xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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  *	from nfs_vnops.c	8.16 (Berkeley) 5/27/95
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 /*
39  * vnode op calls for Sun NFS version 2, 3 and 4
40  */
41 
42 #include "opt_inet.h"
43 
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/resourcevar.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/jail.h>
53 #include <sys/malloc.h>
54 #include <sys/mbuf.h>
55 #include <sys/namei.h>
56 #include <sys/socket.h>
57 #include <sys/vnode.h>
58 #include <sys/dirent.h>
59 #include <sys/fcntl.h>
60 #include <sys/lockf.h>
61 #include <sys/stat.h>
62 #include <sys/sysctl.h>
63 #include <sys/signalvar.h>
64 
65 #include <vm/vm.h>
66 #include <vm/vm_object.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_object.h>
69 
70 
71 #include <fs/nfs/nfsport.h>
72 #include <fs/nfsclient/nfsnode.h>
73 #include <fs/nfsclient/nfsmount.h>
74 #include <fs/nfsclient/nfs.h>
75 #include <fs/nfsclient/nfs_lock.h>
76 
77 #include <net/if.h>
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80 
81 /* Defs */
82 #define	TRUE	1
83 #define	FALSE	0
84 
85 extern struct nfsstats newnfsstats;
86 MALLOC_DECLARE(M_NEWNFSREQ);
87 vop_advlock_t	*ncl_advlock_p = ncl_dolock;
88 
89 /*
90  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
91  * calls are not in getblk() and brelse() so that they would not be necessary
92  * here.
93  */
94 #ifndef B_VMIO
95 #define	vfs_busy_pages(bp, f)
96 #endif
97 
98 static vop_read_t	nfsfifo_read;
99 static vop_write_t	nfsfifo_write;
100 static vop_close_t	nfsfifo_close;
101 static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
102 		    struct thread *);
103 static vop_lookup_t	nfs_lookup;
104 static vop_create_t	nfs_create;
105 static vop_mknod_t	nfs_mknod;
106 static vop_open_t	nfs_open;
107 static vop_close_t	nfs_close;
108 static vop_access_t	nfs_access;
109 static vop_getattr_t	nfs_getattr;
110 static vop_setattr_t	nfs_setattr;
111 static vop_read_t	nfs_read;
112 static vop_fsync_t	nfs_fsync;
113 static vop_remove_t	nfs_remove;
114 static vop_link_t	nfs_link;
115 static vop_rename_t	nfs_rename;
116 static vop_mkdir_t	nfs_mkdir;
117 static vop_rmdir_t	nfs_rmdir;
118 static vop_symlink_t	nfs_symlink;
119 static vop_readdir_t	nfs_readdir;
120 static vop_strategy_t	nfs_strategy;
121 static vop_lock1_t	nfs_lock1;
122 static	int	nfs_lookitup(struct vnode *, char *, int,
123 		    struct ucred *, struct thread *, struct nfsnode **);
124 static	int	nfs_sillyrename(struct vnode *, struct vnode *,
125 		    struct componentname *);
126 static vop_access_t	nfsspec_access;
127 static vop_readlink_t	nfs_readlink;
128 static vop_print_t	nfs_print;
129 static vop_advlock_t	nfs_advlock;
130 static vop_advlockasync_t nfs_advlockasync;
131 #ifdef NFS4_ACL_EXTATTR_NAME
132 static vop_getacl_t nfs_getacl;
133 static vop_setacl_t nfs_setacl;
134 #endif
135 
136 /*
137  * Global vfs data structures for nfs
138  */
139 struct vop_vector newnfs_vnodeops = {
140 	.vop_default =		&default_vnodeops,
141 	.vop_access =		nfs_access,
142 	.vop_advlock =		nfs_advlock,
143 	.vop_advlockasync =	nfs_advlockasync,
144 	.vop_close =		nfs_close,
145 	.vop_create =		nfs_create,
146 	.vop_fsync =		nfs_fsync,
147 	.vop_getattr =		nfs_getattr,
148 	.vop_getpages =		ncl_getpages,
149 	.vop_putpages =		ncl_putpages,
150 	.vop_inactive =		ncl_inactive,
151 	.vop_link =		nfs_link,
152 	.vop_lock1 = 		nfs_lock1,
153 	.vop_lookup =		nfs_lookup,
154 	.vop_mkdir =		nfs_mkdir,
155 	.vop_mknod =		nfs_mknod,
156 	.vop_open =		nfs_open,
157 	.vop_print =		nfs_print,
158 	.vop_read =		nfs_read,
159 	.vop_readdir =		nfs_readdir,
160 	.vop_readlink =		nfs_readlink,
161 	.vop_reclaim =		ncl_reclaim,
162 	.vop_remove =		nfs_remove,
163 	.vop_rename =		nfs_rename,
164 	.vop_rmdir =		nfs_rmdir,
165 	.vop_setattr =		nfs_setattr,
166 	.vop_strategy =		nfs_strategy,
167 	.vop_symlink =		nfs_symlink,
168 	.vop_write =		ncl_write,
169 #ifdef NFS4_ACL_EXTATTR_NAME
170 	.vop_getacl =		nfs_getacl,
171 	.vop_setacl =		nfs_setacl,
172 #endif
173 };
174 
175 struct vop_vector newnfs_fifoops = {
176 	.vop_default =		&fifo_specops,
177 	.vop_access =		nfsspec_access,
178 	.vop_close =		nfsfifo_close,
179 	.vop_fsync =		nfs_fsync,
180 	.vop_getattr =		nfs_getattr,
181 	.vop_inactive =		ncl_inactive,
182 	.vop_print =		nfs_print,
183 	.vop_read =		nfsfifo_read,
184 	.vop_reclaim =		ncl_reclaim,
185 	.vop_setattr =		nfs_setattr,
186 	.vop_write =		nfsfifo_write,
187 };
188 
189 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
190     struct componentname *cnp, struct vattr *vap);
191 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
192     int namelen, struct ucred *cred, struct thread *td);
193 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
194     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
195     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
196 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
197     struct componentname *scnp, struct sillyrename *sp);
198 
199 /*
200  * Global variables
201  */
202 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
203 
204 SYSCTL_DECL(_vfs_newnfs);
205 
206 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
207 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
208 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
209 
210 static int	nfs_prime_access_cache = 0;
211 SYSCTL_INT(_vfs_newnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
212 	   &nfs_prime_access_cache, 0,
213 	   "Prime NFS ACCESS cache when fetching attributes");
214 
215 static int	newnfs_commit_on_close = 0;
216 SYSCTL_INT(_vfs_newnfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
217     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
218 
219 static int	nfs_clean_pages_on_close = 1;
220 SYSCTL_INT(_vfs_newnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
221 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
222 
223 int newnfs_directio_enable = 0;
224 SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_enable, CTLFLAG_RW,
225 	   &newnfs_directio_enable, 0, "Enable NFS directio");
226 
227 static int newnfs_neglookup_enable = 1;
228 SYSCTL_INT(_vfs_newnfs, OID_AUTO, neglookup_enable, CTLFLAG_RW,
229     &newnfs_neglookup_enable, 0, "Enable NFS negative lookup caching");
230 
231 /*
232  * This sysctl allows other processes to mmap a file that has been opened
233  * O_DIRECT by a process.  In general, having processes mmap the file while
234  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
235  * this by default to prevent DoS attacks - to prevent a malicious user from
236  * opening up files O_DIRECT preventing other users from mmap'ing these
237  * files.  "Protected" environments where stricter consistency guarantees are
238  * required can disable this knob.  The process that opened the file O_DIRECT
239  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
240  * meaningful.
241  */
242 int newnfs_directio_allow_mmap = 1;
243 SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_allow_mmap, CTLFLAG_RW,
244 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
245 
246 #if 0
247 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
248 	   &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
249 
250 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
251 	   &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
252 #endif
253 
254 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
255 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
256 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
257 
258 /*
259  * SMP Locking Note :
260  * The list of locks after the description of the lock is the ordering
261  * of other locks acquired with the lock held.
262  * np->n_mtx : Protects the fields in the nfsnode.
263        VM Object Lock
264        VI_MTX (acquired indirectly)
265  * nmp->nm_mtx : Protects the fields in the nfsmount.
266        rep->r_mtx
267  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
268  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
269        nmp->nm_mtx
270        rep->r_mtx
271  * rep->r_mtx : Protects the fields in an nfsreq.
272  */
273 
274 static int
275 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
276     struct ucred *cred, u_int32_t *retmode)
277 {
278 	int error = 0, attrflag, i, lrupos;
279 	u_int32_t rmode;
280 	struct nfsnode *np = VTONFS(vp);
281 	struct nfsvattr nfsva;
282 
283 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
284 	    &rmode, NULL);
285 	if (attrflag)
286 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
287 	if (!error) {
288 		lrupos = 0;
289 		mtx_lock(&np->n_mtx);
290 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
291 			if (np->n_accesscache[i].uid == cred->cr_uid) {
292 				np->n_accesscache[i].mode = rmode;
293 				np->n_accesscache[i].stamp = time_second;
294 				break;
295 			}
296 			if (i > 0 && np->n_accesscache[i].stamp <
297 			    np->n_accesscache[lrupos].stamp)
298 				lrupos = i;
299 		}
300 		if (i == NFS_ACCESSCACHESIZE) {
301 			np->n_accesscache[lrupos].uid = cred->cr_uid;
302 			np->n_accesscache[lrupos].mode = rmode;
303 			np->n_accesscache[lrupos].stamp = time_second;
304 		}
305 		mtx_unlock(&np->n_mtx);
306 		if (retmode != NULL)
307 			*retmode = rmode;
308 	} else if (NFS_ISV4(vp)) {
309 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
310 	}
311 	return (error);
312 }
313 
314 /*
315  * nfs access vnode op.
316  * For nfs version 2, just return ok. File accesses may fail later.
317  * For nfs version 3, use the access rpc to check accessibility. If file modes
318  * are changed on the server, accesses might still fail later.
319  */
320 static int
321 nfs_access(struct vop_access_args *ap)
322 {
323 	struct vnode *vp = ap->a_vp;
324 	int error = 0, i, gotahit;
325 	u_int32_t mode, wmode, rmode;
326 	int v34 = NFS_ISV34(vp);
327 	struct nfsnode *np = VTONFS(vp);
328 
329 	/*
330 	 * Disallow write attempts on filesystems mounted read-only;
331 	 * unless the file is a socket, fifo, or a block or character
332 	 * device resident on the filesystem.
333 	 */
334 	if ((ap->a_accmode & (VWRITE | VAPPEND
335 #ifdef NFS4_ACL_EXTATTR_NAME
336 	    | VWRITE_NAMED_ATTRS | VDELETE_CHILD | VWRITE_ATTRIBUTES |
337 	    VDELETE | VWRITE_ACL | VWRITE_OWNER
338 #endif
339 	    )) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
340 		switch (vp->v_type) {
341 		case VREG:
342 		case VDIR:
343 		case VLNK:
344 			return (EROFS);
345 		default:
346 			break;
347 		}
348 	}
349 	/*
350 	 * For nfs v3 or v4, check to see if we have done this recently, and if
351 	 * so return our cached result instead of making an ACCESS call.
352 	 * If not, do an access rpc, otherwise you are stuck emulating
353 	 * ufs_access() locally using the vattr. This may not be correct,
354 	 * since the server may apply other access criteria such as
355 	 * client uid-->server uid mapping that we do not know about.
356 	 */
357 	if (v34) {
358 		if (ap->a_accmode & VREAD)
359 			mode = NFSACCESS_READ;
360 		else
361 			mode = 0;
362 		if (vp->v_type != VDIR) {
363 			if (ap->a_accmode & VWRITE)
364 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
365 			if (ap->a_accmode & VAPPEND)
366 				mode |= NFSACCESS_EXTEND;
367 			if (ap->a_accmode & VEXEC)
368 				mode |= NFSACCESS_EXECUTE;
369 #ifdef NFS4_ACL_EXTATTR_NAME
370 			if (ap->a_accmode & VDELETE)
371 				mode |= NFSACCESS_DELETE;
372 #endif
373 		} else {
374 			if (ap->a_accmode & VWRITE)
375 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
376 			if (ap->a_accmode & VAPPEND)
377 				mode |= NFSACCESS_EXTEND;
378 			if (ap->a_accmode & VEXEC)
379 				mode |= NFSACCESS_LOOKUP;
380 #ifdef NFS4_ACL_EXTATTR_NAME
381 			if (ap->a_accmode & VDELETE)
382 				mode |= NFSACCESS_DELETE;
383 			if (ap->a_accmode & VDELETE_CHILD)
384 				mode |= NFSACCESS_MODIFY;
385 #endif
386 		}
387 		/* XXX safety belt, only make blanket request if caching */
388 		if (nfsaccess_cache_timeout > 0) {
389 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
390 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
391 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
392 		} else {
393 			wmode = mode;
394 		}
395 
396 		/*
397 		 * Does our cached result allow us to give a definite yes to
398 		 * this request?
399 		 */
400 		gotahit = 0;
401 		mtx_lock(&np->n_mtx);
402 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
403 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
404 			    if (time_second < (np->n_accesscache[i].stamp
405 				+ nfsaccess_cache_timeout) &&
406 				(np->n_accesscache[i].mode & mode) == mode) {
407 				NFSINCRGLOBAL(newnfsstats.accesscache_hits);
408 				gotahit = 1;
409 			    }
410 			    break;
411 			}
412 		}
413 		mtx_unlock(&np->n_mtx);
414 		if (gotahit == 0) {
415 			/*
416 			 * Either a no, or a don't know.  Go to the wire.
417 			 */
418 			NFSINCRGLOBAL(newnfsstats.accesscache_misses);
419 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
420 			    ap->a_cred, &rmode);
421 			if (!error &&
422 			    (rmode & mode) != mode)
423 				error = EACCES;
424 		}
425 		return (error);
426 	} else {
427 		if ((error = nfsspec_access(ap)) != 0) {
428 			return (error);
429 		}
430 		/*
431 		 * Attempt to prevent a mapped root from accessing a file
432 		 * which it shouldn't.  We try to read a byte from the file
433 		 * if the user is root and the file is not zero length.
434 		 * After calling nfsspec_access, we should have the correct
435 		 * file size cached.
436 		 */
437 		mtx_lock(&np->n_mtx);
438 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
439 		    && VTONFS(vp)->n_size > 0) {
440 			struct iovec aiov;
441 			struct uio auio;
442 			char buf[1];
443 
444 			mtx_unlock(&np->n_mtx);
445 			aiov.iov_base = buf;
446 			aiov.iov_len = 1;
447 			auio.uio_iov = &aiov;
448 			auio.uio_iovcnt = 1;
449 			auio.uio_offset = 0;
450 			auio.uio_resid = 1;
451 			auio.uio_segflg = UIO_SYSSPACE;
452 			auio.uio_rw = UIO_READ;
453 			auio.uio_td = ap->a_td;
454 
455 			if (vp->v_type == VREG)
456 				error = ncl_readrpc(vp, &auio, ap->a_cred);
457 			else if (vp->v_type == VDIR) {
458 				char* bp;
459 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
460 				aiov.iov_base = bp;
461 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
462 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
463 				    ap->a_td);
464 				free(bp, M_TEMP);
465 			} else if (vp->v_type == VLNK)
466 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
467 			else
468 				error = EACCES;
469 		} else
470 			mtx_unlock(&np->n_mtx);
471 		return (error);
472 	}
473 }
474 
475 
476 /*
477  * nfs open vnode op
478  * Check to see if the type is ok
479  * and that deletion is not in progress.
480  * For paged in text files, you will need to flush the page cache
481  * if consistency is lost.
482  */
483 /* ARGSUSED */
484 static int
485 nfs_open(struct vop_open_args *ap)
486 {
487 	struct vnode *vp = ap->a_vp;
488 	struct nfsnode *np = VTONFS(vp);
489 	struct vattr vattr;
490 	int error;
491 	int fmode = ap->a_mode;
492 
493 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
494 		return (EOPNOTSUPP);
495 
496 	/*
497 	 * For NFSv4, we need to do the Open Op before cache validation,
498 	 * so that we conform to RFC3530 Sec. 9.3.1.
499 	 */
500 	if (NFS_ISV4(vp)) {
501 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
502 		if (error) {
503 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
504 			    (gid_t)0);
505 			return (error);
506 		}
507 	}
508 
509 	/*
510 	 * Now, if this Open will be doing reading, re-validate/flush the
511 	 * cache, so that Close/Open coherency is maintained.
512 	 */
513 	if ((fmode & FREAD) && (!NFS_ISV4(vp) || nfscl_mustflush(vp))) {
514 		mtx_lock(&np->n_mtx);
515 		if (np->n_flag & NMODIFIED) {
516 			mtx_unlock(&np->n_mtx);
517 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
518 			if (error == EINTR || error == EIO) {
519 				if (NFS_ISV4(vp))
520 					(void) nfsrpc_close(vp, 0, ap->a_td);
521 				return (error);
522 			}
523 			np->n_attrstamp = 0;
524 			if (vp->v_type == VDIR)
525 				np->n_direofoffset = 0;
526 			error = VOP_GETATTR(vp, &vattr, ap->a_cred);
527 			if (error) {
528 				if (NFS_ISV4(vp))
529 					(void) nfsrpc_close(vp, 0, ap->a_td);
530 				return (error);
531 			}
532 			mtx_lock(&np->n_mtx);
533 			np->n_mtime = vattr.va_mtime;
534 			if (NFS_ISV4(vp))
535 				np->n_change = vattr.va_filerev;
536 			mtx_unlock(&np->n_mtx);
537 		} else {
538 			struct thread *td = curthread;
539 
540 			if (np->n_ac_ts_syscalls != td->td_syscalls ||
541 			    np->n_ac_ts_tid != td->td_tid ||
542 			    td->td_proc == NULL ||
543 			    np->n_ac_ts_pid != td->td_proc->p_pid) {
544 				np->n_attrstamp = 0;
545 			}
546 			mtx_unlock(&np->n_mtx);
547 			error = VOP_GETATTR(vp, &vattr, ap->a_cred);
548 			if (error) {
549 				if (NFS_ISV4(vp))
550 					(void) nfsrpc_close(vp, 0, ap->a_td);
551 				return (error);
552 			}
553 			mtx_lock(&np->n_mtx);
554 			if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
555 			    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
556 				if (vp->v_type == VDIR)
557 					np->n_direofoffset = 0;
558 				mtx_unlock(&np->n_mtx);
559 				error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
560 				if (error == EINTR || error == EIO) {
561 					if (NFS_ISV4(vp))
562 						(void) nfsrpc_close(vp, 0,
563 						    ap->a_td);
564 					return (error);
565 				}
566 				mtx_lock(&np->n_mtx);
567 				np->n_mtime = vattr.va_mtime;
568 				if (NFS_ISV4(vp))
569 					np->n_change = vattr.va_filerev;
570 			}
571 			mtx_unlock(&np->n_mtx);
572 		}
573 	}
574 
575 	/*
576 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
577 	 */
578 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
579 		if (np->n_directio_opens == 0) {
580 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
581 			if (error) {
582 				if (NFS_ISV4(vp))
583 					(void) nfsrpc_close(vp, 0, ap->a_td);
584 				return (error);
585 			}
586 			mtx_lock(&np->n_mtx);
587 			np->n_flag |= NNONCACHE;
588 		} else {
589 			mtx_lock(&np->n_mtx);
590 		}
591 		np->n_directio_opens++;
592 		mtx_unlock(&np->n_mtx);
593 	}
594 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
595 	return (0);
596 }
597 
598 /*
599  * nfs close vnode op
600  * What an NFS client should do upon close after writing is a debatable issue.
601  * Most NFS clients push delayed writes to the server upon close, basically for
602  * two reasons:
603  * 1 - So that any write errors may be reported back to the client process
604  *     doing the close system call. By far the two most likely errors are
605  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
606  * 2 - To put a worst case upper bound on cache inconsistency between
607  *     multiple clients for the file.
608  * There is also a consistency problem for Version 2 of the protocol w.r.t.
609  * not being able to tell if other clients are writing a file concurrently,
610  * since there is no way of knowing if the changed modify time in the reply
611  * is only due to the write for this client.
612  * (NFS Version 3 provides weak cache consistency data in the reply that
613  *  should be sufficient to detect and handle this case.)
614  *
615  * The current code does the following:
616  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
617  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
618  *                     or commit them (this satisfies 1 and 2 except for the
619  *                     case where the server crashes after this close but
620  *                     before the commit RPC, which is felt to be "good
621  *                     enough". Changing the last argument to ncl_flush() to
622  *                     a 1 would force a commit operation, if it is felt a
623  *                     commit is necessary now.
624  * for NFS Version 4 - flush the dirty buffers and commit them, if
625  *		       nfscl_mustflush() says this is necessary.
626  *                     It is necessary if there is no write delegation held,
627  *                     in order to satisfy open/close coherency.
628  *                     If the file isn't cached on local stable storage,
629  *                     it may be necessary in order to detect "out of space"
630  *                     errors from the server, if the write delegation
631  *                     issued by the server doesn't allow the file to grow.
632  */
633 /* ARGSUSED */
634 static int
635 nfs_close(struct vop_close_args *ap)
636 {
637 	struct vnode *vp = ap->a_vp;
638 	struct nfsnode *np = VTONFS(vp);
639 	struct nfsvattr nfsva;
640 	struct ucred *cred;
641 	int error = 0, ret, localcred = 0;
642 	int fmode = ap->a_fflag;
643 
644 	if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
645 		return (0);
646 	/*
647 	 * During shutdown, a_cred isn't valid, so just use root.
648 	 */
649 	if (ap->a_cred == NOCRED) {
650 		cred = newnfs_getcred();
651 		localcred = 1;
652 	} else {
653 		cred = ap->a_cred;
654 	}
655 	if (vp->v_type == VREG) {
656 	    /*
657 	     * Examine and clean dirty pages, regardless of NMODIFIED.
658 	     * This closes a major hole in close-to-open consistency.
659 	     * We want to push out all dirty pages (and buffers) on
660 	     * close, regardless of whether they were dirtied by
661 	     * mmap'ed writes or via write().
662 	     */
663 	    if (nfs_clean_pages_on_close && vp->v_object) {
664 		VM_OBJECT_LOCK(vp->v_object);
665 		vm_object_page_clean(vp->v_object, 0, 0, 0);
666 		VM_OBJECT_UNLOCK(vp->v_object);
667 	    }
668 	    mtx_lock(&np->n_mtx);
669 	    if (np->n_flag & NMODIFIED) {
670 		mtx_unlock(&np->n_mtx);
671 		if (NFS_ISV3(vp)) {
672 		    /*
673 		     * Under NFSv3 we have dirty buffers to dispose of.  We
674 		     * must flush them to the NFS server.  We have the option
675 		     * of waiting all the way through the commit rpc or just
676 		     * waiting for the initial write.  The default is to only
677 		     * wait through the initial write so the data is in the
678 		     * server's cache, which is roughly similar to the state
679 		     * a standard disk subsystem leaves the file in on close().
680 		     *
681 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
682 		     * potential races with other processes, and certainly
683 		     * cannot clear it if we don't commit.
684 		     * These races occur when there is no longer the old
685 		     * traditional vnode locking implemented for Vnode Ops.
686 		     */
687 		    int cm = newnfs_commit_on_close ? 1 : 0;
688 		    error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm);
689 		    /* np->n_flag &= ~NMODIFIED; */
690 		} else if (NFS_ISV4(vp)) {
691 			if (nfscl_mustflush(vp)) {
692 				int cm = newnfs_commit_on_close ? 1 : 0;
693 				error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
694 				    cm);
695 				/*
696 				 * as above w.r.t races when clearing
697 				 * NMODIFIED.
698 				 * np->n_flag &= ~NMODIFIED;
699 				 */
700 			}
701 		} else
702 		    error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
703 		mtx_lock(&np->n_mtx);
704 	    }
705  	    /*
706  	     * Invalidate the attribute cache in all cases.
707  	     * An open is going to fetch fresh attrs any way, other procs
708  	     * on this node that have file open will be forced to do an
709  	     * otw attr fetch, but this is safe.
710 	     * --> A user found that their RPC count dropped by 20% when
711 	     *     this was commented out and I can't see any requirement
712 	     *     for it, so I've disabled it when negative lookups are
713 	     *     enabled. (What does this have to do with negative lookup
714 	     *     caching? Well nothing, except it was reported by the
715 	     *     same user that needed negative lookup caching and I wanted
716 	     *     there to be a way to disable it via sysctl to see if it
717 	     *     is the cause of some caching/coherency issue that might
718 	     *     crop up.)
719  	     */
720 	    if (newnfs_neglookup_enable == 0)
721 		    np->n_attrstamp = 0;
722 	    if (np->n_flag & NWRITEERR) {
723 		np->n_flag &= ~NWRITEERR;
724 		error = np->n_error;
725 	    }
726 	    mtx_unlock(&np->n_mtx);
727 	}
728 
729 	if (NFS_ISV4(vp)) {
730 		/*
731 		 * Get attributes so "change" is up to date.
732 		 */
733 		if (!error) {
734 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
735 			    NULL);
736 			if (!ret) {
737 				np->n_change = nfsva.na_filerev;
738 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
739 				    NULL, 0, 0);
740 			}
741 		}
742 
743 		/*
744 		 * and do the close.
745 		 */
746 		ret = nfsrpc_close(vp, 0, ap->a_td);
747 		if (!error && ret)
748 			error = ret;
749 		if (error)
750 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
751 			    (gid_t)0);
752 	}
753 	if (newnfs_directio_enable)
754 		KASSERT((np->n_directio_asyncwr == 0),
755 			("nfs_close: dirty unflushed (%d) directio buffers\n",
756 			 np->n_directio_asyncwr));
757 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
758 		mtx_lock(&np->n_mtx);
759 		KASSERT((np->n_directio_opens > 0),
760 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
761 		np->n_directio_opens--;
762 		if (np->n_directio_opens == 0)
763 			np->n_flag &= ~NNONCACHE;
764 		mtx_unlock(&np->n_mtx);
765 	}
766 	if (localcred)
767 		NFSFREECRED(cred);
768 	return (error);
769 }
770 
771 /*
772  * nfs getattr call from vfs.
773  */
774 static int
775 nfs_getattr(struct vop_getattr_args *ap)
776 {
777 	struct vnode *vp = ap->a_vp;
778 	struct thread *td = curthread;	/* XXX */
779 	struct nfsnode *np = VTONFS(vp);
780 	int error = 0;
781 	struct nfsvattr nfsva;
782 	struct vattr *vap = ap->a_vap;
783 	struct vattr vattr;
784 
785 	/*
786 	 * Update local times for special files.
787 	 */
788 	mtx_lock(&np->n_mtx);
789 	if (np->n_flag & (NACC | NUPD))
790 		np->n_flag |= NCHG;
791 	mtx_unlock(&np->n_mtx);
792 	/*
793 	 * First look in the cache.
794 	 */
795 	if (ncl_getattrcache(vp, &vattr) == 0) {
796 		vap->va_type = vattr.va_type;
797 		vap->va_mode = vattr.va_mode;
798 		vap->va_nlink = vattr.va_nlink;
799 		vap->va_uid = vattr.va_uid;
800 		vap->va_gid = vattr.va_gid;
801 		vap->va_fsid = vattr.va_fsid;
802 		vap->va_fileid = vattr.va_fileid;
803 		vap->va_size = vattr.va_size;
804 		vap->va_blocksize = vattr.va_blocksize;
805 		vap->va_atime = vattr.va_atime;
806 		vap->va_mtime = vattr.va_mtime;
807 		vap->va_ctime = vattr.va_ctime;
808 		vap->va_gen = vattr.va_gen;
809 		vap->va_flags = vattr.va_flags;
810 		vap->va_rdev = vattr.va_rdev;
811 		vap->va_bytes = vattr.va_bytes;
812 		vap->va_filerev = vattr.va_filerev;
813 		/*
814 		 * Get the local modify time for the case of a write
815 		 * delegation.
816 		 */
817 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
818 		return (0);
819 	}
820 
821 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
822 	    nfsaccess_cache_timeout > 0) {
823 		NFSINCRGLOBAL(newnfsstats.accesscache_misses);
824 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
825 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
826 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
827 			return (0);
828 		}
829 	}
830 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
831 	if (!error)
832 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
833 	if (!error) {
834 		/*
835 		 * Get the local modify time for the case of a write
836 		 * delegation.
837 		 */
838 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
839 	} else if (NFS_ISV4(vp)) {
840 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
841 	}
842 	return (error);
843 }
844 
845 /*
846  * nfs setattr call.
847  */
848 static int
849 nfs_setattr(struct vop_setattr_args *ap)
850 {
851 	struct vnode *vp = ap->a_vp;
852 	struct nfsnode *np = VTONFS(vp);
853 	struct thread *td = curthread;	/* XXX */
854 	struct vattr *vap = ap->a_vap;
855 	int error = 0;
856 	u_quad_t tsize;
857 
858 #ifndef nolint
859 	tsize = (u_quad_t)0;
860 #endif
861 
862 	/*
863 	 * Setting of flags and marking of atimes are not supported.
864 	 */
865 	if (vap->va_flags != VNOVAL)
866 		return (EOPNOTSUPP);
867 
868 	/*
869 	 * Disallow write attempts if the filesystem is mounted read-only.
870 	 */
871   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
872 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
873 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
874 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
875 		return (EROFS);
876 	if (vap->va_size != VNOVAL) {
877  		switch (vp->v_type) {
878  		case VDIR:
879  			return (EISDIR);
880  		case VCHR:
881  		case VBLK:
882  		case VSOCK:
883  		case VFIFO:
884 			if (vap->va_mtime.tv_sec == VNOVAL &&
885 			    vap->va_atime.tv_sec == VNOVAL &&
886 			    vap->va_mode == (mode_t)VNOVAL &&
887 			    vap->va_uid == (uid_t)VNOVAL &&
888 			    vap->va_gid == (gid_t)VNOVAL)
889 				return (0);
890  			vap->va_size = VNOVAL;
891  			break;
892  		default:
893 			/*
894 			 * Disallow write attempts if the filesystem is
895 			 * mounted read-only.
896 			 */
897 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
898 				return (EROFS);
899 			/*
900 			 *  We run vnode_pager_setsize() early (why?),
901 			 * we must set np->n_size now to avoid vinvalbuf
902 			 * V_SAVE races that might setsize a lower
903 			 * value.
904 			 */
905 			mtx_lock(&np->n_mtx);
906 			tsize = np->n_size;
907 			mtx_unlock(&np->n_mtx);
908 			error = ncl_meta_setsize(vp, ap->a_cred, td,
909 			    vap->va_size);
910 			mtx_lock(&np->n_mtx);
911  			if (np->n_flag & NMODIFIED) {
912 			    tsize = np->n_size;
913 			    mtx_unlock(&np->n_mtx);
914  			    if (vap->va_size == 0)
915  				error = ncl_vinvalbuf(vp, 0, td, 1);
916  			    else
917  				error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
918  			    if (error) {
919 				vnode_pager_setsize(vp, tsize);
920 				return (error);
921 			    }
922 			    /*
923 			     * Call nfscl_delegmodtime() to set the modify time
924 			     * locally, as required.
925 			     */
926 			    nfscl_delegmodtime(vp);
927  			} else
928 			    mtx_unlock(&np->n_mtx);
929 			/*
930 			 * np->n_size has already been set to vap->va_size
931 			 * in ncl_meta_setsize(). We must set it again since
932 			 * nfs_loadattrcache() could be called through
933 			 * ncl_meta_setsize() and could modify np->n_size.
934 			 */
935 			mtx_lock(&np->n_mtx);
936  			np->n_vattr.na_size = np->n_size = vap->va_size;
937 			mtx_unlock(&np->n_mtx);
938   		};
939   	} else {
940 		mtx_lock(&np->n_mtx);
941 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
942 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
943 			mtx_unlock(&np->n_mtx);
944 			if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
945 			    (error == EINTR || error == EIO))
946 				return (error);
947 		} else
948 			mtx_unlock(&np->n_mtx);
949 	}
950 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
951 	if (error && vap->va_size != VNOVAL) {
952 		mtx_lock(&np->n_mtx);
953 		np->n_size = np->n_vattr.na_size = tsize;
954 		vnode_pager_setsize(vp, tsize);
955 		mtx_unlock(&np->n_mtx);
956 	}
957 	return (error);
958 }
959 
960 /*
961  * Do an nfs setattr rpc.
962  */
963 static int
964 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
965     struct thread *td)
966 {
967 	struct nfsnode *np = VTONFS(vp);
968 	int error, ret, attrflag, i;
969 	struct nfsvattr nfsva;
970 
971 	if (NFS_ISV34(vp)) {
972 		mtx_lock(&np->n_mtx);
973 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
974 			np->n_accesscache[i].stamp = 0;
975 		np->n_flag |= NDELEGMOD;
976 		mtx_unlock(&np->n_mtx);
977 	}
978 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
979 	    NULL);
980 	if (attrflag) {
981 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
982 		if (ret && !error)
983 			error = ret;
984 	}
985 	if (error && NFS_ISV4(vp))
986 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
987 	return (error);
988 }
989 
990 /*
991  * nfs lookup call, one step at a time...
992  * First look in cache
993  * If not found, unlock the directory nfsnode and do the rpc
994  */
995 static int
996 nfs_lookup(struct vop_lookup_args *ap)
997 {
998 	struct componentname *cnp = ap->a_cnp;
999 	struct vnode *dvp = ap->a_dvp;
1000 	struct vnode **vpp = ap->a_vpp;
1001 	struct mount *mp = dvp->v_mount;
1002 	int flags = cnp->cn_flags;
1003 	struct vnode *newvp;
1004 	struct nfsmount *nmp;
1005 	struct nfsnode *np;
1006 	int error = 0, attrflag, dattrflag, ltype;
1007 	struct thread *td = cnp->cn_thread;
1008 	struct nfsfh *nfhp;
1009 	struct nfsvattr dnfsva, nfsva;
1010 
1011 	*vpp = NULLVP;
1012 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1013 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1014 		return (EROFS);
1015 	if (dvp->v_type != VDIR)
1016 		return (ENOTDIR);
1017 	nmp = VFSTONFS(mp);
1018 	np = VTONFS(dvp);
1019 
1020 	/* For NFSv4, wait until any remove is done. */
1021 	mtx_lock(&np->n_mtx);
1022 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1023 		np->n_flag |= NREMOVEWANT;
1024 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1025 	}
1026 	mtx_unlock(&np->n_mtx);
1027 
1028 	if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1029 		return (error);
1030 	if ((error = cache_lookup(dvp, vpp, cnp)) &&
1031 	    (error != ENOENT || newnfs_neglookup_enable != 0)) {
1032 		struct vattr vattr;
1033 
1034 		if (error == ENOENT) {
1035 			if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred) &&
1036 			    vattr.va_mtime.tv_sec == np->n_dmtime) {
1037 			     NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1038 				return (ENOENT);
1039 			}
1040 			cache_purge_negative(dvp);
1041 			np->n_dmtime = 0;
1042 		} else {
1043 			newvp = *vpp;
1044 			if (nfscl_nodeleg(newvp, 0) == 0 ||
1045 			    (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred) &&
1046 			     vattr.va_ctime.tv_sec==VTONFS(newvp)->n_ctime)) {
1047 			     NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1048 			     if (cnp->cn_nameiop != LOOKUP &&
1049 				 (flags & ISLASTCN))
1050 				     cnp->cn_flags |= SAVENAME;
1051 			     return (0);
1052 			}
1053 			cache_purge(newvp);
1054 			if (dvp != newvp)
1055 				vput(newvp);
1056 			else
1057 				vrele(newvp);
1058 			*vpp = NULLVP;
1059 		}
1060 	}
1061 	error = 0;
1062 	newvp = NULLVP;
1063 	NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1064 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1065 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1066 	    NULL);
1067 	if (dattrflag)
1068 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1069 	if (error) {
1070 		if (newnfs_neglookup_enable != 0 &&
1071 		    error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1072 		    cnp->cn_nameiop != CREATE) {
1073 			if (np->n_dmtime == 0)
1074 				np->n_dmtime = np->n_vattr.na_mtime.tv_sec;
1075 			cache_enter(dvp, NULL, cnp);
1076 		}
1077 		if (newvp != NULLVP) {
1078 			vput(newvp);
1079 			*vpp = NULLVP;
1080 		}
1081 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1082 		    (flags & ISLASTCN) && error == ENOENT) {
1083 			if (mp->mnt_flag & MNT_RDONLY)
1084 				error = EROFS;
1085 			else
1086 				error = EJUSTRETURN;
1087 		}
1088 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1089 			cnp->cn_flags |= SAVENAME;
1090 		if (NFS_ISV4(dvp))
1091 			error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1092 		return (error);
1093 	}
1094 
1095 	/*
1096 	 * Handle RENAME case...
1097 	 */
1098 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1099 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1100 			FREE((caddr_t)nfhp, M_NFSFH);
1101 			return (EISDIR);
1102 		}
1103 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1104 		if (error)
1105 			return (error);
1106 		newvp = NFSTOV(np);
1107 		if (attrflag)
1108 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1109 			    0, 1);
1110 		*vpp = newvp;
1111 		cnp->cn_flags |= SAVENAME;
1112 		return (0);
1113 	}
1114 
1115 	if (flags & ISDOTDOT) {
1116 		ltype = VOP_ISLOCKED(dvp);
1117 		error = vfs_busy(mp, MBF_NOWAIT);
1118 		if (error != 0) {
1119 			vfs_ref(mp);
1120 			VOP_UNLOCK(dvp, 0);
1121 			error = vfs_busy(mp, 0);
1122 			vn_lock(dvp, ltype | LK_RETRY);
1123 			vfs_rel(mp);
1124 			if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1125 				vfs_unbusy(mp);
1126 				error = ENOENT;
1127 			}
1128 			if (error != 0)
1129 				return (error);
1130 		}
1131 		VOP_UNLOCK(dvp, 0);
1132 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1133 		if (error == 0)
1134 			newvp = NFSTOV(np);
1135 		vfs_unbusy(mp);
1136 		if (newvp != dvp)
1137 			vn_lock(dvp, ltype | LK_RETRY);
1138 		if (dvp->v_iflag & VI_DOOMED) {
1139 			if (error == 0) {
1140 				if (newvp == dvp)
1141 					vrele(newvp);
1142 				else
1143 					vput(newvp);
1144 			}
1145 			error = ENOENT;
1146 		}
1147 		if (error != 0)
1148 			return (error);
1149 		if (attrflag)
1150 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1151 			    0, 1);
1152 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1153 		FREE((caddr_t)nfhp, M_NFSFH);
1154 		VREF(dvp);
1155 		newvp = dvp;
1156 		if (attrflag)
1157 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1158 			    0, 1);
1159 	} else {
1160 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL);
1161 		if (error)
1162 			return (error);
1163 		newvp = NFSTOV(np);
1164 		if (attrflag)
1165 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1166 			    0, 1);
1167 	}
1168 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1169 		cnp->cn_flags |= SAVENAME;
1170 	if ((cnp->cn_flags & MAKEENTRY) &&
1171 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1172 		np->n_ctime = np->n_vattr.na_vattr.va_ctime.tv_sec;
1173 		cache_enter(dvp, newvp, cnp);
1174 	}
1175 	*vpp = newvp;
1176 	return (0);
1177 }
1178 
1179 /*
1180  * nfs read call.
1181  * Just call ncl_bioread() to do the work.
1182  */
1183 static int
1184 nfs_read(struct vop_read_args *ap)
1185 {
1186 	struct vnode *vp = ap->a_vp;
1187 
1188 	switch (vp->v_type) {
1189 	case VREG:
1190 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1191 	case VDIR:
1192 		return (EISDIR);
1193 	default:
1194 		return (EOPNOTSUPP);
1195 	}
1196 }
1197 
1198 /*
1199  * nfs readlink call
1200  */
1201 static int
1202 nfs_readlink(struct vop_readlink_args *ap)
1203 {
1204 	struct vnode *vp = ap->a_vp;
1205 
1206 	if (vp->v_type != VLNK)
1207 		return (EINVAL);
1208 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1209 }
1210 
1211 /*
1212  * Do a readlink rpc.
1213  * Called by ncl_doio() from below the buffer cache.
1214  */
1215 int
1216 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1217 {
1218 	int error, ret, attrflag;
1219 	struct nfsvattr nfsva;
1220 
1221 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1222 	    &attrflag, NULL);
1223 	if (attrflag) {
1224 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1225 		if (ret && !error)
1226 			error = ret;
1227 	}
1228 	if (error && NFS_ISV4(vp))
1229 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1230 	return (error);
1231 }
1232 
1233 /*
1234  * nfs read rpc call
1235  * Ditto above
1236  */
1237 int
1238 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1239 {
1240 	int error, ret, attrflag;
1241 	struct nfsvattr nfsva;
1242 
1243 	error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1244 	    NULL);
1245 	if (attrflag) {
1246 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1247 		if (ret && !error)
1248 			error = ret;
1249 	}
1250 	if (error && NFS_ISV4(vp))
1251 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1252 	return (error);
1253 }
1254 
1255 /*
1256  * nfs write call
1257  */
1258 int
1259 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1260 	     int *iomode, int *must_commit)
1261 {
1262 	struct nfsvattr nfsva;
1263 	int error = 0, attrflag, ret;
1264 	u_char verf[NFSX_VERF];
1265 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1266 
1267 	*must_commit = 0;
1268 	error = nfsrpc_write(vp, uiop, iomode, verf, cred,
1269 	    uiop->uio_td, &nfsva, &attrflag, NULL);
1270 	NFSLOCKMNT(nmp);
1271 	if (!error && NFSHASWRITEVERF(nmp) &&
1272 	    NFSBCMP(verf, nmp->nm_verf, NFSX_VERF)) {
1273 		*must_commit = 1;
1274 		NFSBCOPY(verf, nmp->nm_verf, NFSX_VERF);
1275 	}
1276 	NFSUNLOCKMNT(nmp);
1277 	if (attrflag) {
1278 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1279 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1280 			    1);
1281 		else
1282 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1283 			    1);
1284 		if (ret && !error)
1285 			error = ret;
1286 	}
1287 	if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1288 		*iomode = NFSWRITE_FILESYNC;
1289 	if (error && NFS_ISV4(vp))
1290 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1291 	return (error);
1292 }
1293 
1294 /*
1295  * nfs mknod rpc
1296  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1297  * mode set to specify the file type and the size field for rdev.
1298  */
1299 static int
1300 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1301     struct vattr *vap)
1302 {
1303 	struct nfsvattr nfsva, dnfsva;
1304 	struct vnode *newvp = NULL;
1305 	struct nfsnode *np = NULL, *dnp;
1306 	struct nfsfh *nfhp;
1307 	struct vattr vattr;
1308 	int error = 0, attrflag, dattrflag;
1309 	u_int32_t rdev;
1310 
1311 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1312 		rdev = vap->va_rdev;
1313 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1314 		rdev = 0xffffffff;
1315 	else
1316 		return (EOPNOTSUPP);
1317 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1318 		return (error);
1319 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1320 	    rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1321 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1322 	if (!error) {
1323 		if (!nfhp)
1324 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1325 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1326 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1327 			    NULL);
1328 		if (nfhp)
1329 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1330 			    cnp->cn_thread, &np, NULL);
1331 	}
1332 	if (dattrflag)
1333 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1334 	if (!error) {
1335 		newvp = NFSTOV(np);
1336 		if (attrflag)
1337 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1338 			    0, 1);
1339 	}
1340 	if (!error) {
1341 		if ((cnp->cn_flags & MAKEENTRY))
1342 			cache_enter(dvp, newvp, cnp);
1343 		*vpp = newvp;
1344 	} else if (NFS_ISV4(dvp)) {
1345 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1346 		    vap->va_gid);
1347 	}
1348 	dnp = VTONFS(dvp);
1349 	mtx_lock(&dnp->n_mtx);
1350 	dnp->n_flag |= NMODIFIED;
1351 	if (!dattrflag)
1352 		dnp->n_attrstamp = 0;
1353 	mtx_unlock(&dnp->n_mtx);
1354 	return (error);
1355 }
1356 
1357 /*
1358  * nfs mknod vop
1359  * just call nfs_mknodrpc() to do the work.
1360  */
1361 /* ARGSUSED */
1362 static int
1363 nfs_mknod(struct vop_mknod_args *ap)
1364 {
1365 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1366 }
1367 
1368 static u_long create_verf;
1369 /*
1370  * nfs file create call
1371  */
1372 static int
1373 nfs_create(struct vop_create_args *ap)
1374 {
1375 	struct vnode *dvp = ap->a_dvp;
1376 	struct vattr *vap = ap->a_vap;
1377 	struct componentname *cnp = ap->a_cnp;
1378 	struct nfsnode *np = NULL, *dnp;
1379 	struct vnode *newvp = NULL;
1380 	struct nfsmount *nmp;
1381 	struct nfsvattr dnfsva, nfsva;
1382 	struct nfsfh *nfhp;
1383 	nfsquad_t cverf;
1384 	int error = 0, attrflag, dattrflag, fmode = 0;
1385 	struct vattr vattr;
1386 
1387 	/*
1388 	 * Oops, not for me..
1389 	 */
1390 	if (vap->va_type == VSOCK)
1391 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1392 
1393 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1394 		return (error);
1395 	if (vap->va_vaflags & VA_EXCLUSIVE)
1396 		fmode |= O_EXCL;
1397 	dnp = VTONFS(dvp);
1398 	nmp = VFSTONFS(vnode_mount(dvp));
1399 again:
1400 	/* For NFSv4, wait until any remove is done. */
1401 	mtx_lock(&dnp->n_mtx);
1402 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1403 		dnp->n_flag |= NREMOVEWANT;
1404 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1405 	}
1406 	mtx_unlock(&dnp->n_mtx);
1407 
1408 	CURVNET_SET(P_TO_VNET(&proc0));
1409 #ifdef INET
1410 	IN_IFADDR_RLOCK();
1411 	if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1412 		cverf.lval[0] = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1413 	else
1414 #endif
1415 		cverf.lval[0] = create_verf;
1416 #ifdef INET
1417 	IN_IFADDR_RUNLOCK();
1418 #endif
1419 	cverf.lval[1] = ++create_verf;
1420 	CURVNET_RESTORE();
1421 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1422 	    vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1423 	    &nfhp, &attrflag, &dattrflag, NULL);
1424 	if (!error) {
1425 		if (nfhp == NULL)
1426 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1427 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1428 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1429 			    NULL);
1430 		if (nfhp != NULL)
1431 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1432 			    cnp->cn_thread, &np, NULL);
1433 	}
1434 	if (dattrflag)
1435 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1436 	if (!error) {
1437 		newvp = NFSTOV(np);
1438 		if (attrflag)
1439 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1440 			    0, 1);
1441 	}
1442 	if (error) {
1443 		if (newvp != NULL) {
1444 			vrele(newvp);
1445 			newvp = NULL;
1446 		}
1447 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1448 		    error == NFSERR_NOTSUPP) {
1449 			fmode &= ~O_EXCL;
1450 			goto again;
1451 		}
1452 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1453 		if (nfscl_checksattr(vap, &nfsva)) {
1454 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1455 			    cnp->cn_thread, &nfsva, &attrflag, NULL);
1456 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1457 			    vap->va_gid != (gid_t)VNOVAL)) {
1458 				/* try again without setting uid/gid */
1459 				vap->va_uid = (uid_t)VNOVAL;
1460 				vap->va_gid = (uid_t)VNOVAL;
1461 				error = nfsrpc_setattr(newvp, vap, NULL,
1462 				    cnp->cn_cred, cnp->cn_thread, &nfsva,
1463 				    &attrflag, NULL);
1464 			}
1465 			if (attrflag)
1466 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1467 				    NULL, 0, 1);
1468 		}
1469 	}
1470 	if (!error) {
1471 		if (cnp->cn_flags & MAKEENTRY)
1472 			cache_enter(dvp, newvp, cnp);
1473 		*ap->a_vpp = newvp;
1474 	} else if (NFS_ISV4(dvp)) {
1475 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1476 		    vap->va_gid);
1477 	}
1478 	mtx_lock(&dnp->n_mtx);
1479 	dnp->n_flag |= NMODIFIED;
1480 	if (!dattrflag)
1481 		dnp->n_attrstamp = 0;
1482 	mtx_unlock(&dnp->n_mtx);
1483 	return (error);
1484 }
1485 
1486 /*
1487  * nfs file remove call
1488  * To try and make nfs semantics closer to ufs semantics, a file that has
1489  * other processes using the vnode is renamed instead of removed and then
1490  * removed later on the last close.
1491  * - If v_usecount > 1
1492  *	  If a rename is not already in the works
1493  *	     call nfs_sillyrename() to set it up
1494  *     else
1495  *	  do the remove rpc
1496  */
1497 static int
1498 nfs_remove(struct vop_remove_args *ap)
1499 {
1500 	struct vnode *vp = ap->a_vp;
1501 	struct vnode *dvp = ap->a_dvp;
1502 	struct componentname *cnp = ap->a_cnp;
1503 	struct nfsnode *np = VTONFS(vp);
1504 	int error = 0;
1505 	struct vattr vattr;
1506 
1507 #ifndef DIAGNOSTIC
1508 	if ((cnp->cn_flags & HASBUF) == 0)
1509 		panic("nfs_remove: no name");
1510 	if (vrefcnt(vp) < 1)
1511 		panic("nfs_remove: bad v_usecount");
1512 #endif
1513 	if (vp->v_type == VDIR)
1514 		error = EPERM;
1515 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1516 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1517 	    vattr.va_nlink > 1)) {
1518 		/*
1519 		 * Purge the name cache so that the chance of a lookup for
1520 		 * the name succeeding while the remove is in progress is
1521 		 * minimized. Without node locking it can still happen, such
1522 		 * that an I/O op returns ESTALE, but since you get this if
1523 		 * another host removes the file..
1524 		 */
1525 		cache_purge(vp);
1526 		/*
1527 		 * throw away biocache buffers, mainly to avoid
1528 		 * unnecessary delayed writes later.
1529 		 */
1530 		error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1531 		/* Do the rpc */
1532 		if (error != EINTR && error != EIO)
1533 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1534 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1535 		/*
1536 		 * Kludge City: If the first reply to the remove rpc is lost..
1537 		 *   the reply to the retransmitted request will be ENOENT
1538 		 *   since the file was in fact removed
1539 		 *   Therefore, we cheat and return success.
1540 		 */
1541 		if (error == ENOENT)
1542 			error = 0;
1543 	} else if (!np->n_sillyrename)
1544 		error = nfs_sillyrename(dvp, vp, cnp);
1545 	np->n_attrstamp = 0;
1546 	return (error);
1547 }
1548 
1549 /*
1550  * nfs file remove rpc called from nfs_inactive
1551  */
1552 int
1553 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1554 {
1555 	/*
1556 	 * Make sure that the directory vnode is still valid.
1557 	 * XXX we should lock sp->s_dvp here.
1558 	 */
1559 	if (sp->s_dvp->v_type == VBAD)
1560 		return (0);
1561 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1562 	    sp->s_cred, NULL));
1563 }
1564 
1565 /*
1566  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1567  */
1568 static int
1569 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1570     int namelen, struct ucred *cred, struct thread *td)
1571 {
1572 	struct nfsvattr dnfsva;
1573 	struct nfsnode *dnp = VTONFS(dvp);
1574 	int error = 0, dattrflag;
1575 
1576 	mtx_lock(&dnp->n_mtx);
1577 	dnp->n_flag |= NREMOVEINPROG;
1578 	mtx_unlock(&dnp->n_mtx);
1579 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1580 	    &dattrflag, NULL);
1581 	mtx_lock(&dnp->n_mtx);
1582 	if ((dnp->n_flag & NREMOVEWANT)) {
1583 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1584 		mtx_unlock(&dnp->n_mtx);
1585 		wakeup((caddr_t)dnp);
1586 	} else {
1587 		dnp->n_flag &= ~NREMOVEINPROG;
1588 		mtx_unlock(&dnp->n_mtx);
1589 	}
1590 	if (dattrflag)
1591 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1592 	mtx_lock(&dnp->n_mtx);
1593 	dnp->n_flag |= NMODIFIED;
1594 	if (!dattrflag)
1595 		dnp->n_attrstamp = 0;
1596 	mtx_unlock(&dnp->n_mtx);
1597 	if (error && NFS_ISV4(dvp))
1598 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1599 	return (error);
1600 }
1601 
1602 /*
1603  * nfs file rename call
1604  */
1605 static int
1606 nfs_rename(struct vop_rename_args *ap)
1607 {
1608 	struct vnode *fvp = ap->a_fvp;
1609 	struct vnode *tvp = ap->a_tvp;
1610 	struct vnode *fdvp = ap->a_fdvp;
1611 	struct vnode *tdvp = ap->a_tdvp;
1612 	struct componentname *tcnp = ap->a_tcnp;
1613 	struct componentname *fcnp = ap->a_fcnp;
1614 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1615 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1616 	struct nfsv4node *newv4 = NULL;
1617 	int error;
1618 
1619 #ifndef DIAGNOSTIC
1620 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1621 	    (fcnp->cn_flags & HASBUF) == 0)
1622 		panic("nfs_rename: no name");
1623 #endif
1624 	/* Check for cross-device rename */
1625 	if ((fvp->v_mount != tdvp->v_mount) ||
1626 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1627 		error = EXDEV;
1628 		goto out;
1629 	}
1630 
1631 	if (fvp == tvp) {
1632 		ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1633 		error = 0;
1634 		goto out;
1635 	}
1636 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)))
1637 		goto out;
1638 
1639 	/*
1640 	 * We have to flush B_DELWRI data prior to renaming
1641 	 * the file.  If we don't, the delayed-write buffers
1642 	 * can be flushed out later after the file has gone stale
1643 	 * under NFSV3.  NFSV2 does not have this problem because
1644 	 * ( as far as I can tell ) it flushes dirty buffers more
1645 	 * often.
1646 	 *
1647 	 * Skip the rename operation if the fsync fails, this can happen
1648 	 * due to the server's volume being full, when we pushed out data
1649 	 * that was written back to our cache earlier. Not checking for
1650 	 * this condition can result in potential (silent) data loss.
1651 	 */
1652 	error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1653 	VOP_UNLOCK(fvp, 0);
1654 	if (!error && tvp)
1655 		error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1656 	if (error)
1657 		goto out;
1658 
1659 	/*
1660 	 * If the tvp exists and is in use, sillyrename it before doing the
1661 	 * rename of the new file over it.
1662 	 * XXX Can't sillyrename a directory.
1663 	 */
1664 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1665 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1666 		vput(tvp);
1667 		tvp = NULL;
1668 	}
1669 
1670 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1671 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1672 	    tcnp->cn_thread);
1673 
1674 	if (!error) {
1675 		/*
1676 		 * For NFSv4, check to see if it is the same name and
1677 		 * replace the name, if it is different.
1678 		 */
1679 		MALLOC(newv4, struct nfsv4node *,
1680 		    sizeof (struct nfsv4node) +
1681 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1682 		    M_NFSV4NODE, M_WAITOK);
1683 		mtx_lock(&tdnp->n_mtx);
1684 		mtx_lock(&fnp->n_mtx);
1685 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1686 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1687 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1688 		      tcnp->cn_namelen) ||
1689 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1690 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1691 			tdnp->n_fhp->nfh_len))) {
1692 #ifdef notdef
1693 { char nnn[100]; int nnnl;
1694 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1695 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1696 nnn[nnnl] = '\0';
1697 printf("ren replace=%s\n",nnn);
1698 }
1699 #endif
1700 			FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1701 			fnp->n_v4 = newv4;
1702 			newv4 = NULL;
1703 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1704 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1705 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1706 			    tdnp->n_fhp->nfh_len);
1707 			NFSBCOPY(tcnp->cn_nameptr,
1708 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1709 		}
1710 		mtx_unlock(&tdnp->n_mtx);
1711 		mtx_unlock(&fnp->n_mtx);
1712 		if (newv4 != NULL)
1713 			FREE((caddr_t)newv4, M_NFSV4NODE);
1714 	}
1715 
1716 	if (fvp->v_type == VDIR) {
1717 		if (tvp != NULL && tvp->v_type == VDIR)
1718 			cache_purge(tdvp);
1719 		cache_purge(fdvp);
1720 	}
1721 
1722 out:
1723 	if (tdvp == tvp)
1724 		vrele(tdvp);
1725 	else
1726 		vput(tdvp);
1727 	if (tvp)
1728 		vput(tvp);
1729 	vrele(fdvp);
1730 	vrele(fvp);
1731 	/*
1732 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1733 	 */
1734 	if (error == ENOENT)
1735 		error = 0;
1736 	return (error);
1737 }
1738 
1739 /*
1740  * nfs file rename rpc called from nfs_remove() above
1741  */
1742 static int
1743 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1744     struct sillyrename *sp)
1745 {
1746 
1747 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1748 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1749 	    scnp->cn_thread));
1750 }
1751 
1752 /*
1753  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1754  */
1755 static int
1756 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1757     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1758     int tnamelen, struct ucred *cred, struct thread *td)
1759 {
1760 	struct nfsvattr fnfsva, tnfsva;
1761 	struct nfsnode *fdnp = VTONFS(fdvp);
1762 	struct nfsnode *tdnp = VTONFS(tdvp);
1763 	int error = 0, fattrflag, tattrflag;
1764 
1765 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1766 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1767 	    &tattrflag, NULL, NULL);
1768 	mtx_lock(&fdnp->n_mtx);
1769 	fdnp->n_flag |= NMODIFIED;
1770 	mtx_unlock(&fdnp->n_mtx);
1771 	mtx_lock(&tdnp->n_mtx);
1772 	tdnp->n_flag |= NMODIFIED;
1773 	mtx_unlock(&tdnp->n_mtx);
1774 	if (fattrflag)
1775 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1776 	else
1777 		fdnp->n_attrstamp = 0;
1778 	if (tattrflag)
1779 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1780 	else
1781 		tdnp->n_attrstamp = 0;
1782 	if (error && NFS_ISV4(fdvp))
1783 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1784 	return (error);
1785 }
1786 
1787 /*
1788  * nfs hard link create call
1789  */
1790 static int
1791 nfs_link(struct vop_link_args *ap)
1792 {
1793 	struct vnode *vp = ap->a_vp;
1794 	struct vnode *tdvp = ap->a_tdvp;
1795 	struct componentname *cnp = ap->a_cnp;
1796 	struct nfsnode *tdnp;
1797 	struct nfsvattr nfsva, dnfsva;
1798 	int error = 0, attrflag, dattrflag;
1799 
1800 	if (vp->v_mount != tdvp->v_mount) {
1801 		return (EXDEV);
1802 	}
1803 
1804 	/*
1805 	 * Push all writes to the server, so that the attribute cache
1806 	 * doesn't get "out of sync" with the server.
1807 	 * XXX There should be a better way!
1808 	 */
1809 	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1810 
1811 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1812 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1813 	    &dattrflag, NULL);
1814 	tdnp = VTONFS(tdvp);
1815 	mtx_lock(&tdnp->n_mtx);
1816 	tdnp->n_flag |= NMODIFIED;
1817 	mtx_unlock(&tdnp->n_mtx);
1818 	if (attrflag)
1819 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1820 	else
1821 		VTONFS(vp)->n_attrstamp = 0;
1822 	if (dattrflag)
1823 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1824 	else
1825 		tdnp->n_attrstamp = 0;
1826 	/*
1827 	 * If negative lookup caching is enabled, I might as well
1828 	 * add an entry for this node. Not necessary for correctness,
1829 	 * but if negative caching is enabled, then the system
1830 	 * must care about lookup caching hit rate, so...
1831 	 */
1832 	if (newnfs_neglookup_enable != 0 &&
1833 	    (cnp->cn_flags & MAKEENTRY))
1834 		cache_enter(tdvp, vp, cnp);
1835 	if (error && NFS_ISV4(vp))
1836 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1837 		    (gid_t)0);
1838 	return (error);
1839 }
1840 
1841 /*
1842  * nfs symbolic link create call
1843  */
1844 static int
1845 nfs_symlink(struct vop_symlink_args *ap)
1846 {
1847 	struct vnode *dvp = ap->a_dvp;
1848 	struct vattr *vap = ap->a_vap;
1849 	struct componentname *cnp = ap->a_cnp;
1850 	struct nfsvattr nfsva, dnfsva;
1851 	struct nfsfh *nfhp;
1852 	struct nfsnode *np = NULL, *dnp;
1853 	struct vnode *newvp = NULL;
1854 	int error = 0, attrflag, dattrflag, ret;
1855 
1856 	vap->va_type = VLNK;
1857 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1858 	    ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1859 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1860 	if (nfhp) {
1861 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
1862 		    &np, NULL);
1863 		if (!ret)
1864 			newvp = NFSTOV(np);
1865 		else if (!error)
1866 			error = ret;
1867 	}
1868 	if (newvp != NULL) {
1869 		if (attrflag)
1870 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1871 			    0, 1);
1872 	} else if (!error) {
1873 		/*
1874 		 * If we do not have an error and we could not extract the
1875 		 * newvp from the response due to the request being NFSv2, we
1876 		 * have to do a lookup in order to obtain a newvp to return.
1877 		 */
1878 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1879 		    cnp->cn_cred, cnp->cn_thread, &np);
1880 		if (!error)
1881 			newvp = NFSTOV(np);
1882 	}
1883 	if (error) {
1884 		if (newvp)
1885 			vput(newvp);
1886 		if (NFS_ISV4(dvp))
1887 			error = nfscl_maperr(cnp->cn_thread, error,
1888 			    vap->va_uid, vap->va_gid);
1889 	} else {
1890 		/*
1891 		 * If negative lookup caching is enabled, I might as well
1892 		 * add an entry for this node. Not necessary for correctness,
1893 		 * but if negative caching is enabled, then the system
1894 		 * must care about lookup caching hit rate, so...
1895 		 */
1896 		if (newnfs_neglookup_enable != 0 &&
1897 		    (cnp->cn_flags & MAKEENTRY))
1898 			cache_enter(dvp, newvp, cnp);
1899 		*ap->a_vpp = newvp;
1900 	}
1901 
1902 	dnp = VTONFS(dvp);
1903 	mtx_lock(&dnp->n_mtx);
1904 	dnp->n_flag |= NMODIFIED;
1905 	mtx_unlock(&dnp->n_mtx);
1906 	if (dattrflag)
1907 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1908 	else
1909 		dnp->n_attrstamp = 0;
1910 	return (error);
1911 }
1912 
1913 /*
1914  * nfs make dir call
1915  */
1916 static int
1917 nfs_mkdir(struct vop_mkdir_args *ap)
1918 {
1919 	struct vnode *dvp = ap->a_dvp;
1920 	struct vattr *vap = ap->a_vap;
1921 	struct componentname *cnp = ap->a_cnp;
1922 	struct nfsnode *np = NULL, *dnp;
1923 	struct vnode *newvp = NULL;
1924 	struct vattr vattr;
1925 	struct nfsfh *nfhp;
1926 	struct nfsvattr nfsva, dnfsva;
1927 	int error = 0, attrflag, dattrflag, ret;
1928 
1929 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1930 		return (error);
1931 	vap->va_type = VDIR;
1932 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1933 	    vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
1934 	    &attrflag, &dattrflag, NULL);
1935 	dnp = VTONFS(dvp);
1936 	mtx_lock(&dnp->n_mtx);
1937 	dnp->n_flag |= NMODIFIED;
1938 	mtx_unlock(&dnp->n_mtx);
1939 	if (dattrflag)
1940 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1941 	else
1942 		dnp->n_attrstamp = 0;
1943 	if (nfhp) {
1944 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
1945 		    &np, NULL);
1946 		if (!ret) {
1947 			newvp = NFSTOV(np);
1948 			if (attrflag)
1949 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1950 				NULL, 0, 1);
1951 		} else if (!error)
1952 			error = ret;
1953 	}
1954 	if (!error && newvp == NULL) {
1955 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1956 		    cnp->cn_cred, cnp->cn_thread, &np);
1957 		if (!error) {
1958 			newvp = NFSTOV(np);
1959 			if (newvp->v_type != VDIR)
1960 				error = EEXIST;
1961 		}
1962 	}
1963 	if (error) {
1964 		if (newvp)
1965 			vput(newvp);
1966 		if (NFS_ISV4(dvp))
1967 			error = nfscl_maperr(cnp->cn_thread, error,
1968 			    vap->va_uid, vap->va_gid);
1969 	} else {
1970 		/*
1971 		 * If negative lookup caching is enabled, I might as well
1972 		 * add an entry for this node. Not necessary for correctness,
1973 		 * but if negative caching is enabled, then the system
1974 		 * must care about lookup caching hit rate, so...
1975 		 */
1976 		if (newnfs_neglookup_enable != 0 &&
1977 		    (cnp->cn_flags & MAKEENTRY))
1978 			cache_enter(dvp, newvp, cnp);
1979 		*ap->a_vpp = newvp;
1980 	}
1981 	return (error);
1982 }
1983 
1984 /*
1985  * nfs remove directory call
1986  */
1987 static int
1988 nfs_rmdir(struct vop_rmdir_args *ap)
1989 {
1990 	struct vnode *vp = ap->a_vp;
1991 	struct vnode *dvp = ap->a_dvp;
1992 	struct componentname *cnp = ap->a_cnp;
1993 	struct nfsnode *dnp;
1994 	struct nfsvattr dnfsva;
1995 	int error, dattrflag;
1996 
1997 	if (dvp == vp)
1998 		return (EINVAL);
1999 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2000 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2001 	dnp = VTONFS(dvp);
2002 	mtx_lock(&dnp->n_mtx);
2003 	dnp->n_flag |= NMODIFIED;
2004 	mtx_unlock(&dnp->n_mtx);
2005 	if (dattrflag)
2006 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2007 	else
2008 		dnp->n_attrstamp = 0;
2009 
2010 	cache_purge(dvp);
2011 	cache_purge(vp);
2012 	if (error && NFS_ISV4(dvp))
2013 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2014 		    (gid_t)0);
2015 	/*
2016 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2017 	 */
2018 	if (error == ENOENT)
2019 		error = 0;
2020 	return (error);
2021 }
2022 
2023 /*
2024  * nfs readdir call
2025  */
2026 static int
2027 nfs_readdir(struct vop_readdir_args *ap)
2028 {
2029 	struct vnode *vp = ap->a_vp;
2030 	struct nfsnode *np = VTONFS(vp);
2031 	struct uio *uio = ap->a_uio;
2032 	int tresid, error = 0;
2033 	struct vattr vattr;
2034 
2035 	if (vp->v_type != VDIR)
2036 		return(EPERM);
2037 
2038 	/*
2039 	 * First, check for hit on the EOF offset cache
2040 	 */
2041 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2042 	    (np->n_flag & NMODIFIED) == 0) {
2043 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2044 			mtx_lock(&np->n_mtx);
2045 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2046 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2047 				mtx_unlock(&np->n_mtx);
2048 				NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2049 				return (0);
2050 			} else
2051 				mtx_unlock(&np->n_mtx);
2052 		}
2053 	}
2054 
2055 	/*
2056 	 * Call ncl_bioread() to do the real work.
2057 	 */
2058 	tresid = uio->uio_resid;
2059 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2060 
2061 	if (!error && uio->uio_resid == tresid)
2062 		NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2063 	return (error);
2064 }
2065 
2066 /*
2067  * Readdir rpc call.
2068  * Called from below the buffer cache by ncl_doio().
2069  */
2070 int
2071 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2072     struct thread *td)
2073 {
2074 	struct nfsvattr nfsva;
2075 	nfsuint64 *cookiep, cookie;
2076 	struct nfsnode *dnp = VTONFS(vp);
2077 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2078 	int error = 0, eof, attrflag;
2079 
2080 #ifndef DIAGNOSTIC
2081 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2082 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2083 		panic("nfs readdirrpc bad uio");
2084 #endif
2085 
2086 	/*
2087 	 * If there is no cookie, assume directory was stale.
2088 	 */
2089 	ncl_dircookie_lock(dnp);
2090 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2091 	if (cookiep) {
2092 		cookie = *cookiep;
2093 		ncl_dircookie_unlock(dnp);
2094 	} else {
2095 		ncl_dircookie_unlock(dnp);
2096 		return (NFSERR_BAD_COOKIE);
2097 	}
2098 
2099 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2100 		(void)ncl_fsinfo(nmp, vp, cred, td);
2101 
2102 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2103 	    &attrflag, &eof, NULL);
2104 	if (attrflag)
2105 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2106 
2107 	if (!error) {
2108 		/*
2109 		 * We are now either at the end of the directory or have filled
2110 		 * the block.
2111 		 */
2112 		if (eof)
2113 			dnp->n_direofoffset = uiop->uio_offset;
2114 		else {
2115 			if (uiop->uio_resid > 0)
2116 				ncl_printf("EEK! readdirrpc resid > 0\n");
2117 			ncl_dircookie_lock(dnp);
2118 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2119 			*cookiep = cookie;
2120 			ncl_dircookie_unlock(dnp);
2121 		}
2122 	} else if (NFS_ISV4(vp)) {
2123 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2124 	}
2125 	return (error);
2126 }
2127 
2128 /*
2129  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2130  */
2131 int
2132 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2133     struct thread *td)
2134 {
2135 	struct nfsvattr nfsva;
2136 	nfsuint64 *cookiep, cookie;
2137 	struct nfsnode *dnp = VTONFS(vp);
2138 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2139 	int error = 0, attrflag, eof;
2140 
2141 #ifndef DIAGNOSTIC
2142 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2143 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2144 		panic("nfs readdirplusrpc bad uio");
2145 #endif
2146 
2147 	/*
2148 	 * If there is no cookie, assume directory was stale.
2149 	 */
2150 	ncl_dircookie_lock(dnp);
2151 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2152 	if (cookiep) {
2153 		cookie = *cookiep;
2154 		ncl_dircookie_unlock(dnp);
2155 	} else {
2156 		ncl_dircookie_unlock(dnp);
2157 		return (NFSERR_BAD_COOKIE);
2158 	}
2159 
2160 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2161 		(void)ncl_fsinfo(nmp, vp, cred, td);
2162 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2163 	    &attrflag, &eof, NULL);
2164 	if (attrflag)
2165 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2166 
2167 	if (!error) {
2168 		/*
2169 		 * We are now either at end of the directory or have filled the
2170 		 * the block.
2171 		 */
2172 		if (eof)
2173 			dnp->n_direofoffset = uiop->uio_offset;
2174 		else {
2175 			if (uiop->uio_resid > 0)
2176 				ncl_printf("EEK! readdirplusrpc resid > 0\n");
2177 			ncl_dircookie_lock(dnp);
2178 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2179 			*cookiep = cookie;
2180 			ncl_dircookie_unlock(dnp);
2181 		}
2182 	} else if (NFS_ISV4(vp)) {
2183 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2184 	}
2185 	return (error);
2186 }
2187 
2188 /*
2189  * Silly rename. To make the NFS filesystem that is stateless look a little
2190  * more like the "ufs" a remove of an active vnode is translated to a rename
2191  * to a funny looking filename that is removed by nfs_inactive on the
2192  * nfsnode. There is the potential for another process on a different client
2193  * to create the same funny name between the nfs_lookitup() fails and the
2194  * nfs_rename() completes, but...
2195  */
2196 static int
2197 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2198 {
2199 	struct sillyrename *sp;
2200 	struct nfsnode *np;
2201 	int error;
2202 	short pid;
2203 	unsigned int lticks;
2204 
2205 	cache_purge(dvp);
2206 	np = VTONFS(vp);
2207 #ifndef DIAGNOSTIC
2208 	if (vp->v_type == VDIR)
2209 		panic("nfs: sillyrename dir");
2210 #endif
2211 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2212 	    M_NEWNFSREQ, M_WAITOK);
2213 	sp->s_cred = crhold(cnp->cn_cred);
2214 	sp->s_dvp = dvp;
2215 	VREF(dvp);
2216 
2217 	/*
2218 	 * Fudge together a funny name.
2219 	 * Changing the format of the funny name to accomodate more
2220 	 * sillynames per directory.
2221 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2222 	 * CPU ticks since boot.
2223 	 */
2224 	pid = cnp->cn_thread->td_proc->p_pid;
2225 	lticks = (unsigned int)ticks;
2226 	for ( ; ; ) {
2227 		sp->s_namlen = sprintf(sp->s_name,
2228 				       ".nfs.%08x.%04x4.4", lticks,
2229 				       pid);
2230 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2231 				 cnp->cn_thread, NULL))
2232 			break;
2233 		lticks++;
2234 	}
2235 	error = nfs_renameit(dvp, vp, cnp, sp);
2236 	if (error)
2237 		goto bad;
2238 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2239 		cnp->cn_thread, &np);
2240 	np->n_sillyrename = sp;
2241 	return (0);
2242 bad:
2243 	vrele(sp->s_dvp);
2244 	crfree(sp->s_cred);
2245 	free((caddr_t)sp, M_NEWNFSREQ);
2246 	return (error);
2247 }
2248 
2249 /*
2250  * Look up a file name and optionally either update the file handle or
2251  * allocate an nfsnode, depending on the value of npp.
2252  * npp == NULL	--> just do the lookup
2253  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2254  *			handled too
2255  * *npp != NULL --> update the file handle in the vnode
2256  */
2257 static int
2258 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2259     struct thread *td, struct nfsnode **npp)
2260 {
2261 	struct vnode *newvp = NULL, *vp;
2262 	struct nfsnode *np, *dnp = VTONFS(dvp);
2263 	struct nfsfh *nfhp, *onfhp;
2264 	struct nfsvattr nfsva, dnfsva;
2265 	struct componentname cn;
2266 	int error = 0, attrflag, dattrflag;
2267 	u_int hash;
2268 
2269 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2270 	    &nfhp, &attrflag, &dattrflag, NULL);
2271 	if (dattrflag)
2272 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2273 	if (npp && !error) {
2274 		if (*npp != NULL) {
2275 		    np = *npp;
2276 		    vp = NFSTOV(np);
2277 		    /*
2278 		     * For NFSv4, check to see if it is the same name and
2279 		     * replace the name, if it is different.
2280 		     */
2281 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2282 			(np->n_v4->n4_namelen != len ||
2283 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2284 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2285 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2286 			 dnp->n_fhp->nfh_len))) {
2287 #ifdef notdef
2288 { char nnn[100]; int nnnl;
2289 nnnl = (len < 100) ? len : 99;
2290 bcopy(name, nnn, nnnl);
2291 nnn[nnnl] = '\0';
2292 printf("replace=%s\n",nnn);
2293 }
2294 #endif
2295 			    FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2296 			    MALLOC(np->n_v4, struct nfsv4node *,
2297 				sizeof (struct nfsv4node) +
2298 				dnp->n_fhp->nfh_len + len - 1,
2299 				M_NFSV4NODE, M_WAITOK);
2300 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2301 			    np->n_v4->n4_namelen = len;
2302 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2303 				dnp->n_fhp->nfh_len);
2304 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2305 		    }
2306 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2307 			FNV1_32_INIT);
2308 		    onfhp = np->n_fhp;
2309 		    /*
2310 		     * Rehash node for new file handle.
2311 		     */
2312 		    vfs_hash_rehash(vp, hash);
2313 		    np->n_fhp = nfhp;
2314 		    if (onfhp != NULL)
2315 			FREE((caddr_t)onfhp, M_NFSFH);
2316 		    newvp = NFSTOV(np);
2317 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2318 		    FREE((caddr_t)nfhp, M_NFSFH);
2319 		    VREF(dvp);
2320 		    newvp = dvp;
2321 		} else {
2322 		    cn.cn_nameptr = name;
2323 		    cn.cn_namelen = len;
2324 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2325 			&np, NULL);
2326 		    if (error)
2327 			return (error);
2328 		    newvp = NFSTOV(np);
2329 		}
2330 		if (!attrflag && *npp == NULL) {
2331 			vrele(newvp);
2332 			return (ENOENT);
2333 		}
2334 		if (attrflag)
2335 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2336 			    0, 1);
2337 	}
2338 	if (npp && *npp == NULL) {
2339 		if (error) {
2340 			if (newvp) {
2341 				if (newvp == dvp)
2342 					vrele(newvp);
2343 				else
2344 					vput(newvp);
2345 			}
2346 		} else
2347 			*npp = np;
2348 	}
2349 	if (error && NFS_ISV4(dvp))
2350 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2351 	return (error);
2352 }
2353 
2354 /*
2355  * Nfs Version 3 and 4 commit rpc
2356  */
2357 int
2358 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2359    struct thread *td)
2360 {
2361 	struct nfsvattr nfsva;
2362 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2363 	int error, attrflag;
2364 	u_char verf[NFSX_VERF];
2365 
2366 	mtx_lock(&nmp->nm_mtx);
2367 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2368 		mtx_unlock(&nmp->nm_mtx);
2369 		return (0);
2370 	}
2371 	mtx_unlock(&nmp->nm_mtx);
2372 	error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2373 	    &attrflag, NULL);
2374 	if (!error) {
2375 		if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2376 			NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2377 			error = NFSERR_STALEWRITEVERF;
2378 		}
2379 		if (!error && attrflag)
2380 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2381 			    0, 1);
2382 	} else if (NFS_ISV4(vp)) {
2383 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2384 	}
2385 	return (error);
2386 }
2387 
2388 /*
2389  * Strategy routine.
2390  * For async requests when nfsiod(s) are running, queue the request by
2391  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2392  * request.
2393  */
2394 static int
2395 nfs_strategy(struct vop_strategy_args *ap)
2396 {
2397 	struct buf *bp = ap->a_bp;
2398 	struct ucred *cr;
2399 
2400 	KASSERT(!(bp->b_flags & B_DONE),
2401 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2402 	BUF_ASSERT_HELD(bp);
2403 
2404 	if (bp->b_iocmd == BIO_READ)
2405 		cr = bp->b_rcred;
2406 	else
2407 		cr = bp->b_wcred;
2408 
2409 	/*
2410 	 * If the op is asynchronous and an i/o daemon is waiting
2411 	 * queue the request, wake it up and wait for completion
2412 	 * otherwise just do it ourselves.
2413 	 */
2414 	if ((bp->b_flags & B_ASYNC) == 0 ||
2415 	    ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2416 		(void)ncl_doio(ap->a_vp, bp, cr, curthread);
2417 	return (0);
2418 }
2419 
2420 /*
2421  * fsync vnode op. Just call ncl_flush() with commit == 1.
2422  */
2423 /* ARGSUSED */
2424 static int
2425 nfs_fsync(struct vop_fsync_args *ap)
2426 {
2427 	return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1));
2428 }
2429 
2430 /*
2431  * Flush all the blocks associated with a vnode.
2432  * 	Walk through the buffer pool and push any dirty pages
2433  *	associated with the vnode.
2434  */
2435 int
2436 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2437     int commit)
2438 {
2439 	struct nfsnode *np = VTONFS(vp);
2440 	struct buf *bp;
2441 	int i;
2442 	struct buf *nbp;
2443 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2444 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2445 	int passone = 1, trycnt = 0;
2446 	u_quad_t off, endoff, toff;
2447 	struct ucred* wcred = NULL;
2448 	struct buf **bvec = NULL;
2449 	struct bufobj *bo;
2450 #ifndef NFS_COMMITBVECSIZ
2451 #define	NFS_COMMITBVECSIZ	20
2452 #endif
2453 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2454 	int bvecsize = 0, bveccount;
2455 
2456 	if (nmp->nm_flag & NFSMNT_INT)
2457 		slpflag = NFS_PCATCH;
2458 	if (!commit)
2459 		passone = 0;
2460 	bo = &vp->v_bufobj;
2461 	/*
2462 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2463 	 * server, but has not been committed to stable storage on the server
2464 	 * yet. On the first pass, the byte range is worked out and the commit
2465 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2466 	 * job.
2467 	 */
2468 again:
2469 	off = (u_quad_t)-1;
2470 	endoff = 0;
2471 	bvecpos = 0;
2472 	if (NFS_ISV34(vp) && commit) {
2473 		if (bvec != NULL && bvec != bvec_on_stack)
2474 			free(bvec, M_TEMP);
2475 		/*
2476 		 * Count up how many buffers waiting for a commit.
2477 		 */
2478 		bveccount = 0;
2479 		BO_LOCK(bo);
2480 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2481 			if (!BUF_ISLOCKED(bp) &&
2482 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2483 				== (B_DELWRI | B_NEEDCOMMIT))
2484 				bveccount++;
2485 		}
2486 		/*
2487 		 * Allocate space to remember the list of bufs to commit.  It is
2488 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2489 		 * If we can't get memory (for whatever reason), we will end up
2490 		 * committing the buffers one-by-one in the loop below.
2491 		 */
2492 		if (bveccount > NFS_COMMITBVECSIZ) {
2493 			/*
2494 			 * Release the vnode interlock to avoid a lock
2495 			 * order reversal.
2496 			 */
2497 			BO_UNLOCK(bo);
2498 			bvec = (struct buf **)
2499 				malloc(bveccount * sizeof(struct buf *),
2500 				       M_TEMP, M_NOWAIT);
2501 			BO_LOCK(bo);
2502 			if (bvec == NULL) {
2503 				bvec = bvec_on_stack;
2504 				bvecsize = NFS_COMMITBVECSIZ;
2505 			} else
2506 				bvecsize = bveccount;
2507 		} else {
2508 			bvec = bvec_on_stack;
2509 			bvecsize = NFS_COMMITBVECSIZ;
2510 		}
2511 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2512 			if (bvecpos >= bvecsize)
2513 				break;
2514 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2515 				nbp = TAILQ_NEXT(bp, b_bobufs);
2516 				continue;
2517 			}
2518 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2519 			    (B_DELWRI | B_NEEDCOMMIT)) {
2520 				BUF_UNLOCK(bp);
2521 				nbp = TAILQ_NEXT(bp, b_bobufs);
2522 				continue;
2523 			}
2524 			BO_UNLOCK(bo);
2525 			bremfree(bp);
2526 			/*
2527 			 * Work out if all buffers are using the same cred
2528 			 * so we can deal with them all with one commit.
2529 			 *
2530 			 * NOTE: we are not clearing B_DONE here, so we have
2531 			 * to do it later on in this routine if we intend to
2532 			 * initiate I/O on the bp.
2533 			 *
2534 			 * Note: to avoid loopback deadlocks, we do not
2535 			 * assign b_runningbufspace.
2536 			 */
2537 			if (wcred == NULL)
2538 				wcred = bp->b_wcred;
2539 			else if (wcred != bp->b_wcred)
2540 				wcred = NOCRED;
2541 			vfs_busy_pages(bp, 1);
2542 
2543 			BO_LOCK(bo);
2544 			/*
2545 			 * bp is protected by being locked, but nbp is not
2546 			 * and vfs_busy_pages() may sleep.  We have to
2547 			 * recalculate nbp.
2548 			 */
2549 			nbp = TAILQ_NEXT(bp, b_bobufs);
2550 
2551 			/*
2552 			 * A list of these buffers is kept so that the
2553 			 * second loop knows which buffers have actually
2554 			 * been committed. This is necessary, since there
2555 			 * may be a race between the commit rpc and new
2556 			 * uncommitted writes on the file.
2557 			 */
2558 			bvec[bvecpos++] = bp;
2559 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2560 				bp->b_dirtyoff;
2561 			if (toff < off)
2562 				off = toff;
2563 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2564 			if (toff > endoff)
2565 				endoff = toff;
2566 		}
2567 		BO_UNLOCK(bo);
2568 	}
2569 	if (bvecpos > 0) {
2570 		/*
2571 		 * Commit data on the server, as required.
2572 		 * If all bufs are using the same wcred, then use that with
2573 		 * one call for all of them, otherwise commit each one
2574 		 * separately.
2575 		 */
2576 		if (wcred != NOCRED)
2577 			retv = ncl_commit(vp, off, (int)(endoff - off),
2578 					  wcred, td);
2579 		else {
2580 			retv = 0;
2581 			for (i = 0; i < bvecpos; i++) {
2582 				off_t off, size;
2583 				bp = bvec[i];
2584 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2585 					bp->b_dirtyoff;
2586 				size = (u_quad_t)(bp->b_dirtyend
2587 						  - bp->b_dirtyoff);
2588 				retv = ncl_commit(vp, off, (int)size,
2589 						  bp->b_wcred, td);
2590 				if (retv) break;
2591 			}
2592 		}
2593 
2594 		if (retv == NFSERR_STALEWRITEVERF)
2595 			ncl_clearcommit(vp->v_mount);
2596 
2597 		/*
2598 		 * Now, either mark the blocks I/O done or mark the
2599 		 * blocks dirty, depending on whether the commit
2600 		 * succeeded.
2601 		 */
2602 		for (i = 0; i < bvecpos; i++) {
2603 			bp = bvec[i];
2604 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2605 			if (retv) {
2606 				/*
2607 				 * Error, leave B_DELWRI intact
2608 				 */
2609 				vfs_unbusy_pages(bp);
2610 				brelse(bp);
2611 			} else {
2612 				/*
2613 				 * Success, remove B_DELWRI ( bundirty() ).
2614 				 *
2615 				 * b_dirtyoff/b_dirtyend seem to be NFS
2616 				 * specific.  We should probably move that
2617 				 * into bundirty(). XXX
2618 				 */
2619 				bufobj_wref(bo);
2620 				bp->b_flags |= B_ASYNC;
2621 				bundirty(bp);
2622 				bp->b_flags &= ~B_DONE;
2623 				bp->b_ioflags &= ~BIO_ERROR;
2624 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2625 				bufdone(bp);
2626 			}
2627 		}
2628 	}
2629 
2630 	/*
2631 	 * Start/do any write(s) that are required.
2632 	 */
2633 loop:
2634 	BO_LOCK(bo);
2635 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2636 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2637 			if (waitfor != MNT_WAIT || passone)
2638 				continue;
2639 
2640 			error = BUF_TIMELOCK(bp,
2641 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2642 			    BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2643 			if (error == 0) {
2644 				BUF_UNLOCK(bp);
2645 				goto loop;
2646 			}
2647 			if (error == ENOLCK) {
2648 				error = 0;
2649 				goto loop;
2650 			}
2651 			if (newnfs_sigintr(nmp, td)) {
2652 				error = EINTR;
2653 				goto done;
2654 			}
2655 			if (slpflag & PCATCH) {
2656 				slpflag = 0;
2657 				slptimeo = 2 * hz;
2658 			}
2659 			goto loop;
2660 		}
2661 		if ((bp->b_flags & B_DELWRI) == 0)
2662 			panic("nfs_fsync: not dirty");
2663 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2664 			BUF_UNLOCK(bp);
2665 			continue;
2666 		}
2667 		BO_UNLOCK(bo);
2668 		bremfree(bp);
2669 		if (passone || !commit)
2670 		    bp->b_flags |= B_ASYNC;
2671 		else
2672 		    bp->b_flags |= B_ASYNC;
2673 		bwrite(bp);
2674 		if (newnfs_sigintr(nmp, td)) {
2675 			error = EINTR;
2676 			goto done;
2677 		}
2678 		goto loop;
2679 	}
2680 	if (passone) {
2681 		passone = 0;
2682 		BO_UNLOCK(bo);
2683 		goto again;
2684 	}
2685 	if (waitfor == MNT_WAIT) {
2686 		while (bo->bo_numoutput) {
2687 			error = bufobj_wwait(bo, slpflag, slptimeo);
2688 			if (error) {
2689 			    BO_UNLOCK(bo);
2690 			    error = newnfs_sigintr(nmp, td);
2691 			    if (error)
2692 				goto done;
2693 			    if (slpflag & PCATCH) {
2694 				slpflag = 0;
2695 				slptimeo = 2 * hz;
2696 			    }
2697 			    BO_LOCK(bo);
2698 			}
2699 		}
2700 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
2701 			BO_UNLOCK(bo);
2702 			goto loop;
2703 		}
2704 		/*
2705 		 * Wait for all the async IO requests to drain
2706 		 */
2707 		BO_UNLOCK(bo);
2708 		mtx_lock(&np->n_mtx);
2709 		while (np->n_directio_asyncwr > 0) {
2710 			np->n_flag |= NFSYNCWAIT;
2711 			error = ncl_msleep(td, (caddr_t)&np->n_directio_asyncwr,
2712 					   &np->n_mtx, slpflag | (PRIBIO + 1),
2713 					   "nfsfsync", 0);
2714 			if (error) {
2715 				if (newnfs_sigintr(nmp, td)) {
2716 					mtx_unlock(&np->n_mtx);
2717 					error = EINTR;
2718 					goto done;
2719 				}
2720 			}
2721 		}
2722 		mtx_unlock(&np->n_mtx);
2723 	} else
2724 		BO_UNLOCK(bo);
2725 	mtx_lock(&np->n_mtx);
2726 	if (np->n_flag & NWRITEERR) {
2727 		error = np->n_error;
2728 		np->n_flag &= ~NWRITEERR;
2729 	}
2730   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
2731 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2732   		np->n_flag &= ~NMODIFIED;
2733 	mtx_unlock(&np->n_mtx);
2734 done:
2735 	if (bvec != NULL && bvec != bvec_on_stack)
2736 		free(bvec, M_TEMP);
2737 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2738 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2739 	     np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2740 		/* try, try again... */
2741 		passone = 1;
2742 		wcred = NULL;
2743 		bvec = NULL;
2744 		bvecsize = 0;
2745 printf("try%d\n", trycnt);
2746 		goto again;
2747 	}
2748 	return (error);
2749 }
2750 
2751 /*
2752  * NFS advisory byte-level locks.
2753  */
2754 static int
2755 nfs_advlock(struct vop_advlock_args *ap)
2756 {
2757 	struct vnode *vp = ap->a_vp;
2758 	struct ucred *cred;
2759 	struct nfsnode *np = VTONFS(ap->a_vp);
2760 	struct proc *p = (struct proc *)ap->a_id;
2761 	struct thread *td = curthread;	/* XXX */
2762 	struct vattr va;
2763 	int ret, error = EOPNOTSUPP;
2764 	u_quad_t size;
2765 
2766 	if (NFS_ISV4(vp) && (ap->a_flags & F_POSIX)) {
2767 		cred = p->p_ucred;
2768 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2769 		if (vp->v_iflag & VI_DOOMED) {
2770 			VOP_UNLOCK(vp, 0);
2771 			return (EBADF);
2772 		}
2773 
2774 		/*
2775 		 * If this is unlocking a write locked region, flush and
2776 		 * commit them before unlocking. This is required by
2777 		 * RFC3530 Sec. 9.3.2.
2778 		 */
2779 		if (ap->a_op == F_UNLCK &&
2780 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td))
2781 			(void) ncl_flush(vp, MNT_WAIT, cred, td, 1);
2782 
2783 		/*
2784 		 * Loop around doing the lock op, while a blocking lock
2785 		 * must wait for the lock op to succeed.
2786 		 */
2787 		do {
2788 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
2789 			    ap->a_fl, 0, cred, td);
2790 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2791 			    ap->a_op == F_SETLK) {
2792 				VOP_UNLOCK(vp, 0);
2793 				error = nfs_catnap(PZERO | PCATCH, "ncladvl");
2794 				if (error)
2795 					return (EINTR);
2796 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2797 				if (vp->v_iflag & VI_DOOMED) {
2798 					VOP_UNLOCK(vp, 0);
2799 					return (EBADF);
2800 				}
2801 			}
2802 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2803 		     ap->a_op == F_SETLK);
2804 		if (ret == NFSERR_DENIED) {
2805 			VOP_UNLOCK(vp, 0);
2806 			return (EAGAIN);
2807 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
2808 			VOP_UNLOCK(vp, 0);
2809 			return (ret);
2810 		} else if (ret != 0) {
2811 			VOP_UNLOCK(vp, 0);
2812 			return (EACCES);
2813 		}
2814 
2815 		/*
2816 		 * Now, if we just got a lock, invalidate data in the buffer
2817 		 * cache, as required, so that the coherency conforms with
2818 		 * RFC3530 Sec. 9.3.2.
2819 		 */
2820 		if (ap->a_op == F_SETLK) {
2821 			if ((np->n_flag & NMODIFIED) == 0) {
2822 				np->n_attrstamp = 0;
2823 				ret = VOP_GETATTR(vp, &va, cred);
2824 			}
2825 			if ((np->n_flag & NMODIFIED) || ret ||
2826 			    np->n_change != va.va_filerev) {
2827 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
2828 				np->n_attrstamp = 0;
2829 				ret = VOP_GETATTR(vp, &va, cred);
2830 				if (!ret) {
2831 					np->n_mtime = va.va_mtime;
2832 					np->n_change = va.va_filerev;
2833 				}
2834 			}
2835 		}
2836 		VOP_UNLOCK(vp, 0);
2837 		return (0);
2838 	} else if (!NFS_ISV4(vp)) {
2839 		error = vn_lock(vp, LK_SHARED);
2840 		if (error)
2841 			return (error);
2842 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2843 			size = VTONFS(vp)->n_size;
2844 			VOP_UNLOCK(vp, 0);
2845 			error = lf_advlock(ap, &(vp->v_lockf), size);
2846 		} else {
2847 			if (ncl_advlock_p)
2848 				error = ncl_advlock_p(ap);
2849 			else
2850 				error = ENOLCK;
2851 		}
2852 	}
2853 	return (error);
2854 }
2855 
2856 /*
2857  * NFS advisory byte-level locks.
2858  */
2859 static int
2860 nfs_advlockasync(struct vop_advlockasync_args *ap)
2861 {
2862 	struct vnode *vp = ap->a_vp;
2863 	u_quad_t size;
2864 	int error;
2865 
2866 	if (NFS_ISV4(vp))
2867 		return (EOPNOTSUPP);
2868 	error = vn_lock(vp, LK_SHARED);
2869 	if (error)
2870 		return (error);
2871 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2872 		size = VTONFS(vp)->n_size;
2873 		VOP_UNLOCK(vp, 0);
2874 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
2875 	} else {
2876 		VOP_UNLOCK(vp, 0);
2877 		error = EOPNOTSUPP;
2878 	}
2879 	return (error);
2880 }
2881 
2882 /*
2883  * Print out the contents of an nfsnode.
2884  */
2885 static int
2886 nfs_print(struct vop_print_args *ap)
2887 {
2888 	struct vnode *vp = ap->a_vp;
2889 	struct nfsnode *np = VTONFS(vp);
2890 
2891 	ncl_printf("\tfileid %ld fsid 0x%x",
2892 	   np->n_vattr.na_fileid, np->n_vattr.na_fsid);
2893 	if (vp->v_type == VFIFO)
2894 		fifo_printinfo(vp);
2895 	printf("\n");
2896 	return (0);
2897 }
2898 
2899 /*
2900  * This is the "real" nfs::bwrite(struct buf*).
2901  * We set B_CACHE if this is a VMIO buffer.
2902  */
2903 int
2904 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
2905 {
2906 	int s;
2907 	int oldflags = bp->b_flags;
2908 #if 0
2909 	int retv = 1;
2910 	off_t off;
2911 #endif
2912 
2913 	BUF_ASSERT_HELD(bp);
2914 
2915 	if (bp->b_flags & B_INVAL) {
2916 		brelse(bp);
2917 		return(0);
2918 	}
2919 
2920 	bp->b_flags |= B_CACHE;
2921 
2922 	/*
2923 	 * Undirty the bp.  We will redirty it later if the I/O fails.
2924 	 */
2925 
2926 	s = splbio();
2927 	bundirty(bp);
2928 	bp->b_flags &= ~B_DONE;
2929 	bp->b_ioflags &= ~BIO_ERROR;
2930 	bp->b_iocmd = BIO_WRITE;
2931 
2932 	bufobj_wref(bp->b_bufobj);
2933 	curthread->td_ru.ru_oublock++;
2934 	splx(s);
2935 
2936 	/*
2937 	 * Note: to avoid loopback deadlocks, we do not
2938 	 * assign b_runningbufspace.
2939 	 */
2940 	vfs_busy_pages(bp, 1);
2941 
2942 	BUF_KERNPROC(bp);
2943 	bp->b_iooffset = dbtob(bp->b_blkno);
2944 	bstrategy(bp);
2945 
2946 	if( (oldflags & B_ASYNC) == 0) {
2947 		int rtval = bufwait(bp);
2948 
2949 		if (oldflags & B_DELWRI) {
2950 			s = splbio();
2951 			reassignbuf(bp);
2952 			splx(s);
2953 		}
2954 		brelse(bp);
2955 		return (rtval);
2956 	}
2957 
2958 	return (0);
2959 }
2960 
2961 /*
2962  * nfs special file access vnode op.
2963  * Essentially just get vattr and then imitate iaccess() since the device is
2964  * local to the client.
2965  */
2966 static int
2967 nfsspec_access(struct vop_access_args *ap)
2968 {
2969 	struct vattr *vap;
2970 	struct ucred *cred = ap->a_cred;
2971 	struct vnode *vp = ap->a_vp;
2972 	accmode_t accmode = ap->a_accmode;
2973 	struct vattr vattr;
2974 	int error;
2975 
2976 	/*
2977 	 * Disallow write attempts on filesystems mounted read-only;
2978 	 * unless the file is a socket, fifo, or a block or character
2979 	 * device resident on the filesystem.
2980 	 */
2981 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
2982 		switch (vp->v_type) {
2983 		case VREG:
2984 		case VDIR:
2985 		case VLNK:
2986 			return (EROFS);
2987 		default:
2988 			break;
2989 		}
2990 	}
2991 	vap = &vattr;
2992 	error = VOP_GETATTR(vp, vap, cred);
2993 	if (error)
2994 		goto out;
2995 	error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
2996 	    accmode, cred, NULL);
2997 out:
2998 	return error;
2999 }
3000 
3001 /*
3002  * Read wrapper for fifos.
3003  */
3004 static int
3005 nfsfifo_read(struct vop_read_args *ap)
3006 {
3007 	struct nfsnode *np = VTONFS(ap->a_vp);
3008 	int error;
3009 
3010 	/*
3011 	 * Set access flag.
3012 	 */
3013 	mtx_lock(&np->n_mtx);
3014 	np->n_flag |= NACC;
3015 	getnanotime(&np->n_atim);
3016 	mtx_unlock(&np->n_mtx);
3017 	error = fifo_specops.vop_read(ap);
3018 	return error;
3019 }
3020 
3021 /*
3022  * Write wrapper for fifos.
3023  */
3024 static int
3025 nfsfifo_write(struct vop_write_args *ap)
3026 {
3027 	struct nfsnode *np = VTONFS(ap->a_vp);
3028 
3029 	/*
3030 	 * Set update flag.
3031 	 */
3032 	mtx_lock(&np->n_mtx);
3033 	np->n_flag |= NUPD;
3034 	getnanotime(&np->n_mtim);
3035 	mtx_unlock(&np->n_mtx);
3036 	return(fifo_specops.vop_write(ap));
3037 }
3038 
3039 /*
3040  * Close wrapper for fifos.
3041  *
3042  * Update the times on the nfsnode then do fifo close.
3043  */
3044 static int
3045 nfsfifo_close(struct vop_close_args *ap)
3046 {
3047 	struct vnode *vp = ap->a_vp;
3048 	struct nfsnode *np = VTONFS(vp);
3049 	struct vattr vattr;
3050 	struct timespec ts;
3051 
3052 	mtx_lock(&np->n_mtx);
3053 	if (np->n_flag & (NACC | NUPD)) {
3054 		getnanotime(&ts);
3055 		if (np->n_flag & NACC)
3056 			np->n_atim = ts;
3057 		if (np->n_flag & NUPD)
3058 			np->n_mtim = ts;
3059 		np->n_flag |= NCHG;
3060 		if (vrefcnt(vp) == 1 &&
3061 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3062 			VATTR_NULL(&vattr);
3063 			if (np->n_flag & NACC)
3064 				vattr.va_atime = np->n_atim;
3065 			if (np->n_flag & NUPD)
3066 				vattr.va_mtime = np->n_mtim;
3067 			mtx_unlock(&np->n_mtx);
3068 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3069 			goto out;
3070 		}
3071 	}
3072 	mtx_unlock(&np->n_mtx);
3073 out:
3074 	return (fifo_specops.vop_close(ap));
3075 }
3076 
3077 /*
3078  * Just call ncl_writebp() with the force argument set to 1.
3079  *
3080  * NOTE: B_DONE may or may not be set in a_bp on call.
3081  */
3082 static int
3083 nfs_bwrite(struct buf *bp)
3084 {
3085 
3086 	return (ncl_writebp(bp, 1, curthread));
3087 }
3088 
3089 struct buf_ops buf_ops_newnfs = {
3090 	.bop_name	=	"buf_ops_nfs",
3091 	.bop_write	=	nfs_bwrite,
3092 	.bop_strategy	=	bufstrategy,
3093 	.bop_sync	=	bufsync,
3094 	.bop_bdflush	=	bufbdflush,
3095 };
3096 
3097 /*
3098  * Cloned from vop_stdlock(), and then the ugly hack added.
3099  */
3100 static int
3101 nfs_lock1(struct vop_lock1_args *ap)
3102 {
3103 	struct vnode *vp = ap->a_vp;
3104 	int error = 0;
3105 
3106 	/*
3107 	 * Since vfs_hash_get() calls vget() and it will no longer work
3108 	 * for FreeBSD8 with flags == 0, I can only think of this horrible
3109 	 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3110 	 * and then handle it here. All I want for this case is a v_usecount
3111 	 * on the vnode to use for recovery, while another thread might
3112 	 * hold a lock on the vnode. I have the other threads blocked, so
3113 	 * there isn't any race problem.
3114 	 */
3115 	if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3116 		if ((ap->a_flags & LK_INTERLOCK) == 0)
3117 			panic("ncllock1");
3118 		if ((vp->v_iflag & VI_DOOMED))
3119 			error = ENOENT;
3120 		VI_UNLOCK(vp);
3121 		return (error);
3122 	}
3123 	return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3124 	    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3125 	    ap->a_line));
3126 }
3127 
3128 #ifdef NFS4_ACL_EXTATTR_NAME
3129 static int
3130 nfs_getacl(struct vop_getacl_args *ap)
3131 {
3132 	int error;
3133 
3134 	if (ap->a_type != ACL_TYPE_NFS4)
3135 		return (EOPNOTSUPP);
3136 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3137 	    NULL);
3138 	if (error > NFSERR_STALE) {
3139 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3140 		error = EPERM;
3141 	}
3142 	return (error);
3143 }
3144 
3145 static int
3146 nfs_setacl(struct vop_setacl_args *ap)
3147 {
3148 	int error;
3149 
3150 	if (ap->a_type != ACL_TYPE_NFS4)
3151 		return (EOPNOTSUPP);
3152 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3153 	    NULL);
3154 	if (error > NFSERR_STALE) {
3155 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3156 		error = EPERM;
3157 	}
3158 	return (error);
3159 }
3160 
3161 #endif	/* NFS4_ACL_EXTATTR_NAME */
3162