xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision 3e65b9c6e6b7b2081d54e1dc40983c3c00eaf738)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include <sys/capability.h>
38 
39 /*
40  * Functions that perform the vfs operations required by the routines in
41  * nfsd_serv.c. It is hoped that this change will make the server more
42  * portable.
43  */
44 
45 #include <fs/nfs/nfsport.h>
46 #include <sys/hash.h>
47 #include <sys/sysctl.h>
48 #include <nlm/nlm_prot.h>
49 #include <nlm/nlm.h>
50 
51 FEATURE(nfsd, "NFSv4 server");
52 
53 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
54 extern int nfsrv_useacl;
55 extern int newnfs_numnfsd;
56 extern struct mount nfsv4root_mnt;
57 extern struct nfsrv_stablefirst nfsrv_stablefirst;
58 extern void (*nfsd_call_servertimer)(void);
59 extern SVCPOOL	*nfsrvd_pool;
60 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
61 NFSDLOCKMUTEX;
62 struct mtx nfs_cache_mutex;
63 struct mtx nfs_v4root_mutex;
64 struct nfsrvfh nfs_rootfh, nfs_pubfh;
65 int nfs_pubfhset = 0, nfs_rootfhset = 0;
66 struct proc *nfsd_master_proc = NULL;
67 static pid_t nfsd_master_pid = (pid_t)-1;
68 static char nfsd_master_comm[MAXCOMLEN + 1];
69 static struct timeval nfsd_master_start;
70 static uint32_t nfsv4_sysid = 0;
71 
72 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
73     struct ucred *);
74 
75 int nfsrv_enable_crossmntpt = 1;
76 static int nfs_commit_blks;
77 static int nfs_commit_miss;
78 extern int nfsrv_issuedelegs;
79 extern int nfsrv_dolocallocks;
80 
81 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server");
82 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
83     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
84 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
85     0, "");
86 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
87     0, "");
88 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
89     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
90 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
91     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
92 
93 #define	MAX_REORDERED_RPC	16
94 #define	NUM_HEURISTIC		1031
95 #define	NHUSE_INIT		64
96 #define	NHUSE_INC		16
97 #define	NHUSE_MAX		2048
98 
99 static struct nfsheur {
100 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
101 	off_t nh_nextoff;	/* next offset for sequential detection */
102 	int nh_use;		/* use count for selection */
103 	int nh_seqcount;	/* heuristic */
104 } nfsheur[NUM_HEURISTIC];
105 
106 
107 /*
108  * Heuristic to detect sequential operation.
109  */
110 static struct nfsheur *
111 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
112 {
113 	struct nfsheur *nh;
114 	int hi, try;
115 
116 	/* Locate best candidate. */
117 	try = 32;
118 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
119 	nh = &nfsheur[hi];
120 	while (try--) {
121 		if (nfsheur[hi].nh_vp == vp) {
122 			nh = &nfsheur[hi];
123 			break;
124 		}
125 		if (nfsheur[hi].nh_use > 0)
126 			--nfsheur[hi].nh_use;
127 		hi = (hi + 1) % NUM_HEURISTIC;
128 		if (nfsheur[hi].nh_use < nh->nh_use)
129 			nh = &nfsheur[hi];
130 	}
131 
132 	/* Initialize hint if this is a new file. */
133 	if (nh->nh_vp != vp) {
134 		nh->nh_vp = vp;
135 		nh->nh_nextoff = uio->uio_offset;
136 		nh->nh_use = NHUSE_INIT;
137 		if (uio->uio_offset == 0)
138 			nh->nh_seqcount = 4;
139 		else
140 			nh->nh_seqcount = 1;
141 	}
142 
143 	/* Calculate heuristic. */
144 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
145 	    uio->uio_offset == nh->nh_nextoff) {
146 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
147 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
148 		if (nh->nh_seqcount > IO_SEQMAX)
149 			nh->nh_seqcount = IO_SEQMAX;
150 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
151 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
152 		/* Probably a reordered RPC, leave seqcount alone. */
153 	} else if (nh->nh_seqcount > 1) {
154 		nh->nh_seqcount /= 2;
155 	} else {
156 		nh->nh_seqcount = 0;
157 	}
158 	nh->nh_use += NHUSE_INC;
159 	if (nh->nh_use > NHUSE_MAX)
160 		nh->nh_use = NHUSE_MAX;
161 	return (nh);
162 }
163 
164 /*
165  * Get attributes into nfsvattr structure.
166  */
167 int
168 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
169     struct thread *p, int vpislocked)
170 {
171 	int error, lockedit = 0;
172 
173 	if (vpislocked == 0) {
174 		/*
175 		 * When vpislocked == 0, the vnode is either exclusively
176 		 * locked by this thread or not locked by this thread.
177 		 * As such, shared lock it, if not exclusively locked.
178 		 */
179 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
180 			lockedit = 1;
181 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
182 		}
183 	}
184 	error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
185 	if (lockedit != 0)
186 		NFSVOPUNLOCK(vp, 0);
187 
188 	NFSEXITCODE(error);
189 	return (error);
190 }
191 
192 /*
193  * Get a file handle for a vnode.
194  */
195 int
196 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
197 {
198 	int error;
199 
200 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
201 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
202 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
203 
204 	NFSEXITCODE(error);
205 	return (error);
206 }
207 
208 /*
209  * Perform access checking for vnodes obtained from file handles that would
210  * refer to files already opened by a Unix client. You cannot just use
211  * vn_writechk() and VOP_ACCESSX() for two reasons.
212  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
213  *     case.
214  * 2 - The owner is to be given access irrespective of mode bits for some
215  *     operations, so that processes that chmod after opening a file don't
216  *     break.
217  */
218 int
219 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
220     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
221     u_int32_t *supportedtypep)
222 {
223 	struct vattr vattr;
224 	int error = 0, getret = 0;
225 
226 	if (vpislocked == 0) {
227 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
228 			error = EPERM;
229 			goto out;
230 		}
231 	}
232 	if (accmode & VWRITE) {
233 		/* Just vn_writechk() changed to check rdonly */
234 		/*
235 		 * Disallow write attempts on read-only file systems;
236 		 * unless the file is a socket or a block or character
237 		 * device resident on the file system.
238 		 */
239 		if (NFSVNO_EXRDONLY(exp) ||
240 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
241 			switch (vp->v_type) {
242 			case VREG:
243 			case VDIR:
244 			case VLNK:
245 				error = EROFS;
246 			default:
247 				break;
248 			}
249 		}
250 		/*
251 		 * If there's shared text associated with
252 		 * the inode, try to free it up once.  If
253 		 * we fail, we can't allow writing.
254 		 */
255 		if ((vp->v_vflag & VV_TEXT) != 0 && error == 0)
256 			error = ETXTBSY;
257 	}
258 	if (error != 0) {
259 		if (vpislocked == 0)
260 			NFSVOPUNLOCK(vp, 0);
261 		goto out;
262 	}
263 
264 	/*
265 	 * Should the override still be applied when ACLs are enabled?
266 	 */
267 	error = VOP_ACCESSX(vp, accmode, cred, p);
268 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
269 		/*
270 		 * Try again with VEXPLICIT_DENY, to see if the test for
271 		 * deletion is supported.
272 		 */
273 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
274 		if (error == 0) {
275 			if (vp->v_type == VDIR) {
276 				accmode &= ~(VDELETE | VDELETE_CHILD);
277 				accmode |= VWRITE;
278 				error = VOP_ACCESSX(vp, accmode, cred, p);
279 			} else if (supportedtypep != NULL) {
280 				*supportedtypep &= ~NFSACCESS_DELETE;
281 			}
282 		}
283 	}
284 
285 	/*
286 	 * Allow certain operations for the owner (reads and writes
287 	 * on files that are already open).
288 	 */
289 	if (override != NFSACCCHK_NOOVERRIDE &&
290 	    (error == EPERM || error == EACCES)) {
291 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
292 			error = 0;
293 		else if (override & NFSACCCHK_ALLOWOWNER) {
294 			getret = VOP_GETATTR(vp, &vattr, cred);
295 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
296 				error = 0;
297 		}
298 	}
299 	if (vpislocked == 0)
300 		NFSVOPUNLOCK(vp, 0);
301 
302 out:
303 	NFSEXITCODE(error);
304 	return (error);
305 }
306 
307 /*
308  * Set attribute(s) vnop.
309  */
310 int
311 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
312     struct thread *p, struct nfsexstuff *exp)
313 {
314 	int error;
315 
316 	error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
317 	NFSEXITCODE(error);
318 	return (error);
319 }
320 
321 /*
322  * Set up nameidata for a lookup() call and do it
323  * For the cases where we are crossing mount points
324  * (looking up the public fh path or the v4 root path when
325  *  not using a pseudo-root fs), set/release the Giant lock,
326  * as required.
327  */
328 int
329 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
330     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
331     struct vnode **retdirp)
332 {
333 	struct componentname *cnp = &ndp->ni_cnd;
334 	int i;
335 	struct iovec aiov;
336 	struct uio auio;
337 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
338 	int error = 0, crossmnt;
339 	char *cp;
340 
341 	*retdirp = NULL;
342 	cnp->cn_nameptr = cnp->cn_pnbuf;
343 	ndp->ni_strictrelative = 0;
344 	/*
345 	 * Extract and set starting directory.
346 	 */
347 	if (dp->v_type != VDIR) {
348 		if (islocked)
349 			vput(dp);
350 		else
351 			vrele(dp);
352 		nfsvno_relpathbuf(ndp);
353 		error = ENOTDIR;
354 		goto out1;
355 	}
356 	if (islocked)
357 		NFSVOPUNLOCK(dp, 0);
358 	VREF(dp);
359 	*retdirp = dp;
360 	if (NFSVNO_EXRDONLY(exp))
361 		cnp->cn_flags |= RDONLY;
362 	ndp->ni_segflg = UIO_SYSSPACE;
363 	crossmnt = 1;
364 
365 	if (nd->nd_flag & ND_PUBLOOKUP) {
366 		ndp->ni_loopcnt = 0;
367 		if (cnp->cn_pnbuf[0] == '/') {
368 			vrele(dp);
369 			/*
370 			 * Check for degenerate pathnames here, since lookup()
371 			 * panics on them.
372 			 */
373 			for (i = 1; i < ndp->ni_pathlen; i++)
374 				if (cnp->cn_pnbuf[i] != '/')
375 					break;
376 			if (i == ndp->ni_pathlen) {
377 				error = NFSERR_ACCES;
378 				goto out;
379 			}
380 			dp = rootvnode;
381 			VREF(dp);
382 		}
383 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
384 	    (nd->nd_flag & ND_NFSV4) == 0) {
385 		/*
386 		 * Only cross mount points for NFSv4 when doing a
387 		 * mount while traversing the file system above
388 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
389 		 */
390 		cnp->cn_flags |= NOCROSSMOUNT;
391 		crossmnt = 0;
392 	}
393 
394 	/*
395 	 * Initialize for scan, set ni_startdir and bump ref on dp again
396 	 * becuase lookup() will dereference ni_startdir.
397 	 */
398 
399 	cnp->cn_thread = p;
400 	ndp->ni_startdir = dp;
401 	ndp->ni_rootdir = rootvnode;
402 
403 	if (!lockleaf)
404 		cnp->cn_flags |= LOCKLEAF;
405 	for (;;) {
406 		cnp->cn_nameptr = cnp->cn_pnbuf;
407 		/*
408 		 * Call lookup() to do the real work.  If an error occurs,
409 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
410 		 * we do not have to dereference anything before returning.
411 		 * In either case ni_startdir will be dereferenced and NULLed
412 		 * out.
413 		 */
414 		error = lookup(ndp);
415 		if (error)
416 			break;
417 
418 		/*
419 		 * Check for encountering a symbolic link.  Trivial
420 		 * termination occurs if no symlink encountered.
421 		 */
422 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
423 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
424 				nfsvno_relpathbuf(ndp);
425 			if (ndp->ni_vp && !lockleaf)
426 				NFSVOPUNLOCK(ndp->ni_vp, 0);
427 			break;
428 		}
429 
430 		/*
431 		 * Validate symlink
432 		 */
433 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
434 			NFSVOPUNLOCK(ndp->ni_dvp, 0);
435 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
436 			error = EINVAL;
437 			goto badlink2;
438 		}
439 
440 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
441 			error = ELOOP;
442 			goto badlink2;
443 		}
444 		if (ndp->ni_pathlen > 1)
445 			cp = uma_zalloc(namei_zone, M_WAITOK);
446 		else
447 			cp = cnp->cn_pnbuf;
448 		aiov.iov_base = cp;
449 		aiov.iov_len = MAXPATHLEN;
450 		auio.uio_iov = &aiov;
451 		auio.uio_iovcnt = 1;
452 		auio.uio_offset = 0;
453 		auio.uio_rw = UIO_READ;
454 		auio.uio_segflg = UIO_SYSSPACE;
455 		auio.uio_td = NULL;
456 		auio.uio_resid = MAXPATHLEN;
457 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
458 		if (error) {
459 		badlink1:
460 			if (ndp->ni_pathlen > 1)
461 				uma_zfree(namei_zone, cp);
462 		badlink2:
463 			vrele(ndp->ni_dvp);
464 			vput(ndp->ni_vp);
465 			break;
466 		}
467 		linklen = MAXPATHLEN - auio.uio_resid;
468 		if (linklen == 0) {
469 			error = ENOENT;
470 			goto badlink1;
471 		}
472 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
473 			error = ENAMETOOLONG;
474 			goto badlink1;
475 		}
476 
477 		/*
478 		 * Adjust or replace path
479 		 */
480 		if (ndp->ni_pathlen > 1) {
481 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
482 			uma_zfree(namei_zone, cnp->cn_pnbuf);
483 			cnp->cn_pnbuf = cp;
484 		} else
485 			cnp->cn_pnbuf[linklen] = '\0';
486 		ndp->ni_pathlen += linklen;
487 
488 		/*
489 		 * Cleanup refs for next loop and check if root directory
490 		 * should replace current directory.  Normally ni_dvp
491 		 * becomes the new base directory and is cleaned up when
492 		 * we loop.  Explicitly null pointers after invalidation
493 		 * to clarify operation.
494 		 */
495 		vput(ndp->ni_vp);
496 		ndp->ni_vp = NULL;
497 
498 		if (cnp->cn_pnbuf[0] == '/') {
499 			vrele(ndp->ni_dvp);
500 			ndp->ni_dvp = ndp->ni_rootdir;
501 			VREF(ndp->ni_dvp);
502 		}
503 		ndp->ni_startdir = ndp->ni_dvp;
504 		ndp->ni_dvp = NULL;
505 	}
506 	if (!lockleaf)
507 		cnp->cn_flags &= ~LOCKLEAF;
508 
509 out:
510 	if (error) {
511 		uma_zfree(namei_zone, cnp->cn_pnbuf);
512 		ndp->ni_vp = NULL;
513 		ndp->ni_dvp = NULL;
514 		ndp->ni_startdir = NULL;
515 		cnp->cn_flags &= ~HASBUF;
516 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
517 		ndp->ni_dvp = NULL;
518 	}
519 
520 out1:
521 	NFSEXITCODE2(error, nd);
522 	return (error);
523 }
524 
525 /*
526  * Set up a pathname buffer and return a pointer to it and, optionally
527  * set a hash pointer.
528  */
529 void
530 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
531 {
532 	struct componentname *cnp = &ndp->ni_cnd;
533 
534 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
535 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
536 	if (hashpp != NULL)
537 		*hashpp = NULL;
538 	*bufpp = cnp->cn_pnbuf;
539 }
540 
541 /*
542  * Release the above path buffer, if not released by nfsvno_namei().
543  */
544 void
545 nfsvno_relpathbuf(struct nameidata *ndp)
546 {
547 
548 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
549 		panic("nfsrelpath");
550 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
551 	ndp->ni_cnd.cn_flags &= ~HASBUF;
552 }
553 
554 /*
555  * Readlink vnode op into an mbuf list.
556  */
557 int
558 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
559     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
560 {
561 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
562 	struct iovec *ivp = iv;
563 	struct uio io, *uiop = &io;
564 	struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
565 	int i, len, tlen, error = 0;
566 
567 	len = 0;
568 	i = 0;
569 	while (len < NFS_MAXPATHLEN) {
570 		NFSMGET(mp);
571 		MCLGET(mp, M_WAIT);
572 		mp->m_len = NFSMSIZ(mp);
573 		if (len == 0) {
574 			mp3 = mp2 = mp;
575 		} else {
576 			mp2->m_next = mp;
577 			mp2 = mp;
578 		}
579 		if ((len + mp->m_len) > NFS_MAXPATHLEN) {
580 			mp->m_len = NFS_MAXPATHLEN - len;
581 			len = NFS_MAXPATHLEN;
582 		} else {
583 			len += mp->m_len;
584 		}
585 		ivp->iov_base = mtod(mp, caddr_t);
586 		ivp->iov_len = mp->m_len;
587 		i++;
588 		ivp++;
589 	}
590 	uiop->uio_iov = iv;
591 	uiop->uio_iovcnt = i;
592 	uiop->uio_offset = 0;
593 	uiop->uio_resid = len;
594 	uiop->uio_rw = UIO_READ;
595 	uiop->uio_segflg = UIO_SYSSPACE;
596 	uiop->uio_td = NULL;
597 	error = VOP_READLINK(vp, uiop, cred);
598 	if (error) {
599 		m_freem(mp3);
600 		*lenp = 0;
601 		goto out;
602 	}
603 	if (uiop->uio_resid > 0) {
604 		len -= uiop->uio_resid;
605 		tlen = NFSM_RNDUP(len);
606 		nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
607 	}
608 	*lenp = len;
609 	*mpp = mp3;
610 	*mpendp = mp;
611 
612 out:
613 	NFSEXITCODE(error);
614 	return (error);
615 }
616 
617 /*
618  * Read vnode op call into mbuf list.
619  */
620 int
621 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
622     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
623 {
624 	struct mbuf *m;
625 	int i;
626 	struct iovec *iv;
627 	struct iovec *iv2;
628 	int error = 0, len, left, siz, tlen, ioflag = 0;
629 	struct mbuf *m2 = NULL, *m3;
630 	struct uio io, *uiop = &io;
631 	struct nfsheur *nh;
632 
633 	len = left = NFSM_RNDUP(cnt);
634 	m3 = NULL;
635 	/*
636 	 * Generate the mbuf list with the uio_iov ref. to it.
637 	 */
638 	i = 0;
639 	while (left > 0) {
640 		NFSMGET(m);
641 		MCLGET(m, M_WAIT);
642 		m->m_len = 0;
643 		siz = min(M_TRAILINGSPACE(m), left);
644 		left -= siz;
645 		i++;
646 		if (m3)
647 			m2->m_next = m;
648 		else
649 			m3 = m;
650 		m2 = m;
651 	}
652 	MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
653 	    M_TEMP, M_WAITOK);
654 	uiop->uio_iov = iv2 = iv;
655 	m = m3;
656 	left = len;
657 	i = 0;
658 	while (left > 0) {
659 		if (m == NULL)
660 			panic("nfsvno_read iov");
661 		siz = min(M_TRAILINGSPACE(m), left);
662 		if (siz > 0) {
663 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
664 			iv->iov_len = siz;
665 			m->m_len += siz;
666 			left -= siz;
667 			iv++;
668 			i++;
669 		}
670 		m = m->m_next;
671 	}
672 	uiop->uio_iovcnt = i;
673 	uiop->uio_offset = off;
674 	uiop->uio_resid = len;
675 	uiop->uio_rw = UIO_READ;
676 	uiop->uio_segflg = UIO_SYSSPACE;
677 	nh = nfsrv_sequential_heuristic(uiop, vp);
678 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
679 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
680 	FREE((caddr_t)iv2, M_TEMP);
681 	if (error) {
682 		m_freem(m3);
683 		*mpp = NULL;
684 		goto out;
685 	}
686 	nh->nh_nextoff = uiop->uio_offset;
687 	tlen = len - uiop->uio_resid;
688 	cnt = cnt < tlen ? cnt : tlen;
689 	tlen = NFSM_RNDUP(cnt);
690 	if (tlen == 0) {
691 		m_freem(m3);
692 		m3 = NULL;
693 	} else if (len != tlen || tlen != cnt)
694 		nfsrv_adj(m3, len - tlen, tlen - cnt);
695 	*mpp = m3;
696 	*mpendp = m2;
697 
698 out:
699 	NFSEXITCODE(error);
700 	return (error);
701 }
702 
703 /*
704  * Write vnode op from an mbuf list.
705  */
706 int
707 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
708     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
709 {
710 	struct iovec *ivp;
711 	int i, len;
712 	struct iovec *iv;
713 	int ioflags, error;
714 	struct uio io, *uiop = &io;
715 	struct nfsheur *nh;
716 
717 	MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
718 	    M_WAITOK);
719 	uiop->uio_iov = iv = ivp;
720 	uiop->uio_iovcnt = cnt;
721 	i = mtod(mp, caddr_t) + mp->m_len - cp;
722 	len = retlen;
723 	while (len > 0) {
724 		if (mp == NULL)
725 			panic("nfsvno_write");
726 		if (i > 0) {
727 			i = min(i, len);
728 			ivp->iov_base = cp;
729 			ivp->iov_len = i;
730 			ivp++;
731 			len -= i;
732 		}
733 		mp = mp->m_next;
734 		if (mp) {
735 			i = mp->m_len;
736 			cp = mtod(mp, caddr_t);
737 		}
738 	}
739 
740 	if (stable == NFSWRITE_UNSTABLE)
741 		ioflags = IO_NODELOCKED;
742 	else
743 		ioflags = (IO_SYNC | IO_NODELOCKED);
744 	uiop->uio_resid = retlen;
745 	uiop->uio_rw = UIO_WRITE;
746 	uiop->uio_segflg = UIO_SYSSPACE;
747 	NFSUIOPROC(uiop, p);
748 	uiop->uio_offset = off;
749 	nh = nfsrv_sequential_heuristic(uiop, vp);
750 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
751 	error = VOP_WRITE(vp, uiop, ioflags, cred);
752 	if (error == 0)
753 		nh->nh_nextoff = uiop->uio_offset;
754 	FREE((caddr_t)iv, M_TEMP);
755 
756 	NFSEXITCODE(error);
757 	return (error);
758 }
759 
760 /*
761  * Common code for creating a regular file (plus special files for V2).
762  */
763 int
764 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
765     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
766     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
767 {
768 	u_quad_t tempsize;
769 	int error;
770 
771 	error = nd->nd_repstat;
772 	if (!error && ndp->ni_vp == NULL) {
773 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
774 			vrele(ndp->ni_startdir);
775 			error = VOP_CREATE(ndp->ni_dvp,
776 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
777 			vput(ndp->ni_dvp);
778 			nfsvno_relpathbuf(ndp);
779 			if (!error) {
780 				if (*exclusive_flagp) {
781 					*exclusive_flagp = 0;
782 					NFSVNO_ATTRINIT(nvap);
783 					nvap->na_atime.tv_sec = cverf[0];
784 					nvap->na_atime.tv_nsec = cverf[1];
785 					error = VOP_SETATTR(ndp->ni_vp,
786 					    &nvap->na_vattr, nd->nd_cred);
787 				}
788 			}
789 		/*
790 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
791 		 * (This implies, just get out on an error.)
792 		 */
793 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
794 			nvap->na_type == VFIFO) {
795 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
796 				nvap->na_type = VFIFO;
797                         if (nvap->na_type != VFIFO &&
798 			    (error = priv_check_cred(nd->nd_cred,
799 			     PRIV_VFS_MKNOD_DEV, 0))) {
800 				vrele(ndp->ni_startdir);
801 				nfsvno_relpathbuf(ndp);
802 				vput(ndp->ni_dvp);
803 				goto out;
804 			}
805 			nvap->na_rdev = rdev;
806 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
807 			    &ndp->ni_cnd, &nvap->na_vattr);
808 			vput(ndp->ni_dvp);
809 			nfsvno_relpathbuf(ndp);
810 			vrele(ndp->ni_startdir);
811 			if (error)
812 				goto out;
813 		} else {
814 			vrele(ndp->ni_startdir);
815 			nfsvno_relpathbuf(ndp);
816 			vput(ndp->ni_dvp);
817 			error = ENXIO;
818 			goto out;
819 		}
820 		*vpp = ndp->ni_vp;
821 	} else {
822 		/*
823 		 * Handle cases where error is already set and/or
824 		 * the file exists.
825 		 * 1 - clean up the lookup
826 		 * 2 - iff !error and na_size set, truncate it
827 		 */
828 		vrele(ndp->ni_startdir);
829 		nfsvno_relpathbuf(ndp);
830 		*vpp = ndp->ni_vp;
831 		if (ndp->ni_dvp == *vpp)
832 			vrele(ndp->ni_dvp);
833 		else
834 			vput(ndp->ni_dvp);
835 		if (!error && nvap->na_size != VNOVAL) {
836 			error = nfsvno_accchk(*vpp, VWRITE,
837 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
838 			    NFSACCCHK_VPISLOCKED, NULL);
839 			if (!error) {
840 				tempsize = nvap->na_size;
841 				NFSVNO_ATTRINIT(nvap);
842 				nvap->na_size = tempsize;
843 				error = VOP_SETATTR(*vpp,
844 				    &nvap->na_vattr, nd->nd_cred);
845 			}
846 		}
847 		if (error)
848 			vput(*vpp);
849 	}
850 
851 out:
852 	NFSEXITCODE(error);
853 	return (error);
854 }
855 
856 /*
857  * Do a mknod vnode op.
858  */
859 int
860 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
861     struct thread *p)
862 {
863 	int error = 0;
864 	enum vtype vtyp;
865 
866 	vtyp = nvap->na_type;
867 	/*
868 	 * Iff doesn't exist, create it.
869 	 */
870 	if (ndp->ni_vp) {
871 		vrele(ndp->ni_startdir);
872 		nfsvno_relpathbuf(ndp);
873 		vput(ndp->ni_dvp);
874 		vrele(ndp->ni_vp);
875 		error = EEXIST;
876 		goto out;
877 	}
878 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
879 		vrele(ndp->ni_startdir);
880 		nfsvno_relpathbuf(ndp);
881 		vput(ndp->ni_dvp);
882 		error = NFSERR_BADTYPE;
883 		goto out;
884 	}
885 	if (vtyp == VSOCK) {
886 		vrele(ndp->ni_startdir);
887 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
888 		    &ndp->ni_cnd, &nvap->na_vattr);
889 		vput(ndp->ni_dvp);
890 		nfsvno_relpathbuf(ndp);
891 	} else {
892 		if (nvap->na_type != VFIFO &&
893 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
894 			vrele(ndp->ni_startdir);
895 			nfsvno_relpathbuf(ndp);
896 			vput(ndp->ni_dvp);
897 			goto out;
898 		}
899 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
900 		    &ndp->ni_cnd, &nvap->na_vattr);
901 		vput(ndp->ni_dvp);
902 		nfsvno_relpathbuf(ndp);
903 		vrele(ndp->ni_startdir);
904 		/*
905 		 * Since VOP_MKNOD returns the ni_vp, I can't
906 		 * see any reason to do the lookup.
907 		 */
908 	}
909 
910 out:
911 	NFSEXITCODE(error);
912 	return (error);
913 }
914 
915 /*
916  * Mkdir vnode op.
917  */
918 int
919 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
920     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
921 {
922 	int error = 0;
923 
924 	if (ndp->ni_vp != NULL) {
925 		if (ndp->ni_dvp == ndp->ni_vp)
926 			vrele(ndp->ni_dvp);
927 		else
928 			vput(ndp->ni_dvp);
929 		vrele(ndp->ni_vp);
930 		nfsvno_relpathbuf(ndp);
931 		error = EEXIST;
932 		goto out;
933 	}
934 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
935 	    &nvap->na_vattr);
936 	vput(ndp->ni_dvp);
937 	nfsvno_relpathbuf(ndp);
938 
939 out:
940 	NFSEXITCODE(error);
941 	return (error);
942 }
943 
944 /*
945  * symlink vnode op.
946  */
947 int
948 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
949     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
950     struct nfsexstuff *exp)
951 {
952 	int error = 0;
953 
954 	if (ndp->ni_vp) {
955 		vrele(ndp->ni_startdir);
956 		nfsvno_relpathbuf(ndp);
957 		if (ndp->ni_dvp == ndp->ni_vp)
958 			vrele(ndp->ni_dvp);
959 		else
960 			vput(ndp->ni_dvp);
961 		vrele(ndp->ni_vp);
962 		error = EEXIST;
963 		goto out;
964 	}
965 
966 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
967 	    &nvap->na_vattr, pathcp);
968 	vput(ndp->ni_dvp);
969 	vrele(ndp->ni_startdir);
970 	nfsvno_relpathbuf(ndp);
971 	/*
972 	 * Although FreeBSD still had the lookup code in
973 	 * it for 7/current, there doesn't seem to be any
974 	 * point, since VOP_SYMLINK() returns the ni_vp.
975 	 * Just vput it for v2.
976 	 */
977 	if (!not_v2 && !error)
978 		vput(ndp->ni_vp);
979 
980 out:
981 	NFSEXITCODE(error);
982 	return (error);
983 }
984 
985 /*
986  * Parse symbolic link arguments.
987  * This function has an ugly side effect. It will MALLOC() an area for
988  * the symlink and set iov_base to point to it, only if it succeeds.
989  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
990  * be FREE'd later.
991  */
992 int
993 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
994     struct thread *p, char **pathcpp, int *lenp)
995 {
996 	u_int32_t *tl;
997 	char *pathcp = NULL;
998 	int error = 0, len;
999 	struct nfsv2_sattr *sp;
1000 
1001 	*pathcpp = NULL;
1002 	*lenp = 0;
1003 	if ((nd->nd_flag & ND_NFSV3) &&
1004 	    (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
1005 		goto nfsmout;
1006 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1007 	len = fxdr_unsigned(int, *tl);
1008 	if (len > NFS_MAXPATHLEN || len <= 0) {
1009 		error = EBADRPC;
1010 		goto nfsmout;
1011 	}
1012 	MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
1013 	error = nfsrv_mtostr(nd, pathcp, len);
1014 	if (error)
1015 		goto nfsmout;
1016 	if (nd->nd_flag & ND_NFSV2) {
1017 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1018 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1019 	}
1020 	*pathcpp = pathcp;
1021 	*lenp = len;
1022 	NFSEXITCODE2(0, nd);
1023 	return (0);
1024 nfsmout:
1025 	if (pathcp)
1026 		free(pathcp, M_TEMP);
1027 	NFSEXITCODE2(error, nd);
1028 	return (error);
1029 }
1030 
1031 /*
1032  * Remove a non-directory object.
1033  */
1034 int
1035 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1036     struct thread *p, struct nfsexstuff *exp)
1037 {
1038 	struct vnode *vp;
1039 	int error = 0;
1040 
1041 	vp = ndp->ni_vp;
1042 	if (vp->v_type == VDIR)
1043 		error = NFSERR_ISDIR;
1044 	else if (is_v4)
1045 		error = nfsrv_checkremove(vp, 1, p);
1046 	if (!error)
1047 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1048 	if (ndp->ni_dvp == vp)
1049 		vrele(ndp->ni_dvp);
1050 	else
1051 		vput(ndp->ni_dvp);
1052 	vput(vp);
1053 	NFSEXITCODE(error);
1054 	return (error);
1055 }
1056 
1057 /*
1058  * Remove a directory.
1059  */
1060 int
1061 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1062     struct thread *p, struct nfsexstuff *exp)
1063 {
1064 	struct vnode *vp;
1065 	int error = 0;
1066 
1067 	vp = ndp->ni_vp;
1068 	if (vp->v_type != VDIR) {
1069 		error = ENOTDIR;
1070 		goto out;
1071 	}
1072 	/*
1073 	 * No rmdir "." please.
1074 	 */
1075 	if (ndp->ni_dvp == vp) {
1076 		error = EINVAL;
1077 		goto out;
1078 	}
1079 	/*
1080 	 * The root of a mounted filesystem cannot be deleted.
1081 	 */
1082 	if (vp->v_vflag & VV_ROOT)
1083 		error = EBUSY;
1084 out:
1085 	if (!error)
1086 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1087 	if (ndp->ni_dvp == vp)
1088 		vrele(ndp->ni_dvp);
1089 	else
1090 		vput(ndp->ni_dvp);
1091 	vput(vp);
1092 	NFSEXITCODE(error);
1093 	return (error);
1094 }
1095 
1096 /*
1097  * Rename vnode op.
1098  */
1099 int
1100 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1101     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1102 {
1103 	struct vnode *fvp, *tvp, *tdvp;
1104 	int error = 0;
1105 
1106 	fvp = fromndp->ni_vp;
1107 	if (ndstat) {
1108 		vrele(fromndp->ni_dvp);
1109 		vrele(fvp);
1110 		error = ndstat;
1111 		goto out1;
1112 	}
1113 	tdvp = tondp->ni_dvp;
1114 	tvp = tondp->ni_vp;
1115 	if (tvp != NULL) {
1116 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1117 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1118 			goto out;
1119 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1120 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1121 			goto out;
1122 		}
1123 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1124 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1125 			goto out;
1126 		}
1127 
1128 		/*
1129 		 * A rename to '.' or '..' results in a prematurely
1130 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1131 		 * here.
1132 		 */
1133 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1134 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1135 		    (tondp->ni_cnd.cn_namelen == 2 &&
1136 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1137 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1138 			error = EINVAL;
1139 			goto out;
1140 		}
1141 	}
1142 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1143 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1144 		goto out;
1145 	}
1146 	if (fvp->v_mount != tdvp->v_mount) {
1147 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1148 		goto out;
1149 	}
1150 	if (fvp == tdvp) {
1151 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1152 		goto out;
1153 	}
1154 	if (fvp == tvp) {
1155 		/*
1156 		 * If source and destination are the same, there is nothing to
1157 		 * do. Set error to -1 to indicate this.
1158 		 */
1159 		error = -1;
1160 		goto out;
1161 	}
1162 	if (ndflag & ND_NFSV4) {
1163 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1164 			error = nfsrv_checkremove(fvp, 0, p);
1165 			NFSVOPUNLOCK(fvp, 0);
1166 		} else
1167 			error = EPERM;
1168 		if (tvp && !error)
1169 			error = nfsrv_checkremove(tvp, 1, p);
1170 	} else {
1171 		/*
1172 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1173 		 * that the NFSv4 client won't be confused by the rename.
1174 		 * Since nfsd_recalldelegation() can only be called on an
1175 		 * unlocked vnode at this point and fvp is the file that will
1176 		 * still exist after the rename, just do fvp.
1177 		 */
1178 		nfsd_recalldelegation(fvp, p);
1179 	}
1180 out:
1181 	if (!error) {
1182 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1183 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1184 		    &tondp->ni_cnd);
1185 	} else {
1186 		if (tdvp == tvp)
1187 			vrele(tdvp);
1188 		else
1189 			vput(tdvp);
1190 		if (tvp)
1191 			vput(tvp);
1192 		vrele(fromndp->ni_dvp);
1193 		vrele(fvp);
1194 		if (error == -1)
1195 			error = 0;
1196 	}
1197 	vrele(tondp->ni_startdir);
1198 	nfsvno_relpathbuf(tondp);
1199 out1:
1200 	vrele(fromndp->ni_startdir);
1201 	nfsvno_relpathbuf(fromndp);
1202 	NFSEXITCODE(error);
1203 	return (error);
1204 }
1205 
1206 /*
1207  * Link vnode op.
1208  */
1209 int
1210 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1211     struct thread *p, struct nfsexstuff *exp)
1212 {
1213 	struct vnode *xp;
1214 	int error = 0;
1215 
1216 	xp = ndp->ni_vp;
1217 	if (xp != NULL) {
1218 		error = EEXIST;
1219 	} else {
1220 		xp = ndp->ni_dvp;
1221 		if (vp->v_mount != xp->v_mount)
1222 			error = EXDEV;
1223 	}
1224 	if (!error) {
1225 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1226 		if ((vp->v_iflag & VI_DOOMED) == 0)
1227 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1228 		else
1229 			error = EPERM;
1230 		if (ndp->ni_dvp == vp)
1231 			vrele(ndp->ni_dvp);
1232 		else
1233 			vput(ndp->ni_dvp);
1234 		NFSVOPUNLOCK(vp, 0);
1235 	} else {
1236 		if (ndp->ni_dvp == ndp->ni_vp)
1237 			vrele(ndp->ni_dvp);
1238 		else
1239 			vput(ndp->ni_dvp);
1240 		if (ndp->ni_vp)
1241 			vrele(ndp->ni_vp);
1242 	}
1243 	nfsvno_relpathbuf(ndp);
1244 	NFSEXITCODE(error);
1245 	return (error);
1246 }
1247 
1248 /*
1249  * Do the fsync() appropriate for the commit.
1250  */
1251 int
1252 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1253     struct thread *td)
1254 {
1255 	int error = 0;
1256 
1257 	/*
1258 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1259 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1260 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1261 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1262 	 */
1263 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT) {
1264 		/*
1265 		 * Give up and do the whole thing
1266 		 */
1267 		if (vp->v_object &&
1268 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1269 			VM_OBJECT_LOCK(vp->v_object);
1270 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1271 			VM_OBJECT_UNLOCK(vp->v_object);
1272 		}
1273 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1274 	} else {
1275 		/*
1276 		 * Locate and synchronously write any buffers that fall
1277 		 * into the requested range.  Note:  we are assuming that
1278 		 * f_iosize is a power of 2.
1279 		 */
1280 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1281 		int iomask = iosize - 1;
1282 		struct bufobj *bo;
1283 		daddr_t lblkno;
1284 
1285 		/*
1286 		 * Align to iosize boundry, super-align to page boundry.
1287 		 */
1288 		if (off & iomask) {
1289 			cnt += off & iomask;
1290 			off &= ~(u_quad_t)iomask;
1291 		}
1292 		if (off & PAGE_MASK) {
1293 			cnt += off & PAGE_MASK;
1294 			off &= ~(u_quad_t)PAGE_MASK;
1295 		}
1296 		lblkno = off / iosize;
1297 
1298 		if (vp->v_object &&
1299 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1300 			VM_OBJECT_LOCK(vp->v_object);
1301 			vm_object_page_clean(vp->v_object, off, off + cnt,
1302 			    OBJPC_SYNC);
1303 			VM_OBJECT_UNLOCK(vp->v_object);
1304 		}
1305 
1306 		bo = &vp->v_bufobj;
1307 		BO_LOCK(bo);
1308 		while (cnt > 0) {
1309 			struct buf *bp;
1310 
1311 			/*
1312 			 * If we have a buffer and it is marked B_DELWRI we
1313 			 * have to lock and write it.  Otherwise the prior
1314 			 * write is assumed to have already been committed.
1315 			 *
1316 			 * gbincore() can return invalid buffers now so we
1317 			 * have to check that bit as well (though B_DELWRI
1318 			 * should not be set if B_INVAL is set there could be
1319 			 * a race here since we haven't locked the buffer).
1320 			 */
1321 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1322 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1323 				    LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
1324 					BO_LOCK(bo);
1325 					continue; /* retry */
1326 				}
1327 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1328 				    B_DELWRI) {
1329 					bremfree(bp);
1330 					bp->b_flags &= ~B_ASYNC;
1331 					bwrite(bp);
1332 					++nfs_commit_miss;
1333 				} else
1334 					BUF_UNLOCK(bp);
1335 				BO_LOCK(bo);
1336 			}
1337 			++nfs_commit_blks;
1338 			if (cnt < iosize)
1339 				break;
1340 			cnt -= iosize;
1341 			++lblkno;
1342 		}
1343 		BO_UNLOCK(bo);
1344 	}
1345 	NFSEXITCODE(error);
1346 	return (error);
1347 }
1348 
1349 /*
1350  * Statfs vnode op.
1351  */
1352 int
1353 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1354 {
1355 	int error;
1356 
1357 	error = VFS_STATFS(vp->v_mount, sf);
1358 	if (error == 0) {
1359 		/*
1360 		 * Since NFS handles these values as unsigned on the
1361 		 * wire, there is no way to represent negative values,
1362 		 * so set them to 0. Without this, they will appear
1363 		 * to be very large positive values for clients like
1364 		 * Solaris10.
1365 		 */
1366 		if (sf->f_bavail < 0)
1367 			sf->f_bavail = 0;
1368 		if (sf->f_ffree < 0)
1369 			sf->f_ffree = 0;
1370 	}
1371 	NFSEXITCODE(error);
1372 	return (error);
1373 }
1374 
1375 /*
1376  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1377  * must handle nfsrv_opencheck() calls after any other access checks.
1378  */
1379 void
1380 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1381     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1382     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1383     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1384     struct nfsexstuff *exp, struct vnode **vpp)
1385 {
1386 	struct vnode *vp = NULL;
1387 	u_quad_t tempsize;
1388 	struct nfsexstuff nes;
1389 
1390 	if (ndp->ni_vp == NULL)
1391 		nd->nd_repstat = nfsrv_opencheck(clientid,
1392 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1393 	if (!nd->nd_repstat) {
1394 		if (ndp->ni_vp == NULL) {
1395 			vrele(ndp->ni_startdir);
1396 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1397 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1398 			vput(ndp->ni_dvp);
1399 			nfsvno_relpathbuf(ndp);
1400 			if (!nd->nd_repstat) {
1401 				if (*exclusive_flagp) {
1402 					*exclusive_flagp = 0;
1403 					NFSVNO_ATTRINIT(nvap);
1404 					nvap->na_atime.tv_sec = cverf[0];
1405 					nvap->na_atime.tv_nsec = cverf[1];
1406 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1407 					    &nvap->na_vattr, cred);
1408 				} else {
1409 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1410 					    aclp, p, attrbitp, exp);
1411 				}
1412 			}
1413 			vp = ndp->ni_vp;
1414 		} else {
1415 			if (ndp->ni_startdir)
1416 				vrele(ndp->ni_startdir);
1417 			nfsvno_relpathbuf(ndp);
1418 			vp = ndp->ni_vp;
1419 			if (create == NFSV4OPEN_CREATE) {
1420 				if (ndp->ni_dvp == vp)
1421 					vrele(ndp->ni_dvp);
1422 				else
1423 					vput(ndp->ni_dvp);
1424 			}
1425 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1426 				if (ndp->ni_cnd.cn_flags & RDONLY)
1427 					NFSVNO_SETEXRDONLY(&nes);
1428 				else
1429 					NFSVNO_EXINIT(&nes);
1430 				nd->nd_repstat = nfsvno_accchk(vp,
1431 				    VWRITE, cred, &nes, p,
1432 				    NFSACCCHK_NOOVERRIDE,
1433 				    NFSACCCHK_VPISLOCKED, NULL);
1434 				nd->nd_repstat = nfsrv_opencheck(clientid,
1435 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1436 				if (!nd->nd_repstat) {
1437 					tempsize = nvap->na_size;
1438 					NFSVNO_ATTRINIT(nvap);
1439 					nvap->na_size = tempsize;
1440 					nd->nd_repstat = VOP_SETATTR(vp,
1441 					    &nvap->na_vattr, cred);
1442 				}
1443 			} else if (vp->v_type == VREG) {
1444 				nd->nd_repstat = nfsrv_opencheck(clientid,
1445 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1446 			}
1447 		}
1448 	} else {
1449 		if (ndp->ni_cnd.cn_flags & HASBUF)
1450 			nfsvno_relpathbuf(ndp);
1451 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1452 			vrele(ndp->ni_startdir);
1453 			if (ndp->ni_dvp == ndp->ni_vp)
1454 				vrele(ndp->ni_dvp);
1455 			else
1456 				vput(ndp->ni_dvp);
1457 			if (ndp->ni_vp)
1458 				vput(ndp->ni_vp);
1459 		}
1460 	}
1461 	*vpp = vp;
1462 
1463 	NFSEXITCODE2(0, nd);
1464 }
1465 
1466 /*
1467  * Updates the file rev and sets the mtime and ctime
1468  * to the current clock time, returning the va_filerev and va_Xtime
1469  * values.
1470  */
1471 void
1472 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1473     struct ucred *cred, struct thread *p)
1474 {
1475 	struct vattr va;
1476 
1477 	VATTR_NULL(&va);
1478 	getnanotime(&va.va_mtime);
1479 	(void) VOP_SETATTR(vp, &va, cred);
1480 	(void) nfsvno_getattr(vp, nvap, cred, p, 1);
1481 }
1482 
1483 /*
1484  * Glue routine to nfsv4_fillattr().
1485  */
1486 int
1487 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1488     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1489     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1490     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1491 {
1492 	int error;
1493 
1494 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1495 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1496 	    mounted_on_fileno);
1497 	NFSEXITCODE2(0, nd);
1498 	return (error);
1499 }
1500 
1501 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1502 /*
1503  * nfs readdir service
1504  * - mallocs what it thinks is enough to read
1505  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1506  * - calls VOP_READDIR()
1507  * - loops around building the reply
1508  *	if the output generated exceeds count break out of loop
1509  *	The NFSM_CLGET macro is used here so that the reply will be packed
1510  *	tightly in mbuf clusters.
1511  * - it trims out records with d_fileno == 0
1512  *	this doesn't matter for Unix clients, but they might confuse clients
1513  *	for other os'.
1514  * - it trims out records with d_type == DT_WHT
1515  *	these cannot be seen through NFS (unless we extend the protocol)
1516  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1517  * PS: The NFS protocol spec. does not clarify what the "count" byte
1518  *	argument is a count of.. just name strings and file id's or the
1519  *	entire reply rpc or ...
1520  *	I tried just file name and id sizes and it confused the Sun client,
1521  *	so I am using the full rpc size now. The "paranoia.." comment refers
1522  *	to including the status longwords that are not a part of the dir.
1523  *	"entry" structures, but are in the rpc.
1524  */
1525 int
1526 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1527     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1528 {
1529 	struct dirent *dp;
1530 	u_int32_t *tl;
1531 	int dirlen;
1532 	char *cpos, *cend, *rbuf;
1533 	struct nfsvattr at;
1534 	int nlen, error = 0, getret = 1;
1535 	int siz, cnt, fullsiz, eofflag, ncookies;
1536 	u_int64_t off, toff, verf;
1537 	u_long *cookies = NULL, *cookiep;
1538 	struct uio io;
1539 	struct iovec iv;
1540 	int not_zfs;
1541 
1542 	if (nd->nd_repstat) {
1543 		nfsrv_postopattr(nd, getret, &at);
1544 		goto out;
1545 	}
1546 	if (nd->nd_flag & ND_NFSV2) {
1547 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1548 		off = fxdr_unsigned(u_quad_t, *tl++);
1549 	} else {
1550 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1551 		off = fxdr_hyper(tl);
1552 		tl += 2;
1553 		verf = fxdr_hyper(tl);
1554 		tl += 2;
1555 	}
1556 	toff = off;
1557 	cnt = fxdr_unsigned(int, *tl);
1558 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1559 		cnt = NFS_SRVMAXDATA(nd);
1560 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1561 	fullsiz = siz;
1562 	if (nd->nd_flag & ND_NFSV3) {
1563 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1564 		    p, 1);
1565 #if 0
1566 		/*
1567 		 * va_filerev is not sufficient as a cookie verifier,
1568 		 * since it is not supposed to change when entries are
1569 		 * removed/added unless that offset cookies returned to
1570 		 * the client are no longer valid.
1571 		 */
1572 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
1573 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1574 #endif
1575 	}
1576 	if (nd->nd_repstat == 0 && cnt == 0) {
1577 		if (nd->nd_flag & ND_NFSV2)
1578 			/* NFSv2 does not have NFSERR_TOOSMALL */
1579 			nd->nd_repstat = EPERM;
1580 		else
1581 			nd->nd_repstat = NFSERR_TOOSMALL;
1582 	}
1583 	if (!nd->nd_repstat)
1584 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1585 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1586 		    NFSACCCHK_VPISLOCKED, NULL);
1587 	if (nd->nd_repstat) {
1588 		vput(vp);
1589 		if (nd->nd_flag & ND_NFSV3)
1590 			nfsrv_postopattr(nd, getret, &at);
1591 		goto out;
1592 	}
1593 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1594 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1595 again:
1596 	eofflag = 0;
1597 	if (cookies) {
1598 		free((caddr_t)cookies, M_TEMP);
1599 		cookies = NULL;
1600 	}
1601 
1602 	iv.iov_base = rbuf;
1603 	iv.iov_len = siz;
1604 	io.uio_iov = &iv;
1605 	io.uio_iovcnt = 1;
1606 	io.uio_offset = (off_t)off;
1607 	io.uio_resid = siz;
1608 	io.uio_segflg = UIO_SYSSPACE;
1609 	io.uio_rw = UIO_READ;
1610 	io.uio_td = NULL;
1611 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1612 	    &cookies);
1613 	off = (u_int64_t)io.uio_offset;
1614 	if (io.uio_resid)
1615 		siz -= io.uio_resid;
1616 
1617 	if (!cookies && !nd->nd_repstat)
1618 		nd->nd_repstat = NFSERR_PERM;
1619 	if (nd->nd_flag & ND_NFSV3) {
1620 		getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1621 		if (!nd->nd_repstat)
1622 			nd->nd_repstat = getret;
1623 	}
1624 
1625 	/*
1626 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
1627 	 */
1628 	if (nd->nd_repstat) {
1629 		vput(vp);
1630 		free((caddr_t)rbuf, M_TEMP);
1631 		if (cookies)
1632 			free((caddr_t)cookies, M_TEMP);
1633 		if (nd->nd_flag & ND_NFSV3)
1634 			nfsrv_postopattr(nd, getret, &at);
1635 		goto out;
1636 	}
1637 	/*
1638 	 * If nothing read, return eof
1639 	 * rpc reply
1640 	 */
1641 	if (siz == 0) {
1642 		vput(vp);
1643 		if (nd->nd_flag & ND_NFSV2) {
1644 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1645 		} else {
1646 			nfsrv_postopattr(nd, getret, &at);
1647 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1648 			txdr_hyper(at.na_filerev, tl);
1649 			tl += 2;
1650 		}
1651 		*tl++ = newnfs_false;
1652 		*tl = newnfs_true;
1653 		FREE((caddr_t)rbuf, M_TEMP);
1654 		FREE((caddr_t)cookies, M_TEMP);
1655 		goto out;
1656 	}
1657 
1658 	/*
1659 	 * Check for degenerate cases of nothing useful read.
1660 	 * If so go try again
1661 	 */
1662 	cpos = rbuf;
1663 	cend = rbuf + siz;
1664 	dp = (struct dirent *)cpos;
1665 	cookiep = cookies;
1666 
1667 	/*
1668 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1669 	 * directory offset up to a block boundary, so it is necessary to
1670 	 * skip over the records that precede the requested offset. This
1671 	 * requires the assumption that file offset cookies monotonically
1672 	 * increase.
1673 	 * Since the offset cookies don't monotonically increase for ZFS,
1674 	 * this is not done when ZFS is the file system.
1675 	 */
1676 	while (cpos < cend && ncookies > 0 &&
1677 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1678 	     (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
1679 		cpos += dp->d_reclen;
1680 		dp = (struct dirent *)cpos;
1681 		cookiep++;
1682 		ncookies--;
1683 	}
1684 	if (cpos >= cend || ncookies == 0) {
1685 		siz = fullsiz;
1686 		toff = off;
1687 		goto again;
1688 	}
1689 	vput(vp);
1690 
1691 	/*
1692 	 * dirlen is the size of the reply, including all XDR and must
1693 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1694 	 * if the XDR should be included in "count", but to be safe, we do.
1695 	 * (Include the two booleans at the end of the reply in dirlen now.)
1696 	 */
1697 	if (nd->nd_flag & ND_NFSV3) {
1698 		nfsrv_postopattr(nd, getret, &at);
1699 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1700 		txdr_hyper(at.na_filerev, tl);
1701 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1702 	} else {
1703 		dirlen = 2 * NFSX_UNSIGNED;
1704 	}
1705 
1706 	/* Loop through the records and build reply */
1707 	while (cpos < cend && ncookies > 0) {
1708 		nlen = dp->d_namlen;
1709 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1710 			nlen <= NFS_MAXNAMLEN) {
1711 			if (nd->nd_flag & ND_NFSV3)
1712 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1713 			else
1714 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1715 			if (dirlen > cnt) {
1716 				eofflag = 0;
1717 				break;
1718 			}
1719 
1720 			/*
1721 			 * Build the directory record xdr from
1722 			 * the dirent entry.
1723 			 */
1724 			if (nd->nd_flag & ND_NFSV3) {
1725 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1726 				*tl++ = newnfs_true;
1727 				*tl++ = 0;
1728 			} else {
1729 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1730 				*tl++ = newnfs_true;
1731 			}
1732 			*tl = txdr_unsigned(dp->d_fileno);
1733 			(void) nfsm_strtom(nd, dp->d_name, nlen);
1734 			if (nd->nd_flag & ND_NFSV3) {
1735 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1736 				*tl++ = 0;
1737 			} else
1738 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1739 			*tl = txdr_unsigned(*cookiep);
1740 		}
1741 		cpos += dp->d_reclen;
1742 		dp = (struct dirent *)cpos;
1743 		cookiep++;
1744 		ncookies--;
1745 	}
1746 	if (cpos < cend)
1747 		eofflag = 0;
1748 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1749 	*tl++ = newnfs_false;
1750 	if (eofflag)
1751 		*tl = newnfs_true;
1752 	else
1753 		*tl = newnfs_false;
1754 	FREE((caddr_t)rbuf, M_TEMP);
1755 	FREE((caddr_t)cookies, M_TEMP);
1756 
1757 out:
1758 	NFSEXITCODE2(0, nd);
1759 	return (0);
1760 nfsmout:
1761 	vput(vp);
1762 	NFSEXITCODE2(error, nd);
1763 	return (error);
1764 }
1765 
1766 /*
1767  * Readdirplus for V3 and Readdir for V4.
1768  */
1769 int
1770 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1771     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1772 {
1773 	struct dirent *dp;
1774 	u_int32_t *tl;
1775 	int dirlen;
1776 	char *cpos, *cend, *rbuf;
1777 	struct vnode *nvp;
1778 	fhandle_t nfh;
1779 	struct nfsvattr nva, at, *nvap = &nva;
1780 	struct mbuf *mb0, *mb1;
1781 	struct nfsreferral *refp;
1782 	int nlen, r, error = 0, getret = 1, usevget = 1;
1783 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1784 	caddr_t bpos0, bpos1;
1785 	u_int64_t off, toff, verf;
1786 	u_long *cookies = NULL, *cookiep;
1787 	nfsattrbit_t attrbits, rderrbits, savbits;
1788 	struct uio io;
1789 	struct iovec iv;
1790 	struct componentname cn;
1791 	int at_root, needs_unbusy, not_zfs, supports_nfsv4acls;
1792 	struct mount *mp, *new_mp;
1793 	uint64_t mounted_on_fileno;
1794 
1795 	if (nd->nd_repstat) {
1796 		nfsrv_postopattr(nd, getret, &at);
1797 		goto out;
1798 	}
1799 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1800 	off = fxdr_hyper(tl);
1801 	toff = off;
1802 	tl += 2;
1803 	verf = fxdr_hyper(tl);
1804 	tl += 2;
1805 	siz = fxdr_unsigned(int, *tl++);
1806 	cnt = fxdr_unsigned(int, *tl);
1807 
1808 	/*
1809 	 * Use the server's maximum data transfer size as the upper bound
1810 	 * on reply datalen.
1811 	 */
1812 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1813 		cnt = NFS_SRVMAXDATA(nd);
1814 
1815 	/*
1816 	 * siz is a "hint" of how much directory information (name, fileid,
1817 	 * cookie) should be in the reply. At least one client "hints" 0,
1818 	 * so I set it to cnt for that case. I also round it up to the
1819 	 * next multiple of DIRBLKSIZ.
1820 	 */
1821 	if (siz <= 0)
1822 		siz = cnt;
1823 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1824 
1825 	if (nd->nd_flag & ND_NFSV4) {
1826 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1827 		if (error)
1828 			goto nfsmout;
1829 		NFSSET_ATTRBIT(&savbits, &attrbits);
1830 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1831 		NFSZERO_ATTRBIT(&rderrbits);
1832 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1833 	} else {
1834 		NFSZERO_ATTRBIT(&attrbits);
1835 	}
1836 	fullsiz = siz;
1837 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1838 	if (!nd->nd_repstat) {
1839 	    if (off && verf != at.na_filerev) {
1840 		/*
1841 		 * va_filerev is not sufficient as a cookie verifier,
1842 		 * since it is not supposed to change when entries are
1843 		 * removed/added unless that offset cookies returned to
1844 		 * the client are no longer valid.
1845 		 */
1846 #if 0
1847 		if (nd->nd_flag & ND_NFSV4) {
1848 			nd->nd_repstat = NFSERR_NOTSAME;
1849 		} else {
1850 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1851 		}
1852 #endif
1853 	    } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1854 		nd->nd_repstat = NFSERR_BAD_COOKIE;
1855 	    }
1856 	}
1857 	if (!nd->nd_repstat && vp->v_type != VDIR)
1858 		nd->nd_repstat = NFSERR_NOTDIR;
1859 	if (!nd->nd_repstat && cnt == 0)
1860 		nd->nd_repstat = NFSERR_TOOSMALL;
1861 	if (!nd->nd_repstat)
1862 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1863 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1864 		    NFSACCCHK_VPISLOCKED, NULL);
1865 	if (nd->nd_repstat) {
1866 		vput(vp);
1867 		if (nd->nd_flag & ND_NFSV3)
1868 			nfsrv_postopattr(nd, getret, &at);
1869 		goto out;
1870 	}
1871 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1872 
1873 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1874 again:
1875 	eofflag = 0;
1876 	if (cookies) {
1877 		free((caddr_t)cookies, M_TEMP);
1878 		cookies = NULL;
1879 	}
1880 
1881 	iv.iov_base = rbuf;
1882 	iv.iov_len = siz;
1883 	io.uio_iov = &iv;
1884 	io.uio_iovcnt = 1;
1885 	io.uio_offset = (off_t)off;
1886 	io.uio_resid = siz;
1887 	io.uio_segflg = UIO_SYSSPACE;
1888 	io.uio_rw = UIO_READ;
1889 	io.uio_td = NULL;
1890 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1891 	    &cookies);
1892 	off = (u_int64_t)io.uio_offset;
1893 	if (io.uio_resid)
1894 		siz -= io.uio_resid;
1895 
1896 	getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1897 
1898 	if (!cookies && !nd->nd_repstat)
1899 		nd->nd_repstat = NFSERR_PERM;
1900 	if (!nd->nd_repstat)
1901 		nd->nd_repstat = getret;
1902 	if (nd->nd_repstat) {
1903 		vput(vp);
1904 		if (cookies)
1905 			free((caddr_t)cookies, M_TEMP);
1906 		free((caddr_t)rbuf, M_TEMP);
1907 		if (nd->nd_flag & ND_NFSV3)
1908 			nfsrv_postopattr(nd, getret, &at);
1909 		goto out;
1910 	}
1911 	/*
1912 	 * If nothing read, return eof
1913 	 * rpc reply
1914 	 */
1915 	if (siz == 0) {
1916 		vput(vp);
1917 		if (nd->nd_flag & ND_NFSV3)
1918 			nfsrv_postopattr(nd, getret, &at);
1919 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1920 		txdr_hyper(at.na_filerev, tl);
1921 		tl += 2;
1922 		*tl++ = newnfs_false;
1923 		*tl = newnfs_true;
1924 		free((caddr_t)cookies, M_TEMP);
1925 		free((caddr_t)rbuf, M_TEMP);
1926 		goto out;
1927 	}
1928 
1929 	/*
1930 	 * Check for degenerate cases of nothing useful read.
1931 	 * If so go try again
1932 	 */
1933 	cpos = rbuf;
1934 	cend = rbuf + siz;
1935 	dp = (struct dirent *)cpos;
1936 	cookiep = cookies;
1937 
1938 	/*
1939 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1940 	 * directory offset up to a block boundary, so it is necessary to
1941 	 * skip over the records that precede the requested offset. This
1942 	 * requires the assumption that file offset cookies monotonically
1943 	 * increase.
1944 	 * Since the offset cookies don't monotonically increase for ZFS,
1945 	 * this is not done when ZFS is the file system.
1946 	 */
1947 	while (cpos < cend && ncookies > 0 &&
1948 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1949 	   (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff) ||
1950 	   ((nd->nd_flag & ND_NFSV4) &&
1951 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1952 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1953 		cpos += dp->d_reclen;
1954 		dp = (struct dirent *)cpos;
1955 		cookiep++;
1956 		ncookies--;
1957 	}
1958 	if (cpos >= cend || ncookies == 0) {
1959 		siz = fullsiz;
1960 		toff = off;
1961 		goto again;
1962 	}
1963 
1964 	/*
1965 	 * Busy the file system so that the mount point won't go away
1966 	 * and, as such, VFS_VGET() can be used safely.
1967 	 */
1968 	mp = vp->v_mount;
1969 	vfs_ref(mp);
1970 	NFSVOPUNLOCK(vp, 0);
1971 	nd->nd_repstat = vfs_busy(mp, 0);
1972 	vfs_rel(mp);
1973 	if (nd->nd_repstat != 0) {
1974 		vrele(vp);
1975 		free(cookies, M_TEMP);
1976 		free(rbuf, M_TEMP);
1977 		if (nd->nd_flag & ND_NFSV3)
1978 			nfsrv_postopattr(nd, getret, &at);
1979 		goto out;
1980 	}
1981 
1982 	/*
1983 	 * Save this position, in case there is an error before one entry
1984 	 * is created.
1985 	 */
1986 	mb0 = nd->nd_mb;
1987 	bpos0 = nd->nd_bpos;
1988 
1989 	/*
1990 	 * Fill in the first part of the reply.
1991 	 * dirlen is the reply length in bytes and cannot exceed cnt.
1992 	 * (Include the two booleans at the end of the reply in dirlen now,
1993 	 *  so we recognize when we have exceeded cnt.)
1994 	 */
1995 	if (nd->nd_flag & ND_NFSV3) {
1996 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1997 		nfsrv_postopattr(nd, getret, &at);
1998 	} else {
1999 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2000 	}
2001 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2002 	txdr_hyper(at.na_filerev, tl);
2003 
2004 	/*
2005 	 * Save this position, in case there is an empty reply needed.
2006 	 */
2007 	mb1 = nd->nd_mb;
2008 	bpos1 = nd->nd_bpos;
2009 
2010 	/* Loop through the records and build reply */
2011 	entrycnt = 0;
2012 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2013 		nlen = dp->d_namlen;
2014 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2015 		    nlen <= NFS_MAXNAMLEN &&
2016 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2017 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2018 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2019 			/*
2020 			 * Save the current position in the reply, in case
2021 			 * this entry exceeds cnt.
2022 			 */
2023 			mb1 = nd->nd_mb;
2024 			bpos1 = nd->nd_bpos;
2025 
2026 			/*
2027 			 * For readdir_and_lookup get the vnode using
2028 			 * the file number.
2029 			 */
2030 			nvp = NULL;
2031 			refp = NULL;
2032 			r = 0;
2033 			at_root = 0;
2034 			needs_unbusy = 0;
2035 			new_mp = mp;
2036 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2037 			if ((nd->nd_flag & ND_NFSV3) ||
2038 			    NFSNONZERO_ATTRBIT(&savbits)) {
2039 				if (nd->nd_flag & ND_NFSV4)
2040 					refp = nfsv4root_getreferral(NULL,
2041 					    vp, dp->d_fileno);
2042 				if (refp == NULL) {
2043 					if (usevget)
2044 						r = VFS_VGET(mp, dp->d_fileno,
2045 						    LK_SHARED, &nvp);
2046 					else
2047 						r = EOPNOTSUPP;
2048 					if (r == EOPNOTSUPP) {
2049 						if (usevget) {
2050 							usevget = 0;
2051 							cn.cn_nameiop = LOOKUP;
2052 							cn.cn_lkflags =
2053 							    LK_SHARED |
2054 							    LK_RETRY;
2055 							cn.cn_cred =
2056 							    nd->nd_cred;
2057 							cn.cn_thread = p;
2058 						}
2059 						cn.cn_nameptr = dp->d_name;
2060 						cn.cn_namelen = nlen;
2061 						cn.cn_flags = ISLASTCN |
2062 						    NOFOLLOW | LOCKLEAF |
2063 						    MPSAFE;
2064 						if (nlen == 2 &&
2065 						    dp->d_name[0] == '.' &&
2066 						    dp->d_name[1] == '.')
2067 							cn.cn_flags |=
2068 							    ISDOTDOT;
2069 						if (NFSVOPLOCK(vp, LK_SHARED)
2070 						    != 0) {
2071 							nd->nd_repstat = EPERM;
2072 							break;
2073 						}
2074 						if ((vp->v_vflag & VV_ROOT) != 0
2075 						    && (cn.cn_flags & ISDOTDOT)
2076 						    != 0) {
2077 							vref(vp);
2078 							nvp = vp;
2079 							r = 0;
2080 						} else {
2081 							r = VOP_LOOKUP(vp, &nvp,
2082 							    &cn);
2083 							if (vp != nvp)
2084 								NFSVOPUNLOCK(vp,
2085 								    0);
2086 						}
2087 					}
2088 
2089 					/*
2090 					 * For NFSv4, check to see if nvp is
2091 					 * a mount point and get the mount
2092 					 * point vnode, as required.
2093 					 */
2094 					if (r == 0 &&
2095 					    nfsrv_enable_crossmntpt != 0 &&
2096 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2097 					    nvp->v_type == VDIR &&
2098 					    nvp->v_mountedhere != NULL) {
2099 						new_mp = nvp->v_mountedhere;
2100 						r = vfs_busy(new_mp, 0);
2101 						vput(nvp);
2102 						nvp = NULL;
2103 						if (r == 0) {
2104 							r = VFS_ROOT(new_mp,
2105 							    LK_SHARED, &nvp);
2106 							needs_unbusy = 1;
2107 							if (r == 0)
2108 								at_root = 1;
2109 						}
2110 					}
2111 				}
2112 				if (!r) {
2113 				    if (refp == NULL &&
2114 					((nd->nd_flag & ND_NFSV3) ||
2115 					 NFSNONZERO_ATTRBIT(&attrbits))) {
2116 					r = nfsvno_getfh(nvp, &nfh, p);
2117 					if (!r)
2118 					    r = nfsvno_getattr(nvp, nvap,
2119 						nd->nd_cred, p, 1);
2120 				    }
2121 				} else {
2122 				    nvp = NULL;
2123 				}
2124 				if (r) {
2125 					if (!NFSISSET_ATTRBIT(&attrbits,
2126 					    NFSATTRBIT_RDATTRERROR)) {
2127 						if (nvp != NULL)
2128 							vput(nvp);
2129 						if (needs_unbusy != 0)
2130 							vfs_unbusy(new_mp);
2131 						nd->nd_repstat = r;
2132 						break;
2133 					}
2134 				}
2135 			}
2136 
2137 			/*
2138 			 * Build the directory record xdr
2139 			 */
2140 			if (nd->nd_flag & ND_NFSV3) {
2141 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2142 				*tl++ = newnfs_true;
2143 				*tl++ = 0;
2144 				*tl = txdr_unsigned(dp->d_fileno);
2145 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2146 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2147 				*tl++ = 0;
2148 				*tl = txdr_unsigned(*cookiep);
2149 				nfsrv_postopattr(nd, 0, nvap);
2150 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2151 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2152 				if (nvp != NULL)
2153 					vput(nvp);
2154 			} else {
2155 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2156 				*tl++ = newnfs_true;
2157 				*tl++ = 0;
2158 				*tl = txdr_unsigned(*cookiep);
2159 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2160 				if (nvp != NULL) {
2161 					supports_nfsv4acls =
2162 					    nfs_supportsnfsv4acls(nvp);
2163 					NFSVOPUNLOCK(nvp, 0);
2164 				} else
2165 					supports_nfsv4acls = 0;
2166 				if (refp != NULL) {
2167 					dirlen += nfsrv_putreferralattr(nd,
2168 					    &savbits, refp, 0,
2169 					    &nd->nd_repstat);
2170 					if (nd->nd_repstat) {
2171 						if (nvp != NULL)
2172 							vrele(nvp);
2173 						if (needs_unbusy != 0)
2174 							vfs_unbusy(new_mp);
2175 						break;
2176 					}
2177 				} else if (r) {
2178 					dirlen += nfsvno_fillattr(nd, new_mp,
2179 					    nvp, nvap, &nfh, r, &rderrbits,
2180 					    nd->nd_cred, p, isdgram, 0,
2181 					    supports_nfsv4acls, at_root,
2182 					    mounted_on_fileno);
2183 				} else {
2184 					dirlen += nfsvno_fillattr(nd, new_mp,
2185 					    nvp, nvap, &nfh, r, &attrbits,
2186 					    nd->nd_cred, p, isdgram, 0,
2187 					    supports_nfsv4acls, at_root,
2188 					    mounted_on_fileno);
2189 				}
2190 				if (nvp != NULL)
2191 					vrele(nvp);
2192 				dirlen += (3 * NFSX_UNSIGNED);
2193 			}
2194 			if (needs_unbusy != 0)
2195 				vfs_unbusy(new_mp);
2196 			if (dirlen <= cnt)
2197 				entrycnt++;
2198 		}
2199 		cpos += dp->d_reclen;
2200 		dp = (struct dirent *)cpos;
2201 		cookiep++;
2202 		ncookies--;
2203 	}
2204 	vrele(vp);
2205 	vfs_unbusy(mp);
2206 
2207 	/*
2208 	 * If dirlen > cnt, we must strip off the last entry. If that
2209 	 * results in an empty reply, report NFSERR_TOOSMALL.
2210 	 */
2211 	if (dirlen > cnt || nd->nd_repstat) {
2212 		if (!nd->nd_repstat && entrycnt == 0)
2213 			nd->nd_repstat = NFSERR_TOOSMALL;
2214 		if (nd->nd_repstat)
2215 			newnfs_trimtrailing(nd, mb0, bpos0);
2216 		else
2217 			newnfs_trimtrailing(nd, mb1, bpos1);
2218 		eofflag = 0;
2219 	} else if (cpos < cend)
2220 		eofflag = 0;
2221 	if (!nd->nd_repstat) {
2222 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2223 		*tl++ = newnfs_false;
2224 		if (eofflag)
2225 			*tl = newnfs_true;
2226 		else
2227 			*tl = newnfs_false;
2228 	}
2229 	FREE((caddr_t)cookies, M_TEMP);
2230 	FREE((caddr_t)rbuf, M_TEMP);
2231 
2232 out:
2233 	NFSEXITCODE2(0, nd);
2234 	return (0);
2235 nfsmout:
2236 	vput(vp);
2237 	NFSEXITCODE2(error, nd);
2238 	return (error);
2239 }
2240 
2241 /*
2242  * Get the settable attributes out of the mbuf list.
2243  * (Return 0 or EBADRPC)
2244  */
2245 int
2246 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2247     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2248 {
2249 	u_int32_t *tl;
2250 	struct nfsv2_sattr *sp;
2251 	struct timeval curtime;
2252 	int error = 0, toclient = 0;
2253 
2254 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2255 	case ND_NFSV2:
2256 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2257 		/*
2258 		 * Some old clients didn't fill in the high order 16bits.
2259 		 * --> check the low order 2 bytes for 0xffff
2260 		 */
2261 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2262 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2263 		if (sp->sa_uid != newnfs_xdrneg1)
2264 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2265 		if (sp->sa_gid != newnfs_xdrneg1)
2266 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2267 		if (sp->sa_size != newnfs_xdrneg1)
2268 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2269 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2270 #ifdef notyet
2271 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2272 #else
2273 			nvap->na_atime.tv_sec =
2274 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2275 			nvap->na_atime.tv_nsec = 0;
2276 #endif
2277 		}
2278 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2279 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2280 		break;
2281 	case ND_NFSV3:
2282 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2283 		if (*tl == newnfs_true) {
2284 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2285 			nvap->na_mode = nfstov_mode(*tl);
2286 		}
2287 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2288 		if (*tl == newnfs_true) {
2289 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2290 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2291 		}
2292 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2293 		if (*tl == newnfs_true) {
2294 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2295 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2296 		}
2297 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2298 		if (*tl == newnfs_true) {
2299 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2300 			nvap->na_size = fxdr_hyper(tl);
2301 		}
2302 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2303 		switch (fxdr_unsigned(int, *tl)) {
2304 		case NFSV3SATTRTIME_TOCLIENT:
2305 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2306 			fxdr_nfsv3time(tl, &nvap->na_atime);
2307 			toclient = 1;
2308 			break;
2309 		case NFSV3SATTRTIME_TOSERVER:
2310 			NFSGETTIME(&curtime);
2311 			nvap->na_atime.tv_sec = curtime.tv_sec;
2312 			nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2313 			nvap->na_vaflags |= VA_UTIMES_NULL;
2314 			break;
2315 		};
2316 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2317 		switch (fxdr_unsigned(int, *tl)) {
2318 		case NFSV3SATTRTIME_TOCLIENT:
2319 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2320 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2321 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2322 			break;
2323 		case NFSV3SATTRTIME_TOSERVER:
2324 			NFSGETTIME(&curtime);
2325 			nvap->na_mtime.tv_sec = curtime.tv_sec;
2326 			nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2327 			if (!toclient)
2328 				nvap->na_vaflags |= VA_UTIMES_NULL;
2329 			break;
2330 		};
2331 		break;
2332 	case ND_NFSV4:
2333 		error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2334 	};
2335 nfsmout:
2336 	NFSEXITCODE2(error, nd);
2337 	return (error);
2338 }
2339 
2340 /*
2341  * Handle the setable attributes for V4.
2342  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2343  */
2344 int
2345 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2346     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2347 {
2348 	u_int32_t *tl;
2349 	int attrsum = 0;
2350 	int i, j;
2351 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2352 	int toclient = 0;
2353 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2354 	uid_t uid;
2355 	gid_t gid;
2356 	struct timeval curtime;
2357 
2358 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2359 	if (error)
2360 		goto nfsmout;
2361 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2362 	attrsize = fxdr_unsigned(int, *tl);
2363 
2364 	/*
2365 	 * Loop around getting the setable attributes. If an unsupported
2366 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2367 	 */
2368 	if (retnotsup) {
2369 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2370 		bitpos = NFSATTRBIT_MAX;
2371 	} else {
2372 		bitpos = 0;
2373 	}
2374 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2375 	    if (attrsum > attrsize) {
2376 		error = NFSERR_BADXDR;
2377 		goto nfsmout;
2378 	    }
2379 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2380 		switch (bitpos) {
2381 		case NFSATTRBIT_SIZE:
2382 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2383 			nvap->na_size = fxdr_hyper(tl);
2384 			attrsum += NFSX_HYPER;
2385 			break;
2386 		case NFSATTRBIT_ACL:
2387 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2388 			    p);
2389 			if (error)
2390 				goto nfsmout;
2391 			if (aceerr && !nd->nd_repstat)
2392 				nd->nd_repstat = aceerr;
2393 			attrsum += aclsize;
2394 			break;
2395 		case NFSATTRBIT_ARCHIVE:
2396 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2397 			if (!nd->nd_repstat)
2398 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2399 			attrsum += NFSX_UNSIGNED;
2400 			break;
2401 		case NFSATTRBIT_HIDDEN:
2402 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2403 			if (!nd->nd_repstat)
2404 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2405 			attrsum += NFSX_UNSIGNED;
2406 			break;
2407 		case NFSATTRBIT_MIMETYPE:
2408 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2409 			i = fxdr_unsigned(int, *tl);
2410 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2411 			if (error)
2412 				goto nfsmout;
2413 			if (!nd->nd_repstat)
2414 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2415 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2416 			break;
2417 		case NFSATTRBIT_MODE:
2418 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2419 			nvap->na_mode = nfstov_mode(*tl);
2420 			attrsum += NFSX_UNSIGNED;
2421 			break;
2422 		case NFSATTRBIT_OWNER:
2423 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2424 			j = fxdr_unsigned(int, *tl);
2425 			if (j < 0) {
2426 				error = NFSERR_BADXDR;
2427 				goto nfsmout;
2428 			}
2429 			if (j > NFSV4_SMALLSTR)
2430 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2431 			else
2432 				cp = namestr;
2433 			error = nfsrv_mtostr(nd, cp, j);
2434 			if (error) {
2435 				if (j > NFSV4_SMALLSTR)
2436 					free(cp, M_NFSSTRING);
2437 				goto nfsmout;
2438 			}
2439 			if (!nd->nd_repstat) {
2440 				nd->nd_repstat = nfsv4_strtouid(cp,j,&uid,p);
2441 				if (!nd->nd_repstat)
2442 					nvap->na_uid = uid;
2443 			}
2444 			if (j > NFSV4_SMALLSTR)
2445 				free(cp, M_NFSSTRING);
2446 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2447 			break;
2448 		case NFSATTRBIT_OWNERGROUP:
2449 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2450 			j = fxdr_unsigned(int, *tl);
2451 			if (j < 0) {
2452 				error = NFSERR_BADXDR;
2453 				goto nfsmout;
2454 			}
2455 			if (j > NFSV4_SMALLSTR)
2456 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2457 			else
2458 				cp = namestr;
2459 			error = nfsrv_mtostr(nd, cp, j);
2460 			if (error) {
2461 				if (j > NFSV4_SMALLSTR)
2462 					free(cp, M_NFSSTRING);
2463 				goto nfsmout;
2464 			}
2465 			if (!nd->nd_repstat) {
2466 				nd->nd_repstat = nfsv4_strtogid(cp,j,&gid,p);
2467 				if (!nd->nd_repstat)
2468 					nvap->na_gid = gid;
2469 			}
2470 			if (j > NFSV4_SMALLSTR)
2471 				free(cp, M_NFSSTRING);
2472 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2473 			break;
2474 		case NFSATTRBIT_SYSTEM:
2475 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2476 			if (!nd->nd_repstat)
2477 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2478 			attrsum += NFSX_UNSIGNED;
2479 			break;
2480 		case NFSATTRBIT_TIMEACCESSSET:
2481 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2482 			attrsum += NFSX_UNSIGNED;
2483 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2484 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2485 			    fxdr_nfsv4time(tl, &nvap->na_atime);
2486 			    toclient = 1;
2487 			    attrsum += NFSX_V4TIME;
2488 			} else {
2489 			    NFSGETTIME(&curtime);
2490 			    nvap->na_atime.tv_sec = curtime.tv_sec;
2491 			    nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2492 			    nvap->na_vaflags |= VA_UTIMES_NULL;
2493 			}
2494 			break;
2495 		case NFSATTRBIT_TIMEBACKUP:
2496 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2497 			if (!nd->nd_repstat)
2498 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2499 			attrsum += NFSX_V4TIME;
2500 			break;
2501 		case NFSATTRBIT_TIMECREATE:
2502 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2503 			if (!nd->nd_repstat)
2504 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2505 			attrsum += NFSX_V4TIME;
2506 			break;
2507 		case NFSATTRBIT_TIMEMODIFYSET:
2508 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2509 			attrsum += NFSX_UNSIGNED;
2510 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2511 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2512 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
2513 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
2514 			    attrsum += NFSX_V4TIME;
2515 			} else {
2516 			    NFSGETTIME(&curtime);
2517 			    nvap->na_mtime.tv_sec = curtime.tv_sec;
2518 			    nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2519 			    if (!toclient)
2520 				nvap->na_vaflags |= VA_UTIMES_NULL;
2521 			}
2522 			break;
2523 		default:
2524 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2525 			/*
2526 			 * set bitpos so we drop out of the loop.
2527 			 */
2528 			bitpos = NFSATTRBIT_MAX;
2529 			break;
2530 		};
2531 	}
2532 
2533 	/*
2534 	 * some clients pad the attrlist, so we need to skip over the
2535 	 * padding.
2536 	 */
2537 	if (attrsum > attrsize) {
2538 		error = NFSERR_BADXDR;
2539 	} else {
2540 		attrsize = NFSM_RNDUP(attrsize);
2541 		if (attrsum < attrsize)
2542 			error = nfsm_advance(nd, attrsize - attrsum, -1);
2543 	}
2544 nfsmout:
2545 	NFSEXITCODE2(error, nd);
2546 	return (error);
2547 }
2548 
2549 /*
2550  * Check/setup export credentials.
2551  */
2552 int
2553 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2554     struct ucred *credanon)
2555 {
2556 	int error = 0;
2557 
2558 	/*
2559 	 * Check/setup credentials.
2560 	 */
2561 	if (nd->nd_flag & ND_GSS)
2562 		exp->nes_exflag &= ~MNT_EXPORTANON;
2563 
2564 	/*
2565 	 * Check to see if the operation is allowed for this security flavor.
2566 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2567 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2568 	 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2569 	 */
2570 	if (nfsvno_testexp(nd, exp) &&
2571 	    nd->nd_procnum != NFSV4OP_SECINFO &&
2572 	    nd->nd_procnum != NFSPROC_FSINFO) {
2573 		if (nd->nd_flag & ND_NFSV4)
2574 			error = NFSERR_WRONGSEC;
2575 		else
2576 			error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2577 		goto out;
2578 	}
2579 
2580 	/*
2581 	 * Check to see if the file system is exported V4 only.
2582 	 */
2583 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2584 		error = NFSERR_PROGNOTV4;
2585 		goto out;
2586 	}
2587 
2588 	/*
2589 	 * Now, map the user credentials.
2590 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
2591 	 *  Fsinfo RPC. If set for anything else, this code might need
2592 	 *  to change.)
2593 	 */
2594 	if (NFSVNO_EXPORTED(exp) &&
2595 	    ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2596 	     NFSVNO_EXPORTANON(exp) ||
2597 	     (nd->nd_flag & ND_AUTHNONE))) {
2598 		nd->nd_cred->cr_uid = credanon->cr_uid;
2599 		nd->nd_cred->cr_gid = credanon->cr_gid;
2600 		crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2601 		    credanon->cr_groups);
2602 	}
2603 
2604 out:
2605 	NFSEXITCODE2(error, nd);
2606 	return (error);
2607 }
2608 
2609 /*
2610  * Check exports.
2611  */
2612 int
2613 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2614     struct ucred **credp)
2615 {
2616 	int i, error, *secflavors;
2617 
2618 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2619 	    &exp->nes_numsecflavor, &secflavors);
2620 	if (error) {
2621 		if (nfs_rootfhset) {
2622 			exp->nes_exflag = 0;
2623 			exp->nes_numsecflavor = 0;
2624 			error = 0;
2625 		}
2626 	} else {
2627 		/* Copy the security flavors. */
2628 		for (i = 0; i < exp->nes_numsecflavor; i++)
2629 			exp->nes_secflavors[i] = secflavors[i];
2630 	}
2631 	NFSEXITCODE(error);
2632 	return (error);
2633 }
2634 
2635 /*
2636  * Get a vnode for a file handle and export stuff.
2637  */
2638 int
2639 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2640     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2641     struct ucred **credp)
2642 {
2643 	int i, error, *secflavors;
2644 
2645 	*credp = NULL;
2646 	exp->nes_numsecflavor = 0;
2647 	if (VFS_NEEDSGIANT(mp))
2648 		error = ESTALE;
2649 	else
2650 		error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2651 	if (error != 0)
2652 		/* Make sure the server replies ESTALE to the client. */
2653 		error = ESTALE;
2654 	if (nam && !error) {
2655 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2656 		    &exp->nes_numsecflavor, &secflavors);
2657 		if (error) {
2658 			if (nfs_rootfhset) {
2659 				exp->nes_exflag = 0;
2660 				exp->nes_numsecflavor = 0;
2661 				error = 0;
2662 			} else {
2663 				vput(*vpp);
2664 			}
2665 		} else {
2666 			/* Copy the security flavors. */
2667 			for (i = 0; i < exp->nes_numsecflavor; i++)
2668 				exp->nes_secflavors[i] = secflavors[i];
2669 		}
2670 	}
2671 	NFSEXITCODE(error);
2672 	return (error);
2673 }
2674 
2675 /*
2676  * nfsd_fhtovp() - convert a fh to a vnode ptr
2677  * 	- look up fsid in mount list (if not found ret error)
2678  *	- get vp and export rights by calling nfsvno_fhtovp()
2679  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2680  *	  for AUTH_SYS
2681  *	- if mpp != NULL, return the mount point so that it can
2682  *	  be used for vn_finished_write() by the caller
2683  */
2684 void
2685 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2686     struct vnode **vpp, struct nfsexstuff *exp,
2687     struct mount **mpp, int startwrite, struct thread *p)
2688 {
2689 	struct mount *mp;
2690 	struct ucred *credanon;
2691 	fhandle_t *fhp;
2692 
2693 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
2694 	/*
2695 	 * Check for the special case of the nfsv4root_fh.
2696 	 */
2697 	mp = vfs_busyfs(&fhp->fh_fsid);
2698 	if (mpp != NULL)
2699 		*mpp = mp;
2700 	if (mp == NULL) {
2701 		*vpp = NULL;
2702 		nd->nd_repstat = ESTALE;
2703 		goto out;
2704 	}
2705 
2706 	if (startwrite)
2707 		vn_start_write(NULL, mpp, V_WAIT);
2708 
2709 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2710 	    &credanon);
2711 	vfs_unbusy(mp);
2712 
2713 	/*
2714 	 * For NFSv4 without a pseudo root fs, unexported file handles
2715 	 * can be returned, so that Lookup works everywhere.
2716 	 */
2717 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2718 	    !(nd->nd_flag & ND_NFSV4)) {
2719 		vput(*vpp);
2720 		nd->nd_repstat = EACCES;
2721 	}
2722 
2723 	/*
2724 	 * Personally, I've never seen any point in requiring a
2725 	 * reserved port#, since only in the rare case where the
2726 	 * clients are all boxes with secure system priviledges,
2727 	 * does it provide any enhanced security, but... some people
2728 	 * believe it to be useful and keep putting this code back in.
2729 	 * (There is also some "security checker" out there that
2730 	 *  complains if the nfs server doesn't enforce this.)
2731 	 * However, note the following:
2732 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
2733 	 *	required.
2734 	 * RFC2623 recommends that, if a reserved port# is checked for,
2735 	 *	that there be a way to turn that off--> ifdef'd.
2736 	 */
2737 #ifdef NFS_REQRSVPORT
2738 	if (!nd->nd_repstat) {
2739 		struct sockaddr_in *saddr;
2740 		struct sockaddr_in6 *saddr6;
2741 
2742 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2743 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2744 		if (!(nd->nd_flag & ND_NFSV4) &&
2745 		    ((saddr->sin_family == AF_INET &&
2746 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2747 		     (saddr6->sin6_family == AF_INET6 &&
2748 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2749 			vput(*vpp);
2750 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2751 		}
2752 	}
2753 #endif	/* NFS_REQRSVPORT */
2754 
2755 	/*
2756 	 * Check/setup credentials.
2757 	 */
2758 	if (!nd->nd_repstat) {
2759 		nd->nd_saveduid = nd->nd_cred->cr_uid;
2760 		nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2761 		if (nd->nd_repstat)
2762 			vput(*vpp);
2763 	}
2764 	if (credanon != NULL)
2765 		crfree(credanon);
2766 	if (nd->nd_repstat) {
2767 		if (startwrite)
2768 			vn_finished_write(mp);
2769 		*vpp = NULL;
2770 		if (mpp != NULL)
2771 			*mpp = NULL;
2772 	}
2773 
2774 out:
2775 	NFSEXITCODE2(0, nd);
2776 }
2777 
2778 /*
2779  * glue for fp.
2780  */
2781 int
2782 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2783 {
2784 	struct filedesc *fdp;
2785 	struct file *fp;
2786 	int error = 0;
2787 
2788 	fdp = p->td_proc->p_fd;
2789 	if (fd >= fdp->fd_nfiles ||
2790 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
2791 		error = EBADF;
2792 		goto out;
2793 	}
2794 	*fpp = fp;
2795 
2796 out:
2797 	NFSEXITCODE(error);
2798 	return (error);
2799 }
2800 
2801 /*
2802  * Called from nfssvc() to update the exports list. Just call
2803  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2804  * in the mount list.
2805  */
2806 int
2807 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2808 {
2809 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2810 	int error = 0;
2811 	struct nameidata nd;
2812 	fhandle_t fh;
2813 
2814 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2815 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2816 		nfs_rootfhset = 0;
2817 	else if (error == 0) {
2818 		if (nfsexargp->fspec == NULL) {
2819 			error = EPERM;
2820 			goto out;
2821 		}
2822 		/*
2823 		 * If fspec != NULL, this is the v4root path.
2824 		 */
2825 		NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_USERSPACE,
2826 		    nfsexargp->fspec, p);
2827 		if ((error = namei(&nd)) != 0)
2828 			goto out;
2829 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
2830 		vrele(nd.ni_vp);
2831 		if (!error) {
2832 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2833 			NFSBCOPY((caddr_t)&fh,
2834 			    nfs_rootfh.nfsrvfh_data,
2835 			    sizeof (fhandle_t));
2836 			nfs_rootfhset = 1;
2837 		}
2838 	}
2839 
2840 out:
2841 	NFSEXITCODE(error);
2842 	return (error);
2843 }
2844 
2845 /*
2846  * Get the tcp socket sequence numbers we need.
2847  * (Maybe this should be moved to the tcp sources?)
2848  */
2849 int
2850 nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap)
2851 {
2852 	struct inpcb *inp;
2853 	struct tcpcb *tp;
2854 	int error = 0;
2855 
2856 	inp = sotoinpcb(so);
2857 	KASSERT(inp != NULL, ("nfsrv_getsocksndseq: inp == NULL"));
2858 	INP_RLOCK(inp);
2859 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2860 		INP_RUNLOCK(inp);
2861 		error = EPIPE;
2862 		goto out;
2863 	}
2864 	tp = intotcpcb(inp);
2865 	if (tp->t_state != TCPS_ESTABLISHED) {
2866 		INP_RUNLOCK(inp);
2867 		error = EPIPE;
2868 		goto out;
2869 	}
2870 	*maxp = tp->snd_max;
2871 	*unap = tp->snd_una;
2872 	INP_RUNLOCK(inp);
2873 
2874 out:
2875 	NFSEXITCODE(error);
2876 	return (error);
2877 }
2878 
2879 /*
2880  * This function needs to test to see if the system is near its limit
2881  * for memory allocation via malloc() or mget() and return True iff
2882  * either of these resources are near their limit.
2883  * XXX (For now, this is just a stub.)
2884  */
2885 int nfsrv_testmalloclimit = 0;
2886 int
2887 nfsrv_mallocmget_limit(void)
2888 {
2889 	static int printmesg = 0;
2890 	static int testval = 1;
2891 
2892 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2893 		if ((printmesg++ % 100) == 0)
2894 			printf("nfsd: malloc/mget near limit\n");
2895 		return (1);
2896 	}
2897 	return (0);
2898 }
2899 
2900 /*
2901  * BSD specific initialization of a mount point.
2902  */
2903 void
2904 nfsd_mntinit(void)
2905 {
2906 	static int inited = 0;
2907 
2908 	if (inited)
2909 		return;
2910 	inited = 1;
2911 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2912 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2913 	nfsv4root_mnt.mnt_export = NULL;
2914 	TAILQ_INIT(&nfsv4root_opt);
2915 	TAILQ_INIT(&nfsv4root_newopt);
2916 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2917 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2918 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
2919 }
2920 
2921 /*
2922  * Get a vnode for a file handle, without checking exports, etc.
2923  */
2924 struct vnode *
2925 nfsvno_getvp(fhandle_t *fhp)
2926 {
2927 	struct mount *mp;
2928 	struct vnode *vp;
2929 	int error;
2930 
2931 	mp = vfs_busyfs(&fhp->fh_fsid);
2932 	if (mp == NULL)
2933 		return (NULL);
2934 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2935 	vfs_unbusy(mp);
2936 	if (error)
2937 		return (NULL);
2938 	return (vp);
2939 }
2940 
2941 /*
2942  * Do a local VOP_ADVLOCK().
2943  */
2944 int
2945 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2946     u_int64_t end, struct thread *td)
2947 {
2948 	int error = 0;
2949 	struct flock fl;
2950 	u_int64_t tlen;
2951 
2952 	if (nfsrv_dolocallocks == 0)
2953 		goto out;
2954 
2955 	/* Check for VI_DOOMED here, so that VOP_ADVLOCK() isn't performed. */
2956 	if ((vp->v_iflag & VI_DOOMED) != 0) {
2957 		error = EPERM;
2958 		goto out;
2959 	}
2960 
2961 	fl.l_whence = SEEK_SET;
2962 	fl.l_type = ftype;
2963 	fl.l_start = (off_t)first;
2964 	if (end == NFS64BITSSET) {
2965 		fl.l_len = 0;
2966 	} else {
2967 		tlen = end - first;
2968 		fl.l_len = (off_t)tlen;
2969 	}
2970 	/*
2971 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2972 	 * values for all calls, so that all locks will be held by the
2973 	 * nfsd server. (The nfsd server handles conflicts between the
2974 	 * various clients.)
2975 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2976 	 * bytes, so it can't be put in l_sysid.
2977 	 */
2978 	if (nfsv4_sysid == 0)
2979 		nfsv4_sysid = nlm_acquire_next_sysid();
2980 	fl.l_pid = (pid_t)0;
2981 	fl.l_sysid = (int)nfsv4_sysid;
2982 
2983 	NFSVOPUNLOCK(vp, 0);
2984 	if (ftype == F_UNLCK)
2985 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
2986 		    (F_POSIX | F_REMOTE));
2987 	else
2988 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
2989 		    (F_POSIX | F_REMOTE));
2990 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2991 
2992 out:
2993 	NFSEXITCODE(error);
2994 	return (error);
2995 }
2996 
2997 /*
2998  * Check the nfsv4 root exports.
2999  */
3000 int
3001 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3002 {
3003 	struct ucred *credanon;
3004 	int exflags, error = 0, numsecflavor, *secflavors, i;
3005 
3006 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3007 	    &credanon, &numsecflavor, &secflavors);
3008 	if (error) {
3009 		error = NFSERR_PROGUNAVAIL;
3010 		goto out;
3011 	}
3012 	if (credanon != NULL)
3013 		crfree(credanon);
3014 	for (i = 0; i < numsecflavor; i++) {
3015 		if (secflavors[i] == AUTH_SYS)
3016 			nd->nd_flag |= ND_EXAUTHSYS;
3017 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3018 			nd->nd_flag |= ND_EXGSS;
3019 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3020 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3021 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3022 			nd->nd_flag |= ND_EXGSSPRIVACY;
3023 	}
3024 
3025 out:
3026 	NFSEXITCODE(error);
3027 	return (error);
3028 }
3029 
3030 /*
3031  * Nfs server psuedo system call for the nfsd's
3032  */
3033 /*
3034  * MPSAFE
3035  */
3036 static int
3037 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3038 {
3039 	struct file *fp;
3040 	struct nfsd_addsock_args sockarg;
3041 	struct nfsd_nfsd_args nfsdarg;
3042 	int error;
3043 
3044 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3045 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3046 		if (error)
3047 			goto out;
3048 		/*
3049 		 * Since we don't know what rights might be required,
3050 		 * pretend that we need them all. It is better to be too
3051 		 * careful than too reckless.
3052 		 */
3053 		if ((error = fget(td, sockarg.sock, CAP_SOCK_ALL, &fp)) != 0)
3054 			goto out;
3055 		if (fp->f_type != DTYPE_SOCKET) {
3056 			fdrop(fp, td);
3057 			error = EPERM;
3058 			goto out;
3059 		}
3060 		error = nfsrvd_addsock(fp);
3061 		fdrop(fp, td);
3062 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3063 		if (uap->argp == NULL) {
3064 			error = EINVAL;
3065 			goto out;
3066 		}
3067 		error = copyin(uap->argp, (caddr_t)&nfsdarg,
3068 		    sizeof (nfsdarg));
3069 		if (error)
3070 			goto out;
3071 		error = nfsrvd_nfsd(td, &nfsdarg);
3072 	} else {
3073 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3074 	}
3075 
3076 out:
3077 	NFSEXITCODE(error);
3078 	return (error);
3079 }
3080 
3081 static int
3082 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3083 {
3084 	struct nfsex_args export;
3085 	struct file *fp = NULL;
3086 	int stablefd, len;
3087 	struct nfsd_clid adminrevoke;
3088 	struct nfsd_dumplist dumplist;
3089 	struct nfsd_dumpclients *dumpclients;
3090 	struct nfsd_dumplocklist dumplocklist;
3091 	struct nfsd_dumplocks *dumplocks;
3092 	struct nameidata nd;
3093 	vnode_t vp;
3094 	int error = EINVAL;
3095 	struct proc *procp;
3096 
3097 	if (uap->flag & NFSSVC_PUBLICFH) {
3098 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3099 		    sizeof (fhandle_t));
3100 		error = copyin(uap->argp,
3101 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3102 		if (!error)
3103 			nfs_pubfhset = 1;
3104 	} else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3105 		error = copyin(uap->argp,(caddr_t)&export,
3106 		    sizeof (struct nfsex_args));
3107 		if (!error)
3108 			error = nfsrv_v4rootexport(&export, cred, p);
3109 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3110 		nfs_pubfhset = 0;
3111 		error = 0;
3112 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3113 		error = copyin(uap->argp, (caddr_t)&stablefd,
3114 		    sizeof (int));
3115 		if (!error)
3116 			error = fp_getfvp(p, stablefd, &fp, &vp);
3117 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3118 			error = EBADF;
3119 		if (!error && newnfs_numnfsd != 0)
3120 			error = EPERM;
3121 		if (!error) {
3122 			nfsrv_stablefirst.nsf_fp = fp;
3123 			nfsrv_setupstable(p);
3124 		}
3125 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3126 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3127 		    sizeof (struct nfsd_clid));
3128 		if (!error)
3129 			error = nfsrv_adminrevoke(&adminrevoke, p);
3130 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3131 		error = copyin(uap->argp, (caddr_t)&dumplist,
3132 		    sizeof (struct nfsd_dumplist));
3133 		if (!error && (dumplist.ndl_size < 1 ||
3134 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3135 			error = EPERM;
3136 		if (!error) {
3137 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3138 		    dumpclients = (struct nfsd_dumpclients *)malloc(len,
3139 			M_TEMP, M_WAITOK);
3140 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3141 		    error = copyout(dumpclients,
3142 			CAST_USER_ADDR_T(dumplist.ndl_list), len);
3143 		    free((caddr_t)dumpclients, M_TEMP);
3144 		}
3145 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3146 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3147 		    sizeof (struct nfsd_dumplocklist));
3148 		if (!error && (dumplocklist.ndllck_size < 1 ||
3149 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3150 			error = EPERM;
3151 		if (!error)
3152 			error = nfsrv_lookupfilename(&nd,
3153 				dumplocklist.ndllck_fname, p);
3154 		if (!error) {
3155 			len = sizeof (struct nfsd_dumplocks) *
3156 				dumplocklist.ndllck_size;
3157 			dumplocks = (struct nfsd_dumplocks *)malloc(len,
3158 				M_TEMP, M_WAITOK);
3159 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3160 			    dumplocklist.ndllck_size, p);
3161 			vput(nd.ni_vp);
3162 			error = copyout(dumplocks,
3163 			    CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3164 			free((caddr_t)dumplocks, M_TEMP);
3165 		}
3166 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3167 		procp = p->td_proc;
3168 		PROC_LOCK(procp);
3169 		nfsd_master_pid = procp->p_pid;
3170 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3171 		nfsd_master_start = procp->p_stats->p_start;
3172 		nfsd_master_proc = procp;
3173 		PROC_UNLOCK(procp);
3174 	}
3175 
3176 	NFSEXITCODE(error);
3177 	return (error);
3178 }
3179 
3180 /*
3181  * Check exports.
3182  * Returns 0 if ok, 1 otherwise.
3183  */
3184 int
3185 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3186 {
3187 	int i;
3188 
3189 	/*
3190 	 * This seems odd, but allow the case where the security flavor
3191 	 * list is empty. This happens when NFSv4 is traversing non-exported
3192 	 * file systems. Exported file systems should always have a non-empty
3193 	 * security flavor list.
3194 	 */
3195 	if (exp->nes_numsecflavor == 0)
3196 		return (0);
3197 
3198 	for (i = 0; i < exp->nes_numsecflavor; i++) {
3199 		/*
3200 		 * The tests for privacy and integrity must be first,
3201 		 * since ND_GSS is set for everything but AUTH_SYS.
3202 		 */
3203 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3204 		    (nd->nd_flag & ND_GSSPRIVACY))
3205 			return (0);
3206 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3207 		    (nd->nd_flag & ND_GSSINTEGRITY))
3208 			return (0);
3209 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3210 		    (nd->nd_flag & ND_GSS))
3211 			return (0);
3212 		if (exp->nes_secflavors[i] == AUTH_SYS &&
3213 		    (nd->nd_flag & ND_GSS) == 0)
3214 			return (0);
3215 	}
3216 	return (1);
3217 }
3218 
3219 /*
3220  * Calculate a hash value for the fid in a file handle.
3221  */
3222 uint32_t
3223 nfsrv_hashfh(fhandle_t *fhp)
3224 {
3225 	uint32_t hashval;
3226 
3227 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3228 	return (hashval);
3229 }
3230 
3231 /*
3232  * Signal the userland master nfsd to backup the stable restart file.
3233  */
3234 void
3235 nfsrv_backupstable(void)
3236 {
3237 	struct proc *procp;
3238 
3239 	if (nfsd_master_proc != NULL) {
3240 		procp = pfind(nfsd_master_pid);
3241 		/* Try to make sure it is the correct process. */
3242 		if (procp == nfsd_master_proc &&
3243 		    procp->p_stats->p_start.tv_sec ==
3244 		    nfsd_master_start.tv_sec &&
3245 		    procp->p_stats->p_start.tv_usec ==
3246 		    nfsd_master_start.tv_usec &&
3247 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
3248 			kern_psignal(procp, SIGUSR2);
3249 		else
3250 			nfsd_master_proc = NULL;
3251 
3252 		if (procp != NULL)
3253 			PROC_UNLOCK(procp);
3254 	}
3255 }
3256 
3257 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3258 
3259 /*
3260  * Called once to initialize data structures...
3261  */
3262 static int
3263 nfsd_modevent(module_t mod, int type, void *data)
3264 {
3265 	int error = 0;
3266 	static int loaded = 0;
3267 
3268 	switch (type) {
3269 	case MOD_LOAD:
3270 		if (loaded)
3271 			goto out;
3272 		newnfs_portinit();
3273 		mtx_init(&nfs_cache_mutex, "nfs_cache_mutex", NULL, MTX_DEF);
3274 		mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3275 		mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3276 		    MTX_DEF);
3277 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3278 		nfsrvd_initcache();
3279 		nfsd_init();
3280 		NFSD_LOCK();
3281 		nfsrvd_init(0);
3282 		NFSD_UNLOCK();
3283 		nfsd_mntinit();
3284 #ifdef VV_DISABLEDELEG
3285 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3286 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3287 #endif
3288 		nfsd_call_servertimer = nfsrv_servertimer;
3289 		nfsd_call_nfsd = nfssvc_nfsd;
3290 		loaded = 1;
3291 		break;
3292 
3293 	case MOD_UNLOAD:
3294 		if (newnfs_numnfsd != 0) {
3295 			error = EBUSY;
3296 			break;
3297 		}
3298 
3299 #ifdef VV_DISABLEDELEG
3300 		vn_deleg_ops.vndeleg_recall = NULL;
3301 		vn_deleg_ops.vndeleg_disable = NULL;
3302 #endif
3303 		nfsd_call_servertimer = NULL;
3304 		nfsd_call_nfsd = NULL;
3305 
3306 		/* Clean out all NFSv4 state. */
3307 		nfsrv_throwawayallstate(curthread);
3308 
3309 		/* Clean the NFS server reply cache */
3310 		nfsrvd_cleancache();
3311 
3312 		/* Free up the krpc server pool. */
3313 		if (nfsrvd_pool != NULL)
3314 			svcpool_destroy(nfsrvd_pool);
3315 
3316 		/* and get rid of the locks */
3317 		mtx_destroy(&nfs_cache_mutex);
3318 		mtx_destroy(&nfs_v4root_mutex);
3319 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3320 		lockdestroy(&nfsv4root_mnt.mnt_explock);
3321 		loaded = 0;
3322 		break;
3323 	default:
3324 		error = EOPNOTSUPP;
3325 		break;
3326 	}
3327 
3328 out:
3329 	NFSEXITCODE(error);
3330 	return (error);
3331 }
3332 static moduledata_t nfsd_mod = {
3333 	"nfsd",
3334 	nfsd_modevent,
3335 	NULL,
3336 };
3337 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3338 
3339 /* So that loader and kldload(2) can find us, wherever we are.. */
3340 MODULE_VERSION(nfsd, 1);
3341 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3342 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3343 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3344 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3345 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3346 
3347