, getret = 0;
161 
162 	if (vpislocked == 0) {
163 		if (vn_lock(vp, LK_SHARED) != 0)
164 			return (EPERM);
165 	}
166 	if (accmode & VWRITE) {
167 		/* Just vn_writechk() changed to check rdonly */
168 		/*
169 		 * Disallow write attempts on read-only file systems;
170 		 * unless the file is a socket or a block or character
171 		 * device resident on the file system.
172 		 */
173 		if (NFSVNO_EXRDONLY(exp) ||
174 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
175 			switch (vp->v_type) {
176 			case VREG:
177 			case VDIR:
178 			case VLNK:
179 				error = EROFS;
180 			default:
181 				break;
182 			}
183 		}
184 		/*
185 		 * If there's shared text associated with
186 		 * the inode, try to free it up once.  If
187 		 * we fail, we can't allow writing.
188 		 */
189 		if ((vp->v_vflag & VV_TEXT) != 0 && error == 0)
190 			error = ETXTBSY;
191 	}
192 	if (error != 0) {
193 		if (vpislocked == 0)
194 			VOP_UNLOCK(vp, 0);
195 		return (error);
196 	}
197 
198 	/*
199 	 * Should the override still be applied when ACLs are enabled?
200 	 */
201 	error = VOP_ACCESSX(vp, accmode, cred, p);
202 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
203 		/*
204 		 * Try again with VEXPLICIT_DENY, to see if the test for
205 		 * deletion is supported.
206 		 */
207 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
208 		if (error == 0) {
209 			if (vp->v_type == VDIR) {
210 				accmode &= ~(VDELETE | VDELETE_CHILD);
211 				accmode |= VWRITE;
212 				error = VOP_ACCESSX(vp, accmode, cred, p);
213 			} else if (supportedtypep != NULL) {
214 				*supportedtypep &= ~NFSACCESS_DELETE;
215 			}
216 		}
217 	}
218 
219 	/*
220 	 * Allow certain operations for the owner (reads and writes
221 	 * on files that are already open).
222 	 */
223 	if (override != NFSACCCHK_NOOVERRIDE &&
224 	    (error == EPERM || error == EACCES)) {
225 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
226 			error = 0;
227 		else if (override & NFSACCCHK_ALLOWOWNER) {
228 			getret = VOP_GETATTR(vp, &vattr, cred);
229 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
230 				error = 0;
231 		}
232 	}
233 	if (vpislocked == 0)
234 		VOP_UNLOCK(vp, 0);
235 	return (error);
236 }
237 
238 /*
239  * Set attribute(s) vnop.
240  */
241 int
242 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
243     struct thread *p, struct nfsexstuff *exp)
244 {
245 	int error;
246 
247 	error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
248 	return (error);
249 }
250 
251 /*
252  * Set up nameidata for a lookup() call and do it
253  * For the cases where we are crossing mount points
254  * (looking up the public fh path or the v4 root path when
255  *  not using a pseudo-root fs), set/release the Giant lock,
256  * as required.
257  */
258 int
259 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
260     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
261     struct vnode **retdirp)
262 {
263 	struct componentname *cnp = &ndp->ni_cnd;
264 	int i;
265 	struct iovec aiov;
266 	struct uio auio;
267 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
268 	int error = 0, crossmnt;
269 	char *cp;
270 
271 	*retdirp = NULL;
272 	cnp->cn_nameptr = cnp->cn_pnbuf;
273 	/*
274 	 * Extract and set starting directory.
275 	 */
276 	if (dp->v_type != VDIR) {
277 		if (islocked)
278 			vput(dp);
279 		else
280 			vrele(dp);
281 		nfsvno_relpathbuf(ndp);
282 		return (ENOTDIR);
283 	}
284 	if (islocked)
285 		NFSVOPUNLOCK(dp, 0, p);
286 	VREF(dp);
287 	*retdirp = dp;
288 	if (NFSVNO_EXRDONLY(exp))
289 		cnp->cn_flags |= RDONLY;
290 	ndp->ni_segflg = UIO_SYSSPACE;
291 	crossmnt = 1;
292 
293 	if (nd->nd_flag & ND_PUBLOOKUP) {
294 		ndp->ni_loopcnt = 0;
295 		if (cnp->cn_pnbuf[0] == '/') {
296 			vrele(dp);
297 			/*
298 			 * Check for degenerate pathnames here, since lookup()
299 			 * panics on them.
300 			 */
301 			for (i = 1; i < ndp->ni_pathlen; i++)
302 				if (cnp->cn_pnbuf[i] != '/')
303 					break;
304 			if (i == ndp->ni_pathlen) {
305 				error = NFSERR_ACCES;
306 				goto out;
307 			}
308 			dp = rootvnode;
309 			VREF(dp);
310 		}
311 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
312 	    (nd->nd_flag & ND_NFSV4) == 0) {
313 		/*
314 		 * Only cross mount points for NFSv4 when doing a
315 		 * mount while traversing the file system above
316 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
317 		 */
318 		cnp->cn_flags |= NOCROSSMOUNT;
319 		crossmnt = 0;
320 	}
321 
322 	/*
323 	 * Initialize for scan, set ni_startdir and bump ref on dp again
324 	 * becuase lookup() will dereference ni_startdir.
325 	 */
326 
327 	cnp->cn_thread = p;
328 	ndp->ni_startdir = dp;
329 	ndp->ni_rootdir = rootvnode;
330 
331 	if (!lockleaf)
332 		cnp->cn_flags |= LOCKLEAF;
333 	for (;;) {
334 		cnp->cn_nameptr = cnp->cn_pnbuf;
335 		/*
336 		 * Call lookup() to do the real work.  If an error occurs,
337 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
338 		 * we do not have to dereference anything before returning.
339 		 * In either case ni_startdir will be dereferenced and NULLed
340 		 * out.
341 		 */
342 		error = lookup(ndp);
343 		if (error)
344 			break;
345 
346 		/*
347 		 * Check for encountering a symbolic link.  Trivial
348 		 * termination occurs if no symlink encountered.
349 		 */
350 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
351 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
352 				nfsvno_relpathbuf(ndp);
353 			if (ndp->ni_vp && !lockleaf)
354 				NFSVOPUNLOCK(ndp->ni_vp, 0, p);
355 			break;
356 		}
357 
358 		/*
359 		 * Validate symlink
360 		 */
361 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
362 			NFSVOPUNLOCK(ndp->ni_dvp, 0, p);
363 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
364 			error = EINVAL;
365 			goto badlink2;
366 		}
367 
368 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
369 			error = ELOOP;
370 			goto badlink2;
371 		}
372 		if (ndp->ni_pathlen > 1)
373 			cp = uma_zalloc(namei_zone, M_WAITOK);
374 		else
375 			cp = cnp->cn_pnbuf;
376 		aiov.iov_base = cp;
377 		aiov.iov_len = MAXPATHLEN;
378 		auio.uio_iov = &aiov;
379 		auio.uio_iovcnt = 1;
380 		auio.uio_offset = 0;
381 		auio.uio_rw = UIO_READ;
382 		auio.uio_segflg = UIO_SYSSPACE;
383 		auio.uio_td = NULL;
384 		auio.uio_resid = MAXPATHLEN;
385 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
386 		if (error) {
387 		badlink1:
388 			if (ndp->ni_pathlen > 1)
389 				uma_zfree(namei_zone, cp);
390 		badlink2:
391 			vrele(ndp->ni_dvp);
392 			vput(ndp->ni_vp);
393 			break;
394 		}
395 		linklen = MAXPATHLEN - auio.uio_resid;
396 		if (linklen == 0) {
397 			error = ENOENT;
398 			goto badlink1;
399 		}
400 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
401 			error = ENAMETOOLONG;
402 			goto badlink1;
403 		}
404 
405 		/*
406 		 * Adjust or replace path
407 		 */
408 		if (ndp->ni_pathlen > 1) {
409 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
410 			uma_zfree(namei_zone, cnp->cn_pnbuf);
411 			cnp->cn_pnbuf = cp;
412 		} else
413 			cnp->cn_pnbuf[linklen] = '\0';
414 		ndp->ni_pathlen += linklen;
415 
416 		/*
417 		 * Cleanup refs for next loop and check if root directory
418 		 * should replace current directory.  Normally ni_dvp
419 		 * becomes the new base directory and is cleaned up when
420 		 * we loop.  Explicitly null pointers after invalidation
421 		 * to clarify operation.
422 		 */
423 		vput(ndp->ni_vp);
424 		ndp->ni_vp = NULL;
425 
426 		if (cnp->cn_pnbuf[0] == '/') {
427 			vrele(ndp->ni_dvp);
428 			ndp->ni_dvp = ndp->ni_rootdir;
429 			VREF(ndp->ni_dvp);
430 		}
431 		ndp->ni_startdir = ndp->ni_dvp;
432 		ndp->ni_dvp = NULL;
433 	}
434 	if (!lockleaf)
435 		cnp->cn_flags &= ~LOCKLEAF;
436 
437 out:
438 	if (error) {
439 		uma_zfree(namei_zone, cnp->cn_pnbuf);
440 		ndp->ni_vp = NULL;
441 		ndp->ni_dvp = NULL;
442 		ndp->ni_startdir = NULL;
443 		cnp->cn_flags &= ~HASBUF;
444 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
445 		ndp->ni_dvp = NULL;
446 	}
447 	return (error);
448 }
449 
450 /*
451  * Set up a pathname buffer and return a pointer to it and, optionally
452  * set a hash pointer.
453  */
454 void
455 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
456 {
457 	struct componentname *cnp = &ndp->ni_cnd;
458 
459 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
460 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
461 	if (hashpp != NULL)
462 		*hashpp = NULL;
463 	*bufpp = cnp->cn_pnbuf;
464 }
465 
466 /*
467  * Release the above path buffer, if not released by nfsvno_namei().
468  */
469 void
470 nfsvno_relpathbuf(struct nameidata *ndp)
471 {
472 
473 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
474 		panic("nfsrelpath");
475 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
476 	ndp->ni_cnd.cn_flags &= ~HASBUF;
477 }
478 
479 /*
480  * Readlink vnode op into an mbuf list.
481  */
482 int
483 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
484     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
485 {
486 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
487 	struct iovec *ivp = iv;
488 	struct uio io, *uiop = &io;
489 	struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
490 	int i, len, tlen, error;
491 
492 	len = 0;
493 	i = 0;
494 	while (len < NFS_MAXPATHLEN) {
495 		NFSMGET(mp);
496 		MCLGET(mp, M_WAIT);
497 		mp->m_len = NFSMSIZ(mp);
498 		if (len == 0) {
499 			mp3 = mp2 = mp;
500 		} else {
501 			mp2->m_next = mp;
502 			mp2 = mp;
503 		}
504 		if ((len + mp->m_len) > NFS_MAXPATHLEN) {
505 			mp->m_len = NFS_MAXPATHLEN - len;
506 			len = NFS_MAXPATHLEN;
507 		} else {
508 			len += mp->m_len;
509 		}
510 		ivp->iov_base = mtod(mp, caddr_t);
511 		ivp->iov_len = mp->m_len;
512 		i++;
513 		ivp++;
514 	}
515 	uiop->uio_iov = iv;
516 	uiop->uio_iovcnt = i;
517 	uiop->uio_offset = 0;
518 	uiop->uio_resid = len;
519 	uiop->uio_rw = UIO_READ;
520 	uiop->uio_segflg = UIO_SYSSPACE;
521 	uiop->uio_td = NULL;
522 	error = VOP_READLINK(vp, uiop, cred);
523 	if (error) {
524 		m_freem(mp3);
525 		*lenp = 0;
526 		return (error);
527 	}
528 	if (uiop->uio_resid > 0) {
529 		len -= uiop->uio_resid;
530 		tlen = NFSM_RNDUP(len);
531 		nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
532 	}
533 	*lenp = len;
534 	*mpp = mp3;
535 	*mpendp = mp;
536 	return (0);
537 }
538 
539 /*
540  * Read vnode op call into mbuf list.
541  */
542 int
543 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
544     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
545 {
546 	struct mbuf *m;
547 	int i;
548 	struct iovec *iv;
549 	struct iovec *iv2;
550 	int error = 0, len, left, siz, tlen, ioflag = 0, hi, try = 32;
551 	struct mbuf *m2 = NULL, *m3;
552 	struct uio io, *uiop = &io;
553 	struct nfsheur *nh;
554 
555 	/*
556 	 * Calculate seqcount for heuristic
557 	 */
558 	/*
559 	 * Locate best candidate
560 	 */
561 
562 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
563 	nh = &nfsheur[hi];
564 
565 	while (try--) {
566 		if (nfsheur[hi].nh_vp == vp) {
567 			nh = &nfsheur[hi];
568 			break;
569 		}
570 		if (nfsheur[hi].nh_use > 0)
571 			--nfsheur[hi].nh_use;
572 		hi = (hi + 1) % NUM_HEURISTIC;
573 		if (nfsheur[hi].nh_use < nh->nh_use)
574 			nh = &nfsheur[hi];
575 	}
576 
577 	if (nh->nh_vp != vp) {
578 		nh->nh_vp = vp;
579 		nh->nh_nextr = off;
580 		nh->nh_use = NHUSE_INIT;
581 		if (off == 0)
582 			nh->nh_seqcount = 4;
583 		else
584 			nh->nh_seqcount = 1;
585 	}
586 
587 	/*
588 	 * Calculate heuristic
589 	 */
590 
591 	if ((off == 0 && nh->nh_seqcount > 0) || off == nh->nh_nextr) {
592 		if (++nh->nh_seqcount > IO_SEQMAX)
593 			nh->nh_seqcount = IO_SEQMAX;
594 	} else if (nh->nh_seqcount > 1) {
595 		nh->nh_seqcount = 1;
596 	} else {
597 		nh->nh_seqcount = 0;
598 	}
599 	nh->nh_use += NHUSE_INC;
600 	if (nh->nh_use > NHUSE_MAX)
601 		nh->nh_use = NHUSE_MAX;
602 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
603 
604 	len = left = NFSM_RNDUP(cnt);
605 	m3 = NULL;
606 	/*
607 	 * Generate the mbuf list with the uio_iov ref. to it.
608 	 */
609 	i = 0;
610 	while (left > 0) {
611 		NFSMGET(m);
612 		MCLGET(m, M_WAIT);
613 		m->m_len = 0;
614 		siz = min(M_TRAILINGSPACE(m), left);
615 		left -= siz;
616 		i++;
617 		if (m3)
618 			m2->m_next = m;
619 		else
620 			m3 = m;
621 		m2 = m;
622 	}
623 	MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
624 	    M_TEMP, M_WAITOK);
625 	uiop->uio_iov = iv2 = iv;
626 	m = m3;
627 	left = len;
628 	i = 0;
629 	while (left > 0) {
630 		if (m == NULL)
631 			panic("nfsvno_read iov");
632 		siz = min(M_TRAILINGSPACE(m), left);
633 		if (siz > 0) {
634 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
635 			iv->iov_len = siz;
636 			m->m_len += siz;
637 			left -= siz;
638 			iv++;
639 			i++;
640 		}
641 		m = m->m_next;
642 	}
643 	uiop->uio_iovcnt = i;
644 	uiop->uio_offset = off;
645 	uiop->uio_resid = len;
646 	uiop->uio_rw = UIO_READ;
647 	uiop->uio_segflg = UIO_SYSSPACE;
648 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
649 	FREE((caddr_t)iv2, M_TEMP);
650 	if (error) {
651 		m_freem(m3);
652 		*mpp = NULL;
653 		return (error);
654 	}
655 	tlen = len - uiop->uio_resid;
656 	cnt = cnt < tlen ? cnt : tlen;
657 	tlen = NFSM_RNDUP(cnt);
658 	if (tlen == 0) {
659 		m_freem(m3);
660 		m3 = NULL;
661 	} else if (len != tlen || tlen != cnt)
662 		nfsrv_adj(m3, len - tlen, tlen - cnt);
663 	*mpp = m3;
664 	*mpendp = m2;
665 	return (0);
666 }
667 
668 /*
669  * Write vnode op from an mbuf list.
670  */
671 int
672 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
673     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
674 {
675 	struct iovec *ivp;
676 	int i, len;
677 	struct iovec *iv;
678 	int ioflags, error;
679 	struct uio io, *uiop = &io;
680 
681 	MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
682 	    M_WAITOK);
683 	uiop->uio_iov = iv = ivp;
684 	uiop->uio_iovcnt = cnt;
685 	i = mtod(mp, caddr_t) + mp->m_len - cp;
686 	len = retlen;
687 	while (len > 0) {
688 		if (mp == NULL)
689 			panic("nfsvno_write");
690 		if (i > 0) {
691 			i = min(i, len);
692 			ivp->iov_base = cp;
693 			ivp->iov_len = i;
694 			ivp++;
695 			len -= i;
696 		}
697 		mp = mp->m_next;
698 		if (mp) {
699 			i = mp->m_len;
700 			cp = mtod(mp, caddr_t);
701 		}
702 	}
703 
704 	if (stable == NFSWRITE_UNSTABLE)
705 		ioflags = IO_NODELOCKED;
706 	else
707 		ioflags = (IO_SYNC | IO_NODELOCKED);
708 	uiop->uio_resid = retlen;
709 	uiop->uio_rw = UIO_WRITE;
710 	uiop->uio_segflg = UIO_SYSSPACE;
711 	NFSUIOPROC(uiop, p);
712 	uiop->uio_offset = off;
713 	error = VOP_WRITE(vp, uiop, ioflags, cred);
714 	FREE((caddr_t)iv, M_TEMP);
715 	return (error);
716 }
717 
718 /*
719  * Common code for creating a regular file (plus special files for V2).
720  */
721 int
722 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
723     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
724     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
725 {
726 	u_quad_t tempsize;
727 	int error;
728 
729 	error = nd->nd_repstat;
730 	if (!error && ndp->ni_vp == NULL) {
731 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
732 			vrele(ndp->ni_startdir);
733 			error = VOP_CREATE(ndp->ni_dvp,
734 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
735 			vput(ndp->ni_dvp);
736 			nfsvno_relpathbuf(ndp);
737 			if (!error) {
738 				if (*exclusive_flagp) {
739 					*exclusive_flagp = 0;
740 					NFSVNO_ATTRINIT(nvap);
741 					nvap->na_atime.tv_sec = cverf[0];
742 					nvap->na_atime.tv_nsec = cverf[1];
743 					error = VOP_SETATTR(ndp->ni_vp,
744 					    &nvap->na_vattr, nd->nd_cred);
745 				}
746 			}
747 		/*
748 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
749 		 * (This implies, just get out on an error.)
750 		 */
751 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
752 			nvap->na_type == VFIFO) {
753 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
754 				nvap->na_type = VFIFO;
755                         if (nvap->na_type != VFIFO &&
756 			    (error = priv_check_cred(nd->nd_cred,
757 			     PRIV_VFS_MKNOD_DEV, 0))) {
758 				vrele(ndp->ni_startdir);
759 				nfsvno_relpathbuf(ndp);
760 				vput(ndp->ni_dvp);
761 				return (error);
762 			}
763 			nvap->na_rdev = rdev;
764 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
765 			    &ndp->ni_cnd, &nvap->na_vattr);
766 			vput(ndp->ni_dvp);
767 			nfsvno_relpathbuf(ndp);
768 			vrele(ndp->ni_startdir);
769 			if (error)
770 				return (error);
771 		} else {
772 			vrele(ndp->ni_startdir);
773 			nfsvno_relpathbuf(ndp);
774 			vput(ndp->ni_dvp);
775 			return (ENXIO);
776 		}
777 		*vpp = ndp->ni_vp;
778 	} else {
779 		/*
780 		 * Handle cases where error is already set and/or
781 		 * the file exists.
782 		 * 1 - clean up the lookup
783 		 * 2 - iff !error and na_size set, truncate it
784 		 */
785 		vrele(ndp->ni_startdir);
786 		nfsvno_relpathbuf(ndp);
787 		*vpp = ndp->ni_vp;
788 		if (ndp->ni_dvp == *vpp)
789 			vrele(ndp->ni_dvp);
790 		else
791 			vput(ndp->ni_dvp);
792 		if (!error && nvap->na_size != VNOVAL) {
793 			error = nfsvno_accchk(*vpp, VWRITE,
794 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
795 			    NFSACCCHK_VPISLOCKED, NULL);
796 			if (!error) {
797 				tempsize = nvap->na_size;
798 				NFSVNO_ATTRINIT(nvap);
799 				nvap->na_size = tempsize;
800 				error = VOP_SETATTR(*vpp,
801 				    &nvap->na_vattr, nd->nd_cred);
802 			}
803 		}
804 		if (error)
805 			vput(*vpp);
806 	}
807 	return (error);
808 }
809 
810 /*
811  * Do a mknod vnode op.
812  */
813 int
814 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
815     struct thread *p)
816 {
817 	int error = 0;
818 	enum vtype vtyp;
819 
820 	vtyp = nvap->na_type;
821 	/*
822 	 * Iff doesn't exist, create it.
823 	 */
824 	if (ndp->ni_vp) {
825 		vrele(ndp->ni_startdir);
826 		nfsvno_relpathbuf(ndp);
827 		vput(ndp->ni_dvp);
828 		vrele(ndp->ni_vp);
829 		return (EEXIST);
830 	}
831 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
832 		vrele(ndp->ni_startdir);
833 		nfsvno_relpathbuf(ndp);
834 		vput(ndp->ni_dvp);
835 		return (NFSERR_BADTYPE);
836 	}
837 	if (vtyp == VSOCK) {
838 		vrele(ndp->ni_startdir);
839 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
840 		    &ndp->ni_cnd, &nvap->na_vattr);
841 		vput(ndp->ni_dvp);
842 		nfsvno_relpathbuf(ndp);
843 	} else {
844 		if (nvap->na_type != VFIFO &&
845 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
846 			vrele(ndp->ni_startdir);
847 			nfsvno_relpathbuf(ndp);
848 			vput(ndp->ni_dvp);
849 			return (error);
850 		}
851 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
852 		    &ndp->ni_cnd, &nvap->na_vattr);
853 		vput(ndp->ni_dvp);
854 		nfsvno_relpathbuf(ndp);
855 		vrele(ndp->ni_startdir);
856 		/*
857 		 * Since VOP_MKNOD returns the ni_vp, I can't
858 		 * see any reason to do the lookup.
859 		 */
860 	}
861 	return (error);
862 }
863 
864 /*
865  * Mkdir vnode op.
866  */
867 int
868 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
869     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
870 {
871 	int error = 0;
872 
873 	if (ndp->ni_vp != NULL) {
874 		if (ndp->ni_dvp == ndp->ni_vp)
875 			vrele(ndp->ni_dvp);
876 		else
877 			vput(ndp->ni_dvp);
878 		vrele(ndp->ni_vp);
879 		nfsvno_relpathbuf(ndp);
880 		return (EEXIST);
881 	}
882 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
883 	    &nvap->na_vattr);
884 	vput(ndp->ni_dvp);
885 	nfsvno_relpathbuf(ndp);
886 	return (error);
887 }
888 
889 /*
890  * symlink vnode op.
891  */
892 int
893 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
894     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
895     struct nfsexstuff *exp)
896 {
897 	int error = 0;
898 
899 	if (ndp->ni_vp) {
900 		vrele(ndp->ni_startdir);
901 		nfsvno_relpathbuf(ndp);
902 		if (ndp->ni_dvp == ndp->ni_vp)
903 			vrele(ndp->ni_dvp);
904 		else
905 			vput(ndp->ni_dvp);
906 		vrele(ndp->ni_vp);
907 		return (EEXIST);
908 	}
909 
910 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
911 	    &nvap->na_vattr, pathcp);
912 	vput(ndp->ni_dvp);
913 	vrele(ndp->ni_startdir);
914 	nfsvno_relpathbuf(ndp);
915 	/*
916 	 * Although FreeBSD still had the lookup code in
917 	 * it for 7/current, there doesn't seem to be any
918 	 * point, since VOP_SYMLINK() returns the ni_vp.
919 	 * Just vput it for v2.
920 	 */
921 	if (!not_v2 && !error)
922 		vput(ndp->ni_vp);
923 	return (error);
924 }
925 
926 /*
927  * Parse symbolic link arguments.
928  * This function has an ugly side effect. It will MALLOC() an area for
929  * the symlink and set iov_base to point to it, only if it succeeds.
930  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
931  * be FREE'd later.
932  */
933 int
934 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
935     struct thread *p, char **pathcpp, int *lenp)
936 {
937 	u_int32_t *tl;
938 	char *pathcp = NULL;
939 	int error = 0, len;
940 	struct nfsv2_sattr *sp;
941 
942 	*pathcpp = NULL;
943 	*lenp = 0;
944 	if ((nd->nd_flag & ND_NFSV3) &&
945 	    (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
946 		goto nfsmout;
947 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
948 	len = fxdr_unsigned(int, *tl);
949 	if (len > NFS_MAXPATHLEN || len <= 0) {
950 		error = EBADRPC;
951 		goto nfsmout;
952 	}
953 	MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
954 	error = nfsrv_mtostr(nd, pathcp, len);
955 	if (error)
956 		goto nfsmout;
957 	if (nd->nd_flag & ND_NFSV2) {
958 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
959 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
960 	}
961 	*pathcpp = pathcp;
962 	*lenp = len;
963 	return (0);
964 nfsmout:
965 	if (pathcp)
966 		free(pathcp, M_TEMP);
967 	return (error);
968 }
969 
970 /*
971  * Remove a non-directory object.
972  */
973 int
974 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
975     struct thread *p, struct nfsexstuff *exp)
976 {
977 	struct vnode *vp;
978 	int error = 0;
979 
980 	vp = ndp->ni_vp;
981 	if (vp->v_type == VDIR)
982 		error = NFSERR_ISDIR;
983 	else if (is_v4)
984 		error = nfsrv_checkremove(vp, 1, p);
985 	if (!error)
986 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
987 	if (ndp->ni_dvp == vp)
988 		vrele(ndp->ni_dvp);
989 	else
990 		vput(ndp->ni_dvp);
991 	vput(vp);
992 	return (error);
993 }
994 
995 /*
996  * Remove a directory.
997  */
998 int
999 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1000     struct thread *p, struct nfsexstuff *exp)
1001 {
1002 	struct vnode *vp;
1003 	int error = 0;
1004 
1005 	vp = ndp->ni_vp;
1006 	if (vp->v_type != VDIR) {
1007 		error = ENOTDIR;
1008 		goto out;
1009 	}
1010 	/*
1011 	 * No rmdir "." please.
1012 	 */
1013 	if (ndp->ni_dvp == vp) {
1014 		error = EINVAL;
1015 		goto out;
1016 	}
1017 	/*
1018 	 * The root of a mounted filesystem cannot be deleted.
1019 	 */
1020 	if (vp->v_vflag & VV_ROOT)
1021 		error = EBUSY;
1022 out:
1023 	if (!error)
1024 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1025 	if (ndp->ni_dvp == vp)
1026 		vrele(ndp->ni_dvp);
1027 	else
1028 		vput(ndp->ni_dvp);
1029 	vput(vp);
1030 	return (error);
1031 }
1032 
1033 /*
1034  * Rename vnode op.
1035  */
1036 int
1037 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1038     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1039 {
1040 	struct vnode *fvp, *tvp, *tdvp;
1041 	int error = 0;
1042 
1043 	fvp = fromndp->ni_vp;
1044 	if (ndstat) {
1045 		vrele(fromndp->ni_dvp);
1046 		vrele(fvp);
1047 		error = ndstat;
1048 		goto out1;
1049 	}
1050 	tdvp = tondp->ni_dvp;
1051 	tvp = tondp->ni_vp;
1052 	if (tvp != NULL) {
1053 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1054 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1055 			goto out;
1056 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1057 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1058 			goto out;
1059 		}
1060 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1061 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1062 			goto out;
1063 		}
1064 
1065 		/*
1066 		 * A rename to '.' or '..' results in a prematurely
1067 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1068 		 * here.
1069 		 */
1070 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1071 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1072 		    (tondp->ni_cnd.cn_namelen == 2 &&
1073 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1074 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1075 			error = EINVAL;
1076 			goto out;
1077 		}
1078 	}
1079 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1080 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1081 		goto out;
1082 	}
1083 	if (fvp->v_mount != tdvp->v_mount) {
1084 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1085 		goto out;
1086 	}
1087 	if (fvp == tdvp) {
1088 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1089 		goto out;
1090 	}
1091 	if (fvp == tvp) {
1092 		/*
1093 		 * If source and destination are the same, there is nothing to
1094 		 * do. Set error to -1 to indicate this.
1095 		 */
1096 		error = -1;
1097 		goto out;
1098 	}
1099 	if (ndflag & ND_NFSV4) {
1100 		if (vn_lock(fvp, LK_EXCLUSIVE) == 0) {
1101 			error = nfsrv_checkremove(fvp, 0, p);
1102 			VOP_UNLOCK(fvp, 0);
1103 		} else
1104 			error = EPERM;
1105 		if (tvp && !error)
1106 			error = nfsrv_checkremove(tvp, 1, p);
1107 	} else {
1108 		/*
1109 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1110 		 * that the NFSv4 client won't be confused by the rename.
1111 		 * Since nfsd_recalldelegation() can only be called on an
1112 		 * unlocked vnode at this point and fvp is the file that will
1113 		 * still exist after the rename, just do fvp.
1114 		 */
1115 		nfsd_recalldelegation(fvp, p);
1116 	}
1117 out:
1118 	if (!error) {
1119 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1120 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1121 		    &tondp->ni_cnd);
1122 	} else {
1123 		if (tdvp == tvp)
1124 			vrele(tdvp);
1125 		else
1126 			vput(tdvp);
1127 		if (tvp)
1128 			vput(tvp);
1129 		vrele(fromndp->ni_dvp);
1130 		vrele(fvp);
1131 		if (error == -1)
1132 			error = 0;
1133 	}
1134 	vrele(tondp->ni_startdir);
1135 	nfsvno_relpathbuf(tondp);
1136 out1:
1137 	vrele(fromndp->ni_startdir);
1138 	nfsvno_relpathbuf(fromndp);
1139 	return (error);
1140 }
1141 
1142 /*
1143  * Link vnode op.
1144  */
1145 int
1146 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1147     struct thread *p, struct nfsexstuff *exp)
1148 {
1149 	struct vnode *xp;
1150 	int error = 0;
1151 
1152 	xp = ndp->ni_vp;
1153 	if (xp != NULL) {
1154 		error = EEXIST;
1155 	} else {
1156 		xp = ndp->ni_dvp;
1157 		if (vp->v_mount != xp->v_mount)
1158 			error = EXDEV;
1159 	}
1160 	if (!error) {
1161 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1162 		if ((vp->v_iflag & VI_DOOMED) == 0)
1163 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1164 		else
1165 			error = EPERM;
1166 		if (ndp->ni_dvp == vp)
1167 			vrele(ndp->ni_dvp);
1168 		else
1169 			vput(ndp->ni_dvp);
1170 		VOP_UNLOCK(vp, 0);
1171 	} else {
1172 		if (ndp->ni_dvp == ndp->ni_vp)
1173 			vrele(ndp->ni_dvp);
1174 		else
1175 			vput(ndp->ni_dvp);
1176 		if (ndp->ni_vp)
1177 			vrele(ndp->ni_vp);
1178 	}
1179 	nfsvno_relpathbuf(ndp);
1180 	return (error);
1181 }
1182 
1183 /*
1184  * Do the fsync() appropriate for the commit.
1185  */
1186 int
1187 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1188     struct thread *td)
1189 {
1190 	int error = 0;
1191 
1192 	if (cnt > MAX_COMMIT_COUNT) {
1193 		/*
1194 		 * Give up and do the whole thing
1195 		 */
1196 		if (vp->v_object &&
1197 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1198 			VM_OBJECT_LOCK(vp->v_object);
1199 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1200 			VM_OBJECT_UNLOCK(vp->v_object);
1201 		}
1202 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1203 	} else {
1204 		/*
1205 		 * Locate and synchronously write any buffers that fall
1206 		 * into the requested range.  Note:  we are assuming that
1207 		 * f_iosize is a power of 2.
1208 		 */
1209 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1210 		int iomask = iosize - 1;
1211 		struct bufobj *bo;
1212 		daddr_t lblkno;
1213 
1214 		/*
1215 		 * Align to iosize boundry, super-align to page boundry.
1216 		 */
1217 		if (off & iomask) {
1218 			cnt += off & iomask;
1219 			off &= ~(u_quad_t)iomask;
1220 		}
1221 		if (off & PAGE_MASK) {
1222 			cnt += off & PAGE_MASK;
1223 			off &= ~(u_quad_t)PAGE_MASK;
1224 		}
1225 		lblkno = off / iosize;
1226 
1227 		if (vp->v_object &&
1228 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1229 			VM_OBJECT_LOCK(vp->v_object);
1230 			vm_object_page_clean(vp->v_object, off, off + cnt,
1231 			    OBJPC_SYNC);
1232 			VM_OBJECT_UNLOCK(vp->v_object);
1233 		}
1234 
1235 		bo = &vp->v_bufobj;
1236 		BO_LOCK(bo);
1237 		while (cnt > 0) {
1238 			struct buf *bp;
1239 
1240 			/*
1241 			 * If we have a buffer and it is marked B_DELWRI we
1242 			 * have to lock and write it.  Otherwise the prior
1243 			 * write is assumed to have already been committed.
1244 			 *
1245 			 * gbincore() can return invalid buffers now so we
1246 			 * have to check that bit as well (though B_DELWRI
1247 			 * should not be set if B_INVAL is set there could be
1248 			 * a race here since we haven't locked the buffer).
1249 			 */
1250 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1251 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1252 				    LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
1253 					BO_LOCK(bo);
1254 					continue; /* retry */
1255 				}
1256 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1257 				    B_DELWRI) {
1258 					bremfree(bp);
1259 					bp->b_flags &= ~B_ASYNC;
1260 					bwrite(bp);
1261 					++nfs_commit_miss;
1262 				} else
1263 					BUF_UNLOCK(bp);
1264 				BO_LOCK(bo);
1265 			}
1266 			++nfs_commit_blks;
1267 			if (cnt < iosize)
1268 				break;
1269 			cnt -= iosize;
1270 			++lblkno;
1271 		}
1272 		BO_UNLOCK(bo);
1273 	}
1274 	return (error);
1275 }
1276 
1277 /*
1278  * Statfs vnode op.
1279  */
1280 int
1281 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1282 {
1283 	int error;
1284 
1285 	error = VFS_STATFS(vp->v_mount, sf);
1286 	if (error == 0) {
1287 		/*
1288 		 * Since NFS handles these values as unsigned on the
1289 		 * wire, there is no way to represent negative values,
1290 		 * so set them to 0. Without this, they will appear
1291 		 * to be very large positive values for clients like
1292 		 * Solaris10.
1293 		 */
1294 		if (sf->f_bavail < 0)
1295 			sf->f_bavail = 0;
1296 		if (sf->f_ffree < 0)
1297 			sf->f_ffree = 0;
1298 	}
1299 	return (error);
1300 }
1301 
1302 /*
1303  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1304  * must handle nfsrv_opencheck() calls after any other access checks.
1305  */
1306 void
1307 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1308     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1309     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1310     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1311     struct nfsexstuff *exp, struct vnode **vpp)
1312 {
1313 	struct vnode *vp = NULL;
1314 	u_quad_t tempsize;
1315 	struct nfsexstuff nes;
1316 
1317 	if (ndp->ni_vp == NULL)
1318 		nd->nd_repstat = nfsrv_opencheck(clientid,
1319 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1320 	if (!nd->nd_repstat) {
1321 		if (ndp->ni_vp == NULL) {
1322 			vrele(ndp->ni_startdir);
1323 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1324 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1325 			vput(ndp->ni_dvp);
1326 			nfsvno_relpathbuf(ndp);
1327 			if (!nd->nd_repstat) {
1328 				if (*exclusive_flagp) {
1329 					*exclusive_flagp = 0;
1330 					NFSVNO_ATTRINIT(nvap);
1331 					nvap->na_atime.tv_sec = cverf[0];
1332 					nvap->na_atime.tv_nsec = cverf[1];
1333 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1334 					    &nvap->na_vattr, cred);
1335 				} else {
1336 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1337 					    aclp, p, attrbitp, exp);
1338 				}
1339 			}
1340 			vp = ndp->ni_vp;
1341 		} else {
1342 			if (ndp->ni_startdir)
1343 				vrele(ndp->ni_startdir);
1344 			nfsvno_relpathbuf(ndp);
1345 			vp = ndp->ni_vp;
1346 			if (create == NFSV4OPEN_CREATE) {
1347 				if (ndp->ni_dvp == vp)
1348 					vrele(ndp->ni_dvp);
1349 				else
1350 					vput(ndp->ni_dvp);
1351 			}
1352 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1353 				if (ndp->ni_cnd.cn_flags & RDONLY)
1354 					NFSVNO_SETEXRDONLY(&nes);
1355 				else
1356 					NFSVNO_EXINIT(&nes);
1357 				nd->nd_repstat = nfsvno_accchk(vp,
1358 				    VWRITE, cred, &nes, p,
1359 				    NFSACCCHK_NOOVERRIDE,
1360 				    NFSACCCHK_VPISLOCKED, NULL);
1361 				nd->nd_repstat = nfsrv_opencheck(clientid,
1362 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1363 				if (!nd->nd_repstat) {
1364 					tempsize = nvap->na_size;
1365 					NFSVNO_ATTRINIT(nvap);
1366 					nvap->na_size = tempsize;
1367 					nd->nd_repstat = VOP_SETATTR(vp,
1368 					    &nvap->na_vattr, cred);
1369 				}
1370 			} else if (vp->v_type == VREG) {
1371 				nd->nd_repstat = nfsrv_opencheck(clientid,
1372 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1373 			}
1374 		}
1375 	} else {
1376 		if (ndp->ni_cnd.cn_flags & HASBUF)
1377 			nfsvno_relpathbuf(ndp);
1378 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1379 			vrele(ndp->ni_startdir);
1380 			if (ndp->ni_dvp == ndp->ni_vp)
1381 				vrele(ndp->ni_dvp);
1382 			else
1383 				vput(ndp->ni_dvp);
1384 			if (ndp->ni_vp)
1385 				vput(ndp->ni_vp);
1386 		}
1387 	}
1388 	*vpp = vp;
1389 }
1390 
1391 /*
1392  * Updates the file rev and sets the mtime and ctime
1393  * to the current clock time, returning the va_filerev and va_Xtime
1394  * values.
1395  */
1396 void
1397 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1398     struct ucred *cred, struct thread *p)
1399 {
1400 	struct vattr va;
1401 
1402 	VATTR_NULL(&va);
1403 	getnanotime(&va.va_mtime);
1404 	(void) VOP_SETATTR(vp, &va, cred);
1405 	(void) nfsvno_getattr(vp, nvap, cred, p, 1);
1406 }
1407 
1408 /*
1409  * Glue routine to nfsv4_fillattr().
1410  */
1411 int
1412 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1413     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1414     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1415     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1416 {
1417 	int error;
1418 
1419 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1420 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1421 	    mounted_on_fileno);
1422 	return (error);
1423 }
1424 
1425 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1426 /*
1427  * nfs readdir service
1428  * - mallocs what it thinks is enough to read
1429  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1430  * - calls VOP_READDIR()
1431  * - loops around building the reply
1432  *	if the output generated exceeds count break out of loop
1433  *	The NFSM_CLGET macro is used here so that the reply will be packed
1434  *	tightly in mbuf clusters.
1435  * - it trims out records with d_fileno == 0
1436  *	this doesn't matter for Unix clients, but they might confuse clients
1437  *	for other os'.
1438  * - it trims out records with d_type == DT_WHT
1439  *	these cannot be seen through NFS (unless we extend the protocol)
1440  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1441  * PS: The NFS protocol spec. does not clarify what the "count" byte
1442  *	argument is a count of.. just name strings and file id's or the
1443  *	entire reply rpc or ...
1444  *	I tried just file name and id sizes and it confused the Sun client,
1445  *	so I am using the full rpc size now. The "paranoia.." comment refers
1446  *	to including the status longwords that are not a part of the dir.
1447  *	"entry" structures, but are in the rpc.
1448  */
1449 int
1450 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1451     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1452 {
1453 	struct dirent *dp;
1454 	u_int32_t *tl;
1455 	int dirlen;
1456 	char *cpos, *cend, *rbuf;
1457 	struct nfsvattr at;
1458 	int nlen, error = 0, getret = 1;
1459 	int siz, cnt, fullsiz, eofflag, ncookies;
1460 	u_int64_t off, toff, verf;
1461 	u_long *cookies = NULL, *cookiep;
1462 	struct uio io;
1463 	struct iovec iv;
1464 	int not_zfs;
1465 
1466 	if (nd->nd_repstat) {
1467 		nfsrv_postopattr(nd, getret, &at);
1468 		return (0);
1469 	}
1470 	if (nd->nd_flag & ND_NFSV2) {
1471 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1472 		off = fxdr_unsigned(u_quad_t, *tl++);
1473 	} else {
1474 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1475 		off = fxdr_hyper(tl);
1476 		tl += 2;
1477 		verf = fxdr_hyper(tl);
1478 		tl += 2;
1479 	}
1480 	toff = off;
1481 	cnt = fxdr_unsigned(int, *tl);
1482 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1483 		cnt = NFS_SRVMAXDATA(nd);
1484 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1485 	fullsiz = siz;
1486 	if (nd->nd_flag & ND_NFSV3) {
1487 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1488 		    p, 1);
1489 #if 0
1490 		/*
1491 		 * va_filerev is not sufficient as a cookie verifier,
1492 		 * since it is not supposed to change when entries are
1493 		 * removed/added unless that offset cookies returned to
1494 		 * the client are no longer valid.
1495 		 */
1496 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
1497 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1498 #endif
1499 	}
1500 	if (nd->nd_repstat == 0 && cnt == 0) {
1501 		if (nd->nd_flag & ND_NFSV2)
1502 			/* NFSv2 does not have NFSERR_TOOSMALL */
1503 			nd->nd_repstat = EPERM;
1504 		else
1505 			nd->nd_repstat = NFSERR_TOOSMALL;
1506 	}
1507 	if (!nd->nd_repstat)
1508 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1509 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1510 		    NFSACCCHK_VPISLOCKED, NULL);
1511 	if (nd->nd_repstat) {
1512 		vput(vp);
1513 		if (nd->nd_flag & ND_NFSV3)
1514 			nfsrv_postopattr(nd, getret, &at);
1515 		return (0);
1516 	}
1517 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1518 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1519 again:
1520 	eofflag = 0;
1521 	if (cookies) {
1522 		free((caddr_t)cookies, M_TEMP);
1523 		cookies = NULL;
1524 	}
1525 
1526 	iv.iov_base = rbuf;
1527 	iv.iov_len = siz;
1528 	io.uio_iov = &iv;
1529 	io.uio_iovcnt = 1;
1530 	io.uio_offset = (off_t)off;
1531 	io.uio_resid = siz;
1532 	io.uio_segflg = UIO_SYSSPACE;
1533 	io.uio_rw = UIO_READ;
1534 	io.uio_td = NULL;
1535 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1536 	    &cookies);
1537 	off = (u_int64_t)io.uio_offset;
1538 	if (io.uio_resid)
1539 		siz -= io.uio_resid;
1540 
1541 	if (!cookies && !nd->nd_repstat)
1542 		nd->nd_repstat = NFSERR_PERM;
1543 	if (nd->nd_flag & ND_NFSV3) {
1544 		getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1545 		if (!nd->nd_repstat)
1546 			nd->nd_repstat = getret;
1547 	}
1548 
1549 	/*
1550 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
1551 	 */
1552 	if (nd->nd_repstat) {
1553 		vput(vp);
1554 		free((caddr_t)rbuf, M_TEMP);
1555 		if (cookies)
1556 			free((caddr_t)cookies, M_TEMP);
1557 		if (nd->nd_flag & ND_NFSV3)
1558 			nfsrv_postopattr(nd, getret, &at);
1559 		return (0);
1560 	}
1561 	/*
1562 	 * If nothing read, return eof
1563 	 * rpc reply
1564 	 */
1565 	if (siz == 0) {
1566 		vput(vp);
1567 		if (nd->nd_flag & ND_NFSV2) {
1568 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1569 		} else {
1570 			nfsrv_postopattr(nd, getret, &at);
1571 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1572 			txdr_hyper(at.na_filerev, tl);
1573 			tl += 2;
1574 		}
1575 		*tl++ = newnfs_false;
1576 		*tl = newnfs_true;
1577 		FREE((caddr_t)rbuf, M_TEMP);
1578 		FREE((caddr_t)cookies, M_TEMP);
1579 		return (0);
1580 	}
1581 
1582 	/*
1583 	 * Check for degenerate cases of nothing useful read.
1584 	 * If so go try again
1585 	 */
1586 	cpos = rbuf;
1587 	cend = rbuf + siz;
1588 	dp = (struct dirent *)cpos;
1589 	cookiep = cookies;
1590 
1591 	/*
1592 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1593 	 * directory offset up to a block boundary, so it is necessary to
1594 	 * skip over the records that precede the requested offset. This
1595 	 * requires the assumption that file offset cookies monotonically
1596 	 * increase.
1597 	 * Since the offset cookies don't monotonically increase for ZFS,
1598 	 * this is not done when ZFS is the file system.
1599 	 */
1600 	while (cpos < cend && ncookies > 0 &&
1601 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1602 	     (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
1603 		cpos += dp->d_reclen;
1604 		dp = (struct dirent *)cpos;
1605 		cookiep++;
1606 		ncookies--;
1607 	}
1608 	if (cpos >= cend || ncookies == 0) {
1609 		siz = fullsiz;
1610 		toff = off;
1611 		goto again;
1612 	}
1613 	vput(vp);
1614 
1615 	/*
1616 	 * dirlen is the size of the reply, including all XDR and must
1617 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1618 	 * if the XDR should be included in "count", but to be safe, we do.
1619 	 * (Include the two booleans at the end of the reply in dirlen now.)
1620 	 */
1621 	if (nd->nd_flag & ND_NFSV3) {
1622 		nfsrv_postopattr(nd, getret, &at);
1623 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1624 		txdr_hyper(at.na_filerev, tl);
1625 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1626 	} else {
1627 		dirlen = 2 * NFSX_UNSIGNED;
1628 	}
1629 
1630 	/* Loop through the records and build reply */
1631 	while (cpos < cend && ncookies > 0) {
1632 		nlen = dp->d_namlen;
1633 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1634 			nlen <= NFS_MAXNAMLEN) {
1635 			if (nd->nd_flag & ND_NFSV3)
1636 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1637 			else
1638 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1639 			if (dirlen > cnt) {
1640 				eofflag = 0;
1641 				break;
1642 			}
1643 
1644 			/*
1645 			 * Build the directory record xdr from
1646 			 * the dirent entry.
1647 			 */
1648 			if (nd->nd_flag & ND_NFSV3) {
1649 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1650 				*tl++ = newnfs_true;
1651 				*tl++ = 0;
1652 			} else {
1653 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1654 				*tl++ = newnfs_true;
1655 			}
1656 			*tl = txdr_unsigned(dp->d_fileno);
1657 			(void) nfsm_strtom(nd, dp->d_name, nlen);
1658 			if (nd->nd_flag & ND_NFSV3) {
1659 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1660 				*tl++ = 0;
1661 			} else
1662 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1663 			*tl = txdr_unsigned(*cookiep);
1664 		}
1665 		cpos += dp->d_reclen;
1666 		dp = (struct dirent *)cpos;
1667 		cookiep++;
1668 		ncookies--;
1669 	}
1670 	if (cpos < cend)
1671 		eofflag = 0;
1672 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1673 	*tl++ = newnfs_false;
1674 	if (eofflag)
1675 		*tl = newnfs_true;
1676 	else
1677 		*tl = newnfs_false;
1678 	FREE((caddr_t)rbuf, M_TEMP);
1679 	FREE((caddr_t)cookies, M_TEMP);
1680 	return (0);
1681 nfsmout:
1682 	vput(vp);
1683 	return (error);
1684 }
1685 
1686 /*
1687  * Readdirplus for V3 and Readdir for V4.
1688  */
1689 int
1690 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1691     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1692 {
1693 	struct dirent *dp;
1694 	u_int32_t *tl;
1695 	int dirlen;
1696 	char *cpos, *cend, *rbuf;
1697 	struct vnode *nvp;
1698 	fhandle_t nfh;
1699 	struct nfsvattr nva, at, *nvap = &nva;
1700 	struct mbuf *mb0, *mb1;
1701 	struct nfsreferral *refp;
1702 	int nlen, r, error = 0, getret = 1, usevget = 1;
1703 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1704 	caddr_t bpos0, bpos1;
1705 	u_int64_t off, toff, verf;
1706 	u_long *cookies = NULL, *cookiep;
1707 	nfsattrbit_t attrbits, rderrbits, savbits;
1708 	struct uio io;
1709 	struct iovec iv;
1710 	struct componentname cn;
1711 	int at_root, needs_unbusy, not_zfs, supports_nfsv4acls;
1712 	struct mount *mp, *new_mp;
1713 	uint64_t mounted_on_fileno;
1714 
1715 	if (nd->nd_repstat) {
1716 		nfsrv_postopattr(nd, getret, &at);
1717 		return (0);
1718 	}
1719 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1720 	off = fxdr_hyper(tl);
1721 	toff = off;
1722 	tl += 2;
1723 	verf = fxdr_hyper(tl);
1724 	tl += 2;
1725 	siz = fxdr_unsigned(int, *tl++);
1726 	cnt = fxdr_unsigned(int, *tl);
1727 
1728 	/*
1729 	 * Use the server's maximum data transfer size as the upper bound
1730 	 * on reply datalen.
1731 	 */
1732 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1733 		cnt = NFS_SRVMAXDATA(nd);
1734 
1735 	/*
1736 	 * siz is a "hint" of how much directory information (name, fileid,
1737 	 * cookie) should be in the reply. At least one client "hints" 0,
1738 	 * so I set it to cnt for that case. I also round it up to the
1739 	 * next multiple of DIRBLKSIZ.
1740 	 */
1741 	if (siz <= 0)
1742 		siz = cnt;
1743 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1744 
1745 	if (nd->nd_flag & ND_NFSV4) {
1746 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1747 		if (error)
1748 			goto nfsmout;
1749 		NFSSET_ATTRBIT(&savbits, &attrbits);
1750 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1751 		NFSZERO_ATTRBIT(&rderrbits);
1752 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1753 	} else {
1754 		NFSZERO_ATTRBIT(&attrbits);
1755 	}
1756 	fullsiz = siz;
1757 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1758 	if (!nd->nd_repstat) {
1759 	    if (off && verf != at.na_filerev) {
1760 		/*
1761 		 * va_filerev is not sufficient as a cookie verifier,
1762 		 * since it is not supposed to change when entries are
1763 		 * removed/added unless that offset cookies returned to
1764 		 * the client are no longer valid.
1765 		 */
1766 #if 0
1767 		if (nd->nd_flag & ND_NFSV4) {
1768 			nd->nd_repstat = NFSERR_NOTSAME;
1769 		} else {
1770 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1771 		}
1772 #endif
1773 	    } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1774 		nd->nd_repstat = NFSERR_BAD_COOKIE;
1775 	    }
1776 	}
1777 	if (!nd->nd_repstat && vp->v_type != VDIR)
1778 		nd->nd_repstat = NFSERR_NOTDIR;
1779 	if (!nd->nd_repstat && cnt == 0)
1780 		nd->nd_repstat = NFSERR_TOOSMALL;
1781 	if (!nd->nd_repstat)
1782 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1783 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1784 		    NFSACCCHK_VPISLOCKED, NULL);
1785 	if (nd->nd_repstat) {
1786 		vput(vp);
1787 		if (nd->nd_flag & ND_NFSV3)
1788 			nfsrv_postopattr(nd, getret, &at);
1789 		return (0);
1790 	}
1791 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1792 
1793 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1794 again:
1795 	eofflag = 0;
1796 	if (cookies) {
1797 		free((caddr_t)cookies, M_TEMP);
1798 		cookies = NULL;
1799 	}
1800 
1801 	iv.iov_base = rbuf;
1802 	iv.iov_len = siz;
1803 	io.uio_iov = &iv;
1804 	io.uio_iovcnt = 1;
1805 	io.uio_offset = (off_t)off;
1806 	io.uio_resid = siz;
1807 	io.uio_segflg = UIO_SYSSPACE;
1808 	io.uio_rw = UIO_READ;
1809 	io.uio_td = NULL;
1810 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1811 	    &cookies);
1812 	off = (u_int64_t)io.uio_offset;
1813 	if (io.uio_resid)
1814 		siz -= io.uio_resid;
1815 
1816 	getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1817 
1818 	if (!cookies && !nd->nd_repstat)
1819 		nd->nd_repstat = NFSERR_PERM;
1820 	if (!nd->nd_repstat)
1821 		nd->nd_repstat = getret;
1822 	if (nd->nd_repstat) {
1823 		vput(vp);
1824 		if (cookies)
1825 			free((caddr_t)cookies, M_TEMP);
1826 		free((caddr_t)rbuf, M_TEMP);
1827 		if (nd->nd_flag & ND_NFSV3)
1828 			nfsrv_postopattr(nd, getret, &at);
1829 		return (0);
1830 	}
1831 	/*
1832 	 * If nothing read, return eof
1833 	 * rpc reply
1834 	 */
1835 	if (siz == 0) {
1836 		vput(vp);
1837 		if (nd->nd_flag & ND_NFSV3)
1838 			nfsrv_postopattr(nd, getret, &at);
1839 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1840 		txdr_hyper(at.na_filerev, tl);
1841 		tl += 2;
1842 		*tl++ = newnfs_false;
1843 		*tl = newnfs_true;
1844 		free((caddr_t)cookies, M_TEMP);
1845 		free((caddr_t)rbuf, M_TEMP);
1846 		return (0);
1847 	}
1848 
1849 	/*
1850 	 * Check for degenerate cases of nothing useful read.
1851 	 * If so go try again
1852 	 */
1853 	cpos = rbuf;
1854 	cend = rbuf + siz;
1855 	dp = (struct dirent *)cpos;
1856 	cookiep = cookies;
1857 
1858 	/*
1859 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1860 	 * directory offset up to a block boundary, so it is necessary to
1861 	 * skip over the records that precede the requested offset. This
1862 	 * requires the assumption that file offset cookies monotonically
1863 	 * increase.
1864 	 * Since the offset cookies don't monotonically increase for ZFS,
1865 	 * this is not done when ZFS is the file system.
1866 	 */
1867 	while (cpos < cend && ncookies > 0 &&
1868 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1869 	   (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff) ||
1870 	   ((nd->nd_flag & ND_NFSV4) &&
1871 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1872 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1873 		cpos += dp->d_reclen;
1874 		dp = (struct dirent *)cpos;
1875 		cookiep++;
1876 		ncookies--;
1877 	}
1878 	if (cpos >= cend || ncookies == 0) {
1879 		siz = fullsiz;
1880 		toff = off;
1881 		goto again;
1882 	}
1883 
1884 	/*
1885 	 * Busy the file system so that the mount point won't go away
1886 	 * and, as such, VFS_VGET() can be used safely.
1887 	 */
1888 	mp = vp->v_mount;
1889 	vfs_ref(mp);
1890 	VOP_UNLOCK(vp, 0);
1891 	nd->nd_repstat = vfs_busy(mp, 0);
1892 	vfs_rel(mp);
1893 	if (nd->nd_repstat != 0) {
1894 		vrele(vp);
1895 		free(cookies, M_TEMP);
1896 		free(rbuf, M_TEMP);
1897 		if (nd->nd_flag & ND_NFSV3)
1898 			nfsrv_postopattr(nd, getret, &at);
1899 		return (0);
1900 	}
1901 
1902 	/*
1903 	 * Save this position, in case there is an error before one entry
1904 	 * is created.
1905 	 */
1906 	mb0 = nd->nd_mb;
1907 	bpos0 = nd->nd_bpos;
1908 
1909 	/*
1910 	 * Fill in the first part of the reply.
1911 	 * dirlen is the reply length in bytes and cannot exceed cnt.
1912 	 * (Include the two booleans at the end of the reply in dirlen now,
1913 	 *  so we recognize when we have exceeded cnt.)
1914 	 */
1915 	if (nd->nd_flag & ND_NFSV3) {
1916 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1917 		nfsrv_postopattr(nd, getret, &at);
1918 	} else {
1919 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
1920 	}
1921 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
1922 	txdr_hyper(at.na_filerev, tl);
1923 
1924 	/*
1925 	 * Save this position, in case there is an empty reply needed.
1926 	 */
1927 	mb1 = nd->nd_mb;
1928 	bpos1 = nd->nd_bpos;
1929 
1930 	/* Loop through the records and build reply */
1931 	entrycnt = 0;
1932 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
1933 		nlen = dp->d_namlen;
1934 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1935 		    nlen <= NFS_MAXNAMLEN &&
1936 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
1937 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
1938 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
1939 			/*
1940 			 * Save the current position in the reply, in case
1941 			 * this entry exceeds cnt.
1942 			 */
1943 			mb1 = nd->nd_mb;
1944 			bpos1 = nd->nd_bpos;
1945 
1946 			/*
1947 			 * For readdir_and_lookup get the vnode using
1948 			 * the file number.
1949 			 */
1950 			nvp = NULL;
1951 			refp = NULL;
1952 			r = 0;
1953 			at_root = 0;
1954 			needs_unbusy = 0;
1955 			new_mp = mp;
1956 			mounted_on_fileno = (uint64_t)dp->d_fileno;
1957 			if ((nd->nd_flag & ND_NFSV3) ||
1958 			    NFSNONZERO_ATTRBIT(&savbits)) {
1959 				if (nd->nd_flag & ND_NFSV4)
1960 					refp = nfsv4root_getreferral(NULL,
1961 					    vp, dp->d_fileno);
1962 				if (refp == NULL) {
1963 					if (usevget)
1964 						r = VFS_VGET(mp, dp->d_fileno,
1965 						    LK_SHARED, &nvp);
1966 					else
1967 						r = EOPNOTSUPP;
1968 					if (r == EOPNOTSUPP) {
1969 						if (usevget) {
1970 							usevget = 0;
1971 							cn.cn_nameiop = LOOKUP;
1972 							cn.cn_lkflags =
1973 							    LK_SHARED |
1974 							    LK_RETRY;
1975 							cn.cn_cred =
1976 							    nd->nd_cred;
1977 							cn.cn_thread = p;
1978 						}
1979 						cn.cn_nameptr = dp->d_name;
1980 						cn.cn_namelen = nlen;
1981 						cn.cn_flags = ISLASTCN |
1982 						    NOFOLLOW | LOCKLEAF |
1983 						    MPSAFE;
1984 						if (nlen == 2 &&
1985 						    dp->d_name[0] == '.' &&
1986 						    dp->d_name[1] == '.')
1987 							cn.cn_flags |=
1988 							    ISDOTDOT;
1989 						if (vn_lock(vp, LK_SHARED)
1990 						    != 0) {
1991 							nd->nd_repstat = EPERM;
1992 							break;
1993 						}
1994 						if ((vp->v_vflag & VV_ROOT) != 0
1995 						    && (cn.cn_flags & ISDOTDOT)
1996 						    != 0) {
1997 							vref(vp);
1998 							nvp = vp;
1999 							r = 0;
2000 						} else {
2001 							r = VOP_LOOKUP(vp, &nvp,
2002 							    &cn);
2003 							if (vp != nvp)
2004 								VOP_UNLOCK(vp,
2005 								    0);
2006 						}
2007 					}
2008 
2009 					/*
2010 					 * For NFSv4, check to see if nvp is
2011 					 * a mount point and get the mount
2012 					 * point vnode, as required.
2013 					 */
2014 					if (r == 0 &&
2015 					    nfsrv_enable_crossmntpt != 0 &&
2016 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2017 					    nvp->v_type == VDIR &&
2018 					    nvp->v_mountedhere != NULL) {
2019 						new_mp = nvp->v_mountedhere;
2020 						r = vfs_busy(new_mp, 0);
2021 						vput(nvp);
2022 						nvp = NULL;
2023 						if (r == 0) {
2024 							r = VFS_ROOT(new_mp,
2025 							    LK_SHARED, &nvp);
2026 							needs_unbusy = 1;
2027 							if (r == 0)
2028 								at_root = 1;
2029 						}
2030 					}
2031 				}
2032 				if (!r) {
2033 				    if (refp == NULL &&
2034 					((nd->nd_flag & ND_NFSV3) ||
2035 					 NFSNONZERO_ATTRBIT(&attrbits))) {
2036 					r = nfsvno_getfh(nvp, &nfh, p);
2037 					if (!r)
2038 					    r = nfsvno_getattr(nvp, nvap,
2039 						nd->nd_cred, p, 1);
2040 				    }
2041 				} else {
2042 				    nvp = NULL;
2043 				}
2044 				if (r) {
2045 					if (!NFSISSET_ATTRBIT(&attrbits,
2046 					    NFSATTRBIT_RDATTRERROR)) {
2047 						if (nvp != NULL)
2048 							vput(nvp);
2049 						if (needs_unbusy != 0)
2050 							vfs_unbusy(new_mp);
2051 						nd->nd_repstat = r;
2052 						break;
2053 					}
2054 				}
2055 			}
2056 
2057 			/*
2058 			 * Build the directory record xdr
2059 			 */
2060 			if (nd->nd_flag & ND_NFSV3) {
2061 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2062 				*tl++ = newnfs_true;
2063 				*tl++ = 0;
2064 				*tl = txdr_unsigned(dp->d_fileno);
2065 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2066 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2067 				*tl++ = 0;
2068 				*tl = txdr_unsigned(*cookiep);
2069 				nfsrv_postopattr(nd, 0, nvap);
2070 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2071 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2072 				if (nvp != NULL)
2073 					vput(nvp);
2074 			} else {
2075 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2076 				*tl++ = newnfs_true;
2077 				*tl++ = 0;
2078 				*tl = txdr_unsigned(*cookiep);
2079 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2080 				if (nvp != NULL) {
2081 					supports_nfsv4acls =
2082 					    nfs_supportsnfsv4acls(nvp);
2083 					VOP_UNLOCK(nvp, 0);
2084 				} else
2085 					supports_nfsv4acls = 0;
2086 				if (refp != NULL) {
2087 					dirlen += nfsrv_putreferralattr(nd,
2088 					    &savbits, refp, 0,
2089 					    &nd->nd_repstat);
2090 					if (nd->nd_repstat) {
2091 						if (nvp != NULL)
2092 							vrele(nvp);
2093 						if (needs_unbusy != 0)
2094 							vfs_unbusy(new_mp);
2095 						break;
2096 					}
2097 				} else if (r) {
2098 					dirlen += nfsvno_fillattr(nd, new_mp,
2099 					    nvp, nvap, &nfh, r, &rderrbits,
2100 					    nd->nd_cred, p, isdgram, 0,
2101 					    supports_nfsv4acls, at_root,
2102 					    mounted_on_fileno);
2103 				} else {
2104 					dirlen += nfsvno_fillattr(nd, new_mp,
2105 					    nvp, nvap, &nfh, r, &attrbits,
2106 					    nd->nd_cred, p, isdgram, 0,
2107 					    supports_nfsv4acls, at_root,
2108 					    mounted_on_fileno);
2109 				}
2110 				if (nvp != NULL)
2111 					vrele(nvp);
2112 				dirlen += (3 * NFSX_UNSIGNED);
2113 			}
2114 			if (needs_unbusy != 0)
2115 				vfs_unbusy(new_mp);
2116 			if (dirlen <= cnt)
2117 				entrycnt++;
2118 		}
2119 		cpos += dp->d_reclen;
2120 		dp = (struct dirent *)cpos;
2121 		cookiep++;
2122 		ncookies--;
2123 	}
2124 	vrele(vp);
2125 	vfs_unbusy(mp);
2126 
2127 	/*
2128 	 * If dirlen > cnt, we must strip off the last entry. If that
2129 	 * results in an empty reply, report NFSERR_TOOSMALL.
2130 	 */
2131 	if (dirlen > cnt || nd->nd_repstat) {
2132 		if (!nd->nd_repstat && entrycnt == 0)
2133 			nd->nd_repstat = NFSERR_TOOSMALL;
2134 		if (nd->nd_repstat)
2135 			newnfs_trimtrailing(nd, mb0, bpos0);
2136 		else
2137 			newnfs_trimtrailing(nd, mb1, bpos1);
2138 		eofflag = 0;
2139 	} else if (cpos < cend)
2140 		eofflag = 0;
2141 	if (!nd->nd_repstat) {
2142 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2143 		*tl++ = newnfs_false;
2144 		if (eofflag)
2145 			*tl = newnfs_true;
2146 		else
2147 			*tl = newnfs_false;
2148 	}
2149 	FREE((caddr_t)cookies, M_TEMP);
2150 	FREE((caddr_t)rbuf, M_TEMP);
2151 	return (0);
2152 nfsmout:
2153 	vput(vp);
2154 	return (error);
2155 }
2156 
2157 /*
2158  * Get the settable attributes out of the mbuf list.
2159  * (Return 0 or EBADRPC)
2160  */
2161 int
2162 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2163     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2164 {
2165 	u_int32_t *tl;
2166 	struct nfsv2_sattr *sp;
2167 	struct timeval curtime;
2168 	int error = 0, toclient = 0;
2169 
2170 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2171 	case ND_NFSV2:
2172 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2173 		/*
2174 		 * Some old clients didn't fill in the high order 16bits.
2175 		 * --> check the low order 2 bytes for 0xffff
2176 		 */
2177 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2178 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2179 		if (sp->sa_uid != newnfs_xdrneg1)
2180 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2181 		if (sp->sa_gid != newnfs_xdrneg1)
2182 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2183 		if (sp->sa_size != newnfs_xdrneg1)
2184 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2185 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2186 #ifdef notyet
2187 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2188 #else
2189 			nvap->na_atime.tv_sec =
2190 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2191 			nvap->na_atime.tv_nsec = 0;
2192 #endif
2193 		}
2194 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2195 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2196 		break;
2197 	case ND_NFSV3:
2198 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2199 		if (*tl == newnfs_true) {
2200 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2201 			nvap->na_mode = nfstov_mode(*tl);
2202 		}
2203 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2204 		if (*tl == newnfs_true) {
2205 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2206 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2207 		}
2208 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2209 		if (*tl == newnfs_true) {
2210 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2211 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2212 		}
2213 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2214 		if (*tl == newnfs_true) {
2215 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2216 			nvap->na_size = fxdr_hyper(tl);
2217 		}
2218 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2219 		switch (fxdr_unsigned(int, *tl)) {
2220 		case NFSV3SATTRTIME_TOCLIENT:
2221 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2222 			fxdr_nfsv3time(tl, &nvap->na_atime);
2223 			toclient = 1;
2224 			break;
2225 		case NFSV3SATTRTIME_TOSERVER:
2226 			NFSGETTIME(&curtime);
2227 			nvap->na_atime.tv_sec = curtime.tv_sec;
2228 			nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2229 			nvap->na_vaflags |= VA_UTIMES_NULL;
2230 			break;
2231 		};
2232 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2233 		switch (fxdr_unsigned(int, *tl)) {
2234 		case NFSV3SATTRTIME_TOCLIENT:
2235 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2236 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2237 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2238 			break;
2239 		case NFSV3SATTRTIME_TOSERVER:
2240 			NFSGETTIME(&curtime);
2241 			nvap->na_mtime.tv_sec = curtime.tv_sec;
2242 			nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2243 			if (!toclient)
2244 				nvap->na_vaflags |= VA_UTIMES_NULL;
2245 			break;
2246 		};
2247 		break;
2248 	case ND_NFSV4:
2249 		error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2250 	};
2251 nfsmout:
2252 	return (error);
2253 }
2254 
2255 /*
2256  * Handle the setable attributes for V4.
2257  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2258  */
2259 int
2260 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2261     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2262 {
2263 	u_int32_t *tl;
2264 	int attrsum = 0;
2265 	int i, j;
2266 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2267 	int toclient = 0;
2268 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2269 	uid_t uid;
2270 	gid_t gid;
2271 	struct timeval curtime;
2272 
2273 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2274 	if (error)
2275 		return (error);
2276 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2277 	attrsize = fxdr_unsigned(int, *tl);
2278 
2279 	/*
2280 	 * Loop around getting the setable attributes. If an unsupported
2281 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2282 	 */
2283 	if (retnotsup) {
2284 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2285 		bitpos = NFSATTRBIT_MAX;
2286 	} else {
2287 		bitpos = 0;
2288 	}
2289 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2290 	    if (attrsum > attrsize) {
2291 		error = NFSERR_BADXDR;
2292 		goto nfsmout;
2293 	    }
2294 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2295 		switch (bitpos) {
2296 		case NFSATTRBIT_SIZE:
2297 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2298 			nvap->na_size = fxdr_hyper(tl);
2299 			attrsum += NFSX_HYPER;
2300 			break;
2301 		case NFSATTRBIT_ACL:
2302 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2303 			    p);
2304 			if (error)
2305 				goto nfsmout;
2306 			if (aceerr && !nd->nd_repstat)
2307 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2308 			attrsum += aclsize;
2309 			break;
2310 		case NFSATTRBIT_ARCHIVE:
2311 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2312 			if (!nd->nd_repstat)
2313 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2314 			attrsum += NFSX_UNSIGNED;
2315 			break;
2316 		case NFSATTRBIT_HIDDEN:
2317 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2318 			if (!nd->nd_repstat)
2319 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2320 			attrsum += NFSX_UNSIGNED;
2321 			break;
2322 		case NFSATTRBIT_MIMETYPE:
2323 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2324 			i = fxdr_unsigned(int, *tl);
2325 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2326 			if (error)
2327 				goto nfsmout;
2328 			if (!nd->nd_repstat)
2329 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2330 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2331 			break;
2332 		case NFSATTRBIT_MODE:
2333 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2334 			nvap->na_mode = nfstov_mode(*tl);
2335 			attrsum += NFSX_UNSIGNED;
2336 			break;
2337 		case NFSATTRBIT_OWNER:
2338 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2339 			j = fxdr_unsigned(int, *tl);
2340 			if (j < 0)
2341 				return (NFSERR_BADXDR);
2342 			if (j > NFSV4_SMALLSTR)
2343 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2344 			else
2345 				cp = namestr;
2346 			error = nfsrv_mtostr(nd, cp, j);
2347 			if (error) {
2348 				if (j > NFSV4_SMALLSTR)
2349 					free(cp, M_NFSSTRING);
2350 				return (error);
2351 			}
2352 			if (!nd->nd_repstat) {
2353 				nd->nd_repstat = nfsv4_strtouid(cp,j,&uid,p);
2354 				if (!nd->nd_repstat)
2355 					nvap->na_uid = uid;
2356 			}
2357 			if (j > NFSV4_SMALLSTR)
2358 				free(cp, M_NFSSTRING);
2359 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2360 			break;
2361 		case NFSATTRBIT_OWNERGROUP:
2362 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2363 			j = fxdr_unsigned(int, *tl);
2364 			if (j < 0)
2365 				return (NFSERR_BADXDR);
2366 			if (j > NFSV4_SMALLSTR)
2367 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2368 			else
2369 				cp = namestr;
2370 			error = nfsrv_mtostr(nd, cp, j);
2371 			if (error) {
2372 				if (j > NFSV4_SMALLSTR)
2373 					free(cp, M_NFSSTRING);
2374 				return (error);
2375 			}
2376 			if (!nd->nd_repstat) {
2377 				nd->nd_repstat = nfsv4_strtogid(cp,j,&gid,p);
2378 				if (!nd->nd_repstat)
2379 					nvap->na_gid = gid;
2380 			}
2381 			if (j > NFSV4_SMALLSTR)
2382 				free(cp, M_NFSSTRING);
2383 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2384 			break;
2385 		case NFSATTRBIT_SYSTEM:
2386 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2387 			if (!nd->nd_repstat)
2388 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2389 			attrsum += NFSX_UNSIGNED;
2390 			break;
2391 		case NFSATTRBIT_TIMEACCESSSET:
2392 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2393 			attrsum += NFSX_UNSIGNED;
2394 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2395 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2396 			    fxdr_nfsv4time(tl, &nvap->na_atime);
2397 			    toclient = 1;
2398 			    attrsum += NFSX_V4TIME;
2399 			} else {
2400 			    NFSGETTIME(&curtime);
2401 			    nvap->na_atime.tv_sec = curtime.tv_sec;
2402 			    nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2403 			    nvap->na_vaflags |= VA_UTIMES_NULL;
2404 			}
2405 			break;
2406 		case NFSATTRBIT_TIMEBACKUP:
2407 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2408 			if (!nd->nd_repstat)
2409 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2410 			attrsum += NFSX_V4TIME;
2411 			break;
2412 		case NFSATTRBIT_TIMECREATE:
2413 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2414 			if (!nd->nd_repstat)
2415 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2416 			attrsum += NFSX_V4TIME;
2417 			break;
2418 		case NFSATTRBIT_TIMEMODIFYSET:
2419 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2420 			attrsum += NFSX_UNSIGNED;
2421 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2422 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2423 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
2424 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
2425 			    attrsum += NFSX_V4TIME;
2426 			} else {
2427 			    NFSGETTIME(&curtime);
2428 			    nvap->na_mtime.tv_sec = curtime.tv_sec;
2429 			    nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2430 			    if (!toclient)
2431 				nvap->na_vaflags |= VA_UTIMES_NULL;
2432 			}
2433 			break;
2434 		default:
2435 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2436 			/*
2437 			 * set bitpos so we drop out of the loop.
2438 			 */
2439 			bitpos = NFSATTRBIT_MAX;
2440 			break;
2441 		};
2442 	}
2443 
2444 	/*
2445 	 * some clients pad the attrlist, so we need to skip over the
2446 	 * padding.
2447 	 */
2448 	if (attrsum > attrsize) {
2449 		error = NFSERR_BADXDR;
2450 	} else {
2451 		attrsize = NFSM_RNDUP(attrsize);
2452 		if (attrsum < attrsize)
2453 			error = nfsm_advance(nd, attrsize - attrsum, -1);
2454 	}
2455 nfsmout:
2456 	return (error);
2457 }
2458 
2459 /*
2460  * Check/setup export credentials.
2461  */
2462 int
2463 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2464     struct ucred *credanon)
2465 {
2466 	int error = 0;
2467 
2468 	/*
2469 	 * Check/setup credentials.
2470 	 */
2471 	if (nd->nd_flag & ND_GSS)
2472 		exp->nes_exflag &= ~MNT_EXPORTANON;
2473 
2474 	/*
2475 	 * Check to see if the operation is allowed for this security flavor.
2476 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2477 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2478 	 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2479 	 */
2480 	if (nfsvno_testexp(nd, exp) &&
2481 	    nd->nd_procnum != NFSV4OP_SECINFO &&
2482 	    nd->nd_procnum != NFSPROC_FSINFO) {
2483 		if (nd->nd_flag & ND_NFSV4)
2484 			error = NFSERR_WRONGSEC;
2485 		else
2486 			error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2487 		return (error);
2488 	}
2489 
2490 	/*
2491 	 * Check to see if the file system is exported V4 only.
2492 	 */
2493 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4))
2494 		return (NFSERR_PROGNOTV4);
2495 
2496 	/*
2497 	 * Now, map the user credentials.
2498 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
2499 	 *  Fsinfo RPC. If set for anything else, this code might need
2500 	 *  to change.)
2501 	 */
2502 	if (NFSVNO_EXPORTED(exp) &&
2503 	    ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2504 	     NFSVNO_EXPORTANON(exp) ||
2505 	     (nd->nd_flag & ND_AUTHNONE))) {
2506 		nd->nd_cred->cr_uid = credanon->cr_uid;
2507 		nd->nd_cred->cr_gid = credanon->cr_gid;
2508 		crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2509 		    credanon->cr_groups);
2510 	}
2511 	return (0);
2512 }
2513 
2514 /*
2515  * Check exports.
2516  */
2517 int
2518 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2519     struct ucred **credp)
2520 {
2521 	int i, error, *secflavors;
2522 
2523 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2524 	    &exp->nes_numsecflavor, &secflavors);
2525 	if (error) {
2526 		if (nfs_rootfhset) {
2527 			exp->nes_exflag = 0;
2528 			exp->nes_numsecflavor = 0;
2529 			error = 0;
2530 		}
2531 	} else {
2532 		/* Copy the security flavors. */
2533 		for (i = 0; i < exp->nes_numsecflavor; i++)
2534 			exp->nes_secflavors[i] = secflavors[i];
2535 	}
2536 	return (error);
2537 }
2538 
2539 /*
2540  * Get a vnode for a file handle and export stuff.
2541  */
2542 int
2543 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2544     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2545     struct ucred **credp)
2546 {
2547 	int i, error, *secflavors;
2548 
2549 	*credp = NULL;
2550 	exp->nes_numsecflavor = 0;
2551 	if (VFS_NEEDSGIANT(mp))
2552 		error = ESTALE;
2553 	else
2554 		error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, vpp);
2555 	if (error != 0)
2556 		/* Make sure the server replies ESTALE to the client. */
2557 		error = ESTALE;
2558 	if (nam && !error) {
2559 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2560 		    &exp->nes_numsecflavor, &secflavors);
2561 		if (error) {
2562 			if (nfs_rootfhset) {
2563 				exp->nes_exflag = 0;
2564 				exp->nes_numsecflavor = 0;
2565 				error = 0;
2566 			} else {
2567 				vput(*vpp);
2568 			}
2569 		} else {
2570 			/* Copy the security flavors. */
2571 			for (i = 0; i < exp->nes_numsecflavor; i++)
2572 				exp->nes_secflavors[i] = secflavors[i];
2573 		}
2574 	}
2575 	if (error == 0 && lktype == LK_SHARED)
2576 		/*
2577 		 * It would be much better to pass lktype to VFS_FHTOVP(),
2578 		 * but this will have to do until VFS_FHTOVP() has a lock
2579 		 * type argument like VFS_VGET().
2580 		 */
2581 		vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
2582 	return (error);
2583 }
2584 
2585 /*
2586  * Do the pathconf vnode op.
2587  */
2588 int
2589 nfsvno_pathconf(struct vnode *vp, int flag, register_t *retf,
2590     struct ucred *cred, struct thread *p)
2591 {
2592 	int error;
2593 
2594 	error = VOP_PATHCONF(vp, flag, retf);
2595 	if (error == EOPNOTSUPP || error == EINVAL) {
2596 		/*
2597 		 * Some file systems return EINVAL for name arguments not
2598 		 * supported and some return EOPNOTSUPP for this case.
2599 		 * So the NFSv3 Pathconf RPC doesn't fail for these cases,
2600 		 * just fake them.
2601 		 */
2602 		switch (flag) {
2603 		case _PC_LINK_MAX:
2604 			*retf = LINK_MAX;
2605 			break;
2606 		case _PC_NAME_MAX:
2607 			*retf = NAME_MAX;
2608 			break;
2609 		case _PC_CHOWN_RESTRICTED:
2610 			*retf = 1;
2611 			break;
2612 		case _PC_NO_TRUNC:
2613 			*retf = 1;
2614 			break;
2615 		default:
2616 			/*
2617 			 * Only happens if a _PC_xxx is added to the server,
2618 			 * but this isn't updated.
2619 			 */
2620 			*retf = 0;
2621 			printf("nfsrvd pathconf flag=%d not supp\n", flag);
2622 		};
2623 		error = 0;
2624 	}
2625 	return (error);
2626 }
2627 
2628 /*
2629  * nfsd_fhtovp() - convert a fh to a vnode ptr
2630  * 	- look up fsid in mount list (if not found ret error)
2631  *	- get vp and export rights by calling nfsvno_fhtovp()
2632  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2633  *	  for AUTH_SYS
2634  *	- if mpp != NULL, return the mount point so that it can
2635  *	  be used for vn_finished_write() by the caller
2636  */
2637 void
2638 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2639     struct vnode **vpp, struct nfsexstuff *exp,
2640     struct mount **mpp, int startwrite, struct thread *p)
2641 {
2642 	struct mount *mp;
2643 	struct ucred *credanon;
2644 	fhandle_t *fhp;
2645 
2646 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
2647 	/*
2648 	 * Check for the special case of the nfsv4root_fh.
2649 	 */
2650 	mp = vfs_busyfs(&fhp->fh_fsid);
2651 	if (mpp != NULL)
2652 		*mpp = mp;
2653 	if (mp == NULL) {
2654 		*vpp = NULL;
2655 		nd->nd_repstat = ESTALE;
2656 		return;
2657 	}
2658 
2659 	if (startwrite)
2660 		vn_start_write(NULL, mpp, V_WAIT);
2661 
2662 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2663 	    &credanon);
2664 	vfs_unbusy(mp);
2665 
2666 	/*
2667 	 * For NFSv4 without a pseudo root fs, unexported file handles
2668 	 * can be returned, so that Lookup works everywhere.
2669 	 */
2670 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2671 	    !(nd->nd_flag & ND_NFSV4)) {
2672 		vput(*vpp);
2673 		nd->nd_repstat = EACCES;
2674 	}
2675 
2676 	/*
2677 	 * Personally, I've never seen any point in requiring a
2678 	 * reserved port#, since only in the rare case where the
2679 	 * clients are all boxes with secure system priviledges,
2680 	 * does it provide any enhanced security, but... some people
2681 	 * believe it to be useful and keep putting this code back in.
2682 	 * (There is also some "security checker" out there that
2683 	 *  complains if the nfs server doesn't enforce this.)
2684 	 * However, note the following:
2685 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
2686 	 *	required.
2687 	 * RFC2623 recommends that, if a reserved port# is checked for,
2688 	 *	that there be a way to turn that off--> ifdef'd.
2689 	 */
2690 #ifdef NFS_REQRSVPORT
2691 	if (!nd->nd_repstat) {
2692 		struct sockaddr_in *saddr;
2693 		struct sockaddr_in6 *saddr6;
2694 
2695 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2696 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2697 		if (!(nd->nd_flag & ND_NFSV4) &&
2698 		    ((saddr->sin_family == AF_INET &&
2699 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2700 		     (saddr6->sin6_family == AF_INET6 &&
2701 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2702 			vput(*vpp);
2703 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2704 		}
2705 	}
2706 #endif	/* NFS_REQRSVPORT */
2707 
2708 	/*
2709 	 * Check/setup credentials.
2710 	 */
2711 	if (!nd->nd_repstat) {
2712 		nd->nd_saveduid = nd->nd_cred->cr_uid;
2713 		nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2714 		if (nd->nd_repstat)
2715 			vput(*vpp);
2716 	}
2717 	if (credanon != NULL)
2718 		crfree(credanon);
2719 	if (nd->nd_repstat) {
2720 		if (startwrite)
2721 			vn_finished_write(mp);
2722 		*vpp = NULL;
2723 		if (mpp != NULL)
2724 			*mpp = NULL;
2725 	}
2726 }
2727 
2728 /*
2729  * glue for fp.
2730  */
2731 int
2732 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2733 {
2734 	struct filedesc *fdp;
2735 	struct file *fp;
2736 
2737 	fdp = p->td_proc->p_fd;
2738 	if (fd >= fdp->fd_nfiles ||
2739 	    (fp = fdp->fd_ofiles[fd]) == NULL)
2740 		return (EBADF);
2741 	*fpp = fp;
2742 	return (0);
2743 }
2744 
2745 /*
2746  * Called from nfssvc() to update the exports list. Just call
2747  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2748  * in the mount list.
2749  */
2750 int
2751 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2752 {
2753 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2754 	int error;
2755 	struct nameidata nd;
2756 	fhandle_t fh;
2757 
2758 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2759 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2760 		nfs_rootfhset = 0;
2761 	else if (error == 0) {
2762 		if (nfsexargp->fspec == NULL)
2763 			return (EPERM);
2764 		/*
2765 		 * If fspec != NULL, this is the v4root path.
2766 		 */
2767 		NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_USERSPACE,
2768 		    nfsexargp->fspec, p);
2769 		if ((error = namei(&nd)) != 0)
2770 			return (error);
2771 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
2772 		vrele(nd.ni_vp);
2773 		if (!error) {
2774 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2775 			NFSBCOPY((caddr_t)&fh,
2776 			    nfs_rootfh.nfsrvfh_data,
2777 			    sizeof (fhandle_t));
2778 			nfs_rootfhset = 1;
2779 		}
2780 	}
2781 	return (error);
2782 }
2783 
2784 /*
2785  * Get the tcp socket sequence numbers we need.
2786  * (Maybe this should be moved to the tcp sources?)
2787  */
2788 int
2789 nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap)
2790 {
2791 	struct inpcb *inp;
2792 	struct tcpcb *tp;
2793 
2794 	inp = sotoinpcb(so);
2795 	KASSERT(inp != NULL, ("nfsrv_getsocksndseq: inp == NULL"));
2796 	INP_RLOCK(inp);
2797 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2798 		INP_RUNLOCK(inp);
2799 		return (EPIPE);
2800 	}
2801 	tp = intotcpcb(inp);
2802 	if (tp->t_state != TCPS_ESTABLISHED) {
2803 		INP_RUNLOCK(inp);
2804 		return (EPIPE);
2805 	}
2806 	*maxp = tp->snd_max;
2807 	*unap = tp->snd_una;
2808 	INP_RUNLOCK(inp);
2809 	return (0);
2810 }
2811 
2812 /*
2813  * This function needs to test to see if the system is near its limit
2814  * for memory allocation via malloc() or mget() and return True iff
2815  * either of these resources are near their limit.
2816  * XXX (For now, this is just a stub.)
2817  */
2818 int nfsrv_testmalloclimit = 0;
2819 int
2820 nfsrv_mallocmget_limit(void)
2821 {
2822 	static int printmesg = 0;
2823 	static int testval = 1;
2824 
2825 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2826 		if ((printmesg++ % 100) == 0)
2827 			printf("nfsd: malloc/mget near limit\n");
2828 		return (1);
2829 	}
2830 	return (0);
2831 }
2832 
2833 /*
2834  * BSD specific initialization of a mount point.
2835  */
2836 void
2837 nfsd_mntinit(void)
2838 {
2839 	static int inited = 0;
2840 
2841 	if (inited)
2842 		return;
2843 	inited = 1;
2844 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2845 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2846 	nfsv4root_mnt.mnt_export = NULL;
2847 	TAILQ_INIT(&nfsv4root_opt);
2848 	TAILQ_INIT(&nfsv4root_newopt);
2849 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2850 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2851 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
2852 }
2853 
2854 /*
2855  * Get a vnode for a file handle, without checking exports, etc.
2856  */
2857 struct vnode *
2858 nfsvno_getvp(fhandle_t *fhp)
2859 {
2860 	struct mount *mp;
2861 	struct vnode *vp;
2862 	int error;
2863 
2864 	mp = vfs_busyfs(&fhp->fh_fsid);
2865 	if (mp == NULL)
2866 		return (NULL);
2867 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2868 	vfs_unbusy(mp);
2869 	if (error)
2870 		return (NULL);
2871 	return (vp);
2872 }
2873 
2874 /*
2875  * Do a local VOP_ADVLOCK().
2876  */
2877 int
2878 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2879     u_int64_t end, struct thread *td)
2880 {
2881 	int error;
2882 	struct flock fl;
2883 	u_int64_t tlen;
2884 
2885 	if (nfsrv_dolocallocks == 0)
2886 		return (0);
2887 
2888 	/* Check for VI_DOOMED here, so that VOP_ADVLOCK() isn't performed. */
2889 	if ((vp->v_iflag & VI_DOOMED) != 0)
2890 		return (EPERM);
2891 
2892 	fl.l_whence = SEEK_SET;
2893 	fl.l_type = ftype;
2894 	fl.l_start = (off_t)first;
2895 	if (end == NFS64BITSSET) {
2896 		fl.l_len = 0;
2897 	} else {
2898 		tlen = end - first;
2899 		fl.l_len = (off_t)tlen;
2900 	}
2901 	/*
2902 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2903 	 * values for all calls, so that all locks will be held by the
2904 	 * nfsd server. (The nfsd server handles conflicts between the
2905 	 * various clients.)
2906 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2907 	 * bytes, so it can't be put in l_sysid.
2908 	 */
2909 	if (nfsv4_sysid == 0)
2910 		nfsv4_sysid = nlm_acquire_next_sysid();
2911 	fl.l_pid = (pid_t)0;
2912 	fl.l_sysid = (int)nfsv4_sysid;
2913 
2914 	NFSVOPUNLOCK(vp, 0, td);
2915 	if (ftype == F_UNLCK)
2916 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
2917 		    (F_POSIX | F_REMOTE));
2918 	else
2919 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
2920 		    (F_POSIX | F_REMOTE));
2921 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, td);
2922 	return (error);
2923 }
2924 
2925 /*
2926  * Check the nfsv4 root exports.
2927  */
2928 int
2929 nfsvno_v4rootexport(struct nfsrv_descript *nd)
2930 {
2931 	struct ucred *credanon;
2932 	int exflags, error, numsecflavor, *secflavors, i;
2933 
2934 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
2935 	    &credanon, &numsecflavor, &secflavors);
2936 	if (error)
2937 		return (NFSERR_PROGUNAVAIL);
2938 	if (credanon != NULL)
2939 		crfree(credanon);
2940 	for (i = 0; i < numsecflavor; i++) {
2941 		if (secflavors[i] == AUTH_SYS)
2942 			nd->nd_flag |= ND_EXAUTHSYS;
2943 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
2944 			nd->nd_flag |= ND_EXGSS;
2945 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
2946 			nd->nd_flag |= ND_EXGSSINTEGRITY;
2947 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
2948 			nd->nd_flag |= ND_EXGSSPRIVACY;
2949 	}
2950 	return (0);
2951 }
2952 
2953 /*
2954  * Nfs server psuedo system call for the nfsd's
2955  */
2956 /*
2957  * MPSAFE
2958  */
2959 static int
2960 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
2961 {
2962 	struct file *fp;
2963 	struct nfsd_addsock_args sockarg;
2964 	struct nfsd_nfsd_args nfsdarg;
2965 	int error;
2966 
2967 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
2968 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
2969 		if (error)
2970 			return (error);
2971 		if ((error = fget(td, sockarg.sock, &fp)) != 0) {
2972 			return (error);
2973 		}
2974 		if (fp->f_type != DTYPE_SOCKET) {
2975 			fdrop(fp, td);
2976 			return (EPERM);
2977 		}
2978 		error = nfsrvd_addsock(fp);
2979 		fdrop(fp, td);
2980 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
2981 		if (uap->argp == NULL)
2982 			return (EINVAL);
2983 		error = copyin(uap->argp, (caddr_t)&nfsdarg,
2984 		    sizeof (nfsdarg));
2985 		if (error)
2986 			return (error);
2987 		error = nfsrvd_nfsd(td, &nfsdarg);
2988 	} else {
2989 		error = nfssvc_srvcall(td, uap, td->td_ucred);
2990 	}
2991 	return (error);
2992 }
2993 
2994 static int
2995 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
2996 {
2997 	struct nfsex_args export;
2998 	struct file *fp = NULL;
2999 	int stablefd, len;
3000 	struct nfsd_clid adminrevoke;
3001 	struct nfsd_dumplist dumplist;
3002 	struct nfsd_dumpclients *dumpclients;
3003 	struct nfsd_dumplocklist dumplocklist;
3004 	struct nfsd_dumplocks *dumplocks;
3005 	struct nameidata nd;
3006 	vnode_t vp;
3007 	int error = EINVAL;
3008 	struct proc *procp;
3009 
3010 	if (uap->flag & NFSSVC_PUBLICFH) {
3011 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3012 		    sizeof (fhandle_t));
3013 		error = copyin(uap->argp,
3014 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3015 		if (!error)
3016 			nfs_pubfhset = 1;
3017 	} else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3018 		error = copyin(uap->argp,(caddr_t)&export,
3019 		    sizeof (struct nfsex_args));
3020 		if (!error)
3021 			error = nfsrv_v4rootexport(&export, cred, p);
3022 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3023 		nfs_pubfhset = 0;
3024 		error = 0;
3025 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3026 		error = copyin(uap->argp, (caddr_t)&stablefd,
3027 		    sizeof (int));
3028 		if (!error)
3029 			error = fp_getfvp(p, stablefd, &fp, &vp);
3030 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3031 			error = EBADF;
3032 		if (!error && newnfs_numnfsd != 0)
3033 			error = EPERM;
3034 		if (!error) {
3035 			nfsrv_stablefirst.nsf_fp = fp;
3036 			nfsrv_setupstable(p);
3037 		}
3038 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3039 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3040 		    sizeof (struct nfsd_clid));
3041 		if (!error)
3042 			error = nfsrv_adminrevoke(&adminrevoke, p);
3043 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3044 		error = copyin(uap->argp, (caddr_t)&dumplist,
3045 		    sizeof (struct nfsd_dumplist));
3046 		if (!error && (dumplist.ndl_size < 1 ||
3047 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3048 			error = EPERM;
3049 		if (!error) {
3050 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3051 		    dumpclients = (struct nfsd_dumpclients *)malloc(len,
3052 			M_TEMP, M_WAITOK);
3053 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3054 		    error = copyout(dumpclients,
3055 			CAST_USER_ADDR_T(dumplist.ndl_list), len);
3056 		    free((caddr_t)dumpclients, M_TEMP);
3057 		}
3058 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3059 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3060 		    sizeof (struct nfsd_dumplocklist));
3061 		if (!error && (dumplocklist.ndllck_size < 1 ||
3062 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3063 			error = EPERM;
3064 		if (!error)
3065 			error = nfsrv_lookupfilename(&nd,
3066 				dumplocklist.ndllck_fname, p);
3067 		if (!error) {
3068 			len = sizeof (struct nfsd_dumplocks) *
3069 				dumplocklist.ndllck_size;
3070 			dumplocks = (struct nfsd_dumplocks *)malloc(len,
3071 				M_TEMP, M_WAITOK);
3072 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3073 			    dumplocklist.ndllck_size, p);
3074 			vput(nd.ni_vp);
3075 			error = copyout(dumplocks,
3076 			    CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3077 			free((caddr_t)dumplocks, M_TEMP);
3078 		}
3079 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3080 		procp = p->td_proc;
3081 		PROC_LOCK(procp);
3082 		nfsd_master_pid = procp->p_pid;
3083 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3084 		nfsd_master_start = procp->p_stats->p_start;
3085 		nfsd_master_proc = procp;
3086 		PROC_UNLOCK(procp);
3087 	}
3088 	return (error);
3089 }
3090 
3091 /*
3092  * Check exports.
3093  * Returns 0 if ok, 1 otherwise.
3094  */
3095 int
3096 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3097 {
3098 	int i;
3099 
3100 	/*
3101 	 * This seems odd, but allow the case where the security flavor
3102 	 * list is empty. This happens when NFSv4 is traversing non-exported
3103 	 * file systems. Exported file systems should always have a non-empty
3104 	 * security flavor list.
3105 	 */
3106 	if (exp->nes_numsecflavor == 0)
3107 		return (0);
3108 
3109 	for (i = 0; i < exp->nes_numsecflavor; i++) {
3110 		/*
3111 		 * The tests for privacy and integrity must be first,
3112 		 * since ND_GSS is set for everything but AUTH_SYS.
3113 		 */
3114 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3115 		    (nd->nd_flag & ND_GSSPRIVACY))
3116 			return (0);
3117 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3118 		    (nd->nd_flag & ND_GSSINTEGRITY))
3119 			return (0);
3120 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3121 		    (nd->nd_flag & ND_GSS))
3122 			return (0);
3123 		if (exp->nes_secflavors[i] == AUTH_SYS &&
3124 		    (nd->nd_flag & ND_GSS) == 0)
3125 			return (0);
3126 	}
3127 	return (1);
3128 }
3129 
3130 /*
3131  * Calculate a hash value for the fid in a file handle.
3132  */
3133 uint32_t
3134 nfsrv_hashfh(fhandle_t *fhp)
3135 {
3136 	uint32_t hashval;
3137 
3138 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3139 	return (hashval);
3140 }
3141 
3142 /*
3143  * Signal the userland master nfsd to backup the stable restart file.
3144  */
3145 void
3146 nfsrv_backupstable(void)
3147 {
3148 	struct proc *procp;
3149 
3150 	if (nfsd_master_proc != NULL) {
3151 		procp = pfind(nfsd_master_pid);
3152 		/* Try to make sure it is the correct process. */
3153 		if (procp == nfsd_master_proc &&
3154 		    procp->p_stats->p_start.tv_sec ==
3155 		    nfsd_master_start.tv_sec &&
3156 		    procp->p_stats->p_start.tv_usec ==
3157 		    nfsd_master_start.tv_usec &&
3158 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
3159 			psignal(procp, SIGUSR2);
3160 		else
3161 			nfsd_master_proc = NULL;
3162 
3163 		if (procp != NULL)
3164 			PROC_UNLOCK(procp);
3165 	}
3166 }
3167 
3168 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3169 
3170 /*
3171  * Called once to initialize data structures...
3172  */
3173 static int
3174 nfsd_modevent(module_t mod, int type, void *data)
3175 {
3176 	int error = 0;
3177 	static int loaded = 0;
3178 
3179 	switch (type) {
3180 	case MOD_LOAD:
3181 		if (loaded)
3182 			return (0);
3183 		newnfs_portinit();
3184 		mtx_init(&nfs_cache_mutex, "nfs_cache_mutex", NULL, MTX_DEF);
3185 		mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3186 		mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3187 		    MTX_DEF);
3188 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3189 		nfsrvd_initcache();
3190 		nfsd_init();
3191 		NFSD_LOCK();
3192 		nfsrvd_init(0);
3193 		NFSD_UNLOCK();
3194 		nfsd_mntinit();
3195 #ifdef VV_DISABLEDELEG
3196 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3197 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3198 #endif
3199 		nfsd_call_servertimer = nfsrv_servertimer;
3200 		nfsd_call_nfsd = nfssvc_nfsd;
3201 		loaded = 1;
3202 		break;
3203 
3204 	case MOD_UNLOAD:
3205 		if (newnfs_numnfsd != 0) {
3206 			error = EBUSY;
3207 			break;
3208 		}
3209 
3210 #ifdef VV_DISABLEDELEG
3211 		vn_deleg_ops.vndeleg_recall = NULL;
3212 		vn_deleg_ops.vndeleg_disable = NULL;
3213 #endif
3214 		nfsd_call_servertimer = NULL;
3215 		nfsd_call_nfsd = NULL;
3216 
3217 		/* Clean out all NFSv4 state. */
3218 		nfsrv_throwawayallstate(curthread);
3219 
3220 		/* Clean the NFS server reply cache */
3221 		nfsrvd_cleancache();
3222 
3223 		/* Free up the krpc server pool. */
3224 		if (nfsrvd_pool != NULL)
3225 			svcpool_destroy(nfsrvd_pool);
3226 
3227 		/* and get rid of the locks */
3228 		mtx_destroy(&nfs_cache_mutex);
3229 		mtx_destroy(&nfs_v4root_mutex);
3230 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3231 		lockdestroy(&nfsv4root_mnt.mnt_explock);
3232 		loaded = 0;
3233 		break;
3234 	default:
3235 		error = EOPNOTSUPP;
3236 		break;
3237 	}
3238 	return error;
3239 }
3240 static moduledata_t nfsd_mod = {
3241 	"nfsd",
3242 	nfsd_modevent,
3243 	NULL,
3244 };
3245 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3246 
3247 /* So that loader and kldload(2) can find us, wherever we are.. */
3248 MODULE_VERSION(nfsd, 1);
3249 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3250 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3251 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3252 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3253 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3254 
3255