xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision ab0b9f6b3073e6c4d1dfbf07444d7db67a189a96)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include <sys/capability.h>
38 
39 /*
40  * Functions that perform the vfs operations required by the routines in
41  * nfsd_serv.c. It is hoped that this change will make the server more
42  * portable.
43  */
44 
45 #include <fs/nfs/nfsport.h>
46 #include <sys/hash.h>
47 #include <sys/sysctl.h>
48 #include <nlm/nlm_prot.h>
49 #include <nlm/nlm.h>
50 
51 FEATURE(nfsd, "NFSv4 server");
52 
53 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
54 extern int nfsrv_useacl;
55 extern int newnfs_numnfsd;
56 extern struct mount nfsv4root_mnt;
57 extern struct nfsrv_stablefirst nfsrv_stablefirst;
58 extern void (*nfsd_call_servertimer)(void);
59 extern SVCPOOL	*nfsrvd_pool;
60 extern struct nfsv4lock nfsd_suspend_lock;
61 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
62 NFSDLOCKMUTEX;
63 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
64 struct mtx nfsrc_udpmtx;
65 struct mtx nfs_v4root_mutex;
66 struct nfsrvfh nfs_rootfh, nfs_pubfh;
67 int nfs_pubfhset = 0, nfs_rootfhset = 0;
68 struct proc *nfsd_master_proc = NULL;
69 static pid_t nfsd_master_pid = (pid_t)-1;
70 static char nfsd_master_comm[MAXCOMLEN + 1];
71 static struct timeval nfsd_master_start;
72 static uint32_t nfsv4_sysid = 0;
73 
74 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
75     struct ucred *);
76 
77 int nfsrv_enable_crossmntpt = 1;
78 static int nfs_commit_blks;
79 static int nfs_commit_miss;
80 extern int nfsrv_issuedelegs;
81 extern int nfsrv_dolocallocks;
82 
83 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server");
84 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
85     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
86 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
87     0, "");
88 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
89     0, "");
90 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
91     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
92 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
93     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
94 
95 #define	MAX_REORDERED_RPC	16
96 #define	NUM_HEURISTIC		1031
97 #define	NHUSE_INIT		64
98 #define	NHUSE_INC		16
99 #define	NHUSE_MAX		2048
100 
101 static struct nfsheur {
102 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
103 	off_t nh_nextoff;	/* next offset for sequential detection */
104 	int nh_use;		/* use count for selection */
105 	int nh_seqcount;	/* heuristic */
106 } nfsheur[NUM_HEURISTIC];
107 
108 
109 /*
110  * Heuristic to detect sequential operation.
111  */
112 static struct nfsheur *
113 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
114 {
115 	struct nfsheur *nh;
116 	int hi, try;
117 
118 	/* Locate best candidate. */
119 	try = 32;
120 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
121 	nh = &nfsheur[hi];
122 	while (try--) {
123 		if (nfsheur[hi].nh_vp == vp) {
124 			nh = &nfsheur[hi];
125 			break;
126 		}
127 		if (nfsheur[hi].nh_use > 0)
128 			--nfsheur[hi].nh_use;
129 		hi = (hi + 1) % NUM_HEURISTIC;
130 		if (nfsheur[hi].nh_use < nh->nh_use)
131 			nh = &nfsheur[hi];
132 	}
133 
134 	/* Initialize hint if this is a new file. */
135 	if (nh->nh_vp != vp) {
136 		nh->nh_vp = vp;
137 		nh->nh_nextoff = uio->uio_offset;
138 		nh->nh_use = NHUSE_INIT;
139 		if (uio->uio_offset == 0)
140 			nh->nh_seqcount = 4;
141 		else
142 			nh->nh_seqcount = 1;
143 	}
144 
145 	/* Calculate heuristic. */
146 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
147 	    uio->uio_offset == nh->nh_nextoff) {
148 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
149 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
150 		if (nh->nh_seqcount > IO_SEQMAX)
151 			nh->nh_seqcount = IO_SEQMAX;
152 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
153 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
154 		/* Probably a reordered RPC, leave seqcount alone. */
155 	} else if (nh->nh_seqcount > 1) {
156 		nh->nh_seqcount /= 2;
157 	} else {
158 		nh->nh_seqcount = 0;
159 	}
160 	nh->nh_use += NHUSE_INC;
161 	if (nh->nh_use > NHUSE_MAX)
162 		nh->nh_use = NHUSE_MAX;
163 	return (nh);
164 }
165 
166 /*
167  * Get attributes into nfsvattr structure.
168  */
169 int
170 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
171     struct thread *p, int vpislocked)
172 {
173 	int error, lockedit = 0;
174 
175 	if (vpislocked == 0) {
176 		/*
177 		 * When vpislocked == 0, the vnode is either exclusively
178 		 * locked by this thread or not locked by this thread.
179 		 * As such, shared lock it, if not exclusively locked.
180 		 */
181 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
182 			lockedit = 1;
183 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
184 		}
185 	}
186 	error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
187 	if (lockedit != 0)
188 		NFSVOPUNLOCK(vp, 0);
189 
190 	NFSEXITCODE(error);
191 	return (error);
192 }
193 
194 /*
195  * Get a file handle for a vnode.
196  */
197 int
198 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
199 {
200 	int error;
201 
202 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
203 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
204 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
205 
206 	NFSEXITCODE(error);
207 	return (error);
208 }
209 
210 /*
211  * Perform access checking for vnodes obtained from file handles that would
212  * refer to files already opened by a Unix client. You cannot just use
213  * vn_writechk() and VOP_ACCESSX() for two reasons.
214  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
215  *     case.
216  * 2 - The owner is to be given access irrespective of mode bits for some
217  *     operations, so that processes that chmod after opening a file don't
218  *     break.
219  */
220 int
221 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
222     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
223     u_int32_t *supportedtypep)
224 {
225 	struct vattr vattr;
226 	int error = 0, getret = 0;
227 
228 	if (vpislocked == 0) {
229 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
230 			error = EPERM;
231 			goto out;
232 		}
233 	}
234 	if (accmode & VWRITE) {
235 		/* Just vn_writechk() changed to check rdonly */
236 		/*
237 		 * Disallow write attempts on read-only file systems;
238 		 * unless the file is a socket or a block or character
239 		 * device resident on the file system.
240 		 */
241 		if (NFSVNO_EXRDONLY(exp) ||
242 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
243 			switch (vp->v_type) {
244 			case VREG:
245 			case VDIR:
246 			case VLNK:
247 				error = EROFS;
248 			default:
249 				break;
250 			}
251 		}
252 		/*
253 		 * If there's shared text associated with
254 		 * the inode, try to free it up once.  If
255 		 * we fail, we can't allow writing.
256 		 */
257 		if (VOP_IS_TEXT(vp) && error == 0)
258 			error = ETXTBSY;
259 	}
260 	if (error != 0) {
261 		if (vpislocked == 0)
262 			NFSVOPUNLOCK(vp, 0);
263 		goto out;
264 	}
265 
266 	/*
267 	 * Should the override still be applied when ACLs are enabled?
268 	 */
269 	error = VOP_ACCESSX(vp, accmode, cred, p);
270 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
271 		/*
272 		 * Try again with VEXPLICIT_DENY, to see if the test for
273 		 * deletion is supported.
274 		 */
275 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
276 		if (error == 0) {
277 			if (vp->v_type == VDIR) {
278 				accmode &= ~(VDELETE | VDELETE_CHILD);
279 				accmode |= VWRITE;
280 				error = VOP_ACCESSX(vp, accmode, cred, p);
281 			} else if (supportedtypep != NULL) {
282 				*supportedtypep &= ~NFSACCESS_DELETE;
283 			}
284 		}
285 	}
286 
287 	/*
288 	 * Allow certain operations for the owner (reads and writes
289 	 * on files that are already open).
290 	 */
291 	if (override != NFSACCCHK_NOOVERRIDE &&
292 	    (error == EPERM || error == EACCES)) {
293 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
294 			error = 0;
295 		else if (override & NFSACCCHK_ALLOWOWNER) {
296 			getret = VOP_GETATTR(vp, &vattr, cred);
297 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
298 				error = 0;
299 		}
300 	}
301 	if (vpislocked == 0)
302 		NFSVOPUNLOCK(vp, 0);
303 
304 out:
305 	NFSEXITCODE(error);
306 	return (error);
307 }
308 
309 /*
310  * Set attribute(s) vnop.
311  */
312 int
313 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
314     struct thread *p, struct nfsexstuff *exp)
315 {
316 	int error;
317 
318 	error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
319 	NFSEXITCODE(error);
320 	return (error);
321 }
322 
323 /*
324  * Set up nameidata for a lookup() call and do it.
325  */
326 int
327 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
328     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
329     struct vnode **retdirp)
330 {
331 	struct componentname *cnp = &ndp->ni_cnd;
332 	int i;
333 	struct iovec aiov;
334 	struct uio auio;
335 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
336 	int error = 0, crossmnt;
337 	char *cp;
338 
339 	*retdirp = NULL;
340 	cnp->cn_nameptr = cnp->cn_pnbuf;
341 	ndp->ni_strictrelative = 0;
342 	/*
343 	 * Extract and set starting directory.
344 	 */
345 	if (dp->v_type != VDIR) {
346 		if (islocked)
347 			vput(dp);
348 		else
349 			vrele(dp);
350 		nfsvno_relpathbuf(ndp);
351 		error = ENOTDIR;
352 		goto out1;
353 	}
354 	if (islocked)
355 		NFSVOPUNLOCK(dp, 0);
356 	VREF(dp);
357 	*retdirp = dp;
358 	if (NFSVNO_EXRDONLY(exp))
359 		cnp->cn_flags |= RDONLY;
360 	ndp->ni_segflg = UIO_SYSSPACE;
361 	crossmnt = 1;
362 
363 	if (nd->nd_flag & ND_PUBLOOKUP) {
364 		ndp->ni_loopcnt = 0;
365 		if (cnp->cn_pnbuf[0] == '/') {
366 			vrele(dp);
367 			/*
368 			 * Check for degenerate pathnames here, since lookup()
369 			 * panics on them.
370 			 */
371 			for (i = 1; i < ndp->ni_pathlen; i++)
372 				if (cnp->cn_pnbuf[i] != '/')
373 					break;
374 			if (i == ndp->ni_pathlen) {
375 				error = NFSERR_ACCES;
376 				goto out;
377 			}
378 			dp = rootvnode;
379 			VREF(dp);
380 		}
381 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
382 	    (nd->nd_flag & ND_NFSV4) == 0) {
383 		/*
384 		 * Only cross mount points for NFSv4 when doing a
385 		 * mount while traversing the file system above
386 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
387 		 */
388 		cnp->cn_flags |= NOCROSSMOUNT;
389 		crossmnt = 0;
390 	}
391 
392 	/*
393 	 * Initialize for scan, set ni_startdir and bump ref on dp again
394 	 * because lookup() will dereference ni_startdir.
395 	 */
396 
397 	cnp->cn_thread = p;
398 	ndp->ni_startdir = dp;
399 	ndp->ni_rootdir = rootvnode;
400 	ndp->ni_topdir = NULL;
401 
402 	if (!lockleaf)
403 		cnp->cn_flags |= LOCKLEAF;
404 	for (;;) {
405 		cnp->cn_nameptr = cnp->cn_pnbuf;
406 		/*
407 		 * Call lookup() to do the real work.  If an error occurs,
408 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
409 		 * we do not have to dereference anything before returning.
410 		 * In either case ni_startdir will be dereferenced and NULLed
411 		 * out.
412 		 */
413 		error = lookup(ndp);
414 		if (error)
415 			break;
416 
417 		/*
418 		 * Check for encountering a symbolic link.  Trivial
419 		 * termination occurs if no symlink encountered.
420 		 */
421 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
422 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
423 				nfsvno_relpathbuf(ndp);
424 			if (ndp->ni_vp && !lockleaf)
425 				NFSVOPUNLOCK(ndp->ni_vp, 0);
426 			break;
427 		}
428 
429 		/*
430 		 * Validate symlink
431 		 */
432 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
433 			NFSVOPUNLOCK(ndp->ni_dvp, 0);
434 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
435 			error = EINVAL;
436 			goto badlink2;
437 		}
438 
439 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
440 			error = ELOOP;
441 			goto badlink2;
442 		}
443 		if (ndp->ni_pathlen > 1)
444 			cp = uma_zalloc(namei_zone, M_WAITOK);
445 		else
446 			cp = cnp->cn_pnbuf;
447 		aiov.iov_base = cp;
448 		aiov.iov_len = MAXPATHLEN;
449 		auio.uio_iov = &aiov;
450 		auio.uio_iovcnt = 1;
451 		auio.uio_offset = 0;
452 		auio.uio_rw = UIO_READ;
453 		auio.uio_segflg = UIO_SYSSPACE;
454 		auio.uio_td = NULL;
455 		auio.uio_resid = MAXPATHLEN;
456 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
457 		if (error) {
458 		badlink1:
459 			if (ndp->ni_pathlen > 1)
460 				uma_zfree(namei_zone, cp);
461 		badlink2:
462 			vrele(ndp->ni_dvp);
463 			vput(ndp->ni_vp);
464 			break;
465 		}
466 		linklen = MAXPATHLEN - auio.uio_resid;
467 		if (linklen == 0) {
468 			error = ENOENT;
469 			goto badlink1;
470 		}
471 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
472 			error = ENAMETOOLONG;
473 			goto badlink1;
474 		}
475 
476 		/*
477 		 * Adjust or replace path
478 		 */
479 		if (ndp->ni_pathlen > 1) {
480 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
481 			uma_zfree(namei_zone, cnp->cn_pnbuf);
482 			cnp->cn_pnbuf = cp;
483 		} else
484 			cnp->cn_pnbuf[linklen] = '\0';
485 		ndp->ni_pathlen += linklen;
486 
487 		/*
488 		 * Cleanup refs for next loop and check if root directory
489 		 * should replace current directory.  Normally ni_dvp
490 		 * becomes the new base directory and is cleaned up when
491 		 * we loop.  Explicitly null pointers after invalidation
492 		 * to clarify operation.
493 		 */
494 		vput(ndp->ni_vp);
495 		ndp->ni_vp = NULL;
496 
497 		if (cnp->cn_pnbuf[0] == '/') {
498 			vrele(ndp->ni_dvp);
499 			ndp->ni_dvp = ndp->ni_rootdir;
500 			VREF(ndp->ni_dvp);
501 		}
502 		ndp->ni_startdir = ndp->ni_dvp;
503 		ndp->ni_dvp = NULL;
504 	}
505 	if (!lockleaf)
506 		cnp->cn_flags &= ~LOCKLEAF;
507 
508 out:
509 	if (error) {
510 		nfsvno_relpathbuf(ndp);
511 		ndp->ni_vp = NULL;
512 		ndp->ni_dvp = NULL;
513 		ndp->ni_startdir = NULL;
514 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
515 		ndp->ni_dvp = NULL;
516 	}
517 
518 out1:
519 	NFSEXITCODE2(error, nd);
520 	return (error);
521 }
522 
523 /*
524  * Set up a pathname buffer and return a pointer to it and, optionally
525  * set a hash pointer.
526  */
527 void
528 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
529 {
530 	struct componentname *cnp = &ndp->ni_cnd;
531 
532 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
533 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
534 	if (hashpp != NULL)
535 		*hashpp = NULL;
536 	*bufpp = cnp->cn_pnbuf;
537 }
538 
539 /*
540  * Release the above path buffer, if not released by nfsvno_namei().
541  */
542 void
543 nfsvno_relpathbuf(struct nameidata *ndp)
544 {
545 
546 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
547 		panic("nfsrelpath");
548 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
549 	ndp->ni_cnd.cn_flags &= ~HASBUF;
550 }
551 
552 /*
553  * Readlink vnode op into an mbuf list.
554  */
555 int
556 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
557     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
558 {
559 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
560 	struct iovec *ivp = iv;
561 	struct uio io, *uiop = &io;
562 	struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
563 	int i, len, tlen, error = 0;
564 
565 	len = 0;
566 	i = 0;
567 	while (len < NFS_MAXPATHLEN) {
568 		NFSMGET(mp);
569 		MCLGET(mp, M_WAITOK);
570 		mp->m_len = NFSMSIZ(mp);
571 		if (len == 0) {
572 			mp3 = mp2 = mp;
573 		} else {
574 			mp2->m_next = mp;
575 			mp2 = mp;
576 		}
577 		if ((len + mp->m_len) > NFS_MAXPATHLEN) {
578 			mp->m_len = NFS_MAXPATHLEN - len;
579 			len = NFS_MAXPATHLEN;
580 		} else {
581 			len += mp->m_len;
582 		}
583 		ivp->iov_base = mtod(mp, caddr_t);
584 		ivp->iov_len = mp->m_len;
585 		i++;
586 		ivp++;
587 	}
588 	uiop->uio_iov = iv;
589 	uiop->uio_iovcnt = i;
590 	uiop->uio_offset = 0;
591 	uiop->uio_resid = len;
592 	uiop->uio_rw = UIO_READ;
593 	uiop->uio_segflg = UIO_SYSSPACE;
594 	uiop->uio_td = NULL;
595 	error = VOP_READLINK(vp, uiop, cred);
596 	if (error) {
597 		m_freem(mp3);
598 		*lenp = 0;
599 		goto out;
600 	}
601 	if (uiop->uio_resid > 0) {
602 		len -= uiop->uio_resid;
603 		tlen = NFSM_RNDUP(len);
604 		nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
605 	}
606 	*lenp = len;
607 	*mpp = mp3;
608 	*mpendp = mp;
609 
610 out:
611 	NFSEXITCODE(error);
612 	return (error);
613 }
614 
615 /*
616  * Read vnode op call into mbuf list.
617  */
618 int
619 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
620     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
621 {
622 	struct mbuf *m;
623 	int i;
624 	struct iovec *iv;
625 	struct iovec *iv2;
626 	int error = 0, len, left, siz, tlen, ioflag = 0;
627 	struct mbuf *m2 = NULL, *m3;
628 	struct uio io, *uiop = &io;
629 	struct nfsheur *nh;
630 
631 	len = left = NFSM_RNDUP(cnt);
632 	m3 = NULL;
633 	/*
634 	 * Generate the mbuf list with the uio_iov ref. to it.
635 	 */
636 	i = 0;
637 	while (left > 0) {
638 		NFSMGET(m);
639 		MCLGET(m, M_WAITOK);
640 		m->m_len = 0;
641 		siz = min(M_TRAILINGSPACE(m), left);
642 		left -= siz;
643 		i++;
644 		if (m3)
645 			m2->m_next = m;
646 		else
647 			m3 = m;
648 		m2 = m;
649 	}
650 	MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
651 	    M_TEMP, M_WAITOK);
652 	uiop->uio_iov = iv2 = iv;
653 	m = m3;
654 	left = len;
655 	i = 0;
656 	while (left > 0) {
657 		if (m == NULL)
658 			panic("nfsvno_read iov");
659 		siz = min(M_TRAILINGSPACE(m), left);
660 		if (siz > 0) {
661 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
662 			iv->iov_len = siz;
663 			m->m_len += siz;
664 			left -= siz;
665 			iv++;
666 			i++;
667 		}
668 		m = m->m_next;
669 	}
670 	uiop->uio_iovcnt = i;
671 	uiop->uio_offset = off;
672 	uiop->uio_resid = len;
673 	uiop->uio_rw = UIO_READ;
674 	uiop->uio_segflg = UIO_SYSSPACE;
675 	nh = nfsrv_sequential_heuristic(uiop, vp);
676 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
677 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
678 	FREE((caddr_t)iv2, M_TEMP);
679 	if (error) {
680 		m_freem(m3);
681 		*mpp = NULL;
682 		goto out;
683 	}
684 	nh->nh_nextoff = uiop->uio_offset;
685 	tlen = len - uiop->uio_resid;
686 	cnt = cnt < tlen ? cnt : tlen;
687 	tlen = NFSM_RNDUP(cnt);
688 	if (tlen == 0) {
689 		m_freem(m3);
690 		m3 = NULL;
691 	} else if (len != tlen || tlen != cnt)
692 		nfsrv_adj(m3, len - tlen, tlen - cnt);
693 	*mpp = m3;
694 	*mpendp = m2;
695 
696 out:
697 	NFSEXITCODE(error);
698 	return (error);
699 }
700 
701 /*
702  * Write vnode op from an mbuf list.
703  */
704 int
705 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
706     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
707 {
708 	struct iovec *ivp;
709 	int i, len;
710 	struct iovec *iv;
711 	int ioflags, error;
712 	struct uio io, *uiop = &io;
713 	struct nfsheur *nh;
714 
715 	MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
716 	    M_WAITOK);
717 	uiop->uio_iov = iv = ivp;
718 	uiop->uio_iovcnt = cnt;
719 	i = mtod(mp, caddr_t) + mp->m_len - cp;
720 	len = retlen;
721 	while (len > 0) {
722 		if (mp == NULL)
723 			panic("nfsvno_write");
724 		if (i > 0) {
725 			i = min(i, len);
726 			ivp->iov_base = cp;
727 			ivp->iov_len = i;
728 			ivp++;
729 			len -= i;
730 		}
731 		mp = mp->m_next;
732 		if (mp) {
733 			i = mp->m_len;
734 			cp = mtod(mp, caddr_t);
735 		}
736 	}
737 
738 	if (stable == NFSWRITE_UNSTABLE)
739 		ioflags = IO_NODELOCKED;
740 	else
741 		ioflags = (IO_SYNC | IO_NODELOCKED);
742 	uiop->uio_resid = retlen;
743 	uiop->uio_rw = UIO_WRITE;
744 	uiop->uio_segflg = UIO_SYSSPACE;
745 	NFSUIOPROC(uiop, p);
746 	uiop->uio_offset = off;
747 	nh = nfsrv_sequential_heuristic(uiop, vp);
748 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
749 	error = VOP_WRITE(vp, uiop, ioflags, cred);
750 	if (error == 0)
751 		nh->nh_nextoff = uiop->uio_offset;
752 	FREE((caddr_t)iv, M_TEMP);
753 
754 	NFSEXITCODE(error);
755 	return (error);
756 }
757 
758 /*
759  * Common code for creating a regular file (plus special files for V2).
760  */
761 int
762 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
763     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
764     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
765 {
766 	u_quad_t tempsize;
767 	int error;
768 
769 	error = nd->nd_repstat;
770 	if (!error && ndp->ni_vp == NULL) {
771 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
772 			vrele(ndp->ni_startdir);
773 			error = VOP_CREATE(ndp->ni_dvp,
774 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
775 			vput(ndp->ni_dvp);
776 			nfsvno_relpathbuf(ndp);
777 			if (!error) {
778 				if (*exclusive_flagp) {
779 					*exclusive_flagp = 0;
780 					NFSVNO_ATTRINIT(nvap);
781 					nvap->na_atime.tv_sec = cverf[0];
782 					nvap->na_atime.tv_nsec = cverf[1];
783 					error = VOP_SETATTR(ndp->ni_vp,
784 					    &nvap->na_vattr, nd->nd_cred);
785 				}
786 			}
787 		/*
788 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
789 		 * (This implies, just get out on an error.)
790 		 */
791 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
792 			nvap->na_type == VFIFO) {
793 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
794 				nvap->na_type = VFIFO;
795                         if (nvap->na_type != VFIFO &&
796 			    (error = priv_check_cred(nd->nd_cred,
797 			     PRIV_VFS_MKNOD_DEV, 0))) {
798 				vrele(ndp->ni_startdir);
799 				nfsvno_relpathbuf(ndp);
800 				vput(ndp->ni_dvp);
801 				goto out;
802 			}
803 			nvap->na_rdev = rdev;
804 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
805 			    &ndp->ni_cnd, &nvap->na_vattr);
806 			vput(ndp->ni_dvp);
807 			nfsvno_relpathbuf(ndp);
808 			vrele(ndp->ni_startdir);
809 			if (error)
810 				goto out;
811 		} else {
812 			vrele(ndp->ni_startdir);
813 			nfsvno_relpathbuf(ndp);
814 			vput(ndp->ni_dvp);
815 			error = ENXIO;
816 			goto out;
817 		}
818 		*vpp = ndp->ni_vp;
819 	} else {
820 		/*
821 		 * Handle cases where error is already set and/or
822 		 * the file exists.
823 		 * 1 - clean up the lookup
824 		 * 2 - iff !error and na_size set, truncate it
825 		 */
826 		vrele(ndp->ni_startdir);
827 		nfsvno_relpathbuf(ndp);
828 		*vpp = ndp->ni_vp;
829 		if (ndp->ni_dvp == *vpp)
830 			vrele(ndp->ni_dvp);
831 		else
832 			vput(ndp->ni_dvp);
833 		if (!error && nvap->na_size != VNOVAL) {
834 			error = nfsvno_accchk(*vpp, VWRITE,
835 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
836 			    NFSACCCHK_VPISLOCKED, NULL);
837 			if (!error) {
838 				tempsize = nvap->na_size;
839 				NFSVNO_ATTRINIT(nvap);
840 				nvap->na_size = tempsize;
841 				error = VOP_SETATTR(*vpp,
842 				    &nvap->na_vattr, nd->nd_cred);
843 			}
844 		}
845 		if (error)
846 			vput(*vpp);
847 	}
848 
849 out:
850 	NFSEXITCODE(error);
851 	return (error);
852 }
853 
854 /*
855  * Do a mknod vnode op.
856  */
857 int
858 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
859     struct thread *p)
860 {
861 	int error = 0;
862 	enum vtype vtyp;
863 
864 	vtyp = nvap->na_type;
865 	/*
866 	 * Iff doesn't exist, create it.
867 	 */
868 	if (ndp->ni_vp) {
869 		vrele(ndp->ni_startdir);
870 		nfsvno_relpathbuf(ndp);
871 		vput(ndp->ni_dvp);
872 		vrele(ndp->ni_vp);
873 		error = EEXIST;
874 		goto out;
875 	}
876 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
877 		vrele(ndp->ni_startdir);
878 		nfsvno_relpathbuf(ndp);
879 		vput(ndp->ni_dvp);
880 		error = NFSERR_BADTYPE;
881 		goto out;
882 	}
883 	if (vtyp == VSOCK) {
884 		vrele(ndp->ni_startdir);
885 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
886 		    &ndp->ni_cnd, &nvap->na_vattr);
887 		vput(ndp->ni_dvp);
888 		nfsvno_relpathbuf(ndp);
889 	} else {
890 		if (nvap->na_type != VFIFO &&
891 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
892 			vrele(ndp->ni_startdir);
893 			nfsvno_relpathbuf(ndp);
894 			vput(ndp->ni_dvp);
895 			goto out;
896 		}
897 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
898 		    &ndp->ni_cnd, &nvap->na_vattr);
899 		vput(ndp->ni_dvp);
900 		nfsvno_relpathbuf(ndp);
901 		vrele(ndp->ni_startdir);
902 		/*
903 		 * Since VOP_MKNOD returns the ni_vp, I can't
904 		 * see any reason to do the lookup.
905 		 */
906 	}
907 
908 out:
909 	NFSEXITCODE(error);
910 	return (error);
911 }
912 
913 /*
914  * Mkdir vnode op.
915  */
916 int
917 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
918     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
919 {
920 	int error = 0;
921 
922 	if (ndp->ni_vp != NULL) {
923 		if (ndp->ni_dvp == ndp->ni_vp)
924 			vrele(ndp->ni_dvp);
925 		else
926 			vput(ndp->ni_dvp);
927 		vrele(ndp->ni_vp);
928 		nfsvno_relpathbuf(ndp);
929 		error = EEXIST;
930 		goto out;
931 	}
932 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
933 	    &nvap->na_vattr);
934 	vput(ndp->ni_dvp);
935 	nfsvno_relpathbuf(ndp);
936 
937 out:
938 	NFSEXITCODE(error);
939 	return (error);
940 }
941 
942 /*
943  * symlink vnode op.
944  */
945 int
946 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
947     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
948     struct nfsexstuff *exp)
949 {
950 	int error = 0;
951 
952 	if (ndp->ni_vp) {
953 		vrele(ndp->ni_startdir);
954 		nfsvno_relpathbuf(ndp);
955 		if (ndp->ni_dvp == ndp->ni_vp)
956 			vrele(ndp->ni_dvp);
957 		else
958 			vput(ndp->ni_dvp);
959 		vrele(ndp->ni_vp);
960 		error = EEXIST;
961 		goto out;
962 	}
963 
964 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
965 	    &nvap->na_vattr, pathcp);
966 	vput(ndp->ni_dvp);
967 	vrele(ndp->ni_startdir);
968 	nfsvno_relpathbuf(ndp);
969 	/*
970 	 * Although FreeBSD still had the lookup code in
971 	 * it for 7/current, there doesn't seem to be any
972 	 * point, since VOP_SYMLINK() returns the ni_vp.
973 	 * Just vput it for v2.
974 	 */
975 	if (!not_v2 && !error)
976 		vput(ndp->ni_vp);
977 
978 out:
979 	NFSEXITCODE(error);
980 	return (error);
981 }
982 
983 /*
984  * Parse symbolic link arguments.
985  * This function has an ugly side effect. It will MALLOC() an area for
986  * the symlink and set iov_base to point to it, only if it succeeds.
987  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
988  * be FREE'd later.
989  */
990 int
991 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
992     struct thread *p, char **pathcpp, int *lenp)
993 {
994 	u_int32_t *tl;
995 	char *pathcp = NULL;
996 	int error = 0, len;
997 	struct nfsv2_sattr *sp;
998 
999 	*pathcpp = NULL;
1000 	*lenp = 0;
1001 	if ((nd->nd_flag & ND_NFSV3) &&
1002 	    (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
1003 		goto nfsmout;
1004 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1005 	len = fxdr_unsigned(int, *tl);
1006 	if (len > NFS_MAXPATHLEN || len <= 0) {
1007 		error = EBADRPC;
1008 		goto nfsmout;
1009 	}
1010 	MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
1011 	error = nfsrv_mtostr(nd, pathcp, len);
1012 	if (error)
1013 		goto nfsmout;
1014 	if (nd->nd_flag & ND_NFSV2) {
1015 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1016 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1017 	}
1018 	*pathcpp = pathcp;
1019 	*lenp = len;
1020 	NFSEXITCODE2(0, nd);
1021 	return (0);
1022 nfsmout:
1023 	if (pathcp)
1024 		free(pathcp, M_TEMP);
1025 	NFSEXITCODE2(error, nd);
1026 	return (error);
1027 }
1028 
1029 /*
1030  * Remove a non-directory object.
1031  */
1032 int
1033 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1034     struct thread *p, struct nfsexstuff *exp)
1035 {
1036 	struct vnode *vp;
1037 	int error = 0;
1038 
1039 	vp = ndp->ni_vp;
1040 	if (vp->v_type == VDIR)
1041 		error = NFSERR_ISDIR;
1042 	else if (is_v4)
1043 		error = nfsrv_checkremove(vp, 1, p);
1044 	if (!error)
1045 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1046 	if (ndp->ni_dvp == vp)
1047 		vrele(ndp->ni_dvp);
1048 	else
1049 		vput(ndp->ni_dvp);
1050 	vput(vp);
1051 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1052 		nfsvno_relpathbuf(ndp);
1053 	NFSEXITCODE(error);
1054 	return (error);
1055 }
1056 
1057 /*
1058  * Remove a directory.
1059  */
1060 int
1061 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1062     struct thread *p, struct nfsexstuff *exp)
1063 {
1064 	struct vnode *vp;
1065 	int error = 0;
1066 
1067 	vp = ndp->ni_vp;
1068 	if (vp->v_type != VDIR) {
1069 		error = ENOTDIR;
1070 		goto out;
1071 	}
1072 	/*
1073 	 * No rmdir "." please.
1074 	 */
1075 	if (ndp->ni_dvp == vp) {
1076 		error = EINVAL;
1077 		goto out;
1078 	}
1079 	/*
1080 	 * The root of a mounted filesystem cannot be deleted.
1081 	 */
1082 	if (vp->v_vflag & VV_ROOT)
1083 		error = EBUSY;
1084 out:
1085 	if (!error)
1086 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1087 	if (ndp->ni_dvp == vp)
1088 		vrele(ndp->ni_dvp);
1089 	else
1090 		vput(ndp->ni_dvp);
1091 	vput(vp);
1092 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1093 		nfsvno_relpathbuf(ndp);
1094 	NFSEXITCODE(error);
1095 	return (error);
1096 }
1097 
1098 /*
1099  * Rename vnode op.
1100  */
1101 int
1102 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1103     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1104 {
1105 	struct vnode *fvp, *tvp, *tdvp;
1106 	int error = 0;
1107 
1108 	fvp = fromndp->ni_vp;
1109 	if (ndstat) {
1110 		vrele(fromndp->ni_dvp);
1111 		vrele(fvp);
1112 		error = ndstat;
1113 		goto out1;
1114 	}
1115 	tdvp = tondp->ni_dvp;
1116 	tvp = tondp->ni_vp;
1117 	if (tvp != NULL) {
1118 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1119 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1120 			goto out;
1121 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1122 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1123 			goto out;
1124 		}
1125 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1126 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1127 			goto out;
1128 		}
1129 
1130 		/*
1131 		 * A rename to '.' or '..' results in a prematurely
1132 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1133 		 * here.
1134 		 */
1135 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1136 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1137 		    (tondp->ni_cnd.cn_namelen == 2 &&
1138 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1139 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1140 			error = EINVAL;
1141 			goto out;
1142 		}
1143 	}
1144 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1145 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1146 		goto out;
1147 	}
1148 	if (fvp->v_mount != tdvp->v_mount) {
1149 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1150 		goto out;
1151 	}
1152 	if (fvp == tdvp) {
1153 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1154 		goto out;
1155 	}
1156 	if (fvp == tvp) {
1157 		/*
1158 		 * If source and destination are the same, there is nothing to
1159 		 * do. Set error to -1 to indicate this.
1160 		 */
1161 		error = -1;
1162 		goto out;
1163 	}
1164 	if (ndflag & ND_NFSV4) {
1165 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1166 			error = nfsrv_checkremove(fvp, 0, p);
1167 			NFSVOPUNLOCK(fvp, 0);
1168 		} else
1169 			error = EPERM;
1170 		if (tvp && !error)
1171 			error = nfsrv_checkremove(tvp, 1, p);
1172 	} else {
1173 		/*
1174 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1175 		 * that the NFSv4 client won't be confused by the rename.
1176 		 * Since nfsd_recalldelegation() can only be called on an
1177 		 * unlocked vnode at this point and fvp is the file that will
1178 		 * still exist after the rename, just do fvp.
1179 		 */
1180 		nfsd_recalldelegation(fvp, p);
1181 	}
1182 out:
1183 	if (!error) {
1184 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1185 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1186 		    &tondp->ni_cnd);
1187 	} else {
1188 		if (tdvp == tvp)
1189 			vrele(tdvp);
1190 		else
1191 			vput(tdvp);
1192 		if (tvp)
1193 			vput(tvp);
1194 		vrele(fromndp->ni_dvp);
1195 		vrele(fvp);
1196 		if (error == -1)
1197 			error = 0;
1198 	}
1199 	vrele(tondp->ni_startdir);
1200 	nfsvno_relpathbuf(tondp);
1201 out1:
1202 	vrele(fromndp->ni_startdir);
1203 	nfsvno_relpathbuf(fromndp);
1204 	NFSEXITCODE(error);
1205 	return (error);
1206 }
1207 
1208 /*
1209  * Link vnode op.
1210  */
1211 int
1212 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1213     struct thread *p, struct nfsexstuff *exp)
1214 {
1215 	struct vnode *xp;
1216 	int error = 0;
1217 
1218 	xp = ndp->ni_vp;
1219 	if (xp != NULL) {
1220 		error = EEXIST;
1221 	} else {
1222 		xp = ndp->ni_dvp;
1223 		if (vp->v_mount != xp->v_mount)
1224 			error = EXDEV;
1225 	}
1226 	if (!error) {
1227 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1228 		if ((vp->v_iflag & VI_DOOMED) == 0)
1229 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1230 		else
1231 			error = EPERM;
1232 		if (ndp->ni_dvp == vp)
1233 			vrele(ndp->ni_dvp);
1234 		else
1235 			vput(ndp->ni_dvp);
1236 		NFSVOPUNLOCK(vp, 0);
1237 	} else {
1238 		if (ndp->ni_dvp == ndp->ni_vp)
1239 			vrele(ndp->ni_dvp);
1240 		else
1241 			vput(ndp->ni_dvp);
1242 		if (ndp->ni_vp)
1243 			vrele(ndp->ni_vp);
1244 	}
1245 	nfsvno_relpathbuf(ndp);
1246 	NFSEXITCODE(error);
1247 	return (error);
1248 }
1249 
1250 /*
1251  * Do the fsync() appropriate for the commit.
1252  */
1253 int
1254 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1255     struct thread *td)
1256 {
1257 	int error = 0;
1258 
1259 	/*
1260 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1261 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1262 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1263 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1264 	 */
1265 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT) {
1266 		/*
1267 		 * Give up and do the whole thing
1268 		 */
1269 		if (vp->v_object &&
1270 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1271 			VM_OBJECT_WLOCK(vp->v_object);
1272 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1273 			VM_OBJECT_WUNLOCK(vp->v_object);
1274 		}
1275 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1276 	} else {
1277 		/*
1278 		 * Locate and synchronously write any buffers that fall
1279 		 * into the requested range.  Note:  we are assuming that
1280 		 * f_iosize is a power of 2.
1281 		 */
1282 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1283 		int iomask = iosize - 1;
1284 		struct bufobj *bo;
1285 		daddr_t lblkno;
1286 
1287 		/*
1288 		 * Align to iosize boundry, super-align to page boundry.
1289 		 */
1290 		if (off & iomask) {
1291 			cnt += off & iomask;
1292 			off &= ~(u_quad_t)iomask;
1293 		}
1294 		if (off & PAGE_MASK) {
1295 			cnt += off & PAGE_MASK;
1296 			off &= ~(u_quad_t)PAGE_MASK;
1297 		}
1298 		lblkno = off / iosize;
1299 
1300 		if (vp->v_object &&
1301 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1302 			VM_OBJECT_WLOCK(vp->v_object);
1303 			vm_object_page_clean(vp->v_object, off, off + cnt,
1304 			    OBJPC_SYNC);
1305 			VM_OBJECT_WUNLOCK(vp->v_object);
1306 		}
1307 
1308 		bo = &vp->v_bufobj;
1309 		BO_LOCK(bo);
1310 		while (cnt > 0) {
1311 			struct buf *bp;
1312 
1313 			/*
1314 			 * If we have a buffer and it is marked B_DELWRI we
1315 			 * have to lock and write it.  Otherwise the prior
1316 			 * write is assumed to have already been committed.
1317 			 *
1318 			 * gbincore() can return invalid buffers now so we
1319 			 * have to check that bit as well (though B_DELWRI
1320 			 * should not be set if B_INVAL is set there could be
1321 			 * a race here since we haven't locked the buffer).
1322 			 */
1323 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1324 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1325 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1326 					BO_LOCK(bo);
1327 					continue; /* retry */
1328 				}
1329 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1330 				    B_DELWRI) {
1331 					bremfree(bp);
1332 					bp->b_flags &= ~B_ASYNC;
1333 					bwrite(bp);
1334 					++nfs_commit_miss;
1335 				} else
1336 					BUF_UNLOCK(bp);
1337 				BO_LOCK(bo);
1338 			}
1339 			++nfs_commit_blks;
1340 			if (cnt < iosize)
1341 				break;
1342 			cnt -= iosize;
1343 			++lblkno;
1344 		}
1345 		BO_UNLOCK(bo);
1346 	}
1347 	NFSEXITCODE(error);
1348 	return (error);
1349 }
1350 
1351 /*
1352  * Statfs vnode op.
1353  */
1354 int
1355 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1356 {
1357 	int error;
1358 
1359 	error = VFS_STATFS(vp->v_mount, sf);
1360 	if (error == 0) {
1361 		/*
1362 		 * Since NFS handles these values as unsigned on the
1363 		 * wire, there is no way to represent negative values,
1364 		 * so set them to 0. Without this, they will appear
1365 		 * to be very large positive values for clients like
1366 		 * Solaris10.
1367 		 */
1368 		if (sf->f_bavail < 0)
1369 			sf->f_bavail = 0;
1370 		if (sf->f_ffree < 0)
1371 			sf->f_ffree = 0;
1372 	}
1373 	NFSEXITCODE(error);
1374 	return (error);
1375 }
1376 
1377 /*
1378  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1379  * must handle nfsrv_opencheck() calls after any other access checks.
1380  */
1381 void
1382 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1383     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1384     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1385     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1386     struct nfsexstuff *exp, struct vnode **vpp)
1387 {
1388 	struct vnode *vp = NULL;
1389 	u_quad_t tempsize;
1390 	struct nfsexstuff nes;
1391 
1392 	if (ndp->ni_vp == NULL)
1393 		nd->nd_repstat = nfsrv_opencheck(clientid,
1394 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1395 	if (!nd->nd_repstat) {
1396 		if (ndp->ni_vp == NULL) {
1397 			vrele(ndp->ni_startdir);
1398 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1399 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1400 			vput(ndp->ni_dvp);
1401 			nfsvno_relpathbuf(ndp);
1402 			if (!nd->nd_repstat) {
1403 				if (*exclusive_flagp) {
1404 					*exclusive_flagp = 0;
1405 					NFSVNO_ATTRINIT(nvap);
1406 					nvap->na_atime.tv_sec = cverf[0];
1407 					nvap->na_atime.tv_nsec = cverf[1];
1408 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1409 					    &nvap->na_vattr, cred);
1410 				} else {
1411 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1412 					    aclp, p, attrbitp, exp);
1413 				}
1414 			}
1415 			vp = ndp->ni_vp;
1416 		} else {
1417 			if (ndp->ni_startdir)
1418 				vrele(ndp->ni_startdir);
1419 			nfsvno_relpathbuf(ndp);
1420 			vp = ndp->ni_vp;
1421 			if (create == NFSV4OPEN_CREATE) {
1422 				if (ndp->ni_dvp == vp)
1423 					vrele(ndp->ni_dvp);
1424 				else
1425 					vput(ndp->ni_dvp);
1426 			}
1427 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1428 				if (ndp->ni_cnd.cn_flags & RDONLY)
1429 					NFSVNO_SETEXRDONLY(&nes);
1430 				else
1431 					NFSVNO_EXINIT(&nes);
1432 				nd->nd_repstat = nfsvno_accchk(vp,
1433 				    VWRITE, cred, &nes, p,
1434 				    NFSACCCHK_NOOVERRIDE,
1435 				    NFSACCCHK_VPISLOCKED, NULL);
1436 				nd->nd_repstat = nfsrv_opencheck(clientid,
1437 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1438 				if (!nd->nd_repstat) {
1439 					tempsize = nvap->na_size;
1440 					NFSVNO_ATTRINIT(nvap);
1441 					nvap->na_size = tempsize;
1442 					nd->nd_repstat = VOP_SETATTR(vp,
1443 					    &nvap->na_vattr, cred);
1444 				}
1445 			} else if (vp->v_type == VREG) {
1446 				nd->nd_repstat = nfsrv_opencheck(clientid,
1447 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1448 			}
1449 		}
1450 	} else {
1451 		if (ndp->ni_cnd.cn_flags & HASBUF)
1452 			nfsvno_relpathbuf(ndp);
1453 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1454 			vrele(ndp->ni_startdir);
1455 			if (ndp->ni_dvp == ndp->ni_vp)
1456 				vrele(ndp->ni_dvp);
1457 			else
1458 				vput(ndp->ni_dvp);
1459 			if (ndp->ni_vp)
1460 				vput(ndp->ni_vp);
1461 		}
1462 	}
1463 	*vpp = vp;
1464 
1465 	NFSEXITCODE2(0, nd);
1466 }
1467 
1468 /*
1469  * Updates the file rev and sets the mtime and ctime
1470  * to the current clock time, returning the va_filerev and va_Xtime
1471  * values.
1472  * Return ESTALE to indicate the vnode is VI_DOOMED.
1473  */
1474 int
1475 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1476     struct ucred *cred, struct thread *p)
1477 {
1478 	struct vattr va;
1479 
1480 	VATTR_NULL(&va);
1481 	vfs_timestamp(&va.va_mtime);
1482 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1483 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1484 		if ((vp->v_iflag & VI_DOOMED) != 0)
1485 			return (ESTALE);
1486 	}
1487 	(void) VOP_SETATTR(vp, &va, cred);
1488 	(void) nfsvno_getattr(vp, nvap, cred, p, 1);
1489 	return (0);
1490 }
1491 
1492 /*
1493  * Glue routine to nfsv4_fillattr().
1494  */
1495 int
1496 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1497     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1498     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1499     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1500 {
1501 	int error;
1502 
1503 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1504 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1505 	    mounted_on_fileno);
1506 	NFSEXITCODE2(0, nd);
1507 	return (error);
1508 }
1509 
1510 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1511 /*
1512  * nfs readdir service
1513  * - mallocs what it thinks is enough to read
1514  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1515  * - calls VOP_READDIR()
1516  * - loops around building the reply
1517  *	if the output generated exceeds count break out of loop
1518  *	The NFSM_CLGET macro is used here so that the reply will be packed
1519  *	tightly in mbuf clusters.
1520  * - it trims out records with d_fileno == 0
1521  *	this doesn't matter for Unix clients, but they might confuse clients
1522  *	for other os'.
1523  * - it trims out records with d_type == DT_WHT
1524  *	these cannot be seen through NFS (unless we extend the protocol)
1525  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1526  * PS: The NFS protocol spec. does not clarify what the "count" byte
1527  *	argument is a count of.. just name strings and file id's or the
1528  *	entire reply rpc or ...
1529  *	I tried just file name and id sizes and it confused the Sun client,
1530  *	so I am using the full rpc size now. The "paranoia.." comment refers
1531  *	to including the status longwords that are not a part of the dir.
1532  *	"entry" structures, but are in the rpc.
1533  */
1534 int
1535 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1536     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1537 {
1538 	struct dirent *dp;
1539 	u_int32_t *tl;
1540 	int dirlen;
1541 	char *cpos, *cend, *rbuf;
1542 	struct nfsvattr at;
1543 	int nlen, error = 0, getret = 1;
1544 	int siz, cnt, fullsiz, eofflag, ncookies;
1545 	u_int64_t off, toff, verf;
1546 	u_long *cookies = NULL, *cookiep;
1547 	struct uio io;
1548 	struct iovec iv;
1549 	int not_zfs;
1550 
1551 	if (nd->nd_repstat) {
1552 		nfsrv_postopattr(nd, getret, &at);
1553 		goto out;
1554 	}
1555 	if (nd->nd_flag & ND_NFSV2) {
1556 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1557 		off = fxdr_unsigned(u_quad_t, *tl++);
1558 	} else {
1559 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1560 		off = fxdr_hyper(tl);
1561 		tl += 2;
1562 		verf = fxdr_hyper(tl);
1563 		tl += 2;
1564 	}
1565 	toff = off;
1566 	cnt = fxdr_unsigned(int, *tl);
1567 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1568 		cnt = NFS_SRVMAXDATA(nd);
1569 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1570 	fullsiz = siz;
1571 	if (nd->nd_flag & ND_NFSV3) {
1572 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1573 		    p, 1);
1574 #if 0
1575 		/*
1576 		 * va_filerev is not sufficient as a cookie verifier,
1577 		 * since it is not supposed to change when entries are
1578 		 * removed/added unless that offset cookies returned to
1579 		 * the client are no longer valid.
1580 		 */
1581 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
1582 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1583 #endif
1584 	}
1585 	if (!nd->nd_repstat && vp->v_type != VDIR)
1586 		nd->nd_repstat = NFSERR_NOTDIR;
1587 	if (nd->nd_repstat == 0 && cnt == 0) {
1588 		if (nd->nd_flag & ND_NFSV2)
1589 			/* NFSv2 does not have NFSERR_TOOSMALL */
1590 			nd->nd_repstat = EPERM;
1591 		else
1592 			nd->nd_repstat = NFSERR_TOOSMALL;
1593 	}
1594 	if (!nd->nd_repstat)
1595 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1596 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1597 		    NFSACCCHK_VPISLOCKED, NULL);
1598 	if (nd->nd_repstat) {
1599 		vput(vp);
1600 		if (nd->nd_flag & ND_NFSV3)
1601 			nfsrv_postopattr(nd, getret, &at);
1602 		goto out;
1603 	}
1604 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1605 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1606 again:
1607 	eofflag = 0;
1608 	if (cookies) {
1609 		free((caddr_t)cookies, M_TEMP);
1610 		cookies = NULL;
1611 	}
1612 
1613 	iv.iov_base = rbuf;
1614 	iv.iov_len = siz;
1615 	io.uio_iov = &iv;
1616 	io.uio_iovcnt = 1;
1617 	io.uio_offset = (off_t)off;
1618 	io.uio_resid = siz;
1619 	io.uio_segflg = UIO_SYSSPACE;
1620 	io.uio_rw = UIO_READ;
1621 	io.uio_td = NULL;
1622 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1623 	    &cookies);
1624 	off = (u_int64_t)io.uio_offset;
1625 	if (io.uio_resid)
1626 		siz -= io.uio_resid;
1627 
1628 	if (!cookies && !nd->nd_repstat)
1629 		nd->nd_repstat = NFSERR_PERM;
1630 	if (nd->nd_flag & ND_NFSV3) {
1631 		getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1632 		if (!nd->nd_repstat)
1633 			nd->nd_repstat = getret;
1634 	}
1635 
1636 	/*
1637 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
1638 	 */
1639 	if (nd->nd_repstat) {
1640 		vput(vp);
1641 		free((caddr_t)rbuf, M_TEMP);
1642 		if (cookies)
1643 			free((caddr_t)cookies, M_TEMP);
1644 		if (nd->nd_flag & ND_NFSV3)
1645 			nfsrv_postopattr(nd, getret, &at);
1646 		goto out;
1647 	}
1648 	/*
1649 	 * If nothing read, return eof
1650 	 * rpc reply
1651 	 */
1652 	if (siz == 0) {
1653 		vput(vp);
1654 		if (nd->nd_flag & ND_NFSV2) {
1655 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1656 		} else {
1657 			nfsrv_postopattr(nd, getret, &at);
1658 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1659 			txdr_hyper(at.na_filerev, tl);
1660 			tl += 2;
1661 		}
1662 		*tl++ = newnfs_false;
1663 		*tl = newnfs_true;
1664 		FREE((caddr_t)rbuf, M_TEMP);
1665 		FREE((caddr_t)cookies, M_TEMP);
1666 		goto out;
1667 	}
1668 
1669 	/*
1670 	 * Check for degenerate cases of nothing useful read.
1671 	 * If so go try again
1672 	 */
1673 	cpos = rbuf;
1674 	cend = rbuf + siz;
1675 	dp = (struct dirent *)cpos;
1676 	cookiep = cookies;
1677 
1678 	/*
1679 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1680 	 * directory offset up to a block boundary, so it is necessary to
1681 	 * skip over the records that precede the requested offset. This
1682 	 * requires the assumption that file offset cookies monotonically
1683 	 * increase.
1684 	 * Since the offset cookies don't monotonically increase for ZFS,
1685 	 * this is not done when ZFS is the file system.
1686 	 */
1687 	while (cpos < cend && ncookies > 0 &&
1688 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1689 	     (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
1690 		cpos += dp->d_reclen;
1691 		dp = (struct dirent *)cpos;
1692 		cookiep++;
1693 		ncookies--;
1694 	}
1695 	if (cpos >= cend || ncookies == 0) {
1696 		siz = fullsiz;
1697 		toff = off;
1698 		goto again;
1699 	}
1700 	vput(vp);
1701 
1702 	/*
1703 	 * dirlen is the size of the reply, including all XDR and must
1704 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1705 	 * if the XDR should be included in "count", but to be safe, we do.
1706 	 * (Include the two booleans at the end of the reply in dirlen now.)
1707 	 */
1708 	if (nd->nd_flag & ND_NFSV3) {
1709 		nfsrv_postopattr(nd, getret, &at);
1710 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1711 		txdr_hyper(at.na_filerev, tl);
1712 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1713 	} else {
1714 		dirlen = 2 * NFSX_UNSIGNED;
1715 	}
1716 
1717 	/* Loop through the records and build reply */
1718 	while (cpos < cend && ncookies > 0) {
1719 		nlen = dp->d_namlen;
1720 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1721 			nlen <= NFS_MAXNAMLEN) {
1722 			if (nd->nd_flag & ND_NFSV3)
1723 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1724 			else
1725 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1726 			if (dirlen > cnt) {
1727 				eofflag = 0;
1728 				break;
1729 			}
1730 
1731 			/*
1732 			 * Build the directory record xdr from
1733 			 * the dirent entry.
1734 			 */
1735 			if (nd->nd_flag & ND_NFSV3) {
1736 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1737 				*tl++ = newnfs_true;
1738 				*tl++ = 0;
1739 			} else {
1740 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1741 				*tl++ = newnfs_true;
1742 			}
1743 			*tl = txdr_unsigned(dp->d_fileno);
1744 			(void) nfsm_strtom(nd, dp->d_name, nlen);
1745 			if (nd->nd_flag & ND_NFSV3) {
1746 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1747 				*tl++ = 0;
1748 			} else
1749 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1750 			*tl = txdr_unsigned(*cookiep);
1751 		}
1752 		cpos += dp->d_reclen;
1753 		dp = (struct dirent *)cpos;
1754 		cookiep++;
1755 		ncookies--;
1756 	}
1757 	if (cpos < cend)
1758 		eofflag = 0;
1759 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1760 	*tl++ = newnfs_false;
1761 	if (eofflag)
1762 		*tl = newnfs_true;
1763 	else
1764 		*tl = newnfs_false;
1765 	FREE((caddr_t)rbuf, M_TEMP);
1766 	FREE((caddr_t)cookies, M_TEMP);
1767 
1768 out:
1769 	NFSEXITCODE2(0, nd);
1770 	return (0);
1771 nfsmout:
1772 	vput(vp);
1773 	NFSEXITCODE2(error, nd);
1774 	return (error);
1775 }
1776 
1777 /*
1778  * Readdirplus for V3 and Readdir for V4.
1779  */
1780 int
1781 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1782     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1783 {
1784 	struct dirent *dp;
1785 	u_int32_t *tl;
1786 	int dirlen;
1787 	char *cpos, *cend, *rbuf;
1788 	struct vnode *nvp;
1789 	fhandle_t nfh;
1790 	struct nfsvattr nva, at, *nvap = &nva;
1791 	struct mbuf *mb0, *mb1;
1792 	struct nfsreferral *refp;
1793 	int nlen, r, error = 0, getret = 1, usevget = 1;
1794 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1795 	caddr_t bpos0, bpos1;
1796 	u_int64_t off, toff, verf;
1797 	u_long *cookies = NULL, *cookiep;
1798 	nfsattrbit_t attrbits, rderrbits, savbits;
1799 	struct uio io;
1800 	struct iovec iv;
1801 	struct componentname cn;
1802 	int at_root, needs_unbusy, not_zfs, supports_nfsv4acls;
1803 	struct mount *mp, *new_mp;
1804 	uint64_t mounted_on_fileno;
1805 
1806 	if (nd->nd_repstat) {
1807 		nfsrv_postopattr(nd, getret, &at);
1808 		goto out;
1809 	}
1810 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1811 	off = fxdr_hyper(tl);
1812 	toff = off;
1813 	tl += 2;
1814 	verf = fxdr_hyper(tl);
1815 	tl += 2;
1816 	siz = fxdr_unsigned(int, *tl++);
1817 	cnt = fxdr_unsigned(int, *tl);
1818 
1819 	/*
1820 	 * Use the server's maximum data transfer size as the upper bound
1821 	 * on reply datalen.
1822 	 */
1823 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1824 		cnt = NFS_SRVMAXDATA(nd);
1825 
1826 	/*
1827 	 * siz is a "hint" of how much directory information (name, fileid,
1828 	 * cookie) should be in the reply. At least one client "hints" 0,
1829 	 * so I set it to cnt for that case. I also round it up to the
1830 	 * next multiple of DIRBLKSIZ.
1831 	 */
1832 	if (siz <= 0)
1833 		siz = cnt;
1834 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1835 
1836 	if (nd->nd_flag & ND_NFSV4) {
1837 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1838 		if (error)
1839 			goto nfsmout;
1840 		NFSSET_ATTRBIT(&savbits, &attrbits);
1841 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1842 		NFSZERO_ATTRBIT(&rderrbits);
1843 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1844 	} else {
1845 		NFSZERO_ATTRBIT(&attrbits);
1846 	}
1847 	fullsiz = siz;
1848 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1849 	if (!nd->nd_repstat) {
1850 	    if (off && verf != at.na_filerev) {
1851 		/*
1852 		 * va_filerev is not sufficient as a cookie verifier,
1853 		 * since it is not supposed to change when entries are
1854 		 * removed/added unless that offset cookies returned to
1855 		 * the client are no longer valid.
1856 		 */
1857 #if 0
1858 		if (nd->nd_flag & ND_NFSV4) {
1859 			nd->nd_repstat = NFSERR_NOTSAME;
1860 		} else {
1861 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1862 		}
1863 #endif
1864 	    } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1865 		nd->nd_repstat = NFSERR_BAD_COOKIE;
1866 	    }
1867 	}
1868 	if (!nd->nd_repstat && vp->v_type != VDIR)
1869 		nd->nd_repstat = NFSERR_NOTDIR;
1870 	if (!nd->nd_repstat && cnt == 0)
1871 		nd->nd_repstat = NFSERR_TOOSMALL;
1872 	if (!nd->nd_repstat)
1873 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1874 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1875 		    NFSACCCHK_VPISLOCKED, NULL);
1876 	if (nd->nd_repstat) {
1877 		vput(vp);
1878 		if (nd->nd_flag & ND_NFSV3)
1879 			nfsrv_postopattr(nd, getret, &at);
1880 		goto out;
1881 	}
1882 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1883 
1884 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1885 again:
1886 	eofflag = 0;
1887 	if (cookies) {
1888 		free((caddr_t)cookies, M_TEMP);
1889 		cookies = NULL;
1890 	}
1891 
1892 	iv.iov_base = rbuf;
1893 	iv.iov_len = siz;
1894 	io.uio_iov = &iv;
1895 	io.uio_iovcnt = 1;
1896 	io.uio_offset = (off_t)off;
1897 	io.uio_resid = siz;
1898 	io.uio_segflg = UIO_SYSSPACE;
1899 	io.uio_rw = UIO_READ;
1900 	io.uio_td = NULL;
1901 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1902 	    &cookies);
1903 	off = (u_int64_t)io.uio_offset;
1904 	if (io.uio_resid)
1905 		siz -= io.uio_resid;
1906 
1907 	getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1908 
1909 	if (!cookies && !nd->nd_repstat)
1910 		nd->nd_repstat = NFSERR_PERM;
1911 	if (!nd->nd_repstat)
1912 		nd->nd_repstat = getret;
1913 	if (nd->nd_repstat) {
1914 		vput(vp);
1915 		if (cookies)
1916 			free((caddr_t)cookies, M_TEMP);
1917 		free((caddr_t)rbuf, M_TEMP);
1918 		if (nd->nd_flag & ND_NFSV3)
1919 			nfsrv_postopattr(nd, getret, &at);
1920 		goto out;
1921 	}
1922 	/*
1923 	 * If nothing read, return eof
1924 	 * rpc reply
1925 	 */
1926 	if (siz == 0) {
1927 		vput(vp);
1928 		if (nd->nd_flag & ND_NFSV3)
1929 			nfsrv_postopattr(nd, getret, &at);
1930 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1931 		txdr_hyper(at.na_filerev, tl);
1932 		tl += 2;
1933 		*tl++ = newnfs_false;
1934 		*tl = newnfs_true;
1935 		free((caddr_t)cookies, M_TEMP);
1936 		free((caddr_t)rbuf, M_TEMP);
1937 		goto out;
1938 	}
1939 
1940 	/*
1941 	 * Check for degenerate cases of nothing useful read.
1942 	 * If so go try again
1943 	 */
1944 	cpos = rbuf;
1945 	cend = rbuf + siz;
1946 	dp = (struct dirent *)cpos;
1947 	cookiep = cookies;
1948 
1949 	/*
1950 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1951 	 * directory offset up to a block boundary, so it is necessary to
1952 	 * skip over the records that precede the requested offset. This
1953 	 * requires the assumption that file offset cookies monotonically
1954 	 * increase.
1955 	 * Since the offset cookies don't monotonically increase for ZFS,
1956 	 * this is not done when ZFS is the file system.
1957 	 */
1958 	while (cpos < cend && ncookies > 0 &&
1959 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1960 	   (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff) ||
1961 	   ((nd->nd_flag & ND_NFSV4) &&
1962 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1963 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1964 		cpos += dp->d_reclen;
1965 		dp = (struct dirent *)cpos;
1966 		cookiep++;
1967 		ncookies--;
1968 	}
1969 	if (cpos >= cend || ncookies == 0) {
1970 		siz = fullsiz;
1971 		toff = off;
1972 		goto again;
1973 	}
1974 
1975 	/*
1976 	 * Busy the file system so that the mount point won't go away
1977 	 * and, as such, VFS_VGET() can be used safely.
1978 	 */
1979 	mp = vp->v_mount;
1980 	vfs_ref(mp);
1981 	NFSVOPUNLOCK(vp, 0);
1982 	nd->nd_repstat = vfs_busy(mp, 0);
1983 	vfs_rel(mp);
1984 	if (nd->nd_repstat != 0) {
1985 		vrele(vp);
1986 		free(cookies, M_TEMP);
1987 		free(rbuf, M_TEMP);
1988 		if (nd->nd_flag & ND_NFSV3)
1989 			nfsrv_postopattr(nd, getret, &at);
1990 		goto out;
1991 	}
1992 
1993 	/*
1994 	 * Check to see if entries in this directory can be safely acquired
1995 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
1996 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
1997 	 * automount of the snapshot directory that is required will
1998 	 * be done.
1999 	 * This needs to be done here for NFSv4, since NFSv4 never does
2000 	 * a VFS_VGET() for "." or "..".
2001 	 */
2002 	if (not_zfs == 0) {
2003 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2004 		if (r == EOPNOTSUPP) {
2005 			usevget = 0;
2006 			cn.cn_nameiop = LOOKUP;
2007 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2008 			cn.cn_cred = nd->nd_cred;
2009 			cn.cn_thread = p;
2010 		} else if (r == 0)
2011 			vput(nvp);
2012 	}
2013 
2014 	/*
2015 	 * Save this position, in case there is an error before one entry
2016 	 * is created.
2017 	 */
2018 	mb0 = nd->nd_mb;
2019 	bpos0 = nd->nd_bpos;
2020 
2021 	/*
2022 	 * Fill in the first part of the reply.
2023 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2024 	 * (Include the two booleans at the end of the reply in dirlen now,
2025 	 *  so we recognize when we have exceeded cnt.)
2026 	 */
2027 	if (nd->nd_flag & ND_NFSV3) {
2028 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2029 		nfsrv_postopattr(nd, getret, &at);
2030 	} else {
2031 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2032 	}
2033 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2034 	txdr_hyper(at.na_filerev, tl);
2035 
2036 	/*
2037 	 * Save this position, in case there is an empty reply needed.
2038 	 */
2039 	mb1 = nd->nd_mb;
2040 	bpos1 = nd->nd_bpos;
2041 
2042 	/* Loop through the records and build reply */
2043 	entrycnt = 0;
2044 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2045 		nlen = dp->d_namlen;
2046 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2047 		    nlen <= NFS_MAXNAMLEN &&
2048 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2049 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2050 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2051 			/*
2052 			 * Save the current position in the reply, in case
2053 			 * this entry exceeds cnt.
2054 			 */
2055 			mb1 = nd->nd_mb;
2056 			bpos1 = nd->nd_bpos;
2057 
2058 			/*
2059 			 * For readdir_and_lookup get the vnode using
2060 			 * the file number.
2061 			 */
2062 			nvp = NULL;
2063 			refp = NULL;
2064 			r = 0;
2065 			at_root = 0;
2066 			needs_unbusy = 0;
2067 			new_mp = mp;
2068 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2069 			if ((nd->nd_flag & ND_NFSV3) ||
2070 			    NFSNONZERO_ATTRBIT(&savbits)) {
2071 				if (nd->nd_flag & ND_NFSV4)
2072 					refp = nfsv4root_getreferral(NULL,
2073 					    vp, dp->d_fileno);
2074 				if (refp == NULL) {
2075 					if (usevget)
2076 						r = VFS_VGET(mp, dp->d_fileno,
2077 						    LK_SHARED, &nvp);
2078 					else
2079 						r = EOPNOTSUPP;
2080 					if (r == EOPNOTSUPP) {
2081 						if (usevget) {
2082 							usevget = 0;
2083 							cn.cn_nameiop = LOOKUP;
2084 							cn.cn_lkflags =
2085 							    LK_SHARED |
2086 							    LK_RETRY;
2087 							cn.cn_cred =
2088 							    nd->nd_cred;
2089 							cn.cn_thread = p;
2090 						}
2091 						cn.cn_nameptr = dp->d_name;
2092 						cn.cn_namelen = nlen;
2093 						cn.cn_flags = ISLASTCN |
2094 						    NOFOLLOW | LOCKLEAF;
2095 						if (nlen == 2 &&
2096 						    dp->d_name[0] == '.' &&
2097 						    dp->d_name[1] == '.')
2098 							cn.cn_flags |=
2099 							    ISDOTDOT;
2100 						if (NFSVOPLOCK(vp, LK_SHARED)
2101 						    != 0) {
2102 							nd->nd_repstat = EPERM;
2103 							break;
2104 						}
2105 						if ((vp->v_vflag & VV_ROOT) != 0
2106 						    && (cn.cn_flags & ISDOTDOT)
2107 						    != 0) {
2108 							vref(vp);
2109 							nvp = vp;
2110 							r = 0;
2111 						} else {
2112 							r = VOP_LOOKUP(vp, &nvp,
2113 							    &cn);
2114 							if (vp != nvp)
2115 								NFSVOPUNLOCK(vp,
2116 								    0);
2117 						}
2118 					}
2119 
2120 					/*
2121 					 * For NFSv4, check to see if nvp is
2122 					 * a mount point and get the mount
2123 					 * point vnode, as required.
2124 					 */
2125 					if (r == 0 &&
2126 					    nfsrv_enable_crossmntpt != 0 &&
2127 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2128 					    nvp->v_type == VDIR &&
2129 					    nvp->v_mountedhere != NULL) {
2130 						new_mp = nvp->v_mountedhere;
2131 						r = vfs_busy(new_mp, 0);
2132 						vput(nvp);
2133 						nvp = NULL;
2134 						if (r == 0) {
2135 							r = VFS_ROOT(new_mp,
2136 							    LK_SHARED, &nvp);
2137 							needs_unbusy = 1;
2138 							if (r == 0)
2139 								at_root = 1;
2140 						}
2141 					}
2142 				}
2143 				if (!r) {
2144 				    if (refp == NULL &&
2145 					((nd->nd_flag & ND_NFSV3) ||
2146 					 NFSNONZERO_ATTRBIT(&attrbits))) {
2147 					r = nfsvno_getfh(nvp, &nfh, p);
2148 					if (!r)
2149 					    r = nfsvno_getattr(nvp, nvap,
2150 						nd->nd_cred, p, 1);
2151 					if (r == 0 && not_zfs == 0 &&
2152 					    nfsrv_enable_crossmntpt != 0 &&
2153 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2154 					    nvp->v_type == VDIR &&
2155 					    vp->v_mount != nvp->v_mount) {
2156 					    /*
2157 					     * For a ZFS snapshot, there is a
2158 					     * pseudo mount that does not set
2159 					     * v_mountedhere, so it needs to
2160 					     * be detected via a different
2161 					     * mount structure.
2162 					     */
2163 					    at_root = 1;
2164 					    if (new_mp == mp)
2165 						new_mp = nvp->v_mount;
2166 					}
2167 				    }
2168 				} else {
2169 				    nvp = NULL;
2170 				}
2171 				if (r) {
2172 					if (!NFSISSET_ATTRBIT(&attrbits,
2173 					    NFSATTRBIT_RDATTRERROR)) {
2174 						if (nvp != NULL)
2175 							vput(nvp);
2176 						if (needs_unbusy != 0)
2177 							vfs_unbusy(new_mp);
2178 						nd->nd_repstat = r;
2179 						break;
2180 					}
2181 				}
2182 			}
2183 
2184 			/*
2185 			 * Build the directory record xdr
2186 			 */
2187 			if (nd->nd_flag & ND_NFSV3) {
2188 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2189 				*tl++ = newnfs_true;
2190 				*tl++ = 0;
2191 				*tl = txdr_unsigned(dp->d_fileno);
2192 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2193 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2194 				*tl++ = 0;
2195 				*tl = txdr_unsigned(*cookiep);
2196 				nfsrv_postopattr(nd, 0, nvap);
2197 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2198 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2199 				if (nvp != NULL)
2200 					vput(nvp);
2201 			} else {
2202 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2203 				*tl++ = newnfs_true;
2204 				*tl++ = 0;
2205 				*tl = txdr_unsigned(*cookiep);
2206 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2207 				if (nvp != NULL) {
2208 					supports_nfsv4acls =
2209 					    nfs_supportsnfsv4acls(nvp);
2210 					NFSVOPUNLOCK(nvp, 0);
2211 				} else
2212 					supports_nfsv4acls = 0;
2213 				if (refp != NULL) {
2214 					dirlen += nfsrv_putreferralattr(nd,
2215 					    &savbits, refp, 0,
2216 					    &nd->nd_repstat);
2217 					if (nd->nd_repstat) {
2218 						if (nvp != NULL)
2219 							vrele(nvp);
2220 						if (needs_unbusy != 0)
2221 							vfs_unbusy(new_mp);
2222 						break;
2223 					}
2224 				} else if (r) {
2225 					dirlen += nfsvno_fillattr(nd, new_mp,
2226 					    nvp, nvap, &nfh, r, &rderrbits,
2227 					    nd->nd_cred, p, isdgram, 0,
2228 					    supports_nfsv4acls, at_root,
2229 					    mounted_on_fileno);
2230 				} else {
2231 					dirlen += nfsvno_fillattr(nd, new_mp,
2232 					    nvp, nvap, &nfh, r, &attrbits,
2233 					    nd->nd_cred, p, isdgram, 0,
2234 					    supports_nfsv4acls, at_root,
2235 					    mounted_on_fileno);
2236 				}
2237 				if (nvp != NULL)
2238 					vrele(nvp);
2239 				dirlen += (3 * NFSX_UNSIGNED);
2240 			}
2241 			if (needs_unbusy != 0)
2242 				vfs_unbusy(new_mp);
2243 			if (dirlen <= cnt)
2244 				entrycnt++;
2245 		}
2246 		cpos += dp->d_reclen;
2247 		dp = (struct dirent *)cpos;
2248 		cookiep++;
2249 		ncookies--;
2250 	}
2251 	vrele(vp);
2252 	vfs_unbusy(mp);
2253 
2254 	/*
2255 	 * If dirlen > cnt, we must strip off the last entry. If that
2256 	 * results in an empty reply, report NFSERR_TOOSMALL.
2257 	 */
2258 	if (dirlen > cnt || nd->nd_repstat) {
2259 		if (!nd->nd_repstat && entrycnt == 0)
2260 			nd->nd_repstat = NFSERR_TOOSMALL;
2261 		if (nd->nd_repstat)
2262 			newnfs_trimtrailing(nd, mb0, bpos0);
2263 		else
2264 			newnfs_trimtrailing(nd, mb1, bpos1);
2265 		eofflag = 0;
2266 	} else if (cpos < cend)
2267 		eofflag = 0;
2268 	if (!nd->nd_repstat) {
2269 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2270 		*tl++ = newnfs_false;
2271 		if (eofflag)
2272 			*tl = newnfs_true;
2273 		else
2274 			*tl = newnfs_false;
2275 	}
2276 	FREE((caddr_t)cookies, M_TEMP);
2277 	FREE((caddr_t)rbuf, M_TEMP);
2278 
2279 out:
2280 	NFSEXITCODE2(0, nd);
2281 	return (0);
2282 nfsmout:
2283 	vput(vp);
2284 	NFSEXITCODE2(error, nd);
2285 	return (error);
2286 }
2287 
2288 /*
2289  * Get the settable attributes out of the mbuf list.
2290  * (Return 0 or EBADRPC)
2291  */
2292 int
2293 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2294     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2295 {
2296 	u_int32_t *tl;
2297 	struct nfsv2_sattr *sp;
2298 	int error = 0, toclient = 0;
2299 
2300 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2301 	case ND_NFSV2:
2302 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2303 		/*
2304 		 * Some old clients didn't fill in the high order 16bits.
2305 		 * --> check the low order 2 bytes for 0xffff
2306 		 */
2307 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2308 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2309 		if (sp->sa_uid != newnfs_xdrneg1)
2310 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2311 		if (sp->sa_gid != newnfs_xdrneg1)
2312 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2313 		if (sp->sa_size != newnfs_xdrneg1)
2314 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2315 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2316 #ifdef notyet
2317 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2318 #else
2319 			nvap->na_atime.tv_sec =
2320 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2321 			nvap->na_atime.tv_nsec = 0;
2322 #endif
2323 		}
2324 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2325 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2326 		break;
2327 	case ND_NFSV3:
2328 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2329 		if (*tl == newnfs_true) {
2330 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2331 			nvap->na_mode = nfstov_mode(*tl);
2332 		}
2333 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2334 		if (*tl == newnfs_true) {
2335 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2336 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2337 		}
2338 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2339 		if (*tl == newnfs_true) {
2340 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2341 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2342 		}
2343 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2344 		if (*tl == newnfs_true) {
2345 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2346 			nvap->na_size = fxdr_hyper(tl);
2347 		}
2348 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2349 		switch (fxdr_unsigned(int, *tl)) {
2350 		case NFSV3SATTRTIME_TOCLIENT:
2351 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2352 			fxdr_nfsv3time(tl, &nvap->na_atime);
2353 			toclient = 1;
2354 			break;
2355 		case NFSV3SATTRTIME_TOSERVER:
2356 			vfs_timestamp(&nvap->na_atime);
2357 			nvap->na_vaflags |= VA_UTIMES_NULL;
2358 			break;
2359 		};
2360 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2361 		switch (fxdr_unsigned(int, *tl)) {
2362 		case NFSV3SATTRTIME_TOCLIENT:
2363 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2364 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2365 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2366 			break;
2367 		case NFSV3SATTRTIME_TOSERVER:
2368 			vfs_timestamp(&nvap->na_mtime);
2369 			if (!toclient)
2370 				nvap->na_vaflags |= VA_UTIMES_NULL;
2371 			break;
2372 		};
2373 		break;
2374 	case ND_NFSV4:
2375 		error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2376 	};
2377 nfsmout:
2378 	NFSEXITCODE2(error, nd);
2379 	return (error);
2380 }
2381 
2382 /*
2383  * Handle the setable attributes for V4.
2384  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2385  */
2386 int
2387 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2388     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2389 {
2390 	u_int32_t *tl;
2391 	int attrsum = 0;
2392 	int i, j;
2393 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2394 	int toclient = 0;
2395 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2396 	uid_t uid;
2397 	gid_t gid;
2398 
2399 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2400 	if (error)
2401 		goto nfsmout;
2402 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2403 	attrsize = fxdr_unsigned(int, *tl);
2404 
2405 	/*
2406 	 * Loop around getting the setable attributes. If an unsupported
2407 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2408 	 */
2409 	if (retnotsup) {
2410 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2411 		bitpos = NFSATTRBIT_MAX;
2412 	} else {
2413 		bitpos = 0;
2414 	}
2415 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2416 	    if (attrsum > attrsize) {
2417 		error = NFSERR_BADXDR;
2418 		goto nfsmout;
2419 	    }
2420 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2421 		switch (bitpos) {
2422 		case NFSATTRBIT_SIZE:
2423 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2424 			nvap->na_size = fxdr_hyper(tl);
2425 			attrsum += NFSX_HYPER;
2426 			break;
2427 		case NFSATTRBIT_ACL:
2428 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2429 			    p);
2430 			if (error)
2431 				goto nfsmout;
2432 			if (aceerr && !nd->nd_repstat)
2433 				nd->nd_repstat = aceerr;
2434 			attrsum += aclsize;
2435 			break;
2436 		case NFSATTRBIT_ARCHIVE:
2437 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2438 			if (!nd->nd_repstat)
2439 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2440 			attrsum += NFSX_UNSIGNED;
2441 			break;
2442 		case NFSATTRBIT_HIDDEN:
2443 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2444 			if (!nd->nd_repstat)
2445 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2446 			attrsum += NFSX_UNSIGNED;
2447 			break;
2448 		case NFSATTRBIT_MIMETYPE:
2449 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2450 			i = fxdr_unsigned(int, *tl);
2451 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2452 			if (error)
2453 				goto nfsmout;
2454 			if (!nd->nd_repstat)
2455 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2456 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2457 			break;
2458 		case NFSATTRBIT_MODE:
2459 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2460 			nvap->na_mode = nfstov_mode(*tl);
2461 			attrsum += NFSX_UNSIGNED;
2462 			break;
2463 		case NFSATTRBIT_OWNER:
2464 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2465 			j = fxdr_unsigned(int, *tl);
2466 			if (j < 0) {
2467 				error = NFSERR_BADXDR;
2468 				goto nfsmout;
2469 			}
2470 			if (j > NFSV4_SMALLSTR)
2471 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2472 			else
2473 				cp = namestr;
2474 			error = nfsrv_mtostr(nd, cp, j);
2475 			if (error) {
2476 				if (j > NFSV4_SMALLSTR)
2477 					free(cp, M_NFSSTRING);
2478 				goto nfsmout;
2479 			}
2480 			if (!nd->nd_repstat) {
2481 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2482 				    p);
2483 				if (!nd->nd_repstat)
2484 					nvap->na_uid = uid;
2485 			}
2486 			if (j > NFSV4_SMALLSTR)
2487 				free(cp, M_NFSSTRING);
2488 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2489 			break;
2490 		case NFSATTRBIT_OWNERGROUP:
2491 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2492 			j = fxdr_unsigned(int, *tl);
2493 			if (j < 0) {
2494 				error = NFSERR_BADXDR;
2495 				goto nfsmout;
2496 			}
2497 			if (j > NFSV4_SMALLSTR)
2498 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2499 			else
2500 				cp = namestr;
2501 			error = nfsrv_mtostr(nd, cp, j);
2502 			if (error) {
2503 				if (j > NFSV4_SMALLSTR)
2504 					free(cp, M_NFSSTRING);
2505 				goto nfsmout;
2506 			}
2507 			if (!nd->nd_repstat) {
2508 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2509 				    p);
2510 				if (!nd->nd_repstat)
2511 					nvap->na_gid = gid;
2512 			}
2513 			if (j > NFSV4_SMALLSTR)
2514 				free(cp, M_NFSSTRING);
2515 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2516 			break;
2517 		case NFSATTRBIT_SYSTEM:
2518 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2519 			if (!nd->nd_repstat)
2520 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2521 			attrsum += NFSX_UNSIGNED;
2522 			break;
2523 		case NFSATTRBIT_TIMEACCESSSET:
2524 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2525 			attrsum += NFSX_UNSIGNED;
2526 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2527 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2528 			    fxdr_nfsv4time(tl, &nvap->na_atime);
2529 			    toclient = 1;
2530 			    attrsum += NFSX_V4TIME;
2531 			} else {
2532 			    vfs_timestamp(&nvap->na_atime);
2533 			    nvap->na_vaflags |= VA_UTIMES_NULL;
2534 			}
2535 			break;
2536 		case NFSATTRBIT_TIMEBACKUP:
2537 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2538 			if (!nd->nd_repstat)
2539 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2540 			attrsum += NFSX_V4TIME;
2541 			break;
2542 		case NFSATTRBIT_TIMECREATE:
2543 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2544 			if (!nd->nd_repstat)
2545 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2546 			attrsum += NFSX_V4TIME;
2547 			break;
2548 		case NFSATTRBIT_TIMEMODIFYSET:
2549 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2550 			attrsum += NFSX_UNSIGNED;
2551 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2552 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2553 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
2554 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
2555 			    attrsum += NFSX_V4TIME;
2556 			} else {
2557 			    vfs_timestamp(&nvap->na_mtime);
2558 			    if (!toclient)
2559 				nvap->na_vaflags |= VA_UTIMES_NULL;
2560 			}
2561 			break;
2562 		default:
2563 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2564 			/*
2565 			 * set bitpos so we drop out of the loop.
2566 			 */
2567 			bitpos = NFSATTRBIT_MAX;
2568 			break;
2569 		};
2570 	}
2571 
2572 	/*
2573 	 * some clients pad the attrlist, so we need to skip over the
2574 	 * padding.
2575 	 */
2576 	if (attrsum > attrsize) {
2577 		error = NFSERR_BADXDR;
2578 	} else {
2579 		attrsize = NFSM_RNDUP(attrsize);
2580 		if (attrsum < attrsize)
2581 			error = nfsm_advance(nd, attrsize - attrsum, -1);
2582 	}
2583 nfsmout:
2584 	NFSEXITCODE2(error, nd);
2585 	return (error);
2586 }
2587 
2588 /*
2589  * Check/setup export credentials.
2590  */
2591 int
2592 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2593     struct ucred *credanon)
2594 {
2595 	int error = 0;
2596 
2597 	/*
2598 	 * Check/setup credentials.
2599 	 */
2600 	if (nd->nd_flag & ND_GSS)
2601 		exp->nes_exflag &= ~MNT_EXPORTANON;
2602 
2603 	/*
2604 	 * Check to see if the operation is allowed for this security flavor.
2605 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2606 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2607 	 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2608 	 */
2609 	if (nfsvno_testexp(nd, exp) &&
2610 	    nd->nd_procnum != NFSV4OP_SECINFO &&
2611 	    nd->nd_procnum != NFSPROC_FSINFO) {
2612 		if (nd->nd_flag & ND_NFSV4)
2613 			error = NFSERR_WRONGSEC;
2614 		else
2615 			error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2616 		goto out;
2617 	}
2618 
2619 	/*
2620 	 * Check to see if the file system is exported V4 only.
2621 	 */
2622 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2623 		error = NFSERR_PROGNOTV4;
2624 		goto out;
2625 	}
2626 
2627 	/*
2628 	 * Now, map the user credentials.
2629 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
2630 	 *  Fsinfo RPC. If set for anything else, this code might need
2631 	 *  to change.)
2632 	 */
2633 	if (NFSVNO_EXPORTED(exp) &&
2634 	    ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2635 	     NFSVNO_EXPORTANON(exp) ||
2636 	     (nd->nd_flag & ND_AUTHNONE))) {
2637 		nd->nd_cred->cr_uid = credanon->cr_uid;
2638 		nd->nd_cred->cr_gid = credanon->cr_gid;
2639 		crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2640 		    credanon->cr_groups);
2641 	}
2642 
2643 out:
2644 	NFSEXITCODE2(error, nd);
2645 	return (error);
2646 }
2647 
2648 /*
2649  * Check exports.
2650  */
2651 int
2652 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2653     struct ucred **credp)
2654 {
2655 	int i, error, *secflavors;
2656 
2657 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2658 	    &exp->nes_numsecflavor, &secflavors);
2659 	if (error) {
2660 		if (nfs_rootfhset) {
2661 			exp->nes_exflag = 0;
2662 			exp->nes_numsecflavor = 0;
2663 			error = 0;
2664 		}
2665 	} else {
2666 		/* Copy the security flavors. */
2667 		for (i = 0; i < exp->nes_numsecflavor; i++)
2668 			exp->nes_secflavors[i] = secflavors[i];
2669 	}
2670 	NFSEXITCODE(error);
2671 	return (error);
2672 }
2673 
2674 /*
2675  * Get a vnode for a file handle and export stuff.
2676  */
2677 int
2678 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2679     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2680     struct ucred **credp)
2681 {
2682 	int i, error, *secflavors;
2683 
2684 	*credp = NULL;
2685 	exp->nes_numsecflavor = 0;
2686 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2687 	if (error != 0)
2688 		/* Make sure the server replies ESTALE to the client. */
2689 		error = ESTALE;
2690 	if (nam && !error) {
2691 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2692 		    &exp->nes_numsecflavor, &secflavors);
2693 		if (error) {
2694 			if (nfs_rootfhset) {
2695 				exp->nes_exflag = 0;
2696 				exp->nes_numsecflavor = 0;
2697 				error = 0;
2698 			} else {
2699 				vput(*vpp);
2700 			}
2701 		} else {
2702 			/* Copy the security flavors. */
2703 			for (i = 0; i < exp->nes_numsecflavor; i++)
2704 				exp->nes_secflavors[i] = secflavors[i];
2705 		}
2706 	}
2707 	NFSEXITCODE(error);
2708 	return (error);
2709 }
2710 
2711 /*
2712  * nfsd_fhtovp() - convert a fh to a vnode ptr
2713  * 	- look up fsid in mount list (if not found ret error)
2714  *	- get vp and export rights by calling nfsvno_fhtovp()
2715  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2716  *	  for AUTH_SYS
2717  *	- if mpp != NULL, return the mount point so that it can
2718  *	  be used for vn_finished_write() by the caller
2719  */
2720 void
2721 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2722     struct vnode **vpp, struct nfsexstuff *exp,
2723     struct mount **mpp, int startwrite, struct thread *p)
2724 {
2725 	struct mount *mp;
2726 	struct ucred *credanon;
2727 	fhandle_t *fhp;
2728 
2729 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
2730 	/*
2731 	 * Check for the special case of the nfsv4root_fh.
2732 	 */
2733 	mp = vfs_busyfs(&fhp->fh_fsid);
2734 	if (mpp != NULL)
2735 		*mpp = mp;
2736 	if (mp == NULL) {
2737 		*vpp = NULL;
2738 		nd->nd_repstat = ESTALE;
2739 		goto out;
2740 	}
2741 
2742 	if (startwrite) {
2743 		vn_start_write(NULL, mpp, V_WAIT);
2744 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
2745 			lktype = LK_EXCLUSIVE;
2746 	}
2747 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2748 	    &credanon);
2749 	vfs_unbusy(mp);
2750 
2751 	/*
2752 	 * For NFSv4 without a pseudo root fs, unexported file handles
2753 	 * can be returned, so that Lookup works everywhere.
2754 	 */
2755 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2756 	    !(nd->nd_flag & ND_NFSV4)) {
2757 		vput(*vpp);
2758 		nd->nd_repstat = EACCES;
2759 	}
2760 
2761 	/*
2762 	 * Personally, I've never seen any point in requiring a
2763 	 * reserved port#, since only in the rare case where the
2764 	 * clients are all boxes with secure system priviledges,
2765 	 * does it provide any enhanced security, but... some people
2766 	 * believe it to be useful and keep putting this code back in.
2767 	 * (There is also some "security checker" out there that
2768 	 *  complains if the nfs server doesn't enforce this.)
2769 	 * However, note the following:
2770 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
2771 	 *	required.
2772 	 * RFC2623 recommends that, if a reserved port# is checked for,
2773 	 *	that there be a way to turn that off--> ifdef'd.
2774 	 */
2775 #ifdef NFS_REQRSVPORT
2776 	if (!nd->nd_repstat) {
2777 		struct sockaddr_in *saddr;
2778 		struct sockaddr_in6 *saddr6;
2779 
2780 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2781 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2782 		if (!(nd->nd_flag & ND_NFSV4) &&
2783 		    ((saddr->sin_family == AF_INET &&
2784 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2785 		     (saddr6->sin6_family == AF_INET6 &&
2786 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2787 			vput(*vpp);
2788 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2789 		}
2790 	}
2791 #endif	/* NFS_REQRSVPORT */
2792 
2793 	/*
2794 	 * Check/setup credentials.
2795 	 */
2796 	if (!nd->nd_repstat) {
2797 		nd->nd_saveduid = nd->nd_cred->cr_uid;
2798 		nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2799 		if (nd->nd_repstat)
2800 			vput(*vpp);
2801 	}
2802 	if (credanon != NULL)
2803 		crfree(credanon);
2804 	if (nd->nd_repstat) {
2805 		if (startwrite)
2806 			vn_finished_write(mp);
2807 		*vpp = NULL;
2808 		if (mpp != NULL)
2809 			*mpp = NULL;
2810 	}
2811 
2812 out:
2813 	NFSEXITCODE2(0, nd);
2814 }
2815 
2816 /*
2817  * glue for fp.
2818  */
2819 static int
2820 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2821 {
2822 	struct filedesc *fdp;
2823 	struct file *fp;
2824 	int error = 0;
2825 
2826 	fdp = p->td_proc->p_fd;
2827 	if (fd < 0 || fd >= fdp->fd_nfiles ||
2828 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
2829 		error = EBADF;
2830 		goto out;
2831 	}
2832 	*fpp = fp;
2833 
2834 out:
2835 	NFSEXITCODE(error);
2836 	return (error);
2837 }
2838 
2839 /*
2840  * Called from nfssvc() to update the exports list. Just call
2841  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2842  * in the mount list.
2843  */
2844 int
2845 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2846 {
2847 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2848 	int error = 0;
2849 	struct nameidata nd;
2850 	fhandle_t fh;
2851 
2852 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2853 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2854 		nfs_rootfhset = 0;
2855 	else if (error == 0) {
2856 		if (nfsexargp->fspec == NULL) {
2857 			error = EPERM;
2858 			goto out;
2859 		}
2860 		/*
2861 		 * If fspec != NULL, this is the v4root path.
2862 		 */
2863 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
2864 		    nfsexargp->fspec, p);
2865 		if ((error = namei(&nd)) != 0)
2866 			goto out;
2867 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
2868 		vrele(nd.ni_vp);
2869 		if (!error) {
2870 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2871 			NFSBCOPY((caddr_t)&fh,
2872 			    nfs_rootfh.nfsrvfh_data,
2873 			    sizeof (fhandle_t));
2874 			nfs_rootfhset = 1;
2875 		}
2876 	}
2877 
2878 out:
2879 	NFSEXITCODE(error);
2880 	return (error);
2881 }
2882 
2883 /*
2884  * Get the tcp socket sequence numbers we need.
2885  * (Maybe this should be moved to the tcp sources?)
2886  */
2887 int
2888 nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap)
2889 {
2890 	struct inpcb *inp;
2891 	struct tcpcb *tp;
2892 	int error = 0;
2893 
2894 	inp = sotoinpcb(so);
2895 	KASSERT(inp != NULL, ("nfsrv_getsocksndseq: inp == NULL"));
2896 	INP_RLOCK(inp);
2897 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2898 		INP_RUNLOCK(inp);
2899 		error = EPIPE;
2900 		goto out;
2901 	}
2902 	tp = intotcpcb(inp);
2903 	if (tp->t_state != TCPS_ESTABLISHED) {
2904 		INP_RUNLOCK(inp);
2905 		error = EPIPE;
2906 		goto out;
2907 	}
2908 	*maxp = tp->snd_max;
2909 	*unap = tp->snd_una;
2910 	INP_RUNLOCK(inp);
2911 
2912 out:
2913 	NFSEXITCODE(error);
2914 	return (error);
2915 }
2916 
2917 /*
2918  * This function needs to test to see if the system is near its limit
2919  * for memory allocation via malloc() or mget() and return True iff
2920  * either of these resources are near their limit.
2921  * XXX (For now, this is just a stub.)
2922  */
2923 int nfsrv_testmalloclimit = 0;
2924 int
2925 nfsrv_mallocmget_limit(void)
2926 {
2927 	static int printmesg = 0;
2928 	static int testval = 1;
2929 
2930 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2931 		if ((printmesg++ % 100) == 0)
2932 			printf("nfsd: malloc/mget near limit\n");
2933 		return (1);
2934 	}
2935 	return (0);
2936 }
2937 
2938 /*
2939  * BSD specific initialization of a mount point.
2940  */
2941 void
2942 nfsd_mntinit(void)
2943 {
2944 	static int inited = 0;
2945 
2946 	if (inited)
2947 		return;
2948 	inited = 1;
2949 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2950 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2951 	TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2952 	nfsv4root_mnt.mnt_export = NULL;
2953 	TAILQ_INIT(&nfsv4root_opt);
2954 	TAILQ_INIT(&nfsv4root_newopt);
2955 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2956 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2957 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
2958 	nfsv4root_mnt.mnt_activevnodelistsize = 0;
2959 }
2960 
2961 /*
2962  * Get a vnode for a file handle, without checking exports, etc.
2963  */
2964 struct vnode *
2965 nfsvno_getvp(fhandle_t *fhp)
2966 {
2967 	struct mount *mp;
2968 	struct vnode *vp;
2969 	int error;
2970 
2971 	mp = vfs_busyfs(&fhp->fh_fsid);
2972 	if (mp == NULL)
2973 		return (NULL);
2974 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2975 	vfs_unbusy(mp);
2976 	if (error)
2977 		return (NULL);
2978 	return (vp);
2979 }
2980 
2981 /*
2982  * Do a local VOP_ADVLOCK().
2983  */
2984 int
2985 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2986     u_int64_t end, struct thread *td)
2987 {
2988 	int error = 0;
2989 	struct flock fl;
2990 	u_int64_t tlen;
2991 
2992 	if (nfsrv_dolocallocks == 0)
2993 		goto out;
2994 
2995 	/* Check for VI_DOOMED here, so that VOP_ADVLOCK() isn't performed. */
2996 	if ((vp->v_iflag & VI_DOOMED) != 0) {
2997 		error = EPERM;
2998 		goto out;
2999 	}
3000 
3001 	fl.l_whence = SEEK_SET;
3002 	fl.l_type = ftype;
3003 	fl.l_start = (off_t)first;
3004 	if (end == NFS64BITSSET) {
3005 		fl.l_len = 0;
3006 	} else {
3007 		tlen = end - first;
3008 		fl.l_len = (off_t)tlen;
3009 	}
3010 	/*
3011 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3012 	 * values for all calls, so that all locks will be held by the
3013 	 * nfsd server. (The nfsd server handles conflicts between the
3014 	 * various clients.)
3015 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3016 	 * bytes, so it can't be put in l_sysid.
3017 	 */
3018 	if (nfsv4_sysid == 0)
3019 		nfsv4_sysid = nlm_acquire_next_sysid();
3020 	fl.l_pid = (pid_t)0;
3021 	fl.l_sysid = (int)nfsv4_sysid;
3022 
3023 	NFSVOPUNLOCK(vp, 0);
3024 	if (ftype == F_UNLCK)
3025 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3026 		    (F_POSIX | F_REMOTE));
3027 	else
3028 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3029 		    (F_POSIX | F_REMOTE));
3030 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3031 
3032 out:
3033 	NFSEXITCODE(error);
3034 	return (error);
3035 }
3036 
3037 /*
3038  * Check the nfsv4 root exports.
3039  */
3040 int
3041 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3042 {
3043 	struct ucred *credanon;
3044 	int exflags, error = 0, numsecflavor, *secflavors, i;
3045 
3046 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3047 	    &credanon, &numsecflavor, &secflavors);
3048 	if (error) {
3049 		error = NFSERR_PROGUNAVAIL;
3050 		goto out;
3051 	}
3052 	if (credanon != NULL)
3053 		crfree(credanon);
3054 	for (i = 0; i < numsecflavor; i++) {
3055 		if (secflavors[i] == AUTH_SYS)
3056 			nd->nd_flag |= ND_EXAUTHSYS;
3057 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3058 			nd->nd_flag |= ND_EXGSS;
3059 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3060 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3061 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3062 			nd->nd_flag |= ND_EXGSSPRIVACY;
3063 	}
3064 
3065 out:
3066 	NFSEXITCODE(error);
3067 	return (error);
3068 }
3069 
3070 /*
3071  * Nfs server psuedo system call for the nfsd's
3072  */
3073 /*
3074  * MPSAFE
3075  */
3076 static int
3077 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3078 {
3079 	struct file *fp;
3080 	struct nfsd_addsock_args sockarg;
3081 	struct nfsd_nfsd_args nfsdarg;
3082 	cap_rights_t rights;
3083 	int error;
3084 
3085 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3086 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3087 		if (error)
3088 			goto out;
3089 		/*
3090 		 * Since we don't know what rights might be required,
3091 		 * pretend that we need them all. It is better to be too
3092 		 * careful than too reckless.
3093 		 */
3094 		error = fget(td, sockarg.sock,
3095 		    cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3096 		if (error != 0)
3097 			goto out;
3098 		if (fp->f_type != DTYPE_SOCKET) {
3099 			fdrop(fp, td);
3100 			error = EPERM;
3101 			goto out;
3102 		}
3103 		error = nfsrvd_addsock(fp);
3104 		fdrop(fp, td);
3105 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3106 		if (uap->argp == NULL) {
3107 			error = EINVAL;
3108 			goto out;
3109 		}
3110 		error = copyin(uap->argp, (caddr_t)&nfsdarg,
3111 		    sizeof (nfsdarg));
3112 		if (error)
3113 			goto out;
3114 		error = nfsrvd_nfsd(td, &nfsdarg);
3115 	} else {
3116 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3117 	}
3118 
3119 out:
3120 	NFSEXITCODE(error);
3121 	return (error);
3122 }
3123 
3124 static int
3125 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3126 {
3127 	struct nfsex_args export;
3128 	struct file *fp = NULL;
3129 	int stablefd, len;
3130 	struct nfsd_clid adminrevoke;
3131 	struct nfsd_dumplist dumplist;
3132 	struct nfsd_dumpclients *dumpclients;
3133 	struct nfsd_dumplocklist dumplocklist;
3134 	struct nfsd_dumplocks *dumplocks;
3135 	struct nameidata nd;
3136 	vnode_t vp;
3137 	int error = EINVAL, igotlock;
3138 	struct proc *procp;
3139 	static int suspend_nfsd = 0;
3140 
3141 	if (uap->flag & NFSSVC_PUBLICFH) {
3142 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3143 		    sizeof (fhandle_t));
3144 		error = copyin(uap->argp,
3145 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3146 		if (!error)
3147 			nfs_pubfhset = 1;
3148 	} else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3149 		error = copyin(uap->argp,(caddr_t)&export,
3150 		    sizeof (struct nfsex_args));
3151 		if (!error)
3152 			error = nfsrv_v4rootexport(&export, cred, p);
3153 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3154 		nfs_pubfhset = 0;
3155 		error = 0;
3156 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3157 		error = copyin(uap->argp, (caddr_t)&stablefd,
3158 		    sizeof (int));
3159 		if (!error)
3160 			error = fp_getfvp(p, stablefd, &fp, &vp);
3161 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3162 			error = EBADF;
3163 		if (!error && newnfs_numnfsd != 0)
3164 			error = EPERM;
3165 		if (!error) {
3166 			nfsrv_stablefirst.nsf_fp = fp;
3167 			nfsrv_setupstable(p);
3168 		}
3169 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3170 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3171 		    sizeof (struct nfsd_clid));
3172 		if (!error)
3173 			error = nfsrv_adminrevoke(&adminrevoke, p);
3174 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3175 		error = copyin(uap->argp, (caddr_t)&dumplist,
3176 		    sizeof (struct nfsd_dumplist));
3177 		if (!error && (dumplist.ndl_size < 1 ||
3178 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3179 			error = EPERM;
3180 		if (!error) {
3181 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3182 		    dumpclients = (struct nfsd_dumpclients *)malloc(len,
3183 			M_TEMP, M_WAITOK);
3184 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3185 		    error = copyout(dumpclients,
3186 			CAST_USER_ADDR_T(dumplist.ndl_list), len);
3187 		    free((caddr_t)dumpclients, M_TEMP);
3188 		}
3189 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3190 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3191 		    sizeof (struct nfsd_dumplocklist));
3192 		if (!error && (dumplocklist.ndllck_size < 1 ||
3193 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3194 			error = EPERM;
3195 		if (!error)
3196 			error = nfsrv_lookupfilename(&nd,
3197 				dumplocklist.ndllck_fname, p);
3198 		if (!error) {
3199 			len = sizeof (struct nfsd_dumplocks) *
3200 				dumplocklist.ndllck_size;
3201 			dumplocks = (struct nfsd_dumplocks *)malloc(len,
3202 				M_TEMP, M_WAITOK);
3203 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3204 			    dumplocklist.ndllck_size, p);
3205 			vput(nd.ni_vp);
3206 			error = copyout(dumplocks,
3207 			    CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3208 			free((caddr_t)dumplocks, M_TEMP);
3209 		}
3210 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3211 		procp = p->td_proc;
3212 		PROC_LOCK(procp);
3213 		nfsd_master_pid = procp->p_pid;
3214 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3215 		nfsd_master_start = procp->p_stats->p_start;
3216 		nfsd_master_proc = procp;
3217 		PROC_UNLOCK(procp);
3218 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3219 		NFSLOCKV4ROOTMUTEX();
3220 		if (suspend_nfsd == 0) {
3221 			/* Lock out all nfsd threads */
3222 			do {
3223 				igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3224 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3225 			} while (igotlock == 0 && suspend_nfsd == 0);
3226 			suspend_nfsd = 1;
3227 		}
3228 		NFSUNLOCKV4ROOTMUTEX();
3229 		error = 0;
3230 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3231 		NFSLOCKV4ROOTMUTEX();
3232 		if (suspend_nfsd != 0) {
3233 			nfsv4_unlock(&nfsd_suspend_lock, 0);
3234 			suspend_nfsd = 0;
3235 		}
3236 		NFSUNLOCKV4ROOTMUTEX();
3237 		error = 0;
3238 	}
3239 
3240 	NFSEXITCODE(error);
3241 	return (error);
3242 }
3243 
3244 /*
3245  * Check exports.
3246  * Returns 0 if ok, 1 otherwise.
3247  */
3248 int
3249 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3250 {
3251 	int i;
3252 
3253 	/*
3254 	 * This seems odd, but allow the case where the security flavor
3255 	 * list is empty. This happens when NFSv4 is traversing non-exported
3256 	 * file systems. Exported file systems should always have a non-empty
3257 	 * security flavor list.
3258 	 */
3259 	if (exp->nes_numsecflavor == 0)
3260 		return (0);
3261 
3262 	for (i = 0; i < exp->nes_numsecflavor; i++) {
3263 		/*
3264 		 * The tests for privacy and integrity must be first,
3265 		 * since ND_GSS is set for everything but AUTH_SYS.
3266 		 */
3267 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3268 		    (nd->nd_flag & ND_GSSPRIVACY))
3269 			return (0);
3270 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3271 		    (nd->nd_flag & ND_GSSINTEGRITY))
3272 			return (0);
3273 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3274 		    (nd->nd_flag & ND_GSS))
3275 			return (0);
3276 		if (exp->nes_secflavors[i] == AUTH_SYS &&
3277 		    (nd->nd_flag & ND_GSS) == 0)
3278 			return (0);
3279 	}
3280 	return (1);
3281 }
3282 
3283 /*
3284  * Calculate a hash value for the fid in a file handle.
3285  */
3286 uint32_t
3287 nfsrv_hashfh(fhandle_t *fhp)
3288 {
3289 	uint32_t hashval;
3290 
3291 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3292 	return (hashval);
3293 }
3294 
3295 /*
3296  * Signal the userland master nfsd to backup the stable restart file.
3297  */
3298 void
3299 nfsrv_backupstable(void)
3300 {
3301 	struct proc *procp;
3302 
3303 	if (nfsd_master_proc != NULL) {
3304 		procp = pfind(nfsd_master_pid);
3305 		/* Try to make sure it is the correct process. */
3306 		if (procp == nfsd_master_proc &&
3307 		    procp->p_stats->p_start.tv_sec ==
3308 		    nfsd_master_start.tv_sec &&
3309 		    procp->p_stats->p_start.tv_usec ==
3310 		    nfsd_master_start.tv_usec &&
3311 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
3312 			kern_psignal(procp, SIGUSR2);
3313 		else
3314 			nfsd_master_proc = NULL;
3315 
3316 		if (procp != NULL)
3317 			PROC_UNLOCK(procp);
3318 	}
3319 }
3320 
3321 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3322 
3323 /*
3324  * Called once to initialize data structures...
3325  */
3326 static int
3327 nfsd_modevent(module_t mod, int type, void *data)
3328 {
3329 	int error = 0, i;
3330 	static int loaded = 0;
3331 
3332 	switch (type) {
3333 	case MOD_LOAD:
3334 		if (loaded)
3335 			goto out;
3336 		newnfs_portinit();
3337 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3338 			snprintf(nfsrchash_table[i].lock_name,
3339 			    sizeof(nfsrchash_table[i].lock_name), "nfsrc_tcp%d",
3340 			    i);
3341 			mtx_init(&nfsrchash_table[i].mtx,
3342 			    nfsrchash_table[i].lock_name, NULL, MTX_DEF);
3343 		}
3344 		mtx_init(&nfsrc_udpmtx, "nfs_udpcache_mutex", NULL, MTX_DEF);
3345 		mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3346 		mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3347 		    MTX_DEF);
3348 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3349 		nfsrvd_initcache();
3350 		nfsd_init();
3351 		NFSD_LOCK();
3352 		nfsrvd_init(0);
3353 		NFSD_UNLOCK();
3354 		nfsd_mntinit();
3355 #ifdef VV_DISABLEDELEG
3356 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3357 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3358 #endif
3359 		nfsd_call_servertimer = nfsrv_servertimer;
3360 		nfsd_call_nfsd = nfssvc_nfsd;
3361 		loaded = 1;
3362 		break;
3363 
3364 	case MOD_UNLOAD:
3365 		if (newnfs_numnfsd != 0) {
3366 			error = EBUSY;
3367 			break;
3368 		}
3369 
3370 #ifdef VV_DISABLEDELEG
3371 		vn_deleg_ops.vndeleg_recall = NULL;
3372 		vn_deleg_ops.vndeleg_disable = NULL;
3373 #endif
3374 		nfsd_call_servertimer = NULL;
3375 		nfsd_call_nfsd = NULL;
3376 
3377 		/* Clean out all NFSv4 state. */
3378 		nfsrv_throwawayallstate(curthread);
3379 
3380 		/* Clean the NFS server reply cache */
3381 		nfsrvd_cleancache();
3382 
3383 		/* Free up the krpc server pool. */
3384 		if (nfsrvd_pool != NULL)
3385 			svcpool_destroy(nfsrvd_pool);
3386 
3387 		/* and get rid of the locks */
3388 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++)
3389 			mtx_destroy(&nfsrchash_table[i].mtx);
3390 		mtx_destroy(&nfsrc_udpmtx);
3391 		mtx_destroy(&nfs_v4root_mutex);
3392 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3393 		lockdestroy(&nfsv4root_mnt.mnt_explock);
3394 		loaded = 0;
3395 		break;
3396 	default:
3397 		error = EOPNOTSUPP;
3398 		break;
3399 	}
3400 
3401 out:
3402 	NFSEXITCODE(error);
3403 	return (error);
3404 }
3405 static moduledata_t nfsd_mod = {
3406 	"nfsd",
3407 	nfsd_modevent,
3408 	NULL,
3409 };
3410 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3411 
3412 /* So that loader and kldload(2) can find us, wherever we are.. */
3413 MODULE_VERSION(nfsd, 1);
3414 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3415 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3416 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3417 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3418 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3419 
3420