xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision a3cf0ef5a295c885c895fabfd56470c0d1db322d)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 /*
38  * Functions that perform the vfs operations required by the routines in
39  * nfsd_serv.c. It is hoped that this change will make the server more
40  * portable.
41  */
42 
43 #include <fs/nfs/nfsport.h>
44 #include <sys/hash.h>
45 #include <sys/sysctl.h>
46 #include <nlm/nlm_prot.h>
47 #include <nlm/nlm.h>
48 
49 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
50 extern int nfsv4root_set;
51 extern int nfsrv_useacl;
52 extern int newnfs_numnfsd;
53 extern struct mount nfsv4root_mnt;
54 extern struct nfsrv_stablefirst nfsrv_stablefirst;
55 extern void (*nfsd_call_servertimer)(void);
56 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
57 NFSDLOCKMUTEX;
58 struct mtx nfs_cache_mutex;
59 struct mtx nfs_v4root_mutex;
60 struct nfsrvfh nfs_rootfh, nfs_pubfh;
61 int nfs_pubfhset = 0, nfs_rootfhset = 0;
62 static uint32_t nfsv4_sysid = 0;
63 
64 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
65     struct ucred *);
66 
67 static int enable_crossmntpt = 1;
68 static int nfs_commit_blks;
69 static int nfs_commit_miss;
70 extern int nfsrv_issuedelegs;
71 extern int nfsrv_dolocallocks;
72 
73 SYSCTL_DECL(_vfs_newnfs);
74 SYSCTL_INT(_vfs_newnfs, OID_AUTO, mirrormnt, CTLFLAG_RW, &enable_crossmntpt,
75     0, "Enable nfsd to cross mount points");
76 SYSCTL_INT(_vfs_newnfs, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
77     0, "");
78 SYSCTL_INT(_vfs_newnfs, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
79     0, "");
80 SYSCTL_INT(_vfs_newnfs, OID_AUTO, issue_delegations, CTLFLAG_RW,
81     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
82 SYSCTL_INT(_vfs_newnfs, OID_AUTO, enable_locallocks, CTLFLAG_RW,
83     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
84 
85 #define	NUM_HEURISTIC		1017
86 #define	NHUSE_INIT		64
87 #define	NHUSE_INC		16
88 #define	NHUSE_MAX		2048
89 
90 static struct nfsheur {
91 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
92 	off_t nh_nextr;		/* next offset for sequential detection */
93 	int nh_use;		/* use count for selection */
94 	int nh_seqcount;	/* heuristic */
95 } nfsheur[NUM_HEURISTIC];
96 
97 
98 /*
99  * Get attributes into nfsvattr structure.
100  */
101 int
102 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
103     struct thread *p)
104 {
105 	int error, lockedit = 0;
106 
107 	/* Since FreeBSD insists the vnode be locked... */
108 	if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
109 		lockedit = 1;
110 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
111 	}
112 	error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
113 	if (lockedit)
114 		NFSVOPUNLOCK(vp, 0, p);
115 	return (error);
116 }
117 
118 /*
119  * Get a file handle for a vnode.
120  */
121 int
122 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
123 {
124 	int error;
125 
126 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
127 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
128 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
129 	return (error);
130 }
131 
132 /*
133  * Perform access checking for vnodes obtained from file handles that would
134  * refer to files already opened by a Unix client. You cannot just use
135  * vn_writechk() and VOP_ACCESSX() for two reasons.
136  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
137  *     case.
138  * 2 - The owner is to be given access irrespective of mode bits for some
139  *     operations, so that processes that chmod after opening a file don't
140  *     break.
141  */
142 int
143 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
144     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
145     u_int32_t *supportedtypep)
146 {
147 	struct vattr vattr;
148 	int error = 0, getret = 0;
149 
150 	if (accmode & VWRITE) {
151 		/* Just vn_writechk() changed to check rdonly */
152 		/*
153 		 * Disallow write attempts on read-only file systems;
154 		 * unless the file is a socket or a block or character
155 		 * device resident on the file system.
156 		 */
157 		if (NFSVNO_EXRDONLY(exp) ||
158 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
159 			switch (vp->v_type) {
160 			case VREG:
161 			case VDIR:
162 			case VLNK:
163 				return (EROFS);
164 			default:
165 				break;
166 			}
167 		}
168 		/*
169 		 * If there's shared text associated with
170 		 * the inode, try to free it up once.  If
171 		 * we fail, we can't allow writing.
172 		 */
173 		if (vp->v_vflag & VV_TEXT)
174 			return (ETXTBSY);
175 	}
176 	if (vpislocked == 0)
177 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
178 
179 	/*
180 	 * Should the override still be applied when ACLs are enabled?
181 	 */
182 	error = VOP_ACCESSX(vp, accmode, cred, p);
183 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
184 		/*
185 		 * Try again with VEXPLICIT_DENY, to see if the test for
186 		 * deletion is supported.
187 		 */
188 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
189 		if (error == 0) {
190 			if (vp->v_type == VDIR) {
191 				accmode &= ~(VDELETE | VDELETE_CHILD);
192 				accmode |= VWRITE;
193 				error = VOP_ACCESSX(vp, accmode, cred, p);
194 			} else if (supportedtypep != NULL) {
195 				*supportedtypep &= ~NFSACCESS_DELETE;
196 			}
197 		}
198 	}
199 
200 	/*
201 	 * Allow certain operations for the owner (reads and writes
202 	 * on files that are already open).
203 	 */
204 	if (override != NFSACCCHK_NOOVERRIDE &&
205 	    (error == EPERM || error == EACCES)) {
206 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
207 			error = 0;
208 		else if (override & NFSACCCHK_ALLOWOWNER) {
209 			getret = VOP_GETATTR(vp, &vattr, cred);
210 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
211 				error = 0;
212 		}
213 	}
214 	if (vpislocked == 0)
215 		NFSVOPUNLOCK(vp, 0, p);
216 	return (error);
217 }
218 
219 /*
220  * Set attribute(s) vnop.
221  */
222 int
223 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
224     struct thread *p, struct nfsexstuff *exp)
225 {
226 	int error;
227 
228 	error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
229 	return (error);
230 }
231 
232 /*
233  * Set up nameidata for a lookup() call and do it
234  * For the cases where we are crossing mount points
235  * (looking up the public fh path or the v4 root path when
236  *  not using a pseudo-root fs), set/release the Giant lock,
237  * as required.
238  */
239 int
240 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
241     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
242     struct vnode **retdirp)
243 {
244 	struct componentname *cnp = &ndp->ni_cnd;
245 	int i;
246 	struct iovec aiov;
247 	struct uio auio;
248 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
249 	int error = 0, crossmnt;
250 	char *cp;
251 
252 	*retdirp = NULL;
253 	cnp->cn_nameptr = cnp->cn_pnbuf;
254 	/*
255 	 * Extract and set starting directory.
256 	 */
257 	if (dp->v_type != VDIR) {
258 		if (islocked)
259 			vput(dp);
260 		else
261 			vrele(dp);
262 		nfsvno_relpathbuf(ndp);
263 		return (ENOTDIR);
264 	}
265 	if (islocked)
266 		NFSVOPUNLOCK(dp, 0, p);
267 	VREF(dp);
268 	*retdirp = dp;
269 	if (NFSVNO_EXRDONLY(exp))
270 		cnp->cn_flags |= RDONLY;
271 	ndp->ni_segflg = UIO_SYSSPACE;
272 	crossmnt = 1;
273 
274 	if (nd->nd_flag & ND_PUBLOOKUP) {
275 		ndp->ni_loopcnt = 0;
276 		if (cnp->cn_pnbuf[0] == '/') {
277 			vrele(dp);
278 			/*
279 			 * Check for degenerate pathnames here, since lookup()
280 			 * panics on them.
281 			 */
282 			for (i = 1; i < ndp->ni_pathlen; i++)
283 				if (cnp->cn_pnbuf[i] != '/')
284 					break;
285 			if (i == ndp->ni_pathlen) {
286 				error = NFSERR_ACCES;
287 				goto out;
288 			}
289 			dp = rootvnode;
290 			VREF(dp);
291 		}
292 	} else if ((enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
293 	    (nd->nd_flag & ND_NFSV4) == 0) {
294 		/*
295 		 * Only cross mount points for NFSv4 when doing a
296 		 * mount while traversing the file system above
297 		 * the mount point, unless enable_crossmntpt is set.
298 		 */
299 		cnp->cn_flags |= NOCROSSMOUNT;
300 		crossmnt = 0;
301 	}
302 
303 	/*
304 	 * Initialize for scan, set ni_startdir and bump ref on dp again
305 	 * becuase lookup() will dereference ni_startdir.
306 	 */
307 
308 	cnp->cn_thread = p;
309 	ndp->ni_startdir = dp;
310 	ndp->ni_rootdir = rootvnode;
311 
312 	if (!lockleaf)
313 		cnp->cn_flags |= LOCKLEAF;
314 	for (;;) {
315 		cnp->cn_nameptr = cnp->cn_pnbuf;
316 		/*
317 		 * Call lookup() to do the real work.  If an error occurs,
318 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
319 		 * we do not have to dereference anything before returning.
320 		 * In either case ni_startdir will be dereferenced and NULLed
321 		 * out.
322 		 */
323 		if (exp->nes_vfslocked)
324 			ndp->ni_cnd.cn_flags |= GIANTHELD;
325 		error = lookup(ndp);
326 		/*
327 		 * The Giant lock should only change when
328 		 * crossing mount points.
329 		 */
330 		if (crossmnt) {
331 			exp->nes_vfslocked =
332 			    (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
333 			ndp->ni_cnd.cn_flags &= ~GIANTHELD;
334 		}
335 		if (error)
336 			break;
337 
338 		/*
339 		 * Check for encountering a symbolic link.  Trivial
340 		 * termination occurs if no symlink encountered.
341 		 */
342 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
343 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
344 				nfsvno_relpathbuf(ndp);
345 			if (ndp->ni_vp && !lockleaf)
346 				NFSVOPUNLOCK(ndp->ni_vp, 0, p);
347 			break;
348 		}
349 
350 		/*
351 		 * Validate symlink
352 		 */
353 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
354 			NFSVOPUNLOCK(ndp->ni_dvp, 0, p);
355 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
356 			error = EINVAL;
357 			goto badlink2;
358 		}
359 
360 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
361 			error = ELOOP;
362 			goto badlink2;
363 		}
364 		if (ndp->ni_pathlen > 1)
365 			cp = uma_zalloc(namei_zone, M_WAITOK);
366 		else
367 			cp = cnp->cn_pnbuf;
368 		aiov.iov_base = cp;
369 		aiov.iov_len = MAXPATHLEN;
370 		auio.uio_iov = &aiov;
371 		auio.uio_iovcnt = 1;
372 		auio.uio_offset = 0;
373 		auio.uio_rw = UIO_READ;
374 		auio.uio_segflg = UIO_SYSSPACE;
375 		auio.uio_td = NULL;
376 		auio.uio_resid = MAXPATHLEN;
377 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
378 		if (error) {
379 		badlink1:
380 			if (ndp->ni_pathlen > 1)
381 				uma_zfree(namei_zone, cp);
382 		badlink2:
383 			vrele(ndp->ni_dvp);
384 			vput(ndp->ni_vp);
385 			break;
386 		}
387 		linklen = MAXPATHLEN - auio.uio_resid;
388 		if (linklen == 0) {
389 			error = ENOENT;
390 			goto badlink1;
391 		}
392 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
393 			error = ENAMETOOLONG;
394 			goto badlink1;
395 		}
396 
397 		/*
398 		 * Adjust or replace path
399 		 */
400 		if (ndp->ni_pathlen > 1) {
401 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
402 			uma_zfree(namei_zone, cnp->cn_pnbuf);
403 			cnp->cn_pnbuf = cp;
404 		} else
405 			cnp->cn_pnbuf[linklen] = '\0';
406 		ndp->ni_pathlen += linklen;
407 
408 		/*
409 		 * Cleanup refs for next loop and check if root directory
410 		 * should replace current directory.  Normally ni_dvp
411 		 * becomes the new base directory and is cleaned up when
412 		 * we loop.  Explicitly null pointers after invalidation
413 		 * to clarify operation.
414 		 */
415 		vput(ndp->ni_vp);
416 		ndp->ni_vp = NULL;
417 
418 		if (cnp->cn_pnbuf[0] == '/') {
419 			vrele(ndp->ni_dvp);
420 			ndp->ni_dvp = ndp->ni_rootdir;
421 			VREF(ndp->ni_dvp);
422 		}
423 		ndp->ni_startdir = ndp->ni_dvp;
424 		ndp->ni_dvp = NULL;
425 	}
426 	if (!lockleaf)
427 		cnp->cn_flags &= ~LOCKLEAF;
428 
429 out:
430 	if (error) {
431 		uma_zfree(namei_zone, cnp->cn_pnbuf);
432 		ndp->ni_vp = NULL;
433 		ndp->ni_dvp = NULL;
434 		ndp->ni_startdir = NULL;
435 		cnp->cn_flags &= ~HASBUF;
436 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
437 		ndp->ni_dvp = NULL;
438 	}
439 	return (error);
440 }
441 
442 /*
443  * Set up a pathname buffer and return a pointer to it and, optionally
444  * set a hash pointer.
445  */
446 void
447 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
448 {
449 	struct componentname *cnp = &ndp->ni_cnd;
450 
451 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
452 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
453 	if (hashpp != NULL)
454 		*hashpp = NULL;
455 	*bufpp = cnp->cn_pnbuf;
456 }
457 
458 /*
459  * Release the above path buffer, if not released by nfsvno_namei().
460  */
461 void
462 nfsvno_relpathbuf(struct nameidata *ndp)
463 {
464 
465 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
466 		panic("nfsrelpath");
467 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
468 	ndp->ni_cnd.cn_flags &= ~HASBUF;
469 }
470 
471 /*
472  * Readlink vnode op into an mbuf list.
473  */
474 int
475 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
476     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
477 {
478 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
479 	struct iovec *ivp = iv;
480 	struct uio io, *uiop = &io;
481 	struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
482 	int i, len, tlen, error;
483 
484 	len = 0;
485 	i = 0;
486 	while (len < NFS_MAXPATHLEN) {
487 		NFSMGET(mp);
488 		MCLGET(mp, M_WAIT);
489 		mp->m_len = NFSMSIZ(mp);
490 		if (len == 0) {
491 			mp3 = mp2 = mp;
492 		} else {
493 			mp2->m_next = mp;
494 			mp2 = mp;
495 		}
496 		if ((len + mp->m_len) > NFS_MAXPATHLEN) {
497 			mp->m_len = NFS_MAXPATHLEN - len;
498 			len = NFS_MAXPATHLEN;
499 		} else {
500 			len += mp->m_len;
501 		}
502 		ivp->iov_base = mtod(mp, caddr_t);
503 		ivp->iov_len = mp->m_len;
504 		i++;
505 		ivp++;
506 	}
507 	uiop->uio_iov = iv;
508 	uiop->uio_iovcnt = i;
509 	uiop->uio_offset = 0;
510 	uiop->uio_resid = len;
511 	uiop->uio_rw = UIO_READ;
512 	uiop->uio_segflg = UIO_SYSSPACE;
513 	uiop->uio_td = NULL;
514 	error = VOP_READLINK(vp, uiop, cred);
515 	if (error) {
516 		m_freem(mp3);
517 		*lenp = 0;
518 		return (error);
519 	}
520 	if (uiop->uio_resid > 0) {
521 		len -= uiop->uio_resid;
522 		tlen = NFSM_RNDUP(len);
523 		nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
524 	}
525 	*lenp = len;
526 	*mpp = mp3;
527 	*mpendp = mp;
528 	return (0);
529 }
530 
531 /*
532  * Read vnode op call into mbuf list.
533  */
534 int
535 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
536     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
537 {
538 	struct mbuf *m;
539 	int i;
540 	struct iovec *iv;
541 	struct iovec *iv2;
542 	int error = 0, len, left, siz, tlen, ioflag = 0, hi, try = 32;
543 	struct mbuf *m2 = NULL, *m3;
544 	struct uio io, *uiop = &io;
545 	struct nfsheur *nh;
546 
547 	/*
548 	 * Calculate seqcount for heuristic
549 	 */
550 	/*
551 	 * Locate best candidate
552 	 */
553 
554 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
555 	nh = &nfsheur[hi];
556 
557 	while (try--) {
558 		if (nfsheur[hi].nh_vp == vp) {
559 			nh = &nfsheur[hi];
560 			break;
561 		}
562 		if (nfsheur[hi].nh_use > 0)
563 			--nfsheur[hi].nh_use;
564 		hi = (hi + 1) % NUM_HEURISTIC;
565 		if (nfsheur[hi].nh_use < nh->nh_use)
566 			nh = &nfsheur[hi];
567 	}
568 
569 	if (nh->nh_vp != vp) {
570 		nh->nh_vp = vp;
571 		nh->nh_nextr = off;
572 		nh->nh_use = NHUSE_INIT;
573 		if (off == 0)
574 			nh->nh_seqcount = 4;
575 		else
576 			nh->nh_seqcount = 1;
577 	}
578 
579 	/*
580 	 * Calculate heuristic
581 	 */
582 
583 	if ((off == 0 && nh->nh_seqcount > 0) || off == nh->nh_nextr) {
584 		if (++nh->nh_seqcount > IO_SEQMAX)
585 			nh->nh_seqcount = IO_SEQMAX;
586 	} else if (nh->nh_seqcount > 1) {
587 		nh->nh_seqcount = 1;
588 	} else {
589 		nh->nh_seqcount = 0;
590 	}
591 	nh->nh_use += NHUSE_INC;
592 	if (nh->nh_use > NHUSE_MAX)
593 		nh->nh_use = NHUSE_MAX;
594 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
595 
596 	len = left = NFSM_RNDUP(cnt);
597 	m3 = NULL;
598 	/*
599 	 * Generate the mbuf list with the uio_iov ref. to it.
600 	 */
601 	i = 0;
602 	while (left > 0) {
603 		NFSMGET(m);
604 		MCLGET(m, M_WAIT);
605 		m->m_len = 0;
606 		siz = min(M_TRAILINGSPACE(m), left);
607 		left -= siz;
608 		i++;
609 		if (m3)
610 			m2->m_next = m;
611 		else
612 			m3 = m;
613 		m2 = m;
614 	}
615 	MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
616 	    M_TEMP, M_WAITOK);
617 	uiop->uio_iov = iv2 = iv;
618 	m = m3;
619 	left = len;
620 	i = 0;
621 	while (left > 0) {
622 		if (m == NULL)
623 			panic("nfsvno_read iov");
624 		siz = min(M_TRAILINGSPACE(m), left);
625 		if (siz > 0) {
626 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
627 			iv->iov_len = siz;
628 			m->m_len += siz;
629 			left -= siz;
630 			iv++;
631 			i++;
632 		}
633 		m = m->m_next;
634 	}
635 	uiop->uio_iovcnt = i;
636 	uiop->uio_offset = off;
637 	uiop->uio_resid = len;
638 	uiop->uio_rw = UIO_READ;
639 	uiop->uio_segflg = UIO_SYSSPACE;
640 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
641 	FREE((caddr_t)iv2, M_TEMP);
642 	if (error) {
643 		m_freem(m3);
644 		*mpp = NULL;
645 		return (error);
646 	}
647 	tlen = len - uiop->uio_resid;
648 	cnt = cnt < tlen ? cnt : tlen;
649 	tlen = NFSM_RNDUP(cnt);
650 	if (tlen == 0) {
651 		m_freem(m3);
652 		m3 = NULL;
653 	} else if (len != tlen || tlen != cnt)
654 		nfsrv_adj(m3, len - tlen, tlen - cnt);
655 	*mpp = m3;
656 	*mpendp = m2;
657 	return (0);
658 }
659 
660 /*
661  * Write vnode op from an mbuf list.
662  */
663 int
664 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
665     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
666 {
667 	struct iovec *ivp;
668 	int i, len;
669 	struct iovec *iv;
670 	int ioflags, error;
671 	struct uio io, *uiop = &io;
672 
673 	MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
674 	    M_WAITOK);
675 	uiop->uio_iov = iv = ivp;
676 	uiop->uio_iovcnt = cnt;
677 	i = mtod(mp, caddr_t) + mp->m_len - cp;
678 	len = retlen;
679 	while (len > 0) {
680 		if (mp == NULL)
681 			panic("nfsvno_write");
682 		if (i > 0) {
683 			i = min(i, len);
684 			ivp->iov_base = cp;
685 			ivp->iov_len = i;
686 			ivp++;
687 			len -= i;
688 		}
689 		mp = mp->m_next;
690 		if (mp) {
691 			i = mp->m_len;
692 			cp = mtod(mp, caddr_t);
693 		}
694 	}
695 
696 	if (stable == NFSWRITE_UNSTABLE)
697 		ioflags = IO_NODELOCKED;
698 	else
699 		ioflags = (IO_SYNC | IO_NODELOCKED);
700 	uiop->uio_resid = retlen;
701 	uiop->uio_rw = UIO_WRITE;
702 	uiop->uio_segflg = UIO_SYSSPACE;
703 	NFSUIOPROC(uiop, p);
704 	uiop->uio_offset = off;
705 	error = VOP_WRITE(vp, uiop, ioflags, cred);
706 	FREE((caddr_t)iv, M_TEMP);
707 	return (error);
708 }
709 
710 /*
711  * Common code for creating a regular file (plus special files for V2).
712  */
713 int
714 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
715     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
716     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
717 {
718 	u_quad_t tempsize;
719 	int error;
720 
721 	error = nd->nd_repstat;
722 	if (!error && ndp->ni_vp == NULL) {
723 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
724 			vrele(ndp->ni_startdir);
725 			error = VOP_CREATE(ndp->ni_dvp,
726 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
727 			vput(ndp->ni_dvp);
728 			nfsvno_relpathbuf(ndp);
729 			if (!error) {
730 				if (*exclusive_flagp) {
731 					*exclusive_flagp = 0;
732 					NFSVNO_ATTRINIT(nvap);
733 					nvap->na_atime.tv_sec = cverf[0];
734 					nvap->na_atime.tv_nsec = cverf[1];
735 					error = VOP_SETATTR(ndp->ni_vp,
736 					    &nvap->na_vattr, nd->nd_cred);
737 				}
738 			}
739 		/*
740 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
741 		 * (This implies, just get out on an error.)
742 		 */
743 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
744 			nvap->na_type == VFIFO) {
745 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
746 				nvap->na_type = VFIFO;
747                         if (nvap->na_type != VFIFO &&
748 			    (error = priv_check_cred(nd->nd_cred,
749 			     PRIV_VFS_MKNOD_DEV, 0))) {
750 				vrele(ndp->ni_startdir);
751 				nfsvno_relpathbuf(ndp);
752 				vput(ndp->ni_dvp);
753 				return (error);
754 			}
755 			nvap->na_rdev = rdev;
756 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
757 			    &ndp->ni_cnd, &nvap->na_vattr);
758 			vput(ndp->ni_dvp);
759 			nfsvno_relpathbuf(ndp);
760 			if (error) {
761 				vrele(ndp->ni_startdir);
762 				return (error);
763 			}
764 		} else {
765 			vrele(ndp->ni_startdir);
766 			nfsvno_relpathbuf(ndp);
767 			vput(ndp->ni_dvp);
768 			return (ENXIO);
769 		}
770 		*vpp = ndp->ni_vp;
771 	} else {
772 		/*
773 		 * Handle cases where error is already set and/or
774 		 * the file exists.
775 		 * 1 - clean up the lookup
776 		 * 2 - iff !error and na_size set, truncate it
777 		 */
778 		vrele(ndp->ni_startdir);
779 		nfsvno_relpathbuf(ndp);
780 		*vpp = ndp->ni_vp;
781 		if (ndp->ni_dvp == *vpp)
782 			vrele(ndp->ni_dvp);
783 		else
784 			vput(ndp->ni_dvp);
785 		if (!error && nvap->na_size != VNOVAL) {
786 			error = nfsvno_accchk(*vpp, VWRITE,
787 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
788 			    NFSACCCHK_VPISLOCKED, NULL);
789 			if (!error) {
790 				tempsize = nvap->na_size;
791 				NFSVNO_ATTRINIT(nvap);
792 				nvap->na_size = tempsize;
793 				error = VOP_SETATTR(*vpp,
794 				    &nvap->na_vattr, nd->nd_cred);
795 			}
796 		}
797 		if (error)
798 			vput(*vpp);
799 	}
800 	return (error);
801 }
802 
803 /*
804  * Do a mknod vnode op.
805  */
806 int
807 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
808     struct thread *p)
809 {
810 	int error = 0;
811 	enum vtype vtyp;
812 
813 	vtyp = nvap->na_type;
814 	/*
815 	 * Iff doesn't exist, create it.
816 	 */
817 	if (ndp->ni_vp) {
818 		vrele(ndp->ni_startdir);
819 		nfsvno_relpathbuf(ndp);
820 		vput(ndp->ni_dvp);
821 		vrele(ndp->ni_vp);
822 		return (EEXIST);
823 	}
824 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
825 		vrele(ndp->ni_startdir);
826 		nfsvno_relpathbuf(ndp);
827 		vput(ndp->ni_dvp);
828 		return (NFSERR_BADTYPE);
829 	}
830 	if (vtyp == VSOCK) {
831 		vrele(ndp->ni_startdir);
832 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
833 		    &ndp->ni_cnd, &nvap->na_vattr);
834 		vput(ndp->ni_dvp);
835 		nfsvno_relpathbuf(ndp);
836 	} else {
837 		if (nvap->na_type != VFIFO &&
838 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
839 			vrele(ndp->ni_startdir);
840 			nfsvno_relpathbuf(ndp);
841 			vput(ndp->ni_dvp);
842 			return (error);
843 		}
844 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
845 		    &ndp->ni_cnd, &nvap->na_vattr);
846 		vput(ndp->ni_dvp);
847 		nfsvno_relpathbuf(ndp);
848 		if (error)
849 			vrele(ndp->ni_startdir);
850 		/*
851 		 * Since VOP_MKNOD returns the ni_vp, I can't
852 		 * see any reason to do the lookup.
853 		 */
854 	}
855 	return (error);
856 }
857 
858 /*
859  * Mkdir vnode op.
860  */
861 int
862 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
863     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
864 {
865 	int error = 0;
866 
867 	if (ndp->ni_vp != NULL) {
868 		if (ndp->ni_dvp == ndp->ni_vp)
869 			vrele(ndp->ni_dvp);
870 		else
871 			vput(ndp->ni_dvp);
872 		vrele(ndp->ni_vp);
873 		nfsvno_relpathbuf(ndp);
874 		return (EEXIST);
875 	}
876 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
877 	    &nvap->na_vattr);
878 	vput(ndp->ni_dvp);
879 	nfsvno_relpathbuf(ndp);
880 	return (error);
881 }
882 
883 /*
884  * symlink vnode op.
885  */
886 int
887 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
888     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
889     struct nfsexstuff *exp)
890 {
891 	int error = 0;
892 
893 	if (ndp->ni_vp) {
894 		vrele(ndp->ni_startdir);
895 		nfsvno_relpathbuf(ndp);
896 		if (ndp->ni_dvp == ndp->ni_vp)
897 			vrele(ndp->ni_dvp);
898 		else
899 			vput(ndp->ni_dvp);
900 		vrele(ndp->ni_vp);
901 		return (EEXIST);
902 	}
903 
904 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
905 	    &nvap->na_vattr, pathcp);
906 	vput(ndp->ni_dvp);
907 	vrele(ndp->ni_startdir);
908 	nfsvno_relpathbuf(ndp);
909 	/*
910 	 * Although FreeBSD still had the lookup code in
911 	 * it for 7/current, there doesn't seem to be any
912 	 * point, since VOP_SYMLINK() returns the ni_vp.
913 	 * Just vput it for v2.
914 	 */
915 	if (!not_v2 && !error)
916 		vput(ndp->ni_vp);
917 	return (error);
918 }
919 
920 /*
921  * Parse symbolic link arguments.
922  * This function has an ugly side effect. It will MALLOC() an area for
923  * the symlink and set iov_base to point to it, only if it succeeds.
924  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
925  * be FREE'd later.
926  */
927 int
928 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
929     struct thread *p, char **pathcpp, int *lenp)
930 {
931 	u_int32_t *tl;
932 	char *pathcp = NULL;
933 	int error = 0, len;
934 	struct nfsv2_sattr *sp;
935 
936 	*pathcpp = NULL;
937 	*lenp = 0;
938 	if ((nd->nd_flag & ND_NFSV3) &&
939 	    (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
940 		goto nfsmout;
941 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
942 	len = fxdr_unsigned(int, *tl);
943 	if (len > NFS_MAXPATHLEN || len <= 0) {
944 		error = EBADRPC;
945 		goto nfsmout;
946 	}
947 	MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
948 	error = nfsrv_mtostr(nd, pathcp, len);
949 	if (error)
950 		goto nfsmout;
951 	if (nd->nd_flag & ND_NFSV2) {
952 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
953 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
954 	}
955 	*pathcpp = pathcp;
956 	*lenp = len;
957 	return (0);
958 nfsmout:
959 	if (pathcp)
960 		free(pathcp, M_TEMP);
961 	return (error);
962 }
963 
964 /*
965  * Remove a non-directory object.
966  */
967 int
968 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
969     struct thread *p, struct nfsexstuff *exp)
970 {
971 	struct vnode *vp;
972 	int error = 0;
973 
974 	vp = ndp->ni_vp;
975 	if (vp->v_type == VDIR)
976 		error = NFSERR_ISDIR;
977 	else if (is_v4)
978 		error = nfsrv_checkremove(vp, 1, p);
979 	if (!error)
980 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
981 	if (ndp->ni_dvp == vp)
982 		vrele(ndp->ni_dvp);
983 	else
984 		vput(ndp->ni_dvp);
985 	vput(vp);
986 	return (error);
987 }
988 
989 /*
990  * Remove a directory.
991  */
992 int
993 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
994     struct thread *p, struct nfsexstuff *exp)
995 {
996 	struct vnode *vp;
997 	int error = 0;
998 
999 	vp = ndp->ni_vp;
1000 	if (vp->v_type != VDIR) {
1001 		error = ENOTDIR;
1002 		goto out;
1003 	}
1004 	/*
1005 	 * No rmdir "." please.
1006 	 */
1007 	if (ndp->ni_dvp == vp) {
1008 		error = EINVAL;
1009 		goto out;
1010 	}
1011 	/*
1012 	 * The root of a mounted filesystem cannot be deleted.
1013 	 */
1014 	if (vp->v_vflag & VV_ROOT)
1015 		error = EBUSY;
1016 out:
1017 	if (!error)
1018 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1019 	if (ndp->ni_dvp == vp)
1020 		vrele(ndp->ni_dvp);
1021 	else
1022 		vput(ndp->ni_dvp);
1023 	vput(vp);
1024 	return (error);
1025 }
1026 
1027 /*
1028  * Rename vnode op.
1029  */
1030 int
1031 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1032     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1033 {
1034 	struct vnode *fvp, *tvp, *tdvp;
1035 	int error = 0;
1036 
1037 	fvp = fromndp->ni_vp;
1038 	if (ndstat) {
1039 		vrele(fromndp->ni_dvp);
1040 		vrele(fvp);
1041 		error = ndstat;
1042 		goto out1;
1043 	}
1044 	tdvp = tondp->ni_dvp;
1045 	tvp = tondp->ni_vp;
1046 	if (tvp != NULL) {
1047 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1048 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1049 			goto out;
1050 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1051 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1052 			goto out;
1053 		}
1054 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1055 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1056 			goto out;
1057 		}
1058 
1059 		/*
1060 		 * A rename to '.' or '..' results in a prematurely
1061 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1062 		 * here.
1063 		 */
1064 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1065 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1066 		    (tondp->ni_cnd.cn_namelen == 2 &&
1067 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1068 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1069 			error = EINVAL;
1070 			goto out;
1071 		}
1072 	}
1073 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1074 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1075 		goto out;
1076 	}
1077 	if (fvp->v_mount != tdvp->v_mount) {
1078 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1079 		goto out;
1080 	}
1081 	if (fvp == tdvp) {
1082 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1083 		goto out;
1084 	}
1085 	if (fvp == tvp) {
1086 		/*
1087 		 * If source and destination are the same, there is nothing to
1088 		 * do. Set error to -1 to indicate this.
1089 		 */
1090 		error = -1;
1091 		goto out;
1092 	}
1093 	if (ndflag & ND_NFSV4) {
1094 		NFSVOPLOCK(fvp, LK_EXCLUSIVE | LK_RETRY, p);
1095 		error = nfsrv_checkremove(fvp, 0, p);
1096 		NFSVOPUNLOCK(fvp, 0, p);
1097 		if (tvp && !error)
1098 			error = nfsrv_checkremove(tvp, 1, p);
1099 	} else {
1100 		/*
1101 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1102 		 * that the NFSv4 client won't be confused by the rename.
1103 		 * Since nfsd_recalldelegation() can only be called on an
1104 		 * unlocked vnode at this point and fvp is the file that will
1105 		 * still exist after the rename, just do fvp.
1106 		 */
1107 		nfsd_recalldelegation(fvp, p);
1108 	}
1109 out:
1110 	if (!error) {
1111 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1112 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1113 		    &tondp->ni_cnd);
1114 	} else {
1115 		if (tdvp == tvp)
1116 			vrele(tdvp);
1117 		else
1118 			vput(tdvp);
1119 		if (tvp)
1120 			vput(tvp);
1121 		vrele(fromndp->ni_dvp);
1122 		vrele(fvp);
1123 		if (error == -1)
1124 			error = 0;
1125 	}
1126 	vrele(tondp->ni_startdir);
1127 	nfsvno_relpathbuf(tondp);
1128 out1:
1129 	vrele(fromndp->ni_startdir);
1130 	nfsvno_relpathbuf(fromndp);
1131 	return (error);
1132 }
1133 
1134 /*
1135  * Link vnode op.
1136  */
1137 int
1138 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1139     struct thread *p, struct nfsexstuff *exp)
1140 {
1141 	struct vnode *xp;
1142 	int error = 0;
1143 
1144 	xp = ndp->ni_vp;
1145 	if (xp != NULL) {
1146 		error = EEXIST;
1147 	} else {
1148 		xp = ndp->ni_dvp;
1149 		if (vp->v_mount != xp->v_mount)
1150 			error = EXDEV;
1151 	}
1152 	if (!error) {
1153 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
1154 		error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1155 		if (ndp->ni_dvp == vp)
1156 			vrele(ndp->ni_dvp);
1157 		else
1158 			vput(ndp->ni_dvp);
1159 		NFSVOPUNLOCK(vp, 0, p);
1160 	} else {
1161 		if (ndp->ni_dvp == ndp->ni_vp)
1162 			vrele(ndp->ni_dvp);
1163 		else
1164 			vput(ndp->ni_dvp);
1165 		if (ndp->ni_vp)
1166 			vrele(ndp->ni_vp);
1167 	}
1168 	nfsvno_relpathbuf(ndp);
1169 	return (error);
1170 }
1171 
1172 /*
1173  * Do the fsync() appropriate for the commit.
1174  */
1175 int
1176 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1177     struct thread *td)
1178 {
1179 	int error = 0;
1180 
1181 	if (cnt > MAX_COMMIT_COUNT) {
1182 		/*
1183 		 * Give up and do the whole thing
1184 		 */
1185 		if (vp->v_object &&
1186 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1187 			VM_OBJECT_LOCK(vp->v_object);
1188 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1189 			VM_OBJECT_UNLOCK(vp->v_object);
1190 		}
1191 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1192 	} else {
1193 		/*
1194 		 * Locate and synchronously write any buffers that fall
1195 		 * into the requested range.  Note:  we are assuming that
1196 		 * f_iosize is a power of 2.
1197 		 */
1198 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1199 		int iomask = iosize - 1;
1200 		struct bufobj *bo;
1201 		daddr_t lblkno;
1202 
1203 		/*
1204 		 * Align to iosize boundry, super-align to page boundry.
1205 		 */
1206 		if (off & iomask) {
1207 			cnt += off & iomask;
1208 			off &= ~(u_quad_t)iomask;
1209 		}
1210 		if (off & PAGE_MASK) {
1211 			cnt += off & PAGE_MASK;
1212 			off &= ~(u_quad_t)PAGE_MASK;
1213 		}
1214 		lblkno = off / iosize;
1215 
1216 		if (vp->v_object &&
1217 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1218 			VM_OBJECT_LOCK(vp->v_object);
1219 			vm_object_page_clean(vp->v_object, off / PAGE_SIZE, (cnt + PAGE_MASK) / PAGE_SIZE, OBJPC_SYNC);
1220 			VM_OBJECT_UNLOCK(vp->v_object);
1221 		}
1222 
1223 		bo = &vp->v_bufobj;
1224 		BO_LOCK(bo);
1225 		while (cnt > 0) {
1226 			struct buf *bp;
1227 
1228 			/*
1229 			 * If we have a buffer and it is marked B_DELWRI we
1230 			 * have to lock and write it.  Otherwise the prior
1231 			 * write is assumed to have already been committed.
1232 			 *
1233 			 * gbincore() can return invalid buffers now so we
1234 			 * have to check that bit as well (though B_DELWRI
1235 			 * should not be set if B_INVAL is set there could be
1236 			 * a race here since we haven't locked the buffer).
1237 			 */
1238 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1239 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1240 				    LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
1241 					BO_LOCK(bo);
1242 					continue; /* retry */
1243 				}
1244 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1245 				    B_DELWRI) {
1246 					bremfree(bp);
1247 					bp->b_flags &= ~B_ASYNC;
1248 					bwrite(bp);
1249 					++nfs_commit_miss;
1250 				} else
1251 					BUF_UNLOCK(bp);
1252 				BO_LOCK(bo);
1253 			}
1254 			++nfs_commit_blks;
1255 			if (cnt < iosize)
1256 				break;
1257 			cnt -= iosize;
1258 			++lblkno;
1259 		}
1260 		BO_UNLOCK(bo);
1261 	}
1262 	return (error);
1263 }
1264 
1265 /*
1266  * Statfs vnode op.
1267  */
1268 int
1269 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1270 {
1271 
1272 	return (VFS_STATFS(vp->v_mount, sf));
1273 }
1274 
1275 /*
1276  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1277  * must handle nfsrv_opencheck() calls after any other access checks.
1278  */
1279 void
1280 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1281     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1282     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1283     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1284     struct nfsexstuff *exp, struct vnode **vpp)
1285 {
1286 	struct vnode *vp = NULL;
1287 	u_quad_t tempsize;
1288 	struct nfsexstuff nes;
1289 
1290 	if (ndp->ni_vp == NULL)
1291 		nd->nd_repstat = nfsrv_opencheck(clientid,
1292 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1293 	if (!nd->nd_repstat) {
1294 		if (ndp->ni_vp == NULL) {
1295 			vrele(ndp->ni_startdir);
1296 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1297 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1298 			vput(ndp->ni_dvp);
1299 			nfsvno_relpathbuf(ndp);
1300 			if (!nd->nd_repstat) {
1301 				if (*exclusive_flagp) {
1302 					*exclusive_flagp = 0;
1303 					NFSVNO_ATTRINIT(nvap);
1304 					nvap->na_atime.tv_sec = cverf[0];
1305 					nvap->na_atime.tv_nsec = cverf[1];
1306 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1307 					    &nvap->na_vattr, cred);
1308 				} else {
1309 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1310 					    aclp, p, attrbitp, exp);
1311 				}
1312 			}
1313 			vp = ndp->ni_vp;
1314 		} else {
1315 			if (ndp->ni_startdir)
1316 				vrele(ndp->ni_startdir);
1317 			nfsvno_relpathbuf(ndp);
1318 			vp = ndp->ni_vp;
1319 			if (create == NFSV4OPEN_CREATE) {
1320 				if (ndp->ni_dvp == vp)
1321 					vrele(ndp->ni_dvp);
1322 				else
1323 					vput(ndp->ni_dvp);
1324 			}
1325 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1326 				if (ndp->ni_cnd.cn_flags & RDONLY)
1327 					NFSVNO_SETEXRDONLY(&nes);
1328 				else
1329 					NFSVNO_EXINIT(&nes);
1330 				nd->nd_repstat = nfsvno_accchk(vp,
1331 				    VWRITE, cred, &nes, p,
1332 				    NFSACCCHK_NOOVERRIDE,
1333 				    NFSACCCHK_VPISLOCKED, NULL);
1334 				nd->nd_repstat = nfsrv_opencheck(clientid,
1335 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1336 				if (!nd->nd_repstat) {
1337 					tempsize = nvap->na_size;
1338 					NFSVNO_ATTRINIT(nvap);
1339 					nvap->na_size = tempsize;
1340 					nd->nd_repstat = VOP_SETATTR(vp,
1341 					    &nvap->na_vattr, cred);
1342 				}
1343 			} else if (vp->v_type == VREG) {
1344 				nd->nd_repstat = nfsrv_opencheck(clientid,
1345 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1346 			}
1347 		}
1348 	} else {
1349 		if (ndp->ni_cnd.cn_flags & HASBUF)
1350 			nfsvno_relpathbuf(ndp);
1351 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1352 			vrele(ndp->ni_startdir);
1353 			if (ndp->ni_dvp == ndp->ni_vp)
1354 				vrele(ndp->ni_dvp);
1355 			else
1356 				vput(ndp->ni_dvp);
1357 			if (ndp->ni_vp)
1358 				vput(ndp->ni_vp);
1359 		}
1360 	}
1361 	*vpp = vp;
1362 }
1363 
1364 /*
1365  * Updates the file rev and sets the mtime and ctime
1366  * to the current clock time, returning the va_filerev and va_Xtime
1367  * values.
1368  */
1369 void
1370 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1371     struct ucred *cred, struct thread *p)
1372 {
1373 	struct vattr va;
1374 
1375 	VATTR_NULL(&va);
1376 	getnanotime(&va.va_mtime);
1377 	(void) VOP_SETATTR(vp, &va, cred);
1378 	(void) nfsvno_getattr(vp, nvap, cred, p);
1379 }
1380 
1381 /*
1382  * Glue routine to nfsv4_fillattr().
1383  */
1384 int
1385 nfsvno_fillattr(struct nfsrv_descript *nd, struct vnode *vp,
1386     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1387     struct ucred *cred, struct thread *p, int isdgram, int reterr)
1388 {
1389 	int error;
1390 
1391 	error = nfsv4_fillattr(nd, vp, NULL, &nvap->na_vattr, fhp, rderror,
1392 	    attrbitp, cred, p, isdgram, reterr);
1393 	return (error);
1394 }
1395 
1396 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1397 /*
1398  * nfs readdir service
1399  * - mallocs what it thinks is enough to read
1400  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1401  * - calls VOP_READDIR()
1402  * - loops around building the reply
1403  *	if the output generated exceeds count break out of loop
1404  *	The NFSM_CLGET macro is used here so that the reply will be packed
1405  *	tightly in mbuf clusters.
1406  * - it trims out records with d_fileno == 0
1407  *	this doesn't matter for Unix clients, but they might confuse clients
1408  *	for other os'.
1409  * - it trims out records with d_type == DT_WHT
1410  *	these cannot be seen through NFS (unless we extend the protocol)
1411  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1412  * PS: The NFS protocol spec. does not clarify what the "count" byte
1413  *	argument is a count of.. just name strings and file id's or the
1414  *	entire reply rpc or ...
1415  *	I tried just file name and id sizes and it confused the Sun client,
1416  *	so I am using the full rpc size now. The "paranoia.." comment refers
1417  *	to including the status longwords that are not a part of the dir.
1418  *	"entry" structures, but are in the rpc.
1419  */
1420 int
1421 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1422     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1423 {
1424 	struct dirent *dp;
1425 	u_int32_t *tl;
1426 	int dirlen;
1427 	char *cpos, *cend, *rbuf;
1428 	struct nfsvattr at;
1429 	int nlen, error = 0, getret = 1;
1430 	int siz, cnt, fullsiz, eofflag, ncookies;
1431 	u_int64_t off, toff, verf;
1432 	u_long *cookies = NULL, *cookiep;
1433 	struct uio io;
1434 	struct iovec iv;
1435 
1436 	if (nd->nd_repstat) {
1437 		nfsrv_postopattr(nd, getret, &at);
1438 		return (0);
1439 	}
1440 	if (nd->nd_flag & ND_NFSV2) {
1441 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1442 		off = fxdr_unsigned(u_quad_t, *tl++);
1443 	} else {
1444 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1445 		off = fxdr_hyper(tl);
1446 		tl += 2;
1447 		verf = fxdr_hyper(tl);
1448 		tl += 2;
1449 	}
1450 	toff = off;
1451 	cnt = fxdr_unsigned(int, *tl);
1452 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1453 		cnt = NFS_SRVMAXDATA(nd);
1454 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1455 	fullsiz = siz;
1456 	if (nd->nd_flag & ND_NFSV3) {
1457 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1458 		    p);
1459 #if 0
1460 		/*
1461 		 * va_filerev is not sufficient as a cookie verifier,
1462 		 * since it is not supposed to change when entries are
1463 		 * removed/added unless that offset cookies returned to
1464 		 * the client are no longer valid.
1465 		 */
1466 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
1467 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1468 #endif
1469 	}
1470 	if (nd->nd_repstat == 0 && cnt == 0) {
1471 		if (nd->nd_flag & ND_NFSV2)
1472 			/* NFSv2 does not have NFSERR_TOOSMALL */
1473 			nd->nd_repstat = EPERM;
1474 		else
1475 			nd->nd_repstat = NFSERR_TOOSMALL;
1476 	}
1477 	if (!nd->nd_repstat)
1478 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1479 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1480 		    NFSACCCHK_VPISLOCKED, NULL);
1481 	if (nd->nd_repstat) {
1482 		vput(vp);
1483 		if (nd->nd_flag & ND_NFSV3)
1484 			nfsrv_postopattr(nd, getret, &at);
1485 		return (0);
1486 	}
1487 	NFSVOPUNLOCK(vp, 0, p);
1488 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1489 again:
1490 	eofflag = 0;
1491 	if (cookies) {
1492 		free((caddr_t)cookies, M_TEMP);
1493 		cookies = NULL;
1494 	}
1495 
1496 	iv.iov_base = rbuf;
1497 	iv.iov_len = siz;
1498 	io.uio_iov = &iv;
1499 	io.uio_iovcnt = 1;
1500 	io.uio_offset = (off_t)off;
1501 	io.uio_resid = siz;
1502 	io.uio_segflg = UIO_SYSSPACE;
1503 	io.uio_rw = UIO_READ;
1504 	io.uio_td = NULL;
1505 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
1506 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1507 	    &cookies);
1508 	NFSVOPUNLOCK(vp, 0, p);
1509 	off = (u_int64_t)io.uio_offset;
1510 	if (io.uio_resid)
1511 		siz -= io.uio_resid;
1512 
1513 	if (!cookies && !nd->nd_repstat)
1514 		nd->nd_repstat = NFSERR_PERM;
1515 	if (nd->nd_flag & ND_NFSV3) {
1516 		getret = nfsvno_getattr(vp, &at, nd->nd_cred, p);
1517 		if (!nd->nd_repstat)
1518 			nd->nd_repstat = getret;
1519 	}
1520 
1521 	/*
1522 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
1523 	 */
1524 	if (nd->nd_repstat) {
1525 		vrele(vp);
1526 		free((caddr_t)rbuf, M_TEMP);
1527 		if (cookies)
1528 			free((caddr_t)cookies, M_TEMP);
1529 		if (nd->nd_flag & ND_NFSV3)
1530 			nfsrv_postopattr(nd, getret, &at);
1531 		return (0);
1532 	}
1533 	/*
1534 	 * If nothing read, return eof
1535 	 * rpc reply
1536 	 */
1537 	if (siz == 0) {
1538 		vrele(vp);
1539 		if (nd->nd_flag & ND_NFSV2) {
1540 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1541 		} else {
1542 			nfsrv_postopattr(nd, getret, &at);
1543 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1544 			txdr_hyper(at.na_filerev, tl);
1545 			tl += 2;
1546 		}
1547 		*tl++ = newnfs_false;
1548 		*tl = newnfs_true;
1549 		FREE((caddr_t)rbuf, M_TEMP);
1550 		FREE((caddr_t)cookies, M_TEMP);
1551 		return (0);
1552 	}
1553 
1554 	/*
1555 	 * Check for degenerate cases of nothing useful read.
1556 	 * If so go try again
1557 	 */
1558 	cpos = rbuf;
1559 	cend = rbuf + siz;
1560 	dp = (struct dirent *)cpos;
1561 	cookiep = cookies;
1562 
1563 	/*
1564 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1565 	 * directory offset up to a block boundary, so it is necessary to
1566 	 * skip over the records that precede the requested offset. This
1567 	 * requires the assumption that file offset cookies monotonically
1568 	 * increase.
1569 	 */
1570 	while (cpos < cend && ncookies > 0 &&
1571 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1572 	     ((u_quad_t)(*cookiep)) <= toff)) {
1573 		cpos += dp->d_reclen;
1574 		dp = (struct dirent *)cpos;
1575 		cookiep++;
1576 		ncookies--;
1577 	}
1578 	if (cpos >= cend || ncookies == 0) {
1579 		siz = fullsiz;
1580 		toff = off;
1581 		goto again;
1582 	}
1583 
1584 	/*
1585 	 * dirlen is the size of the reply, including all XDR and must
1586 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1587 	 * if the XDR should be included in "count", but to be safe, we do.
1588 	 * (Include the two booleans at the end of the reply in dirlen now.)
1589 	 */
1590 	if (nd->nd_flag & ND_NFSV3) {
1591 		nfsrv_postopattr(nd, getret, &at);
1592 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1593 		txdr_hyper(at.na_filerev, tl);
1594 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1595 	} else {
1596 		dirlen = 2 * NFSX_UNSIGNED;
1597 	}
1598 
1599 	/* Loop through the records and build reply */
1600 	while (cpos < cend && ncookies > 0) {
1601 		nlen = dp->d_namlen;
1602 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1603 			nlen <= NFS_MAXNAMLEN) {
1604 			if (nd->nd_flag & ND_NFSV3)
1605 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1606 			else
1607 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1608 			if (dirlen > cnt) {
1609 				eofflag = 0;
1610 				break;
1611 			}
1612 
1613 			/*
1614 			 * Build the directory record xdr from
1615 			 * the dirent entry.
1616 			 */
1617 			if (nd->nd_flag & ND_NFSV3) {
1618 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1619 				*tl++ = newnfs_true;
1620 				*tl++ = 0;
1621 			} else {
1622 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1623 				*tl++ = newnfs_true;
1624 			}
1625 			*tl = txdr_unsigned(dp->d_fileno);
1626 			(void) nfsm_strtom(nd, dp->d_name, nlen);
1627 			if (nd->nd_flag & ND_NFSV3) {
1628 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1629 				*tl++ = 0;
1630 			} else
1631 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1632 			*tl = txdr_unsigned(*cookiep);
1633 		}
1634 		cpos += dp->d_reclen;
1635 		dp = (struct dirent *)cpos;
1636 		cookiep++;
1637 		ncookies--;
1638 	}
1639 	if (cpos < cend)
1640 		eofflag = 0;
1641 	vrele(vp);
1642 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1643 	*tl++ = newnfs_false;
1644 	if (eofflag)
1645 		*tl = newnfs_true;
1646 	else
1647 		*tl = newnfs_false;
1648 	FREE((caddr_t)rbuf, M_TEMP);
1649 	FREE((caddr_t)cookies, M_TEMP);
1650 	return (0);
1651 nfsmout:
1652 	vput(vp);
1653 	return (error);
1654 }
1655 
1656 /*
1657  * Readdirplus for V3 and Readdir for V4.
1658  */
1659 int
1660 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1661     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1662 {
1663 	struct dirent *dp;
1664 	u_int32_t *tl;
1665 	int dirlen;
1666 	char *cpos, *cend, *rbuf;
1667 	struct vnode *nvp;
1668 	fhandle_t nfh;
1669 	struct nfsvattr nva, at, *nvap = &nva;
1670 	struct mbuf *mb0, *mb1;
1671 	struct nfsreferral *refp;
1672 	int nlen, r, error = 0, getret = 1, usevget = 1;
1673 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1674 	caddr_t bpos0, bpos1;
1675 	u_int64_t off, toff, verf;
1676 	u_long *cookies = NULL, *cookiep;
1677 	nfsattrbit_t attrbits, rderrbits, savbits;
1678 	struct uio io;
1679 	struct iovec iv;
1680 	struct componentname cn;
1681 
1682 	if (nd->nd_repstat) {
1683 		nfsrv_postopattr(nd, getret, &at);
1684 		return (0);
1685 	}
1686 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1687 	off = fxdr_hyper(tl);
1688 	toff = off;
1689 	tl += 2;
1690 	verf = fxdr_hyper(tl);
1691 	tl += 2;
1692 	siz = fxdr_unsigned(int, *tl++);
1693 	cnt = fxdr_unsigned(int, *tl);
1694 
1695 	/*
1696 	 * Use the server's maximum data transfer size as the upper bound
1697 	 * on reply datalen.
1698 	 */
1699 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1700 		cnt = NFS_SRVMAXDATA(nd);
1701 
1702 	/*
1703 	 * siz is a "hint" of how much directory information (name, fileid,
1704 	 * cookie) should be in the reply. At least one client "hints" 0,
1705 	 * so I set it to cnt for that case. I also round it up to the
1706 	 * next multiple of DIRBLKSIZ.
1707 	 */
1708 	if (siz <= 0)
1709 		siz = cnt;
1710 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1711 
1712 	if (nd->nd_flag & ND_NFSV4) {
1713 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1714 		if (error)
1715 			goto nfsmout;
1716 		NFSSET_ATTRBIT(&savbits, &attrbits);
1717 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1718 		NFSZERO_ATTRBIT(&rderrbits);
1719 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1720 	} else {
1721 		NFSZERO_ATTRBIT(&attrbits);
1722 	}
1723 	fullsiz = siz;
1724 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p);
1725 	if (!nd->nd_repstat) {
1726 	    if (off && verf != at.na_filerev) {
1727 		/*
1728 		 * va_filerev is not sufficient as a cookie verifier,
1729 		 * since it is not supposed to change when entries are
1730 		 * removed/added unless that offset cookies returned to
1731 		 * the client are no longer valid.
1732 		 */
1733 #if 0
1734 		if (nd->nd_flag & ND_NFSV4) {
1735 			nd->nd_repstat = NFSERR_NOTSAME;
1736 		} else {
1737 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1738 		}
1739 #endif
1740 	    } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1741 		nd->nd_repstat = NFSERR_BAD_COOKIE;
1742 	    }
1743 	}
1744 	if (!nd->nd_repstat && vp->v_type != VDIR)
1745 		nd->nd_repstat = NFSERR_NOTDIR;
1746 	if (!nd->nd_repstat && cnt == 0)
1747 		nd->nd_repstat = NFSERR_TOOSMALL;
1748 	if (!nd->nd_repstat)
1749 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1750 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1751 		    NFSACCCHK_VPISLOCKED, NULL);
1752 	if (nd->nd_repstat) {
1753 		vput(vp);
1754 		if (nd->nd_flag & ND_NFSV3)
1755 			nfsrv_postopattr(nd, getret, &at);
1756 		return (0);
1757 	}
1758 
1759 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1760 again:
1761 	eofflag = 0;
1762 	if (cookies) {
1763 		free((caddr_t)cookies, M_TEMP);
1764 		cookies = NULL;
1765 	}
1766 
1767 	iv.iov_base = rbuf;
1768 	iv.iov_len = siz;
1769 	io.uio_iov = &iv;
1770 	io.uio_iovcnt = 1;
1771 	io.uio_offset = (off_t)off;
1772 	io.uio_resid = siz;
1773 	io.uio_segflg = UIO_SYSSPACE;
1774 	io.uio_rw = UIO_READ;
1775 	io.uio_td = NULL;
1776 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1777 	    &cookies);
1778 	off = (u_int64_t)io.uio_offset;
1779 	if (io.uio_resid)
1780 		siz -= io.uio_resid;
1781 
1782 	getret = nfsvno_getattr(vp, &at, nd->nd_cred, p);
1783 
1784 	if (!cookies && !nd->nd_repstat)
1785 		nd->nd_repstat = NFSERR_PERM;
1786 	if (!nd->nd_repstat)
1787 		nd->nd_repstat = getret;
1788 	if (nd->nd_repstat) {
1789 		vput(vp);
1790 		if (cookies)
1791 			free((caddr_t)cookies, M_TEMP);
1792 		free((caddr_t)rbuf, M_TEMP);
1793 		if (nd->nd_flag & ND_NFSV3)
1794 			nfsrv_postopattr(nd, getret, &at);
1795 		return (0);
1796 	}
1797 	/*
1798 	 * If nothing read, return eof
1799 	 * rpc reply
1800 	 */
1801 	if (siz == 0) {
1802 		vput(vp);
1803 		if (nd->nd_flag & ND_NFSV3)
1804 			nfsrv_postopattr(nd, getret, &at);
1805 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1806 		txdr_hyper(at.na_filerev, tl);
1807 		tl += 2;
1808 		*tl++ = newnfs_false;
1809 		*tl = newnfs_true;
1810 		free((caddr_t)cookies, M_TEMP);
1811 		free((caddr_t)rbuf, M_TEMP);
1812 		return (0);
1813 	}
1814 
1815 	/*
1816 	 * Check for degenerate cases of nothing useful read.
1817 	 * If so go try again
1818 	 */
1819 	cpos = rbuf;
1820 	cend = rbuf + siz;
1821 	dp = (struct dirent *)cpos;
1822 	cookiep = cookies;
1823 
1824 	/*
1825 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1826 	 * directory offset up to a block boundary, so it is necessary to
1827 	 * skip over the records that precede the requested offset. This
1828 	 * requires the assumption that file offset cookies monotonically
1829 	 * increase.
1830 	 */
1831 	while (cpos < cend && ncookies > 0 &&
1832 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1833 	   ((u_quad_t)(*cookiep)) <= toff ||
1834 	   ((nd->nd_flag & ND_NFSV4) &&
1835 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1836 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1837 		cpos += dp->d_reclen;
1838 		dp = (struct dirent *)cpos;
1839 		cookiep++;
1840 		ncookies--;
1841 	}
1842 	if (cpos >= cend || ncookies == 0) {
1843 		siz = fullsiz;
1844 		toff = off;
1845 		goto again;
1846 	}
1847 	NFSVOPUNLOCK(vp, 0, p);
1848 
1849 	/*
1850 	 * Save this position, in case there is an error before one entry
1851 	 * is created.
1852 	 */
1853 	mb0 = nd->nd_mb;
1854 	bpos0 = nd->nd_bpos;
1855 
1856 	/*
1857 	 * Fill in the first part of the reply.
1858 	 * dirlen is the reply length in bytes and cannot exceed cnt.
1859 	 * (Include the two booleans at the end of the reply in dirlen now,
1860 	 *  so we recognize when we have exceeded cnt.)
1861 	 */
1862 	if (nd->nd_flag & ND_NFSV3) {
1863 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1864 		nfsrv_postopattr(nd, getret, &at);
1865 	} else {
1866 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
1867 	}
1868 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
1869 	txdr_hyper(at.na_filerev, tl);
1870 
1871 	/*
1872 	 * Save this position, in case there is an empty reply needed.
1873 	 */
1874 	mb1 = nd->nd_mb;
1875 	bpos1 = nd->nd_bpos;
1876 
1877 	/* Loop through the records and build reply */
1878 	entrycnt = 0;
1879 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
1880 		nlen = dp->d_namlen;
1881 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1882 		    nlen <= NFS_MAXNAMLEN &&
1883 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
1884 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
1885 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
1886 			/*
1887 			 * Save the current position in the reply, in case
1888 			 * this entry exceeds cnt.
1889 			 */
1890 			mb1 = nd->nd_mb;
1891 			bpos1 = nd->nd_bpos;
1892 
1893 			/*
1894 			 * For readdir_and_lookup get the vnode using
1895 			 * the file number.
1896 			 */
1897 			nvp = NULL;
1898 			refp = NULL;
1899 			r = 0;
1900 			if ((nd->nd_flag & ND_NFSV3) ||
1901 			    NFSNONZERO_ATTRBIT(&savbits)) {
1902 				if (nd->nd_flag & ND_NFSV4)
1903 					refp = nfsv4root_getreferral(NULL,
1904 					    vp, dp->d_fileno);
1905 				if (refp == NULL) {
1906 					if (usevget)
1907 						r = VFS_VGET(vp->v_mount,
1908 						    dp->d_fileno, LK_EXCLUSIVE,
1909 						    &nvp);
1910 					else
1911 						r = EOPNOTSUPP;
1912 					if (r == EOPNOTSUPP) {
1913 						if (usevget) {
1914 							usevget = 0;
1915 							cn.cn_nameiop = LOOKUP;
1916 							cn.cn_lkflags =
1917 							    LK_EXCLUSIVE |
1918 							    LK_RETRY;
1919 							cn.cn_cred =
1920 							    nd->nd_cred;
1921 							cn.cn_thread = p;
1922 						}
1923 						cn.cn_nameptr = dp->d_name;
1924 						cn.cn_namelen = nlen;
1925 						cn.cn_flags = ISLASTCN |
1926 						    NOFOLLOW | LOCKLEAF |
1927 						    MPSAFE;
1928 						if (nlen == 2 &&
1929 						    dp->d_name[0] == '.' &&
1930 						    dp->d_name[1] == '.')
1931 							cn.cn_flags |=
1932 							    ISDOTDOT;
1933 						if (!VOP_ISLOCKED(vp))
1934 							vn_lock(vp,
1935 							    LK_EXCLUSIVE |
1936 							    LK_RETRY);
1937 						if ((vp->v_vflag & VV_ROOT) != 0
1938 						    && (cn.cn_flags & ISDOTDOT)
1939 						    != 0) {
1940 							vref(vp);
1941 							nvp = vp;
1942 							r = 0;
1943 						} else
1944 							r = VOP_LOOKUP(vp, &nvp,
1945 							    &cn);
1946 					}
1947 				}
1948 				if (!r) {
1949 				    if (refp == NULL &&
1950 					((nd->nd_flag & ND_NFSV3) ||
1951 					 NFSNONZERO_ATTRBIT(&attrbits))) {
1952 					r = nfsvno_getfh(nvp, &nfh, p);
1953 					if (!r)
1954 					    r = nfsvno_getattr(nvp, nvap,
1955 						nd->nd_cred, p);
1956 				    }
1957 				} else {
1958 				    nvp = NULL;
1959 				}
1960 				if (r) {
1961 					if (!NFSISSET_ATTRBIT(&attrbits,
1962 					    NFSATTRBIT_RDATTRERROR)) {
1963 						if (nvp != NULL)
1964 							vput(nvp);
1965 						nd->nd_repstat = r;
1966 						break;
1967 					}
1968 				}
1969 			}
1970 
1971 			/*
1972 			 * Build the directory record xdr
1973 			 */
1974 			if (nd->nd_flag & ND_NFSV3) {
1975 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1976 				*tl++ = newnfs_true;
1977 				*tl++ = 0;
1978 				*tl = txdr_unsigned(dp->d_fileno);
1979 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
1980 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1981 				*tl++ = 0;
1982 				*tl = txdr_unsigned(*cookiep);
1983 				nfsrv_postopattr(nd, 0, nvap);
1984 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
1985 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
1986 				if (nvp != NULL)
1987 					vput(nvp);
1988 			} else {
1989 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1990 				*tl++ = newnfs_true;
1991 				*tl++ = 0;
1992 				*tl = txdr_unsigned(*cookiep);
1993 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
1994 				if (nvp != NULL)
1995 					NFSVOPUNLOCK(nvp, 0, p);
1996 				if (refp != NULL) {
1997 					dirlen += nfsrv_putreferralattr(nd,
1998 					    &savbits, refp, 0,
1999 					    &nd->nd_repstat);
2000 					if (nd->nd_repstat) {
2001 						if (nvp != NULL)
2002 							vrele(nvp);
2003 						break;
2004 					}
2005 				} else if (r) {
2006 					dirlen += nfsvno_fillattr(nd, nvp, nvap,
2007 					    &nfh, r, &rderrbits, nd->nd_cred,
2008 					    p, isdgram, 0);
2009 				} else {
2010 					dirlen += nfsvno_fillattr(nd, nvp, nvap,
2011 					    &nfh, r, &attrbits, nd->nd_cred,
2012 					    p, isdgram, 0);
2013 				}
2014 				if (nvp != NULL)
2015 					vrele(nvp);
2016 				dirlen += (3 * NFSX_UNSIGNED);
2017 			}
2018 			if (dirlen <= cnt)
2019 				entrycnt++;
2020 		}
2021 		cpos += dp->d_reclen;
2022 		dp = (struct dirent *)cpos;
2023 		cookiep++;
2024 		ncookies--;
2025 	}
2026 	if (!usevget && VOP_ISLOCKED(vp))
2027 		vput(vp);
2028 	else
2029 		vrele(vp);
2030 
2031 	/*
2032 	 * If dirlen > cnt, we must strip off the last entry. If that
2033 	 * results in an empty reply, report NFSERR_TOOSMALL.
2034 	 */
2035 	if (dirlen > cnt || nd->nd_repstat) {
2036 		if (!nd->nd_repstat && entrycnt == 0)
2037 			nd->nd_repstat = NFSERR_TOOSMALL;
2038 		if (nd->nd_repstat)
2039 			newnfs_trimtrailing(nd, mb0, bpos0);
2040 		else
2041 			newnfs_trimtrailing(nd, mb1, bpos1);
2042 		eofflag = 0;
2043 	} else if (cpos < cend)
2044 		eofflag = 0;
2045 	if (!nd->nd_repstat) {
2046 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2047 		*tl++ = newnfs_false;
2048 		if (eofflag)
2049 			*tl = newnfs_true;
2050 		else
2051 			*tl = newnfs_false;
2052 	}
2053 	FREE((caddr_t)cookies, M_TEMP);
2054 	FREE((caddr_t)rbuf, M_TEMP);
2055 	return (0);
2056 nfsmout:
2057 	vput(vp);
2058 	return (error);
2059 }
2060 
2061 /*
2062  * Get the settable attributes out of the mbuf list.
2063  * (Return 0 or EBADRPC)
2064  */
2065 int
2066 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2067     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2068 {
2069 	u_int32_t *tl;
2070 	struct nfsv2_sattr *sp;
2071 	struct timeval curtime;
2072 	int error = 0, toclient = 0;
2073 
2074 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2075 	case ND_NFSV2:
2076 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2077 		/*
2078 		 * Some old clients didn't fill in the high order 16bits.
2079 		 * --> check the low order 2 bytes for 0xffff
2080 		 */
2081 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2082 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2083 		if (sp->sa_uid != newnfs_xdrneg1)
2084 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2085 		if (sp->sa_gid != newnfs_xdrneg1)
2086 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2087 		if (sp->sa_size != newnfs_xdrneg1)
2088 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2089 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2090 #ifdef notyet
2091 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2092 #else
2093 			nvap->na_atime.tv_sec =
2094 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2095 			nvap->na_atime.tv_nsec = 0;
2096 #endif
2097 		}
2098 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2099 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2100 		break;
2101 	case ND_NFSV3:
2102 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2103 		if (*tl == newnfs_true) {
2104 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2105 			nvap->na_mode = nfstov_mode(*tl);
2106 		}
2107 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2108 		if (*tl == newnfs_true) {
2109 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2110 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2111 		}
2112 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2113 		if (*tl == newnfs_true) {
2114 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2115 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2116 		}
2117 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2118 		if (*tl == newnfs_true) {
2119 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2120 			nvap->na_size = fxdr_hyper(tl);
2121 		}
2122 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2123 		switch (fxdr_unsigned(int, *tl)) {
2124 		case NFSV3SATTRTIME_TOCLIENT:
2125 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2126 			fxdr_nfsv3time(tl, &nvap->na_atime);
2127 			toclient = 1;
2128 			break;
2129 		case NFSV3SATTRTIME_TOSERVER:
2130 			NFSGETTIME(&curtime);
2131 			nvap->na_atime.tv_sec = curtime.tv_sec;
2132 			nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2133 			nvap->na_vaflags |= VA_UTIMES_NULL;
2134 			break;
2135 		};
2136 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2137 		switch (fxdr_unsigned(int, *tl)) {
2138 		case NFSV3SATTRTIME_TOCLIENT:
2139 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2140 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2141 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2142 			break;
2143 		case NFSV3SATTRTIME_TOSERVER:
2144 			NFSGETTIME(&curtime);
2145 			nvap->na_mtime.tv_sec = curtime.tv_sec;
2146 			nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2147 			if (!toclient)
2148 				nvap->na_vaflags |= VA_UTIMES_NULL;
2149 			break;
2150 		};
2151 		break;
2152 	case ND_NFSV4:
2153 		error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2154 	};
2155 nfsmout:
2156 	return (error);
2157 }
2158 
2159 /*
2160  * Handle the setable attributes for V4.
2161  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2162  */
2163 int
2164 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2165     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2166 {
2167 	u_int32_t *tl;
2168 	int attrsum = 0;
2169 	int i, j;
2170 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2171 	int toclient = 0;
2172 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2173 	uid_t uid;
2174 	gid_t gid;
2175 	struct timeval curtime;
2176 
2177 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2178 	if (error)
2179 		return (error);
2180 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2181 	attrsize = fxdr_unsigned(int, *tl);
2182 
2183 	/*
2184 	 * Loop around getting the setable attributes. If an unsupported
2185 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2186 	 */
2187 	if (retnotsup) {
2188 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2189 		bitpos = NFSATTRBIT_MAX;
2190 	} else {
2191 		bitpos = 0;
2192 	}
2193 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2194 	    if (attrsum > attrsize) {
2195 		error = NFSERR_BADXDR;
2196 		goto nfsmout;
2197 	    }
2198 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2199 		switch (bitpos) {
2200 		case NFSATTRBIT_SIZE:
2201 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2202 			nvap->na_size = fxdr_hyper(tl);
2203 			attrsum += NFSX_HYPER;
2204 			break;
2205 		case NFSATTRBIT_ACL:
2206 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2207 			    p);
2208 			if (error)
2209 				goto nfsmout;
2210 			if (aceerr && !nd->nd_repstat)
2211 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2212 			attrsum += aclsize;
2213 			break;
2214 		case NFSATTRBIT_ARCHIVE:
2215 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2216 			if (!nd->nd_repstat)
2217 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2218 			attrsum += NFSX_UNSIGNED;
2219 			break;
2220 		case NFSATTRBIT_HIDDEN:
2221 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2222 			if (!nd->nd_repstat)
2223 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2224 			attrsum += NFSX_UNSIGNED;
2225 			break;
2226 		case NFSATTRBIT_MIMETYPE:
2227 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2228 			i = fxdr_unsigned(int, *tl);
2229 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2230 			if (error)
2231 				goto nfsmout;
2232 			if (!nd->nd_repstat)
2233 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2234 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2235 			break;
2236 		case NFSATTRBIT_MODE:
2237 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2238 			nvap->na_mode = nfstov_mode(*tl);
2239 			attrsum += NFSX_UNSIGNED;
2240 			break;
2241 		case NFSATTRBIT_OWNER:
2242 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2243 			j = fxdr_unsigned(int, *tl);
2244 			if (j < 0)
2245 				return (NFSERR_BADXDR);
2246 			if (j > NFSV4_SMALLSTR)
2247 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2248 			else
2249 				cp = namestr;
2250 			error = nfsrv_mtostr(nd, cp, j);
2251 			if (error) {
2252 				if (j > NFSV4_SMALLSTR)
2253 					free(cp, M_NFSSTRING);
2254 				return (error);
2255 			}
2256 			if (!nd->nd_repstat) {
2257 				nd->nd_repstat = nfsv4_strtouid(cp,j,&uid,p);
2258 				if (!nd->nd_repstat)
2259 					nvap->na_uid = uid;
2260 			}
2261 			if (j > NFSV4_SMALLSTR)
2262 				free(cp, M_NFSSTRING);
2263 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2264 			break;
2265 		case NFSATTRBIT_OWNERGROUP:
2266 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2267 			j = fxdr_unsigned(int, *tl);
2268 			if (j < 0)
2269 				return (NFSERR_BADXDR);
2270 			if (j > NFSV4_SMALLSTR)
2271 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2272 			else
2273 				cp = namestr;
2274 			error = nfsrv_mtostr(nd, cp, j);
2275 			if (error) {
2276 				if (j > NFSV4_SMALLSTR)
2277 					free(cp, M_NFSSTRING);
2278 				return (error);
2279 			}
2280 			if (!nd->nd_repstat) {
2281 				nd->nd_repstat = nfsv4_strtogid(cp,j,&gid,p);
2282 				if (!nd->nd_repstat)
2283 					nvap->na_gid = gid;
2284 			}
2285 			if (j > NFSV4_SMALLSTR)
2286 				free(cp, M_NFSSTRING);
2287 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2288 			break;
2289 		case NFSATTRBIT_SYSTEM:
2290 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2291 			if (!nd->nd_repstat)
2292 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2293 			attrsum += NFSX_UNSIGNED;
2294 			break;
2295 		case NFSATTRBIT_TIMEACCESSSET:
2296 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2297 			attrsum += NFSX_UNSIGNED;
2298 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2299 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2300 			    fxdr_nfsv4time(tl, &nvap->na_atime);
2301 			    toclient = 1;
2302 			    attrsum += NFSX_V4TIME;
2303 			} else {
2304 			    NFSGETTIME(&curtime);
2305 			    nvap->na_atime.tv_sec = curtime.tv_sec;
2306 			    nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2307 			    nvap->na_vaflags |= VA_UTIMES_NULL;
2308 			}
2309 			break;
2310 		case NFSATTRBIT_TIMEBACKUP:
2311 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2312 			if (!nd->nd_repstat)
2313 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2314 			attrsum += NFSX_V4TIME;
2315 			break;
2316 		case NFSATTRBIT_TIMECREATE:
2317 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2318 			if (!nd->nd_repstat)
2319 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2320 			attrsum += NFSX_V4TIME;
2321 			break;
2322 		case NFSATTRBIT_TIMEMODIFYSET:
2323 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2324 			attrsum += NFSX_UNSIGNED;
2325 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2326 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2327 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
2328 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
2329 			    attrsum += NFSX_V4TIME;
2330 			} else {
2331 			    NFSGETTIME(&curtime);
2332 			    nvap->na_mtime.tv_sec = curtime.tv_sec;
2333 			    nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2334 			    if (!toclient)
2335 				nvap->na_vaflags |= VA_UTIMES_NULL;
2336 			}
2337 			break;
2338 		default:
2339 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2340 			/*
2341 			 * set bitpos so we drop out of the loop.
2342 			 */
2343 			bitpos = NFSATTRBIT_MAX;
2344 			break;
2345 		};
2346 	}
2347 
2348 	/*
2349 	 * some clients pad the attrlist, so we need to skip over the
2350 	 * padding.
2351 	 */
2352 	if (attrsum > attrsize) {
2353 		error = NFSERR_BADXDR;
2354 	} else {
2355 		attrsize = NFSM_RNDUP(attrsize);
2356 		if (attrsum < attrsize)
2357 			error = nfsm_advance(nd, attrsize - attrsum, -1);
2358 	}
2359 nfsmout:
2360 	return (error);
2361 }
2362 
2363 /*
2364  * Check/setup export credentials.
2365  */
2366 int
2367 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2368     struct ucred *credanon)
2369 {
2370 	int error = 0;
2371 
2372 	/*
2373 	 * Check/setup credentials.
2374 	 */
2375 	if (nd->nd_flag & ND_GSS)
2376 		exp->nes_exflag &= ~MNT_EXPORTANON;
2377 
2378 	/*
2379 	 * Check to see if the operation is allowed for this security flavor.
2380 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2381 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2382 	 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2383 	 */
2384 	if (nfsvno_testexp(nd, exp) &&
2385 	    nd->nd_procnum != NFSV4OP_SECINFO &&
2386 	    nd->nd_procnum != NFSPROC_FSINFO) {
2387 		if (nd->nd_flag & ND_NFSV4)
2388 			error = NFSERR_WRONGSEC;
2389 		else
2390 			error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2391 		return (error);
2392 	}
2393 
2394 	/*
2395 	 * Check to see if the file system is exported V4 only.
2396 	 */
2397 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4))
2398 		return (NFSERR_PROGNOTV4);
2399 
2400 	/*
2401 	 * Now, map the user credentials.
2402 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
2403 	 *  Fsinfo RPC. If set for anything else, this code might need
2404 	 *  to change.)
2405 	 */
2406 	if (NFSVNO_EXPORTED(exp) &&
2407 	    ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2408 	     NFSVNO_EXPORTANON(exp) ||
2409 	     (nd->nd_flag & ND_AUTHNONE))) {
2410 		nd->nd_cred->cr_uid = credanon->cr_uid;
2411 		nd->nd_cred->cr_gid = credanon->cr_gid;
2412 		crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2413 		    credanon->cr_groups);
2414 	}
2415 	return (0);
2416 }
2417 
2418 /*
2419  * Check exports.
2420  */
2421 int
2422 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2423     struct ucred **credp)
2424 {
2425 	int i, error, *secflavors;
2426 
2427 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2428 	    &exp->nes_numsecflavor, &secflavors);
2429 	if (error) {
2430 		if (nfs_rootfhset) {
2431 			exp->nes_exflag = 0;
2432 			exp->nes_numsecflavor = 0;
2433 			error = 0;
2434 		}
2435 	} else {
2436 		/* Copy the security flavors. */
2437 		for (i = 0; i < exp->nes_numsecflavor; i++)
2438 			exp->nes_secflavors[i] = secflavors[i];
2439 	}
2440 	return (error);
2441 }
2442 
2443 /*
2444  * Get a vnode for a file handle and export stuff.
2445  */
2446 int
2447 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2448     struct vnode **vpp, struct nfsexstuff *exp, struct ucred **credp)
2449 {
2450 	int i, error, *secflavors;
2451 
2452 	*credp = NULL;
2453 	exp->nes_numsecflavor = 0;
2454 	error = VFS_FHTOVP(mp, &fhp->fh_fid, vpp);
2455 	if (error != 0)
2456 		/* Make sure the server replies ESTALE to the client. */
2457 		error = ESTALE;
2458 	if (nam && !error) {
2459 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2460 		    &exp->nes_numsecflavor, &secflavors);
2461 		if (error) {
2462 			if (nfs_rootfhset) {
2463 				exp->nes_exflag = 0;
2464 				exp->nes_numsecflavor = 0;
2465 				error = 0;
2466 			} else {
2467 				vput(*vpp);
2468 			}
2469 		} else {
2470 			/* Copy the security flavors. */
2471 			for (i = 0; i < exp->nes_numsecflavor; i++)
2472 				exp->nes_secflavors[i] = secflavors[i];
2473 		}
2474 	}
2475 	return (error);
2476 }
2477 
2478 /*
2479  * Do the pathconf vnode op.
2480  */
2481 int
2482 nfsvno_pathconf(struct vnode *vp, int flag, register_t *retf,
2483     struct ucred *cred, struct thread *p)
2484 {
2485 	int error;
2486 
2487 	error = VOP_PATHCONF(vp, flag, retf);
2488 	return (error);
2489 }
2490 
2491 /*
2492  * nfsd_fhtovp() - convert a fh to a vnode ptr
2493  * 	- look up fsid in mount list (if not found ret error)
2494  *	- get vp and export rights by calling nfsvno_fhtovp()
2495  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2496  *	  for AUTH_SYS
2497  * Also handle getting the Giant lock for the file system,
2498  * as required:
2499  * - if same mount point as *mpp
2500  *       do nothing
2501  *   else if *mpp == NULL
2502  *       if already locked
2503  *           leave it locked
2504  *       else
2505  *           call VFS_LOCK_GIANT()
2506  *   else
2507  *       if already locked
2508  *            unlock Giant
2509  *       call VFS_LOCK_GIANT()
2510  */
2511 void
2512 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp,
2513     struct vnode **vpp, struct nfsexstuff *exp,
2514     struct mount **mpp, int startwrite, struct thread *p)
2515 {
2516 	struct mount *mp;
2517 	struct ucred *credanon;
2518 	fhandle_t *fhp;
2519 
2520 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
2521 	/*
2522 	 * Check for the special case of the nfsv4root_fh.
2523 	 */
2524 	mp = vfs_getvfs(&fhp->fh_fsid);
2525 	if (!mp) {
2526 		*vpp = NULL;
2527 		nd->nd_repstat = ESTALE;
2528 		if (*mpp && exp->nes_vfslocked)
2529 			VFS_UNLOCK_GIANT(*mpp);
2530 		*mpp = NULL;
2531 		exp->nes_vfslocked = 0;
2532 		return;
2533 	}
2534 
2535 	/*
2536 	 * Now, handle Giant for the file system.
2537 	 */
2538 	if (*mpp != NULL && *mpp != mp && exp->nes_vfslocked) {
2539 		VFS_UNLOCK_GIANT(*mpp);
2540 		exp->nes_vfslocked = 0;
2541 	}
2542 	if (!exp->nes_vfslocked && *mpp != mp)
2543 		exp->nes_vfslocked = VFS_LOCK_GIANT(mp);
2544 
2545 	*mpp = mp;
2546 	if (startwrite)
2547 		vn_start_write(NULL, mpp, V_WAIT);
2548 
2549 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, vpp, exp,
2550 	    &credanon);
2551 
2552 	/*
2553 	 * For NFSv4 without a pseudo root fs, unexported file handles
2554 	 * can be returned, so that Lookup works everywhere.
2555 	 */
2556 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2557 	    !(nd->nd_flag & ND_NFSV4)) {
2558 		vput(*vpp);
2559 		nd->nd_repstat = EACCES;
2560 	}
2561 
2562 	/*
2563 	 * Personally, I've never seen any point in requiring a
2564 	 * reserved port#, since only in the rare case where the
2565 	 * clients are all boxes with secure system priviledges,
2566 	 * does it provide any enhanced security, but... some people
2567 	 * believe it to be useful and keep putting this code back in.
2568 	 * (There is also some "security checker" out there that
2569 	 *  complains if the nfs server doesn't enforce this.)
2570 	 * However, note the following:
2571 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
2572 	 *	required.
2573 	 * RFC2623 recommends that, if a reserved port# is checked for,
2574 	 *	that there be a way to turn that off--> ifdef'd.
2575 	 */
2576 #ifdef NFS_REQRSVPORT
2577 	if (!nd->nd_repstat) {
2578 		struct sockaddr_in *saddr;
2579 		struct sockaddr_in6 *saddr6;
2580 
2581 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2582 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2583 		if (!(nd->nd_flag & ND_NFSV4) &&
2584 		    ((saddr->sin_family == AF_INET &&
2585 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2586 		     (saddr6->sin6_family == AF_INET6 &&
2587 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2588 			vput(*vpp);
2589 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2590 		}
2591 	}
2592 #endif	/* NFS_REQRSVPORT */
2593 
2594 	/*
2595 	 * Check/setup credentials.
2596 	 */
2597 	if (!nd->nd_repstat) {
2598 		nd->nd_saveduid = nd->nd_cred->cr_uid;
2599 		nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2600 		if (nd->nd_repstat)
2601 			vput(*vpp);
2602 	}
2603 	if (credanon != NULL)
2604 		crfree(credanon);
2605 	if (nd->nd_repstat) {
2606 		if (startwrite)
2607 			vn_finished_write(mp);
2608 		if (exp->nes_vfslocked) {
2609 			VFS_UNLOCK_GIANT(mp);
2610 			exp->nes_vfslocked = 0;
2611 		}
2612 		vfs_rel(mp);
2613 		*vpp = NULL;
2614 		*mpp = NULL;
2615 	} else {
2616 		vfs_rel(mp);
2617 	}
2618 }
2619 
2620 /*
2621  * glue for fp.
2622  */
2623 int
2624 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2625 {
2626 	struct filedesc *fdp;
2627 	struct file *fp;
2628 
2629 	fdp = p->td_proc->p_fd;
2630 	if (fd >= fdp->fd_nfiles ||
2631 	    (fp = fdp->fd_ofiles[fd]) == NULL)
2632 		return (EBADF);
2633 	*fpp = fp;
2634 	return (0);
2635 }
2636 
2637 /*
2638  * Called from nfssvc() to update the exports list. Just call
2639  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2640  * in the mount list.
2641  */
2642 int
2643 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2644 {
2645 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2646 	int error;
2647 	struct nameidata nd;
2648 	fhandle_t fh;
2649 
2650 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2651 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT)) {
2652 		nfs_rootfhset = 0;
2653 		nfsv4root_set = 0;
2654 	} else if (error == 0) {
2655 		if (nfsexargp->fspec == NULL)
2656 			return (EPERM);
2657 		/*
2658 		 * If fspec != NULL, this is the v4root path.
2659 		 */
2660 		NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_USERSPACE,
2661 		    nfsexargp->fspec, p);
2662 		if ((error = namei(&nd)) != 0)
2663 			return (error);
2664 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
2665 		vrele(nd.ni_vp);
2666 		if (!error) {
2667 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2668 			NFSBCOPY((caddr_t)&fh,
2669 			    nfs_rootfh.nfsrvfh_data,
2670 			    sizeof (fhandle_t));
2671 			nfs_rootfhset = 1;
2672 		}
2673 	}
2674 	return (error);
2675 }
2676 
2677 /*
2678  * Get the tcp socket sequence numbers we need.
2679  * (Maybe this should be moved to the tcp sources?)
2680  */
2681 int
2682 nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap)
2683 {
2684 	struct inpcb *inp;
2685 	struct tcpcb *tp;
2686 
2687 	inp = sotoinpcb(so);
2688 	KASSERT(inp != NULL, ("nfsrv_getsocksndseq: inp == NULL"));
2689 	INP_RLOCK(inp);
2690 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2691 		INP_RUNLOCK(inp);
2692 		return (EPIPE);
2693 	}
2694 	tp = intotcpcb(inp);
2695 	if (tp->t_state != TCPS_ESTABLISHED) {
2696 		INP_RUNLOCK(inp);
2697 		return (EPIPE);
2698 	}
2699 	*maxp = tp->snd_max;
2700 	*unap = tp->snd_una;
2701 	INP_RUNLOCK(inp);
2702 	return (0);
2703 }
2704 
2705 /*
2706  * This function needs to test to see if the system is near its limit
2707  * for memory allocation via malloc() or mget() and return True iff
2708  * either of these resources are near their limit.
2709  * XXX (For now, this is just a stub.)
2710  */
2711 int nfsrv_testmalloclimit = 0;
2712 int
2713 nfsrv_mallocmget_limit(void)
2714 {
2715 	static int printmesg = 0;
2716 	static int testval = 1;
2717 
2718 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2719 		if ((printmesg++ % 100) == 0)
2720 			printf("nfsd: malloc/mget near limit\n");
2721 		return (1);
2722 	}
2723 	return (0);
2724 }
2725 
2726 /*
2727  * BSD specific initialization of a mount point.
2728  */
2729 void
2730 nfsd_mntinit(void)
2731 {
2732 	static int inited = 0;
2733 
2734 	if (inited)
2735 		return;
2736 	inited = 1;
2737 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2738 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2739 	nfsv4root_mnt.mnt_export = NULL;
2740 	TAILQ_INIT(&nfsv4root_opt);
2741 	TAILQ_INIT(&nfsv4root_newopt);
2742 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2743 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2744 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
2745 }
2746 
2747 /*
2748  * Get a vnode for a file handle, without checking exports, etc.
2749  */
2750 struct vnode *
2751 nfsvno_getvp(fhandle_t *fhp)
2752 {
2753 	struct mount *mp;
2754 	struct vnode *vp;
2755 	int error;
2756 
2757 	mp = vfs_getvfs(&fhp->fh_fsid);
2758 	if (mp == NULL)
2759 		return (NULL);
2760 	error = VFS_FHTOVP(mp, &fhp->fh_fid, &vp);
2761 	if (error)
2762 		return (NULL);
2763 	return (vp);
2764 }
2765 
2766 /*
2767  * Check to see it a byte range lock held by a process running
2768  * locally on the server conflicts with the new lock.
2769  */
2770 int
2771 nfsvno_localconflict(struct vnode *vp, int ftype, u_int64_t first,
2772     u_int64_t end, struct nfslockconflict *cfp, struct thread *td)
2773 {
2774 	int error;
2775 	struct flock fl;
2776 
2777 	if (!nfsrv_dolocallocks)
2778 		return (0);
2779 	fl.l_whence = SEEK_SET;
2780 	fl.l_type = ftype;
2781 	fl.l_start = (off_t)first;
2782 	if (end == NFS64BITSSET)
2783 		fl.l_len = 0;
2784 	else
2785 		fl.l_len = (off_t)(end - first);
2786 	/*
2787 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2788 	 * values for all calls, so that all locks will be held by the
2789 	 * nfsd server. (The nfsd server handles conflicts between the
2790 	 * various clients.)
2791 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2792 	 * bytes, so it can't be put in l_sysid.
2793 	 */
2794 	if (nfsv4_sysid == 0)
2795 		nfsv4_sysid = nlm_acquire_next_sysid();
2796 	fl.l_pid = (pid_t)0;
2797 	fl.l_sysid = (int)nfsv4_sysid;
2798 
2799 	NFSVOPUNLOCK(vp, 0, td);
2800 	error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_GETLK, &fl,
2801 	    (F_POSIX | F_REMOTE));
2802 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, td);
2803 	if (error)
2804 		return (error);
2805 	if (fl.l_type == F_UNLCK)
2806 		return (0);
2807 	if (cfp != NULL) {
2808 		cfp->cl_clientid.lval[0] = cfp->cl_clientid.lval[1] = 0;
2809 		cfp->cl_first = (u_int64_t)fl.l_start;
2810 		if (fl.l_len == 0)
2811 			cfp->cl_end = NFS64BITSSET;
2812 		else
2813 			cfp->cl_end = (u_int64_t)
2814 			    (fl.l_start + fl.l_len);
2815 		if (fl.l_type == F_WRLCK)
2816 			cfp->cl_flags = NFSLCK_WRITE;
2817 		else
2818 			cfp->cl_flags = NFSLCK_READ;
2819 		sprintf(cfp->cl_owner, "LOCALID%d", fl.l_pid);
2820 		cfp->cl_ownerlen = strlen(cfp->cl_owner);
2821 		return (NFSERR_DENIED);
2822 	}
2823 	return (NFSERR_INVAL);
2824 }
2825 
2826 /*
2827  * Do a local VOP_ADVLOCK().
2828  */
2829 int
2830 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2831     u_int64_t end, struct thread *td)
2832 {
2833 	int error;
2834 	struct flock fl;
2835 	u_int64_t tlen;
2836 
2837 	if (nfsrv_dolocallocks == 0)
2838 		return (0);
2839 	fl.l_whence = SEEK_SET;
2840 	fl.l_type = ftype;
2841 	fl.l_start = (off_t)first;
2842 	if (end == NFS64BITSSET) {
2843 		fl.l_len = 0;
2844 	} else {
2845 		tlen = end - first;
2846 		fl.l_len = (off_t)tlen;
2847 	}
2848 	/*
2849 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2850 	 * values for all calls, so that all locks will be held by the
2851 	 * nfsd server. (The nfsd server handles conflicts between the
2852 	 * various clients.)
2853 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2854 	 * bytes, so it can't be put in l_sysid.
2855 	 */
2856 	if (nfsv4_sysid == 0)
2857 		nfsv4_sysid = nlm_acquire_next_sysid();
2858 	fl.l_pid = (pid_t)0;
2859 	fl.l_sysid = (int)nfsv4_sysid;
2860 
2861 	NFSVOPUNLOCK(vp, 0, td);
2862 	if (ftype == F_UNLCK)
2863 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
2864 		    (F_POSIX | F_REMOTE));
2865 	else
2866 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
2867 		    (F_POSIX | F_REMOTE));
2868 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, td);
2869 	return (error);
2870 }
2871 
2872 /*
2873  * Unlock an underlying local file system.
2874  */
2875 void
2876 nfsvno_unlockvfs(struct mount *mp)
2877 {
2878 
2879 	VFS_UNLOCK_GIANT(mp);
2880 }
2881 
2882 /*
2883  * Lock an underlying file system, as required, and return
2884  * whether or not it is locked.
2885  */
2886 int
2887 nfsvno_lockvfs(struct mount *mp)
2888 {
2889 	int ret;
2890 
2891 	ret = VFS_LOCK_GIANT(mp);
2892 	return (ret);
2893 }
2894 
2895 /*
2896  * Check the nfsv4 root exports.
2897  */
2898 int
2899 nfsvno_v4rootexport(struct nfsrv_descript *nd)
2900 {
2901 	struct ucred *credanon;
2902 	int exflags, error, numsecflavor, *secflavors, i;
2903 
2904 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
2905 	    &credanon, &numsecflavor, &secflavors);
2906 	if (error)
2907 		return (NFSERR_PROGUNAVAIL);
2908 	if (credanon != NULL)
2909 		crfree(credanon);
2910 	for (i = 0; i < numsecflavor; i++) {
2911 		if (secflavors[i] == AUTH_SYS)
2912 			nd->nd_flag |= ND_EXAUTHSYS;
2913 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
2914 			nd->nd_flag |= ND_EXGSS;
2915 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
2916 			nd->nd_flag |= ND_EXGSSINTEGRITY;
2917 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
2918 			nd->nd_flag |= ND_EXGSSPRIVACY;
2919 	}
2920 	return (0);
2921 }
2922 
2923 /*
2924  * Nfs server psuedo system call for the nfsd's
2925  */
2926 /*
2927  * MPSAFE
2928  */
2929 static int
2930 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
2931 {
2932 	struct file *fp;
2933 	struct nfsd_addsock_args sockarg;
2934 	struct nfsd_nfsd_args nfsdarg;
2935 	int error;
2936 
2937 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
2938 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
2939 		if (error)
2940 			return (error);
2941 		if ((error = fget(td, sockarg.sock, &fp)) != 0) {
2942 			return (error);
2943 		}
2944 		if (fp->f_type != DTYPE_SOCKET) {
2945 			fdrop(fp, td);
2946 			return (EPERM);
2947 		}
2948 		error = nfsrvd_addsock(fp);
2949 		fdrop(fp, td);
2950 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
2951 		if (uap->argp == NULL)
2952 			return (EINVAL);
2953 		error = copyin(uap->argp, (caddr_t)&nfsdarg,
2954 		    sizeof (nfsdarg));
2955 		if (error)
2956 			return (error);
2957 		error = nfsrvd_nfsd(td, &nfsdarg);
2958 	} else {
2959 		error = nfssvc_srvcall(td, uap, td->td_ucred);
2960 	}
2961 	return (error);
2962 }
2963 
2964 static int
2965 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
2966 {
2967 	struct nfsex_args export;
2968 	struct file *fp = NULL;
2969 	int stablefd, len;
2970 	struct nfsd_clid adminrevoke;
2971 	struct nfsd_dumplist dumplist;
2972 	struct nfsd_dumpclients *dumpclients;
2973 	struct nfsd_dumplocklist dumplocklist;
2974 	struct nfsd_dumplocks *dumplocks;
2975 	struct nameidata nd;
2976 	vnode_t vp;
2977 	int error = EINVAL;
2978 
2979 	if (uap->flag & NFSSVC_PUBLICFH) {
2980 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
2981 		    sizeof (fhandle_t));
2982 		error = copyin(uap->argp,
2983 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
2984 		if (!error)
2985 			nfs_pubfhset = 1;
2986 	} else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
2987 		error = copyin(uap->argp,(caddr_t)&export,
2988 		    sizeof (struct nfsex_args));
2989 		if (!error)
2990 			error = nfsrv_v4rootexport(&export, cred, p);
2991 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
2992 		nfs_pubfhset = 0;
2993 		error = 0;
2994 	} else if (uap->flag & NFSSVC_STABLERESTART) {
2995 		error = copyin(uap->argp, (caddr_t)&stablefd,
2996 		    sizeof (int));
2997 		if (!error)
2998 			error = fp_getfvp(p, stablefd, &fp, &vp);
2999 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3000 			error = EBADF;
3001 		if (!error && newnfs_numnfsd != 0)
3002 			error = EPERM;
3003 		if (!error) {
3004 			nfsrv_stablefirst.nsf_fp = fp;
3005 			nfsrv_setupstable(p);
3006 		}
3007 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3008 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3009 		    sizeof (struct nfsd_clid));
3010 		if (!error)
3011 			error = nfsrv_adminrevoke(&adminrevoke, p);
3012 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3013 		error = copyin(uap->argp, (caddr_t)&dumplist,
3014 		    sizeof (struct nfsd_dumplist));
3015 		if (!error && (dumplist.ndl_size < 1 ||
3016 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3017 			error = EPERM;
3018 		if (!error) {
3019 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3020 		    dumpclients = (struct nfsd_dumpclients *)malloc(len,
3021 			M_TEMP, M_WAITOK);
3022 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3023 		    error = copyout(dumpclients,
3024 			CAST_USER_ADDR_T(dumplist.ndl_list), len);
3025 		    free((caddr_t)dumpclients, M_TEMP);
3026 		}
3027 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3028 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3029 		    sizeof (struct nfsd_dumplocklist));
3030 		if (!error && (dumplocklist.ndllck_size < 1 ||
3031 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3032 			error = EPERM;
3033 		if (!error)
3034 			error = nfsrv_lookupfilename(&nd,
3035 				dumplocklist.ndllck_fname, p);
3036 		if (!error) {
3037 			len = sizeof (struct nfsd_dumplocks) *
3038 				dumplocklist.ndllck_size;
3039 			dumplocks = (struct nfsd_dumplocks *)malloc(len,
3040 				M_TEMP, M_WAITOK);
3041 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3042 			    dumplocklist.ndllck_size, p);
3043 			vput(nd.ni_vp);
3044 			error = copyout(dumplocks,
3045 			    CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3046 			free((caddr_t)dumplocks, M_TEMP);
3047 		}
3048 	}
3049 	return (error);
3050 }
3051 
3052 /*
3053  * Check exports.
3054  * Returns 0 if ok, 1 otherwise.
3055  */
3056 int
3057 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3058 {
3059 	int i;
3060 
3061 	/*
3062 	 * This seems odd, but allow the case where the security flavor
3063 	 * list is empty. This happens when NFSv4 is traversing non-exported
3064 	 * file systems. Exported file systems should always have a non-empty
3065 	 * security flavor list.
3066 	 */
3067 	if (exp->nes_numsecflavor == 0)
3068 		return (0);
3069 
3070 	for (i = 0; i < exp->nes_numsecflavor; i++) {
3071 		/*
3072 		 * The tests for privacy and integrity must be first,
3073 		 * since ND_GSS is set for everything but AUTH_SYS.
3074 		 */
3075 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3076 		    (nd->nd_flag & ND_GSSPRIVACY))
3077 			return (0);
3078 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3079 		    (nd->nd_flag & ND_GSSINTEGRITY))
3080 			return (0);
3081 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3082 		    (nd->nd_flag & ND_GSS))
3083 			return (0);
3084 		if (exp->nes_secflavors[i] == AUTH_SYS &&
3085 		    (nd->nd_flag & ND_GSS) == 0)
3086 			return (0);
3087 	}
3088 	return (1);
3089 }
3090 
3091 /*
3092  * Calculate a hash value for the fid in a file handle.
3093  */
3094 uint32_t
3095 nfsrv_hashfh(fhandle_t *fhp)
3096 {
3097 	uint32_t hashval;
3098 
3099 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3100 	return (hashval);
3101 }
3102 
3103 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3104 
3105 /*
3106  * Called once to initialize data structures...
3107  */
3108 static int
3109 nfsd_modevent(module_t mod, int type, void *data)
3110 {
3111 	int error = 0;
3112 	static int loaded = 0;
3113 
3114 	switch (type) {
3115 	case MOD_LOAD:
3116 		if (loaded)
3117 			return (0);
3118 		newnfs_portinit();
3119 		mtx_init(&nfs_cache_mutex, "nfs_cache_mutex", NULL, MTX_DEF);
3120 		mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3121 		mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3122 		    MTX_DEF);
3123 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3124 		nfsrvd_initcache();
3125 		nfsd_init();
3126 		NFSD_LOCK();
3127 		nfsrvd_init(0);
3128 		NFSD_UNLOCK();
3129 		nfsd_mntinit();
3130 #ifdef VV_DISABLEDELEG
3131 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3132 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3133 #endif
3134 		nfsd_call_servertimer = nfsrv_servertimer;
3135 		nfsd_call_nfsd = nfssvc_nfsd;
3136 		loaded = 1;
3137 		break;
3138 
3139 	case MOD_UNLOAD:
3140 		if (newnfs_numnfsd != 0) {
3141 			error = EBUSY;
3142 			break;
3143 		}
3144 
3145 #ifdef VV_DISABLEDELEG
3146 		vn_deleg_ops.vndeleg_recall = NULL;
3147 		vn_deleg_ops.vndeleg_disable = NULL;
3148 #endif
3149 		nfsd_call_servertimer = NULL;
3150 		nfsd_call_nfsd = NULL;
3151 		/* and get rid of the locks */
3152 		mtx_destroy(&nfs_cache_mutex);
3153 		mtx_destroy(&nfs_v4root_mutex);
3154 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3155 		lockdestroy(&nfsv4root_mnt.mnt_explock);
3156 		loaded = 0;
3157 		break;
3158 	default:
3159 		error = EOPNOTSUPP;
3160 		break;
3161 	}
3162 	return error;
3163 }
3164 static moduledata_t nfsd_mod = {
3165 	"nfsd",
3166 	nfsd_modevent,
3167 	NULL,
3168 };
3169 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3170 
3171 /* So that loader and kldload(2) can find us, wherever we are.. */
3172 MODULE_VERSION(nfsd, 1);
3173 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3174 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3175 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3176 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3177 
3178