xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision 0572ccaa4543b0abef8ef81e384c1d04de9f3da1)
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/capsicum.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 nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
65 struct mtx nfsrc_udpmtx;
66 struct mtx nfs_v4root_mutex;
67 struct nfsrvfh nfs_rootfh, nfs_pubfh;
68 int nfs_pubfhset = 0, nfs_rootfhset = 0;
69 struct proc *nfsd_master_proc = NULL;
70 static pid_t nfsd_master_pid = (pid_t)-1;
71 static char nfsd_master_comm[MAXCOMLEN + 1];
72 static struct timeval nfsd_master_start;
73 static uint32_t nfsv4_sysid = 0;
74 
75 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
76     struct ucred *);
77 
78 int nfsrv_enable_crossmntpt = 1;
79 static int nfs_commit_blks;
80 static int nfs_commit_miss;
81 extern int nfsrv_issuedelegs;
82 extern int nfsrv_dolocallocks;
83 extern int nfsd_enable_stringtouid;
84 
85 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server");
86 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
87     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
88 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
89     0, "");
90 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
91     0, "");
92 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
93     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
94 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
95     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
96 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
97     &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
98 
99 #define	MAX_REORDERED_RPC	16
100 #define	NUM_HEURISTIC		1031
101 #define	NHUSE_INIT		64
102 #define	NHUSE_INC		16
103 #define	NHUSE_MAX		2048
104 
105 static struct nfsheur {
106 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
107 	off_t nh_nextoff;	/* next offset for sequential detection */
108 	int nh_use;		/* use count for selection */
109 	int nh_seqcount;	/* heuristic */
110 } nfsheur[NUM_HEURISTIC];
111 
112 
113 /*
114  * Heuristic to detect sequential operation.
115  */
116 static struct nfsheur *
117 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
118 {
119 	struct nfsheur *nh;
120 	int hi, try;
121 
122 	/* Locate best candidate. */
123 	try = 32;
124 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
125 	nh = &nfsheur[hi];
126 	while (try--) {
127 		if (nfsheur[hi].nh_vp == vp) {
128 			nh = &nfsheur[hi];
129 			break;
130 		}
131 		if (nfsheur[hi].nh_use > 0)
132 			--nfsheur[hi].nh_use;
133 		hi = (hi + 1) % NUM_HEURISTIC;
134 		if (nfsheur[hi].nh_use < nh->nh_use)
135 			nh = &nfsheur[hi];
136 	}
137 
138 	/* Initialize hint if this is a new file. */
139 	if (nh->nh_vp != vp) {
140 		nh->nh_vp = vp;
141 		nh->nh_nextoff = uio->uio_offset;
142 		nh->nh_use = NHUSE_INIT;
143 		if (uio->uio_offset == 0)
144 			nh->nh_seqcount = 4;
145 		else
146 			nh->nh_seqcount = 1;
147 	}
148 
149 	/* Calculate heuristic. */
150 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
151 	    uio->uio_offset == nh->nh_nextoff) {
152 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
153 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
154 		if (nh->nh_seqcount > IO_SEQMAX)
155 			nh->nh_seqcount = IO_SEQMAX;
156 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
157 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
158 		/* Probably a reordered RPC, leave seqcount alone. */
159 	} else if (nh->nh_seqcount > 1) {
160 		nh->nh_seqcount /= 2;
161 	} else {
162 		nh->nh_seqcount = 0;
163 	}
164 	nh->nh_use += NHUSE_INC;
165 	if (nh->nh_use > NHUSE_MAX)
166 		nh->nh_use = NHUSE_MAX;
167 	return (nh);
168 }
169 
170 /*
171  * Get attributes into nfsvattr structure.
172  */
173 int
174 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
175     struct thread *p, int vpislocked)
176 {
177 	int error, lockedit = 0;
178 
179 	if (vpislocked == 0) {
180 		/*
181 		 * When vpislocked == 0, the vnode is either exclusively
182 		 * locked by this thread or not locked by this thread.
183 		 * As such, shared lock it, if not exclusively locked.
184 		 */
185 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
186 			lockedit = 1;
187 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
188 		}
189 	}
190 	error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
191 	if (lockedit != 0)
192 		NFSVOPUNLOCK(vp, 0);
193 
194 	NFSEXITCODE(error);
195 	return (error);
196 }
197 
198 /*
199  * Get a file handle for a vnode.
200  */
201 int
202 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
203 {
204 	int error;
205 
206 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
207 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
208 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
209 
210 	NFSEXITCODE(error);
211 	return (error);
212 }
213 
214 /*
215  * Perform access checking for vnodes obtained from file handles that would
216  * refer to files already opened by a Unix client. You cannot just use
217  * vn_writechk() and VOP_ACCESSX() for two reasons.
218  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
219  *     case.
220  * 2 - The owner is to be given access irrespective of mode bits for some
221  *     operations, so that processes that chmod after opening a file don't
222  *     break.
223  */
224 int
225 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
226     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
227     u_int32_t *supportedtypep)
228 {
229 	struct vattr vattr;
230 	int error = 0, getret = 0;
231 
232 	if (vpislocked == 0) {
233 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
234 			error = EPERM;
235 			goto out;
236 		}
237 	}
238 	if (accmode & VWRITE) {
239 		/* Just vn_writechk() changed to check rdonly */
240 		/*
241 		 * Disallow write attempts on read-only file systems;
242 		 * unless the file is a socket or a block or character
243 		 * device resident on the file system.
244 		 */
245 		if (NFSVNO_EXRDONLY(exp) ||
246 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
247 			switch (vp->v_type) {
248 			case VREG:
249 			case VDIR:
250 			case VLNK:
251 				error = EROFS;
252 			default:
253 				break;
254 			}
255 		}
256 		/*
257 		 * If there's shared text associated with
258 		 * the inode, try to free it up once.  If
259 		 * we fail, we can't allow writing.
260 		 */
261 		if (VOP_IS_TEXT(vp) && error == 0)
262 			error = ETXTBSY;
263 	}
264 	if (error != 0) {
265 		if (vpislocked == 0)
266 			NFSVOPUNLOCK(vp, 0);
267 		goto out;
268 	}
269 
270 	/*
271 	 * Should the override still be applied when ACLs are enabled?
272 	 */
273 	error = VOP_ACCESSX(vp, accmode, cred, p);
274 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
275 		/*
276 		 * Try again with VEXPLICIT_DENY, to see if the test for
277 		 * deletion is supported.
278 		 */
279 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
280 		if (error == 0) {
281 			if (vp->v_type == VDIR) {
282 				accmode &= ~(VDELETE | VDELETE_CHILD);
283 				accmode |= VWRITE;
284 				error = VOP_ACCESSX(vp, accmode, cred, p);
285 			} else if (supportedtypep != NULL) {
286 				*supportedtypep &= ~NFSACCESS_DELETE;
287 			}
288 		}
289 	}
290 
291 	/*
292 	 * Allow certain operations for the owner (reads and writes
293 	 * on files that are already open).
294 	 */
295 	if (override != NFSACCCHK_NOOVERRIDE &&
296 	    (error == EPERM || error == EACCES)) {
297 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
298 			error = 0;
299 		else if (override & NFSACCCHK_ALLOWOWNER) {
300 			getret = VOP_GETATTR(vp, &vattr, cred);
301 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
302 				error = 0;
303 		}
304 	}
305 	if (vpislocked == 0)
306 		NFSVOPUNLOCK(vp, 0);
307 
308 out:
309 	NFSEXITCODE(error);
310 	return (error);
311 }
312 
313 /*
314  * Set attribute(s) vnop.
315  */
316 int
317 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
318     struct thread *p, struct nfsexstuff *exp)
319 {
320 	int error;
321 
322 	error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
323 	NFSEXITCODE(error);
324 	return (error);
325 }
326 
327 /*
328  * Set up nameidata for a lookup() call and do it.
329  */
330 int
331 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
332     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
333     struct vnode **retdirp)
334 {
335 	struct componentname *cnp = &ndp->ni_cnd;
336 	int i;
337 	struct iovec aiov;
338 	struct uio auio;
339 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
340 	int error = 0, crossmnt;
341 	char *cp;
342 
343 	*retdirp = NULL;
344 	cnp->cn_nameptr = cnp->cn_pnbuf;
345 	ndp->ni_strictrelative = 0;
346 	/*
347 	 * Extract and set starting directory.
348 	 */
349 	if (dp->v_type != VDIR) {
350 		if (islocked)
351 			vput(dp);
352 		else
353 			vrele(dp);
354 		nfsvno_relpathbuf(ndp);
355 		error = ENOTDIR;
356 		goto out1;
357 	}
358 	if (islocked)
359 		NFSVOPUNLOCK(dp, 0);
360 	VREF(dp);
361 	*retdirp = dp;
362 	if (NFSVNO_EXRDONLY(exp))
363 		cnp->cn_flags |= RDONLY;
364 	ndp->ni_segflg = UIO_SYSSPACE;
365 	crossmnt = 1;
366 
367 	if (nd->nd_flag & ND_PUBLOOKUP) {
368 		ndp->ni_loopcnt = 0;
369 		if (cnp->cn_pnbuf[0] == '/') {
370 			vrele(dp);
371 			/*
372 			 * Check for degenerate pathnames here, since lookup()
373 			 * panics on them.
374 			 */
375 			for (i = 1; i < ndp->ni_pathlen; i++)
376 				if (cnp->cn_pnbuf[i] != '/')
377 					break;
378 			if (i == ndp->ni_pathlen) {
379 				error = NFSERR_ACCES;
380 				goto out;
381 			}
382 			dp = rootvnode;
383 			VREF(dp);
384 		}
385 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
386 	    (nd->nd_flag & ND_NFSV4) == 0) {
387 		/*
388 		 * Only cross mount points for NFSv4 when doing a
389 		 * mount while traversing the file system above
390 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
391 		 */
392 		cnp->cn_flags |= NOCROSSMOUNT;
393 		crossmnt = 0;
394 	}
395 
396 	/*
397 	 * Initialize for scan, set ni_startdir and bump ref on dp again
398 	 * because lookup() will dereference ni_startdir.
399 	 */
400 
401 	cnp->cn_thread = p;
402 	ndp->ni_startdir = dp;
403 	ndp->ni_rootdir = rootvnode;
404 	ndp->ni_topdir = NULL;
405 
406 	if (!lockleaf)
407 		cnp->cn_flags |= LOCKLEAF;
408 	for (;;) {
409 		cnp->cn_nameptr = cnp->cn_pnbuf;
410 		/*
411 		 * Call lookup() to do the real work.  If an error occurs,
412 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
413 		 * we do not have to dereference anything before returning.
414 		 * In either case ni_startdir will be dereferenced and NULLed
415 		 * out.
416 		 */
417 		error = lookup(ndp);
418 		if (error)
419 			break;
420 
421 		/*
422 		 * Check for encountering a symbolic link.  Trivial
423 		 * termination occurs if no symlink encountered.
424 		 */
425 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
426 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
427 				nfsvno_relpathbuf(ndp);
428 			if (ndp->ni_vp && !lockleaf)
429 				NFSVOPUNLOCK(ndp->ni_vp, 0);
430 			break;
431 		}
432 
433 		/*
434 		 * Validate symlink
435 		 */
436 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
437 			NFSVOPUNLOCK(ndp->ni_dvp, 0);
438 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
439 			error = EINVAL;
440 			goto badlink2;
441 		}
442 
443 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
444 			error = ELOOP;
445 			goto badlink2;
446 		}
447 		if (ndp->ni_pathlen > 1)
448 			cp = uma_zalloc(namei_zone, M_WAITOK);
449 		else
450 			cp = cnp->cn_pnbuf;
451 		aiov.iov_base = cp;
452 		aiov.iov_len = MAXPATHLEN;
453 		auio.uio_iov = &aiov;
454 		auio.uio_iovcnt = 1;
455 		auio.uio_offset = 0;
456 		auio.uio_rw = UIO_READ;
457 		auio.uio_segflg = UIO_SYSSPACE;
458 		auio.uio_td = NULL;
459 		auio.uio_resid = MAXPATHLEN;
460 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
461 		if (error) {
462 		badlink1:
463 			if (ndp->ni_pathlen > 1)
464 				uma_zfree(namei_zone, cp);
465 		badlink2:
466 			vrele(ndp->ni_dvp);
467 			vput(ndp->ni_vp);
468 			break;
469 		}
470 		linklen = MAXPATHLEN - auio.uio_resid;
471 		if (linklen == 0) {
472 			error = ENOENT;
473 			goto badlink1;
474 		}
475 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
476 			error = ENAMETOOLONG;
477 			goto badlink1;
478 		}
479 
480 		/*
481 		 * Adjust or replace path
482 		 */
483 		if (ndp->ni_pathlen > 1) {
484 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
485 			uma_zfree(namei_zone, cnp->cn_pnbuf);
486 			cnp->cn_pnbuf = cp;
487 		} else
488 			cnp->cn_pnbuf[linklen] = '\0';
489 		ndp->ni_pathlen += linklen;
490 
491 		/*
492 		 * Cleanup refs for next loop and check if root directory
493 		 * should replace current directory.  Normally ni_dvp
494 		 * becomes the new base directory and is cleaned up when
495 		 * we loop.  Explicitly null pointers after invalidation
496 		 * to clarify operation.
497 		 */
498 		vput(ndp->ni_vp);
499 		ndp->ni_vp = NULL;
500 
501 		if (cnp->cn_pnbuf[0] == '/') {
502 			vrele(ndp->ni_dvp);
503 			ndp->ni_dvp = ndp->ni_rootdir;
504 			VREF(ndp->ni_dvp);
505 		}
506 		ndp->ni_startdir = ndp->ni_dvp;
507 		ndp->ni_dvp = NULL;
508 	}
509 	if (!lockleaf)
510 		cnp->cn_flags &= ~LOCKLEAF;
511 
512 out:
513 	if (error) {
514 		nfsvno_relpathbuf(ndp);
515 		ndp->ni_vp = NULL;
516 		ndp->ni_dvp = NULL;
517 		ndp->ni_startdir = NULL;
518 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
519 		ndp->ni_dvp = NULL;
520 	}
521 
522 out1:
523 	NFSEXITCODE2(error, nd);
524 	return (error);
525 }
526 
527 /*
528  * Set up a pathname buffer and return a pointer to it and, optionally
529  * set a hash pointer.
530  */
531 void
532 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
533 {
534 	struct componentname *cnp = &ndp->ni_cnd;
535 
536 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
537 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
538 	if (hashpp != NULL)
539 		*hashpp = NULL;
540 	*bufpp = cnp->cn_pnbuf;
541 }
542 
543 /*
544  * Release the above path buffer, if not released by nfsvno_namei().
545  */
546 void
547 nfsvno_relpathbuf(struct nameidata *ndp)
548 {
549 
550 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
551 		panic("nfsrelpath");
552 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
553 	ndp->ni_cnd.cn_flags &= ~HASBUF;
554 }
555 
556 /*
557  * Readlink vnode op into an mbuf list.
558  */
559 int
560 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
561     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
562 {
563 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
564 	struct iovec *ivp = iv;
565 	struct uio io, *uiop = &io;
566 	struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
567 	int i, len, tlen, error = 0;
568 
569 	len = 0;
570 	i = 0;
571 	while (len < NFS_MAXPATHLEN) {
572 		NFSMGET(mp);
573 		MCLGET(mp, M_WAITOK);
574 		mp->m_len = NFSMSIZ(mp);
575 		if (len == 0) {
576 			mp3 = mp2 = mp;
577 		} else {
578 			mp2->m_next = mp;
579 			mp2 = mp;
580 		}
581 		if ((len + mp->m_len) > NFS_MAXPATHLEN) {
582 			mp->m_len = NFS_MAXPATHLEN - len;
583 			len = NFS_MAXPATHLEN;
584 		} else {
585 			len += mp->m_len;
586 		}
587 		ivp->iov_base = mtod(mp, caddr_t);
588 		ivp->iov_len = mp->m_len;
589 		i++;
590 		ivp++;
591 	}
592 	uiop->uio_iov = iv;
593 	uiop->uio_iovcnt = i;
594 	uiop->uio_offset = 0;
595 	uiop->uio_resid = len;
596 	uiop->uio_rw = UIO_READ;
597 	uiop->uio_segflg = UIO_SYSSPACE;
598 	uiop->uio_td = NULL;
599 	error = VOP_READLINK(vp, uiop, cred);
600 	if (error) {
601 		m_freem(mp3);
602 		*lenp = 0;
603 		goto out;
604 	}
605 	if (uiop->uio_resid > 0) {
606 		len -= uiop->uio_resid;
607 		tlen = NFSM_RNDUP(len);
608 		nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
609 	}
610 	*lenp = len;
611 	*mpp = mp3;
612 	*mpendp = mp;
613 
614 out:
615 	NFSEXITCODE(error);
616 	return (error);
617 }
618 
619 /*
620  * Read vnode op call into mbuf list.
621  */
622 int
623 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
624     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
625 {
626 	struct mbuf *m;
627 	int i;
628 	struct iovec *iv;
629 	struct iovec *iv2;
630 	int error = 0, len, left, siz, tlen, ioflag = 0;
631 	struct mbuf *m2 = NULL, *m3;
632 	struct uio io, *uiop = &io;
633 	struct nfsheur *nh;
634 
635 	len = left = NFSM_RNDUP(cnt);
636 	m3 = NULL;
637 	/*
638 	 * Generate the mbuf list with the uio_iov ref. to it.
639 	 */
640 	i = 0;
641 	while (left > 0) {
642 		NFSMGET(m);
643 		MCLGET(m, M_WAITOK);
644 		m->m_len = 0;
645 		siz = min(M_TRAILINGSPACE(m), left);
646 		left -= siz;
647 		i++;
648 		if (m3)
649 			m2->m_next = m;
650 		else
651 			m3 = m;
652 		m2 = m;
653 	}
654 	MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
655 	    M_TEMP, M_WAITOK);
656 	uiop->uio_iov = iv2 = iv;
657 	m = m3;
658 	left = len;
659 	i = 0;
660 	while (left > 0) {
661 		if (m == NULL)
662 			panic("nfsvno_read iov");
663 		siz = min(M_TRAILINGSPACE(m), left);
664 		if (siz > 0) {
665 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
666 			iv->iov_len = siz;
667 			m->m_len += siz;
668 			left -= siz;
669 			iv++;
670 			i++;
671 		}
672 		m = m->m_next;
673 	}
674 	uiop->uio_iovcnt = i;
675 	uiop->uio_offset = off;
676 	uiop->uio_resid = len;
677 	uiop->uio_rw = UIO_READ;
678 	uiop->uio_segflg = UIO_SYSSPACE;
679 	uiop->uio_td = NULL;
680 	nh = nfsrv_sequential_heuristic(uiop, vp);
681 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
682 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
683 	FREE((caddr_t)iv2, M_TEMP);
684 	if (error) {
685 		m_freem(m3);
686 		*mpp = NULL;
687 		goto out;
688 	}
689 	nh->nh_nextoff = uiop->uio_offset;
690 	tlen = len - uiop->uio_resid;
691 	cnt = cnt < tlen ? cnt : tlen;
692 	tlen = NFSM_RNDUP(cnt);
693 	if (tlen == 0) {
694 		m_freem(m3);
695 		m3 = NULL;
696 	} else if (len != tlen || tlen != cnt)
697 		nfsrv_adj(m3, len - tlen, tlen - cnt);
698 	*mpp = m3;
699 	*mpendp = m2;
700 
701 out:
702 	NFSEXITCODE(error);
703 	return (error);
704 }
705 
706 /*
707  * Write vnode op from an mbuf list.
708  */
709 int
710 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
711     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
712 {
713 	struct iovec *ivp;
714 	int i, len;
715 	struct iovec *iv;
716 	int ioflags, error;
717 	struct uio io, *uiop = &io;
718 	struct nfsheur *nh;
719 
720 	MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
721 	    M_WAITOK);
722 	uiop->uio_iov = iv = ivp;
723 	uiop->uio_iovcnt = cnt;
724 	i = mtod(mp, caddr_t) + mp->m_len - cp;
725 	len = retlen;
726 	while (len > 0) {
727 		if (mp == NULL)
728 			panic("nfsvno_write");
729 		if (i > 0) {
730 			i = min(i, len);
731 			ivp->iov_base = cp;
732 			ivp->iov_len = i;
733 			ivp++;
734 			len -= i;
735 		}
736 		mp = mp->m_next;
737 		if (mp) {
738 			i = mp->m_len;
739 			cp = mtod(mp, caddr_t);
740 		}
741 	}
742 
743 	if (stable == NFSWRITE_UNSTABLE)
744 		ioflags = IO_NODELOCKED;
745 	else
746 		ioflags = (IO_SYNC | IO_NODELOCKED);
747 	uiop->uio_resid = retlen;
748 	uiop->uio_rw = UIO_WRITE;
749 	uiop->uio_segflg = UIO_SYSSPACE;
750 	NFSUIOPROC(uiop, p);
751 	uiop->uio_offset = off;
752 	nh = nfsrv_sequential_heuristic(uiop, vp);
753 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
754 	error = VOP_WRITE(vp, uiop, ioflags, cred);
755 	if (error == 0)
756 		nh->nh_nextoff = uiop->uio_offset;
757 	FREE((caddr_t)iv, M_TEMP);
758 
759 	NFSEXITCODE(error);
760 	return (error);
761 }
762 
763 /*
764  * Common code for creating a regular file (plus special files for V2).
765  */
766 int
767 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
768     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
769     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
770 {
771 	u_quad_t tempsize;
772 	int error;
773 
774 	error = nd->nd_repstat;
775 	if (!error && ndp->ni_vp == NULL) {
776 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
777 			vrele(ndp->ni_startdir);
778 			error = VOP_CREATE(ndp->ni_dvp,
779 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
780 			vput(ndp->ni_dvp);
781 			nfsvno_relpathbuf(ndp);
782 			if (!error) {
783 				if (*exclusive_flagp) {
784 					*exclusive_flagp = 0;
785 					NFSVNO_ATTRINIT(nvap);
786 					nvap->na_atime.tv_sec = cverf[0];
787 					nvap->na_atime.tv_nsec = cverf[1];
788 					error = VOP_SETATTR(ndp->ni_vp,
789 					    &nvap->na_vattr, nd->nd_cred);
790 				}
791 			}
792 		/*
793 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
794 		 * (This implies, just get out on an error.)
795 		 */
796 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
797 			nvap->na_type == VFIFO) {
798 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
799 				nvap->na_type = VFIFO;
800                         if (nvap->na_type != VFIFO &&
801 			    (error = priv_check_cred(nd->nd_cred,
802 			     PRIV_VFS_MKNOD_DEV, 0))) {
803 				vrele(ndp->ni_startdir);
804 				nfsvno_relpathbuf(ndp);
805 				vput(ndp->ni_dvp);
806 				goto out;
807 			}
808 			nvap->na_rdev = rdev;
809 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
810 			    &ndp->ni_cnd, &nvap->na_vattr);
811 			vput(ndp->ni_dvp);
812 			nfsvno_relpathbuf(ndp);
813 			vrele(ndp->ni_startdir);
814 			if (error)
815 				goto out;
816 		} else {
817 			vrele(ndp->ni_startdir);
818 			nfsvno_relpathbuf(ndp);
819 			vput(ndp->ni_dvp);
820 			error = ENXIO;
821 			goto out;
822 		}
823 		*vpp = ndp->ni_vp;
824 	} else {
825 		/*
826 		 * Handle cases where error is already set and/or
827 		 * the file exists.
828 		 * 1 - clean up the lookup
829 		 * 2 - iff !error and na_size set, truncate it
830 		 */
831 		vrele(ndp->ni_startdir);
832 		nfsvno_relpathbuf(ndp);
833 		*vpp = ndp->ni_vp;
834 		if (ndp->ni_dvp == *vpp)
835 			vrele(ndp->ni_dvp);
836 		else
837 			vput(ndp->ni_dvp);
838 		if (!error && nvap->na_size != VNOVAL) {
839 			error = nfsvno_accchk(*vpp, VWRITE,
840 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
841 			    NFSACCCHK_VPISLOCKED, NULL);
842 			if (!error) {
843 				tempsize = nvap->na_size;
844 				NFSVNO_ATTRINIT(nvap);
845 				nvap->na_size = tempsize;
846 				error = VOP_SETATTR(*vpp,
847 				    &nvap->na_vattr, nd->nd_cred);
848 			}
849 		}
850 		if (error)
851 			vput(*vpp);
852 	}
853 
854 out:
855 	NFSEXITCODE(error);
856 	return (error);
857 }
858 
859 /*
860  * Do a mknod vnode op.
861  */
862 int
863 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
864     struct thread *p)
865 {
866 	int error = 0;
867 	enum vtype vtyp;
868 
869 	vtyp = nvap->na_type;
870 	/*
871 	 * Iff doesn't exist, create it.
872 	 */
873 	if (ndp->ni_vp) {
874 		vrele(ndp->ni_startdir);
875 		nfsvno_relpathbuf(ndp);
876 		vput(ndp->ni_dvp);
877 		vrele(ndp->ni_vp);
878 		error = EEXIST;
879 		goto out;
880 	}
881 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
882 		vrele(ndp->ni_startdir);
883 		nfsvno_relpathbuf(ndp);
884 		vput(ndp->ni_dvp);
885 		error = NFSERR_BADTYPE;
886 		goto out;
887 	}
888 	if (vtyp == VSOCK) {
889 		vrele(ndp->ni_startdir);
890 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
891 		    &ndp->ni_cnd, &nvap->na_vattr);
892 		vput(ndp->ni_dvp);
893 		nfsvno_relpathbuf(ndp);
894 	} else {
895 		if (nvap->na_type != VFIFO &&
896 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
897 			vrele(ndp->ni_startdir);
898 			nfsvno_relpathbuf(ndp);
899 			vput(ndp->ni_dvp);
900 			goto out;
901 		}
902 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
903 		    &ndp->ni_cnd, &nvap->na_vattr);
904 		vput(ndp->ni_dvp);
905 		nfsvno_relpathbuf(ndp);
906 		vrele(ndp->ni_startdir);
907 		/*
908 		 * Since VOP_MKNOD returns the ni_vp, I can't
909 		 * see any reason to do the lookup.
910 		 */
911 	}
912 
913 out:
914 	NFSEXITCODE(error);
915 	return (error);
916 }
917 
918 /*
919  * Mkdir vnode op.
920  */
921 int
922 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
923     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
924 {
925 	int error = 0;
926 
927 	if (ndp->ni_vp != NULL) {
928 		if (ndp->ni_dvp == ndp->ni_vp)
929 			vrele(ndp->ni_dvp);
930 		else
931 			vput(ndp->ni_dvp);
932 		vrele(ndp->ni_vp);
933 		nfsvno_relpathbuf(ndp);
934 		error = EEXIST;
935 		goto out;
936 	}
937 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
938 	    &nvap->na_vattr);
939 	vput(ndp->ni_dvp);
940 	nfsvno_relpathbuf(ndp);
941 
942 out:
943 	NFSEXITCODE(error);
944 	return (error);
945 }
946 
947 /*
948  * symlink vnode op.
949  */
950 int
951 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
952     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
953     struct nfsexstuff *exp)
954 {
955 	int error = 0;
956 
957 	if (ndp->ni_vp) {
958 		vrele(ndp->ni_startdir);
959 		nfsvno_relpathbuf(ndp);
960 		if (ndp->ni_dvp == ndp->ni_vp)
961 			vrele(ndp->ni_dvp);
962 		else
963 			vput(ndp->ni_dvp);
964 		vrele(ndp->ni_vp);
965 		error = EEXIST;
966 		goto out;
967 	}
968 
969 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
970 	    &nvap->na_vattr, pathcp);
971 	vput(ndp->ni_dvp);
972 	vrele(ndp->ni_startdir);
973 	nfsvno_relpathbuf(ndp);
974 	/*
975 	 * Although FreeBSD still had the lookup code in
976 	 * it for 7/current, there doesn't seem to be any
977 	 * point, since VOP_SYMLINK() returns the ni_vp.
978 	 * Just vput it for v2.
979 	 */
980 	if (!not_v2 && !error)
981 		vput(ndp->ni_vp);
982 
983 out:
984 	NFSEXITCODE(error);
985 	return (error);
986 }
987 
988 /*
989  * Parse symbolic link arguments.
990  * This function has an ugly side effect. It will MALLOC() an area for
991  * the symlink and set iov_base to point to it, only if it succeeds.
992  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
993  * be FREE'd later.
994  */
995 int
996 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
997     struct thread *p, char **pathcpp, int *lenp)
998 {
999 	u_int32_t *tl;
1000 	char *pathcp = NULL;
1001 	int error = 0, len;
1002 	struct nfsv2_sattr *sp;
1003 
1004 	*pathcpp = NULL;
1005 	*lenp = 0;
1006 	if ((nd->nd_flag & ND_NFSV3) &&
1007 	    (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
1008 		goto nfsmout;
1009 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1010 	len = fxdr_unsigned(int, *tl);
1011 	if (len > NFS_MAXPATHLEN || len <= 0) {
1012 		error = EBADRPC;
1013 		goto nfsmout;
1014 	}
1015 	MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
1016 	error = nfsrv_mtostr(nd, pathcp, len);
1017 	if (error)
1018 		goto nfsmout;
1019 	if (nd->nd_flag & ND_NFSV2) {
1020 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1021 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1022 	}
1023 	*pathcpp = pathcp;
1024 	*lenp = len;
1025 	NFSEXITCODE2(0, nd);
1026 	return (0);
1027 nfsmout:
1028 	if (pathcp)
1029 		free(pathcp, M_TEMP);
1030 	NFSEXITCODE2(error, nd);
1031 	return (error);
1032 }
1033 
1034 /*
1035  * Remove a non-directory object.
1036  */
1037 int
1038 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1039     struct thread *p, struct nfsexstuff *exp)
1040 {
1041 	struct vnode *vp;
1042 	int error = 0;
1043 
1044 	vp = ndp->ni_vp;
1045 	if (vp->v_type == VDIR)
1046 		error = NFSERR_ISDIR;
1047 	else if (is_v4)
1048 		error = nfsrv_checkremove(vp, 1, p);
1049 	if (!error)
1050 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1051 	if (ndp->ni_dvp == vp)
1052 		vrele(ndp->ni_dvp);
1053 	else
1054 		vput(ndp->ni_dvp);
1055 	vput(vp);
1056 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1057 		nfsvno_relpathbuf(ndp);
1058 	NFSEXITCODE(error);
1059 	return (error);
1060 }
1061 
1062 /*
1063  * Remove a directory.
1064  */
1065 int
1066 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1067     struct thread *p, struct nfsexstuff *exp)
1068 {
1069 	struct vnode *vp;
1070 	int error = 0;
1071 
1072 	vp = ndp->ni_vp;
1073 	if (vp->v_type != VDIR) {
1074 		error = ENOTDIR;
1075 		goto out;
1076 	}
1077 	/*
1078 	 * No rmdir "." please.
1079 	 */
1080 	if (ndp->ni_dvp == vp) {
1081 		error = EINVAL;
1082 		goto out;
1083 	}
1084 	/*
1085 	 * The root of a mounted filesystem cannot be deleted.
1086 	 */
1087 	if (vp->v_vflag & VV_ROOT)
1088 		error = EBUSY;
1089 out:
1090 	if (!error)
1091 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1092 	if (ndp->ni_dvp == vp)
1093 		vrele(ndp->ni_dvp);
1094 	else
1095 		vput(ndp->ni_dvp);
1096 	vput(vp);
1097 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1098 		nfsvno_relpathbuf(ndp);
1099 	NFSEXITCODE(error);
1100 	return (error);
1101 }
1102 
1103 /*
1104  * Rename vnode op.
1105  */
1106 int
1107 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1108     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1109 {
1110 	struct vnode *fvp, *tvp, *tdvp;
1111 	int error = 0;
1112 
1113 	fvp = fromndp->ni_vp;
1114 	if (ndstat) {
1115 		vrele(fromndp->ni_dvp);
1116 		vrele(fvp);
1117 		error = ndstat;
1118 		goto out1;
1119 	}
1120 	tdvp = tondp->ni_dvp;
1121 	tvp = tondp->ni_vp;
1122 	if (tvp != NULL) {
1123 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1124 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1125 			goto out;
1126 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1127 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1128 			goto out;
1129 		}
1130 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1131 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1132 			goto out;
1133 		}
1134 
1135 		/*
1136 		 * A rename to '.' or '..' results in a prematurely
1137 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1138 		 * here.
1139 		 */
1140 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1141 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1142 		    (tondp->ni_cnd.cn_namelen == 2 &&
1143 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1144 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1145 			error = EINVAL;
1146 			goto out;
1147 		}
1148 	}
1149 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1150 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1151 		goto out;
1152 	}
1153 	if (fvp->v_mount != tdvp->v_mount) {
1154 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1155 		goto out;
1156 	}
1157 	if (fvp == tdvp) {
1158 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1159 		goto out;
1160 	}
1161 	if (fvp == tvp) {
1162 		/*
1163 		 * If source and destination are the same, there is nothing to
1164 		 * do. Set error to -1 to indicate this.
1165 		 */
1166 		error = -1;
1167 		goto out;
1168 	}
1169 	if (ndflag & ND_NFSV4) {
1170 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1171 			error = nfsrv_checkremove(fvp, 0, p);
1172 			NFSVOPUNLOCK(fvp, 0);
1173 		} else
1174 			error = EPERM;
1175 		if (tvp && !error)
1176 			error = nfsrv_checkremove(tvp, 1, p);
1177 	} else {
1178 		/*
1179 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1180 		 * that the NFSv4 client won't be confused by the rename.
1181 		 * Since nfsd_recalldelegation() can only be called on an
1182 		 * unlocked vnode at this point and fvp is the file that will
1183 		 * still exist after the rename, just do fvp.
1184 		 */
1185 		nfsd_recalldelegation(fvp, p);
1186 	}
1187 out:
1188 	if (!error) {
1189 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1190 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1191 		    &tondp->ni_cnd);
1192 	} else {
1193 		if (tdvp == tvp)
1194 			vrele(tdvp);
1195 		else
1196 			vput(tdvp);
1197 		if (tvp)
1198 			vput(tvp);
1199 		vrele(fromndp->ni_dvp);
1200 		vrele(fvp);
1201 		if (error == -1)
1202 			error = 0;
1203 	}
1204 	vrele(tondp->ni_startdir);
1205 	nfsvno_relpathbuf(tondp);
1206 out1:
1207 	vrele(fromndp->ni_startdir);
1208 	nfsvno_relpathbuf(fromndp);
1209 	NFSEXITCODE(error);
1210 	return (error);
1211 }
1212 
1213 /*
1214  * Link vnode op.
1215  */
1216 int
1217 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1218     struct thread *p, struct nfsexstuff *exp)
1219 {
1220 	struct vnode *xp;
1221 	int error = 0;
1222 
1223 	xp = ndp->ni_vp;
1224 	if (xp != NULL) {
1225 		error = EEXIST;
1226 	} else {
1227 		xp = ndp->ni_dvp;
1228 		if (vp->v_mount != xp->v_mount)
1229 			error = EXDEV;
1230 	}
1231 	if (!error) {
1232 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1233 		if ((vp->v_iflag & VI_DOOMED) == 0)
1234 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1235 		else
1236 			error = EPERM;
1237 		if (ndp->ni_dvp == vp)
1238 			vrele(ndp->ni_dvp);
1239 		else
1240 			vput(ndp->ni_dvp);
1241 		NFSVOPUNLOCK(vp, 0);
1242 	} else {
1243 		if (ndp->ni_dvp == ndp->ni_vp)
1244 			vrele(ndp->ni_dvp);
1245 		else
1246 			vput(ndp->ni_dvp);
1247 		if (ndp->ni_vp)
1248 			vrele(ndp->ni_vp);
1249 	}
1250 	nfsvno_relpathbuf(ndp);
1251 	NFSEXITCODE(error);
1252 	return (error);
1253 }
1254 
1255 /*
1256  * Do the fsync() appropriate for the commit.
1257  */
1258 int
1259 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1260     struct thread *td)
1261 {
1262 	int error = 0;
1263 
1264 	/*
1265 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1266 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1267 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1268 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1269 	 */
1270 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT) {
1271 		/*
1272 		 * Give up and do the whole thing
1273 		 */
1274 		if (vp->v_object &&
1275 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1276 			VM_OBJECT_WLOCK(vp->v_object);
1277 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1278 			VM_OBJECT_WUNLOCK(vp->v_object);
1279 		}
1280 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1281 	} else {
1282 		/*
1283 		 * Locate and synchronously write any buffers that fall
1284 		 * into the requested range.  Note:  we are assuming that
1285 		 * f_iosize is a power of 2.
1286 		 */
1287 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1288 		int iomask = iosize - 1;
1289 		struct bufobj *bo;
1290 		daddr_t lblkno;
1291 
1292 		/*
1293 		 * Align to iosize boundry, super-align to page boundry.
1294 		 */
1295 		if (off & iomask) {
1296 			cnt += off & iomask;
1297 			off &= ~(u_quad_t)iomask;
1298 		}
1299 		if (off & PAGE_MASK) {
1300 			cnt += off & PAGE_MASK;
1301 			off &= ~(u_quad_t)PAGE_MASK;
1302 		}
1303 		lblkno = off / iosize;
1304 
1305 		if (vp->v_object &&
1306 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1307 			VM_OBJECT_WLOCK(vp->v_object);
1308 			vm_object_page_clean(vp->v_object, off, off + cnt,
1309 			    OBJPC_SYNC);
1310 			VM_OBJECT_WUNLOCK(vp->v_object);
1311 		}
1312 
1313 		bo = &vp->v_bufobj;
1314 		BO_LOCK(bo);
1315 		while (cnt > 0) {
1316 			struct buf *bp;
1317 
1318 			/*
1319 			 * If we have a buffer and it is marked B_DELWRI we
1320 			 * have to lock and write it.  Otherwise the prior
1321 			 * write is assumed to have already been committed.
1322 			 *
1323 			 * gbincore() can return invalid buffers now so we
1324 			 * have to check that bit as well (though B_DELWRI
1325 			 * should not be set if B_INVAL is set there could be
1326 			 * a race here since we haven't locked the buffer).
1327 			 */
1328 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1329 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1330 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1331 					BO_LOCK(bo);
1332 					continue; /* retry */
1333 				}
1334 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1335 				    B_DELWRI) {
1336 					bremfree(bp);
1337 					bp->b_flags &= ~B_ASYNC;
1338 					bwrite(bp);
1339 					++nfs_commit_miss;
1340 				} else
1341 					BUF_UNLOCK(bp);
1342 				BO_LOCK(bo);
1343 			}
1344 			++nfs_commit_blks;
1345 			if (cnt < iosize)
1346 				break;
1347 			cnt -= iosize;
1348 			++lblkno;
1349 		}
1350 		BO_UNLOCK(bo);
1351 	}
1352 	NFSEXITCODE(error);
1353 	return (error);
1354 }
1355 
1356 /*
1357  * Statfs vnode op.
1358  */
1359 int
1360 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1361 {
1362 	int error;
1363 
1364 	error = VFS_STATFS(vp->v_mount, sf);
1365 	if (error == 0) {
1366 		/*
1367 		 * Since NFS handles these values as unsigned on the
1368 		 * wire, there is no way to represent negative values,
1369 		 * so set them to 0. Without this, they will appear
1370 		 * to be very large positive values for clients like
1371 		 * Solaris10.
1372 		 */
1373 		if (sf->f_bavail < 0)
1374 			sf->f_bavail = 0;
1375 		if (sf->f_ffree < 0)
1376 			sf->f_ffree = 0;
1377 	}
1378 	NFSEXITCODE(error);
1379 	return (error);
1380 }
1381 
1382 /*
1383  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1384  * must handle nfsrv_opencheck() calls after any other access checks.
1385  */
1386 void
1387 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1388     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1389     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1390     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1391     struct nfsexstuff *exp, struct vnode **vpp)
1392 {
1393 	struct vnode *vp = NULL;
1394 	u_quad_t tempsize;
1395 	struct nfsexstuff nes;
1396 
1397 	if (ndp->ni_vp == NULL)
1398 		nd->nd_repstat = nfsrv_opencheck(clientid,
1399 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1400 	if (!nd->nd_repstat) {
1401 		if (ndp->ni_vp == NULL) {
1402 			vrele(ndp->ni_startdir);
1403 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1404 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1405 			vput(ndp->ni_dvp);
1406 			nfsvno_relpathbuf(ndp);
1407 			if (!nd->nd_repstat) {
1408 				if (*exclusive_flagp) {
1409 					*exclusive_flagp = 0;
1410 					NFSVNO_ATTRINIT(nvap);
1411 					nvap->na_atime.tv_sec = cverf[0];
1412 					nvap->na_atime.tv_nsec = cverf[1];
1413 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1414 					    &nvap->na_vattr, cred);
1415 				} else {
1416 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1417 					    aclp, p, attrbitp, exp);
1418 				}
1419 			}
1420 			vp = ndp->ni_vp;
1421 		} else {
1422 			if (ndp->ni_startdir)
1423 				vrele(ndp->ni_startdir);
1424 			nfsvno_relpathbuf(ndp);
1425 			vp = ndp->ni_vp;
1426 			if (create == NFSV4OPEN_CREATE) {
1427 				if (ndp->ni_dvp == vp)
1428 					vrele(ndp->ni_dvp);
1429 				else
1430 					vput(ndp->ni_dvp);
1431 			}
1432 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1433 				if (ndp->ni_cnd.cn_flags & RDONLY)
1434 					NFSVNO_SETEXRDONLY(&nes);
1435 				else
1436 					NFSVNO_EXINIT(&nes);
1437 				nd->nd_repstat = nfsvno_accchk(vp,
1438 				    VWRITE, cred, &nes, p,
1439 				    NFSACCCHK_NOOVERRIDE,
1440 				    NFSACCCHK_VPISLOCKED, NULL);
1441 				nd->nd_repstat = nfsrv_opencheck(clientid,
1442 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1443 				if (!nd->nd_repstat) {
1444 					tempsize = nvap->na_size;
1445 					NFSVNO_ATTRINIT(nvap);
1446 					nvap->na_size = tempsize;
1447 					nd->nd_repstat = VOP_SETATTR(vp,
1448 					    &nvap->na_vattr, cred);
1449 				}
1450 			} else if (vp->v_type == VREG) {
1451 				nd->nd_repstat = nfsrv_opencheck(clientid,
1452 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1453 			}
1454 		}
1455 	} else {
1456 		if (ndp->ni_cnd.cn_flags & HASBUF)
1457 			nfsvno_relpathbuf(ndp);
1458 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1459 			vrele(ndp->ni_startdir);
1460 			if (ndp->ni_dvp == ndp->ni_vp)
1461 				vrele(ndp->ni_dvp);
1462 			else
1463 				vput(ndp->ni_dvp);
1464 			if (ndp->ni_vp)
1465 				vput(ndp->ni_vp);
1466 		}
1467 	}
1468 	*vpp = vp;
1469 
1470 	NFSEXITCODE2(0, nd);
1471 }
1472 
1473 /*
1474  * Updates the file rev and sets the mtime and ctime
1475  * to the current clock time, returning the va_filerev and va_Xtime
1476  * values.
1477  * Return ESTALE to indicate the vnode is VI_DOOMED.
1478  */
1479 int
1480 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1481     struct ucred *cred, struct thread *p)
1482 {
1483 	struct vattr va;
1484 
1485 	VATTR_NULL(&va);
1486 	vfs_timestamp(&va.va_mtime);
1487 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1488 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1489 		if ((vp->v_iflag & VI_DOOMED) != 0)
1490 			return (ESTALE);
1491 	}
1492 	(void) VOP_SETATTR(vp, &va, cred);
1493 	(void) nfsvno_getattr(vp, nvap, cred, p, 1);
1494 	return (0);
1495 }
1496 
1497 /*
1498  * Glue routine to nfsv4_fillattr().
1499  */
1500 int
1501 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1502     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1503     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1504     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1505 {
1506 	int error;
1507 
1508 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1509 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1510 	    mounted_on_fileno);
1511 	NFSEXITCODE2(0, nd);
1512 	return (error);
1513 }
1514 
1515 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1516 /*
1517  * nfs readdir service
1518  * - mallocs what it thinks is enough to read
1519  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1520  * - calls VOP_READDIR()
1521  * - loops around building the reply
1522  *	if the output generated exceeds count break out of loop
1523  *	The NFSM_CLGET macro is used here so that the reply will be packed
1524  *	tightly in mbuf clusters.
1525  * - it trims out records with d_fileno == 0
1526  *	this doesn't matter for Unix clients, but they might confuse clients
1527  *	for other os'.
1528  * - it trims out records with d_type == DT_WHT
1529  *	these cannot be seen through NFS (unless we extend the protocol)
1530  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1531  * PS: The NFS protocol spec. does not clarify what the "count" byte
1532  *	argument is a count of.. just name strings and file id's or the
1533  *	entire reply rpc or ...
1534  *	I tried just file name and id sizes and it confused the Sun client,
1535  *	so I am using the full rpc size now. The "paranoia.." comment refers
1536  *	to including the status longwords that are not a part of the dir.
1537  *	"entry" structures, but are in the rpc.
1538  */
1539 int
1540 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1541     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1542 {
1543 	struct dirent *dp;
1544 	u_int32_t *tl;
1545 	int dirlen;
1546 	char *cpos, *cend, *rbuf;
1547 	struct nfsvattr at;
1548 	int nlen, error = 0, getret = 1;
1549 	int siz, cnt, fullsiz, eofflag, ncookies;
1550 	u_int64_t off, toff, verf;
1551 	u_long *cookies = NULL, *cookiep;
1552 	struct uio io;
1553 	struct iovec iv;
1554 	int not_zfs;
1555 
1556 	if (nd->nd_repstat) {
1557 		nfsrv_postopattr(nd, getret, &at);
1558 		goto out;
1559 	}
1560 	if (nd->nd_flag & ND_NFSV2) {
1561 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1562 		off = fxdr_unsigned(u_quad_t, *tl++);
1563 	} else {
1564 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1565 		off = fxdr_hyper(tl);
1566 		tl += 2;
1567 		verf = fxdr_hyper(tl);
1568 		tl += 2;
1569 	}
1570 	toff = off;
1571 	cnt = fxdr_unsigned(int, *tl);
1572 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1573 		cnt = NFS_SRVMAXDATA(nd);
1574 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1575 	fullsiz = siz;
1576 	if (nd->nd_flag & ND_NFSV3) {
1577 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1578 		    p, 1);
1579 #if 0
1580 		/*
1581 		 * va_filerev is not sufficient as a cookie verifier,
1582 		 * since it is not supposed to change when entries are
1583 		 * removed/added unless that offset cookies returned to
1584 		 * the client are no longer valid.
1585 		 */
1586 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
1587 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1588 #endif
1589 	}
1590 	if (!nd->nd_repstat && vp->v_type != VDIR)
1591 		nd->nd_repstat = NFSERR_NOTDIR;
1592 	if (nd->nd_repstat == 0 && cnt == 0) {
1593 		if (nd->nd_flag & ND_NFSV2)
1594 			/* NFSv2 does not have NFSERR_TOOSMALL */
1595 			nd->nd_repstat = EPERM;
1596 		else
1597 			nd->nd_repstat = NFSERR_TOOSMALL;
1598 	}
1599 	if (!nd->nd_repstat)
1600 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1601 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1602 		    NFSACCCHK_VPISLOCKED, NULL);
1603 	if (nd->nd_repstat) {
1604 		vput(vp);
1605 		if (nd->nd_flag & ND_NFSV3)
1606 			nfsrv_postopattr(nd, getret, &at);
1607 		goto out;
1608 	}
1609 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1610 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1611 again:
1612 	eofflag = 0;
1613 	if (cookies) {
1614 		free((caddr_t)cookies, M_TEMP);
1615 		cookies = NULL;
1616 	}
1617 
1618 	iv.iov_base = rbuf;
1619 	iv.iov_len = siz;
1620 	io.uio_iov = &iv;
1621 	io.uio_iovcnt = 1;
1622 	io.uio_offset = (off_t)off;
1623 	io.uio_resid = siz;
1624 	io.uio_segflg = UIO_SYSSPACE;
1625 	io.uio_rw = UIO_READ;
1626 	io.uio_td = NULL;
1627 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1628 	    &cookies);
1629 	off = (u_int64_t)io.uio_offset;
1630 	if (io.uio_resid)
1631 		siz -= io.uio_resid;
1632 
1633 	if (!cookies && !nd->nd_repstat)
1634 		nd->nd_repstat = NFSERR_PERM;
1635 	if (nd->nd_flag & ND_NFSV3) {
1636 		getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1637 		if (!nd->nd_repstat)
1638 			nd->nd_repstat = getret;
1639 	}
1640 
1641 	/*
1642 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
1643 	 */
1644 	if (nd->nd_repstat) {
1645 		vput(vp);
1646 		free((caddr_t)rbuf, M_TEMP);
1647 		if (cookies)
1648 			free((caddr_t)cookies, M_TEMP);
1649 		if (nd->nd_flag & ND_NFSV3)
1650 			nfsrv_postopattr(nd, getret, &at);
1651 		goto out;
1652 	}
1653 	/*
1654 	 * If nothing read, return eof
1655 	 * rpc reply
1656 	 */
1657 	if (siz == 0) {
1658 		vput(vp);
1659 		if (nd->nd_flag & ND_NFSV2) {
1660 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1661 		} else {
1662 			nfsrv_postopattr(nd, getret, &at);
1663 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1664 			txdr_hyper(at.na_filerev, tl);
1665 			tl += 2;
1666 		}
1667 		*tl++ = newnfs_false;
1668 		*tl = newnfs_true;
1669 		FREE((caddr_t)rbuf, M_TEMP);
1670 		FREE((caddr_t)cookies, M_TEMP);
1671 		goto out;
1672 	}
1673 
1674 	/*
1675 	 * Check for degenerate cases of nothing useful read.
1676 	 * If so go try again
1677 	 */
1678 	cpos = rbuf;
1679 	cend = rbuf + siz;
1680 	dp = (struct dirent *)cpos;
1681 	cookiep = cookies;
1682 
1683 	/*
1684 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1685 	 * directory offset up to a block boundary, so it is necessary to
1686 	 * skip over the records that precede the requested offset. This
1687 	 * requires the assumption that file offset cookies monotonically
1688 	 * increase.
1689 	 * Since the offset cookies don't monotonically increase for ZFS,
1690 	 * this is not done when ZFS is the file system.
1691 	 */
1692 	while (cpos < cend && ncookies > 0 &&
1693 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1694 	     (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
1695 		cpos += dp->d_reclen;
1696 		dp = (struct dirent *)cpos;
1697 		cookiep++;
1698 		ncookies--;
1699 	}
1700 	if (cpos >= cend || ncookies == 0) {
1701 		siz = fullsiz;
1702 		toff = off;
1703 		goto again;
1704 	}
1705 	vput(vp);
1706 
1707 	/*
1708 	 * dirlen is the size of the reply, including all XDR and must
1709 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1710 	 * if the XDR should be included in "count", but to be safe, we do.
1711 	 * (Include the two booleans at the end of the reply in dirlen now.)
1712 	 */
1713 	if (nd->nd_flag & ND_NFSV3) {
1714 		nfsrv_postopattr(nd, getret, &at);
1715 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1716 		txdr_hyper(at.na_filerev, tl);
1717 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1718 	} else {
1719 		dirlen = 2 * NFSX_UNSIGNED;
1720 	}
1721 
1722 	/* Loop through the records and build reply */
1723 	while (cpos < cend && ncookies > 0) {
1724 		nlen = dp->d_namlen;
1725 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1726 			nlen <= NFS_MAXNAMLEN) {
1727 			if (nd->nd_flag & ND_NFSV3)
1728 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1729 			else
1730 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1731 			if (dirlen > cnt) {
1732 				eofflag = 0;
1733 				break;
1734 			}
1735 
1736 			/*
1737 			 * Build the directory record xdr from
1738 			 * the dirent entry.
1739 			 */
1740 			if (nd->nd_flag & ND_NFSV3) {
1741 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1742 				*tl++ = newnfs_true;
1743 				*tl++ = 0;
1744 			} else {
1745 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1746 				*tl++ = newnfs_true;
1747 			}
1748 			*tl = txdr_unsigned(dp->d_fileno);
1749 			(void) nfsm_strtom(nd, dp->d_name, nlen);
1750 			if (nd->nd_flag & ND_NFSV3) {
1751 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1752 				*tl++ = 0;
1753 			} else
1754 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1755 			*tl = txdr_unsigned(*cookiep);
1756 		}
1757 		cpos += dp->d_reclen;
1758 		dp = (struct dirent *)cpos;
1759 		cookiep++;
1760 		ncookies--;
1761 	}
1762 	if (cpos < cend)
1763 		eofflag = 0;
1764 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1765 	*tl++ = newnfs_false;
1766 	if (eofflag)
1767 		*tl = newnfs_true;
1768 	else
1769 		*tl = newnfs_false;
1770 	FREE((caddr_t)rbuf, M_TEMP);
1771 	FREE((caddr_t)cookies, M_TEMP);
1772 
1773 out:
1774 	NFSEXITCODE2(0, nd);
1775 	return (0);
1776 nfsmout:
1777 	vput(vp);
1778 	NFSEXITCODE2(error, nd);
1779 	return (error);
1780 }
1781 
1782 /*
1783  * Readdirplus for V3 and Readdir for V4.
1784  */
1785 int
1786 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1787     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1788 {
1789 	struct dirent *dp;
1790 	u_int32_t *tl;
1791 	int dirlen;
1792 	char *cpos, *cend, *rbuf;
1793 	struct vnode *nvp;
1794 	fhandle_t nfh;
1795 	struct nfsvattr nva, at, *nvap = &nva;
1796 	struct mbuf *mb0, *mb1;
1797 	struct nfsreferral *refp;
1798 	int nlen, r, error = 0, getret = 1, usevget = 1;
1799 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1800 	caddr_t bpos0, bpos1;
1801 	u_int64_t off, toff, verf;
1802 	u_long *cookies = NULL, *cookiep;
1803 	nfsattrbit_t attrbits, rderrbits, savbits;
1804 	struct uio io;
1805 	struct iovec iv;
1806 	struct componentname cn;
1807 	int at_root, needs_unbusy, not_zfs, supports_nfsv4acls;
1808 	struct mount *mp, *new_mp;
1809 	uint64_t mounted_on_fileno;
1810 
1811 	if (nd->nd_repstat) {
1812 		nfsrv_postopattr(nd, getret, &at);
1813 		goto out;
1814 	}
1815 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1816 	off = fxdr_hyper(tl);
1817 	toff = off;
1818 	tl += 2;
1819 	verf = fxdr_hyper(tl);
1820 	tl += 2;
1821 	siz = fxdr_unsigned(int, *tl++);
1822 	cnt = fxdr_unsigned(int, *tl);
1823 
1824 	/*
1825 	 * Use the server's maximum data transfer size as the upper bound
1826 	 * on reply datalen.
1827 	 */
1828 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1829 		cnt = NFS_SRVMAXDATA(nd);
1830 
1831 	/*
1832 	 * siz is a "hint" of how much directory information (name, fileid,
1833 	 * cookie) should be in the reply. At least one client "hints" 0,
1834 	 * so I set it to cnt for that case. I also round it up to the
1835 	 * next multiple of DIRBLKSIZ.
1836 	 */
1837 	if (siz <= 0)
1838 		siz = cnt;
1839 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1840 
1841 	if (nd->nd_flag & ND_NFSV4) {
1842 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1843 		if (error)
1844 			goto nfsmout;
1845 		NFSSET_ATTRBIT(&savbits, &attrbits);
1846 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1847 		NFSZERO_ATTRBIT(&rderrbits);
1848 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1849 	} else {
1850 		NFSZERO_ATTRBIT(&attrbits);
1851 	}
1852 	fullsiz = siz;
1853 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1854 	if (!nd->nd_repstat) {
1855 	    if (off && verf != at.na_filerev) {
1856 		/*
1857 		 * va_filerev is not sufficient as a cookie verifier,
1858 		 * since it is not supposed to change when entries are
1859 		 * removed/added unless that offset cookies returned to
1860 		 * the client are no longer valid.
1861 		 */
1862 #if 0
1863 		if (nd->nd_flag & ND_NFSV4) {
1864 			nd->nd_repstat = NFSERR_NOTSAME;
1865 		} else {
1866 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1867 		}
1868 #endif
1869 	    } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1870 		nd->nd_repstat = NFSERR_BAD_COOKIE;
1871 	    }
1872 	}
1873 	if (!nd->nd_repstat && vp->v_type != VDIR)
1874 		nd->nd_repstat = NFSERR_NOTDIR;
1875 	if (!nd->nd_repstat && cnt == 0)
1876 		nd->nd_repstat = NFSERR_TOOSMALL;
1877 	if (!nd->nd_repstat)
1878 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1879 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1880 		    NFSACCCHK_VPISLOCKED, NULL);
1881 	if (nd->nd_repstat) {
1882 		vput(vp);
1883 		if (nd->nd_flag & ND_NFSV3)
1884 			nfsrv_postopattr(nd, getret, &at);
1885 		goto out;
1886 	}
1887 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1888 
1889 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1890 again:
1891 	eofflag = 0;
1892 	if (cookies) {
1893 		free((caddr_t)cookies, M_TEMP);
1894 		cookies = NULL;
1895 	}
1896 
1897 	iv.iov_base = rbuf;
1898 	iv.iov_len = siz;
1899 	io.uio_iov = &iv;
1900 	io.uio_iovcnt = 1;
1901 	io.uio_offset = (off_t)off;
1902 	io.uio_resid = siz;
1903 	io.uio_segflg = UIO_SYSSPACE;
1904 	io.uio_rw = UIO_READ;
1905 	io.uio_td = NULL;
1906 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1907 	    &cookies);
1908 	off = (u_int64_t)io.uio_offset;
1909 	if (io.uio_resid)
1910 		siz -= io.uio_resid;
1911 
1912 	getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1913 
1914 	if (!cookies && !nd->nd_repstat)
1915 		nd->nd_repstat = NFSERR_PERM;
1916 	if (!nd->nd_repstat)
1917 		nd->nd_repstat = getret;
1918 	if (nd->nd_repstat) {
1919 		vput(vp);
1920 		if (cookies)
1921 			free((caddr_t)cookies, M_TEMP);
1922 		free((caddr_t)rbuf, M_TEMP);
1923 		if (nd->nd_flag & ND_NFSV3)
1924 			nfsrv_postopattr(nd, getret, &at);
1925 		goto out;
1926 	}
1927 	/*
1928 	 * If nothing read, return eof
1929 	 * rpc reply
1930 	 */
1931 	if (siz == 0) {
1932 		vput(vp);
1933 		if (nd->nd_flag & ND_NFSV3)
1934 			nfsrv_postopattr(nd, getret, &at);
1935 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1936 		txdr_hyper(at.na_filerev, tl);
1937 		tl += 2;
1938 		*tl++ = newnfs_false;
1939 		*tl = newnfs_true;
1940 		free((caddr_t)cookies, M_TEMP);
1941 		free((caddr_t)rbuf, M_TEMP);
1942 		goto out;
1943 	}
1944 
1945 	/*
1946 	 * Check for degenerate cases of nothing useful read.
1947 	 * If so go try again
1948 	 */
1949 	cpos = rbuf;
1950 	cend = rbuf + siz;
1951 	dp = (struct dirent *)cpos;
1952 	cookiep = cookies;
1953 
1954 	/*
1955 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1956 	 * directory offset up to a block boundary, so it is necessary to
1957 	 * skip over the records that precede the requested offset. This
1958 	 * requires the assumption that file offset cookies monotonically
1959 	 * increase.
1960 	 * Since the offset cookies don't monotonically increase for ZFS,
1961 	 * this is not done when ZFS is the file system.
1962 	 */
1963 	while (cpos < cend && ncookies > 0 &&
1964 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1965 	   (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff) ||
1966 	   ((nd->nd_flag & ND_NFSV4) &&
1967 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1968 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1969 		cpos += dp->d_reclen;
1970 		dp = (struct dirent *)cpos;
1971 		cookiep++;
1972 		ncookies--;
1973 	}
1974 	if (cpos >= cend || ncookies == 0) {
1975 		siz = fullsiz;
1976 		toff = off;
1977 		goto again;
1978 	}
1979 
1980 	/*
1981 	 * Busy the file system so that the mount point won't go away
1982 	 * and, as such, VFS_VGET() can be used safely.
1983 	 */
1984 	mp = vp->v_mount;
1985 	vfs_ref(mp);
1986 	NFSVOPUNLOCK(vp, 0);
1987 	nd->nd_repstat = vfs_busy(mp, 0);
1988 	vfs_rel(mp);
1989 	if (nd->nd_repstat != 0) {
1990 		vrele(vp);
1991 		free(cookies, M_TEMP);
1992 		free(rbuf, M_TEMP);
1993 		if (nd->nd_flag & ND_NFSV3)
1994 			nfsrv_postopattr(nd, getret, &at);
1995 		goto out;
1996 	}
1997 
1998 	/*
1999 	 * Check to see if entries in this directory can be safely acquired
2000 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2001 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2002 	 * automount of the snapshot directory that is required will
2003 	 * be done.
2004 	 * This needs to be done here for NFSv4, since NFSv4 never does
2005 	 * a VFS_VGET() for "." or "..".
2006 	 */
2007 	if (not_zfs == 0) {
2008 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2009 		if (r == EOPNOTSUPP) {
2010 			usevget = 0;
2011 			cn.cn_nameiop = LOOKUP;
2012 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2013 			cn.cn_cred = nd->nd_cred;
2014 			cn.cn_thread = p;
2015 		} else if (r == 0)
2016 			vput(nvp);
2017 	}
2018 
2019 	/*
2020 	 * Save this position, in case there is an error before one entry
2021 	 * is created.
2022 	 */
2023 	mb0 = nd->nd_mb;
2024 	bpos0 = nd->nd_bpos;
2025 
2026 	/*
2027 	 * Fill in the first part of the reply.
2028 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2029 	 * (Include the two booleans at the end of the reply in dirlen now,
2030 	 *  so we recognize when we have exceeded cnt.)
2031 	 */
2032 	if (nd->nd_flag & ND_NFSV3) {
2033 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2034 		nfsrv_postopattr(nd, getret, &at);
2035 	} else {
2036 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2037 	}
2038 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2039 	txdr_hyper(at.na_filerev, tl);
2040 
2041 	/*
2042 	 * Save this position, in case there is an empty reply needed.
2043 	 */
2044 	mb1 = nd->nd_mb;
2045 	bpos1 = nd->nd_bpos;
2046 
2047 	/* Loop through the records and build reply */
2048 	entrycnt = 0;
2049 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2050 		nlen = dp->d_namlen;
2051 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2052 		    nlen <= NFS_MAXNAMLEN &&
2053 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2054 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2055 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2056 			/*
2057 			 * Save the current position in the reply, in case
2058 			 * this entry exceeds cnt.
2059 			 */
2060 			mb1 = nd->nd_mb;
2061 			bpos1 = nd->nd_bpos;
2062 
2063 			/*
2064 			 * For readdir_and_lookup get the vnode using
2065 			 * the file number.
2066 			 */
2067 			nvp = NULL;
2068 			refp = NULL;
2069 			r = 0;
2070 			at_root = 0;
2071 			needs_unbusy = 0;
2072 			new_mp = mp;
2073 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2074 			if ((nd->nd_flag & ND_NFSV3) ||
2075 			    NFSNONZERO_ATTRBIT(&savbits)) {
2076 				if (nd->nd_flag & ND_NFSV4)
2077 					refp = nfsv4root_getreferral(NULL,
2078 					    vp, dp->d_fileno);
2079 				if (refp == NULL) {
2080 					if (usevget)
2081 						r = VFS_VGET(mp, dp->d_fileno,
2082 						    LK_SHARED, &nvp);
2083 					else
2084 						r = EOPNOTSUPP;
2085 					if (r == EOPNOTSUPP) {
2086 						if (usevget) {
2087 							usevget = 0;
2088 							cn.cn_nameiop = LOOKUP;
2089 							cn.cn_lkflags =
2090 							    LK_SHARED |
2091 							    LK_RETRY;
2092 							cn.cn_cred =
2093 							    nd->nd_cred;
2094 							cn.cn_thread = p;
2095 						}
2096 						cn.cn_nameptr = dp->d_name;
2097 						cn.cn_namelen = nlen;
2098 						cn.cn_flags = ISLASTCN |
2099 						    NOFOLLOW | LOCKLEAF;
2100 						if (nlen == 2 &&
2101 						    dp->d_name[0] == '.' &&
2102 						    dp->d_name[1] == '.')
2103 							cn.cn_flags |=
2104 							    ISDOTDOT;
2105 						if (NFSVOPLOCK(vp, LK_SHARED)
2106 						    != 0) {
2107 							nd->nd_repstat = EPERM;
2108 							break;
2109 						}
2110 						if ((vp->v_vflag & VV_ROOT) != 0
2111 						    && (cn.cn_flags & ISDOTDOT)
2112 						    != 0) {
2113 							vref(vp);
2114 							nvp = vp;
2115 							r = 0;
2116 						} else {
2117 							r = VOP_LOOKUP(vp, &nvp,
2118 							    &cn);
2119 							if (vp != nvp)
2120 								NFSVOPUNLOCK(vp,
2121 								    0);
2122 						}
2123 					}
2124 
2125 					/*
2126 					 * For NFSv4, check to see if nvp is
2127 					 * a mount point and get the mount
2128 					 * point vnode, as required.
2129 					 */
2130 					if (r == 0 &&
2131 					    nfsrv_enable_crossmntpt != 0 &&
2132 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2133 					    nvp->v_type == VDIR &&
2134 					    nvp->v_mountedhere != NULL) {
2135 						new_mp = nvp->v_mountedhere;
2136 						r = vfs_busy(new_mp, 0);
2137 						vput(nvp);
2138 						nvp = NULL;
2139 						if (r == 0) {
2140 							r = VFS_ROOT(new_mp,
2141 							    LK_SHARED, &nvp);
2142 							needs_unbusy = 1;
2143 							if (r == 0)
2144 								at_root = 1;
2145 						}
2146 					}
2147 				}
2148 				if (!r) {
2149 				    if (refp == NULL &&
2150 					((nd->nd_flag & ND_NFSV3) ||
2151 					 NFSNONZERO_ATTRBIT(&attrbits))) {
2152 					r = nfsvno_getfh(nvp, &nfh, p);
2153 					if (!r)
2154 					    r = nfsvno_getattr(nvp, nvap,
2155 						nd->nd_cred, p, 1);
2156 					if (r == 0 && not_zfs == 0 &&
2157 					    nfsrv_enable_crossmntpt != 0 &&
2158 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2159 					    nvp->v_type == VDIR &&
2160 					    vp->v_mount != nvp->v_mount) {
2161 					    /*
2162 					     * For a ZFS snapshot, there is a
2163 					     * pseudo mount that does not set
2164 					     * v_mountedhere, so it needs to
2165 					     * be detected via a different
2166 					     * mount structure.
2167 					     */
2168 					    at_root = 1;
2169 					    if (new_mp == mp)
2170 						new_mp = nvp->v_mount;
2171 					}
2172 				    }
2173 				} else {
2174 				    nvp = NULL;
2175 				}
2176 				if (r) {
2177 					if (!NFSISSET_ATTRBIT(&attrbits,
2178 					    NFSATTRBIT_RDATTRERROR)) {
2179 						if (nvp != NULL)
2180 							vput(nvp);
2181 						if (needs_unbusy != 0)
2182 							vfs_unbusy(new_mp);
2183 						nd->nd_repstat = r;
2184 						break;
2185 					}
2186 				}
2187 			}
2188 
2189 			/*
2190 			 * Build the directory record xdr
2191 			 */
2192 			if (nd->nd_flag & ND_NFSV3) {
2193 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2194 				*tl++ = newnfs_true;
2195 				*tl++ = 0;
2196 				*tl = txdr_unsigned(dp->d_fileno);
2197 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2198 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2199 				*tl++ = 0;
2200 				*tl = txdr_unsigned(*cookiep);
2201 				nfsrv_postopattr(nd, 0, nvap);
2202 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2203 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2204 				if (nvp != NULL)
2205 					vput(nvp);
2206 			} else {
2207 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2208 				*tl++ = newnfs_true;
2209 				*tl++ = 0;
2210 				*tl = txdr_unsigned(*cookiep);
2211 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2212 				if (nvp != NULL) {
2213 					supports_nfsv4acls =
2214 					    nfs_supportsnfsv4acls(nvp);
2215 					NFSVOPUNLOCK(nvp, 0);
2216 				} else
2217 					supports_nfsv4acls = 0;
2218 				if (refp != NULL) {
2219 					dirlen += nfsrv_putreferralattr(nd,
2220 					    &savbits, refp, 0,
2221 					    &nd->nd_repstat);
2222 					if (nd->nd_repstat) {
2223 						if (nvp != NULL)
2224 							vrele(nvp);
2225 						if (needs_unbusy != 0)
2226 							vfs_unbusy(new_mp);
2227 						break;
2228 					}
2229 				} else if (r) {
2230 					dirlen += nfsvno_fillattr(nd, new_mp,
2231 					    nvp, nvap, &nfh, r, &rderrbits,
2232 					    nd->nd_cred, p, isdgram, 0,
2233 					    supports_nfsv4acls, at_root,
2234 					    mounted_on_fileno);
2235 				} else {
2236 					dirlen += nfsvno_fillattr(nd, new_mp,
2237 					    nvp, nvap, &nfh, r, &attrbits,
2238 					    nd->nd_cred, p, isdgram, 0,
2239 					    supports_nfsv4acls, at_root,
2240 					    mounted_on_fileno);
2241 				}
2242 				if (nvp != NULL)
2243 					vrele(nvp);
2244 				dirlen += (3 * NFSX_UNSIGNED);
2245 			}
2246 			if (needs_unbusy != 0)
2247 				vfs_unbusy(new_mp);
2248 			if (dirlen <= cnt)
2249 				entrycnt++;
2250 		}
2251 		cpos += dp->d_reclen;
2252 		dp = (struct dirent *)cpos;
2253 		cookiep++;
2254 		ncookies--;
2255 	}
2256 	vrele(vp);
2257 	vfs_unbusy(mp);
2258 
2259 	/*
2260 	 * If dirlen > cnt, we must strip off the last entry. If that
2261 	 * results in an empty reply, report NFSERR_TOOSMALL.
2262 	 */
2263 	if (dirlen > cnt || nd->nd_repstat) {
2264 		if (!nd->nd_repstat && entrycnt == 0)
2265 			nd->nd_repstat = NFSERR_TOOSMALL;
2266 		if (nd->nd_repstat)
2267 			newnfs_trimtrailing(nd, mb0, bpos0);
2268 		else
2269 			newnfs_trimtrailing(nd, mb1, bpos1);
2270 		eofflag = 0;
2271 	} else if (cpos < cend)
2272 		eofflag = 0;
2273 	if (!nd->nd_repstat) {
2274 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2275 		*tl++ = newnfs_false;
2276 		if (eofflag)
2277 			*tl = newnfs_true;
2278 		else
2279 			*tl = newnfs_false;
2280 	}
2281 	FREE((caddr_t)cookies, M_TEMP);
2282 	FREE((caddr_t)rbuf, M_TEMP);
2283 
2284 out:
2285 	NFSEXITCODE2(0, nd);
2286 	return (0);
2287 nfsmout:
2288 	vput(vp);
2289 	NFSEXITCODE2(error, nd);
2290 	return (error);
2291 }
2292 
2293 /*
2294  * Get the settable attributes out of the mbuf list.
2295  * (Return 0 or EBADRPC)
2296  */
2297 int
2298 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2299     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2300 {
2301 	u_int32_t *tl;
2302 	struct nfsv2_sattr *sp;
2303 	int error = 0, toclient = 0;
2304 
2305 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2306 	case ND_NFSV2:
2307 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2308 		/*
2309 		 * Some old clients didn't fill in the high order 16bits.
2310 		 * --> check the low order 2 bytes for 0xffff
2311 		 */
2312 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2313 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2314 		if (sp->sa_uid != newnfs_xdrneg1)
2315 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2316 		if (sp->sa_gid != newnfs_xdrneg1)
2317 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2318 		if (sp->sa_size != newnfs_xdrneg1)
2319 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2320 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2321 #ifdef notyet
2322 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2323 #else
2324 			nvap->na_atime.tv_sec =
2325 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2326 			nvap->na_atime.tv_nsec = 0;
2327 #endif
2328 		}
2329 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2330 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2331 		break;
2332 	case ND_NFSV3:
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_mode = nfstov_mode(*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_uid = fxdr_unsigned(uid_t, *tl);
2342 		}
2343 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2344 		if (*tl == newnfs_true) {
2345 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2346 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2347 		}
2348 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2349 		if (*tl == newnfs_true) {
2350 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2351 			nvap->na_size = fxdr_hyper(tl);
2352 		}
2353 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2354 		switch (fxdr_unsigned(int, *tl)) {
2355 		case NFSV3SATTRTIME_TOCLIENT:
2356 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2357 			fxdr_nfsv3time(tl, &nvap->na_atime);
2358 			toclient = 1;
2359 			break;
2360 		case NFSV3SATTRTIME_TOSERVER:
2361 			vfs_timestamp(&nvap->na_atime);
2362 			nvap->na_vaflags |= VA_UTIMES_NULL;
2363 			break;
2364 		};
2365 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2366 		switch (fxdr_unsigned(int, *tl)) {
2367 		case NFSV3SATTRTIME_TOCLIENT:
2368 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2369 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2370 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2371 			break;
2372 		case NFSV3SATTRTIME_TOSERVER:
2373 			vfs_timestamp(&nvap->na_mtime);
2374 			if (!toclient)
2375 				nvap->na_vaflags |= VA_UTIMES_NULL;
2376 			break;
2377 		};
2378 		break;
2379 	case ND_NFSV4:
2380 		error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2381 	};
2382 nfsmout:
2383 	NFSEXITCODE2(error, nd);
2384 	return (error);
2385 }
2386 
2387 /*
2388  * Handle the setable attributes for V4.
2389  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2390  */
2391 int
2392 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2393     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2394 {
2395 	u_int32_t *tl;
2396 	int attrsum = 0;
2397 	int i, j;
2398 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2399 	int toclient = 0;
2400 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2401 	uid_t uid;
2402 	gid_t gid;
2403 
2404 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2405 	if (error)
2406 		goto nfsmout;
2407 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2408 	attrsize = fxdr_unsigned(int, *tl);
2409 
2410 	/*
2411 	 * Loop around getting the setable attributes. If an unsupported
2412 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2413 	 */
2414 	if (retnotsup) {
2415 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2416 		bitpos = NFSATTRBIT_MAX;
2417 	} else {
2418 		bitpos = 0;
2419 	}
2420 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2421 	    if (attrsum > attrsize) {
2422 		error = NFSERR_BADXDR;
2423 		goto nfsmout;
2424 	    }
2425 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2426 		switch (bitpos) {
2427 		case NFSATTRBIT_SIZE:
2428 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2429 			nvap->na_size = fxdr_hyper(tl);
2430 			attrsum += NFSX_HYPER;
2431 			break;
2432 		case NFSATTRBIT_ACL:
2433 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2434 			    p);
2435 			if (error)
2436 				goto nfsmout;
2437 			if (aceerr && !nd->nd_repstat)
2438 				nd->nd_repstat = aceerr;
2439 			attrsum += aclsize;
2440 			break;
2441 		case NFSATTRBIT_ARCHIVE:
2442 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2443 			if (!nd->nd_repstat)
2444 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2445 			attrsum += NFSX_UNSIGNED;
2446 			break;
2447 		case NFSATTRBIT_HIDDEN:
2448 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2449 			if (!nd->nd_repstat)
2450 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2451 			attrsum += NFSX_UNSIGNED;
2452 			break;
2453 		case NFSATTRBIT_MIMETYPE:
2454 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2455 			i = fxdr_unsigned(int, *tl);
2456 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2457 			if (error)
2458 				goto nfsmout;
2459 			if (!nd->nd_repstat)
2460 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2461 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2462 			break;
2463 		case NFSATTRBIT_MODE:
2464 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2465 			nvap->na_mode = nfstov_mode(*tl);
2466 			attrsum += NFSX_UNSIGNED;
2467 			break;
2468 		case NFSATTRBIT_OWNER:
2469 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2470 			j = fxdr_unsigned(int, *tl);
2471 			if (j < 0) {
2472 				error = NFSERR_BADXDR;
2473 				goto nfsmout;
2474 			}
2475 			if (j > NFSV4_SMALLSTR)
2476 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2477 			else
2478 				cp = namestr;
2479 			error = nfsrv_mtostr(nd, cp, j);
2480 			if (error) {
2481 				if (j > NFSV4_SMALLSTR)
2482 					free(cp, M_NFSSTRING);
2483 				goto nfsmout;
2484 			}
2485 			if (!nd->nd_repstat) {
2486 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2487 				    p);
2488 				if (!nd->nd_repstat)
2489 					nvap->na_uid = uid;
2490 			}
2491 			if (j > NFSV4_SMALLSTR)
2492 				free(cp, M_NFSSTRING);
2493 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2494 			break;
2495 		case NFSATTRBIT_OWNERGROUP:
2496 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2497 			j = fxdr_unsigned(int, *tl);
2498 			if (j < 0) {
2499 				error = NFSERR_BADXDR;
2500 				goto nfsmout;
2501 			}
2502 			if (j > NFSV4_SMALLSTR)
2503 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2504 			else
2505 				cp = namestr;
2506 			error = nfsrv_mtostr(nd, cp, j);
2507 			if (error) {
2508 				if (j > NFSV4_SMALLSTR)
2509 					free(cp, M_NFSSTRING);
2510 				goto nfsmout;
2511 			}
2512 			if (!nd->nd_repstat) {
2513 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2514 				    p);
2515 				if (!nd->nd_repstat)
2516 					nvap->na_gid = gid;
2517 			}
2518 			if (j > NFSV4_SMALLSTR)
2519 				free(cp, M_NFSSTRING);
2520 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2521 			break;
2522 		case NFSATTRBIT_SYSTEM:
2523 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2524 			if (!nd->nd_repstat)
2525 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2526 			attrsum += NFSX_UNSIGNED;
2527 			break;
2528 		case NFSATTRBIT_TIMEACCESSSET:
2529 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2530 			attrsum += NFSX_UNSIGNED;
2531 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2532 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2533 			    fxdr_nfsv4time(tl, &nvap->na_atime);
2534 			    toclient = 1;
2535 			    attrsum += NFSX_V4TIME;
2536 			} else {
2537 			    vfs_timestamp(&nvap->na_atime);
2538 			    nvap->na_vaflags |= VA_UTIMES_NULL;
2539 			}
2540 			break;
2541 		case NFSATTRBIT_TIMEBACKUP:
2542 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2543 			if (!nd->nd_repstat)
2544 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2545 			attrsum += NFSX_V4TIME;
2546 			break;
2547 		case NFSATTRBIT_TIMECREATE:
2548 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2549 			if (!nd->nd_repstat)
2550 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2551 			attrsum += NFSX_V4TIME;
2552 			break;
2553 		case NFSATTRBIT_TIMEMODIFYSET:
2554 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2555 			attrsum += NFSX_UNSIGNED;
2556 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2557 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2558 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
2559 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
2560 			    attrsum += NFSX_V4TIME;
2561 			} else {
2562 			    vfs_timestamp(&nvap->na_mtime);
2563 			    if (!toclient)
2564 				nvap->na_vaflags |= VA_UTIMES_NULL;
2565 			}
2566 			break;
2567 		default:
2568 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2569 			/*
2570 			 * set bitpos so we drop out of the loop.
2571 			 */
2572 			bitpos = NFSATTRBIT_MAX;
2573 			break;
2574 		};
2575 	}
2576 
2577 	/*
2578 	 * some clients pad the attrlist, so we need to skip over the
2579 	 * padding.
2580 	 */
2581 	if (attrsum > attrsize) {
2582 		error = NFSERR_BADXDR;
2583 	} else {
2584 		attrsize = NFSM_RNDUP(attrsize);
2585 		if (attrsum < attrsize)
2586 			error = nfsm_advance(nd, attrsize - attrsum, -1);
2587 	}
2588 nfsmout:
2589 	NFSEXITCODE2(error, nd);
2590 	return (error);
2591 }
2592 
2593 /*
2594  * Check/setup export credentials.
2595  */
2596 int
2597 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2598     struct ucred *credanon)
2599 {
2600 	int error = 0;
2601 
2602 	/*
2603 	 * Check/setup credentials.
2604 	 */
2605 	if (nd->nd_flag & ND_GSS)
2606 		exp->nes_exflag &= ~MNT_EXPORTANON;
2607 
2608 	/*
2609 	 * Check to see if the operation is allowed for this security flavor.
2610 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2611 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2612 	 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2613 	 */
2614 	if (nfsvno_testexp(nd, exp) &&
2615 	    nd->nd_procnum != NFSV4OP_SECINFO &&
2616 	    nd->nd_procnum != NFSPROC_FSINFO) {
2617 		if (nd->nd_flag & ND_NFSV4)
2618 			error = NFSERR_WRONGSEC;
2619 		else
2620 			error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2621 		goto out;
2622 	}
2623 
2624 	/*
2625 	 * Check to see if the file system is exported V4 only.
2626 	 */
2627 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2628 		error = NFSERR_PROGNOTV4;
2629 		goto out;
2630 	}
2631 
2632 	/*
2633 	 * Now, map the user credentials.
2634 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
2635 	 *  Fsinfo RPC. If set for anything else, this code might need
2636 	 *  to change.)
2637 	 */
2638 	if (NFSVNO_EXPORTED(exp) &&
2639 	    ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2640 	     NFSVNO_EXPORTANON(exp) ||
2641 	     (nd->nd_flag & ND_AUTHNONE))) {
2642 		nd->nd_cred->cr_uid = credanon->cr_uid;
2643 		nd->nd_cred->cr_gid = credanon->cr_gid;
2644 		crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2645 		    credanon->cr_groups);
2646 	}
2647 
2648 out:
2649 	NFSEXITCODE2(error, nd);
2650 	return (error);
2651 }
2652 
2653 /*
2654  * Check exports.
2655  */
2656 int
2657 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2658     struct ucred **credp)
2659 {
2660 	int i, error, *secflavors;
2661 
2662 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2663 	    &exp->nes_numsecflavor, &secflavors);
2664 	if (error) {
2665 		if (nfs_rootfhset) {
2666 			exp->nes_exflag = 0;
2667 			exp->nes_numsecflavor = 0;
2668 			error = 0;
2669 		}
2670 	} else {
2671 		/* Copy the security flavors. */
2672 		for (i = 0; i < exp->nes_numsecflavor; i++)
2673 			exp->nes_secflavors[i] = secflavors[i];
2674 	}
2675 	NFSEXITCODE(error);
2676 	return (error);
2677 }
2678 
2679 /*
2680  * Get a vnode for a file handle and export stuff.
2681  */
2682 int
2683 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2684     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2685     struct ucred **credp)
2686 {
2687 	int i, error, *secflavors;
2688 
2689 	*credp = NULL;
2690 	exp->nes_numsecflavor = 0;
2691 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2692 	if (error != 0)
2693 		/* Make sure the server replies ESTALE to the client. */
2694 		error = ESTALE;
2695 	if (nam && !error) {
2696 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2697 		    &exp->nes_numsecflavor, &secflavors);
2698 		if (error) {
2699 			if (nfs_rootfhset) {
2700 				exp->nes_exflag = 0;
2701 				exp->nes_numsecflavor = 0;
2702 				error = 0;
2703 			} else {
2704 				vput(*vpp);
2705 			}
2706 		} else {
2707 			/* Copy the security flavors. */
2708 			for (i = 0; i < exp->nes_numsecflavor; i++)
2709 				exp->nes_secflavors[i] = secflavors[i];
2710 		}
2711 	}
2712 	NFSEXITCODE(error);
2713 	return (error);
2714 }
2715 
2716 /*
2717  * nfsd_fhtovp() - convert a fh to a vnode ptr
2718  * 	- look up fsid in mount list (if not found ret error)
2719  *	- get vp and export rights by calling nfsvno_fhtovp()
2720  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2721  *	  for AUTH_SYS
2722  *	- if mpp != NULL, return the mount point so that it can
2723  *	  be used for vn_finished_write() by the caller
2724  */
2725 void
2726 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2727     struct vnode **vpp, struct nfsexstuff *exp,
2728     struct mount **mpp, int startwrite, struct thread *p)
2729 {
2730 	struct mount *mp;
2731 	struct ucred *credanon;
2732 	fhandle_t *fhp;
2733 
2734 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
2735 	/*
2736 	 * Check for the special case of the nfsv4root_fh.
2737 	 */
2738 	mp = vfs_busyfs(&fhp->fh_fsid);
2739 	if (mpp != NULL)
2740 		*mpp = mp;
2741 	if (mp == NULL) {
2742 		*vpp = NULL;
2743 		nd->nd_repstat = ESTALE;
2744 		goto out;
2745 	}
2746 
2747 	if (startwrite) {
2748 		vn_start_write(NULL, mpp, V_WAIT);
2749 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
2750 			lktype = LK_EXCLUSIVE;
2751 	}
2752 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2753 	    &credanon);
2754 	vfs_unbusy(mp);
2755 
2756 	/*
2757 	 * For NFSv4 without a pseudo root fs, unexported file handles
2758 	 * can be returned, so that Lookup works everywhere.
2759 	 */
2760 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2761 	    !(nd->nd_flag & ND_NFSV4)) {
2762 		vput(*vpp);
2763 		nd->nd_repstat = EACCES;
2764 	}
2765 
2766 	/*
2767 	 * Personally, I've never seen any point in requiring a
2768 	 * reserved port#, since only in the rare case where the
2769 	 * clients are all boxes with secure system priviledges,
2770 	 * does it provide any enhanced security, but... some people
2771 	 * believe it to be useful and keep putting this code back in.
2772 	 * (There is also some "security checker" out there that
2773 	 *  complains if the nfs server doesn't enforce this.)
2774 	 * However, note the following:
2775 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
2776 	 *	required.
2777 	 * RFC2623 recommends that, if a reserved port# is checked for,
2778 	 *	that there be a way to turn that off--> ifdef'd.
2779 	 */
2780 #ifdef NFS_REQRSVPORT
2781 	if (!nd->nd_repstat) {
2782 		struct sockaddr_in *saddr;
2783 		struct sockaddr_in6 *saddr6;
2784 
2785 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2786 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2787 		if (!(nd->nd_flag & ND_NFSV4) &&
2788 		    ((saddr->sin_family == AF_INET &&
2789 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2790 		     (saddr6->sin6_family == AF_INET6 &&
2791 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2792 			vput(*vpp);
2793 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2794 		}
2795 	}
2796 #endif	/* NFS_REQRSVPORT */
2797 
2798 	/*
2799 	 * Check/setup credentials.
2800 	 */
2801 	if (!nd->nd_repstat) {
2802 		nd->nd_saveduid = nd->nd_cred->cr_uid;
2803 		nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2804 		if (nd->nd_repstat)
2805 			vput(*vpp);
2806 	}
2807 	if (credanon != NULL)
2808 		crfree(credanon);
2809 	if (nd->nd_repstat) {
2810 		if (startwrite)
2811 			vn_finished_write(mp);
2812 		*vpp = NULL;
2813 		if (mpp != NULL)
2814 			*mpp = NULL;
2815 	}
2816 
2817 out:
2818 	NFSEXITCODE2(0, nd);
2819 }
2820 
2821 /*
2822  * glue for fp.
2823  */
2824 static int
2825 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2826 {
2827 	struct filedesc *fdp;
2828 	struct file *fp;
2829 	int error = 0;
2830 
2831 	fdp = p->td_proc->p_fd;
2832 	if (fd < 0 || fd >= fdp->fd_nfiles ||
2833 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
2834 		error = EBADF;
2835 		goto out;
2836 	}
2837 	*fpp = fp;
2838 
2839 out:
2840 	NFSEXITCODE(error);
2841 	return (error);
2842 }
2843 
2844 /*
2845  * Called from nfssvc() to update the exports list. Just call
2846  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2847  * in the mount list.
2848  */
2849 int
2850 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2851 {
2852 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2853 	int error = 0;
2854 	struct nameidata nd;
2855 	fhandle_t fh;
2856 
2857 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2858 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2859 		nfs_rootfhset = 0;
2860 	else if (error == 0) {
2861 		if (nfsexargp->fspec == NULL) {
2862 			error = EPERM;
2863 			goto out;
2864 		}
2865 		/*
2866 		 * If fspec != NULL, this is the v4root path.
2867 		 */
2868 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
2869 		    nfsexargp->fspec, p);
2870 		if ((error = namei(&nd)) != 0)
2871 			goto out;
2872 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
2873 		vrele(nd.ni_vp);
2874 		if (!error) {
2875 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2876 			NFSBCOPY((caddr_t)&fh,
2877 			    nfs_rootfh.nfsrvfh_data,
2878 			    sizeof (fhandle_t));
2879 			nfs_rootfhset = 1;
2880 		}
2881 	}
2882 
2883 out:
2884 	NFSEXITCODE(error);
2885 	return (error);
2886 }
2887 
2888 /*
2889  * This function needs to test to see if the system is near its limit
2890  * for memory allocation via malloc() or mget() and return True iff
2891  * either of these resources are near their limit.
2892  * XXX (For now, this is just a stub.)
2893  */
2894 int nfsrv_testmalloclimit = 0;
2895 int
2896 nfsrv_mallocmget_limit(void)
2897 {
2898 	static int printmesg = 0;
2899 	static int testval = 1;
2900 
2901 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2902 		if ((printmesg++ % 100) == 0)
2903 			printf("nfsd: malloc/mget near limit\n");
2904 		return (1);
2905 	}
2906 	return (0);
2907 }
2908 
2909 /*
2910  * BSD specific initialization of a mount point.
2911  */
2912 void
2913 nfsd_mntinit(void)
2914 {
2915 	static int inited = 0;
2916 
2917 	if (inited)
2918 		return;
2919 	inited = 1;
2920 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2921 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2922 	TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2923 	nfsv4root_mnt.mnt_export = NULL;
2924 	TAILQ_INIT(&nfsv4root_opt);
2925 	TAILQ_INIT(&nfsv4root_newopt);
2926 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2927 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2928 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
2929 	nfsv4root_mnt.mnt_activevnodelistsize = 0;
2930 }
2931 
2932 /*
2933  * Get a vnode for a file handle, without checking exports, etc.
2934  */
2935 struct vnode *
2936 nfsvno_getvp(fhandle_t *fhp)
2937 {
2938 	struct mount *mp;
2939 	struct vnode *vp;
2940 	int error;
2941 
2942 	mp = vfs_busyfs(&fhp->fh_fsid);
2943 	if (mp == NULL)
2944 		return (NULL);
2945 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2946 	vfs_unbusy(mp);
2947 	if (error)
2948 		return (NULL);
2949 	return (vp);
2950 }
2951 
2952 /*
2953  * Do a local VOP_ADVLOCK().
2954  */
2955 int
2956 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2957     u_int64_t end, struct thread *td)
2958 {
2959 	int error = 0;
2960 	struct flock fl;
2961 	u_int64_t tlen;
2962 
2963 	if (nfsrv_dolocallocks == 0)
2964 		goto out;
2965 
2966 	/* Check for VI_DOOMED here, so that VOP_ADVLOCK() isn't performed. */
2967 	if ((vp->v_iflag & VI_DOOMED) != 0) {
2968 		error = EPERM;
2969 		goto out;
2970 	}
2971 
2972 	fl.l_whence = SEEK_SET;
2973 	fl.l_type = ftype;
2974 	fl.l_start = (off_t)first;
2975 	if (end == NFS64BITSSET) {
2976 		fl.l_len = 0;
2977 	} else {
2978 		tlen = end - first;
2979 		fl.l_len = (off_t)tlen;
2980 	}
2981 	/*
2982 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2983 	 * values for all calls, so that all locks will be held by the
2984 	 * nfsd server. (The nfsd server handles conflicts between the
2985 	 * various clients.)
2986 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2987 	 * bytes, so it can't be put in l_sysid.
2988 	 */
2989 	if (nfsv4_sysid == 0)
2990 		nfsv4_sysid = nlm_acquire_next_sysid();
2991 	fl.l_pid = (pid_t)0;
2992 	fl.l_sysid = (int)nfsv4_sysid;
2993 
2994 	NFSVOPUNLOCK(vp, 0);
2995 	if (ftype == F_UNLCK)
2996 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
2997 		    (F_POSIX | F_REMOTE));
2998 	else
2999 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3000 		    (F_POSIX | F_REMOTE));
3001 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3002 
3003 out:
3004 	NFSEXITCODE(error);
3005 	return (error);
3006 }
3007 
3008 /*
3009  * Check the nfsv4 root exports.
3010  */
3011 int
3012 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3013 {
3014 	struct ucred *credanon;
3015 	int exflags, error = 0, numsecflavor, *secflavors, i;
3016 
3017 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3018 	    &credanon, &numsecflavor, &secflavors);
3019 	if (error) {
3020 		error = NFSERR_PROGUNAVAIL;
3021 		goto out;
3022 	}
3023 	if (credanon != NULL)
3024 		crfree(credanon);
3025 	for (i = 0; i < numsecflavor; i++) {
3026 		if (secflavors[i] == AUTH_SYS)
3027 			nd->nd_flag |= ND_EXAUTHSYS;
3028 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3029 			nd->nd_flag |= ND_EXGSS;
3030 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3031 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3032 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3033 			nd->nd_flag |= ND_EXGSSPRIVACY;
3034 	}
3035 
3036 out:
3037 	NFSEXITCODE(error);
3038 	return (error);
3039 }
3040 
3041 /*
3042  * Nfs server psuedo system call for the nfsd's
3043  */
3044 /*
3045  * MPSAFE
3046  */
3047 static int
3048 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3049 {
3050 	struct file *fp;
3051 	struct nfsd_addsock_args sockarg;
3052 	struct nfsd_nfsd_args nfsdarg;
3053 	cap_rights_t rights;
3054 	int error;
3055 
3056 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3057 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3058 		if (error)
3059 			goto out;
3060 		/*
3061 		 * Since we don't know what rights might be required,
3062 		 * pretend that we need them all. It is better to be too
3063 		 * careful than too reckless.
3064 		 */
3065 		error = fget(td, sockarg.sock,
3066 		    cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3067 		if (error != 0)
3068 			goto out;
3069 		if (fp->f_type != DTYPE_SOCKET) {
3070 			fdrop(fp, td);
3071 			error = EPERM;
3072 			goto out;
3073 		}
3074 		error = nfsrvd_addsock(fp);
3075 		fdrop(fp, td);
3076 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3077 		if (uap->argp == NULL) {
3078 			error = EINVAL;
3079 			goto out;
3080 		}
3081 		error = copyin(uap->argp, (caddr_t)&nfsdarg,
3082 		    sizeof (nfsdarg));
3083 		if (error)
3084 			goto out;
3085 		error = nfsrvd_nfsd(td, &nfsdarg);
3086 	} else {
3087 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3088 	}
3089 
3090 out:
3091 	NFSEXITCODE(error);
3092 	return (error);
3093 }
3094 
3095 static int
3096 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3097 {
3098 	struct nfsex_args export;
3099 	struct file *fp = NULL;
3100 	int stablefd, len;
3101 	struct nfsd_clid adminrevoke;
3102 	struct nfsd_dumplist dumplist;
3103 	struct nfsd_dumpclients *dumpclients;
3104 	struct nfsd_dumplocklist dumplocklist;
3105 	struct nfsd_dumplocks *dumplocks;
3106 	struct nameidata nd;
3107 	vnode_t vp;
3108 	int error = EINVAL, igotlock;
3109 	struct proc *procp;
3110 	static int suspend_nfsd = 0;
3111 
3112 	if (uap->flag & NFSSVC_PUBLICFH) {
3113 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3114 		    sizeof (fhandle_t));
3115 		error = copyin(uap->argp,
3116 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3117 		if (!error)
3118 			nfs_pubfhset = 1;
3119 	} else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3120 		error = copyin(uap->argp,(caddr_t)&export,
3121 		    sizeof (struct nfsex_args));
3122 		if (!error)
3123 			error = nfsrv_v4rootexport(&export, cred, p);
3124 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3125 		nfs_pubfhset = 0;
3126 		error = 0;
3127 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3128 		error = copyin(uap->argp, (caddr_t)&stablefd,
3129 		    sizeof (int));
3130 		if (!error)
3131 			error = fp_getfvp(p, stablefd, &fp, &vp);
3132 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3133 			error = EBADF;
3134 		if (!error && newnfs_numnfsd != 0)
3135 			error = EPERM;
3136 		if (!error) {
3137 			nfsrv_stablefirst.nsf_fp = fp;
3138 			nfsrv_setupstable(p);
3139 		}
3140 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3141 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3142 		    sizeof (struct nfsd_clid));
3143 		if (!error)
3144 			error = nfsrv_adminrevoke(&adminrevoke, p);
3145 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3146 		error = copyin(uap->argp, (caddr_t)&dumplist,
3147 		    sizeof (struct nfsd_dumplist));
3148 		if (!error && (dumplist.ndl_size < 1 ||
3149 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3150 			error = EPERM;
3151 		if (!error) {
3152 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3153 		    dumpclients = (struct nfsd_dumpclients *)malloc(len,
3154 			M_TEMP, M_WAITOK);
3155 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3156 		    error = copyout(dumpclients,
3157 			CAST_USER_ADDR_T(dumplist.ndl_list), len);
3158 		    free((caddr_t)dumpclients, M_TEMP);
3159 		}
3160 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3161 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3162 		    sizeof (struct nfsd_dumplocklist));
3163 		if (!error && (dumplocklist.ndllck_size < 1 ||
3164 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3165 			error = EPERM;
3166 		if (!error)
3167 			error = nfsrv_lookupfilename(&nd,
3168 				dumplocklist.ndllck_fname, p);
3169 		if (!error) {
3170 			len = sizeof (struct nfsd_dumplocks) *
3171 				dumplocklist.ndllck_size;
3172 			dumplocks = (struct nfsd_dumplocks *)malloc(len,
3173 				M_TEMP, M_WAITOK);
3174 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3175 			    dumplocklist.ndllck_size, p);
3176 			vput(nd.ni_vp);
3177 			error = copyout(dumplocks,
3178 			    CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3179 			free((caddr_t)dumplocks, M_TEMP);
3180 		}
3181 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3182 		procp = p->td_proc;
3183 		PROC_LOCK(procp);
3184 		nfsd_master_pid = procp->p_pid;
3185 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3186 		nfsd_master_start = procp->p_stats->p_start;
3187 		nfsd_master_proc = procp;
3188 		PROC_UNLOCK(procp);
3189 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3190 		NFSLOCKV4ROOTMUTEX();
3191 		if (suspend_nfsd == 0) {
3192 			/* Lock out all nfsd threads */
3193 			do {
3194 				igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3195 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3196 			} while (igotlock == 0 && suspend_nfsd == 0);
3197 			suspend_nfsd = 1;
3198 		}
3199 		NFSUNLOCKV4ROOTMUTEX();
3200 		error = 0;
3201 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3202 		NFSLOCKV4ROOTMUTEX();
3203 		if (suspend_nfsd != 0) {
3204 			nfsv4_unlock(&nfsd_suspend_lock, 0);
3205 			suspend_nfsd = 0;
3206 		}
3207 		NFSUNLOCKV4ROOTMUTEX();
3208 		error = 0;
3209 	}
3210 
3211 	NFSEXITCODE(error);
3212 	return (error);
3213 }
3214 
3215 /*
3216  * Check exports.
3217  * Returns 0 if ok, 1 otherwise.
3218  */
3219 int
3220 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3221 {
3222 	int i;
3223 
3224 	/*
3225 	 * This seems odd, but allow the case where the security flavor
3226 	 * list is empty. This happens when NFSv4 is traversing non-exported
3227 	 * file systems. Exported file systems should always have a non-empty
3228 	 * security flavor list.
3229 	 */
3230 	if (exp->nes_numsecflavor == 0)
3231 		return (0);
3232 
3233 	for (i = 0; i < exp->nes_numsecflavor; i++) {
3234 		/*
3235 		 * The tests for privacy and integrity must be first,
3236 		 * since ND_GSS is set for everything but AUTH_SYS.
3237 		 */
3238 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3239 		    (nd->nd_flag & ND_GSSPRIVACY))
3240 			return (0);
3241 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3242 		    (nd->nd_flag & ND_GSSINTEGRITY))
3243 			return (0);
3244 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3245 		    (nd->nd_flag & ND_GSS))
3246 			return (0);
3247 		if (exp->nes_secflavors[i] == AUTH_SYS &&
3248 		    (nd->nd_flag & ND_GSS) == 0)
3249 			return (0);
3250 	}
3251 	return (1);
3252 }
3253 
3254 /*
3255  * Calculate a hash value for the fid in a file handle.
3256  */
3257 uint32_t
3258 nfsrv_hashfh(fhandle_t *fhp)
3259 {
3260 	uint32_t hashval;
3261 
3262 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3263 	return (hashval);
3264 }
3265 
3266 /*
3267  * Signal the userland master nfsd to backup the stable restart file.
3268  */
3269 void
3270 nfsrv_backupstable(void)
3271 {
3272 	struct proc *procp;
3273 
3274 	if (nfsd_master_proc != NULL) {
3275 		procp = pfind(nfsd_master_pid);
3276 		/* Try to make sure it is the correct process. */
3277 		if (procp == nfsd_master_proc &&
3278 		    procp->p_stats->p_start.tv_sec ==
3279 		    nfsd_master_start.tv_sec &&
3280 		    procp->p_stats->p_start.tv_usec ==
3281 		    nfsd_master_start.tv_usec &&
3282 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
3283 			kern_psignal(procp, SIGUSR2);
3284 		else
3285 			nfsd_master_proc = NULL;
3286 
3287 		if (procp != NULL)
3288 			PROC_UNLOCK(procp);
3289 	}
3290 }
3291 
3292 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3293 
3294 /*
3295  * Called once to initialize data structures...
3296  */
3297 static int
3298 nfsd_modevent(module_t mod, int type, void *data)
3299 {
3300 	int error = 0, i;
3301 	static int loaded = 0;
3302 
3303 	switch (type) {
3304 	case MOD_LOAD:
3305 		if (loaded)
3306 			goto out;
3307 		newnfs_portinit();
3308 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3309 			snprintf(nfsrchash_table[i].lock_name,
3310 			    sizeof(nfsrchash_table[i].lock_name), "nfsrc_tcp%d",
3311 			    i);
3312 			mtx_init(&nfsrchash_table[i].mtx,
3313 			    nfsrchash_table[i].lock_name, NULL, MTX_DEF);
3314 			snprintf(nfsrcahash_table[i].lock_name,
3315 			    sizeof(nfsrcahash_table[i].lock_name), "nfsrc_tcpa%d",
3316 			    i);
3317 			mtx_init(&nfsrcahash_table[i].mtx,
3318 			    nfsrcahash_table[i].lock_name, NULL, MTX_DEF);
3319 		}
3320 		mtx_init(&nfsrc_udpmtx, "nfs_udpcache_mutex", NULL, MTX_DEF);
3321 		mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3322 		mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3323 		    MTX_DEF);
3324 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3325 		nfsrvd_initcache();
3326 		nfsd_init();
3327 		NFSD_LOCK();
3328 		nfsrvd_init(0);
3329 		NFSD_UNLOCK();
3330 		nfsd_mntinit();
3331 #ifdef VV_DISABLEDELEG
3332 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3333 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3334 #endif
3335 		nfsd_call_servertimer = nfsrv_servertimer;
3336 		nfsd_call_nfsd = nfssvc_nfsd;
3337 		loaded = 1;
3338 		break;
3339 
3340 	case MOD_UNLOAD:
3341 		if (newnfs_numnfsd != 0) {
3342 			error = EBUSY;
3343 			break;
3344 		}
3345 
3346 #ifdef VV_DISABLEDELEG
3347 		vn_deleg_ops.vndeleg_recall = NULL;
3348 		vn_deleg_ops.vndeleg_disable = NULL;
3349 #endif
3350 		nfsd_call_servertimer = NULL;
3351 		nfsd_call_nfsd = NULL;
3352 
3353 		/* Clean out all NFSv4 state. */
3354 		nfsrv_throwawayallstate(curthread);
3355 
3356 		/* Clean the NFS server reply cache */
3357 		nfsrvd_cleancache();
3358 
3359 		/* Free up the krpc server pool. */
3360 		if (nfsrvd_pool != NULL)
3361 			svcpool_destroy(nfsrvd_pool);
3362 
3363 		/* and get rid of the locks */
3364 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3365 			mtx_destroy(&nfsrchash_table[i].mtx);
3366 			mtx_destroy(&nfsrcahash_table[i].mtx);
3367 		}
3368 		mtx_destroy(&nfsrc_udpmtx);
3369 		mtx_destroy(&nfs_v4root_mutex);
3370 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3371 		lockdestroy(&nfsv4root_mnt.mnt_explock);
3372 		loaded = 0;
3373 		break;
3374 	default:
3375 		error = EOPNOTSUPP;
3376 		break;
3377 	}
3378 
3379 out:
3380 	NFSEXITCODE(error);
3381 	return (error);
3382 }
3383 static moduledata_t nfsd_mod = {
3384 	"nfsd",
3385 	nfsd_modevent,
3386 	NULL,
3387 };
3388 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3389 
3390 /* So that loader and kldload(2) can find us, wherever we are.. */
3391 MODULE_VERSION(nfsd, 1);
3392 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3393 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3394 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3395 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3396 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3397 
3398