xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision 10b59a9b4add0320d52c15ce057dd697261e7dfc)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include <sys/capability.h>
38 
39 /*
40  * Functions that perform the vfs operations required by the routines in
41  * nfsd_serv.c. It is hoped that this change will make the server more
42  * portable.
43  */
44 
45 #include <fs/nfs/nfsport.h>
46 #include <sys/hash.h>
47 #include <sys/sysctl.h>
48 #include <nlm/nlm_prot.h>
49 #include <nlm/nlm.h>
50 
51 FEATURE(nfsd, "NFSv4 server");
52 
53 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
54 extern int nfsrv_useacl;
55 extern int newnfs_numnfsd;
56 extern struct mount nfsv4root_mnt;
57 extern struct nfsrv_stablefirst nfsrv_stablefirst;
58 extern void (*nfsd_call_servertimer)(void);
59 extern SVCPOOL	*nfsrvd_pool;
60 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
61 NFSDLOCKMUTEX;
62 struct mtx nfs_cache_mutex;
63 struct mtx nfs_v4root_mutex;
64 struct nfsrvfh nfs_rootfh, nfs_pubfh;
65 int nfs_pubfhset = 0, nfs_rootfhset = 0;
66 struct proc *nfsd_master_proc = NULL;
67 static pid_t nfsd_master_pid = (pid_t)-1;
68 static char nfsd_master_comm[MAXCOMLEN + 1];
69 static struct timeval nfsd_master_start;
70 static uint32_t nfsv4_sysid = 0;
71 
72 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
73     struct ucred *);
74 
75 int nfsrv_enable_crossmntpt = 1;
76 static int nfs_commit_blks;
77 static int nfs_commit_miss;
78 extern int nfsrv_issuedelegs;
79 extern int nfsrv_dolocallocks;
80 
81 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server");
82 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
83     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
84 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
85     0, "");
86 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
87     0, "");
88 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
89     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
90 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
91     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
92 
93 #define	NUM_HEURISTIC		1017
94 #define	NHUSE_INIT		64
95 #define	NHUSE_INC		16
96 #define	NHUSE_MAX		2048
97 
98 static struct nfsheur {
99 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
100 	off_t nh_nextr;		/* next offset for sequential detection */
101 	int nh_use;		/* use count for selection */
102 	int nh_seqcount;	/* heuristic */
103 } nfsheur[NUM_HEURISTIC];
104 
105 
106 /*
107  * Get attributes into nfsvattr structure.
108  */
109 int
110 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
111     struct thread *p, int vpislocked)
112 {
113 	int error, lockedit = 0;
114 
115 	if (vpislocked == 0) {
116 		/*
117 		 * When vpislocked == 0, the vnode is either exclusively
118 		 * locked by this thread or not locked by this thread.
119 		 * As such, shared lock it, if not exclusively locked.
120 		 */
121 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
122 			lockedit = 1;
123 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
124 		}
125 	}
126 	error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
127 	if (lockedit != 0)
128 		NFSVOPUNLOCK(vp, 0);
129 
130 	NFSEXITCODE(error);
131 	return (error);
132 }
133 
134 /*
135  * Get a file handle for a vnode.
136  */
137 int
138 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
139 {
140 	int error;
141 
142 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
143 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
144 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
145 
146 	NFSEXITCODE(error);
147 	return (error);
148 }
149 
150 /*
151  * Perform access checking for vnodes obtained from file handles that would
152  * refer to files already opened by a Unix client. You cannot just use
153  * vn_writechk() and VOP_ACCESSX() for two reasons.
154  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
155  *     case.
156  * 2 - The owner is to be given access irrespective of mode bits for some
157  *     operations, so that processes that chmod after opening a file don't
158  *     break.
159  */
160 int
161 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
162     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
163     u_int32_t *supportedtypep)
164 {
165 	struct vattr vattr;
166 	int error = 0, getret = 0;
167 
168 	if (vpislocked == 0) {
169 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
170 			error = EPERM;
171 			goto out;
172 		}
173 	}
174 	if (accmode & VWRITE) {
175 		/* Just vn_writechk() changed to check rdonly */
176 		/*
177 		 * Disallow write attempts on read-only file systems;
178 		 * unless the file is a socket or a block or character
179 		 * device resident on the file system.
180 		 */
181 		if (NFSVNO_EXRDONLY(exp) ||
182 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
183 			switch (vp->v_type) {
184 			case VREG:
185 			case VDIR:
186 			case VLNK:
187 				error = EROFS;
188 			default:
189 				break;
190 			}
191 		}
192 		/*
193 		 * If there's shared text associated with
194 		 * the inode, try to free it up once.  If
195 		 * we fail, we can't allow writing.
196 		 */
197 		if ((vp->v_vflag & VV_TEXT) != 0 && error == 0)
198 			error = ETXTBSY;
199 	}
200 	if (error != 0) {
201 		if (vpislocked == 0)
202 			NFSVOPUNLOCK(vp, 0);
203 		goto out;
204 	}
205 
206 	/*
207 	 * Should the override still be applied when ACLs are enabled?
208 	 */
209 	error = VOP_ACCESSX(vp, accmode, cred, p);
210 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
211 		/*
212 		 * Try again with VEXPLICIT_DENY, to see if the test for
213 		 * deletion is supported.
214 		 */
215 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
216 		if (error == 0) {
217 			if (vp->v_type == VDIR) {
218 				accmode &= ~(VDELETE | VDELETE_CHILD);
219 				accmode |= VWRITE;
220 				error = VOP_ACCESSX(vp, accmode, cred, p);
221 			} else if (supportedtypep != NULL) {
222 				*supportedtypep &= ~NFSACCESS_DELETE;
223 			}
224 		}
225 	}
226 
227 	/*
228 	 * Allow certain operations for the owner (reads and writes
229 	 * on files that are already open).
230 	 */
231 	if (override != NFSACCCHK_NOOVERRIDE &&
232 	    (error == EPERM || error == EACCES)) {
233 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
234 			error = 0;
235 		else if (override & NFSACCCHK_ALLOWOWNER) {
236 			getret = VOP_GETATTR(vp, &vattr, cred);
237 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
238 				error = 0;
239 		}
240 	}
241 	if (vpislocked == 0)
242 		NFSVOPUNLOCK(vp, 0);
243 
244 out:
245 	NFSEXITCODE(error);
246 	return (error);
247 }
248 
249 /*
250  * Set attribute(s) vnop.
251  */
252 int
253 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
254     struct thread *p, struct nfsexstuff *exp)
255 {
256 	int error;
257 
258 	error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
259 	NFSEXITCODE(error);
260 	return (error);
261 }
262 
263 /*
264  * Set up nameidata for a lookup() call and do it
265  * For the cases where we are crossing mount points
266  * (looking up the public fh path or the v4 root path when
267  *  not using a pseudo-root fs), set/release the Giant lock,
268  * as required.
269  */
270 int
271 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
272     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
273     struct vnode **retdirp)
274 {
275 	struct componentname *cnp = &ndp->ni_cnd;
276 	int i;
277 	struct iovec aiov;
278 	struct uio auio;
279 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
280 	int error = 0, crossmnt;
281 	char *cp;
282 
283 	*retdirp = NULL;
284 	cnp->cn_nameptr = cnp->cn_pnbuf;
285 	ndp->ni_strictrelative = 0;
286 	/*
287 	 * Extract and set starting directory.
288 	 */
289 	if (dp->v_type != VDIR) {
290 		if (islocked)
291 			vput(dp);
292 		else
293 			vrele(dp);
294 		nfsvno_relpathbuf(ndp);
295 		error = ENOTDIR;
296 		goto out1;
297 	}
298 	if (islocked)
299 		NFSVOPUNLOCK(dp, 0);
300 	VREF(dp);
301 	*retdirp = dp;
302 	if (NFSVNO_EXRDONLY(exp))
303 		cnp->cn_flags |= RDONLY;
304 	ndp->ni_segflg = UIO_SYSSPACE;
305 	crossmnt = 1;
306 
307 	if (nd->nd_flag & ND_PUBLOOKUP) {
308 		ndp->ni_loopcnt = 0;
309 		if (cnp->cn_pnbuf[0] == '/') {
310 			vrele(dp);
311 			/*
312 			 * Check for degenerate pathnames here, since lookup()
313 			 * panics on them.
314 			 */
315 			for (i = 1; i < ndp->ni_pathlen; i++)
316 				if (cnp->cn_pnbuf[i] != '/')
317 					break;
318 			if (i == ndp->ni_pathlen) {
319 				error = NFSERR_ACCES;
320 				goto out;
321 			}
322 			dp = rootvnode;
323 			VREF(dp);
324 		}
325 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
326 	    (nd->nd_flag & ND_NFSV4) == 0) {
327 		/*
328 		 * Only cross mount points for NFSv4 when doing a
329 		 * mount while traversing the file system above
330 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
331 		 */
332 		cnp->cn_flags |= NOCROSSMOUNT;
333 		crossmnt = 0;
334 	}
335 
336 	/*
337 	 * Initialize for scan, set ni_startdir and bump ref on dp again
338 	 * becuase lookup() will dereference ni_startdir.
339 	 */
340 
341 	cnp->cn_thread = p;
342 	ndp->ni_startdir = dp;
343 	ndp->ni_rootdir = rootvnode;
344 
345 	if (!lockleaf)
346 		cnp->cn_flags |= LOCKLEAF;
347 	for (;;) {
348 		cnp->cn_nameptr = cnp->cn_pnbuf;
349 		/*
350 		 * Call lookup() to do the real work.  If an error occurs,
351 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
352 		 * we do not have to dereference anything before returning.
353 		 * In either case ni_startdir will be dereferenced and NULLed
354 		 * out.
355 		 */
356 		error = lookup(ndp);
357 		if (error)
358 			break;
359 
360 		/*
361 		 * Check for encountering a symbolic link.  Trivial
362 		 * termination occurs if no symlink encountered.
363 		 */
364 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
365 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
366 				nfsvno_relpathbuf(ndp);
367 			if (ndp->ni_vp && !lockleaf)
368 				NFSVOPUNLOCK(ndp->ni_vp, 0);
369 			break;
370 		}
371 
372 		/*
373 		 * Validate symlink
374 		 */
375 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
376 			NFSVOPUNLOCK(ndp->ni_dvp, 0);
377 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
378 			error = EINVAL;
379 			goto badlink2;
380 		}
381 
382 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
383 			error = ELOOP;
384 			goto badlink2;
385 		}
386 		if (ndp->ni_pathlen > 1)
387 			cp = uma_zalloc(namei_zone, M_WAITOK);
388 		else
389 			cp = cnp->cn_pnbuf;
390 		aiov.iov_base = cp;
391 		aiov.iov_len = MAXPATHLEN;
392 		auio.uio_iov = &aiov;
393 		auio.uio_iovcnt = 1;
394 		auio.uio_offset = 0;
395 		auio.uio_rw = UIO_READ;
396 		auio.uio_segflg = UIO_SYSSPACE;
397 		auio.uio_td = NULL;
398 		auio.uio_resid = MAXPATHLEN;
399 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
400 		if (error) {
401 		badlink1:
402 			if (ndp->ni_pathlen > 1)
403 				uma_zfree(namei_zone, cp);
404 		badlink2:
405 			vrele(ndp->ni_dvp);
406 			vput(ndp->ni_vp);
407 			break;
408 		}
409 		linklen = MAXPATHLEN - auio.uio_resid;
410 		if (linklen == 0) {
411 			error = ENOENT;
412 			goto badlink1;
413 		}
414 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
415 			error = ENAMETOOLONG;
416 			goto badlink1;
417 		}
418 
419 		/*
420 		 * Adjust or replace path
421 		 */
422 		if (ndp->ni_pathlen > 1) {
423 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
424 			uma_zfree(namei_zone, cnp->cn_pnbuf);
425 			cnp->cn_pnbuf = cp;
426 		} else
427 			cnp->cn_pnbuf[linklen] = '\0';
428 		ndp->ni_pathlen += linklen;
429 
430 		/*
431 		 * Cleanup refs for next loop and check if root directory
432 		 * should replace current directory.  Normally ni_dvp
433 		 * becomes the new base directory and is cleaned up when
434 		 * we loop.  Explicitly null pointers after invalidation
435 		 * to clarify operation.
436 		 */
437 		vput(ndp->ni_vp);
438 		ndp->ni_vp = NULL;
439 
440 		if (cnp->cn_pnbuf[0] == '/') {
441 			vrele(ndp->ni_dvp);
442 			ndp->ni_dvp = ndp->ni_rootdir;
443 			VREF(ndp->ni_dvp);
444 		}
445 		ndp->ni_startdir = ndp->ni_dvp;
446 		ndp->ni_dvp = NULL;
447 	}
448 	if (!lockleaf)
449 		cnp->cn_flags &= ~LOCKLEAF;
450 
451 out:
452 	if (error) {
453 		uma_zfree(namei_zone, cnp->cn_pnbuf);
454 		ndp->ni_vp = NULL;
455 		ndp->ni_dvp = NULL;
456 		ndp->ni_startdir = NULL;
457 		cnp->cn_flags &= ~HASBUF;
458 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
459 		ndp->ni_dvp = NULL;
460 	}
461 
462 out1:
463 	NFSEXITCODE2(error, nd);
464 	return (error);
465 }
466 
467 /*
468  * Set up a pathname buffer and return a pointer to it and, optionally
469  * set a hash pointer.
470  */
471 void
472 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
473 {
474 	struct componentname *cnp = &ndp->ni_cnd;
475 
476 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
477 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
478 	if (hashpp != NULL)
479 		*hashpp = NULL;
480 	*bufpp = cnp->cn_pnbuf;
481 }
482 
483 /*
484  * Release the above path buffer, if not released by nfsvno_namei().
485  */
486 void
487 nfsvno_relpathbuf(struct nameidata *ndp)
488 {
489 
490 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
491 		panic("nfsrelpath");
492 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
493 	ndp->ni_cnd.cn_flags &= ~HASBUF;
494 }
495 
496 /*
497  * Readlink vnode op into an mbuf list.
498  */
499 int
500 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
501     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
502 {
503 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
504 	struct iovec *ivp = iv;
505 	struct uio io, *uiop = &io;
506 	struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
507 	int i, len, tlen, error = 0;
508 
509 	len = 0;
510 	i = 0;
511 	while (len < NFS_MAXPATHLEN) {
512 		NFSMGET(mp);
513 		MCLGET(mp, M_WAIT);
514 		mp->m_len = NFSMSIZ(mp);
515 		if (len == 0) {
516 			mp3 = mp2 = mp;
517 		} else {
518 			mp2->m_next = mp;
519 			mp2 = mp;
520 		}
521 		if ((len + mp->m_len) > NFS_MAXPATHLEN) {
522 			mp->m_len = NFS_MAXPATHLEN - len;
523 			len = NFS_MAXPATHLEN;
524 		} else {
525 			len += mp->m_len;
526 		}
527 		ivp->iov_base = mtod(mp, caddr_t);
528 		ivp->iov_len = mp->m_len;
529 		i++;
530 		ivp++;
531 	}
532 	uiop->uio_iov = iv;
533 	uiop->uio_iovcnt = i;
534 	uiop->uio_offset = 0;
535 	uiop->uio_resid = len;
536 	uiop->uio_rw = UIO_READ;
537 	uiop->uio_segflg = UIO_SYSSPACE;
538 	uiop->uio_td = NULL;
539 	error = VOP_READLINK(vp, uiop, cred);
540 	if (error) {
541 		m_freem(mp3);
542 		*lenp = 0;
543 		goto out;
544 	}
545 	if (uiop->uio_resid > 0) {
546 		len -= uiop->uio_resid;
547 		tlen = NFSM_RNDUP(len);
548 		nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
549 	}
550 	*lenp = len;
551 	*mpp = mp3;
552 	*mpendp = mp;
553 
554 out:
555 	NFSEXITCODE(error);
556 	return (error);
557 }
558 
559 /*
560  * Read vnode op call into mbuf list.
561  */
562 int
563 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
564     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
565 {
566 	struct mbuf *m;
567 	int i;
568 	struct iovec *iv;
569 	struct iovec *iv2;
570 	int error = 0, len, left, siz, tlen, ioflag = 0, hi, try = 32;
571 	struct mbuf *m2 = NULL, *m3;
572 	struct uio io, *uiop = &io;
573 	struct nfsheur *nh;
574 
575 	/*
576 	 * Calculate seqcount for heuristic
577 	 */
578 	/*
579 	 * Locate best candidate
580 	 */
581 
582 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
583 	nh = &nfsheur[hi];
584 
585 	while (try--) {
586 		if (nfsheur[hi].nh_vp == vp) {
587 			nh = &nfsheur[hi];
588 			break;
589 		}
590 		if (nfsheur[hi].nh_use > 0)
591 			--nfsheur[hi].nh_use;
592 		hi = (hi + 1) % NUM_HEURISTIC;
593 		if (nfsheur[hi].nh_use < nh->nh_use)
594 			nh = &nfsheur[hi];
595 	}
596 
597 	if (nh->nh_vp != vp) {
598 		nh->nh_vp = vp;
599 		nh->nh_nextr = off;
600 		nh->nh_use = NHUSE_INIT;
601 		if (off == 0)
602 			nh->nh_seqcount = 4;
603 		else
604 			nh->nh_seqcount = 1;
605 	}
606 
607 	/*
608 	 * Calculate heuristic
609 	 */
610 
611 	if ((off == 0 && nh->nh_seqcount > 0) || off == nh->nh_nextr) {
612 		if (++nh->nh_seqcount > IO_SEQMAX)
613 			nh->nh_seqcount = IO_SEQMAX;
614 	} else if (nh->nh_seqcount > 1) {
615 		nh->nh_seqcount = 1;
616 	} else {
617 		nh->nh_seqcount = 0;
618 	}
619 	nh->nh_use += NHUSE_INC;
620 	if (nh->nh_use > NHUSE_MAX)
621 		nh->nh_use = NHUSE_MAX;
622 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
623 
624 	len = left = NFSM_RNDUP(cnt);
625 	m3 = NULL;
626 	/*
627 	 * Generate the mbuf list with the uio_iov ref. to it.
628 	 */
629 	i = 0;
630 	while (left > 0) {
631 		NFSMGET(m);
632 		MCLGET(m, M_WAIT);
633 		m->m_len = 0;
634 		siz = min(M_TRAILINGSPACE(m), left);
635 		left -= siz;
636 		i++;
637 		if (m3)
638 			m2->m_next = m;
639 		else
640 			m3 = m;
641 		m2 = m;
642 	}
643 	MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
644 	    M_TEMP, M_WAITOK);
645 	uiop->uio_iov = iv2 = iv;
646 	m = m3;
647 	left = len;
648 	i = 0;
649 	while (left > 0) {
650 		if (m == NULL)
651 			panic("nfsvno_read iov");
652 		siz = min(M_TRAILINGSPACE(m), left);
653 		if (siz > 0) {
654 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
655 			iv->iov_len = siz;
656 			m->m_len += siz;
657 			left -= siz;
658 			iv++;
659 			i++;
660 		}
661 		m = m->m_next;
662 	}
663 	uiop->uio_iovcnt = i;
664 	uiop->uio_offset = off;
665 	uiop->uio_resid = len;
666 	uiop->uio_rw = UIO_READ;
667 	uiop->uio_segflg = UIO_SYSSPACE;
668 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
669 	FREE((caddr_t)iv2, M_TEMP);
670 	if (error) {
671 		m_freem(m3);
672 		*mpp = NULL;
673 		goto out;
674 	}
675 	tlen = len - uiop->uio_resid;
676 	cnt = cnt < tlen ? cnt : tlen;
677 	tlen = NFSM_RNDUP(cnt);
678 	if (tlen == 0) {
679 		m_freem(m3);
680 		m3 = NULL;
681 	} else if (len != tlen || tlen != cnt)
682 		nfsrv_adj(m3, len - tlen, tlen - cnt);
683 	*mpp = m3;
684 	*mpendp = m2;
685 
686 out:
687 	NFSEXITCODE(error);
688 	return (error);
689 }
690 
691 /*
692  * Write vnode op from an mbuf list.
693  */
694 int
695 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
696     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
697 {
698 	struct iovec *ivp;
699 	int i, len;
700 	struct iovec *iv;
701 	int ioflags, error;
702 	struct uio io, *uiop = &io;
703 
704 	MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
705 	    M_WAITOK);
706 	uiop->uio_iov = iv = ivp;
707 	uiop->uio_iovcnt = cnt;
708 	i = mtod(mp, caddr_t) + mp->m_len - cp;
709 	len = retlen;
710 	while (len > 0) {
711 		if (mp == NULL)
712 			panic("nfsvno_write");
713 		if (i > 0) {
714 			i = min(i, len);
715 			ivp->iov_base = cp;
716 			ivp->iov_len = i;
717 			ivp++;
718 			len -= i;
719 		}
720 		mp = mp->m_next;
721 		if (mp) {
722 			i = mp->m_len;
723 			cp = mtod(mp, caddr_t);
724 		}
725 	}
726 
727 	if (stable == NFSWRITE_UNSTABLE)
728 		ioflags = IO_NODELOCKED;
729 	else
730 		ioflags = (IO_SYNC | IO_NODELOCKED);
731 	uiop->uio_resid = retlen;
732 	uiop->uio_rw = UIO_WRITE;
733 	uiop->uio_segflg = UIO_SYSSPACE;
734 	NFSUIOPROC(uiop, p);
735 	uiop->uio_offset = off;
736 	error = VOP_WRITE(vp, uiop, ioflags, cred);
737 	FREE((caddr_t)iv, M_TEMP);
738 
739 	NFSEXITCODE(error);
740 	return (error);
741 }
742 
743 /*
744  * Common code for creating a regular file (plus special files for V2).
745  */
746 int
747 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
748     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
749     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
750 {
751 	u_quad_t tempsize;
752 	int error;
753 
754 	error = nd->nd_repstat;
755 	if (!error && ndp->ni_vp == NULL) {
756 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
757 			vrele(ndp->ni_startdir);
758 			error = VOP_CREATE(ndp->ni_dvp,
759 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
760 			vput(ndp->ni_dvp);
761 			nfsvno_relpathbuf(ndp);
762 			if (!error) {
763 				if (*exclusive_flagp) {
764 					*exclusive_flagp = 0;
765 					NFSVNO_ATTRINIT(nvap);
766 					nvap->na_atime.tv_sec = cverf[0];
767 					nvap->na_atime.tv_nsec = cverf[1];
768 					error = VOP_SETATTR(ndp->ni_vp,
769 					    &nvap->na_vattr, nd->nd_cred);
770 				}
771 			}
772 		/*
773 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
774 		 * (This implies, just get out on an error.)
775 		 */
776 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
777 			nvap->na_type == VFIFO) {
778 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
779 				nvap->na_type = VFIFO;
780                         if (nvap->na_type != VFIFO &&
781 			    (error = priv_check_cred(nd->nd_cred,
782 			     PRIV_VFS_MKNOD_DEV, 0))) {
783 				vrele(ndp->ni_startdir);
784 				nfsvno_relpathbuf(ndp);
785 				vput(ndp->ni_dvp);
786 				goto out;
787 			}
788 			nvap->na_rdev = rdev;
789 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
790 			    &ndp->ni_cnd, &nvap->na_vattr);
791 			vput(ndp->ni_dvp);
792 			nfsvno_relpathbuf(ndp);
793 			vrele(ndp->ni_startdir);
794 			if (error)
795 				goto out;
796 		} else {
797 			vrele(ndp->ni_startdir);
798 			nfsvno_relpathbuf(ndp);
799 			vput(ndp->ni_dvp);
800 			error = ENXIO;
801 			goto out;
802 		}
803 		*vpp = ndp->ni_vp;
804 	} else {
805 		/*
806 		 * Handle cases where error is already set and/or
807 		 * the file exists.
808 		 * 1 - clean up the lookup
809 		 * 2 - iff !error and na_size set, truncate it
810 		 */
811 		vrele(ndp->ni_startdir);
812 		nfsvno_relpathbuf(ndp);
813 		*vpp = ndp->ni_vp;
814 		if (ndp->ni_dvp == *vpp)
815 			vrele(ndp->ni_dvp);
816 		else
817 			vput(ndp->ni_dvp);
818 		if (!error && nvap->na_size != VNOVAL) {
819 			error = nfsvno_accchk(*vpp, VWRITE,
820 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
821 			    NFSACCCHK_VPISLOCKED, NULL);
822 			if (!error) {
823 				tempsize = nvap->na_size;
824 				NFSVNO_ATTRINIT(nvap);
825 				nvap->na_size = tempsize;
826 				error = VOP_SETATTR(*vpp,
827 				    &nvap->na_vattr, nd->nd_cred);
828 			}
829 		}
830 		if (error)
831 			vput(*vpp);
832 	}
833 
834 out:
835 	NFSEXITCODE(error);
836 	return (error);
837 }
838 
839 /*
840  * Do a mknod vnode op.
841  */
842 int
843 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
844     struct thread *p)
845 {
846 	int error = 0;
847 	enum vtype vtyp;
848 
849 	vtyp = nvap->na_type;
850 	/*
851 	 * Iff doesn't exist, create it.
852 	 */
853 	if (ndp->ni_vp) {
854 		vrele(ndp->ni_startdir);
855 		nfsvno_relpathbuf(ndp);
856 		vput(ndp->ni_dvp);
857 		vrele(ndp->ni_vp);
858 		error = EEXIST;
859 		goto out;
860 	}
861 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
862 		vrele(ndp->ni_startdir);
863 		nfsvno_relpathbuf(ndp);
864 		vput(ndp->ni_dvp);
865 		error = NFSERR_BADTYPE;
866 		goto out;
867 	}
868 	if (vtyp == VSOCK) {
869 		vrele(ndp->ni_startdir);
870 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
871 		    &ndp->ni_cnd, &nvap->na_vattr);
872 		vput(ndp->ni_dvp);
873 		nfsvno_relpathbuf(ndp);
874 	} else {
875 		if (nvap->na_type != VFIFO &&
876 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
877 			vrele(ndp->ni_startdir);
878 			nfsvno_relpathbuf(ndp);
879 			vput(ndp->ni_dvp);
880 			goto out;
881 		}
882 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
883 		    &ndp->ni_cnd, &nvap->na_vattr);
884 		vput(ndp->ni_dvp);
885 		nfsvno_relpathbuf(ndp);
886 		vrele(ndp->ni_startdir);
887 		/*
888 		 * Since VOP_MKNOD returns the ni_vp, I can't
889 		 * see any reason to do the lookup.
890 		 */
891 	}
892 
893 out:
894 	NFSEXITCODE(error);
895 	return (error);
896 }
897 
898 /*
899  * Mkdir vnode op.
900  */
901 int
902 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
903     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
904 {
905 	int error = 0;
906 
907 	if (ndp->ni_vp != NULL) {
908 		if (ndp->ni_dvp == ndp->ni_vp)
909 			vrele(ndp->ni_dvp);
910 		else
911 			vput(ndp->ni_dvp);
912 		vrele(ndp->ni_vp);
913 		nfsvno_relpathbuf(ndp);
914 		error = EEXIST;
915 		goto out;
916 	}
917 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
918 	    &nvap->na_vattr);
919 	vput(ndp->ni_dvp);
920 	nfsvno_relpathbuf(ndp);
921 
922 out:
923 	NFSEXITCODE(error);
924 	return (error);
925 }
926 
927 /*
928  * symlink vnode op.
929  */
930 int
931 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
932     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
933     struct nfsexstuff *exp)
934 {
935 	int error = 0;
936 
937 	if (ndp->ni_vp) {
938 		vrele(ndp->ni_startdir);
939 		nfsvno_relpathbuf(ndp);
940 		if (ndp->ni_dvp == ndp->ni_vp)
941 			vrele(ndp->ni_dvp);
942 		else
943 			vput(ndp->ni_dvp);
944 		vrele(ndp->ni_vp);
945 		error = EEXIST;
946 		goto out;
947 	}
948 
949 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
950 	    &nvap->na_vattr, pathcp);
951 	vput(ndp->ni_dvp);
952 	vrele(ndp->ni_startdir);
953 	nfsvno_relpathbuf(ndp);
954 	/*
955 	 * Although FreeBSD still had the lookup code in
956 	 * it for 7/current, there doesn't seem to be any
957 	 * point, since VOP_SYMLINK() returns the ni_vp.
958 	 * Just vput it for v2.
959 	 */
960 	if (!not_v2 && !error)
961 		vput(ndp->ni_vp);
962 
963 out:
964 	NFSEXITCODE(error);
965 	return (error);
966 }
967 
968 /*
969  * Parse symbolic link arguments.
970  * This function has an ugly side effect. It will MALLOC() an area for
971  * the symlink and set iov_base to point to it, only if it succeeds.
972  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
973  * be FREE'd later.
974  */
975 int
976 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
977     struct thread *p, char **pathcpp, int *lenp)
978 {
979 	u_int32_t *tl;
980 	char *pathcp = NULL;
981 	int error = 0, len;
982 	struct nfsv2_sattr *sp;
983 
984 	*pathcpp = NULL;
985 	*lenp = 0;
986 	if ((nd->nd_flag & ND_NFSV3) &&
987 	    (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
988 		goto nfsmout;
989 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
990 	len = fxdr_unsigned(int, *tl);
991 	if (len > NFS_MAXPATHLEN || len <= 0) {
992 		error = EBADRPC;
993 		goto nfsmout;
994 	}
995 	MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
996 	error = nfsrv_mtostr(nd, pathcp, len);
997 	if (error)
998 		goto nfsmout;
999 	if (nd->nd_flag & ND_NFSV2) {
1000 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1001 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1002 	}
1003 	*pathcpp = pathcp;
1004 	*lenp = len;
1005 	NFSEXITCODE2(0, nd);
1006 	return (0);
1007 nfsmout:
1008 	if (pathcp)
1009 		free(pathcp, M_TEMP);
1010 	NFSEXITCODE2(error, nd);
1011 	return (error);
1012 }
1013 
1014 /*
1015  * Remove a non-directory object.
1016  */
1017 int
1018 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1019     struct thread *p, struct nfsexstuff *exp)
1020 {
1021 	struct vnode *vp;
1022 	int error = 0;
1023 
1024 	vp = ndp->ni_vp;
1025 	if (vp->v_type == VDIR)
1026 		error = NFSERR_ISDIR;
1027 	else if (is_v4)
1028 		error = nfsrv_checkremove(vp, 1, p);
1029 	if (!error)
1030 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1031 	if (ndp->ni_dvp == vp)
1032 		vrele(ndp->ni_dvp);
1033 	else
1034 		vput(ndp->ni_dvp);
1035 	vput(vp);
1036 	NFSEXITCODE(error);
1037 	return (error);
1038 }
1039 
1040 /*
1041  * Remove a directory.
1042  */
1043 int
1044 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1045     struct thread *p, struct nfsexstuff *exp)
1046 {
1047 	struct vnode *vp;
1048 	int error = 0;
1049 
1050 	vp = ndp->ni_vp;
1051 	if (vp->v_type != VDIR) {
1052 		error = ENOTDIR;
1053 		goto out;
1054 	}
1055 	/*
1056 	 * No rmdir "." please.
1057 	 */
1058 	if (ndp->ni_dvp == vp) {
1059 		error = EINVAL;
1060 		goto out;
1061 	}
1062 	/*
1063 	 * The root of a mounted filesystem cannot be deleted.
1064 	 */
1065 	if (vp->v_vflag & VV_ROOT)
1066 		error = EBUSY;
1067 out:
1068 	if (!error)
1069 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1070 	if (ndp->ni_dvp == vp)
1071 		vrele(ndp->ni_dvp);
1072 	else
1073 		vput(ndp->ni_dvp);
1074 	vput(vp);
1075 	NFSEXITCODE(error);
1076 	return (error);
1077 }
1078 
1079 /*
1080  * Rename vnode op.
1081  */
1082 int
1083 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1084     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1085 {
1086 	struct vnode *fvp, *tvp, *tdvp;
1087 	int error = 0;
1088 
1089 	fvp = fromndp->ni_vp;
1090 	if (ndstat) {
1091 		vrele(fromndp->ni_dvp);
1092 		vrele(fvp);
1093 		error = ndstat;
1094 		goto out1;
1095 	}
1096 	tdvp = tondp->ni_dvp;
1097 	tvp = tondp->ni_vp;
1098 	if (tvp != NULL) {
1099 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1100 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1101 			goto out;
1102 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1103 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1104 			goto out;
1105 		}
1106 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1107 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1108 			goto out;
1109 		}
1110 
1111 		/*
1112 		 * A rename to '.' or '..' results in a prematurely
1113 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1114 		 * here.
1115 		 */
1116 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1117 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1118 		    (tondp->ni_cnd.cn_namelen == 2 &&
1119 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1120 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1121 			error = EINVAL;
1122 			goto out;
1123 		}
1124 	}
1125 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1126 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1127 		goto out;
1128 	}
1129 	if (fvp->v_mount != tdvp->v_mount) {
1130 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1131 		goto out;
1132 	}
1133 	if (fvp == tdvp) {
1134 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1135 		goto out;
1136 	}
1137 	if (fvp == tvp) {
1138 		/*
1139 		 * If source and destination are the same, there is nothing to
1140 		 * do. Set error to -1 to indicate this.
1141 		 */
1142 		error = -1;
1143 		goto out;
1144 	}
1145 	if (ndflag & ND_NFSV4) {
1146 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1147 			error = nfsrv_checkremove(fvp, 0, p);
1148 			NFSVOPUNLOCK(fvp, 0);
1149 		} else
1150 			error = EPERM;
1151 		if (tvp && !error)
1152 			error = nfsrv_checkremove(tvp, 1, p);
1153 	} else {
1154 		/*
1155 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1156 		 * that the NFSv4 client won't be confused by the rename.
1157 		 * Since nfsd_recalldelegation() can only be called on an
1158 		 * unlocked vnode at this point and fvp is the file that will
1159 		 * still exist after the rename, just do fvp.
1160 		 */
1161 		nfsd_recalldelegation(fvp, p);
1162 	}
1163 out:
1164 	if (!error) {
1165 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1166 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1167 		    &tondp->ni_cnd);
1168 	} else {
1169 		if (tdvp == tvp)
1170 			vrele(tdvp);
1171 		else
1172 			vput(tdvp);
1173 		if (tvp)
1174 			vput(tvp);
1175 		vrele(fromndp->ni_dvp);
1176 		vrele(fvp);
1177 		if (error == -1)
1178 			error = 0;
1179 	}
1180 	vrele(tondp->ni_startdir);
1181 	nfsvno_relpathbuf(tondp);
1182 out1:
1183 	vrele(fromndp->ni_startdir);
1184 	nfsvno_relpathbuf(fromndp);
1185 	NFSEXITCODE(error);
1186 	return (error);
1187 }
1188 
1189 /*
1190  * Link vnode op.
1191  */
1192 int
1193 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1194     struct thread *p, struct nfsexstuff *exp)
1195 {
1196 	struct vnode *xp;
1197 	int error = 0;
1198 
1199 	xp = ndp->ni_vp;
1200 	if (xp != NULL) {
1201 		error = EEXIST;
1202 	} else {
1203 		xp = ndp->ni_dvp;
1204 		if (vp->v_mount != xp->v_mount)
1205 			error = EXDEV;
1206 	}
1207 	if (!error) {
1208 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1209 		if ((vp->v_iflag & VI_DOOMED) == 0)
1210 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1211 		else
1212 			error = EPERM;
1213 		if (ndp->ni_dvp == vp)
1214 			vrele(ndp->ni_dvp);
1215 		else
1216 			vput(ndp->ni_dvp);
1217 		NFSVOPUNLOCK(vp, 0);
1218 	} else {
1219 		if (ndp->ni_dvp == ndp->ni_vp)
1220 			vrele(ndp->ni_dvp);
1221 		else
1222 			vput(ndp->ni_dvp);
1223 		if (ndp->ni_vp)
1224 			vrele(ndp->ni_vp);
1225 	}
1226 	nfsvno_relpathbuf(ndp);
1227 	NFSEXITCODE(error);
1228 	return (error);
1229 }
1230 
1231 /*
1232  * Do the fsync() appropriate for the commit.
1233  */
1234 int
1235 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1236     struct thread *td)
1237 {
1238 	int error = 0;
1239 
1240 	if (cnt > MAX_COMMIT_COUNT) {
1241 		/*
1242 		 * Give up and do the whole thing
1243 		 */
1244 		if (vp->v_object &&
1245 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1246 			VM_OBJECT_LOCK(vp->v_object);
1247 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1248 			VM_OBJECT_UNLOCK(vp->v_object);
1249 		}
1250 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1251 	} else {
1252 		/*
1253 		 * Locate and synchronously write any buffers that fall
1254 		 * into the requested range.  Note:  we are assuming that
1255 		 * f_iosize is a power of 2.
1256 		 */
1257 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1258 		int iomask = iosize - 1;
1259 		struct bufobj *bo;
1260 		daddr_t lblkno;
1261 
1262 		/*
1263 		 * Align to iosize boundry, super-align to page boundry.
1264 		 */
1265 		if (off & iomask) {
1266 			cnt += off & iomask;
1267 			off &= ~(u_quad_t)iomask;
1268 		}
1269 		if (off & PAGE_MASK) {
1270 			cnt += off & PAGE_MASK;
1271 			off &= ~(u_quad_t)PAGE_MASK;
1272 		}
1273 		lblkno = off / iosize;
1274 
1275 		if (vp->v_object &&
1276 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1277 			VM_OBJECT_LOCK(vp->v_object);
1278 			vm_object_page_clean(vp->v_object, off, off + cnt,
1279 			    OBJPC_SYNC);
1280 			VM_OBJECT_UNLOCK(vp->v_object);
1281 		}
1282 
1283 		bo = &vp->v_bufobj;
1284 		BO_LOCK(bo);
1285 		while (cnt > 0) {
1286 			struct buf *bp;
1287 
1288 			/*
1289 			 * If we have a buffer and it is marked B_DELWRI we
1290 			 * have to lock and write it.  Otherwise the prior
1291 			 * write is assumed to have already been committed.
1292 			 *
1293 			 * gbincore() can return invalid buffers now so we
1294 			 * have to check that bit as well (though B_DELWRI
1295 			 * should not be set if B_INVAL is set there could be
1296 			 * a race here since we haven't locked the buffer).
1297 			 */
1298 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1299 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1300 				    LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
1301 					BO_LOCK(bo);
1302 					continue; /* retry */
1303 				}
1304 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1305 				    B_DELWRI) {
1306 					bremfree(bp);
1307 					bp->b_flags &= ~B_ASYNC;
1308 					bwrite(bp);
1309 					++nfs_commit_miss;
1310 				} else
1311 					BUF_UNLOCK(bp);
1312 				BO_LOCK(bo);
1313 			}
1314 			++nfs_commit_blks;
1315 			if (cnt < iosize)
1316 				break;
1317 			cnt -= iosize;
1318 			++lblkno;
1319 		}
1320 		BO_UNLOCK(bo);
1321 	}
1322 	NFSEXITCODE(error);
1323 	return (error);
1324 }
1325 
1326 /*
1327  * Statfs vnode op.
1328  */
1329 int
1330 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1331 {
1332 	int error;
1333 
1334 	error = VFS_STATFS(vp->v_mount, sf);
1335 	if (error == 0) {
1336 		/*
1337 		 * Since NFS handles these values as unsigned on the
1338 		 * wire, there is no way to represent negative values,
1339 		 * so set them to 0. Without this, they will appear
1340 		 * to be very large positive values for clients like
1341 		 * Solaris10.
1342 		 */
1343 		if (sf->f_bavail < 0)
1344 			sf->f_bavail = 0;
1345 		if (sf->f_ffree < 0)
1346 			sf->f_ffree = 0;
1347 	}
1348 	NFSEXITCODE(error);
1349 	return (error);
1350 }
1351 
1352 /*
1353  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1354  * must handle nfsrv_opencheck() calls after any other access checks.
1355  */
1356 void
1357 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1358     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1359     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1360     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1361     struct nfsexstuff *exp, struct vnode **vpp)
1362 {
1363 	struct vnode *vp = NULL;
1364 	u_quad_t tempsize;
1365 	struct nfsexstuff nes;
1366 
1367 	if (ndp->ni_vp == NULL)
1368 		nd->nd_repstat = nfsrv_opencheck(clientid,
1369 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1370 	if (!nd->nd_repstat) {
1371 		if (ndp->ni_vp == NULL) {
1372 			vrele(ndp->ni_startdir);
1373 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1374 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1375 			vput(ndp->ni_dvp);
1376 			nfsvno_relpathbuf(ndp);
1377 			if (!nd->nd_repstat) {
1378 				if (*exclusive_flagp) {
1379 					*exclusive_flagp = 0;
1380 					NFSVNO_ATTRINIT(nvap);
1381 					nvap->na_atime.tv_sec = cverf[0];
1382 					nvap->na_atime.tv_nsec = cverf[1];
1383 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1384 					    &nvap->na_vattr, cred);
1385 				} else {
1386 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1387 					    aclp, p, attrbitp, exp);
1388 				}
1389 			}
1390 			vp = ndp->ni_vp;
1391 		} else {
1392 			if (ndp->ni_startdir)
1393 				vrele(ndp->ni_startdir);
1394 			nfsvno_relpathbuf(ndp);
1395 			vp = ndp->ni_vp;
1396 			if (create == NFSV4OPEN_CREATE) {
1397 				if (ndp->ni_dvp == vp)
1398 					vrele(ndp->ni_dvp);
1399 				else
1400 					vput(ndp->ni_dvp);
1401 			}
1402 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1403 				if (ndp->ni_cnd.cn_flags & RDONLY)
1404 					NFSVNO_SETEXRDONLY(&nes);
1405 				else
1406 					NFSVNO_EXINIT(&nes);
1407 				nd->nd_repstat = nfsvno_accchk(vp,
1408 				    VWRITE, cred, &nes, p,
1409 				    NFSACCCHK_NOOVERRIDE,
1410 				    NFSACCCHK_VPISLOCKED, NULL);
1411 				nd->nd_repstat = nfsrv_opencheck(clientid,
1412 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1413 				if (!nd->nd_repstat) {
1414 					tempsize = nvap->na_size;
1415 					NFSVNO_ATTRINIT(nvap);
1416 					nvap->na_size = tempsize;
1417 					nd->nd_repstat = VOP_SETATTR(vp,
1418 					    &nvap->na_vattr, cred);
1419 				}
1420 			} else if (vp->v_type == VREG) {
1421 				nd->nd_repstat = nfsrv_opencheck(clientid,
1422 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1423 			}
1424 		}
1425 	} else {
1426 		if (ndp->ni_cnd.cn_flags & HASBUF)
1427 			nfsvno_relpathbuf(ndp);
1428 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1429 			vrele(ndp->ni_startdir);
1430 			if (ndp->ni_dvp == ndp->ni_vp)
1431 				vrele(ndp->ni_dvp);
1432 			else
1433 				vput(ndp->ni_dvp);
1434 			if (ndp->ni_vp)
1435 				vput(ndp->ni_vp);
1436 		}
1437 	}
1438 	*vpp = vp;
1439 
1440 	NFSEXITCODE2(0, nd);
1441 }
1442 
1443 /*
1444  * Updates the file rev and sets the mtime and ctime
1445  * to the current clock time, returning the va_filerev and va_Xtime
1446  * values.
1447  */
1448 void
1449 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1450     struct ucred *cred, struct thread *p)
1451 {
1452 	struct vattr va;
1453 
1454 	VATTR_NULL(&va);
1455 	getnanotime(&va.va_mtime);
1456 	(void) VOP_SETATTR(vp, &va, cred);
1457 	(void) nfsvno_getattr(vp, nvap, cred, p, 1);
1458 }
1459 
1460 /*
1461  * Glue routine to nfsv4_fillattr().
1462  */
1463 int
1464 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1465     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1466     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1467     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1468 {
1469 	int error;
1470 
1471 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1472 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1473 	    mounted_on_fileno);
1474 	NFSEXITCODE2(0, nd);
1475 	return (error);
1476 }
1477 
1478 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1479 /*
1480  * nfs readdir service
1481  * - mallocs what it thinks is enough to read
1482  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1483  * - calls VOP_READDIR()
1484  * - loops around building the reply
1485  *	if the output generated exceeds count break out of loop
1486  *	The NFSM_CLGET macro is used here so that the reply will be packed
1487  *	tightly in mbuf clusters.
1488  * - it trims out records with d_fileno == 0
1489  *	this doesn't matter for Unix clients, but they might confuse clients
1490  *	for other os'.
1491  * - it trims out records with d_type == DT_WHT
1492  *	these cannot be seen through NFS (unless we extend the protocol)
1493  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1494  * PS: The NFS protocol spec. does not clarify what the "count" byte
1495  *	argument is a count of.. just name strings and file id's or the
1496  *	entire reply rpc or ...
1497  *	I tried just file name and id sizes and it confused the Sun client,
1498  *	so I am using the full rpc size now. The "paranoia.." comment refers
1499  *	to including the status longwords that are not a part of the dir.
1500  *	"entry" structures, but are in the rpc.
1501  */
1502 int
1503 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1504     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1505 {
1506 	struct dirent *dp;
1507 	u_int32_t *tl;
1508 	int dirlen;
1509 	char *cpos, *cend, *rbuf;
1510 	struct nfsvattr at;
1511 	int nlen, error = 0, getret = 1;
1512 	int siz, cnt, fullsiz, eofflag, ncookies;
1513 	u_int64_t off, toff, verf;
1514 	u_long *cookies = NULL, *cookiep;
1515 	struct uio io;
1516 	struct iovec iv;
1517 	int not_zfs;
1518 
1519 	if (nd->nd_repstat) {
1520 		nfsrv_postopattr(nd, getret, &at);
1521 		goto out;
1522 	}
1523 	if (nd->nd_flag & ND_NFSV2) {
1524 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1525 		off = fxdr_unsigned(u_quad_t, *tl++);
1526 	} else {
1527 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1528 		off = fxdr_hyper(tl);
1529 		tl += 2;
1530 		verf = fxdr_hyper(tl);
1531 		tl += 2;
1532 	}
1533 	toff = off;
1534 	cnt = fxdr_unsigned(int, *tl);
1535 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1536 		cnt = NFS_SRVMAXDATA(nd);
1537 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1538 	fullsiz = siz;
1539 	if (nd->nd_flag & ND_NFSV3) {
1540 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1541 		    p, 1);
1542 #if 0
1543 		/*
1544 		 * va_filerev is not sufficient as a cookie verifier,
1545 		 * since it is not supposed to change when entries are
1546 		 * removed/added unless that offset cookies returned to
1547 		 * the client are no longer valid.
1548 		 */
1549 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
1550 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1551 #endif
1552 	}
1553 	if (nd->nd_repstat == 0 && cnt == 0) {
1554 		if (nd->nd_flag & ND_NFSV2)
1555 			/* NFSv2 does not have NFSERR_TOOSMALL */
1556 			nd->nd_repstat = EPERM;
1557 		else
1558 			nd->nd_repstat = NFSERR_TOOSMALL;
1559 	}
1560 	if (!nd->nd_repstat)
1561 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1562 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1563 		    NFSACCCHK_VPISLOCKED, NULL);
1564 	if (nd->nd_repstat) {
1565 		vput(vp);
1566 		if (nd->nd_flag & ND_NFSV3)
1567 			nfsrv_postopattr(nd, getret, &at);
1568 		goto out;
1569 	}
1570 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1571 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1572 again:
1573 	eofflag = 0;
1574 	if (cookies) {
1575 		free((caddr_t)cookies, M_TEMP);
1576 		cookies = NULL;
1577 	}
1578 
1579 	iv.iov_base = rbuf;
1580 	iv.iov_len = siz;
1581 	io.uio_iov = &iv;
1582 	io.uio_iovcnt = 1;
1583 	io.uio_offset = (off_t)off;
1584 	io.uio_resid = siz;
1585 	io.uio_segflg = UIO_SYSSPACE;
1586 	io.uio_rw = UIO_READ;
1587 	io.uio_td = NULL;
1588 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1589 	    &cookies);
1590 	off = (u_int64_t)io.uio_offset;
1591 	if (io.uio_resid)
1592 		siz -= io.uio_resid;
1593 
1594 	if (!cookies && !nd->nd_repstat)
1595 		nd->nd_repstat = NFSERR_PERM;
1596 	if (nd->nd_flag & ND_NFSV3) {
1597 		getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1598 		if (!nd->nd_repstat)
1599 			nd->nd_repstat = getret;
1600 	}
1601 
1602 	/*
1603 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
1604 	 */
1605 	if (nd->nd_repstat) {
1606 		vput(vp);
1607 		free((caddr_t)rbuf, M_TEMP);
1608 		if (cookies)
1609 			free((caddr_t)cookies, M_TEMP);
1610 		if (nd->nd_flag & ND_NFSV3)
1611 			nfsrv_postopattr(nd, getret, &at);
1612 		goto out;
1613 	}
1614 	/*
1615 	 * If nothing read, return eof
1616 	 * rpc reply
1617 	 */
1618 	if (siz == 0) {
1619 		vput(vp);
1620 		if (nd->nd_flag & ND_NFSV2) {
1621 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1622 		} else {
1623 			nfsrv_postopattr(nd, getret, &at);
1624 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1625 			txdr_hyper(at.na_filerev, tl);
1626 			tl += 2;
1627 		}
1628 		*tl++ = newnfs_false;
1629 		*tl = newnfs_true;
1630 		FREE((caddr_t)rbuf, M_TEMP);
1631 		FREE((caddr_t)cookies, M_TEMP);
1632 		goto out;
1633 	}
1634 
1635 	/*
1636 	 * Check for degenerate cases of nothing useful read.
1637 	 * If so go try again
1638 	 */
1639 	cpos = rbuf;
1640 	cend = rbuf + siz;
1641 	dp = (struct dirent *)cpos;
1642 	cookiep = cookies;
1643 
1644 	/*
1645 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1646 	 * directory offset up to a block boundary, so it is necessary to
1647 	 * skip over the records that precede the requested offset. This
1648 	 * requires the assumption that file offset cookies monotonically
1649 	 * increase.
1650 	 * Since the offset cookies don't monotonically increase for ZFS,
1651 	 * this is not done when ZFS is the file system.
1652 	 */
1653 	while (cpos < cend && ncookies > 0 &&
1654 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1655 	     (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
1656 		cpos += dp->d_reclen;
1657 		dp = (struct dirent *)cpos;
1658 		cookiep++;
1659 		ncookies--;
1660 	}
1661 	if (cpos >= cend || ncookies == 0) {
1662 		siz = fullsiz;
1663 		toff = off;
1664 		goto again;
1665 	}
1666 	vput(vp);
1667 
1668 	/*
1669 	 * dirlen is the size of the reply, including all XDR and must
1670 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1671 	 * if the XDR should be included in "count", but to be safe, we do.
1672 	 * (Include the two booleans at the end of the reply in dirlen now.)
1673 	 */
1674 	if (nd->nd_flag & ND_NFSV3) {
1675 		nfsrv_postopattr(nd, getret, &at);
1676 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1677 		txdr_hyper(at.na_filerev, tl);
1678 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1679 	} else {
1680 		dirlen = 2 * NFSX_UNSIGNED;
1681 	}
1682 
1683 	/* Loop through the records and build reply */
1684 	while (cpos < cend && ncookies > 0) {
1685 		nlen = dp->d_namlen;
1686 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1687 			nlen <= NFS_MAXNAMLEN) {
1688 			if (nd->nd_flag & ND_NFSV3)
1689 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1690 			else
1691 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1692 			if (dirlen > cnt) {
1693 				eofflag = 0;
1694 				break;
1695 			}
1696 
1697 			/*
1698 			 * Build the directory record xdr from
1699 			 * the dirent entry.
1700 			 */
1701 			if (nd->nd_flag & ND_NFSV3) {
1702 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1703 				*tl++ = newnfs_true;
1704 				*tl++ = 0;
1705 			} else {
1706 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1707 				*tl++ = newnfs_true;
1708 			}
1709 			*tl = txdr_unsigned(dp->d_fileno);
1710 			(void) nfsm_strtom(nd, dp->d_name, nlen);
1711 			if (nd->nd_flag & ND_NFSV3) {
1712 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1713 				*tl++ = 0;
1714 			} else
1715 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1716 			*tl = txdr_unsigned(*cookiep);
1717 		}
1718 		cpos += dp->d_reclen;
1719 		dp = (struct dirent *)cpos;
1720 		cookiep++;
1721 		ncookies--;
1722 	}
1723 	if (cpos < cend)
1724 		eofflag = 0;
1725 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1726 	*tl++ = newnfs_false;
1727 	if (eofflag)
1728 		*tl = newnfs_true;
1729 	else
1730 		*tl = newnfs_false;
1731 	FREE((caddr_t)rbuf, M_TEMP);
1732 	FREE((caddr_t)cookies, M_TEMP);
1733 
1734 out:
1735 	NFSEXITCODE2(0, nd);
1736 	return (0);
1737 nfsmout:
1738 	vput(vp);
1739 	NFSEXITCODE2(error, nd);
1740 	return (error);
1741 }
1742 
1743 /*
1744  * Readdirplus for V3 and Readdir for V4.
1745  */
1746 int
1747 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1748     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1749 {
1750 	struct dirent *dp;
1751 	u_int32_t *tl;
1752 	int dirlen;
1753 	char *cpos, *cend, *rbuf;
1754 	struct vnode *nvp;
1755 	fhandle_t nfh;
1756 	struct nfsvattr nva, at, *nvap = &nva;
1757 	struct mbuf *mb0, *mb1;
1758 	struct nfsreferral *refp;
1759 	int nlen, r, error = 0, getret = 1, usevget = 1;
1760 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1761 	caddr_t bpos0, bpos1;
1762 	u_int64_t off, toff, verf;
1763 	u_long *cookies = NULL, *cookiep;
1764 	nfsattrbit_t attrbits, rderrbits, savbits;
1765 	struct uio io;
1766 	struct iovec iv;
1767 	struct componentname cn;
1768 	int at_root, needs_unbusy, not_zfs, supports_nfsv4acls;
1769 	struct mount *mp, *new_mp;
1770 	uint64_t mounted_on_fileno;
1771 
1772 	if (nd->nd_repstat) {
1773 		nfsrv_postopattr(nd, getret, &at);
1774 		goto out;
1775 	}
1776 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1777 	off = fxdr_hyper(tl);
1778 	toff = off;
1779 	tl += 2;
1780 	verf = fxdr_hyper(tl);
1781 	tl += 2;
1782 	siz = fxdr_unsigned(int, *tl++);
1783 	cnt = fxdr_unsigned(int, *tl);
1784 
1785 	/*
1786 	 * Use the server's maximum data transfer size as the upper bound
1787 	 * on reply datalen.
1788 	 */
1789 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1790 		cnt = NFS_SRVMAXDATA(nd);
1791 
1792 	/*
1793 	 * siz is a "hint" of how much directory information (name, fileid,
1794 	 * cookie) should be in the reply. At least one client "hints" 0,
1795 	 * so I set it to cnt for that case. I also round it up to the
1796 	 * next multiple of DIRBLKSIZ.
1797 	 */
1798 	if (siz <= 0)
1799 		siz = cnt;
1800 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1801 
1802 	if (nd->nd_flag & ND_NFSV4) {
1803 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1804 		if (error)
1805 			goto nfsmout;
1806 		NFSSET_ATTRBIT(&savbits, &attrbits);
1807 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1808 		NFSZERO_ATTRBIT(&rderrbits);
1809 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1810 	} else {
1811 		NFSZERO_ATTRBIT(&attrbits);
1812 	}
1813 	fullsiz = siz;
1814 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1815 	if (!nd->nd_repstat) {
1816 	    if (off && verf != at.na_filerev) {
1817 		/*
1818 		 * va_filerev is not sufficient as a cookie verifier,
1819 		 * since it is not supposed to change when entries are
1820 		 * removed/added unless that offset cookies returned to
1821 		 * the client are no longer valid.
1822 		 */
1823 #if 0
1824 		if (nd->nd_flag & ND_NFSV4) {
1825 			nd->nd_repstat = NFSERR_NOTSAME;
1826 		} else {
1827 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1828 		}
1829 #endif
1830 	    } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1831 		nd->nd_repstat = NFSERR_BAD_COOKIE;
1832 	    }
1833 	}
1834 	if (!nd->nd_repstat && vp->v_type != VDIR)
1835 		nd->nd_repstat = NFSERR_NOTDIR;
1836 	if (!nd->nd_repstat && cnt == 0)
1837 		nd->nd_repstat = NFSERR_TOOSMALL;
1838 	if (!nd->nd_repstat)
1839 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1840 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1841 		    NFSACCCHK_VPISLOCKED, NULL);
1842 	if (nd->nd_repstat) {
1843 		vput(vp);
1844 		if (nd->nd_flag & ND_NFSV3)
1845 			nfsrv_postopattr(nd, getret, &at);
1846 		goto out;
1847 	}
1848 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1849 
1850 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1851 again:
1852 	eofflag = 0;
1853 	if (cookies) {
1854 		free((caddr_t)cookies, M_TEMP);
1855 		cookies = NULL;
1856 	}
1857 
1858 	iv.iov_base = rbuf;
1859 	iv.iov_len = siz;
1860 	io.uio_iov = &iv;
1861 	io.uio_iovcnt = 1;
1862 	io.uio_offset = (off_t)off;
1863 	io.uio_resid = siz;
1864 	io.uio_segflg = UIO_SYSSPACE;
1865 	io.uio_rw = UIO_READ;
1866 	io.uio_td = NULL;
1867 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1868 	    &cookies);
1869 	off = (u_int64_t)io.uio_offset;
1870 	if (io.uio_resid)
1871 		siz -= io.uio_resid;
1872 
1873 	getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1874 
1875 	if (!cookies && !nd->nd_repstat)
1876 		nd->nd_repstat = NFSERR_PERM;
1877 	if (!nd->nd_repstat)
1878 		nd->nd_repstat = getret;
1879 	if (nd->nd_repstat) {
1880 		vput(vp);
1881 		if (cookies)
1882 			free((caddr_t)cookies, M_TEMP);
1883 		free((caddr_t)rbuf, M_TEMP);
1884 		if (nd->nd_flag & ND_NFSV3)
1885 			nfsrv_postopattr(nd, getret, &at);
1886 		goto out;
1887 	}
1888 	/*
1889 	 * If nothing read, return eof
1890 	 * rpc reply
1891 	 */
1892 	if (siz == 0) {
1893 		vput(vp);
1894 		if (nd->nd_flag & ND_NFSV3)
1895 			nfsrv_postopattr(nd, getret, &at);
1896 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1897 		txdr_hyper(at.na_filerev, tl);
1898 		tl += 2;
1899 		*tl++ = newnfs_false;
1900 		*tl = newnfs_true;
1901 		free((caddr_t)cookies, M_TEMP);
1902 		free((caddr_t)rbuf, M_TEMP);
1903 		goto out;
1904 	}
1905 
1906 	/*
1907 	 * Check for degenerate cases of nothing useful read.
1908 	 * If so go try again
1909 	 */
1910 	cpos = rbuf;
1911 	cend = rbuf + siz;
1912 	dp = (struct dirent *)cpos;
1913 	cookiep = cookies;
1914 
1915 	/*
1916 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1917 	 * directory offset up to a block boundary, so it is necessary to
1918 	 * skip over the records that precede the requested offset. This
1919 	 * requires the assumption that file offset cookies monotonically
1920 	 * increase.
1921 	 * Since the offset cookies don't monotonically increase for ZFS,
1922 	 * this is not done when ZFS is the file system.
1923 	 */
1924 	while (cpos < cend && ncookies > 0 &&
1925 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1926 	   (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff) ||
1927 	   ((nd->nd_flag & ND_NFSV4) &&
1928 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1929 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1930 		cpos += dp->d_reclen;
1931 		dp = (struct dirent *)cpos;
1932 		cookiep++;
1933 		ncookies--;
1934 	}
1935 	if (cpos >= cend || ncookies == 0) {
1936 		siz = fullsiz;
1937 		toff = off;
1938 		goto again;
1939 	}
1940 
1941 	/*
1942 	 * Busy the file system so that the mount point won't go away
1943 	 * and, as such, VFS_VGET() can be used safely.
1944 	 */
1945 	mp = vp->v_mount;
1946 	vfs_ref(mp);
1947 	NFSVOPUNLOCK(vp, 0);
1948 	nd->nd_repstat = vfs_busy(mp, 0);
1949 	vfs_rel(mp);
1950 	if (nd->nd_repstat != 0) {
1951 		vrele(vp);
1952 		free(cookies, M_TEMP);
1953 		free(rbuf, M_TEMP);
1954 		if (nd->nd_flag & ND_NFSV3)
1955 			nfsrv_postopattr(nd, getret, &at);
1956 		goto out;
1957 	}
1958 
1959 	/*
1960 	 * Save this position, in case there is an error before one entry
1961 	 * is created.
1962 	 */
1963 	mb0 = nd->nd_mb;
1964 	bpos0 = nd->nd_bpos;
1965 
1966 	/*
1967 	 * Fill in the first part of the reply.
1968 	 * dirlen is the reply length in bytes and cannot exceed cnt.
1969 	 * (Include the two booleans at the end of the reply in dirlen now,
1970 	 *  so we recognize when we have exceeded cnt.)
1971 	 */
1972 	if (nd->nd_flag & ND_NFSV3) {
1973 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1974 		nfsrv_postopattr(nd, getret, &at);
1975 	} else {
1976 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
1977 	}
1978 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
1979 	txdr_hyper(at.na_filerev, tl);
1980 
1981 	/*
1982 	 * Save this position, in case there is an empty reply needed.
1983 	 */
1984 	mb1 = nd->nd_mb;
1985 	bpos1 = nd->nd_bpos;
1986 
1987 	/* Loop through the records and build reply */
1988 	entrycnt = 0;
1989 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
1990 		nlen = dp->d_namlen;
1991 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1992 		    nlen <= NFS_MAXNAMLEN &&
1993 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
1994 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
1995 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
1996 			/*
1997 			 * Save the current position in the reply, in case
1998 			 * this entry exceeds cnt.
1999 			 */
2000 			mb1 = nd->nd_mb;
2001 			bpos1 = nd->nd_bpos;
2002 
2003 			/*
2004 			 * For readdir_and_lookup get the vnode using
2005 			 * the file number.
2006 			 */
2007 			nvp = NULL;
2008 			refp = NULL;
2009 			r = 0;
2010 			at_root = 0;
2011 			needs_unbusy = 0;
2012 			new_mp = mp;
2013 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2014 			if ((nd->nd_flag & ND_NFSV3) ||
2015 			    NFSNONZERO_ATTRBIT(&savbits)) {
2016 				if (nd->nd_flag & ND_NFSV4)
2017 					refp = nfsv4root_getreferral(NULL,
2018 					    vp, dp->d_fileno);
2019 				if (refp == NULL) {
2020 					if (usevget)
2021 						r = VFS_VGET(mp, dp->d_fileno,
2022 						    LK_SHARED, &nvp);
2023 					else
2024 						r = EOPNOTSUPP;
2025 					if (r == EOPNOTSUPP) {
2026 						if (usevget) {
2027 							usevget = 0;
2028 							cn.cn_nameiop = LOOKUP;
2029 							cn.cn_lkflags =
2030 							    LK_SHARED |
2031 							    LK_RETRY;
2032 							cn.cn_cred =
2033 							    nd->nd_cred;
2034 							cn.cn_thread = p;
2035 						}
2036 						cn.cn_nameptr = dp->d_name;
2037 						cn.cn_namelen = nlen;
2038 						cn.cn_flags = ISLASTCN |
2039 						    NOFOLLOW | LOCKLEAF |
2040 						    MPSAFE;
2041 						if (nlen == 2 &&
2042 						    dp->d_name[0] == '.' &&
2043 						    dp->d_name[1] == '.')
2044 							cn.cn_flags |=
2045 							    ISDOTDOT;
2046 						if (NFSVOPLOCK(vp, LK_SHARED)
2047 						    != 0) {
2048 							nd->nd_repstat = EPERM;
2049 							break;
2050 						}
2051 						if ((vp->v_vflag & VV_ROOT) != 0
2052 						    && (cn.cn_flags & ISDOTDOT)
2053 						    != 0) {
2054 							vref(vp);
2055 							nvp = vp;
2056 							r = 0;
2057 						} else {
2058 							r = VOP_LOOKUP(vp, &nvp,
2059 							    &cn);
2060 							if (vp != nvp)
2061 								NFSVOPUNLOCK(vp,
2062 								    0);
2063 						}
2064 					}
2065 
2066 					/*
2067 					 * For NFSv4, check to see if nvp is
2068 					 * a mount point and get the mount
2069 					 * point vnode, as required.
2070 					 */
2071 					if (r == 0 &&
2072 					    nfsrv_enable_crossmntpt != 0 &&
2073 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2074 					    nvp->v_type == VDIR &&
2075 					    nvp->v_mountedhere != NULL) {
2076 						new_mp = nvp->v_mountedhere;
2077 						r = vfs_busy(new_mp, 0);
2078 						vput(nvp);
2079 						nvp = NULL;
2080 						if (r == 0) {
2081 							r = VFS_ROOT(new_mp,
2082 							    LK_SHARED, &nvp);
2083 							needs_unbusy = 1;
2084 							if (r == 0)
2085 								at_root = 1;
2086 						}
2087 					}
2088 				}
2089 				if (!r) {
2090 				    if (refp == NULL &&
2091 					((nd->nd_flag & ND_NFSV3) ||
2092 					 NFSNONZERO_ATTRBIT(&attrbits))) {
2093 					r = nfsvno_getfh(nvp, &nfh, p);
2094 					if (!r)
2095 					    r = nfsvno_getattr(nvp, nvap,
2096 						nd->nd_cred, p, 1);
2097 				    }
2098 				} else {
2099 				    nvp = NULL;
2100 				}
2101 				if (r) {
2102 					if (!NFSISSET_ATTRBIT(&attrbits,
2103 					    NFSATTRBIT_RDATTRERROR)) {
2104 						if (nvp != NULL)
2105 							vput(nvp);
2106 						if (needs_unbusy != 0)
2107 							vfs_unbusy(new_mp);
2108 						nd->nd_repstat = r;
2109 						break;
2110 					}
2111 				}
2112 			}
2113 
2114 			/*
2115 			 * Build the directory record xdr
2116 			 */
2117 			if (nd->nd_flag & ND_NFSV3) {
2118 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2119 				*tl++ = newnfs_true;
2120 				*tl++ = 0;
2121 				*tl = txdr_unsigned(dp->d_fileno);
2122 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2123 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2124 				*tl++ = 0;
2125 				*tl = txdr_unsigned(*cookiep);
2126 				nfsrv_postopattr(nd, 0, nvap);
2127 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2128 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2129 				if (nvp != NULL)
2130 					vput(nvp);
2131 			} else {
2132 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2133 				*tl++ = newnfs_true;
2134 				*tl++ = 0;
2135 				*tl = txdr_unsigned(*cookiep);
2136 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2137 				if (nvp != NULL) {
2138 					supports_nfsv4acls =
2139 					    nfs_supportsnfsv4acls(nvp);
2140 					NFSVOPUNLOCK(nvp, 0);
2141 				} else
2142 					supports_nfsv4acls = 0;
2143 				if (refp != NULL) {
2144 					dirlen += nfsrv_putreferralattr(nd,
2145 					    &savbits, refp, 0,
2146 					    &nd->nd_repstat);
2147 					if (nd->nd_repstat) {
2148 						if (nvp != NULL)
2149 							vrele(nvp);
2150 						if (needs_unbusy != 0)
2151 							vfs_unbusy(new_mp);
2152 						break;
2153 					}
2154 				} else if (r) {
2155 					dirlen += nfsvno_fillattr(nd, new_mp,
2156 					    nvp, nvap, &nfh, r, &rderrbits,
2157 					    nd->nd_cred, p, isdgram, 0,
2158 					    supports_nfsv4acls, at_root,
2159 					    mounted_on_fileno);
2160 				} else {
2161 					dirlen += nfsvno_fillattr(nd, new_mp,
2162 					    nvp, nvap, &nfh, r, &attrbits,
2163 					    nd->nd_cred, p, isdgram, 0,
2164 					    supports_nfsv4acls, at_root,
2165 					    mounted_on_fileno);
2166 				}
2167 				if (nvp != NULL)
2168 					vrele(nvp);
2169 				dirlen += (3 * NFSX_UNSIGNED);
2170 			}
2171 			if (needs_unbusy != 0)
2172 				vfs_unbusy(new_mp);
2173 			if (dirlen <= cnt)
2174 				entrycnt++;
2175 		}
2176 		cpos += dp->d_reclen;
2177 		dp = (struct dirent *)cpos;
2178 		cookiep++;
2179 		ncookies--;
2180 	}
2181 	vrele(vp);
2182 	vfs_unbusy(mp);
2183 
2184 	/*
2185 	 * If dirlen > cnt, we must strip off the last entry. If that
2186 	 * results in an empty reply, report NFSERR_TOOSMALL.
2187 	 */
2188 	if (dirlen > cnt || nd->nd_repstat) {
2189 		if (!nd->nd_repstat && entrycnt == 0)
2190 			nd->nd_repstat = NFSERR_TOOSMALL;
2191 		if (nd->nd_repstat)
2192 			newnfs_trimtrailing(nd, mb0, bpos0);
2193 		else
2194 			newnfs_trimtrailing(nd, mb1, bpos1);
2195 		eofflag = 0;
2196 	} else if (cpos < cend)
2197 		eofflag = 0;
2198 	if (!nd->nd_repstat) {
2199 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2200 		*tl++ = newnfs_false;
2201 		if (eofflag)
2202 			*tl = newnfs_true;
2203 		else
2204 			*tl = newnfs_false;
2205 	}
2206 	FREE((caddr_t)cookies, M_TEMP);
2207 	FREE((caddr_t)rbuf, M_TEMP);
2208 
2209 out:
2210 	NFSEXITCODE2(0, nd);
2211 	return (0);
2212 nfsmout:
2213 	vput(vp);
2214 	NFSEXITCODE2(error, nd);
2215 	return (error);
2216 }
2217 
2218 /*
2219  * Get the settable attributes out of the mbuf list.
2220  * (Return 0 or EBADRPC)
2221  */
2222 int
2223 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2224     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2225 {
2226 	u_int32_t *tl;
2227 	struct nfsv2_sattr *sp;
2228 	struct timeval curtime;
2229 	int error = 0, toclient = 0;
2230 
2231 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2232 	case ND_NFSV2:
2233 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2234 		/*
2235 		 * Some old clients didn't fill in the high order 16bits.
2236 		 * --> check the low order 2 bytes for 0xffff
2237 		 */
2238 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2239 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2240 		if (sp->sa_uid != newnfs_xdrneg1)
2241 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2242 		if (sp->sa_gid != newnfs_xdrneg1)
2243 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2244 		if (sp->sa_size != newnfs_xdrneg1)
2245 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2246 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2247 #ifdef notyet
2248 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2249 #else
2250 			nvap->na_atime.tv_sec =
2251 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2252 			nvap->na_atime.tv_nsec = 0;
2253 #endif
2254 		}
2255 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2256 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2257 		break;
2258 	case ND_NFSV3:
2259 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2260 		if (*tl == newnfs_true) {
2261 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2262 			nvap->na_mode = nfstov_mode(*tl);
2263 		}
2264 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2265 		if (*tl == newnfs_true) {
2266 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2267 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2268 		}
2269 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2270 		if (*tl == newnfs_true) {
2271 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2272 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2273 		}
2274 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2275 		if (*tl == newnfs_true) {
2276 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2277 			nvap->na_size = fxdr_hyper(tl);
2278 		}
2279 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2280 		switch (fxdr_unsigned(int, *tl)) {
2281 		case NFSV3SATTRTIME_TOCLIENT:
2282 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2283 			fxdr_nfsv3time(tl, &nvap->na_atime);
2284 			toclient = 1;
2285 			break;
2286 		case NFSV3SATTRTIME_TOSERVER:
2287 			NFSGETTIME(&curtime);
2288 			nvap->na_atime.tv_sec = curtime.tv_sec;
2289 			nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2290 			nvap->na_vaflags |= VA_UTIMES_NULL;
2291 			break;
2292 		};
2293 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2294 		switch (fxdr_unsigned(int, *tl)) {
2295 		case NFSV3SATTRTIME_TOCLIENT:
2296 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2297 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2298 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2299 			break;
2300 		case NFSV3SATTRTIME_TOSERVER:
2301 			NFSGETTIME(&curtime);
2302 			nvap->na_mtime.tv_sec = curtime.tv_sec;
2303 			nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2304 			if (!toclient)
2305 				nvap->na_vaflags |= VA_UTIMES_NULL;
2306 			break;
2307 		};
2308 		break;
2309 	case ND_NFSV4:
2310 		error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2311 	};
2312 nfsmout:
2313 	NFSEXITCODE2(error, nd);
2314 	return (error);
2315 }
2316 
2317 /*
2318  * Handle the setable attributes for V4.
2319  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2320  */
2321 int
2322 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2323     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2324 {
2325 	u_int32_t *tl;
2326 	int attrsum = 0;
2327 	int i, j;
2328 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2329 	int toclient = 0;
2330 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2331 	uid_t uid;
2332 	gid_t gid;
2333 	struct timeval curtime;
2334 
2335 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2336 	if (error)
2337 		goto nfsmout;
2338 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2339 	attrsize = fxdr_unsigned(int, *tl);
2340 
2341 	/*
2342 	 * Loop around getting the setable attributes. If an unsupported
2343 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2344 	 */
2345 	if (retnotsup) {
2346 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2347 		bitpos = NFSATTRBIT_MAX;
2348 	} else {
2349 		bitpos = 0;
2350 	}
2351 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2352 	    if (attrsum > attrsize) {
2353 		error = NFSERR_BADXDR;
2354 		goto nfsmout;
2355 	    }
2356 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2357 		switch (bitpos) {
2358 		case NFSATTRBIT_SIZE:
2359 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2360 			nvap->na_size = fxdr_hyper(tl);
2361 			attrsum += NFSX_HYPER;
2362 			break;
2363 		case NFSATTRBIT_ACL:
2364 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2365 			    p);
2366 			if (error)
2367 				goto nfsmout;
2368 			if (aceerr && !nd->nd_repstat)
2369 				nd->nd_repstat = aceerr;
2370 			attrsum += aclsize;
2371 			break;
2372 		case NFSATTRBIT_ARCHIVE:
2373 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2374 			if (!nd->nd_repstat)
2375 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2376 			attrsum += NFSX_UNSIGNED;
2377 			break;
2378 		case NFSATTRBIT_HIDDEN:
2379 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2380 			if (!nd->nd_repstat)
2381 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2382 			attrsum += NFSX_UNSIGNED;
2383 			break;
2384 		case NFSATTRBIT_MIMETYPE:
2385 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2386 			i = fxdr_unsigned(int, *tl);
2387 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2388 			if (error)
2389 				goto nfsmout;
2390 			if (!nd->nd_repstat)
2391 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2392 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2393 			break;
2394 		case NFSATTRBIT_MODE:
2395 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2396 			nvap->na_mode = nfstov_mode(*tl);
2397 			attrsum += NFSX_UNSIGNED;
2398 			break;
2399 		case NFSATTRBIT_OWNER:
2400 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2401 			j = fxdr_unsigned(int, *tl);
2402 			if (j < 0) {
2403 				error = NFSERR_BADXDR;
2404 				goto nfsmout;
2405 			}
2406 			if (j > NFSV4_SMALLSTR)
2407 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2408 			else
2409 				cp = namestr;
2410 			error = nfsrv_mtostr(nd, cp, j);
2411 			if (error) {
2412 				if (j > NFSV4_SMALLSTR)
2413 					free(cp, M_NFSSTRING);
2414 				goto nfsmout;
2415 			}
2416 			if (!nd->nd_repstat) {
2417 				nd->nd_repstat = nfsv4_strtouid(cp,j,&uid,p);
2418 				if (!nd->nd_repstat)
2419 					nvap->na_uid = uid;
2420 			}
2421 			if (j > NFSV4_SMALLSTR)
2422 				free(cp, M_NFSSTRING);
2423 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2424 			break;
2425 		case NFSATTRBIT_OWNERGROUP:
2426 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2427 			j = fxdr_unsigned(int, *tl);
2428 			if (j < 0) {
2429 				error = NFSERR_BADXDR;
2430 				goto nfsmout;
2431 			}
2432 			if (j > NFSV4_SMALLSTR)
2433 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2434 			else
2435 				cp = namestr;
2436 			error = nfsrv_mtostr(nd, cp, j);
2437 			if (error) {
2438 				if (j > NFSV4_SMALLSTR)
2439 					free(cp, M_NFSSTRING);
2440 				goto nfsmout;
2441 			}
2442 			if (!nd->nd_repstat) {
2443 				nd->nd_repstat = nfsv4_strtogid(cp,j,&gid,p);
2444 				if (!nd->nd_repstat)
2445 					nvap->na_gid = gid;
2446 			}
2447 			if (j > NFSV4_SMALLSTR)
2448 				free(cp, M_NFSSTRING);
2449 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2450 			break;
2451 		case NFSATTRBIT_SYSTEM:
2452 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2453 			if (!nd->nd_repstat)
2454 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2455 			attrsum += NFSX_UNSIGNED;
2456 			break;
2457 		case NFSATTRBIT_TIMEACCESSSET:
2458 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2459 			attrsum += NFSX_UNSIGNED;
2460 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2461 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2462 			    fxdr_nfsv4time(tl, &nvap->na_atime);
2463 			    toclient = 1;
2464 			    attrsum += NFSX_V4TIME;
2465 			} else {
2466 			    NFSGETTIME(&curtime);
2467 			    nvap->na_atime.tv_sec = curtime.tv_sec;
2468 			    nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2469 			    nvap->na_vaflags |= VA_UTIMES_NULL;
2470 			}
2471 			break;
2472 		case NFSATTRBIT_TIMEBACKUP:
2473 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2474 			if (!nd->nd_repstat)
2475 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2476 			attrsum += NFSX_V4TIME;
2477 			break;
2478 		case NFSATTRBIT_TIMECREATE:
2479 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2480 			if (!nd->nd_repstat)
2481 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2482 			attrsum += NFSX_V4TIME;
2483 			break;
2484 		case NFSATTRBIT_TIMEMODIFYSET:
2485 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2486 			attrsum += NFSX_UNSIGNED;
2487 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2488 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2489 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
2490 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
2491 			    attrsum += NFSX_V4TIME;
2492 			} else {
2493 			    NFSGETTIME(&curtime);
2494 			    nvap->na_mtime.tv_sec = curtime.tv_sec;
2495 			    nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2496 			    if (!toclient)
2497 				nvap->na_vaflags |= VA_UTIMES_NULL;
2498 			}
2499 			break;
2500 		default:
2501 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2502 			/*
2503 			 * set bitpos so we drop out of the loop.
2504 			 */
2505 			bitpos = NFSATTRBIT_MAX;
2506 			break;
2507 		};
2508 	}
2509 
2510 	/*
2511 	 * some clients pad the attrlist, so we need to skip over the
2512 	 * padding.
2513 	 */
2514 	if (attrsum > attrsize) {
2515 		error = NFSERR_BADXDR;
2516 	} else {
2517 		attrsize = NFSM_RNDUP(attrsize);
2518 		if (attrsum < attrsize)
2519 			error = nfsm_advance(nd, attrsize - attrsum, -1);
2520 	}
2521 nfsmout:
2522 	NFSEXITCODE2(error, nd);
2523 	return (error);
2524 }
2525 
2526 /*
2527  * Check/setup export credentials.
2528  */
2529 int
2530 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2531     struct ucred *credanon)
2532 {
2533 	int error = 0;
2534 
2535 	/*
2536 	 * Check/setup credentials.
2537 	 */
2538 	if (nd->nd_flag & ND_GSS)
2539 		exp->nes_exflag &= ~MNT_EXPORTANON;
2540 
2541 	/*
2542 	 * Check to see if the operation is allowed for this security flavor.
2543 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2544 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2545 	 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2546 	 */
2547 	if (nfsvno_testexp(nd, exp) &&
2548 	    nd->nd_procnum != NFSV4OP_SECINFO &&
2549 	    nd->nd_procnum != NFSPROC_FSINFO) {
2550 		if (nd->nd_flag & ND_NFSV4)
2551 			error = NFSERR_WRONGSEC;
2552 		else
2553 			error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2554 		goto out;
2555 	}
2556 
2557 	/*
2558 	 * Check to see if the file system is exported V4 only.
2559 	 */
2560 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2561 		error = NFSERR_PROGNOTV4;
2562 		goto out;
2563 	}
2564 
2565 	/*
2566 	 * Now, map the user credentials.
2567 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
2568 	 *  Fsinfo RPC. If set for anything else, this code might need
2569 	 *  to change.)
2570 	 */
2571 	if (NFSVNO_EXPORTED(exp) &&
2572 	    ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2573 	     NFSVNO_EXPORTANON(exp) ||
2574 	     (nd->nd_flag & ND_AUTHNONE))) {
2575 		nd->nd_cred->cr_uid = credanon->cr_uid;
2576 		nd->nd_cred->cr_gid = credanon->cr_gid;
2577 		crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2578 		    credanon->cr_groups);
2579 	}
2580 
2581 out:
2582 	NFSEXITCODE2(error, nd);
2583 	return (error);
2584 }
2585 
2586 /*
2587  * Check exports.
2588  */
2589 int
2590 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2591     struct ucred **credp)
2592 {
2593 	int i, error, *secflavors;
2594 
2595 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2596 	    &exp->nes_numsecflavor, &secflavors);
2597 	if (error) {
2598 		if (nfs_rootfhset) {
2599 			exp->nes_exflag = 0;
2600 			exp->nes_numsecflavor = 0;
2601 			error = 0;
2602 		}
2603 	} else {
2604 		/* Copy the security flavors. */
2605 		for (i = 0; i < exp->nes_numsecflavor; i++)
2606 			exp->nes_secflavors[i] = secflavors[i];
2607 	}
2608 	NFSEXITCODE(error);
2609 	return (error);
2610 }
2611 
2612 /*
2613  * Get a vnode for a file handle and export stuff.
2614  */
2615 int
2616 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2617     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2618     struct ucred **credp)
2619 {
2620 	int i, error, *secflavors;
2621 
2622 	*credp = NULL;
2623 	exp->nes_numsecflavor = 0;
2624 	if (VFS_NEEDSGIANT(mp))
2625 		error = ESTALE;
2626 	else
2627 		error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, vpp);
2628 	if (error != 0)
2629 		/* Make sure the server replies ESTALE to the client. */
2630 		error = ESTALE;
2631 	if (nam && !error) {
2632 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2633 		    &exp->nes_numsecflavor, &secflavors);
2634 		if (error) {
2635 			if (nfs_rootfhset) {
2636 				exp->nes_exflag = 0;
2637 				exp->nes_numsecflavor = 0;
2638 				error = 0;
2639 			} else {
2640 				vput(*vpp);
2641 			}
2642 		} else {
2643 			/* Copy the security flavors. */
2644 			for (i = 0; i < exp->nes_numsecflavor; i++)
2645 				exp->nes_secflavors[i] = secflavors[i];
2646 		}
2647 	}
2648 	if (error == 0 && lktype == LK_SHARED)
2649 		/*
2650 		 * It would be much better to pass lktype to VFS_FHTOVP(),
2651 		 * but this will have to do until VFS_FHTOVP() has a lock
2652 		 * type argument like VFS_VGET().
2653 		 */
2654 		NFSVOPLOCK(*vpp, LK_DOWNGRADE | LK_RETRY);
2655 
2656 	NFSEXITCODE(error);
2657 	return (error);
2658 }
2659 
2660 /*
2661  * nfsd_fhtovp() - convert a fh to a vnode ptr
2662  * 	- look up fsid in mount list (if not found ret error)
2663  *	- get vp and export rights by calling nfsvno_fhtovp()
2664  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2665  *	  for AUTH_SYS
2666  *	- if mpp != NULL, return the mount point so that it can
2667  *	  be used for vn_finished_write() by the caller
2668  */
2669 void
2670 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2671     struct vnode **vpp, struct nfsexstuff *exp,
2672     struct mount **mpp, int startwrite, struct thread *p)
2673 {
2674 	struct mount *mp;
2675 	struct ucred *credanon;
2676 	fhandle_t *fhp;
2677 
2678 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
2679 	/*
2680 	 * Check for the special case of the nfsv4root_fh.
2681 	 */
2682 	mp = vfs_busyfs(&fhp->fh_fsid);
2683 	if (mpp != NULL)
2684 		*mpp = mp;
2685 	if (mp == NULL) {
2686 		*vpp = NULL;
2687 		nd->nd_repstat = ESTALE;
2688 		goto out;
2689 	}
2690 
2691 	if (startwrite)
2692 		vn_start_write(NULL, mpp, V_WAIT);
2693 
2694 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2695 	    &credanon);
2696 	vfs_unbusy(mp);
2697 
2698 	/*
2699 	 * For NFSv4 without a pseudo root fs, unexported file handles
2700 	 * can be returned, so that Lookup works everywhere.
2701 	 */
2702 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2703 	    !(nd->nd_flag & ND_NFSV4)) {
2704 		vput(*vpp);
2705 		nd->nd_repstat = EACCES;
2706 	}
2707 
2708 	/*
2709 	 * Personally, I've never seen any point in requiring a
2710 	 * reserved port#, since only in the rare case where the
2711 	 * clients are all boxes with secure system priviledges,
2712 	 * does it provide any enhanced security, but... some people
2713 	 * believe it to be useful and keep putting this code back in.
2714 	 * (There is also some "security checker" out there that
2715 	 *  complains if the nfs server doesn't enforce this.)
2716 	 * However, note the following:
2717 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
2718 	 *	required.
2719 	 * RFC2623 recommends that, if a reserved port# is checked for,
2720 	 *	that there be a way to turn that off--> ifdef'd.
2721 	 */
2722 #ifdef NFS_REQRSVPORT
2723 	if (!nd->nd_repstat) {
2724 		struct sockaddr_in *saddr;
2725 		struct sockaddr_in6 *saddr6;
2726 
2727 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2728 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2729 		if (!(nd->nd_flag & ND_NFSV4) &&
2730 		    ((saddr->sin_family == AF_INET &&
2731 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2732 		     (saddr6->sin6_family == AF_INET6 &&
2733 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2734 			vput(*vpp);
2735 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2736 		}
2737 	}
2738 #endif	/* NFS_REQRSVPORT */
2739 
2740 	/*
2741 	 * Check/setup credentials.
2742 	 */
2743 	if (!nd->nd_repstat) {
2744 		nd->nd_saveduid = nd->nd_cred->cr_uid;
2745 		nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2746 		if (nd->nd_repstat)
2747 			vput(*vpp);
2748 	}
2749 	if (credanon != NULL)
2750 		crfree(credanon);
2751 	if (nd->nd_repstat) {
2752 		if (startwrite)
2753 			vn_finished_write(mp);
2754 		*vpp = NULL;
2755 		if (mpp != NULL)
2756 			*mpp = NULL;
2757 	}
2758 
2759 out:
2760 	NFSEXITCODE2(0, nd);
2761 }
2762 
2763 /*
2764  * glue for fp.
2765  */
2766 int
2767 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2768 {
2769 	struct filedesc *fdp;
2770 	struct file *fp;
2771 	int error = 0;
2772 
2773 	fdp = p->td_proc->p_fd;
2774 	if (fd >= fdp->fd_nfiles ||
2775 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
2776 		error = EBADF;
2777 		goto out;
2778 	}
2779 	*fpp = fp;
2780 
2781 out:
2782 	NFSEXITCODE(error);
2783 	return (error);
2784 }
2785 
2786 /*
2787  * Called from nfssvc() to update the exports list. Just call
2788  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2789  * in the mount list.
2790  */
2791 int
2792 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2793 {
2794 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2795 	int error = 0;
2796 	struct nameidata nd;
2797 	fhandle_t fh;
2798 
2799 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2800 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2801 		nfs_rootfhset = 0;
2802 	else if (error == 0) {
2803 		if (nfsexargp->fspec == NULL) {
2804 			error = EPERM;
2805 			goto out;
2806 		}
2807 		/*
2808 		 * If fspec != NULL, this is the v4root path.
2809 		 */
2810 		NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_USERSPACE,
2811 		    nfsexargp->fspec, p);
2812 		if ((error = namei(&nd)) != 0)
2813 			goto out;
2814 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
2815 		vrele(nd.ni_vp);
2816 		if (!error) {
2817 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2818 			NFSBCOPY((caddr_t)&fh,
2819 			    nfs_rootfh.nfsrvfh_data,
2820 			    sizeof (fhandle_t));
2821 			nfs_rootfhset = 1;
2822 		}
2823 	}
2824 
2825 out:
2826 	NFSEXITCODE(error);
2827 	return (error);
2828 }
2829 
2830 /*
2831  * Get the tcp socket sequence numbers we need.
2832  * (Maybe this should be moved to the tcp sources?)
2833  */
2834 int
2835 nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap)
2836 {
2837 	struct inpcb *inp;
2838 	struct tcpcb *tp;
2839 	int error = 0;
2840 
2841 	inp = sotoinpcb(so);
2842 	KASSERT(inp != NULL, ("nfsrv_getsocksndseq: inp == NULL"));
2843 	INP_RLOCK(inp);
2844 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2845 		INP_RUNLOCK(inp);
2846 		error = EPIPE;
2847 		goto out;
2848 	}
2849 	tp = intotcpcb(inp);
2850 	if (tp->t_state != TCPS_ESTABLISHED) {
2851 		INP_RUNLOCK(inp);
2852 		error = EPIPE;
2853 		goto out;
2854 	}
2855 	*maxp = tp->snd_max;
2856 	*unap = tp->snd_una;
2857 	INP_RUNLOCK(inp);
2858 
2859 out:
2860 	NFSEXITCODE(error);
2861 	return (error);
2862 }
2863 
2864 /*
2865  * This function needs to test to see if the system is near its limit
2866  * for memory allocation via malloc() or mget() and return True iff
2867  * either of these resources are near their limit.
2868  * XXX (For now, this is just a stub.)
2869  */
2870 int nfsrv_testmalloclimit = 0;
2871 int
2872 nfsrv_mallocmget_limit(void)
2873 {
2874 	static int printmesg = 0;
2875 	static int testval = 1;
2876 
2877 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2878 		if ((printmesg++ % 100) == 0)
2879 			printf("nfsd: malloc/mget near limit\n");
2880 		return (1);
2881 	}
2882 	return (0);
2883 }
2884 
2885 /*
2886  * BSD specific initialization of a mount point.
2887  */
2888 void
2889 nfsd_mntinit(void)
2890 {
2891 	static int inited = 0;
2892 
2893 	if (inited)
2894 		return;
2895 	inited = 1;
2896 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2897 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2898 	nfsv4root_mnt.mnt_export = NULL;
2899 	TAILQ_INIT(&nfsv4root_opt);
2900 	TAILQ_INIT(&nfsv4root_newopt);
2901 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2902 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2903 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
2904 }
2905 
2906 /*
2907  * Get a vnode for a file handle, without checking exports, etc.
2908  */
2909 struct vnode *
2910 nfsvno_getvp(fhandle_t *fhp)
2911 {
2912 	struct mount *mp;
2913 	struct vnode *vp;
2914 	int error;
2915 
2916 	mp = vfs_busyfs(&fhp->fh_fsid);
2917 	if (mp == NULL)
2918 		return (NULL);
2919 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2920 	vfs_unbusy(mp);
2921 	if (error)
2922 		return (NULL);
2923 	return (vp);
2924 }
2925 
2926 /*
2927  * Do a local VOP_ADVLOCK().
2928  */
2929 int
2930 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2931     u_int64_t end, struct thread *td)
2932 {
2933 	int error = 0;
2934 	struct flock fl;
2935 	u_int64_t tlen;
2936 
2937 	if (nfsrv_dolocallocks == 0)
2938 		goto out;
2939 
2940 	/* Check for VI_DOOMED here, so that VOP_ADVLOCK() isn't performed. */
2941 	if ((vp->v_iflag & VI_DOOMED) != 0) {
2942 		error = EPERM;
2943 		goto out;
2944 	}
2945 
2946 	fl.l_whence = SEEK_SET;
2947 	fl.l_type = ftype;
2948 	fl.l_start = (off_t)first;
2949 	if (end == NFS64BITSSET) {
2950 		fl.l_len = 0;
2951 	} else {
2952 		tlen = end - first;
2953 		fl.l_len = (off_t)tlen;
2954 	}
2955 	/*
2956 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2957 	 * values for all calls, so that all locks will be held by the
2958 	 * nfsd server. (The nfsd server handles conflicts between the
2959 	 * various clients.)
2960 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2961 	 * bytes, so it can't be put in l_sysid.
2962 	 */
2963 	if (nfsv4_sysid == 0)
2964 		nfsv4_sysid = nlm_acquire_next_sysid();
2965 	fl.l_pid = (pid_t)0;
2966 	fl.l_sysid = (int)nfsv4_sysid;
2967 
2968 	NFSVOPUNLOCK(vp, 0);
2969 	if (ftype == F_UNLCK)
2970 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
2971 		    (F_POSIX | F_REMOTE));
2972 	else
2973 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
2974 		    (F_POSIX | F_REMOTE));
2975 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2976 
2977 out:
2978 	NFSEXITCODE(error);
2979 	return (error);
2980 }
2981 
2982 /*
2983  * Check the nfsv4 root exports.
2984  */
2985 int
2986 nfsvno_v4rootexport(struct nfsrv_descript *nd)
2987 {
2988 	struct ucred *credanon;
2989 	int exflags, error = 0, numsecflavor, *secflavors, i;
2990 
2991 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
2992 	    &credanon, &numsecflavor, &secflavors);
2993 	if (error) {
2994 		error = NFSERR_PROGUNAVAIL;
2995 		goto out;
2996 	}
2997 	if (credanon != NULL)
2998 		crfree(credanon);
2999 	for (i = 0; i < numsecflavor; i++) {
3000 		if (secflavors[i] == AUTH_SYS)
3001 			nd->nd_flag |= ND_EXAUTHSYS;
3002 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3003 			nd->nd_flag |= ND_EXGSS;
3004 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3005 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3006 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3007 			nd->nd_flag |= ND_EXGSSPRIVACY;
3008 	}
3009 
3010 out:
3011 	NFSEXITCODE(error);
3012 	return (error);
3013 }
3014 
3015 /*
3016  * Nfs server psuedo system call for the nfsd's
3017  */
3018 /*
3019  * MPSAFE
3020  */
3021 static int
3022 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3023 {
3024 	struct file *fp;
3025 	struct nfsd_addsock_args sockarg;
3026 	struct nfsd_nfsd_args nfsdarg;
3027 	int error;
3028 
3029 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3030 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3031 		if (error)
3032 			goto out;
3033 		/*
3034 		 * Since we don't know what rights might be required,
3035 		 * pretend that we need them all. It is better to be too
3036 		 * careful than too reckless.
3037 		 */
3038 		if ((error = fget(td, sockarg.sock, CAP_SOCK_ALL, &fp)) != 0)
3039 			goto out;
3040 		if (fp->f_type != DTYPE_SOCKET) {
3041 			fdrop(fp, td);
3042 			error = EPERM;
3043 			goto out;
3044 		}
3045 		error = nfsrvd_addsock(fp);
3046 		fdrop(fp, td);
3047 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3048 		if (uap->argp == NULL) {
3049 			error = EINVAL;
3050 			goto out;
3051 		}
3052 		error = copyin(uap->argp, (caddr_t)&nfsdarg,
3053 		    sizeof (nfsdarg));
3054 		if (error)
3055 			goto out;
3056 		error = nfsrvd_nfsd(td, &nfsdarg);
3057 	} else {
3058 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3059 	}
3060 
3061 out:
3062 	NFSEXITCODE(error);
3063 	return (error);
3064 }
3065 
3066 static int
3067 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3068 {
3069 	struct nfsex_args export;
3070 	struct file *fp = NULL;
3071 	int stablefd, len;
3072 	struct nfsd_clid adminrevoke;
3073 	struct nfsd_dumplist dumplist;
3074 	struct nfsd_dumpclients *dumpclients;
3075 	struct nfsd_dumplocklist dumplocklist;
3076 	struct nfsd_dumplocks *dumplocks;
3077 	struct nameidata nd;
3078 	vnode_t vp;
3079 	int error = EINVAL;
3080 	struct proc *procp;
3081 
3082 	if (uap->flag & NFSSVC_PUBLICFH) {
3083 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3084 		    sizeof (fhandle_t));
3085 		error = copyin(uap->argp,
3086 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3087 		if (!error)
3088 			nfs_pubfhset = 1;
3089 	} else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3090 		error = copyin(uap->argp,(caddr_t)&export,
3091 		    sizeof (struct nfsex_args));
3092 		if (!error)
3093 			error = nfsrv_v4rootexport(&export, cred, p);
3094 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3095 		nfs_pubfhset = 0;
3096 		error = 0;
3097 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3098 		error = copyin(uap->argp, (caddr_t)&stablefd,
3099 		    sizeof (int));
3100 		if (!error)
3101 			error = fp_getfvp(p, stablefd, &fp, &vp);
3102 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3103 			error = EBADF;
3104 		if (!error && newnfs_numnfsd != 0)
3105 			error = EPERM;
3106 		if (!error) {
3107 			nfsrv_stablefirst.nsf_fp = fp;
3108 			nfsrv_setupstable(p);
3109 		}
3110 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3111 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3112 		    sizeof (struct nfsd_clid));
3113 		if (!error)
3114 			error = nfsrv_adminrevoke(&adminrevoke, p);
3115 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3116 		error = copyin(uap->argp, (caddr_t)&dumplist,
3117 		    sizeof (struct nfsd_dumplist));
3118 		if (!error && (dumplist.ndl_size < 1 ||
3119 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3120 			error = EPERM;
3121 		if (!error) {
3122 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3123 		    dumpclients = (struct nfsd_dumpclients *)malloc(len,
3124 			M_TEMP, M_WAITOK);
3125 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3126 		    error = copyout(dumpclients,
3127 			CAST_USER_ADDR_T(dumplist.ndl_list), len);
3128 		    free((caddr_t)dumpclients, M_TEMP);
3129 		}
3130 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3131 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3132 		    sizeof (struct nfsd_dumplocklist));
3133 		if (!error && (dumplocklist.ndllck_size < 1 ||
3134 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3135 			error = EPERM;
3136 		if (!error)
3137 			error = nfsrv_lookupfilename(&nd,
3138 				dumplocklist.ndllck_fname, p);
3139 		if (!error) {
3140 			len = sizeof (struct nfsd_dumplocks) *
3141 				dumplocklist.ndllck_size;
3142 			dumplocks = (struct nfsd_dumplocks *)malloc(len,
3143 				M_TEMP, M_WAITOK);
3144 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3145 			    dumplocklist.ndllck_size, p);
3146 			vput(nd.ni_vp);
3147 			error = copyout(dumplocks,
3148 			    CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3149 			free((caddr_t)dumplocks, M_TEMP);
3150 		}
3151 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3152 		procp = p->td_proc;
3153 		PROC_LOCK(procp);
3154 		nfsd_master_pid = procp->p_pid;
3155 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3156 		nfsd_master_start = procp->p_stats->p_start;
3157 		nfsd_master_proc = procp;
3158 		PROC_UNLOCK(procp);
3159 	}
3160 
3161 	NFSEXITCODE(error);
3162 	return (error);
3163 }
3164 
3165 /*
3166  * Check exports.
3167  * Returns 0 if ok, 1 otherwise.
3168  */
3169 int
3170 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3171 {
3172 	int i;
3173 
3174 	/*
3175 	 * This seems odd, but allow the case where the security flavor
3176 	 * list is empty. This happens when NFSv4 is traversing non-exported
3177 	 * file systems. Exported file systems should always have a non-empty
3178 	 * security flavor list.
3179 	 */
3180 	if (exp->nes_numsecflavor == 0)
3181 		return (0);
3182 
3183 	for (i = 0; i < exp->nes_numsecflavor; i++) {
3184 		/*
3185 		 * The tests for privacy and integrity must be first,
3186 		 * since ND_GSS is set for everything but AUTH_SYS.
3187 		 */
3188 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3189 		    (nd->nd_flag & ND_GSSPRIVACY))
3190 			return (0);
3191 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3192 		    (nd->nd_flag & ND_GSSINTEGRITY))
3193 			return (0);
3194 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3195 		    (nd->nd_flag & ND_GSS))
3196 			return (0);
3197 		if (exp->nes_secflavors[i] == AUTH_SYS &&
3198 		    (nd->nd_flag & ND_GSS) == 0)
3199 			return (0);
3200 	}
3201 	return (1);
3202 }
3203 
3204 /*
3205  * Calculate a hash value for the fid in a file handle.
3206  */
3207 uint32_t
3208 nfsrv_hashfh(fhandle_t *fhp)
3209 {
3210 	uint32_t hashval;
3211 
3212 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3213 	return (hashval);
3214 }
3215 
3216 /*
3217  * Signal the userland master nfsd to backup the stable restart file.
3218  */
3219 void
3220 nfsrv_backupstable(void)
3221 {
3222 	struct proc *procp;
3223 
3224 	if (nfsd_master_proc != NULL) {
3225 		procp = pfind(nfsd_master_pid);
3226 		/* Try to make sure it is the correct process. */
3227 		if (procp == nfsd_master_proc &&
3228 		    procp->p_stats->p_start.tv_sec ==
3229 		    nfsd_master_start.tv_sec &&
3230 		    procp->p_stats->p_start.tv_usec ==
3231 		    nfsd_master_start.tv_usec &&
3232 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
3233 			kern_psignal(procp, SIGUSR2);
3234 		else
3235 			nfsd_master_proc = NULL;
3236 
3237 		if (procp != NULL)
3238 			PROC_UNLOCK(procp);
3239 	}
3240 }
3241 
3242 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3243 
3244 /*
3245  * Called once to initialize data structures...
3246  */
3247 static int
3248 nfsd_modevent(module_t mod, int type, void *data)
3249 {
3250 	int error = 0;
3251 	static int loaded = 0;
3252 
3253 	switch (type) {
3254 	case MOD_LOAD:
3255 		if (loaded)
3256 			goto out;
3257 		newnfs_portinit();
3258 		mtx_init(&nfs_cache_mutex, "nfs_cache_mutex", NULL, MTX_DEF);
3259 		mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3260 		mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3261 		    MTX_DEF);
3262 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3263 		nfsrvd_initcache();
3264 		nfsd_init();
3265 		NFSD_LOCK();
3266 		nfsrvd_init(0);
3267 		NFSD_UNLOCK();
3268 		nfsd_mntinit();
3269 #ifdef VV_DISABLEDELEG
3270 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3271 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3272 #endif
3273 		nfsd_call_servertimer = nfsrv_servertimer;
3274 		nfsd_call_nfsd = nfssvc_nfsd;
3275 		loaded = 1;
3276 		break;
3277 
3278 	case MOD_UNLOAD:
3279 		if (newnfs_numnfsd != 0) {
3280 			error = EBUSY;
3281 			break;
3282 		}
3283 
3284 #ifdef VV_DISABLEDELEG
3285 		vn_deleg_ops.vndeleg_recall = NULL;
3286 		vn_deleg_ops.vndeleg_disable = NULL;
3287 #endif
3288 		nfsd_call_servertimer = NULL;
3289 		nfsd_call_nfsd = NULL;
3290 
3291 		/* Clean out all NFSv4 state. */
3292 		nfsrv_throwawayallstate(curthread);
3293 
3294 		/* Clean the NFS server reply cache */
3295 		nfsrvd_cleancache();
3296 
3297 		/* Free up the krpc server pool. */
3298 		if (nfsrvd_pool != NULL)
3299 			svcpool_destroy(nfsrvd_pool);
3300 
3301 		/* and get rid of the locks */
3302 		mtx_destroy(&nfs_cache_mutex);
3303 		mtx_destroy(&nfs_v4root_mutex);
3304 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3305 		lockdestroy(&nfsv4root_mnt.mnt_explock);
3306 		loaded = 0;
3307 		break;
3308 	default:
3309 		error = EOPNOTSUPP;
3310 		break;
3311 	}
3312 
3313 out:
3314 	NFSEXITCODE(error);
3315 	return (error);
3316 }
3317 static moduledata_t nfsd_mod = {
3318 	"nfsd",
3319 	nfsd_modevent,
3320 	NULL,
3321 };
3322 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3323 
3324 /* So that loader and kldload(2) can find us, wherever we are.. */
3325 MODULE_VERSION(nfsd, 1);
3326 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3327 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3328 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3329 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3330 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3331 
3332