xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision bb15ca603fa442c72dde3f3cb8b46db6970e3950)
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 	if (cnt > MAX_COMMIT_COUNT) {
1258 		/*
1259 		 * Give up and do the whole thing
1260 		 */
1261 		if (vp->v_object &&
1262 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1263 			VM_OBJECT_LOCK(vp->v_object);
1264 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1265 			VM_OBJECT_UNLOCK(vp->v_object);
1266 		}
1267 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1268 	} else {
1269 		/*
1270 		 * Locate and synchronously write any buffers that fall
1271 		 * into the requested range.  Note:  we are assuming that
1272 		 * f_iosize is a power of 2.
1273 		 */
1274 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1275 		int iomask = iosize - 1;
1276 		struct bufobj *bo;
1277 		daddr_t lblkno;
1278 
1279 		/*
1280 		 * Align to iosize boundry, super-align to page boundry.
1281 		 */
1282 		if (off & iomask) {
1283 			cnt += off & iomask;
1284 			off &= ~(u_quad_t)iomask;
1285 		}
1286 		if (off & PAGE_MASK) {
1287 			cnt += off & PAGE_MASK;
1288 			off &= ~(u_quad_t)PAGE_MASK;
1289 		}
1290 		lblkno = off / iosize;
1291 
1292 		if (vp->v_object &&
1293 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1294 			VM_OBJECT_LOCK(vp->v_object);
1295 			vm_object_page_clean(vp->v_object, off, off + cnt,
1296 			    OBJPC_SYNC);
1297 			VM_OBJECT_UNLOCK(vp->v_object);
1298 		}
1299 
1300 		bo = &vp->v_bufobj;
1301 		BO_LOCK(bo);
1302 		while (cnt > 0) {
1303 			struct buf *bp;
1304 
1305 			/*
1306 			 * If we have a buffer and it is marked B_DELWRI we
1307 			 * have to lock and write it.  Otherwise the prior
1308 			 * write is assumed to have already been committed.
1309 			 *
1310 			 * gbincore() can return invalid buffers now so we
1311 			 * have to check that bit as well (though B_DELWRI
1312 			 * should not be set if B_INVAL is set there could be
1313 			 * a race here since we haven't locked the buffer).
1314 			 */
1315 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1316 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1317 				    LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
1318 					BO_LOCK(bo);
1319 					continue; /* retry */
1320 				}
1321 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1322 				    B_DELWRI) {
1323 					bremfree(bp);
1324 					bp->b_flags &= ~B_ASYNC;
1325 					bwrite(bp);
1326 					++nfs_commit_miss;
1327 				} else
1328 					BUF_UNLOCK(bp);
1329 				BO_LOCK(bo);
1330 			}
1331 			++nfs_commit_blks;
1332 			if (cnt < iosize)
1333 				break;
1334 			cnt -= iosize;
1335 			++lblkno;
1336 		}
1337 		BO_UNLOCK(bo);
1338 	}
1339 	NFSEXITCODE(error);
1340 	return (error);
1341 }
1342 
1343 /*
1344  * Statfs vnode op.
1345  */
1346 int
1347 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1348 {
1349 	int error;
1350 
1351 	error = VFS_STATFS(vp->v_mount, sf);
1352 	if (error == 0) {
1353 		/*
1354 		 * Since NFS handles these values as unsigned on the
1355 		 * wire, there is no way to represent negative values,
1356 		 * so set them to 0. Without this, they will appear
1357 		 * to be very large positive values for clients like
1358 		 * Solaris10.
1359 		 */
1360 		if (sf->f_bavail < 0)
1361 			sf->f_bavail = 0;
1362 		if (sf->f_ffree < 0)
1363 			sf->f_ffree = 0;
1364 	}
1365 	NFSEXITCODE(error);
1366 	return (error);
1367 }
1368 
1369 /*
1370  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1371  * must handle nfsrv_opencheck() calls after any other access checks.
1372  */
1373 void
1374 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1375     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1376     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1377     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1378     struct nfsexstuff *exp, struct vnode **vpp)
1379 {
1380 	struct vnode *vp = NULL;
1381 	u_quad_t tempsize;
1382 	struct nfsexstuff nes;
1383 
1384 	if (ndp->ni_vp == NULL)
1385 		nd->nd_repstat = nfsrv_opencheck(clientid,
1386 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1387 	if (!nd->nd_repstat) {
1388 		if (ndp->ni_vp == NULL) {
1389 			vrele(ndp->ni_startdir);
1390 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1391 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1392 			vput(ndp->ni_dvp);
1393 			nfsvno_relpathbuf(ndp);
1394 			if (!nd->nd_repstat) {
1395 				if (*exclusive_flagp) {
1396 					*exclusive_flagp = 0;
1397 					NFSVNO_ATTRINIT(nvap);
1398 					nvap->na_atime.tv_sec = cverf[0];
1399 					nvap->na_atime.tv_nsec = cverf[1];
1400 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1401 					    &nvap->na_vattr, cred);
1402 				} else {
1403 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1404 					    aclp, p, attrbitp, exp);
1405 				}
1406 			}
1407 			vp = ndp->ni_vp;
1408 		} else {
1409 			if (ndp->ni_startdir)
1410 				vrele(ndp->ni_startdir);
1411 			nfsvno_relpathbuf(ndp);
1412 			vp = ndp->ni_vp;
1413 			if (create == NFSV4OPEN_CREATE) {
1414 				if (ndp->ni_dvp == vp)
1415 					vrele(ndp->ni_dvp);
1416 				else
1417 					vput(ndp->ni_dvp);
1418 			}
1419 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1420 				if (ndp->ni_cnd.cn_flags & RDONLY)
1421 					NFSVNO_SETEXRDONLY(&nes);
1422 				else
1423 					NFSVNO_EXINIT(&nes);
1424 				nd->nd_repstat = nfsvno_accchk(vp,
1425 				    VWRITE, cred, &nes, p,
1426 				    NFSACCCHK_NOOVERRIDE,
1427 				    NFSACCCHK_VPISLOCKED, NULL);
1428 				nd->nd_repstat = nfsrv_opencheck(clientid,
1429 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1430 				if (!nd->nd_repstat) {
1431 					tempsize = nvap->na_size;
1432 					NFSVNO_ATTRINIT(nvap);
1433 					nvap->na_size = tempsize;
1434 					nd->nd_repstat = VOP_SETATTR(vp,
1435 					    &nvap->na_vattr, cred);
1436 				}
1437 			} else if (vp->v_type == VREG) {
1438 				nd->nd_repstat = nfsrv_opencheck(clientid,
1439 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1440 			}
1441 		}
1442 	} else {
1443 		if (ndp->ni_cnd.cn_flags & HASBUF)
1444 			nfsvno_relpathbuf(ndp);
1445 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1446 			vrele(ndp->ni_startdir);
1447 			if (ndp->ni_dvp == ndp->ni_vp)
1448 				vrele(ndp->ni_dvp);
1449 			else
1450 				vput(ndp->ni_dvp);
1451 			if (ndp->ni_vp)
1452 				vput(ndp->ni_vp);
1453 		}
1454 	}
1455 	*vpp = vp;
1456 
1457 	NFSEXITCODE2(0, nd);
1458 }
1459 
1460 /*
1461  * Updates the file rev and sets the mtime and ctime
1462  * to the current clock time, returning the va_filerev and va_Xtime
1463  * values.
1464  */
1465 void
1466 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1467     struct ucred *cred, struct thread *p)
1468 {
1469 	struct vattr va;
1470 
1471 	VATTR_NULL(&va);
1472 	getnanotime(&va.va_mtime);
1473 	(void) VOP_SETATTR(vp, &va, cred);
1474 	(void) nfsvno_getattr(vp, nvap, cred, p, 1);
1475 }
1476 
1477 /*
1478  * Glue routine to nfsv4_fillattr().
1479  */
1480 int
1481 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1482     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1483     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1484     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1485 {
1486 	int error;
1487 
1488 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1489 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1490 	    mounted_on_fileno);
1491 	NFSEXITCODE2(0, nd);
1492 	return (error);
1493 }
1494 
1495 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1496 /*
1497  * nfs readdir service
1498  * - mallocs what it thinks is enough to read
1499  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1500  * - calls VOP_READDIR()
1501  * - loops around building the reply
1502  *	if the output generated exceeds count break out of loop
1503  *	The NFSM_CLGET macro is used here so that the reply will be packed
1504  *	tightly in mbuf clusters.
1505  * - it trims out records with d_fileno == 0
1506  *	this doesn't matter for Unix clients, but they might confuse clients
1507  *	for other os'.
1508  * - it trims out records with d_type == DT_WHT
1509  *	these cannot be seen through NFS (unless we extend the protocol)
1510  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1511  * PS: The NFS protocol spec. does not clarify what the "count" byte
1512  *	argument is a count of.. just name strings and file id's or the
1513  *	entire reply rpc or ...
1514  *	I tried just file name and id sizes and it confused the Sun client,
1515  *	so I am using the full rpc size now. The "paranoia.." comment refers
1516  *	to including the status longwords that are not a part of the dir.
1517  *	"entry" structures, but are in the rpc.
1518  */
1519 int
1520 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1521     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1522 {
1523 	struct dirent *dp;
1524 	u_int32_t *tl;
1525 	int dirlen;
1526 	char *cpos, *cend, *rbuf;
1527 	struct nfsvattr at;
1528 	int nlen, error = 0, getret = 1;
1529 	int siz, cnt, fullsiz, eofflag, ncookies;
1530 	u_int64_t off, toff, verf;
1531 	u_long *cookies = NULL, *cookiep;
1532 	struct uio io;
1533 	struct iovec iv;
1534 	int not_zfs;
1535 
1536 	if (nd->nd_repstat) {
1537 		nfsrv_postopattr(nd, getret, &at);
1538 		goto out;
1539 	}
1540 	if (nd->nd_flag & ND_NFSV2) {
1541 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1542 		off = fxdr_unsigned(u_quad_t, *tl++);
1543 	} else {
1544 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1545 		off = fxdr_hyper(tl);
1546 		tl += 2;
1547 		verf = fxdr_hyper(tl);
1548 		tl += 2;
1549 	}
1550 	toff = off;
1551 	cnt = fxdr_unsigned(int, *tl);
1552 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1553 		cnt = NFS_SRVMAXDATA(nd);
1554 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1555 	fullsiz = siz;
1556 	if (nd->nd_flag & ND_NFSV3) {
1557 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1558 		    p, 1);
1559 #if 0
1560 		/*
1561 		 * va_filerev is not sufficient as a cookie verifier,
1562 		 * since it is not supposed to change when entries are
1563 		 * removed/added unless that offset cookies returned to
1564 		 * the client are no longer valid.
1565 		 */
1566 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
1567 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1568 #endif
1569 	}
1570 	if (nd->nd_repstat == 0 && cnt == 0) {
1571 		if (nd->nd_flag & ND_NFSV2)
1572 			/* NFSv2 does not have NFSERR_TOOSMALL */
1573 			nd->nd_repstat = EPERM;
1574 		else
1575 			nd->nd_repstat = NFSERR_TOOSMALL;
1576 	}
1577 	if (!nd->nd_repstat)
1578 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1579 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1580 		    NFSACCCHK_VPISLOCKED, NULL);
1581 	if (nd->nd_repstat) {
1582 		vput(vp);
1583 		if (nd->nd_flag & ND_NFSV3)
1584 			nfsrv_postopattr(nd, getret, &at);
1585 		goto out;
1586 	}
1587 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1588 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1589 again:
1590 	eofflag = 0;
1591 	if (cookies) {
1592 		free((caddr_t)cookies, M_TEMP);
1593 		cookies = NULL;
1594 	}
1595 
1596 	iv.iov_base = rbuf;
1597 	iv.iov_len = siz;
1598 	io.uio_iov = &iv;
1599 	io.uio_iovcnt = 1;
1600 	io.uio_offset = (off_t)off;
1601 	io.uio_resid = siz;
1602 	io.uio_segflg = UIO_SYSSPACE;
1603 	io.uio_rw = UIO_READ;
1604 	io.uio_td = NULL;
1605 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1606 	    &cookies);
1607 	off = (u_int64_t)io.uio_offset;
1608 	if (io.uio_resid)
1609 		siz -= io.uio_resid;
1610 
1611 	if (!cookies && !nd->nd_repstat)
1612 		nd->nd_repstat = NFSERR_PERM;
1613 	if (nd->nd_flag & ND_NFSV3) {
1614 		getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1615 		if (!nd->nd_repstat)
1616 			nd->nd_repstat = getret;
1617 	}
1618 
1619 	/*
1620 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
1621 	 */
1622 	if (nd->nd_repstat) {
1623 		vput(vp);
1624 		free((caddr_t)rbuf, M_TEMP);
1625 		if (cookies)
1626 			free((caddr_t)cookies, M_TEMP);
1627 		if (nd->nd_flag & ND_NFSV3)
1628 			nfsrv_postopattr(nd, getret, &at);
1629 		goto out;
1630 	}
1631 	/*
1632 	 * If nothing read, return eof
1633 	 * rpc reply
1634 	 */
1635 	if (siz == 0) {
1636 		vput(vp);
1637 		if (nd->nd_flag & ND_NFSV2) {
1638 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1639 		} else {
1640 			nfsrv_postopattr(nd, getret, &at);
1641 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1642 			txdr_hyper(at.na_filerev, tl);
1643 			tl += 2;
1644 		}
1645 		*tl++ = newnfs_false;
1646 		*tl = newnfs_true;
1647 		FREE((caddr_t)rbuf, M_TEMP);
1648 		FREE((caddr_t)cookies, M_TEMP);
1649 		goto out;
1650 	}
1651 
1652 	/*
1653 	 * Check for degenerate cases of nothing useful read.
1654 	 * If so go try again
1655 	 */
1656 	cpos = rbuf;
1657 	cend = rbuf + siz;
1658 	dp = (struct dirent *)cpos;
1659 	cookiep = cookies;
1660 
1661 	/*
1662 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1663 	 * directory offset up to a block boundary, so it is necessary to
1664 	 * skip over the records that precede the requested offset. This
1665 	 * requires the assumption that file offset cookies monotonically
1666 	 * increase.
1667 	 * Since the offset cookies don't monotonically increase for ZFS,
1668 	 * this is not done when ZFS is the file system.
1669 	 */
1670 	while (cpos < cend && ncookies > 0 &&
1671 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1672 	     (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
1673 		cpos += dp->d_reclen;
1674 		dp = (struct dirent *)cpos;
1675 		cookiep++;
1676 		ncookies--;
1677 	}
1678 	if (cpos >= cend || ncookies == 0) {
1679 		siz = fullsiz;
1680 		toff = off;
1681 		goto again;
1682 	}
1683 	vput(vp);
1684 
1685 	/*
1686 	 * dirlen is the size of the reply, including all XDR and must
1687 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1688 	 * if the XDR should be included in "count", but to be safe, we do.
1689 	 * (Include the two booleans at the end of the reply in dirlen now.)
1690 	 */
1691 	if (nd->nd_flag & ND_NFSV3) {
1692 		nfsrv_postopattr(nd, getret, &at);
1693 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1694 		txdr_hyper(at.na_filerev, tl);
1695 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1696 	} else {
1697 		dirlen = 2 * NFSX_UNSIGNED;
1698 	}
1699 
1700 	/* Loop through the records and build reply */
1701 	while (cpos < cend && ncookies > 0) {
1702 		nlen = dp->d_namlen;
1703 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1704 			nlen <= NFS_MAXNAMLEN) {
1705 			if (nd->nd_flag & ND_NFSV3)
1706 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1707 			else
1708 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1709 			if (dirlen > cnt) {
1710 				eofflag = 0;
1711 				break;
1712 			}
1713 
1714 			/*
1715 			 * Build the directory record xdr from
1716 			 * the dirent entry.
1717 			 */
1718 			if (nd->nd_flag & ND_NFSV3) {
1719 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1720 				*tl++ = newnfs_true;
1721 				*tl++ = 0;
1722 			} else {
1723 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1724 				*tl++ = newnfs_true;
1725 			}
1726 			*tl = txdr_unsigned(dp->d_fileno);
1727 			(void) nfsm_strtom(nd, dp->d_name, nlen);
1728 			if (nd->nd_flag & ND_NFSV3) {
1729 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1730 				*tl++ = 0;
1731 			} else
1732 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1733 			*tl = txdr_unsigned(*cookiep);
1734 		}
1735 		cpos += dp->d_reclen;
1736 		dp = (struct dirent *)cpos;
1737 		cookiep++;
1738 		ncookies--;
1739 	}
1740 	if (cpos < cend)
1741 		eofflag = 0;
1742 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1743 	*tl++ = newnfs_false;
1744 	if (eofflag)
1745 		*tl = newnfs_true;
1746 	else
1747 		*tl = newnfs_false;
1748 	FREE((caddr_t)rbuf, M_TEMP);
1749 	FREE((caddr_t)cookies, M_TEMP);
1750 
1751 out:
1752 	NFSEXITCODE2(0, nd);
1753 	return (0);
1754 nfsmout:
1755 	vput(vp);
1756 	NFSEXITCODE2(error, nd);
1757 	return (error);
1758 }
1759 
1760 /*
1761  * Readdirplus for V3 and Readdir for V4.
1762  */
1763 int
1764 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1765     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1766 {
1767 	struct dirent *dp;
1768 	u_int32_t *tl;
1769 	int dirlen;
1770 	char *cpos, *cend, *rbuf;
1771 	struct vnode *nvp;
1772 	fhandle_t nfh;
1773 	struct nfsvattr nva, at, *nvap = &nva;
1774 	struct mbuf *mb0, *mb1;
1775 	struct nfsreferral *refp;
1776 	int nlen, r, error = 0, getret = 1, usevget = 1;
1777 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1778 	caddr_t bpos0, bpos1;
1779 	u_int64_t off, toff, verf;
1780 	u_long *cookies = NULL, *cookiep;
1781 	nfsattrbit_t attrbits, rderrbits, savbits;
1782 	struct uio io;
1783 	struct iovec iv;
1784 	struct componentname cn;
1785 	int at_root, needs_unbusy, not_zfs, supports_nfsv4acls;
1786 	struct mount *mp, *new_mp;
1787 	uint64_t mounted_on_fileno;
1788 
1789 	if (nd->nd_repstat) {
1790 		nfsrv_postopattr(nd, getret, &at);
1791 		goto out;
1792 	}
1793 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1794 	off = fxdr_hyper(tl);
1795 	toff = off;
1796 	tl += 2;
1797 	verf = fxdr_hyper(tl);
1798 	tl += 2;
1799 	siz = fxdr_unsigned(int, *tl++);
1800 	cnt = fxdr_unsigned(int, *tl);
1801 
1802 	/*
1803 	 * Use the server's maximum data transfer size as the upper bound
1804 	 * on reply datalen.
1805 	 */
1806 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1807 		cnt = NFS_SRVMAXDATA(nd);
1808 
1809 	/*
1810 	 * siz is a "hint" of how much directory information (name, fileid,
1811 	 * cookie) should be in the reply. At least one client "hints" 0,
1812 	 * so I set it to cnt for that case. I also round it up to the
1813 	 * next multiple of DIRBLKSIZ.
1814 	 */
1815 	if (siz <= 0)
1816 		siz = cnt;
1817 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1818 
1819 	if (nd->nd_flag & ND_NFSV4) {
1820 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1821 		if (error)
1822 			goto nfsmout;
1823 		NFSSET_ATTRBIT(&savbits, &attrbits);
1824 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1825 		NFSZERO_ATTRBIT(&rderrbits);
1826 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1827 	} else {
1828 		NFSZERO_ATTRBIT(&attrbits);
1829 	}
1830 	fullsiz = siz;
1831 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1832 	if (!nd->nd_repstat) {
1833 	    if (off && verf != at.na_filerev) {
1834 		/*
1835 		 * va_filerev is not sufficient as a cookie verifier,
1836 		 * since it is not supposed to change when entries are
1837 		 * removed/added unless that offset cookies returned to
1838 		 * the client are no longer valid.
1839 		 */
1840 #if 0
1841 		if (nd->nd_flag & ND_NFSV4) {
1842 			nd->nd_repstat = NFSERR_NOTSAME;
1843 		} else {
1844 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1845 		}
1846 #endif
1847 	    } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1848 		nd->nd_repstat = NFSERR_BAD_COOKIE;
1849 	    }
1850 	}
1851 	if (!nd->nd_repstat && vp->v_type != VDIR)
1852 		nd->nd_repstat = NFSERR_NOTDIR;
1853 	if (!nd->nd_repstat && cnt == 0)
1854 		nd->nd_repstat = NFSERR_TOOSMALL;
1855 	if (!nd->nd_repstat)
1856 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1857 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1858 		    NFSACCCHK_VPISLOCKED, NULL);
1859 	if (nd->nd_repstat) {
1860 		vput(vp);
1861 		if (nd->nd_flag & ND_NFSV3)
1862 			nfsrv_postopattr(nd, getret, &at);
1863 		goto out;
1864 	}
1865 	not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1866 
1867 	MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1868 again:
1869 	eofflag = 0;
1870 	if (cookies) {
1871 		free((caddr_t)cookies, M_TEMP);
1872 		cookies = NULL;
1873 	}
1874 
1875 	iv.iov_base = rbuf;
1876 	iv.iov_len = siz;
1877 	io.uio_iov = &iv;
1878 	io.uio_iovcnt = 1;
1879 	io.uio_offset = (off_t)off;
1880 	io.uio_resid = siz;
1881 	io.uio_segflg = UIO_SYSSPACE;
1882 	io.uio_rw = UIO_READ;
1883 	io.uio_td = NULL;
1884 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1885 	    &cookies);
1886 	off = (u_int64_t)io.uio_offset;
1887 	if (io.uio_resid)
1888 		siz -= io.uio_resid;
1889 
1890 	getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1891 
1892 	if (!cookies && !nd->nd_repstat)
1893 		nd->nd_repstat = NFSERR_PERM;
1894 	if (!nd->nd_repstat)
1895 		nd->nd_repstat = getret;
1896 	if (nd->nd_repstat) {
1897 		vput(vp);
1898 		if (cookies)
1899 			free((caddr_t)cookies, M_TEMP);
1900 		free((caddr_t)rbuf, M_TEMP);
1901 		if (nd->nd_flag & ND_NFSV3)
1902 			nfsrv_postopattr(nd, getret, &at);
1903 		goto out;
1904 	}
1905 	/*
1906 	 * If nothing read, return eof
1907 	 * rpc reply
1908 	 */
1909 	if (siz == 0) {
1910 		vput(vp);
1911 		if (nd->nd_flag & ND_NFSV3)
1912 			nfsrv_postopattr(nd, getret, &at);
1913 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1914 		txdr_hyper(at.na_filerev, tl);
1915 		tl += 2;
1916 		*tl++ = newnfs_false;
1917 		*tl = newnfs_true;
1918 		free((caddr_t)cookies, M_TEMP);
1919 		free((caddr_t)rbuf, M_TEMP);
1920 		goto out;
1921 	}
1922 
1923 	/*
1924 	 * Check for degenerate cases of nothing useful read.
1925 	 * If so go try again
1926 	 */
1927 	cpos = rbuf;
1928 	cend = rbuf + siz;
1929 	dp = (struct dirent *)cpos;
1930 	cookiep = cookies;
1931 
1932 	/*
1933 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1934 	 * directory offset up to a block boundary, so it is necessary to
1935 	 * skip over the records that precede the requested offset. This
1936 	 * requires the assumption that file offset cookies monotonically
1937 	 * increase.
1938 	 * Since the offset cookies don't monotonically increase for ZFS,
1939 	 * this is not done when ZFS is the file system.
1940 	 */
1941 	while (cpos < cend && ncookies > 0 &&
1942 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1943 	   (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff) ||
1944 	   ((nd->nd_flag & ND_NFSV4) &&
1945 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1946 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1947 		cpos += dp->d_reclen;
1948 		dp = (struct dirent *)cpos;
1949 		cookiep++;
1950 		ncookies--;
1951 	}
1952 	if (cpos >= cend || ncookies == 0) {
1953 		siz = fullsiz;
1954 		toff = off;
1955 		goto again;
1956 	}
1957 
1958 	/*
1959 	 * Busy the file system so that the mount point won't go away
1960 	 * and, as such, VFS_VGET() can be used safely.
1961 	 */
1962 	mp = vp->v_mount;
1963 	vfs_ref(mp);
1964 	NFSVOPUNLOCK(vp, 0);
1965 	nd->nd_repstat = vfs_busy(mp, 0);
1966 	vfs_rel(mp);
1967 	if (nd->nd_repstat != 0) {
1968 		vrele(vp);
1969 		free(cookies, M_TEMP);
1970 		free(rbuf, M_TEMP);
1971 		if (nd->nd_flag & ND_NFSV3)
1972 			nfsrv_postopattr(nd, getret, &at);
1973 		goto out;
1974 	}
1975 
1976 	/*
1977 	 * Save this position, in case there is an error before one entry
1978 	 * is created.
1979 	 */
1980 	mb0 = nd->nd_mb;
1981 	bpos0 = nd->nd_bpos;
1982 
1983 	/*
1984 	 * Fill in the first part of the reply.
1985 	 * dirlen is the reply length in bytes and cannot exceed cnt.
1986 	 * (Include the two booleans at the end of the reply in dirlen now,
1987 	 *  so we recognize when we have exceeded cnt.)
1988 	 */
1989 	if (nd->nd_flag & ND_NFSV3) {
1990 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1991 		nfsrv_postopattr(nd, getret, &at);
1992 	} else {
1993 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
1994 	}
1995 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
1996 	txdr_hyper(at.na_filerev, tl);
1997 
1998 	/*
1999 	 * Save this position, in case there is an empty reply needed.
2000 	 */
2001 	mb1 = nd->nd_mb;
2002 	bpos1 = nd->nd_bpos;
2003 
2004 	/* Loop through the records and build reply */
2005 	entrycnt = 0;
2006 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2007 		nlen = dp->d_namlen;
2008 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2009 		    nlen <= NFS_MAXNAMLEN &&
2010 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2011 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2012 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2013 			/*
2014 			 * Save the current position in the reply, in case
2015 			 * this entry exceeds cnt.
2016 			 */
2017 			mb1 = nd->nd_mb;
2018 			bpos1 = nd->nd_bpos;
2019 
2020 			/*
2021 			 * For readdir_and_lookup get the vnode using
2022 			 * the file number.
2023 			 */
2024 			nvp = NULL;
2025 			refp = NULL;
2026 			r = 0;
2027 			at_root = 0;
2028 			needs_unbusy = 0;
2029 			new_mp = mp;
2030 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2031 			if ((nd->nd_flag & ND_NFSV3) ||
2032 			    NFSNONZERO_ATTRBIT(&savbits)) {
2033 				if (nd->nd_flag & ND_NFSV4)
2034 					refp = nfsv4root_getreferral(NULL,
2035 					    vp, dp->d_fileno);
2036 				if (refp == NULL) {
2037 					if (usevget)
2038 						r = VFS_VGET(mp, dp->d_fileno,
2039 						    LK_SHARED, &nvp);
2040 					else
2041 						r = EOPNOTSUPP;
2042 					if (r == EOPNOTSUPP) {
2043 						if (usevget) {
2044 							usevget = 0;
2045 							cn.cn_nameiop = LOOKUP;
2046 							cn.cn_lkflags =
2047 							    LK_SHARED |
2048 							    LK_RETRY;
2049 							cn.cn_cred =
2050 							    nd->nd_cred;
2051 							cn.cn_thread = p;
2052 						}
2053 						cn.cn_nameptr = dp->d_name;
2054 						cn.cn_namelen = nlen;
2055 						cn.cn_flags = ISLASTCN |
2056 						    NOFOLLOW | LOCKLEAF |
2057 						    MPSAFE;
2058 						if (nlen == 2 &&
2059 						    dp->d_name[0] == '.' &&
2060 						    dp->d_name[1] == '.')
2061 							cn.cn_flags |=
2062 							    ISDOTDOT;
2063 						if (NFSVOPLOCK(vp, LK_SHARED)
2064 						    != 0) {
2065 							nd->nd_repstat = EPERM;
2066 							break;
2067 						}
2068 						if ((vp->v_vflag & VV_ROOT) != 0
2069 						    && (cn.cn_flags & ISDOTDOT)
2070 						    != 0) {
2071 							vref(vp);
2072 							nvp = vp;
2073 							r = 0;
2074 						} else {
2075 							r = VOP_LOOKUP(vp, &nvp,
2076 							    &cn);
2077 							if (vp != nvp)
2078 								NFSVOPUNLOCK(vp,
2079 								    0);
2080 						}
2081 					}
2082 
2083 					/*
2084 					 * For NFSv4, check to see if nvp is
2085 					 * a mount point and get the mount
2086 					 * point vnode, as required.
2087 					 */
2088 					if (r == 0 &&
2089 					    nfsrv_enable_crossmntpt != 0 &&
2090 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2091 					    nvp->v_type == VDIR &&
2092 					    nvp->v_mountedhere != NULL) {
2093 						new_mp = nvp->v_mountedhere;
2094 						r = vfs_busy(new_mp, 0);
2095 						vput(nvp);
2096 						nvp = NULL;
2097 						if (r == 0) {
2098 							r = VFS_ROOT(new_mp,
2099 							    LK_SHARED, &nvp);
2100 							needs_unbusy = 1;
2101 							if (r == 0)
2102 								at_root = 1;
2103 						}
2104 					}
2105 				}
2106 				if (!r) {
2107 				    if (refp == NULL &&
2108 					((nd->nd_flag & ND_NFSV3) ||
2109 					 NFSNONZERO_ATTRBIT(&attrbits))) {
2110 					r = nfsvno_getfh(nvp, &nfh, p);
2111 					if (!r)
2112 					    r = nfsvno_getattr(nvp, nvap,
2113 						nd->nd_cred, p, 1);
2114 				    }
2115 				} else {
2116 				    nvp = NULL;
2117 				}
2118 				if (r) {
2119 					if (!NFSISSET_ATTRBIT(&attrbits,
2120 					    NFSATTRBIT_RDATTRERROR)) {
2121 						if (nvp != NULL)
2122 							vput(nvp);
2123 						if (needs_unbusy != 0)
2124 							vfs_unbusy(new_mp);
2125 						nd->nd_repstat = r;
2126 						break;
2127 					}
2128 				}
2129 			}
2130 
2131 			/*
2132 			 * Build the directory record xdr
2133 			 */
2134 			if (nd->nd_flag & ND_NFSV3) {
2135 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2136 				*tl++ = newnfs_true;
2137 				*tl++ = 0;
2138 				*tl = txdr_unsigned(dp->d_fileno);
2139 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2140 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2141 				*tl++ = 0;
2142 				*tl = txdr_unsigned(*cookiep);
2143 				nfsrv_postopattr(nd, 0, nvap);
2144 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2145 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2146 				if (nvp != NULL)
2147 					vput(nvp);
2148 			} else {
2149 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2150 				*tl++ = newnfs_true;
2151 				*tl++ = 0;
2152 				*tl = txdr_unsigned(*cookiep);
2153 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2154 				if (nvp != NULL) {
2155 					supports_nfsv4acls =
2156 					    nfs_supportsnfsv4acls(nvp);
2157 					NFSVOPUNLOCK(nvp, 0);
2158 				} else
2159 					supports_nfsv4acls = 0;
2160 				if (refp != NULL) {
2161 					dirlen += nfsrv_putreferralattr(nd,
2162 					    &savbits, refp, 0,
2163 					    &nd->nd_repstat);
2164 					if (nd->nd_repstat) {
2165 						if (nvp != NULL)
2166 							vrele(nvp);
2167 						if (needs_unbusy != 0)
2168 							vfs_unbusy(new_mp);
2169 						break;
2170 					}
2171 				} else if (r) {
2172 					dirlen += nfsvno_fillattr(nd, new_mp,
2173 					    nvp, nvap, &nfh, r, &rderrbits,
2174 					    nd->nd_cred, p, isdgram, 0,
2175 					    supports_nfsv4acls, at_root,
2176 					    mounted_on_fileno);
2177 				} else {
2178 					dirlen += nfsvno_fillattr(nd, new_mp,
2179 					    nvp, nvap, &nfh, r, &attrbits,
2180 					    nd->nd_cred, p, isdgram, 0,
2181 					    supports_nfsv4acls, at_root,
2182 					    mounted_on_fileno);
2183 				}
2184 				if (nvp != NULL)
2185 					vrele(nvp);
2186 				dirlen += (3 * NFSX_UNSIGNED);
2187 			}
2188 			if (needs_unbusy != 0)
2189 				vfs_unbusy(new_mp);
2190 			if (dirlen <= cnt)
2191 				entrycnt++;
2192 		}
2193 		cpos += dp->d_reclen;
2194 		dp = (struct dirent *)cpos;
2195 		cookiep++;
2196 		ncookies--;
2197 	}
2198 	vrele(vp);
2199 	vfs_unbusy(mp);
2200 
2201 	/*
2202 	 * If dirlen > cnt, we must strip off the last entry. If that
2203 	 * results in an empty reply, report NFSERR_TOOSMALL.
2204 	 */
2205 	if (dirlen > cnt || nd->nd_repstat) {
2206 		if (!nd->nd_repstat && entrycnt == 0)
2207 			nd->nd_repstat = NFSERR_TOOSMALL;
2208 		if (nd->nd_repstat)
2209 			newnfs_trimtrailing(nd, mb0, bpos0);
2210 		else
2211 			newnfs_trimtrailing(nd, mb1, bpos1);
2212 		eofflag = 0;
2213 	} else if (cpos < cend)
2214 		eofflag = 0;
2215 	if (!nd->nd_repstat) {
2216 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2217 		*tl++ = newnfs_false;
2218 		if (eofflag)
2219 			*tl = newnfs_true;
2220 		else
2221 			*tl = newnfs_false;
2222 	}
2223 	FREE((caddr_t)cookies, M_TEMP);
2224 	FREE((caddr_t)rbuf, M_TEMP);
2225 
2226 out:
2227 	NFSEXITCODE2(0, nd);
2228 	return (0);
2229 nfsmout:
2230 	vput(vp);
2231 	NFSEXITCODE2(error, nd);
2232 	return (error);
2233 }
2234 
2235 /*
2236  * Get the settable attributes out of the mbuf list.
2237  * (Return 0 or EBADRPC)
2238  */
2239 int
2240 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2241     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2242 {
2243 	u_int32_t *tl;
2244 	struct nfsv2_sattr *sp;
2245 	struct timeval curtime;
2246 	int error = 0, toclient = 0;
2247 
2248 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2249 	case ND_NFSV2:
2250 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2251 		/*
2252 		 * Some old clients didn't fill in the high order 16bits.
2253 		 * --> check the low order 2 bytes for 0xffff
2254 		 */
2255 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2256 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2257 		if (sp->sa_uid != newnfs_xdrneg1)
2258 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2259 		if (sp->sa_gid != newnfs_xdrneg1)
2260 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2261 		if (sp->sa_size != newnfs_xdrneg1)
2262 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2263 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2264 #ifdef notyet
2265 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2266 #else
2267 			nvap->na_atime.tv_sec =
2268 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2269 			nvap->na_atime.tv_nsec = 0;
2270 #endif
2271 		}
2272 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2273 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2274 		break;
2275 	case ND_NFSV3:
2276 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2277 		if (*tl == newnfs_true) {
2278 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2279 			nvap->na_mode = nfstov_mode(*tl);
2280 		}
2281 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2282 		if (*tl == newnfs_true) {
2283 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2284 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2285 		}
2286 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2287 		if (*tl == newnfs_true) {
2288 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2289 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2290 		}
2291 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2292 		if (*tl == newnfs_true) {
2293 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2294 			nvap->na_size = fxdr_hyper(tl);
2295 		}
2296 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2297 		switch (fxdr_unsigned(int, *tl)) {
2298 		case NFSV3SATTRTIME_TOCLIENT:
2299 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2300 			fxdr_nfsv3time(tl, &nvap->na_atime);
2301 			toclient = 1;
2302 			break;
2303 		case NFSV3SATTRTIME_TOSERVER:
2304 			NFSGETTIME(&curtime);
2305 			nvap->na_atime.tv_sec = curtime.tv_sec;
2306 			nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2307 			nvap->na_vaflags |= VA_UTIMES_NULL;
2308 			break;
2309 		};
2310 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2311 		switch (fxdr_unsigned(int, *tl)) {
2312 		case NFSV3SATTRTIME_TOCLIENT:
2313 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2314 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2315 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2316 			break;
2317 		case NFSV3SATTRTIME_TOSERVER:
2318 			NFSGETTIME(&curtime);
2319 			nvap->na_mtime.tv_sec = curtime.tv_sec;
2320 			nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2321 			if (!toclient)
2322 				nvap->na_vaflags |= VA_UTIMES_NULL;
2323 			break;
2324 		};
2325 		break;
2326 	case ND_NFSV4:
2327 		error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2328 	};
2329 nfsmout:
2330 	NFSEXITCODE2(error, nd);
2331 	return (error);
2332 }
2333 
2334 /*
2335  * Handle the setable attributes for V4.
2336  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2337  */
2338 int
2339 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2340     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2341 {
2342 	u_int32_t *tl;
2343 	int attrsum = 0;
2344 	int i, j;
2345 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2346 	int toclient = 0;
2347 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2348 	uid_t uid;
2349 	gid_t gid;
2350 	struct timeval curtime;
2351 
2352 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2353 	if (error)
2354 		goto nfsmout;
2355 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2356 	attrsize = fxdr_unsigned(int, *tl);
2357 
2358 	/*
2359 	 * Loop around getting the setable attributes. If an unsupported
2360 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2361 	 */
2362 	if (retnotsup) {
2363 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2364 		bitpos = NFSATTRBIT_MAX;
2365 	} else {
2366 		bitpos = 0;
2367 	}
2368 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2369 	    if (attrsum > attrsize) {
2370 		error = NFSERR_BADXDR;
2371 		goto nfsmout;
2372 	    }
2373 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2374 		switch (bitpos) {
2375 		case NFSATTRBIT_SIZE:
2376 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2377 			nvap->na_size = fxdr_hyper(tl);
2378 			attrsum += NFSX_HYPER;
2379 			break;
2380 		case NFSATTRBIT_ACL:
2381 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2382 			    p);
2383 			if (error)
2384 				goto nfsmout;
2385 			if (aceerr && !nd->nd_repstat)
2386 				nd->nd_repstat = aceerr;
2387 			attrsum += aclsize;
2388 			break;
2389 		case NFSATTRBIT_ARCHIVE:
2390 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2391 			if (!nd->nd_repstat)
2392 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2393 			attrsum += NFSX_UNSIGNED;
2394 			break;
2395 		case NFSATTRBIT_HIDDEN:
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_MIMETYPE:
2402 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2403 			i = fxdr_unsigned(int, *tl);
2404 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2405 			if (error)
2406 				goto nfsmout;
2407 			if (!nd->nd_repstat)
2408 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2409 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2410 			break;
2411 		case NFSATTRBIT_MODE:
2412 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2413 			nvap->na_mode = nfstov_mode(*tl);
2414 			attrsum += NFSX_UNSIGNED;
2415 			break;
2416 		case NFSATTRBIT_OWNER:
2417 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2418 			j = fxdr_unsigned(int, *tl);
2419 			if (j < 0) {
2420 				error = NFSERR_BADXDR;
2421 				goto nfsmout;
2422 			}
2423 			if (j > NFSV4_SMALLSTR)
2424 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2425 			else
2426 				cp = namestr;
2427 			error = nfsrv_mtostr(nd, cp, j);
2428 			if (error) {
2429 				if (j > NFSV4_SMALLSTR)
2430 					free(cp, M_NFSSTRING);
2431 				goto nfsmout;
2432 			}
2433 			if (!nd->nd_repstat) {
2434 				nd->nd_repstat = nfsv4_strtouid(cp,j,&uid,p);
2435 				if (!nd->nd_repstat)
2436 					nvap->na_uid = uid;
2437 			}
2438 			if (j > NFSV4_SMALLSTR)
2439 				free(cp, M_NFSSTRING);
2440 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2441 			break;
2442 		case NFSATTRBIT_OWNERGROUP:
2443 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2444 			j = fxdr_unsigned(int, *tl);
2445 			if (j < 0) {
2446 				error = NFSERR_BADXDR;
2447 				goto nfsmout;
2448 			}
2449 			if (j > NFSV4_SMALLSTR)
2450 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2451 			else
2452 				cp = namestr;
2453 			error = nfsrv_mtostr(nd, cp, j);
2454 			if (error) {
2455 				if (j > NFSV4_SMALLSTR)
2456 					free(cp, M_NFSSTRING);
2457 				goto nfsmout;
2458 			}
2459 			if (!nd->nd_repstat) {
2460 				nd->nd_repstat = nfsv4_strtogid(cp,j,&gid,p);
2461 				if (!nd->nd_repstat)
2462 					nvap->na_gid = gid;
2463 			}
2464 			if (j > NFSV4_SMALLSTR)
2465 				free(cp, M_NFSSTRING);
2466 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2467 			break;
2468 		case NFSATTRBIT_SYSTEM:
2469 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2470 			if (!nd->nd_repstat)
2471 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2472 			attrsum += NFSX_UNSIGNED;
2473 			break;
2474 		case NFSATTRBIT_TIMEACCESSSET:
2475 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2476 			attrsum += NFSX_UNSIGNED;
2477 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2478 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2479 			    fxdr_nfsv4time(tl, &nvap->na_atime);
2480 			    toclient = 1;
2481 			    attrsum += NFSX_V4TIME;
2482 			} else {
2483 			    NFSGETTIME(&curtime);
2484 			    nvap->na_atime.tv_sec = curtime.tv_sec;
2485 			    nvap->na_atime.tv_nsec = curtime.tv_usec * 1000;
2486 			    nvap->na_vaflags |= VA_UTIMES_NULL;
2487 			}
2488 			break;
2489 		case NFSATTRBIT_TIMEBACKUP:
2490 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2491 			if (!nd->nd_repstat)
2492 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2493 			attrsum += NFSX_V4TIME;
2494 			break;
2495 		case NFSATTRBIT_TIMECREATE:
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_TIMEMODIFYSET:
2502 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2503 			attrsum += NFSX_UNSIGNED;
2504 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2505 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2506 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
2507 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
2508 			    attrsum += NFSX_V4TIME;
2509 			} else {
2510 			    NFSGETTIME(&curtime);
2511 			    nvap->na_mtime.tv_sec = curtime.tv_sec;
2512 			    nvap->na_mtime.tv_nsec = curtime.tv_usec * 1000;
2513 			    if (!toclient)
2514 				nvap->na_vaflags |= VA_UTIMES_NULL;
2515 			}
2516 			break;
2517 		default:
2518 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2519 			/*
2520 			 * set bitpos so we drop out of the loop.
2521 			 */
2522 			bitpos = NFSATTRBIT_MAX;
2523 			break;
2524 		};
2525 	}
2526 
2527 	/*
2528 	 * some clients pad the attrlist, so we need to skip over the
2529 	 * padding.
2530 	 */
2531 	if (attrsum > attrsize) {
2532 		error = NFSERR_BADXDR;
2533 	} else {
2534 		attrsize = NFSM_RNDUP(attrsize);
2535 		if (attrsum < attrsize)
2536 			error = nfsm_advance(nd, attrsize - attrsum, -1);
2537 	}
2538 nfsmout:
2539 	NFSEXITCODE2(error, nd);
2540 	return (error);
2541 }
2542 
2543 /*
2544  * Check/setup export credentials.
2545  */
2546 int
2547 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2548     struct ucred *credanon)
2549 {
2550 	int error = 0;
2551 
2552 	/*
2553 	 * Check/setup credentials.
2554 	 */
2555 	if (nd->nd_flag & ND_GSS)
2556 		exp->nes_exflag &= ~MNT_EXPORTANON;
2557 
2558 	/*
2559 	 * Check to see if the operation is allowed for this security flavor.
2560 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2561 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2562 	 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2563 	 */
2564 	if (nfsvno_testexp(nd, exp) &&
2565 	    nd->nd_procnum != NFSV4OP_SECINFO &&
2566 	    nd->nd_procnum != NFSPROC_FSINFO) {
2567 		if (nd->nd_flag & ND_NFSV4)
2568 			error = NFSERR_WRONGSEC;
2569 		else
2570 			error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2571 		goto out;
2572 	}
2573 
2574 	/*
2575 	 * Check to see if the file system is exported V4 only.
2576 	 */
2577 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2578 		error = NFSERR_PROGNOTV4;
2579 		goto out;
2580 	}
2581 
2582 	/*
2583 	 * Now, map the user credentials.
2584 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
2585 	 *  Fsinfo RPC. If set for anything else, this code might need
2586 	 *  to change.)
2587 	 */
2588 	if (NFSVNO_EXPORTED(exp) &&
2589 	    ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2590 	     NFSVNO_EXPORTANON(exp) ||
2591 	     (nd->nd_flag & ND_AUTHNONE))) {
2592 		nd->nd_cred->cr_uid = credanon->cr_uid;
2593 		nd->nd_cred->cr_gid = credanon->cr_gid;
2594 		crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2595 		    credanon->cr_groups);
2596 	}
2597 
2598 out:
2599 	NFSEXITCODE2(error, nd);
2600 	return (error);
2601 }
2602 
2603 /*
2604  * Check exports.
2605  */
2606 int
2607 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2608     struct ucred **credp)
2609 {
2610 	int i, error, *secflavors;
2611 
2612 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2613 	    &exp->nes_numsecflavor, &secflavors);
2614 	if (error) {
2615 		if (nfs_rootfhset) {
2616 			exp->nes_exflag = 0;
2617 			exp->nes_numsecflavor = 0;
2618 			error = 0;
2619 		}
2620 	} else {
2621 		/* Copy the security flavors. */
2622 		for (i = 0; i < exp->nes_numsecflavor; i++)
2623 			exp->nes_secflavors[i] = secflavors[i];
2624 	}
2625 	NFSEXITCODE(error);
2626 	return (error);
2627 }
2628 
2629 /*
2630  * Get a vnode for a file handle and export stuff.
2631  */
2632 int
2633 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2634     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2635     struct ucred **credp)
2636 {
2637 	int i, error, *secflavors;
2638 
2639 	*credp = NULL;
2640 	exp->nes_numsecflavor = 0;
2641 	if (VFS_NEEDSGIANT(mp))
2642 		error = ESTALE;
2643 	else
2644 		error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2645 	if (error != 0)
2646 		/* Make sure the server replies ESTALE to the client. */
2647 		error = ESTALE;
2648 	if (nam && !error) {
2649 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2650 		    &exp->nes_numsecflavor, &secflavors);
2651 		if (error) {
2652 			if (nfs_rootfhset) {
2653 				exp->nes_exflag = 0;
2654 				exp->nes_numsecflavor = 0;
2655 				error = 0;
2656 			} else {
2657 				vput(*vpp);
2658 			}
2659 		} else {
2660 			/* Copy the security flavors. */
2661 			for (i = 0; i < exp->nes_numsecflavor; i++)
2662 				exp->nes_secflavors[i] = secflavors[i];
2663 		}
2664 	}
2665 	NFSEXITCODE(error);
2666 	return (error);
2667 }
2668 
2669 /*
2670  * nfsd_fhtovp() - convert a fh to a vnode ptr
2671  * 	- look up fsid in mount list (if not found ret error)
2672  *	- get vp and export rights by calling nfsvno_fhtovp()
2673  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2674  *	  for AUTH_SYS
2675  *	- if mpp != NULL, return the mount point so that it can
2676  *	  be used for vn_finished_write() by the caller
2677  */
2678 void
2679 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2680     struct vnode **vpp, struct nfsexstuff *exp,
2681     struct mount **mpp, int startwrite, struct thread *p)
2682 {
2683 	struct mount *mp;
2684 	struct ucred *credanon;
2685 	fhandle_t *fhp;
2686 
2687 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
2688 	/*
2689 	 * Check for the special case of the nfsv4root_fh.
2690 	 */
2691 	mp = vfs_busyfs(&fhp->fh_fsid);
2692 	if (mpp != NULL)
2693 		*mpp = mp;
2694 	if (mp == NULL) {
2695 		*vpp = NULL;
2696 		nd->nd_repstat = ESTALE;
2697 		goto out;
2698 	}
2699 
2700 	if (startwrite)
2701 		vn_start_write(NULL, mpp, V_WAIT);
2702 
2703 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2704 	    &credanon);
2705 	vfs_unbusy(mp);
2706 
2707 	/*
2708 	 * For NFSv4 without a pseudo root fs, unexported file handles
2709 	 * can be returned, so that Lookup works everywhere.
2710 	 */
2711 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2712 	    !(nd->nd_flag & ND_NFSV4)) {
2713 		vput(*vpp);
2714 		nd->nd_repstat = EACCES;
2715 	}
2716 
2717 	/*
2718 	 * Personally, I've never seen any point in requiring a
2719 	 * reserved port#, since only in the rare case where the
2720 	 * clients are all boxes with secure system priviledges,
2721 	 * does it provide any enhanced security, but... some people
2722 	 * believe it to be useful and keep putting this code back in.
2723 	 * (There is also some "security checker" out there that
2724 	 *  complains if the nfs server doesn't enforce this.)
2725 	 * However, note the following:
2726 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
2727 	 *	required.
2728 	 * RFC2623 recommends that, if a reserved port# is checked for,
2729 	 *	that there be a way to turn that off--> ifdef'd.
2730 	 */
2731 #ifdef NFS_REQRSVPORT
2732 	if (!nd->nd_repstat) {
2733 		struct sockaddr_in *saddr;
2734 		struct sockaddr_in6 *saddr6;
2735 
2736 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2737 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2738 		if (!(nd->nd_flag & ND_NFSV4) &&
2739 		    ((saddr->sin_family == AF_INET &&
2740 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2741 		     (saddr6->sin6_family == AF_INET6 &&
2742 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2743 			vput(*vpp);
2744 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2745 		}
2746 	}
2747 #endif	/* NFS_REQRSVPORT */
2748 
2749 	/*
2750 	 * Check/setup credentials.
2751 	 */
2752 	if (!nd->nd_repstat) {
2753 		nd->nd_saveduid = nd->nd_cred->cr_uid;
2754 		nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2755 		if (nd->nd_repstat)
2756 			vput(*vpp);
2757 	}
2758 	if (credanon != NULL)
2759 		crfree(credanon);
2760 	if (nd->nd_repstat) {
2761 		if (startwrite)
2762 			vn_finished_write(mp);
2763 		*vpp = NULL;
2764 		if (mpp != NULL)
2765 			*mpp = NULL;
2766 	}
2767 
2768 out:
2769 	NFSEXITCODE2(0, nd);
2770 }
2771 
2772 /*
2773  * glue for fp.
2774  */
2775 int
2776 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2777 {
2778 	struct filedesc *fdp;
2779 	struct file *fp;
2780 	int error = 0;
2781 
2782 	fdp = p->td_proc->p_fd;
2783 	if (fd >= fdp->fd_nfiles ||
2784 	    (fp = fdp->fd_ofiles[fd]) == NULL) {
2785 		error = EBADF;
2786 		goto out;
2787 	}
2788 	*fpp = fp;
2789 
2790 out:
2791 	NFSEXITCODE(error);
2792 	return (error);
2793 }
2794 
2795 /*
2796  * Called from nfssvc() to update the exports list. Just call
2797  * vfs_export(). This has to be done, since the v4 root fake fs isn't
2798  * in the mount list.
2799  */
2800 int
2801 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2802 {
2803 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2804 	int error = 0;
2805 	struct nameidata nd;
2806 	fhandle_t fh;
2807 
2808 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2809 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2810 		nfs_rootfhset = 0;
2811 	else if (error == 0) {
2812 		if (nfsexargp->fspec == NULL) {
2813 			error = EPERM;
2814 			goto out;
2815 		}
2816 		/*
2817 		 * If fspec != NULL, this is the v4root path.
2818 		 */
2819 		NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_USERSPACE,
2820 		    nfsexargp->fspec, p);
2821 		if ((error = namei(&nd)) != 0)
2822 			goto out;
2823 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
2824 		vrele(nd.ni_vp);
2825 		if (!error) {
2826 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2827 			NFSBCOPY((caddr_t)&fh,
2828 			    nfs_rootfh.nfsrvfh_data,
2829 			    sizeof (fhandle_t));
2830 			nfs_rootfhset = 1;
2831 		}
2832 	}
2833 
2834 out:
2835 	NFSEXITCODE(error);
2836 	return (error);
2837 }
2838 
2839 /*
2840  * Get the tcp socket sequence numbers we need.
2841  * (Maybe this should be moved to the tcp sources?)
2842  */
2843 int
2844 nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap)
2845 {
2846 	struct inpcb *inp;
2847 	struct tcpcb *tp;
2848 	int error = 0;
2849 
2850 	inp = sotoinpcb(so);
2851 	KASSERT(inp != NULL, ("nfsrv_getsocksndseq: inp == NULL"));
2852 	INP_RLOCK(inp);
2853 	if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2854 		INP_RUNLOCK(inp);
2855 		error = EPIPE;
2856 		goto out;
2857 	}
2858 	tp = intotcpcb(inp);
2859 	if (tp->t_state != TCPS_ESTABLISHED) {
2860 		INP_RUNLOCK(inp);
2861 		error = EPIPE;
2862 		goto out;
2863 	}
2864 	*maxp = tp->snd_max;
2865 	*unap = tp->snd_una;
2866 	INP_RUNLOCK(inp);
2867 
2868 out:
2869 	NFSEXITCODE(error);
2870 	return (error);
2871 }
2872 
2873 /*
2874  * This function needs to test to see if the system is near its limit
2875  * for memory allocation via malloc() or mget() and return True iff
2876  * either of these resources are near their limit.
2877  * XXX (For now, this is just a stub.)
2878  */
2879 int nfsrv_testmalloclimit = 0;
2880 int
2881 nfsrv_mallocmget_limit(void)
2882 {
2883 	static int printmesg = 0;
2884 	static int testval = 1;
2885 
2886 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2887 		if ((printmesg++ % 100) == 0)
2888 			printf("nfsd: malloc/mget near limit\n");
2889 		return (1);
2890 	}
2891 	return (0);
2892 }
2893 
2894 /*
2895  * BSD specific initialization of a mount point.
2896  */
2897 void
2898 nfsd_mntinit(void)
2899 {
2900 	static int inited = 0;
2901 
2902 	if (inited)
2903 		return;
2904 	inited = 1;
2905 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2906 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2907 	nfsv4root_mnt.mnt_export = NULL;
2908 	TAILQ_INIT(&nfsv4root_opt);
2909 	TAILQ_INIT(&nfsv4root_newopt);
2910 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2911 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2912 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
2913 }
2914 
2915 /*
2916  * Get a vnode for a file handle, without checking exports, etc.
2917  */
2918 struct vnode *
2919 nfsvno_getvp(fhandle_t *fhp)
2920 {
2921 	struct mount *mp;
2922 	struct vnode *vp;
2923 	int error;
2924 
2925 	mp = vfs_busyfs(&fhp->fh_fsid);
2926 	if (mp == NULL)
2927 		return (NULL);
2928 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2929 	vfs_unbusy(mp);
2930 	if (error)
2931 		return (NULL);
2932 	return (vp);
2933 }
2934 
2935 /*
2936  * Do a local VOP_ADVLOCK().
2937  */
2938 int
2939 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2940     u_int64_t end, struct thread *td)
2941 {
2942 	int error = 0;
2943 	struct flock fl;
2944 	u_int64_t tlen;
2945 
2946 	if (nfsrv_dolocallocks == 0)
2947 		goto out;
2948 
2949 	/* Check for VI_DOOMED here, so that VOP_ADVLOCK() isn't performed. */
2950 	if ((vp->v_iflag & VI_DOOMED) != 0) {
2951 		error = EPERM;
2952 		goto out;
2953 	}
2954 
2955 	fl.l_whence = SEEK_SET;
2956 	fl.l_type = ftype;
2957 	fl.l_start = (off_t)first;
2958 	if (end == NFS64BITSSET) {
2959 		fl.l_len = 0;
2960 	} else {
2961 		tlen = end - first;
2962 		fl.l_len = (off_t)tlen;
2963 	}
2964 	/*
2965 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2966 	 * values for all calls, so that all locks will be held by the
2967 	 * nfsd server. (The nfsd server handles conflicts between the
2968 	 * various clients.)
2969 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2970 	 * bytes, so it can't be put in l_sysid.
2971 	 */
2972 	if (nfsv4_sysid == 0)
2973 		nfsv4_sysid = nlm_acquire_next_sysid();
2974 	fl.l_pid = (pid_t)0;
2975 	fl.l_sysid = (int)nfsv4_sysid;
2976 
2977 	NFSVOPUNLOCK(vp, 0);
2978 	if (ftype == F_UNLCK)
2979 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
2980 		    (F_POSIX | F_REMOTE));
2981 	else
2982 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
2983 		    (F_POSIX | F_REMOTE));
2984 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2985 
2986 out:
2987 	NFSEXITCODE(error);
2988 	return (error);
2989 }
2990 
2991 /*
2992  * Check the nfsv4 root exports.
2993  */
2994 int
2995 nfsvno_v4rootexport(struct nfsrv_descript *nd)
2996 {
2997 	struct ucred *credanon;
2998 	int exflags, error = 0, numsecflavor, *secflavors, i;
2999 
3000 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3001 	    &credanon, &numsecflavor, &secflavors);
3002 	if (error) {
3003 		error = NFSERR_PROGUNAVAIL;
3004 		goto out;
3005 	}
3006 	if (credanon != NULL)
3007 		crfree(credanon);
3008 	for (i = 0; i < numsecflavor; i++) {
3009 		if (secflavors[i] == AUTH_SYS)
3010 			nd->nd_flag |= ND_EXAUTHSYS;
3011 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3012 			nd->nd_flag |= ND_EXGSS;
3013 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3014 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3015 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3016 			nd->nd_flag |= ND_EXGSSPRIVACY;
3017 	}
3018 
3019 out:
3020 	NFSEXITCODE(error);
3021 	return (error);
3022 }
3023 
3024 /*
3025  * Nfs server psuedo system call for the nfsd's
3026  */
3027 /*
3028  * MPSAFE
3029  */
3030 static int
3031 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3032 {
3033 	struct file *fp;
3034 	struct nfsd_addsock_args sockarg;
3035 	struct nfsd_nfsd_args nfsdarg;
3036 	int error;
3037 
3038 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3039 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3040 		if (error)
3041 			goto out;
3042 		/*
3043 		 * Since we don't know what rights might be required,
3044 		 * pretend that we need them all. It is better to be too
3045 		 * careful than too reckless.
3046 		 */
3047 		if ((error = fget(td, sockarg.sock, CAP_SOCK_ALL, &fp)) != 0)
3048 			goto out;
3049 		if (fp->f_type != DTYPE_SOCKET) {
3050 			fdrop(fp, td);
3051 			error = EPERM;
3052 			goto out;
3053 		}
3054 		error = nfsrvd_addsock(fp);
3055 		fdrop(fp, td);
3056 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3057 		if (uap->argp == NULL) {
3058 			error = EINVAL;
3059 			goto out;
3060 		}
3061 		error = copyin(uap->argp, (caddr_t)&nfsdarg,
3062 		    sizeof (nfsdarg));
3063 		if (error)
3064 			goto out;
3065 		error = nfsrvd_nfsd(td, &nfsdarg);
3066 	} else {
3067 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3068 	}
3069 
3070 out:
3071 	NFSEXITCODE(error);
3072 	return (error);
3073 }
3074 
3075 static int
3076 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3077 {
3078 	struct nfsex_args export;
3079 	struct file *fp = NULL;
3080 	int stablefd, len;
3081 	struct nfsd_clid adminrevoke;
3082 	struct nfsd_dumplist dumplist;
3083 	struct nfsd_dumpclients *dumpclients;
3084 	struct nfsd_dumplocklist dumplocklist;
3085 	struct nfsd_dumplocks *dumplocks;
3086 	struct nameidata nd;
3087 	vnode_t vp;
3088 	int error = EINVAL;
3089 	struct proc *procp;
3090 
3091 	if (uap->flag & NFSSVC_PUBLICFH) {
3092 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3093 		    sizeof (fhandle_t));
3094 		error = copyin(uap->argp,
3095 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3096 		if (!error)
3097 			nfs_pubfhset = 1;
3098 	} else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3099 		error = copyin(uap->argp,(caddr_t)&export,
3100 		    sizeof (struct nfsex_args));
3101 		if (!error)
3102 			error = nfsrv_v4rootexport(&export, cred, p);
3103 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3104 		nfs_pubfhset = 0;
3105 		error = 0;
3106 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3107 		error = copyin(uap->argp, (caddr_t)&stablefd,
3108 		    sizeof (int));
3109 		if (!error)
3110 			error = fp_getfvp(p, stablefd, &fp, &vp);
3111 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3112 			error = EBADF;
3113 		if (!error && newnfs_numnfsd != 0)
3114 			error = EPERM;
3115 		if (!error) {
3116 			nfsrv_stablefirst.nsf_fp = fp;
3117 			nfsrv_setupstable(p);
3118 		}
3119 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3120 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3121 		    sizeof (struct nfsd_clid));
3122 		if (!error)
3123 			error = nfsrv_adminrevoke(&adminrevoke, p);
3124 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3125 		error = copyin(uap->argp, (caddr_t)&dumplist,
3126 		    sizeof (struct nfsd_dumplist));
3127 		if (!error && (dumplist.ndl_size < 1 ||
3128 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3129 			error = EPERM;
3130 		if (!error) {
3131 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3132 		    dumpclients = (struct nfsd_dumpclients *)malloc(len,
3133 			M_TEMP, M_WAITOK);
3134 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3135 		    error = copyout(dumpclients,
3136 			CAST_USER_ADDR_T(dumplist.ndl_list), len);
3137 		    free((caddr_t)dumpclients, M_TEMP);
3138 		}
3139 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3140 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3141 		    sizeof (struct nfsd_dumplocklist));
3142 		if (!error && (dumplocklist.ndllck_size < 1 ||
3143 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3144 			error = EPERM;
3145 		if (!error)
3146 			error = nfsrv_lookupfilename(&nd,
3147 				dumplocklist.ndllck_fname, p);
3148 		if (!error) {
3149 			len = sizeof (struct nfsd_dumplocks) *
3150 				dumplocklist.ndllck_size;
3151 			dumplocks = (struct nfsd_dumplocks *)malloc(len,
3152 				M_TEMP, M_WAITOK);
3153 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3154 			    dumplocklist.ndllck_size, p);
3155 			vput(nd.ni_vp);
3156 			error = copyout(dumplocks,
3157 			    CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3158 			free((caddr_t)dumplocks, M_TEMP);
3159 		}
3160 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3161 		procp = p->td_proc;
3162 		PROC_LOCK(procp);
3163 		nfsd_master_pid = procp->p_pid;
3164 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3165 		nfsd_master_start = procp->p_stats->p_start;
3166 		nfsd_master_proc = procp;
3167 		PROC_UNLOCK(procp);
3168 	}
3169 
3170 	NFSEXITCODE(error);
3171 	return (error);
3172 }
3173 
3174 /*
3175  * Check exports.
3176  * Returns 0 if ok, 1 otherwise.
3177  */
3178 int
3179 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3180 {
3181 	int i;
3182 
3183 	/*
3184 	 * This seems odd, but allow the case where the security flavor
3185 	 * list is empty. This happens when NFSv4 is traversing non-exported
3186 	 * file systems. Exported file systems should always have a non-empty
3187 	 * security flavor list.
3188 	 */
3189 	if (exp->nes_numsecflavor == 0)
3190 		return (0);
3191 
3192 	for (i = 0; i < exp->nes_numsecflavor; i++) {
3193 		/*
3194 		 * The tests for privacy and integrity must be first,
3195 		 * since ND_GSS is set for everything but AUTH_SYS.
3196 		 */
3197 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3198 		    (nd->nd_flag & ND_GSSPRIVACY))
3199 			return (0);
3200 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3201 		    (nd->nd_flag & ND_GSSINTEGRITY))
3202 			return (0);
3203 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3204 		    (nd->nd_flag & ND_GSS))
3205 			return (0);
3206 		if (exp->nes_secflavors[i] == AUTH_SYS &&
3207 		    (nd->nd_flag & ND_GSS) == 0)
3208 			return (0);
3209 	}
3210 	return (1);
3211 }
3212 
3213 /*
3214  * Calculate a hash value for the fid in a file handle.
3215  */
3216 uint32_t
3217 nfsrv_hashfh(fhandle_t *fhp)
3218 {
3219 	uint32_t hashval;
3220 
3221 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3222 	return (hashval);
3223 }
3224 
3225 /*
3226  * Signal the userland master nfsd to backup the stable restart file.
3227  */
3228 void
3229 nfsrv_backupstable(void)
3230 {
3231 	struct proc *procp;
3232 
3233 	if (nfsd_master_proc != NULL) {
3234 		procp = pfind(nfsd_master_pid);
3235 		/* Try to make sure it is the correct process. */
3236 		if (procp == nfsd_master_proc &&
3237 		    procp->p_stats->p_start.tv_sec ==
3238 		    nfsd_master_start.tv_sec &&
3239 		    procp->p_stats->p_start.tv_usec ==
3240 		    nfsd_master_start.tv_usec &&
3241 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
3242 			kern_psignal(procp, SIGUSR2);
3243 		else
3244 			nfsd_master_proc = NULL;
3245 
3246 		if (procp != NULL)
3247 			PROC_UNLOCK(procp);
3248 	}
3249 }
3250 
3251 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3252 
3253 /*
3254  * Called once to initialize data structures...
3255  */
3256 static int
3257 nfsd_modevent(module_t mod, int type, void *data)
3258 {
3259 	int error = 0;
3260 	static int loaded = 0;
3261 
3262 	switch (type) {
3263 	case MOD_LOAD:
3264 		if (loaded)
3265 			goto out;
3266 		newnfs_portinit();
3267 		mtx_init(&nfs_cache_mutex, "nfs_cache_mutex", NULL, MTX_DEF);
3268 		mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3269 		mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3270 		    MTX_DEF);
3271 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3272 		nfsrvd_initcache();
3273 		nfsd_init();
3274 		NFSD_LOCK();
3275 		nfsrvd_init(0);
3276 		NFSD_UNLOCK();
3277 		nfsd_mntinit();
3278 #ifdef VV_DISABLEDELEG
3279 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3280 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3281 #endif
3282 		nfsd_call_servertimer = nfsrv_servertimer;
3283 		nfsd_call_nfsd = nfssvc_nfsd;
3284 		loaded = 1;
3285 		break;
3286 
3287 	case MOD_UNLOAD:
3288 		if (newnfs_numnfsd != 0) {
3289 			error = EBUSY;
3290 			break;
3291 		}
3292 
3293 #ifdef VV_DISABLEDELEG
3294 		vn_deleg_ops.vndeleg_recall = NULL;
3295 		vn_deleg_ops.vndeleg_disable = NULL;
3296 #endif
3297 		nfsd_call_servertimer = NULL;
3298 		nfsd_call_nfsd = NULL;
3299 
3300 		/* Clean out all NFSv4 state. */
3301 		nfsrv_throwawayallstate(curthread);
3302 
3303 		/* Clean the NFS server reply cache */
3304 		nfsrvd_cleancache();
3305 
3306 		/* Free up the krpc server pool. */
3307 		if (nfsrvd_pool != NULL)
3308 			svcpool_destroy(nfsrvd_pool);
3309 
3310 		/* and get rid of the locks */
3311 		mtx_destroy(&nfs_cache_mutex);
3312 		mtx_destroy(&nfs_v4root_mutex);
3313 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3314 		lockdestroy(&nfsv4root_mnt.mnt_explock);
3315 		loaded = 0;
3316 		break;
3317 	default:
3318 		error = EOPNOTSUPP;
3319 		break;
3320 	}
3321 
3322 out:
3323 	NFSEXITCODE(error);
3324 	return (error);
3325 }
3326 static moduledata_t nfsd_mod = {
3327 	"nfsd",
3328 	nfsd_modevent,
3329 	NULL,
3330 };
3331 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3332 
3333 /* So that loader and kldload(2) can find us, wherever we are.. */
3334 MODULE_VERSION(nfsd, 1);
3335 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3336 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3337 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3338 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3339 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3340 
3341