xref: /netbsd-src/sys/fs/nfs/client/nfs_clvnops.c (revision 946379e7b37692fc43f68eb0d1c10daa0a7f3b6c)
1 /*	$NetBSD: nfs_clvnops.c,v 1.2 2014/03/25 16:30:28 christos Exp $	*/
2 /*-
3  * Copyright (c) 1989, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * Rick Macklem at The University of Guelph.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	from nfs_vnops.c	8.16 (Berkeley) 5/27/95
34  */
35 
36 #include <sys/cdefs.h>
37 /* __FBSDID("FreeBSD: head/sys/fs/nfsclient/nfs_clvnops.c 252072 2013-06-21 22:26:18Z rmacklem "); */
38 __RCSID("$NetBSD: nfs_clvnops.c,v 1.2 2014/03/25 16:30:28 christos Exp $");
39 
40 /*
41  * vnode op calls for Sun NFS version 2, 3 and 4
42  */
43 
44 #include "opt_kdtrace.h"
45 #include "opt_inet.h"
46 
47 #include <sys/param.h>
48 #include <sys/kernel.h>
49 #include <sys/systm.h>
50 #include <sys/resourcevar.h>
51 #include <sys/proc.h>
52 #include <sys/mount.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/jail.h>
56 #include <sys/malloc.h>
57 #include <sys/mbuf.h>
58 #include <sys/namei.h>
59 #include <sys/socket.h>
60 #include <sys/vnode.h>
61 #include <sys/dirent.h>
62 #include <sys/fcntl.h>
63 #include <sys/lockf.h>
64 #include <sys/stat.h>
65 #include <sys/sysctl.h>
66 #include <sys/signalvar.h>
67 
68 #include <vm/vm.h>
69 #include <vm/vm_extern.h>
70 #include <vm/vm_object.h>
71 
72 #include <fs/nfs/nfsport.h>
73 #include <fs/nfsclient/nfsnode.h>
74 #include <fs/nfsclient/nfsmount.h>
75 #include <fs/nfsclient/nfs.h>
76 #include <fs/nfsclient/nfs_kdtrace.h>
77 
78 #include <net/if.h>
79 #include <netinet/in.h>
80 #include <netinet/in_var.h>
81 
82 #include <nfs/nfs_lock.h>
83 
84 #ifdef KDTRACE_HOOKS
85 #include <sys/dtrace_bsd.h>
86 
87 dtrace_nfsclient_accesscache_flush_probe_func_t
88 		dtrace_nfscl_accesscache_flush_done_probe;
89 uint32_t	nfscl_accesscache_flush_done_id;
90 
91 dtrace_nfsclient_accesscache_get_probe_func_t
92 		dtrace_nfscl_accesscache_get_hit_probe,
93 		dtrace_nfscl_accesscache_get_miss_probe;
94 uint32_t	nfscl_accesscache_get_hit_id;
95 uint32_t	nfscl_accesscache_get_miss_id;
96 
97 dtrace_nfsclient_accesscache_load_probe_func_t
98 		dtrace_nfscl_accesscache_load_done_probe;
99 uint32_t	nfscl_accesscache_load_done_id;
100 #endif /* !KDTRACE_HOOKS */
101 
102 /* Defs */
103 #define	TRUE	1
104 #define	FALSE	0
105 
106 extern struct nfsstats newnfsstats;
107 extern int nfsrv_useacl;
108 extern int nfscl_debuglevel;
109 MALLOC_DECLARE(M_NEWNFSREQ);
110 
111 /*
112  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
113  * calls are not in getblk() and brelse() so that they would not be necessary
114  * here.
115  */
116 #ifndef B_VMIO
117 #define	vfs_busy_pages(bp, f)
118 #endif
119 
120 static vop_read_t	nfsfifo_read;
121 static vop_write_t	nfsfifo_write;
122 static vop_close_t	nfsfifo_close;
123 static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
124 		    struct thread *);
125 static vop_lookup_t	nfs_lookup;
126 static vop_create_t	nfs_create;
127 static vop_mknod_t	nfs_mknod;
128 static vop_open_t	nfs_open;
129 static vop_pathconf_t	nfs_pathconf;
130 static vop_close_t	nfs_close;
131 static vop_access_t	nfs_access;
132 static vop_getattr_t	nfs_getattr;
133 static vop_setattr_t	nfs_setattr;
134 static vop_read_t	nfs_read;
135 static vop_fsync_t	nfs_fsync;
136 static vop_remove_t	nfs_remove;
137 static vop_link_t	nfs_link;
138 static vop_rename_t	nfs_rename;
139 static vop_mkdir_t	nfs_mkdir;
140 static vop_rmdir_t	nfs_rmdir;
141 static vop_symlink_t	nfs_symlink;
142 static vop_readdir_t	nfs_readdir;
143 static vop_strategy_t	nfs_strategy;
144 static vop_lock1_t	nfs_lock1;
145 static	int	nfs_lookitup(struct vnode *, char *, int,
146 		    struct ucred *, struct thread *, struct nfsnode **);
147 static	int	nfs_sillyrename(struct vnode *, struct vnode *,
148 		    struct componentname *);
149 static vop_access_t	nfsspec_access;
150 static vop_readlink_t	nfs_readlink;
151 static vop_print_t	nfs_print;
152 static vop_advlock_t	nfs_advlock;
153 static vop_advlockasync_t nfs_advlockasync;
154 static vop_getacl_t nfs_getacl;
155 static vop_setacl_t nfs_setacl;
156 
157 /*
158  * Global vfs data structures for nfs
159  */
160 struct vop_vector newnfs_vnodeops = {
161 	.vop_default =		&default_vnodeops,
162 	.vop_access =		nfs_access,
163 	.vop_advlock =		nfs_advlock,
164 	.vop_advlockasync =	nfs_advlockasync,
165 	.vop_close =		nfs_close,
166 	.vop_create =		nfs_create,
167 	.vop_fsync =		nfs_fsync,
168 	.vop_getattr =		nfs_getattr,
169 	.vop_getpages =		ncl_getpages,
170 	.vop_putpages =		ncl_putpages,
171 	.vop_inactive =		ncl_inactive,
172 	.vop_link =		nfs_link,
173 	.vop_lock1 = 		nfs_lock1,
174 	.vop_lookup =		nfs_lookup,
175 	.vop_mkdir =		nfs_mkdir,
176 	.vop_mknod =		nfs_mknod,
177 	.vop_open =		nfs_open,
178 	.vop_pathconf =		nfs_pathconf,
179 	.vop_print =		nfs_print,
180 	.vop_read =		nfs_read,
181 	.vop_readdir =		nfs_readdir,
182 	.vop_readlink =		nfs_readlink,
183 	.vop_reclaim =		ncl_reclaim,
184 	.vop_remove =		nfs_remove,
185 	.vop_rename =		nfs_rename,
186 	.vop_rmdir =		nfs_rmdir,
187 	.vop_setattr =		nfs_setattr,
188 	.vop_strategy =		nfs_strategy,
189 	.vop_symlink =		nfs_symlink,
190 	.vop_write =		ncl_write,
191 	.vop_getacl =		nfs_getacl,
192 	.vop_setacl =		nfs_setacl,
193 };
194 
195 struct vop_vector newnfs_fifoops = {
196 	.vop_default =		&fifo_specops,
197 	.vop_access =		nfsspec_access,
198 	.vop_close =		nfsfifo_close,
199 	.vop_fsync =		nfs_fsync,
200 	.vop_getattr =		nfs_getattr,
201 	.vop_inactive =		ncl_inactive,
202 	.vop_print =		nfs_print,
203 	.vop_read =		nfsfifo_read,
204 	.vop_reclaim =		ncl_reclaim,
205 	.vop_setattr =		nfs_setattr,
206 	.vop_write =		nfsfifo_write,
207 };
208 
209 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
210     struct componentname *cnp, struct vattr *vap);
211 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
212     int namelen, struct ucred *cred, struct thread *td);
213 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
214     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
215     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
216 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
217     struct componentname *scnp, struct sillyrename *sp);
218 
219 /*
220  * Global variables
221  */
222 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
223 
224 SYSCTL_DECL(_vfs_nfs);
225 
226 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
227 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
228 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
229 
230 static int	nfs_prime_access_cache = 0;
231 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
232 	   &nfs_prime_access_cache, 0,
233 	   "Prime NFS ACCESS cache when fetching attributes");
234 
235 static int	newnfs_commit_on_close = 0;
236 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
237     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
238 
239 static int	nfs_clean_pages_on_close = 1;
240 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
241 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
242 
243 int newnfs_directio_enable = 0;
244 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
245 	   &newnfs_directio_enable, 0, "Enable NFS directio");
246 
247 int nfs_keep_dirty_on_error;
248 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
249     &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
250 
251 /*
252  * This sysctl allows other processes to mmap a file that has been opened
253  * O_DIRECT by a process.  In general, having processes mmap the file while
254  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
255  * this by default to prevent DoS attacks - to prevent a malicious user from
256  * opening up files O_DIRECT preventing other users from mmap'ing these
257  * files.  "Protected" environments where stricter consistency guarantees are
258  * required can disable this knob.  The process that opened the file O_DIRECT
259  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
260  * meaningful.
261  */
262 int newnfs_directio_allow_mmap = 1;
263 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
264 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
265 
266 #if 0
267 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
268 	   &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
269 
270 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
271 	   &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
272 #endif
273 
274 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
275 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
276 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
277 
278 /*
279  * SMP Locking Note :
280  * The list of locks after the description of the lock is the ordering
281  * of other locks acquired with the lock held.
282  * np->n_mtx : Protects the fields in the nfsnode.
283        VM Object Lock
284        VI_MTX (acquired indirectly)
285  * nmp->nm_mtx : Protects the fields in the nfsmount.
286        rep->r_mtx
287  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
288  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
289        nmp->nm_mtx
290        rep->r_mtx
291  * rep->r_mtx : Protects the fields in an nfsreq.
292  */
293 
294 static int
295 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
296     struct ucred *cred, u_int32_t *retmode)
297 {
298 	int error = 0, attrflag, i, lrupos;
299 	u_int32_t rmode;
300 	struct nfsnode *np = VTONFS(vp);
301 	struct nfsvattr nfsva;
302 
303 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
304 	    &rmode, NULL);
305 	if (attrflag)
306 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
307 	if (!error) {
308 		lrupos = 0;
309 		mtx_lock(&np->n_mtx);
310 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
311 			if (np->n_accesscache[i].uid == cred->cr_uid) {
312 				np->n_accesscache[i].mode = rmode;
313 				np->n_accesscache[i].stamp = time_second;
314 				break;
315 			}
316 			if (i > 0 && np->n_accesscache[i].stamp <
317 			    np->n_accesscache[lrupos].stamp)
318 				lrupos = i;
319 		}
320 		if (i == NFS_ACCESSCACHESIZE) {
321 			np->n_accesscache[lrupos].uid = cred->cr_uid;
322 			np->n_accesscache[lrupos].mode = rmode;
323 			np->n_accesscache[lrupos].stamp = time_second;
324 		}
325 		mtx_unlock(&np->n_mtx);
326 		if (retmode != NULL)
327 			*retmode = rmode;
328 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
329 	} else if (NFS_ISV4(vp)) {
330 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
331 	}
332 #ifdef KDTRACE_HOOKS
333 	if (error != 0)
334 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
335 		    error);
336 #endif
337 	return (error);
338 }
339 
340 /*
341  * nfs access vnode op.
342  * For nfs version 2, just return ok. File accesses may fail later.
343  * For nfs version 3, use the access rpc to check accessibility. If file modes
344  * are changed on the server, accesses might still fail later.
345  */
346 static int
347 nfs_access(struct vop_access_args *ap)
348 {
349 	struct vnode *vp = ap->a_vp;
350 	int error = 0, i, gotahit;
351 	u_int32_t mode, wmode, rmode;
352 	int v34 = NFS_ISV34(vp);
353 	struct nfsnode *np = VTONFS(vp);
354 
355 	/*
356 	 * Disallow write attempts on filesystems mounted read-only;
357 	 * unless the file is a socket, fifo, or a block or character
358 	 * device resident on the filesystem.
359 	 */
360 	if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
361 	    VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
362 	    VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
363 		switch (vp->v_type) {
364 		case VREG:
365 		case VDIR:
366 		case VLNK:
367 			return (EROFS);
368 		default:
369 			break;
370 		}
371 	}
372 	/*
373 	 * For nfs v3 or v4, check to see if we have done this recently, and if
374 	 * so return our cached result instead of making an ACCESS call.
375 	 * If not, do an access rpc, otherwise you are stuck emulating
376 	 * ufs_access() locally using the vattr. This may not be correct,
377 	 * since the server may apply other access criteria such as
378 	 * client uid-->server uid mapping that we do not know about.
379 	 */
380 	if (v34) {
381 		if (ap->a_accmode & VREAD)
382 			mode = NFSACCESS_READ;
383 		else
384 			mode = 0;
385 		if (vp->v_type != VDIR) {
386 			if (ap->a_accmode & VWRITE)
387 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
388 			if (ap->a_accmode & VAPPEND)
389 				mode |= NFSACCESS_EXTEND;
390 			if (ap->a_accmode & VEXEC)
391 				mode |= NFSACCESS_EXECUTE;
392 			if (ap->a_accmode & VDELETE)
393 				mode |= NFSACCESS_DELETE;
394 		} else {
395 			if (ap->a_accmode & VWRITE)
396 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
397 			if (ap->a_accmode & VAPPEND)
398 				mode |= NFSACCESS_EXTEND;
399 			if (ap->a_accmode & VEXEC)
400 				mode |= NFSACCESS_LOOKUP;
401 			if (ap->a_accmode & VDELETE)
402 				mode |= NFSACCESS_DELETE;
403 			if (ap->a_accmode & VDELETE_CHILD)
404 				mode |= NFSACCESS_MODIFY;
405 		}
406 		/* XXX safety belt, only make blanket request if caching */
407 		if (nfsaccess_cache_timeout > 0) {
408 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
409 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
410 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
411 		} else {
412 			wmode = mode;
413 		}
414 
415 		/*
416 		 * Does our cached result allow us to give a definite yes to
417 		 * this request?
418 		 */
419 		gotahit = 0;
420 		mtx_lock(&np->n_mtx);
421 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
422 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
423 			    if (time_second < (np->n_accesscache[i].stamp
424 				+ nfsaccess_cache_timeout) &&
425 				(np->n_accesscache[i].mode & mode) == mode) {
426 				NFSINCRGLOBAL(newnfsstats.accesscache_hits);
427 				gotahit = 1;
428 			    }
429 			    break;
430 			}
431 		}
432 		mtx_unlock(&np->n_mtx);
433 #ifdef KDTRACE_HOOKS
434 		if (gotahit != 0)
435 			KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
436 			    ap->a_cred->cr_uid, mode);
437 		else
438 			KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
439 			    ap->a_cred->cr_uid, mode);
440 #endif
441 		if (gotahit == 0) {
442 			/*
443 			 * Either a no, or a don't know.  Go to the wire.
444 			 */
445 			NFSINCRGLOBAL(newnfsstats.accesscache_misses);
446 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
447 			    ap->a_cred, &rmode);
448 			if (!error &&
449 			    (rmode & mode) != mode)
450 				error = EACCES;
451 		}
452 		return (error);
453 	} else {
454 		if ((error = nfsspec_access(ap)) != 0) {
455 			return (error);
456 		}
457 		/*
458 		 * Attempt to prevent a mapped root from accessing a file
459 		 * which it shouldn't.  We try to read a byte from the file
460 		 * if the user is root and the file is not zero length.
461 		 * After calling nfsspec_access, we should have the correct
462 		 * file size cached.
463 		 */
464 		mtx_lock(&np->n_mtx);
465 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
466 		    && VTONFS(vp)->n_size > 0) {
467 			struct iovec aiov;
468 			struct uio auio;
469 			char buf[1];
470 
471 			mtx_unlock(&np->n_mtx);
472 			aiov.iov_base = buf;
473 			aiov.iov_len = 1;
474 			auio.uio_iov = &aiov;
475 			auio.uio_iovcnt = 1;
476 			auio.uio_offset = 0;
477 			auio.uio_resid = 1;
478 			auio.uio_segflg = UIO_SYSSPACE;
479 			auio.uio_rw = UIO_READ;
480 			auio.uio_td = ap->a_td;
481 
482 			if (vp->v_type == VREG)
483 				error = ncl_readrpc(vp, &auio, ap->a_cred);
484 			else if (vp->v_type == VDIR) {
485 				char* bp;
486 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
487 				aiov.iov_base = bp;
488 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
489 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
490 				    ap->a_td);
491 				free(bp, M_TEMP);
492 			} else if (vp->v_type == VLNK)
493 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
494 			else
495 				error = EACCES;
496 		} else
497 			mtx_unlock(&np->n_mtx);
498 		return (error);
499 	}
500 }
501 
502 
503 /*
504  * nfs open vnode op
505  * Check to see if the type is ok
506  * and that deletion is not in progress.
507  * For paged in text files, you will need to flush the page cache
508  * if consistency is lost.
509  */
510 /* ARGSUSED */
511 static int
512 nfs_open(struct vop_open_args *ap)
513 {
514 	struct vnode *vp = ap->a_vp;
515 	struct nfsnode *np = VTONFS(vp);
516 	struct vattr vattr;
517 	int error;
518 	int fmode = ap->a_mode;
519 	struct ucred *cred;
520 
521 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
522 		return (EOPNOTSUPP);
523 
524 	/*
525 	 * For NFSv4, we need to do the Open Op before cache validation,
526 	 * so that we conform to RFC3530 Sec. 9.3.1.
527 	 */
528 	if (NFS_ISV4(vp)) {
529 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
530 		if (error) {
531 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
532 			    (gid_t)0);
533 			return (error);
534 		}
535 	}
536 
537 	/*
538 	 * Now, if this Open will be doing reading, re-validate/flush the
539 	 * cache, so that Close/Open coherency is maintained.
540 	 */
541 	mtx_lock(&np->n_mtx);
542 	if (np->n_flag & NMODIFIED) {
543 		mtx_unlock(&np->n_mtx);
544 		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
545 		if (error == EINTR || error == EIO) {
546 			if (NFS_ISV4(vp))
547 				(void) nfsrpc_close(vp, 0, ap->a_td);
548 			return (error);
549 		}
550 		mtx_lock(&np->n_mtx);
551 		np->n_attrstamp = 0;
552 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
553 		if (vp->v_type == VDIR)
554 			np->n_direofoffset = 0;
555 		mtx_unlock(&np->n_mtx);
556 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
557 		if (error) {
558 			if (NFS_ISV4(vp))
559 				(void) nfsrpc_close(vp, 0, ap->a_td);
560 			return (error);
561 		}
562 		mtx_lock(&np->n_mtx);
563 		np->n_mtime = vattr.va_mtime;
564 		if (NFS_ISV4(vp))
565 			np->n_change = vattr.va_filerev;
566 	} else {
567 		mtx_unlock(&np->n_mtx);
568 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
569 		if (error) {
570 			if (NFS_ISV4(vp))
571 				(void) nfsrpc_close(vp, 0, ap->a_td);
572 			return (error);
573 		}
574 		mtx_lock(&np->n_mtx);
575 		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
576 		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
577 			if (vp->v_type == VDIR)
578 				np->n_direofoffset = 0;
579 			mtx_unlock(&np->n_mtx);
580 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
581 			if (error == EINTR || error == EIO) {
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_mtime = vattr.va_mtime;
588 			if (NFS_ISV4(vp))
589 				np->n_change = vattr.va_filerev;
590 		}
591 	}
592 
593 	/*
594 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
595 	 */
596 	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
597 	    (vp->v_type == VREG)) {
598 		if (np->n_directio_opens == 0) {
599 			mtx_unlock(&np->n_mtx);
600 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
601 			if (error) {
602 				if (NFS_ISV4(vp))
603 					(void) nfsrpc_close(vp, 0, ap->a_td);
604 				return (error);
605 			}
606 			mtx_lock(&np->n_mtx);
607 			np->n_flag |= NNONCACHE;
608 		}
609 		np->n_directio_opens++;
610 	}
611 
612 	/* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
613 	if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
614 		np->n_flag |= NWRITEOPENED;
615 
616 	/*
617 	 * If this is an open for writing, capture a reference to the
618 	 * credentials, so they can be used by ncl_putpages(). Using
619 	 * these write credentials is preferable to the credentials of
620 	 * whatever thread happens to be doing the VOP_PUTPAGES() since
621 	 * the write RPCs are less likely to fail with EACCES.
622 	 */
623 	if ((fmode & FWRITE) != 0) {
624 		cred = np->n_writecred;
625 		np->n_writecred = crhold(ap->a_cred);
626 	} else
627 		cred = NULL;
628 	mtx_unlock(&np->n_mtx);
629 
630 	if (cred != NULL)
631 		crfree(cred);
632 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
633 	return (0);
634 }
635 
636 /*
637  * nfs close vnode op
638  * What an NFS client should do upon close after writing is a debatable issue.
639  * Most NFS clients push delayed writes to the server upon close, basically for
640  * two reasons:
641  * 1 - So that any write errors may be reported back to the client process
642  *     doing the close system call. By far the two most likely errors are
643  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
644  * 2 - To put a worst case upper bound on cache inconsistency between
645  *     multiple clients for the file.
646  * There is also a consistency problem for Version 2 of the protocol w.r.t.
647  * not being able to tell if other clients are writing a file concurrently,
648  * since there is no way of knowing if the changed modify time in the reply
649  * is only due to the write for this client.
650  * (NFS Version 3 provides weak cache consistency data in the reply that
651  *  should be sufficient to detect and handle this case.)
652  *
653  * The current code does the following:
654  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
655  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
656  *                     or commit them (this satisfies 1 and 2 except for the
657  *                     case where the server crashes after this close but
658  *                     before the commit RPC, which is felt to be "good
659  *                     enough". Changing the last argument to ncl_flush() to
660  *                     a 1 would force a commit operation, if it is felt a
661  *                     commit is necessary now.
662  * for NFS Version 4 - flush the dirty buffers and commit them, if
663  *		       nfscl_mustflush() says this is necessary.
664  *                     It is necessary if there is no write delegation held,
665  *                     in order to satisfy open/close coherency.
666  *                     If the file isn't cached on local stable storage,
667  *                     it may be necessary in order to detect "out of space"
668  *                     errors from the server, if the write delegation
669  *                     issued by the server doesn't allow the file to grow.
670  */
671 /* ARGSUSED */
672 static int
673 nfs_close(struct vop_close_args *ap)
674 {
675 	struct vnode *vp = ap->a_vp;
676 	struct nfsnode *np = VTONFS(vp);
677 	struct nfsvattr nfsva;
678 	struct ucred *cred;
679 	int error = 0, ret, localcred = 0;
680 	int fmode = ap->a_fflag;
681 
682 	if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
683 		return (0);
684 	/*
685 	 * During shutdown, a_cred isn't valid, so just use root.
686 	 */
687 	if (ap->a_cred == NOCRED) {
688 		cred = newnfs_getcred();
689 		localcred = 1;
690 	} else {
691 		cred = ap->a_cred;
692 	}
693 	if (vp->v_type == VREG) {
694 	    /*
695 	     * Examine and clean dirty pages, regardless of NMODIFIED.
696 	     * This closes a major hole in close-to-open consistency.
697 	     * We want to push out all dirty pages (and buffers) on
698 	     * close, regardless of whether they were dirtied by
699 	     * mmap'ed writes or via write().
700 	     */
701 	    if (nfs_clean_pages_on_close && vp->v_object) {
702 		VM_OBJECT_WLOCK(vp->v_object);
703 		vm_object_page_clean(vp->v_object, 0, 0, 0);
704 		VM_OBJECT_WUNLOCK(vp->v_object);
705 	    }
706 	    mtx_lock(&np->n_mtx);
707 	    if (np->n_flag & NMODIFIED) {
708 		mtx_unlock(&np->n_mtx);
709 		if (NFS_ISV3(vp)) {
710 		    /*
711 		     * Under NFSv3 we have dirty buffers to dispose of.  We
712 		     * must flush them to the NFS server.  We have the option
713 		     * of waiting all the way through the commit rpc or just
714 		     * waiting for the initial write.  The default is to only
715 		     * wait through the initial write so the data is in the
716 		     * server's cache, which is roughly similar to the state
717 		     * a standard disk subsystem leaves the file in on close().
718 		     *
719 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
720 		     * potential races with other processes, and certainly
721 		     * cannot clear it if we don't commit.
722 		     * These races occur when there is no longer the old
723 		     * traditional vnode locking implemented for Vnode Ops.
724 		     */
725 		    int cm = newnfs_commit_on_close ? 1 : 0;
726 		    error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
727 		    /* np->n_flag &= ~NMODIFIED; */
728 		} else if (NFS_ISV4(vp)) {
729 			if (nfscl_mustflush(vp) != 0) {
730 				int cm = newnfs_commit_on_close ? 1 : 0;
731 				error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
732 				    cm, 0);
733 				/*
734 				 * as above w.r.t races when clearing
735 				 * NMODIFIED.
736 				 * np->n_flag &= ~NMODIFIED;
737 				 */
738 			}
739 		} else
740 		    error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
741 		mtx_lock(&np->n_mtx);
742 	    }
743  	    /*
744  	     * Invalidate the attribute cache in all cases.
745  	     * An open is going to fetch fresh attrs any way, other procs
746  	     * on this node that have file open will be forced to do an
747  	     * otw attr fetch, but this is safe.
748 	     * --> A user found that their RPC count dropped by 20% when
749 	     *     this was commented out and I can't see any requirement
750 	     *     for it, so I've disabled it when negative lookups are
751 	     *     enabled. (What does this have to do with negative lookup
752 	     *     caching? Well nothing, except it was reported by the
753 	     *     same user that needed negative lookup caching and I wanted
754 	     *     there to be a way to disable it to see if it
755 	     *     is the cause of some caching/coherency issue that might
756 	     *     crop up.)
757  	     */
758 	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
759 		    np->n_attrstamp = 0;
760 		    KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
761 	    }
762 	    if (np->n_flag & NWRITEERR) {
763 		np->n_flag &= ~NWRITEERR;
764 		error = np->n_error;
765 	    }
766 	    mtx_unlock(&np->n_mtx);
767 	}
768 
769 	if (NFS_ISV4(vp)) {
770 		/*
771 		 * Get attributes so "change" is up to date.
772 		 */
773 		if (error == 0 && nfscl_mustflush(vp) != 0) {
774 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
775 			    NULL);
776 			if (!ret) {
777 				np->n_change = nfsva.na_filerev;
778 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
779 				    NULL, 0, 0);
780 			}
781 		}
782 
783 		/*
784 		 * and do the close.
785 		 */
786 		ret = nfsrpc_close(vp, 0, ap->a_td);
787 		if (!error && ret)
788 			error = ret;
789 		if (error)
790 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
791 			    (gid_t)0);
792 	}
793 	if (newnfs_directio_enable)
794 		KASSERT((np->n_directio_asyncwr == 0),
795 			("nfs_close: dirty unflushed (%d) directio buffers\n",
796 			 np->n_directio_asyncwr));
797 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
798 		mtx_lock(&np->n_mtx);
799 		KASSERT((np->n_directio_opens > 0),
800 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
801 		np->n_directio_opens--;
802 		if (np->n_directio_opens == 0)
803 			np->n_flag &= ~NNONCACHE;
804 		mtx_unlock(&np->n_mtx);
805 	}
806 	if (localcred)
807 		NFSFREECRED(cred);
808 	return (error);
809 }
810 
811 /*
812  * nfs getattr call from vfs.
813  */
814 static int
815 nfs_getattr(struct vop_getattr_args *ap)
816 {
817 	struct vnode *vp = ap->a_vp;
818 	struct thread *td = curthread;	/* XXX */
819 	struct nfsnode *np = VTONFS(vp);
820 	int error = 0;
821 	struct nfsvattr nfsva;
822 	struct vattr *vap = ap->a_vap;
823 	struct vattr vattr;
824 
825 	/*
826 	 * Update local times for special files.
827 	 */
828 	mtx_lock(&np->n_mtx);
829 	if (np->n_flag & (NACC | NUPD))
830 		np->n_flag |= NCHG;
831 	mtx_unlock(&np->n_mtx);
832 	/*
833 	 * First look in the cache.
834 	 */
835 	if (ncl_getattrcache(vp, &vattr) == 0) {
836 		vap->va_type = vattr.va_type;
837 		vap->va_mode = vattr.va_mode;
838 		vap->va_nlink = vattr.va_nlink;
839 		vap->va_uid = vattr.va_uid;
840 		vap->va_gid = vattr.va_gid;
841 		vap->va_fsid = vattr.va_fsid;
842 		vap->va_fileid = vattr.va_fileid;
843 		vap->va_size = vattr.va_size;
844 		vap->va_blocksize = vattr.va_blocksize;
845 		vap->va_atime = vattr.va_atime;
846 		vap->va_mtime = vattr.va_mtime;
847 		vap->va_ctime = vattr.va_ctime;
848 		vap->va_gen = vattr.va_gen;
849 		vap->va_flags = vattr.va_flags;
850 		vap->va_rdev = vattr.va_rdev;
851 		vap->va_bytes = vattr.va_bytes;
852 		vap->va_filerev = vattr.va_filerev;
853 		/*
854 		 * Get the local modify time for the case of a write
855 		 * delegation.
856 		 */
857 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
858 		return (0);
859 	}
860 
861 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
862 	    nfsaccess_cache_timeout > 0) {
863 		NFSINCRGLOBAL(newnfsstats.accesscache_misses);
864 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
865 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
866 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
867 			return (0);
868 		}
869 	}
870 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
871 	if (!error)
872 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
873 	if (!error) {
874 		/*
875 		 * Get the local modify time for the case of a write
876 		 * delegation.
877 		 */
878 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
879 	} else if (NFS_ISV4(vp)) {
880 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
881 	}
882 	return (error);
883 }
884 
885 /*
886  * nfs setattr call.
887  */
888 static int
889 nfs_setattr(struct vop_setattr_args *ap)
890 {
891 	struct vnode *vp = ap->a_vp;
892 	struct nfsnode *np = VTONFS(vp);
893 	struct thread *td = curthread;	/* XXX */
894 	struct vattr *vap = ap->a_vap;
895 	int error = 0;
896 	u_quad_t tsize;
897 
898 #ifndef nolint
899 	tsize = (u_quad_t)0;
900 #endif
901 
902 	/*
903 	 * Setting of flags and marking of atimes are not supported.
904 	 */
905 	if (vap->va_flags != VNOVAL)
906 		return (EOPNOTSUPP);
907 
908 	/*
909 	 * Disallow write attempts if the filesystem is mounted read-only.
910 	 */
911   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
912 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
913 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
914 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
915 		return (EROFS);
916 	if (vap->va_size != VNOVAL) {
917  		switch (vp->v_type) {
918  		case VDIR:
919  			return (EISDIR);
920  		case VCHR:
921  		case VBLK:
922  		case VSOCK:
923  		case VFIFO:
924 			if (vap->va_mtime.tv_sec == VNOVAL &&
925 			    vap->va_atime.tv_sec == VNOVAL &&
926 			    vap->va_mode == (mode_t)VNOVAL &&
927 			    vap->va_uid == (uid_t)VNOVAL &&
928 			    vap->va_gid == (gid_t)VNOVAL)
929 				return (0);
930  			vap->va_size = VNOVAL;
931  			break;
932  		default:
933 			/*
934 			 * Disallow write attempts if the filesystem is
935 			 * mounted read-only.
936 			 */
937 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
938 				return (EROFS);
939 			/*
940 			 *  We run vnode_pager_setsize() early (why?),
941 			 * we must set np->n_size now to avoid vinvalbuf
942 			 * V_SAVE races that might setsize a lower
943 			 * value.
944 			 */
945 			mtx_lock(&np->n_mtx);
946 			tsize = np->n_size;
947 			mtx_unlock(&np->n_mtx);
948 			error = ncl_meta_setsize(vp, ap->a_cred, td,
949 			    vap->va_size);
950 			mtx_lock(&np->n_mtx);
951  			if (np->n_flag & NMODIFIED) {
952 			    tsize = np->n_size;
953 			    mtx_unlock(&np->n_mtx);
954  			    if (vap->va_size == 0)
955  				error = ncl_vinvalbuf(vp, 0, td, 1);
956  			    else
957  				error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
958  			    if (error) {
959 				vnode_pager_setsize(vp, tsize);
960 				return (error);
961 			    }
962 			    /*
963 			     * Call nfscl_delegmodtime() to set the modify time
964 			     * locally, as required.
965 			     */
966 			    nfscl_delegmodtime(vp);
967  			} else
968 			    mtx_unlock(&np->n_mtx);
969 			/*
970 			 * np->n_size has already been set to vap->va_size
971 			 * in ncl_meta_setsize(). We must set it again since
972 			 * nfs_loadattrcache() could be called through
973 			 * ncl_meta_setsize() and could modify np->n_size.
974 			 */
975 			mtx_lock(&np->n_mtx);
976  			np->n_vattr.na_size = np->n_size = vap->va_size;
977 			mtx_unlock(&np->n_mtx);
978   		};
979   	} else {
980 		mtx_lock(&np->n_mtx);
981 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
982 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
983 			mtx_unlock(&np->n_mtx);
984 			if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
985 			    (error == EINTR || error == EIO))
986 				return (error);
987 		} else
988 			mtx_unlock(&np->n_mtx);
989 	}
990 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
991 	if (error && vap->va_size != VNOVAL) {
992 		mtx_lock(&np->n_mtx);
993 		np->n_size = np->n_vattr.na_size = tsize;
994 		vnode_pager_setsize(vp, tsize);
995 		mtx_unlock(&np->n_mtx);
996 	}
997 	return (error);
998 }
999 
1000 /*
1001  * Do an nfs setattr rpc.
1002  */
1003 static int
1004 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1005     struct thread *td)
1006 {
1007 	struct nfsnode *np = VTONFS(vp);
1008 	int error, ret, attrflag, i;
1009 	struct nfsvattr nfsva;
1010 
1011 	if (NFS_ISV34(vp)) {
1012 		mtx_lock(&np->n_mtx);
1013 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1014 			np->n_accesscache[i].stamp = 0;
1015 		np->n_flag |= NDELEGMOD;
1016 		mtx_unlock(&np->n_mtx);
1017 		KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1018 	}
1019 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1020 	    NULL);
1021 	if (attrflag) {
1022 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1023 		if (ret && !error)
1024 			error = ret;
1025 	}
1026 	if (error && NFS_ISV4(vp))
1027 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1028 	return (error);
1029 }
1030 
1031 /*
1032  * nfs lookup call, one step at a time...
1033  * First look in cache
1034  * If not found, unlock the directory nfsnode and do the rpc
1035  */
1036 static int
1037 nfs_lookup(struct vop_lookup_args *ap)
1038 {
1039 	struct componentname *cnp = ap->a_cnp;
1040 	struct vnode *dvp = ap->a_dvp;
1041 	struct vnode **vpp = ap->a_vpp;
1042 	struct mount *mp = dvp->v_mount;
1043 	int flags = cnp->cn_flags;
1044 	struct vnode *newvp;
1045 	struct nfsmount *nmp;
1046 	struct nfsnode *np, *newnp;
1047 	int error = 0, attrflag, dattrflag, ltype, ncticks;
1048 	struct thread *td = cnp->cn_thread;
1049 	struct nfsfh *nfhp;
1050 	struct nfsvattr dnfsva, nfsva;
1051 	struct vattr vattr;
1052 	struct timespec nctime;
1053 
1054 	*vpp = NULLVP;
1055 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1056 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1057 		return (EROFS);
1058 	if (dvp->v_type != VDIR)
1059 		return (ENOTDIR);
1060 	nmp = VFSTONFS(mp);
1061 	np = VTONFS(dvp);
1062 
1063 	/* For NFSv4, wait until any remove is done. */
1064 	mtx_lock(&np->n_mtx);
1065 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1066 		np->n_flag |= NREMOVEWANT;
1067 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1068 	}
1069 	mtx_unlock(&np->n_mtx);
1070 
1071 	if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1072 		return (error);
1073 	error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1074 	if (error > 0 && error != ENOENT)
1075 		return (error);
1076 	if (error == -1) {
1077 		/*
1078 		 * Lookups of "." are special and always return the
1079 		 * current directory.  cache_lookup() already handles
1080 		 * associated locking bookkeeping, etc.
1081 		 */
1082 		if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1083 			/* XXX: Is this really correct? */
1084 			if (cnp->cn_nameiop != LOOKUP &&
1085 			    (flags & ISLASTCN))
1086 				cnp->cn_flags |= SAVENAME;
1087 			return (0);
1088 		}
1089 
1090 		/*
1091 		 * We only accept a positive hit in the cache if the
1092 		 * change time of the file matches our cached copy.
1093 		 * Otherwise, we discard the cache entry and fallback
1094 		 * to doing a lookup RPC.  We also only trust cache
1095 		 * entries for less than nm_nametimeo seconds.
1096 		 *
1097 		 * To better handle stale file handles and attributes,
1098 		 * clear the attribute cache of this node if it is a
1099 		 * leaf component, part of an open() call, and not
1100 		 * locally modified before fetching the attributes.
1101 		 * This should allow stale file handles to be detected
1102 		 * here where we can fall back to a LOOKUP RPC to
1103 		 * recover rather than having nfs_open() detect the
1104 		 * stale file handle and failing open(2) with ESTALE.
1105 		 */
1106 		newvp = *vpp;
1107 		newnp = VTONFS(newvp);
1108 		if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1109 		    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1110 		    !(newnp->n_flag & NMODIFIED)) {
1111 			mtx_lock(&newnp->n_mtx);
1112 			newnp->n_attrstamp = 0;
1113 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1114 			mtx_unlock(&newnp->n_mtx);
1115 		}
1116 		if (nfscl_nodeleg(newvp, 0) == 0 ||
1117 		    ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1118 		    VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1119 		    timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1120 			NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1121 			if (cnp->cn_nameiop != LOOKUP &&
1122 			    (flags & ISLASTCN))
1123 				cnp->cn_flags |= SAVENAME;
1124 			return (0);
1125 		}
1126 		cache_purge(newvp);
1127 		if (dvp != newvp)
1128 			vput(newvp);
1129 		else
1130 			vrele(newvp);
1131 		*vpp = NULLVP;
1132 	} else if (error == ENOENT) {
1133 		if (dvp->v_iflag & VI_DOOMED)
1134 			return (ENOENT);
1135 		/*
1136 		 * We only accept a negative hit in the cache if the
1137 		 * modification time of the parent directory matches
1138 		 * the cached copy in the name cache entry.
1139 		 * Otherwise, we discard all of the negative cache
1140 		 * entries for this directory.  We also only trust
1141 		 * negative cache entries for up to nm_negnametimeo
1142 		 * seconds.
1143 		 */
1144 		if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1145 		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1146 		    timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1147 			NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1148 			return (ENOENT);
1149 		}
1150 		cache_purge_negative(dvp);
1151 	}
1152 
1153 	error = 0;
1154 	newvp = NULLVP;
1155 	NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1156 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1157 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1158 	    NULL);
1159 	if (dattrflag)
1160 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1161 	if (error) {
1162 		if (newvp != NULLVP) {
1163 			vput(newvp);
1164 			*vpp = NULLVP;
1165 		}
1166 
1167 		if (error != ENOENT) {
1168 			if (NFS_ISV4(dvp))
1169 				error = nfscl_maperr(td, error, (uid_t)0,
1170 				    (gid_t)0);
1171 			return (error);
1172 		}
1173 
1174 		/* The requested file was not found. */
1175 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1176 		    (flags & ISLASTCN)) {
1177 			/*
1178 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1179 			 * VWRITE) here instead of just checking
1180 			 * MNT_RDONLY.
1181 			 */
1182 			if (mp->mnt_flag & MNT_RDONLY)
1183 				return (EROFS);
1184 			cnp->cn_flags |= SAVENAME;
1185 			return (EJUSTRETURN);
1186 		}
1187 
1188 		if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
1189 		    dattrflag) {
1190 			/*
1191 			 * Cache the modification time of the parent
1192 			 * directory from the post-op attributes in
1193 			 * the name cache entry.  The negative cache
1194 			 * entry will be ignored once the directory
1195 			 * has changed.  Don't bother adding the entry
1196 			 * if the directory has already changed.
1197 			 */
1198 			mtx_lock(&np->n_mtx);
1199 			if (timespeccmp(&np->n_vattr.na_mtime,
1200 			    &dnfsva.na_mtime, ==)) {
1201 				mtx_unlock(&np->n_mtx);
1202 				cache_enter_time(dvp, NULL, cnp,
1203 				    &dnfsva.na_mtime, NULL);
1204 			} else
1205 				mtx_unlock(&np->n_mtx);
1206 		}
1207 		return (ENOENT);
1208 	}
1209 
1210 	/*
1211 	 * Handle RENAME case...
1212 	 */
1213 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1214 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1215 			FREE((caddr_t)nfhp, M_NFSFH);
1216 			return (EISDIR);
1217 		}
1218 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1219 		    LK_EXCLUSIVE);
1220 		if (error)
1221 			return (error);
1222 		newvp = NFSTOV(np);
1223 		if (attrflag)
1224 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1225 			    0, 1);
1226 		*vpp = newvp;
1227 		cnp->cn_flags |= SAVENAME;
1228 		return (0);
1229 	}
1230 
1231 	if (flags & ISDOTDOT) {
1232 		ltype = NFSVOPISLOCKED(dvp);
1233 		error = vfs_busy(mp, MBF_NOWAIT);
1234 		if (error != 0) {
1235 			vfs_ref(mp);
1236 			NFSVOPUNLOCK(dvp, 0);
1237 			error = vfs_busy(mp, 0);
1238 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1239 			vfs_rel(mp);
1240 			if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1241 				vfs_unbusy(mp);
1242 				error = ENOENT;
1243 			}
1244 			if (error != 0)
1245 				return (error);
1246 		}
1247 		NFSVOPUNLOCK(dvp, 0);
1248 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1249 		    cnp->cn_lkflags);
1250 		if (error == 0)
1251 			newvp = NFSTOV(np);
1252 		vfs_unbusy(mp);
1253 		if (newvp != dvp)
1254 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1255 		if (dvp->v_iflag & VI_DOOMED) {
1256 			if (error == 0) {
1257 				if (newvp == dvp)
1258 					vrele(newvp);
1259 				else
1260 					vput(newvp);
1261 			}
1262 			error = ENOENT;
1263 		}
1264 		if (error != 0)
1265 			return (error);
1266 		if (attrflag)
1267 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1268 			    0, 1);
1269 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1270 		FREE((caddr_t)nfhp, M_NFSFH);
1271 		VREF(dvp);
1272 		newvp = dvp;
1273 		if (attrflag)
1274 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1275 			    0, 1);
1276 	} else {
1277 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1278 		    cnp->cn_lkflags);
1279 		if (error)
1280 			return (error);
1281 		newvp = NFSTOV(np);
1282 		if (attrflag)
1283 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1284 			    0, 1);
1285 		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1286 		    !(np->n_flag & NMODIFIED)) {
1287 			/*
1288 			 * Flush the attribute cache when opening a
1289 			 * leaf node to ensure that fresh attributes
1290 			 * are fetched in nfs_open() since we did not
1291 			 * fetch attributes from the LOOKUP reply.
1292 			 */
1293 			mtx_lock(&np->n_mtx);
1294 			np->n_attrstamp = 0;
1295 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1296 			mtx_unlock(&np->n_mtx);
1297 		}
1298 	}
1299 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1300 		cnp->cn_flags |= SAVENAME;
1301 	if ((cnp->cn_flags & MAKEENTRY) &&
1302 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1303 	    attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1304 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1305 		    newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1306 	*vpp = newvp;
1307 	return (0);
1308 }
1309 
1310 /*
1311  * nfs read call.
1312  * Just call ncl_bioread() to do the work.
1313  */
1314 static int
1315 nfs_read(struct vop_read_args *ap)
1316 {
1317 	struct vnode *vp = ap->a_vp;
1318 
1319 	switch (vp->v_type) {
1320 	case VREG:
1321 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1322 	case VDIR:
1323 		return (EISDIR);
1324 	default:
1325 		return (EOPNOTSUPP);
1326 	}
1327 }
1328 
1329 /*
1330  * nfs readlink call
1331  */
1332 static int
1333 nfs_readlink(struct vop_readlink_args *ap)
1334 {
1335 	struct vnode *vp = ap->a_vp;
1336 
1337 	if (vp->v_type != VLNK)
1338 		return (EINVAL);
1339 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1340 }
1341 
1342 /*
1343  * Do a readlink rpc.
1344  * Called by ncl_doio() from below the buffer cache.
1345  */
1346 int
1347 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1348 {
1349 	int error, ret, attrflag;
1350 	struct nfsvattr nfsva;
1351 
1352 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1353 	    &attrflag, NULL);
1354 	if (attrflag) {
1355 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1356 		if (ret && !error)
1357 			error = ret;
1358 	}
1359 	if (error && NFS_ISV4(vp))
1360 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1361 	return (error);
1362 }
1363 
1364 /*
1365  * nfs read rpc call
1366  * Ditto above
1367  */
1368 int
1369 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1370 {
1371 	int error, ret, attrflag;
1372 	struct nfsvattr nfsva;
1373 	struct nfsmount *nmp;
1374 
1375 	nmp = VFSTONFS(vnode_mount(vp));
1376 	error = EIO;
1377 	attrflag = 0;
1378 	if (NFSHASPNFS(nmp))
1379 		error = nfscl_doiods(vp, uiop, NULL, NULL,
1380 		    NFSV4OPEN_ACCESSREAD, cred, uiop->uio_td);
1381 	NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1382 	if (error != 0)
1383 		error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1384 		    &attrflag, NULL);
1385 	if (attrflag) {
1386 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1387 		if (ret && !error)
1388 			error = ret;
1389 	}
1390 	if (error && NFS_ISV4(vp))
1391 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1392 	return (error);
1393 }
1394 
1395 /*
1396  * nfs write call
1397  */
1398 int
1399 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1400     int *iomode, int *must_commit, int called_from_strategy)
1401 {
1402 	struct nfsvattr nfsva;
1403 	int error, attrflag, ret;
1404 	struct nfsmount *nmp;
1405 
1406 	nmp = VFSTONFS(vnode_mount(vp));
1407 	error = EIO;
1408 	attrflag = 0;
1409 	if (NFSHASPNFS(nmp))
1410 		error = nfscl_doiods(vp, uiop, iomode, must_commit,
1411 		    NFSV4OPEN_ACCESSWRITE, cred, uiop->uio_td);
1412 	NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1413 	if (error != 0)
1414 		error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1415 		    uiop->uio_td, &nfsva, &attrflag, NULL,
1416 		    called_from_strategy);
1417 	if (attrflag) {
1418 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1419 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1420 			    1);
1421 		else
1422 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1423 			    1);
1424 		if (ret && !error)
1425 			error = ret;
1426 	}
1427 	if (DOINGASYNC(vp))
1428 		*iomode = NFSWRITE_FILESYNC;
1429 	if (error && NFS_ISV4(vp))
1430 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1431 	return (error);
1432 }
1433 
1434 /*
1435  * nfs mknod rpc
1436  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1437  * mode set to specify the file type and the size field for rdev.
1438  */
1439 static int
1440 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1441     struct vattr *vap)
1442 {
1443 	struct nfsvattr nfsva, dnfsva;
1444 	struct vnode *newvp = NULL;
1445 	struct nfsnode *np = NULL, *dnp;
1446 	struct nfsfh *nfhp;
1447 	struct vattr vattr;
1448 	int error = 0, attrflag, dattrflag;
1449 	u_int32_t rdev;
1450 
1451 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1452 		rdev = vap->va_rdev;
1453 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1454 		rdev = 0xffffffff;
1455 	else
1456 		return (EOPNOTSUPP);
1457 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1458 		return (error);
1459 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1460 	    rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1461 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1462 	if (!error) {
1463 		if (!nfhp)
1464 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1465 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1466 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1467 			    NULL);
1468 		if (nfhp)
1469 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1470 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1471 	}
1472 	if (dattrflag)
1473 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1474 	if (!error) {
1475 		newvp = NFSTOV(np);
1476 		if (attrflag != 0) {
1477 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1478 			    0, 1);
1479 			if (error != 0)
1480 				vput(newvp);
1481 		}
1482 	}
1483 	if (!error) {
1484 		*vpp = newvp;
1485 	} else if (NFS_ISV4(dvp)) {
1486 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1487 		    vap->va_gid);
1488 	}
1489 	dnp = VTONFS(dvp);
1490 	mtx_lock(&dnp->n_mtx);
1491 	dnp->n_flag |= NMODIFIED;
1492 	if (!dattrflag) {
1493 		dnp->n_attrstamp = 0;
1494 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1495 	}
1496 	mtx_unlock(&dnp->n_mtx);
1497 	return (error);
1498 }
1499 
1500 /*
1501  * nfs mknod vop
1502  * just call nfs_mknodrpc() to do the work.
1503  */
1504 /* ARGSUSED */
1505 static int
1506 nfs_mknod(struct vop_mknod_args *ap)
1507 {
1508 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1509 }
1510 
1511 static struct mtx nfs_cverf_mtx;
1512 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1513     MTX_DEF);
1514 
1515 static nfsquad_t
1516 nfs_get_cverf(void)
1517 {
1518 	static nfsquad_t cverf;
1519 	nfsquad_t ret;
1520 	static int cverf_initialized = 0;
1521 
1522 	mtx_lock(&nfs_cverf_mtx);
1523 	if (cverf_initialized == 0) {
1524 		cverf.lval[0] = arc4random();
1525 		cverf.lval[1] = arc4random();
1526 		cverf_initialized = 1;
1527 	} else
1528 		cverf.qval++;
1529 	ret = cverf;
1530 	mtx_unlock(&nfs_cverf_mtx);
1531 
1532 	return (ret);
1533 }
1534 
1535 /*
1536  * nfs file create call
1537  */
1538 static int
1539 nfs_create(struct vop_create_args *ap)
1540 {
1541 	struct vnode *dvp = ap->a_dvp;
1542 	struct vattr *vap = ap->a_vap;
1543 	struct componentname *cnp = ap->a_cnp;
1544 	struct nfsnode *np = NULL, *dnp;
1545 	struct vnode *newvp = NULL;
1546 	struct nfsmount *nmp;
1547 	struct nfsvattr dnfsva, nfsva;
1548 	struct nfsfh *nfhp;
1549 	nfsquad_t cverf;
1550 	int error = 0, attrflag, dattrflag, fmode = 0;
1551 	struct vattr vattr;
1552 
1553 	/*
1554 	 * Oops, not for me..
1555 	 */
1556 	if (vap->va_type == VSOCK)
1557 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1558 
1559 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1560 		return (error);
1561 	if (vap->va_vaflags & VA_EXCLUSIVE)
1562 		fmode |= O_EXCL;
1563 	dnp = VTONFS(dvp);
1564 	nmp = VFSTONFS(vnode_mount(dvp));
1565 again:
1566 	/* For NFSv4, wait until any remove is done. */
1567 	mtx_lock(&dnp->n_mtx);
1568 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1569 		dnp->n_flag |= NREMOVEWANT;
1570 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1571 	}
1572 	mtx_unlock(&dnp->n_mtx);
1573 
1574 	cverf = nfs_get_cverf();
1575 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1576 	    vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1577 	    &nfhp, &attrflag, &dattrflag, NULL);
1578 	if (!error) {
1579 		if (nfhp == NULL)
1580 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1581 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1582 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1583 			    NULL);
1584 		if (nfhp != NULL)
1585 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1586 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1587 	}
1588 	if (dattrflag)
1589 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1590 	if (!error) {
1591 		newvp = NFSTOV(np);
1592 		if (attrflag == 0)
1593 			error = nfsrpc_getattr(newvp, cnp->cn_cred,
1594 			    cnp->cn_thread, &nfsva, NULL);
1595 		if (error == 0)
1596 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1597 			    0, 1);
1598 	}
1599 	if (error) {
1600 		if (newvp != NULL) {
1601 			vput(newvp);
1602 			newvp = NULL;
1603 		}
1604 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1605 		    error == NFSERR_NOTSUPP) {
1606 			fmode &= ~O_EXCL;
1607 			goto again;
1608 		}
1609 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1610 		if (nfscl_checksattr(vap, &nfsva)) {
1611 			/*
1612 			 * We are normally called with only a partially
1613 			 * initialized VAP. Since the NFSv3 spec says that
1614 			 * the server may use the file attributes to
1615 			 * store the verifier, the spec requires us to do a
1616 			 * SETATTR RPC. FreeBSD servers store the verifier in
1617 			 * atime, but we can't really assume that all servers
1618 			 * will so we ensure that our SETATTR sets both atime
1619 			 * and mtime.
1620 			 */
1621 			if (vap->va_mtime.tv_sec == VNOVAL)
1622 				vfs_timestamp(&vap->va_mtime);
1623 			if (vap->va_atime.tv_sec == VNOVAL)
1624 				vap->va_atime = vap->va_mtime;
1625 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1626 			    cnp->cn_thread, &nfsva, &attrflag, NULL);
1627 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1628 			    vap->va_gid != (gid_t)VNOVAL)) {
1629 				/* try again without setting uid/gid */
1630 				vap->va_uid = (uid_t)VNOVAL;
1631 				vap->va_gid = (uid_t)VNOVAL;
1632 				error = nfsrpc_setattr(newvp, vap, NULL,
1633 				    cnp->cn_cred, cnp->cn_thread, &nfsva,
1634 				    &attrflag, NULL);
1635 			}
1636 			if (attrflag)
1637 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1638 				    NULL, 0, 1);
1639 			if (error != 0)
1640 				vput(newvp);
1641 		}
1642 	}
1643 	if (!error) {
1644 		if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1645 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1646 			    NULL);
1647 		*ap->a_vpp = newvp;
1648 	} else if (NFS_ISV4(dvp)) {
1649 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1650 		    vap->va_gid);
1651 	}
1652 	mtx_lock(&dnp->n_mtx);
1653 	dnp->n_flag |= NMODIFIED;
1654 	if (!dattrflag) {
1655 		dnp->n_attrstamp = 0;
1656 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1657 	}
1658 	mtx_unlock(&dnp->n_mtx);
1659 	return (error);
1660 }
1661 
1662 /*
1663  * nfs file remove call
1664  * To try and make nfs semantics closer to ufs semantics, a file that has
1665  * other processes using the vnode is renamed instead of removed and then
1666  * removed later on the last close.
1667  * - If v_usecount > 1
1668  *	  If a rename is not already in the works
1669  *	     call nfs_sillyrename() to set it up
1670  *     else
1671  *	  do the remove rpc
1672  */
1673 static int
1674 nfs_remove(struct vop_remove_args *ap)
1675 {
1676 	struct vnode *vp = ap->a_vp;
1677 	struct vnode *dvp = ap->a_dvp;
1678 	struct componentname *cnp = ap->a_cnp;
1679 	struct nfsnode *np = VTONFS(vp);
1680 	int error = 0;
1681 	struct vattr vattr;
1682 
1683 	KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1684 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1685 	if (vp->v_type == VDIR)
1686 		error = EPERM;
1687 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1688 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1689 	    vattr.va_nlink > 1)) {
1690 		/*
1691 		 * Purge the name cache so that the chance of a lookup for
1692 		 * the name succeeding while the remove is in progress is
1693 		 * minimized. Without node locking it can still happen, such
1694 		 * that an I/O op returns ESTALE, but since you get this if
1695 		 * another host removes the file..
1696 		 */
1697 		cache_purge(vp);
1698 		/*
1699 		 * throw away biocache buffers, mainly to avoid
1700 		 * unnecessary delayed writes later.
1701 		 */
1702 		error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1703 		/* Do the rpc */
1704 		if (error != EINTR && error != EIO)
1705 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1706 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1707 		/*
1708 		 * Kludge City: If the first reply to the remove rpc is lost..
1709 		 *   the reply to the retransmitted request will be ENOENT
1710 		 *   since the file was in fact removed
1711 		 *   Therefore, we cheat and return success.
1712 		 */
1713 		if (error == ENOENT)
1714 			error = 0;
1715 	} else if (!np->n_sillyrename)
1716 		error = nfs_sillyrename(dvp, vp, cnp);
1717 	mtx_lock(&np->n_mtx);
1718 	np->n_attrstamp = 0;
1719 	mtx_unlock(&np->n_mtx);
1720 	KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1721 	return (error);
1722 }
1723 
1724 /*
1725  * nfs file remove rpc called from nfs_inactive
1726  */
1727 int
1728 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1729 {
1730 	/*
1731 	 * Make sure that the directory vnode is still valid.
1732 	 * XXX we should lock sp->s_dvp here.
1733 	 */
1734 	if (sp->s_dvp->v_type == VBAD)
1735 		return (0);
1736 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1737 	    sp->s_cred, NULL));
1738 }
1739 
1740 /*
1741  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1742  */
1743 static int
1744 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1745     int namelen, struct ucred *cred, struct thread *td)
1746 {
1747 	struct nfsvattr dnfsva;
1748 	struct nfsnode *dnp = VTONFS(dvp);
1749 	int error = 0, dattrflag;
1750 
1751 	mtx_lock(&dnp->n_mtx);
1752 	dnp->n_flag |= NREMOVEINPROG;
1753 	mtx_unlock(&dnp->n_mtx);
1754 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1755 	    &dattrflag, NULL);
1756 	mtx_lock(&dnp->n_mtx);
1757 	if ((dnp->n_flag & NREMOVEWANT)) {
1758 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1759 		mtx_unlock(&dnp->n_mtx);
1760 		wakeup((caddr_t)dnp);
1761 	} else {
1762 		dnp->n_flag &= ~NREMOVEINPROG;
1763 		mtx_unlock(&dnp->n_mtx);
1764 	}
1765 	if (dattrflag)
1766 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1767 	mtx_lock(&dnp->n_mtx);
1768 	dnp->n_flag |= NMODIFIED;
1769 	if (!dattrflag) {
1770 		dnp->n_attrstamp = 0;
1771 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1772 	}
1773 	mtx_unlock(&dnp->n_mtx);
1774 	if (error && NFS_ISV4(dvp))
1775 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1776 	return (error);
1777 }
1778 
1779 /*
1780  * nfs file rename call
1781  */
1782 static int
1783 nfs_rename(struct vop_rename_args *ap)
1784 {
1785 	struct vnode *fvp = ap->a_fvp;
1786 	struct vnode *tvp = ap->a_tvp;
1787 	struct vnode *fdvp = ap->a_fdvp;
1788 	struct vnode *tdvp = ap->a_tdvp;
1789 	struct componentname *tcnp = ap->a_tcnp;
1790 	struct componentname *fcnp = ap->a_fcnp;
1791 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1792 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1793 	struct nfsv4node *newv4 = NULL;
1794 	int error;
1795 
1796 	KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1797 	    (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1798 	/* Check for cross-device rename */
1799 	if ((fvp->v_mount != tdvp->v_mount) ||
1800 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1801 		error = EXDEV;
1802 		goto out;
1803 	}
1804 
1805 	if (fvp == tvp) {
1806 		ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1807 		error = 0;
1808 		goto out;
1809 	}
1810 	if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1811 		goto out;
1812 
1813 	/*
1814 	 * We have to flush B_DELWRI data prior to renaming
1815 	 * the file.  If we don't, the delayed-write buffers
1816 	 * can be flushed out later after the file has gone stale
1817 	 * under NFSV3.  NFSV2 does not have this problem because
1818 	 * ( as far as I can tell ) it flushes dirty buffers more
1819 	 * often.
1820 	 *
1821 	 * Skip the rename operation if the fsync fails, this can happen
1822 	 * due to the server's volume being full, when we pushed out data
1823 	 * that was written back to our cache earlier. Not checking for
1824 	 * this condition can result in potential (silent) data loss.
1825 	 */
1826 	error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1827 	NFSVOPUNLOCK(fvp, 0);
1828 	if (!error && tvp)
1829 		error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1830 	if (error)
1831 		goto out;
1832 
1833 	/*
1834 	 * If the tvp exists and is in use, sillyrename it before doing the
1835 	 * rename of the new file over it.
1836 	 * XXX Can't sillyrename a directory.
1837 	 */
1838 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1839 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1840 		vput(tvp);
1841 		tvp = NULL;
1842 	}
1843 
1844 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1845 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1846 	    tcnp->cn_thread);
1847 
1848 	if (error == 0 && NFS_ISV4(tdvp)) {
1849 		/*
1850 		 * For NFSv4, check to see if it is the same name and
1851 		 * replace the name, if it is different.
1852 		 */
1853 		MALLOC(newv4, struct nfsv4node *,
1854 		    sizeof (struct nfsv4node) +
1855 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1856 		    M_NFSV4NODE, M_WAITOK);
1857 		mtx_lock(&tdnp->n_mtx);
1858 		mtx_lock(&fnp->n_mtx);
1859 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1860 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1861 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1862 		      tcnp->cn_namelen) ||
1863 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1864 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1865 			tdnp->n_fhp->nfh_len))) {
1866 #ifdef notdef
1867 { char nnn[100]; int nnnl;
1868 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1869 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1870 nnn[nnnl] = '\0';
1871 printf("ren replace=%s\n",nnn);
1872 }
1873 #endif
1874 			FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1875 			fnp->n_v4 = newv4;
1876 			newv4 = NULL;
1877 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1878 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1879 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1880 			    tdnp->n_fhp->nfh_len);
1881 			NFSBCOPY(tcnp->cn_nameptr,
1882 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1883 		}
1884 		mtx_unlock(&tdnp->n_mtx);
1885 		mtx_unlock(&fnp->n_mtx);
1886 		if (newv4 != NULL)
1887 			FREE((caddr_t)newv4, M_NFSV4NODE);
1888 	}
1889 
1890 	if (fvp->v_type == VDIR) {
1891 		if (tvp != NULL && tvp->v_type == VDIR)
1892 			cache_purge(tdvp);
1893 		cache_purge(fdvp);
1894 	}
1895 
1896 out:
1897 	if (tdvp == tvp)
1898 		vrele(tdvp);
1899 	else
1900 		vput(tdvp);
1901 	if (tvp)
1902 		vput(tvp);
1903 	vrele(fdvp);
1904 	vrele(fvp);
1905 	/*
1906 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1907 	 */
1908 	if (error == ENOENT)
1909 		error = 0;
1910 	return (error);
1911 }
1912 
1913 /*
1914  * nfs file rename rpc called from nfs_remove() above
1915  */
1916 static int
1917 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1918     struct sillyrename *sp)
1919 {
1920 
1921 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1922 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1923 	    scnp->cn_thread));
1924 }
1925 
1926 /*
1927  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1928  */
1929 static int
1930 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1931     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1932     int tnamelen, struct ucred *cred, struct thread *td)
1933 {
1934 	struct nfsvattr fnfsva, tnfsva;
1935 	struct nfsnode *fdnp = VTONFS(fdvp);
1936 	struct nfsnode *tdnp = VTONFS(tdvp);
1937 	int error = 0, fattrflag, tattrflag;
1938 
1939 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1940 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1941 	    &tattrflag, NULL, NULL);
1942 	mtx_lock(&fdnp->n_mtx);
1943 	fdnp->n_flag |= NMODIFIED;
1944 	if (fattrflag != 0) {
1945 		mtx_unlock(&fdnp->n_mtx);
1946 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1947 	} else {
1948 		fdnp->n_attrstamp = 0;
1949 		mtx_unlock(&fdnp->n_mtx);
1950 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1951 	}
1952 	mtx_lock(&tdnp->n_mtx);
1953 	tdnp->n_flag |= NMODIFIED;
1954 	if (tattrflag != 0) {
1955 		mtx_unlock(&tdnp->n_mtx);
1956 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1957 	} else {
1958 		tdnp->n_attrstamp = 0;
1959 		mtx_unlock(&tdnp->n_mtx);
1960 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1961 	}
1962 	if (error && NFS_ISV4(fdvp))
1963 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1964 	return (error);
1965 }
1966 
1967 /*
1968  * nfs hard link create call
1969  */
1970 static int
1971 nfs_link(struct vop_link_args *ap)
1972 {
1973 	struct vnode *vp = ap->a_vp;
1974 	struct vnode *tdvp = ap->a_tdvp;
1975 	struct componentname *cnp = ap->a_cnp;
1976 	struct nfsnode *np, *tdnp;
1977 	struct nfsvattr nfsva, dnfsva;
1978 	int error = 0, attrflag, dattrflag;
1979 
1980 	if (vp->v_mount != tdvp->v_mount) {
1981 		return (EXDEV);
1982 	}
1983 
1984 	/*
1985 	 * Push all writes to the server, so that the attribute cache
1986 	 * doesn't get "out of sync" with the server.
1987 	 * XXX There should be a better way!
1988 	 */
1989 	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1990 
1991 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1992 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1993 	    &dattrflag, NULL);
1994 	tdnp = VTONFS(tdvp);
1995 	mtx_lock(&tdnp->n_mtx);
1996 	tdnp->n_flag |= NMODIFIED;
1997 	if (dattrflag != 0) {
1998 		mtx_unlock(&tdnp->n_mtx);
1999 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
2000 	} else {
2001 		tdnp->n_attrstamp = 0;
2002 		mtx_unlock(&tdnp->n_mtx);
2003 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2004 	}
2005 	if (attrflag)
2006 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2007 	else {
2008 		np = VTONFS(vp);
2009 		mtx_lock(&np->n_mtx);
2010 		np->n_attrstamp = 0;
2011 		mtx_unlock(&np->n_mtx);
2012 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2013 	}
2014 	/*
2015 	 * If negative lookup caching is enabled, I might as well
2016 	 * add an entry for this node. Not necessary for correctness,
2017 	 * but if negative caching is enabled, then the system
2018 	 * must care about lookup caching hit rate, so...
2019 	 */
2020 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2021 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2022 		cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2023 	}
2024 	if (error && NFS_ISV4(vp))
2025 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2026 		    (gid_t)0);
2027 	return (error);
2028 }
2029 
2030 /*
2031  * nfs symbolic link create call
2032  */
2033 static int
2034 nfs_symlink(struct vop_symlink_args *ap)
2035 {
2036 	struct vnode *dvp = ap->a_dvp;
2037 	struct vattr *vap = ap->a_vap;
2038 	struct componentname *cnp = ap->a_cnp;
2039 	struct nfsvattr nfsva, dnfsva;
2040 	struct nfsfh *nfhp;
2041 	struct nfsnode *np = NULL, *dnp;
2042 	struct vnode *newvp = NULL;
2043 	int error = 0, attrflag, dattrflag, ret;
2044 
2045 	vap->va_type = VLNK;
2046 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2047 	    ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2048 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2049 	if (nfhp) {
2050 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2051 		    &np, NULL, LK_EXCLUSIVE);
2052 		if (!ret)
2053 			newvp = NFSTOV(np);
2054 		else if (!error)
2055 			error = ret;
2056 	}
2057 	if (newvp != NULL) {
2058 		if (attrflag)
2059 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2060 			    0, 1);
2061 	} else if (!error) {
2062 		/*
2063 		 * If we do not have an error and we could not extract the
2064 		 * newvp from the response due to the request being NFSv2, we
2065 		 * have to do a lookup in order to obtain a newvp to return.
2066 		 */
2067 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2068 		    cnp->cn_cred, cnp->cn_thread, &np);
2069 		if (!error)
2070 			newvp = NFSTOV(np);
2071 	}
2072 	if (error) {
2073 		if (newvp)
2074 			vput(newvp);
2075 		if (NFS_ISV4(dvp))
2076 			error = nfscl_maperr(cnp->cn_thread, error,
2077 			    vap->va_uid, vap->va_gid);
2078 	} else {
2079 		*ap->a_vpp = newvp;
2080 	}
2081 
2082 	dnp = VTONFS(dvp);
2083 	mtx_lock(&dnp->n_mtx);
2084 	dnp->n_flag |= NMODIFIED;
2085 	if (dattrflag != 0) {
2086 		mtx_unlock(&dnp->n_mtx);
2087 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2088 	} else {
2089 		dnp->n_attrstamp = 0;
2090 		mtx_unlock(&dnp->n_mtx);
2091 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2092 	}
2093 	/*
2094 	 * If negative lookup caching is enabled, I might as well
2095 	 * add an entry for this node. Not necessary for correctness,
2096 	 * but if negative caching is enabled, then the system
2097 	 * must care about lookup caching hit rate, so...
2098 	 */
2099 	if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2100 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2101 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2102 	}
2103 	return (error);
2104 }
2105 
2106 /*
2107  * nfs make dir call
2108  */
2109 static int
2110 nfs_mkdir(struct vop_mkdir_args *ap)
2111 {
2112 	struct vnode *dvp = ap->a_dvp;
2113 	struct vattr *vap = ap->a_vap;
2114 	struct componentname *cnp = ap->a_cnp;
2115 	struct nfsnode *np = NULL, *dnp;
2116 	struct vnode *newvp = NULL;
2117 	struct vattr vattr;
2118 	struct nfsfh *nfhp;
2119 	struct nfsvattr nfsva, dnfsva;
2120 	int error = 0, attrflag, dattrflag, ret;
2121 
2122 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2123 		return (error);
2124 	vap->va_type = VDIR;
2125 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2126 	    vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2127 	    &attrflag, &dattrflag, NULL);
2128 	dnp = VTONFS(dvp);
2129 	mtx_lock(&dnp->n_mtx);
2130 	dnp->n_flag |= NMODIFIED;
2131 	if (dattrflag != 0) {
2132 		mtx_unlock(&dnp->n_mtx);
2133 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2134 	} else {
2135 		dnp->n_attrstamp = 0;
2136 		mtx_unlock(&dnp->n_mtx);
2137 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2138 	}
2139 	if (nfhp) {
2140 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2141 		    &np, NULL, LK_EXCLUSIVE);
2142 		if (!ret) {
2143 			newvp = NFSTOV(np);
2144 			if (attrflag)
2145 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2146 				NULL, 0, 1);
2147 		} else if (!error)
2148 			error = ret;
2149 	}
2150 	if (!error && newvp == NULL) {
2151 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2152 		    cnp->cn_cred, cnp->cn_thread, &np);
2153 		if (!error) {
2154 			newvp = NFSTOV(np);
2155 			if (newvp->v_type != VDIR)
2156 				error = EEXIST;
2157 		}
2158 	}
2159 	if (error) {
2160 		if (newvp)
2161 			vput(newvp);
2162 		if (NFS_ISV4(dvp))
2163 			error = nfscl_maperr(cnp->cn_thread, error,
2164 			    vap->va_uid, vap->va_gid);
2165 	} else {
2166 		/*
2167 		 * If negative lookup caching is enabled, I might as well
2168 		 * add an entry for this node. Not necessary for correctness,
2169 		 * but if negative caching is enabled, then the system
2170 		 * must care about lookup caching hit rate, so...
2171 		 */
2172 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2173 		    (cnp->cn_flags & MAKEENTRY) &&
2174 		    attrflag != 0 && dattrflag != 0)
2175 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2176 			    &dnfsva.na_ctime);
2177 		*ap->a_vpp = newvp;
2178 	}
2179 	return (error);
2180 }
2181 
2182 /*
2183  * nfs remove directory call
2184  */
2185 static int
2186 nfs_rmdir(struct vop_rmdir_args *ap)
2187 {
2188 	struct vnode *vp = ap->a_vp;
2189 	struct vnode *dvp = ap->a_dvp;
2190 	struct componentname *cnp = ap->a_cnp;
2191 	struct nfsnode *dnp;
2192 	struct nfsvattr dnfsva;
2193 	int error, dattrflag;
2194 
2195 	if (dvp == vp)
2196 		return (EINVAL);
2197 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2198 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2199 	dnp = VTONFS(dvp);
2200 	mtx_lock(&dnp->n_mtx);
2201 	dnp->n_flag |= NMODIFIED;
2202 	if (dattrflag != 0) {
2203 		mtx_unlock(&dnp->n_mtx);
2204 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2205 	} else {
2206 		dnp->n_attrstamp = 0;
2207 		mtx_unlock(&dnp->n_mtx);
2208 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2209 	}
2210 
2211 	cache_purge(dvp);
2212 	cache_purge(vp);
2213 	if (error && NFS_ISV4(dvp))
2214 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2215 		    (gid_t)0);
2216 	/*
2217 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2218 	 */
2219 	if (error == ENOENT)
2220 		error = 0;
2221 	return (error);
2222 }
2223 
2224 /*
2225  * nfs readdir call
2226  */
2227 static int
2228 nfs_readdir(struct vop_readdir_args *ap)
2229 {
2230 	struct vnode *vp = ap->a_vp;
2231 	struct nfsnode *np = VTONFS(vp);
2232 	struct uio *uio = ap->a_uio;
2233 	ssize_t tresid;
2234 	int error = 0;
2235 	struct vattr vattr;
2236 
2237 	if (ap->a_eofflag != NULL)
2238 		*ap->a_eofflag = 0;
2239 	if (vp->v_type != VDIR)
2240 		return(EPERM);
2241 
2242 	/*
2243 	 * First, check for hit on the EOF offset cache
2244 	 */
2245 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2246 	    (np->n_flag & NMODIFIED) == 0) {
2247 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2248 			mtx_lock(&np->n_mtx);
2249 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2250 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2251 				mtx_unlock(&np->n_mtx);
2252 				NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2253 				if (ap->a_eofflag != NULL)
2254 					*ap->a_eofflag = 1;
2255 				return (0);
2256 			} else
2257 				mtx_unlock(&np->n_mtx);
2258 		}
2259 	}
2260 
2261 	/*
2262 	 * Call ncl_bioread() to do the real work.
2263 	 */
2264 	tresid = uio->uio_resid;
2265 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2266 
2267 	if (!error && uio->uio_resid == tresid) {
2268 		NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2269 		if (ap->a_eofflag != NULL)
2270 			*ap->a_eofflag = 1;
2271 	}
2272 	return (error);
2273 }
2274 
2275 /*
2276  * Readdir rpc call.
2277  * Called from below the buffer cache by ncl_doio().
2278  */
2279 int
2280 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2281     struct thread *td)
2282 {
2283 	struct nfsvattr nfsva;
2284 	nfsuint64 *cookiep, cookie;
2285 	struct nfsnode *dnp = VTONFS(vp);
2286 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2287 	int error = 0, eof, attrflag;
2288 
2289 	KASSERT(uiop->uio_iovcnt == 1 &&
2290 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2291 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2292 	    ("nfs readdirrpc bad uio"));
2293 
2294 	/*
2295 	 * If there is no cookie, assume directory was stale.
2296 	 */
2297 	ncl_dircookie_lock(dnp);
2298 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2299 	if (cookiep) {
2300 		cookie = *cookiep;
2301 		ncl_dircookie_unlock(dnp);
2302 	} else {
2303 		ncl_dircookie_unlock(dnp);
2304 		return (NFSERR_BAD_COOKIE);
2305 	}
2306 
2307 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2308 		(void)ncl_fsinfo(nmp, vp, cred, td);
2309 
2310 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2311 	    &attrflag, &eof, NULL);
2312 	if (attrflag)
2313 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2314 
2315 	if (!error) {
2316 		/*
2317 		 * We are now either at the end of the directory or have filled
2318 		 * the block.
2319 		 */
2320 		if (eof)
2321 			dnp->n_direofoffset = uiop->uio_offset;
2322 		else {
2323 			if (uiop->uio_resid > 0)
2324 				ncl_printf("EEK! readdirrpc resid > 0\n");
2325 			ncl_dircookie_lock(dnp);
2326 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2327 			*cookiep = cookie;
2328 			ncl_dircookie_unlock(dnp);
2329 		}
2330 	} else if (NFS_ISV4(vp)) {
2331 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2332 	}
2333 	return (error);
2334 }
2335 
2336 /*
2337  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2338  */
2339 int
2340 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2341     struct thread *td)
2342 {
2343 	struct nfsvattr nfsva;
2344 	nfsuint64 *cookiep, cookie;
2345 	struct nfsnode *dnp = VTONFS(vp);
2346 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2347 	int error = 0, attrflag, eof;
2348 
2349 	KASSERT(uiop->uio_iovcnt == 1 &&
2350 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2351 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2352 	    ("nfs readdirplusrpc bad uio"));
2353 
2354 	/*
2355 	 * If there is no cookie, assume directory was stale.
2356 	 */
2357 	ncl_dircookie_lock(dnp);
2358 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2359 	if (cookiep) {
2360 		cookie = *cookiep;
2361 		ncl_dircookie_unlock(dnp);
2362 	} else {
2363 		ncl_dircookie_unlock(dnp);
2364 		return (NFSERR_BAD_COOKIE);
2365 	}
2366 
2367 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2368 		(void)ncl_fsinfo(nmp, vp, cred, td);
2369 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2370 	    &attrflag, &eof, NULL);
2371 	if (attrflag)
2372 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2373 
2374 	if (!error) {
2375 		/*
2376 		 * We are now either at end of the directory or have filled the
2377 		 * the block.
2378 		 */
2379 		if (eof)
2380 			dnp->n_direofoffset = uiop->uio_offset;
2381 		else {
2382 			if (uiop->uio_resid > 0)
2383 				ncl_printf("EEK! readdirplusrpc resid > 0\n");
2384 			ncl_dircookie_lock(dnp);
2385 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2386 			*cookiep = cookie;
2387 			ncl_dircookie_unlock(dnp);
2388 		}
2389 	} else if (NFS_ISV4(vp)) {
2390 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2391 	}
2392 	return (error);
2393 }
2394 
2395 /*
2396  * Silly rename. To make the NFS filesystem that is stateless look a little
2397  * more like the "ufs" a remove of an active vnode is translated to a rename
2398  * to a funny looking filename that is removed by nfs_inactive on the
2399  * nfsnode. There is the potential for another process on a different client
2400  * to create the same funny name between the nfs_lookitup() fails and the
2401  * nfs_rename() completes, but...
2402  */
2403 static int
2404 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2405 {
2406 	struct sillyrename *sp;
2407 	struct nfsnode *np;
2408 	int error;
2409 	short pid;
2410 	unsigned int lticks;
2411 
2412 	cache_purge(dvp);
2413 	np = VTONFS(vp);
2414 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2415 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2416 	    M_NEWNFSREQ, M_WAITOK);
2417 	sp->s_cred = crhold(cnp->cn_cred);
2418 	sp->s_dvp = dvp;
2419 	VREF(dvp);
2420 
2421 	/*
2422 	 * Fudge together a funny name.
2423 	 * Changing the format of the funny name to accomodate more
2424 	 * sillynames per directory.
2425 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2426 	 * CPU ticks since boot.
2427 	 */
2428 	pid = cnp->cn_thread->td_proc->p_pid;
2429 	lticks = (unsigned int)ticks;
2430 	for ( ; ; ) {
2431 		sp->s_namlen = snprintf(sp->s_name, sizeof(sp->s_name),
2432 				       ".nfs.%08x.%04x4.4", lticks,
2433 				       pid);
2434 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2435 				 cnp->cn_thread, NULL))
2436 			break;
2437 		lticks++;
2438 	}
2439 	error = nfs_renameit(dvp, vp, cnp, sp);
2440 	if (error)
2441 		goto bad;
2442 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2443 		cnp->cn_thread, &np);
2444 	np->n_sillyrename = sp;
2445 	return (0);
2446 bad:
2447 	vrele(sp->s_dvp);
2448 	crfree(sp->s_cred);
2449 	free((caddr_t)sp, M_NEWNFSREQ);
2450 	return (error);
2451 }
2452 
2453 /*
2454  * Look up a file name and optionally either update the file handle or
2455  * allocate an nfsnode, depending on the value of npp.
2456  * npp == NULL	--> just do the lookup
2457  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2458  *			handled too
2459  * *npp != NULL --> update the file handle in the vnode
2460  */
2461 static int
2462 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2463     struct thread *td, struct nfsnode **npp)
2464 {
2465 	struct vnode *newvp = NULL, *vp;
2466 	struct nfsnode *np, *dnp = VTONFS(dvp);
2467 	struct nfsfh *nfhp, *onfhp;
2468 	struct nfsvattr nfsva, dnfsva;
2469 	struct componentname cn;
2470 	int error = 0, attrflag, dattrflag;
2471 	u_int hash;
2472 
2473 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2474 	    &nfhp, &attrflag, &dattrflag, NULL);
2475 	if (dattrflag)
2476 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2477 	if (npp && !error) {
2478 		if (*npp != NULL) {
2479 		    np = *npp;
2480 		    vp = NFSTOV(np);
2481 		    /*
2482 		     * For NFSv4, check to see if it is the same name and
2483 		     * replace the name, if it is different.
2484 		     */
2485 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2486 			(np->n_v4->n4_namelen != len ||
2487 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2488 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2489 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2490 			 dnp->n_fhp->nfh_len))) {
2491 #ifdef notdef
2492 { char nnn[100]; int nnnl;
2493 nnnl = (len < 100) ? len : 99;
2494 bcopy(name, nnn, nnnl);
2495 nnn[nnnl] = '\0';
2496 printf("replace=%s\n",nnn);
2497 }
2498 #endif
2499 			    FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2500 			    MALLOC(np->n_v4, struct nfsv4node *,
2501 				sizeof (struct nfsv4node) +
2502 				dnp->n_fhp->nfh_len + len - 1,
2503 				M_NFSV4NODE, M_WAITOK);
2504 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2505 			    np->n_v4->n4_namelen = len;
2506 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2507 				dnp->n_fhp->nfh_len);
2508 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2509 		    }
2510 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2511 			FNV1_32_INIT);
2512 		    onfhp = np->n_fhp;
2513 		    /*
2514 		     * Rehash node for new file handle.
2515 		     */
2516 		    vfs_hash_rehash(vp, hash);
2517 		    np->n_fhp = nfhp;
2518 		    if (onfhp != NULL)
2519 			FREE((caddr_t)onfhp, M_NFSFH);
2520 		    newvp = NFSTOV(np);
2521 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2522 		    FREE((caddr_t)nfhp, M_NFSFH);
2523 		    VREF(dvp);
2524 		    newvp = dvp;
2525 		} else {
2526 		    cn.cn_nameptr = name;
2527 		    cn.cn_namelen = len;
2528 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2529 			&np, NULL, LK_EXCLUSIVE);
2530 		    if (error)
2531 			return (error);
2532 		    newvp = NFSTOV(np);
2533 		}
2534 		if (!attrflag && *npp == NULL) {
2535 			if (newvp == dvp)
2536 				vrele(newvp);
2537 			else
2538 				vput(newvp);
2539 			return (ENOENT);
2540 		}
2541 		if (attrflag)
2542 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2543 			    0, 1);
2544 	}
2545 	if (npp && *npp == NULL) {
2546 		if (error) {
2547 			if (newvp) {
2548 				if (newvp == dvp)
2549 					vrele(newvp);
2550 				else
2551 					vput(newvp);
2552 			}
2553 		} else
2554 			*npp = np;
2555 	}
2556 	if (error && NFS_ISV4(dvp))
2557 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2558 	return (error);
2559 }
2560 
2561 /*
2562  * Nfs Version 3 and 4 commit rpc
2563  */
2564 int
2565 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2566    struct thread *td)
2567 {
2568 	struct nfsvattr nfsva;
2569 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2570 	int error, attrflag;
2571 
2572 	mtx_lock(&nmp->nm_mtx);
2573 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2574 		mtx_unlock(&nmp->nm_mtx);
2575 		return (0);
2576 	}
2577 	mtx_unlock(&nmp->nm_mtx);
2578 	error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2579 	    &attrflag, NULL);
2580 	if (attrflag != 0)
2581 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2582 		    0, 1);
2583 	if (error != 0 && NFS_ISV4(vp))
2584 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2585 	return (error);
2586 }
2587 
2588 /*
2589  * Strategy routine.
2590  * For async requests when nfsiod(s) are running, queue the request by
2591  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2592  * request.
2593  */
2594 static int
2595 nfs_strategy(struct vop_strategy_args *ap)
2596 {
2597 	struct buf *bp = ap->a_bp;
2598 	struct ucred *cr;
2599 
2600 	KASSERT(!(bp->b_flags & B_DONE),
2601 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2602 	BUF_ASSERT_HELD(bp);
2603 
2604 	if (bp->b_iocmd == BIO_READ)
2605 		cr = bp->b_rcred;
2606 	else
2607 		cr = bp->b_wcred;
2608 
2609 	/*
2610 	 * If the op is asynchronous and an i/o daemon is waiting
2611 	 * queue the request, wake it up and wait for completion
2612 	 * otherwise just do it ourselves.
2613 	 */
2614 	if ((bp->b_flags & B_ASYNC) == 0 ||
2615 	    ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2616 		(void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2617 	return (0);
2618 }
2619 
2620 /*
2621  * fsync vnode op. Just call ncl_flush() with commit == 1.
2622  */
2623 /* ARGSUSED */
2624 static int
2625 nfs_fsync(struct vop_fsync_args *ap)
2626 {
2627 
2628 	if (ap->a_vp->v_type != VREG) {
2629 		/*
2630 		 * For NFS, metadata is changed synchronously on the server,
2631 		 * so there is nothing to flush. Also, ncl_flush() clears
2632 		 * the NMODIFIED flag and that shouldn't be done here for
2633 		 * directories.
2634 		 */
2635 		return (0);
2636 	}
2637 	return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2638 }
2639 
2640 /*
2641  * Flush all the blocks associated with a vnode.
2642  * 	Walk through the buffer pool and push any dirty pages
2643  *	associated with the vnode.
2644  * If the called_from_renewthread argument is TRUE, it has been called
2645  * from the NFSv4 renew thread and, as such, cannot block indefinitely
2646  * waiting for a buffer write to complete.
2647  */
2648 int
2649 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2650     int commit, int called_from_renewthread)
2651 {
2652 	struct nfsnode *np = VTONFS(vp);
2653 	struct buf *bp;
2654 	int i;
2655 	struct buf *nbp;
2656 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2657 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2658 	int passone = 1, trycnt = 0;
2659 	u_quad_t off, endoff, toff;
2660 	struct ucred* wcred = NULL;
2661 	struct buf **bvec = NULL;
2662 	struct bufobj *bo;
2663 #ifndef NFS_COMMITBVECSIZ
2664 #define	NFS_COMMITBVECSIZ	20
2665 #endif
2666 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2667 	int bvecsize = 0, bveccount;
2668 
2669 	if (called_from_renewthread != 0)
2670 		slptimeo = hz;
2671 	if (nmp->nm_flag & NFSMNT_INT)
2672 		slpflag = PCATCH;
2673 	if (!commit)
2674 		passone = 0;
2675 	bo = &vp->v_bufobj;
2676 	/*
2677 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2678 	 * server, but has not been committed to stable storage on the server
2679 	 * yet. On the first pass, the byte range is worked out and the commit
2680 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2681 	 * job.
2682 	 */
2683 again:
2684 	off = (u_quad_t)-1;
2685 	endoff = 0;
2686 	bvecpos = 0;
2687 	if (NFS_ISV34(vp) && commit) {
2688 		if (bvec != NULL && bvec != bvec_on_stack)
2689 			free(bvec, M_TEMP);
2690 		/*
2691 		 * Count up how many buffers waiting for a commit.
2692 		 */
2693 		bveccount = 0;
2694 		BO_LOCK(bo);
2695 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2696 			if (!BUF_ISLOCKED(bp) &&
2697 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2698 				== (B_DELWRI | B_NEEDCOMMIT))
2699 				bveccount++;
2700 		}
2701 		/*
2702 		 * Allocate space to remember the list of bufs to commit.  It is
2703 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2704 		 * If we can't get memory (for whatever reason), we will end up
2705 		 * committing the buffers one-by-one in the loop below.
2706 		 */
2707 		if (bveccount > NFS_COMMITBVECSIZ) {
2708 			/*
2709 			 * Release the vnode interlock to avoid a lock
2710 			 * order reversal.
2711 			 */
2712 			BO_UNLOCK(bo);
2713 			bvec = (struct buf **)
2714 				malloc(bveccount * sizeof(struct buf *),
2715 				       M_TEMP, M_NOWAIT);
2716 			BO_LOCK(bo);
2717 			if (bvec == NULL) {
2718 				bvec = bvec_on_stack;
2719 				bvecsize = NFS_COMMITBVECSIZ;
2720 			} else
2721 				bvecsize = bveccount;
2722 		} else {
2723 			bvec = bvec_on_stack;
2724 			bvecsize = NFS_COMMITBVECSIZ;
2725 		}
2726 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2727 			if (bvecpos >= bvecsize)
2728 				break;
2729 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2730 				nbp = TAILQ_NEXT(bp, b_bobufs);
2731 				continue;
2732 			}
2733 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2734 			    (B_DELWRI | B_NEEDCOMMIT)) {
2735 				BUF_UNLOCK(bp);
2736 				nbp = TAILQ_NEXT(bp, b_bobufs);
2737 				continue;
2738 			}
2739 			BO_UNLOCK(bo);
2740 			bremfree(bp);
2741 			/*
2742 			 * Work out if all buffers are using the same cred
2743 			 * so we can deal with them all with one commit.
2744 			 *
2745 			 * NOTE: we are not clearing B_DONE here, so we have
2746 			 * to do it later on in this routine if we intend to
2747 			 * initiate I/O on the bp.
2748 			 *
2749 			 * Note: to avoid loopback deadlocks, we do not
2750 			 * assign b_runningbufspace.
2751 			 */
2752 			if (wcred == NULL)
2753 				wcred = bp->b_wcred;
2754 			else if (wcred != bp->b_wcred)
2755 				wcred = NOCRED;
2756 			vfs_busy_pages(bp, 1);
2757 
2758 			BO_LOCK(bo);
2759 			/*
2760 			 * bp is protected by being locked, but nbp is not
2761 			 * and vfs_busy_pages() may sleep.  We have to
2762 			 * recalculate nbp.
2763 			 */
2764 			nbp = TAILQ_NEXT(bp, b_bobufs);
2765 
2766 			/*
2767 			 * A list of these buffers is kept so that the
2768 			 * second loop knows which buffers have actually
2769 			 * been committed. This is necessary, since there
2770 			 * may be a race between the commit rpc and new
2771 			 * uncommitted writes on the file.
2772 			 */
2773 			bvec[bvecpos++] = bp;
2774 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2775 				bp->b_dirtyoff;
2776 			if (toff < off)
2777 				off = toff;
2778 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2779 			if (toff > endoff)
2780 				endoff = toff;
2781 		}
2782 		BO_UNLOCK(bo);
2783 	}
2784 	if (bvecpos > 0) {
2785 		/*
2786 		 * Commit data on the server, as required.
2787 		 * If all bufs are using the same wcred, then use that with
2788 		 * one call for all of them, otherwise commit each one
2789 		 * separately.
2790 		 */
2791 		if (wcred != NOCRED)
2792 			retv = ncl_commit(vp, off, (int)(endoff - off),
2793 					  wcred, td);
2794 		else {
2795 			retv = 0;
2796 			for (i = 0; i < bvecpos; i++) {
2797 				off_t off, size;
2798 				bp = bvec[i];
2799 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2800 					bp->b_dirtyoff;
2801 				size = (u_quad_t)(bp->b_dirtyend
2802 						  - bp->b_dirtyoff);
2803 				retv = ncl_commit(vp, off, (int)size,
2804 						  bp->b_wcred, td);
2805 				if (retv) break;
2806 			}
2807 		}
2808 
2809 		if (retv == NFSERR_STALEWRITEVERF)
2810 			ncl_clearcommit(vp->v_mount);
2811 
2812 		/*
2813 		 * Now, either mark the blocks I/O done or mark the
2814 		 * blocks dirty, depending on whether the commit
2815 		 * succeeded.
2816 		 */
2817 		for (i = 0; i < bvecpos; i++) {
2818 			bp = bvec[i];
2819 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2820 			if (retv) {
2821 				/*
2822 				 * Error, leave B_DELWRI intact
2823 				 */
2824 				vfs_unbusy_pages(bp);
2825 				brelse(bp);
2826 			} else {
2827 				/*
2828 				 * Success, remove B_DELWRI ( bundirty() ).
2829 				 *
2830 				 * b_dirtyoff/b_dirtyend seem to be NFS
2831 				 * specific.  We should probably move that
2832 				 * into bundirty(). XXX
2833 				 */
2834 				bufobj_wref(bo);
2835 				bp->b_flags |= B_ASYNC;
2836 				bundirty(bp);
2837 				bp->b_flags &= ~B_DONE;
2838 				bp->b_ioflags &= ~BIO_ERROR;
2839 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2840 				bufdone(bp);
2841 			}
2842 		}
2843 	}
2844 
2845 	/*
2846 	 * Start/do any write(s) that are required.
2847 	 */
2848 loop:
2849 	BO_LOCK(bo);
2850 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2851 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2852 			if (waitfor != MNT_WAIT || passone)
2853 				continue;
2854 
2855 			error = BUF_TIMELOCK(bp,
2856 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2857 			    BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
2858 			if (error == 0) {
2859 				BUF_UNLOCK(bp);
2860 				goto loop;
2861 			}
2862 			if (error == ENOLCK) {
2863 				error = 0;
2864 				goto loop;
2865 			}
2866 			if (called_from_renewthread != 0) {
2867 				/*
2868 				 * Return EIO so the flush will be retried
2869 				 * later.
2870 				 */
2871 				error = EIO;
2872 				goto done;
2873 			}
2874 			if (newnfs_sigintr(nmp, td)) {
2875 				error = EINTR;
2876 				goto done;
2877 			}
2878 			if (slpflag == PCATCH) {
2879 				slpflag = 0;
2880 				slptimeo = 2 * hz;
2881 			}
2882 			goto loop;
2883 		}
2884 		if ((bp->b_flags & B_DELWRI) == 0)
2885 			panic("nfs_fsync: not dirty");
2886 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2887 			BUF_UNLOCK(bp);
2888 			continue;
2889 		}
2890 		BO_UNLOCK(bo);
2891 		bremfree(bp);
2892 		if (passone || !commit)
2893 		    bp->b_flags |= B_ASYNC;
2894 		else
2895 		    bp->b_flags |= B_ASYNC;
2896 		bwrite(bp);
2897 		if (newnfs_sigintr(nmp, td)) {
2898 			error = EINTR;
2899 			goto done;
2900 		}
2901 		goto loop;
2902 	}
2903 	if (passone) {
2904 		passone = 0;
2905 		BO_UNLOCK(bo);
2906 		goto again;
2907 	}
2908 	if (waitfor == MNT_WAIT) {
2909 		while (bo->bo_numoutput) {
2910 			error = bufobj_wwait(bo, slpflag, slptimeo);
2911 			if (error) {
2912 			    BO_UNLOCK(bo);
2913 			    if (called_from_renewthread != 0) {
2914 				/*
2915 				 * Return EIO so that the flush will be
2916 				 * retried later.
2917 				 */
2918 				error = EIO;
2919 				goto done;
2920 			    }
2921 			    error = newnfs_sigintr(nmp, td);
2922 			    if (error)
2923 				goto done;
2924 			    if (slpflag == PCATCH) {
2925 				slpflag = 0;
2926 				slptimeo = 2 * hz;
2927 			    }
2928 			    BO_LOCK(bo);
2929 			}
2930 		}
2931 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
2932 			BO_UNLOCK(bo);
2933 			goto loop;
2934 		}
2935 		/*
2936 		 * Wait for all the async IO requests to drain
2937 		 */
2938 		BO_UNLOCK(bo);
2939 		mtx_lock(&np->n_mtx);
2940 		while (np->n_directio_asyncwr > 0) {
2941 			np->n_flag |= NFSYNCWAIT;
2942 			error = newnfs_msleep(td, &np->n_directio_asyncwr,
2943 			    &np->n_mtx, slpflag | (PRIBIO + 1),
2944 			    "nfsfsync", 0);
2945 			if (error) {
2946 				if (newnfs_sigintr(nmp, td)) {
2947 					mtx_unlock(&np->n_mtx);
2948 					error = EINTR;
2949 					goto done;
2950 				}
2951 			}
2952 		}
2953 		mtx_unlock(&np->n_mtx);
2954 	} else
2955 		BO_UNLOCK(bo);
2956 	if (NFSHASPNFS(nmp)) {
2957 		nfscl_layoutcommit(vp, td);
2958 		/*
2959 		 * Invalidate the attribute cache, since writes to a DS
2960 		 * won't update the size attribute.
2961 		 */
2962 		mtx_lock(&np->n_mtx);
2963 		np->n_attrstamp = 0;
2964 	} else
2965 		mtx_lock(&np->n_mtx);
2966 	if (np->n_flag & NWRITEERR) {
2967 		error = np->n_error;
2968 		np->n_flag &= ~NWRITEERR;
2969 	}
2970   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
2971 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2972   		np->n_flag &= ~NMODIFIED;
2973 	mtx_unlock(&np->n_mtx);
2974 done:
2975 	if (bvec != NULL && bvec != bvec_on_stack)
2976 		free(bvec, M_TEMP);
2977 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2978 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2979 	     np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2980 		/* try, try again... */
2981 		passone = 1;
2982 		wcred = NULL;
2983 		bvec = NULL;
2984 		bvecsize = 0;
2985 printf("try%d\n", trycnt);
2986 		goto again;
2987 	}
2988 	return (error);
2989 }
2990 
2991 /*
2992  * NFS advisory byte-level locks.
2993  */
2994 static int
2995 nfs_advlock(struct vop_advlock_args *ap)
2996 {
2997 	struct vnode *vp = ap->a_vp;
2998 	struct ucred *cred;
2999 	struct nfsnode *np = VTONFS(ap->a_vp);
3000 	struct proc *p = (struct proc *)ap->a_id;
3001 	struct thread *td = curthread;	/* XXX */
3002 	struct vattr va;
3003 	int ret, error = EOPNOTSUPP;
3004 	u_quad_t size;
3005 
3006 	if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3007 		if (vp->v_type != VREG)
3008 			return (EINVAL);
3009 		if ((ap->a_flags & F_POSIX) != 0)
3010 			cred = p->p_ucred;
3011 		else
3012 			cred = td->td_ucred;
3013 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3014 		if (vp->v_iflag & VI_DOOMED) {
3015 			NFSVOPUNLOCK(vp, 0);
3016 			return (EBADF);
3017 		}
3018 
3019 		/*
3020 		 * If this is unlocking a write locked region, flush and
3021 		 * commit them before unlocking. This is required by
3022 		 * RFC3530 Sec. 9.3.2.
3023 		 */
3024 		if (ap->a_op == F_UNLCK &&
3025 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3026 		    ap->a_flags))
3027 			(void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
3028 
3029 		/*
3030 		 * Loop around doing the lock op, while a blocking lock
3031 		 * must wait for the lock op to succeed.
3032 		 */
3033 		do {
3034 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3035 			    ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3036 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3037 			    ap->a_op == F_SETLK) {
3038 				NFSVOPUNLOCK(vp, 0);
3039 				error = nfs_catnap(PZERO | PCATCH, ret,
3040 				    "ncladvl");
3041 				if (error)
3042 					return (EINTR);
3043 				NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3044 				if (vp->v_iflag & VI_DOOMED) {
3045 					NFSVOPUNLOCK(vp, 0);
3046 					return (EBADF);
3047 				}
3048 			}
3049 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3050 		     ap->a_op == F_SETLK);
3051 		if (ret == NFSERR_DENIED) {
3052 			NFSVOPUNLOCK(vp, 0);
3053 			return (EAGAIN);
3054 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3055 			NFSVOPUNLOCK(vp, 0);
3056 			return (ret);
3057 		} else if (ret != 0) {
3058 			NFSVOPUNLOCK(vp, 0);
3059 			return (EACCES);
3060 		}
3061 
3062 		/*
3063 		 * Now, if we just got a lock, invalidate data in the buffer
3064 		 * cache, as required, so that the coherency conforms with
3065 		 * RFC3530 Sec. 9.3.2.
3066 		 */
3067 		if (ap->a_op == F_SETLK) {
3068 			if ((np->n_flag & NMODIFIED) == 0) {
3069 				np->n_attrstamp = 0;
3070 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3071 				ret = VOP_GETATTR(vp, &va, cred);
3072 			}
3073 			if ((np->n_flag & NMODIFIED) || ret ||
3074 			    np->n_change != va.va_filerev) {
3075 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3076 				np->n_attrstamp = 0;
3077 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3078 				ret = VOP_GETATTR(vp, &va, cred);
3079 				if (!ret) {
3080 					np->n_mtime = va.va_mtime;
3081 					np->n_change = va.va_filerev;
3082 				}
3083 			}
3084 		}
3085 		NFSVOPUNLOCK(vp, 0);
3086 		return (0);
3087 	} else if (!NFS_ISV4(vp)) {
3088 		error = NFSVOPLOCK(vp, LK_SHARED);
3089 		if (error)
3090 			return (error);
3091 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3092 			size = VTONFS(vp)->n_size;
3093 			NFSVOPUNLOCK(vp, 0);
3094 			error = lf_advlock(ap, &(vp->v_lockf), size);
3095 		} else {
3096 			if (nfs_advlock_p != NULL)
3097 				error = nfs_advlock_p(ap);
3098 			else {
3099 				NFSVOPUNLOCK(vp, 0);
3100 				error = ENOLCK;
3101 			}
3102 		}
3103 	}
3104 	return (error);
3105 }
3106 
3107 /*
3108  * NFS advisory byte-level locks.
3109  */
3110 static int
3111 nfs_advlockasync(struct vop_advlockasync_args *ap)
3112 {
3113 	struct vnode *vp = ap->a_vp;
3114 	u_quad_t size;
3115 	int error;
3116 
3117 	if (NFS_ISV4(vp))
3118 		return (EOPNOTSUPP);
3119 	error = NFSVOPLOCK(vp, LK_SHARED);
3120 	if (error)
3121 		return (error);
3122 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3123 		size = VTONFS(vp)->n_size;
3124 		NFSVOPUNLOCK(vp, 0);
3125 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
3126 	} else {
3127 		NFSVOPUNLOCK(vp, 0);
3128 		error = EOPNOTSUPP;
3129 	}
3130 	return (error);
3131 }
3132 
3133 /*
3134  * Print out the contents of an nfsnode.
3135  */
3136 static int
3137 nfs_print(struct vop_print_args *ap)
3138 {
3139 	struct vnode *vp = ap->a_vp;
3140 	struct nfsnode *np = VTONFS(vp);
3141 
3142 	ncl_printf("\tfileid %ld fsid 0x%x",
3143 	   np->n_vattr.na_fileid, np->n_vattr.na_fsid);
3144 	if (vp->v_type == VFIFO)
3145 		fifo_printinfo(vp);
3146 	printf("\n");
3147 	return (0);
3148 }
3149 
3150 /*
3151  * This is the "real" nfs::bwrite(struct buf*).
3152  * We set B_CACHE if this is a VMIO buffer.
3153  */
3154 int
3155 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3156 {
3157 	int s;
3158 	int oldflags = bp->b_flags;
3159 #if 0
3160 	int retv = 1;
3161 	off_t off;
3162 #endif
3163 
3164 	BUF_ASSERT_HELD(bp);
3165 
3166 	if (bp->b_flags & B_INVAL) {
3167 		brelse(bp);
3168 		return(0);
3169 	}
3170 
3171 	bp->b_flags |= B_CACHE;
3172 
3173 	/*
3174 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3175 	 */
3176 
3177 	s = splbio();
3178 	bundirty(bp);
3179 	bp->b_flags &= ~B_DONE;
3180 	bp->b_ioflags &= ~BIO_ERROR;
3181 	bp->b_iocmd = BIO_WRITE;
3182 
3183 	bufobj_wref(bp->b_bufobj);
3184 	curthread->td_ru.ru_oublock++;
3185 	splx(s);
3186 
3187 	/*
3188 	 * Note: to avoid loopback deadlocks, we do not
3189 	 * assign b_runningbufspace.
3190 	 */
3191 	vfs_busy_pages(bp, 1);
3192 
3193 	BUF_KERNPROC(bp);
3194 	bp->b_iooffset = dbtob(bp->b_blkno);
3195 	bstrategy(bp);
3196 
3197 	if( (oldflags & B_ASYNC) == 0) {
3198 		int rtval = bufwait(bp);
3199 
3200 		if (oldflags & B_DELWRI) {
3201 			s = splbio();
3202 			reassignbuf(bp);
3203 			splx(s);
3204 		}
3205 		brelse(bp);
3206 		return (rtval);
3207 	}
3208 
3209 	return (0);
3210 }
3211 
3212 /*
3213  * nfs special file access vnode op.
3214  * Essentially just get vattr and then imitate iaccess() since the device is
3215  * local to the client.
3216  */
3217 static int
3218 nfsspec_access(struct vop_access_args *ap)
3219 {
3220 	struct vattr *vap;
3221 	struct ucred *cred = ap->a_cred;
3222 	struct vnode *vp = ap->a_vp;
3223 	accmode_t accmode = ap->a_accmode;
3224 	struct vattr vattr;
3225 	int error;
3226 
3227 	/*
3228 	 * Disallow write attempts on filesystems mounted read-only;
3229 	 * unless the file is a socket, fifo, or a block or character
3230 	 * device resident on the filesystem.
3231 	 */
3232 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3233 		switch (vp->v_type) {
3234 		case VREG:
3235 		case VDIR:
3236 		case VLNK:
3237 			return (EROFS);
3238 		default:
3239 			break;
3240 		}
3241 	}
3242 	vap = &vattr;
3243 	error = VOP_GETATTR(vp, vap, cred);
3244 	if (error)
3245 		goto out;
3246 	error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3247 	    accmode, cred, NULL);
3248 out:
3249 	return error;
3250 }
3251 
3252 /*
3253  * Read wrapper for fifos.
3254  */
3255 static int
3256 nfsfifo_read(struct vop_read_args *ap)
3257 {
3258 	struct nfsnode *np = VTONFS(ap->a_vp);
3259 	int error;
3260 
3261 	/*
3262 	 * Set access flag.
3263 	 */
3264 	mtx_lock(&np->n_mtx);
3265 	np->n_flag |= NACC;
3266 	vfs_timestamp(&np->n_atim);
3267 	mtx_unlock(&np->n_mtx);
3268 	error = fifo_specops.vop_read(ap);
3269 	return error;
3270 }
3271 
3272 /*
3273  * Write wrapper for fifos.
3274  */
3275 static int
3276 nfsfifo_write(struct vop_write_args *ap)
3277 {
3278 	struct nfsnode *np = VTONFS(ap->a_vp);
3279 
3280 	/*
3281 	 * Set update flag.
3282 	 */
3283 	mtx_lock(&np->n_mtx);
3284 	np->n_flag |= NUPD;
3285 	vfs_timestamp(&np->n_mtim);
3286 	mtx_unlock(&np->n_mtx);
3287 	return(fifo_specops.vop_write(ap));
3288 }
3289 
3290 /*
3291  * Close wrapper for fifos.
3292  *
3293  * Update the times on the nfsnode then do fifo close.
3294  */
3295 static int
3296 nfsfifo_close(struct vop_close_args *ap)
3297 {
3298 	struct vnode *vp = ap->a_vp;
3299 	struct nfsnode *np = VTONFS(vp);
3300 	struct vattr vattr;
3301 	struct timespec ts;
3302 
3303 	mtx_lock(&np->n_mtx);
3304 	if (np->n_flag & (NACC | NUPD)) {
3305 		vfs_timestamp(&ts);
3306 		if (np->n_flag & NACC)
3307 			np->n_atim = ts;
3308 		if (np->n_flag & NUPD)
3309 			np->n_mtim = ts;
3310 		np->n_flag |= NCHG;
3311 		if (vrefcnt(vp) == 1 &&
3312 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3313 			VATTR_NULL(&vattr);
3314 			if (np->n_flag & NACC)
3315 				vattr.va_atime = np->n_atim;
3316 			if (np->n_flag & NUPD)
3317 				vattr.va_mtime = np->n_mtim;
3318 			mtx_unlock(&np->n_mtx);
3319 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3320 			goto out;
3321 		}
3322 	}
3323 	mtx_unlock(&np->n_mtx);
3324 out:
3325 	return (fifo_specops.vop_close(ap));
3326 }
3327 
3328 /*
3329  * Just call ncl_writebp() with the force argument set to 1.
3330  *
3331  * NOTE: B_DONE may or may not be set in a_bp on call.
3332  */
3333 static int
3334 nfs_bwrite(struct buf *bp)
3335 {
3336 
3337 	return (ncl_writebp(bp, 1, curthread));
3338 }
3339 
3340 struct buf_ops buf_ops_newnfs = {
3341 	.bop_name	=	"buf_ops_nfs",
3342 	.bop_write	=	nfs_bwrite,
3343 	.bop_strategy	=	bufstrategy,
3344 	.bop_sync	=	bufsync,
3345 	.bop_bdflush	=	bufbdflush,
3346 };
3347 
3348 /*
3349  * Cloned from vop_stdlock(), and then the ugly hack added.
3350  */
3351 static int
3352 nfs_lock1(struct vop_lock1_args *ap)
3353 {
3354 	struct vnode *vp = ap->a_vp;
3355 	int error = 0;
3356 
3357 	/*
3358 	 * Since vfs_hash_get() calls vget() and it will no longer work
3359 	 * for FreeBSD8 with flags == 0, I can only think of this horrible
3360 	 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3361 	 * and then handle it here. All I want for this case is a v_usecount
3362 	 * on the vnode to use for recovery, while another thread might
3363 	 * hold a lock on the vnode. I have the other threads blocked, so
3364 	 * there isn't any race problem.
3365 	 */
3366 	if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3367 		if ((ap->a_flags & LK_INTERLOCK) == 0)
3368 			panic("ncllock1");
3369 		if ((vp->v_iflag & VI_DOOMED))
3370 			error = ENOENT;
3371 		VI_UNLOCK(vp);
3372 		return (error);
3373 	}
3374 	return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3375 	    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3376 	    ap->a_line));
3377 }
3378 
3379 static int
3380 nfs_getacl(struct vop_getacl_args *ap)
3381 {
3382 	int error;
3383 
3384 	if (ap->a_type != ACL_TYPE_NFS4)
3385 		return (EOPNOTSUPP);
3386 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3387 	    NULL);
3388 	if (error > NFSERR_STALE) {
3389 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3390 		error = EPERM;
3391 	}
3392 	return (error);
3393 }
3394 
3395 static int
3396 nfs_setacl(struct vop_setacl_args *ap)
3397 {
3398 	int error;
3399 
3400 	if (ap->a_type != ACL_TYPE_NFS4)
3401 		return (EOPNOTSUPP);
3402 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3403 	    NULL);
3404 	if (error > NFSERR_STALE) {
3405 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3406 		error = EPERM;
3407 	}
3408 	return (error);
3409 }
3410 
3411 /*
3412  * Return POSIX pathconf information applicable to nfs filesystems.
3413  */
3414 static int
3415 nfs_pathconf(struct vop_pathconf_args *ap)
3416 {
3417 	struct nfsv3_pathconf pc;
3418 	struct nfsvattr nfsva;
3419 	struct vnode *vp = ap->a_vp;
3420 	struct thread *td = curthread;
3421 	int attrflag, error;
3422 
3423 	if (NFS_ISV4(vp) || (NFS_ISV3(vp) && (ap->a_name == _PC_LINK_MAX ||
3424 	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3425 	    ap->a_name == _PC_NO_TRUNC))) {
3426 		/*
3427 		 * Since only the above 4 a_names are returned by the NFSv3
3428 		 * Pathconf RPC, there is no point in doing it for others.
3429 		 */
3430 		error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3431 		    &attrflag, NULL);
3432 		if (attrflag != 0)
3433 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3434 			    1);
3435 		if (error != 0)
3436 			return (error);
3437 	} else {
3438 		/*
3439 		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3440 		 * just fake them.
3441 		 */
3442 		pc.pc_linkmax = LINK_MAX;
3443 		pc.pc_namemax = NFS_MAXNAMLEN;
3444 		pc.pc_notrunc = 1;
3445 		pc.pc_chownrestricted = 1;
3446 		pc.pc_caseinsensitive = 0;
3447 		pc.pc_casepreserving = 1;
3448 		error = 0;
3449 	}
3450 	switch (ap->a_name) {
3451 	case _PC_LINK_MAX:
3452 		*ap->a_retval = pc.pc_linkmax;
3453 		break;
3454 	case _PC_NAME_MAX:
3455 		*ap->a_retval = pc.pc_namemax;
3456 		break;
3457 	case _PC_PATH_MAX:
3458 		*ap->a_retval = PATH_MAX;
3459 		break;
3460 	case _PC_PIPE_BUF:
3461 		*ap->a_retval = PIPE_BUF;
3462 		break;
3463 	case _PC_CHOWN_RESTRICTED:
3464 		*ap->a_retval = pc.pc_chownrestricted;
3465 		break;
3466 	case _PC_NO_TRUNC:
3467 		*ap->a_retval = pc.pc_notrunc;
3468 		break;
3469 	case _PC_ACL_EXTENDED:
3470 		*ap->a_retval = 0;
3471 		break;
3472 	case _PC_ACL_NFS4:
3473 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3474 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3475 			*ap->a_retval = 1;
3476 		else
3477 			*ap->a_retval = 0;
3478 		break;
3479 	case _PC_ACL_PATH_MAX:
3480 		if (NFS_ISV4(vp))
3481 			*ap->a_retval = ACL_MAX_ENTRIES;
3482 		else
3483 			*ap->a_retval = 3;
3484 		break;
3485 	case _PC_MAC_PRESENT:
3486 		*ap->a_retval = 0;
3487 		break;
3488 	case _PC_ASYNC_IO:
3489 		/* _PC_ASYNC_IO should have been handled by upper layers. */
3490 		KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3491 		error = EINVAL;
3492 		break;
3493 	case _PC_PRIO_IO:
3494 		*ap->a_retval = 0;
3495 		break;
3496 	case _PC_SYNC_IO:
3497 		*ap->a_retval = 0;
3498 		break;
3499 	case _PC_ALLOC_SIZE_MIN:
3500 		*ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3501 		break;
3502 	case _PC_FILESIZEBITS:
3503 		if (NFS_ISV34(vp))
3504 			*ap->a_retval = 64;
3505 		else
3506 			*ap->a_retval = 32;
3507 		break;
3508 	case _PC_REC_INCR_XFER_SIZE:
3509 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3510 		break;
3511 	case _PC_REC_MAX_XFER_SIZE:
3512 		*ap->a_retval = -1; /* means ``unlimited'' */
3513 		break;
3514 	case _PC_REC_MIN_XFER_SIZE:
3515 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3516 		break;
3517 	case _PC_REC_XFER_ALIGN:
3518 		*ap->a_retval = PAGE_SIZE;
3519 		break;
3520 	case _PC_SYMLINK_MAX:
3521 		*ap->a_retval = NFS_MAXPATHLEN;
3522 		break;
3523 
3524 	default:
3525 		error = EINVAL;
3526 		break;
3527 	}
3528 	return (error);
3529 }
3530 
3531