xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision bb92cd7bcd16f3f36cdbda18d8193619892715fb)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	from nfs_vnops.c	8.16 (Berkeley) 5/27/95
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 /*
41  * vnode op calls for Sun NFS version 2, 3 and 4
42  */
43 
44 #include "opt_inet.h"
45 
46 #include <sys/param.h>
47 #include <sys/kernel.h>
48 #include <sys/systm.h>
49 #include <sys/resourcevar.h>
50 #include <sys/proc.h>
51 #include <sys/mount.h>
52 #include <sys/bio.h>
53 #include <sys/buf.h>
54 #include <sys/extattr.h>
55 #include <sys/filio.h>
56 #include <sys/jail.h>
57 #include <sys/malloc.h>
58 #include <sys/mbuf.h>
59 #include <sys/namei.h>
60 #include <sys/socket.h>
61 #include <sys/vnode.h>
62 #include <sys/dirent.h>
63 #include <sys/fcntl.h>
64 #include <sys/lockf.h>
65 #include <sys/stat.h>
66 #include <sys/sysctl.h>
67 #include <sys/signalvar.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_object.h>
72 
73 #include <fs/nfs/nfsport.h>
74 #include <fs/nfsclient/nfsnode.h>
75 #include <fs/nfsclient/nfsmount.h>
76 #include <fs/nfsclient/nfs.h>
77 #include <fs/nfsclient/nfs_kdtrace.h>
78 
79 #include <net/if.h>
80 #include <netinet/in.h>
81 #include <netinet/in_var.h>
82 
83 #include <nfs/nfs_lock.h>
84 
85 #ifdef KDTRACE_HOOKS
86 #include <sys/dtrace_bsd.h>
87 
88 dtrace_nfsclient_accesscache_flush_probe_func_t
89 		dtrace_nfscl_accesscache_flush_done_probe;
90 uint32_t	nfscl_accesscache_flush_done_id;
91 
92 dtrace_nfsclient_accesscache_get_probe_func_t
93 		dtrace_nfscl_accesscache_get_hit_probe,
94 		dtrace_nfscl_accesscache_get_miss_probe;
95 uint32_t	nfscl_accesscache_get_hit_id;
96 uint32_t	nfscl_accesscache_get_miss_id;
97 
98 dtrace_nfsclient_accesscache_load_probe_func_t
99 		dtrace_nfscl_accesscache_load_done_probe;
100 uint32_t	nfscl_accesscache_load_done_id;
101 #endif /* !KDTRACE_HOOKS */
102 
103 /* Defs */
104 #define	TRUE	1
105 #define	FALSE	0
106 
107 extern struct nfsstatsv1 nfsstatsv1;
108 extern int nfsrv_useacl;
109 extern int nfscl_debuglevel;
110 MALLOC_DECLARE(M_NEWNFSREQ);
111 
112 static vop_read_t	nfsfifo_read;
113 static vop_write_t	nfsfifo_write;
114 static vop_close_t	nfsfifo_close;
115 static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
116 		    struct thread *);
117 static vop_lookup_t	nfs_lookup;
118 static vop_create_t	nfs_create;
119 static vop_mknod_t	nfs_mknod;
120 static vop_open_t	nfs_open;
121 static vop_pathconf_t	nfs_pathconf;
122 static vop_close_t	nfs_close;
123 static vop_access_t	nfs_access;
124 static vop_getattr_t	nfs_getattr;
125 static vop_setattr_t	nfs_setattr;
126 static vop_read_t	nfs_read;
127 static vop_fsync_t	nfs_fsync;
128 static vop_remove_t	nfs_remove;
129 static vop_link_t	nfs_link;
130 static vop_rename_t	nfs_rename;
131 static vop_mkdir_t	nfs_mkdir;
132 static vop_rmdir_t	nfs_rmdir;
133 static vop_symlink_t	nfs_symlink;
134 static vop_readdir_t	nfs_readdir;
135 static vop_strategy_t	nfs_strategy;
136 static	int	nfs_lookitup(struct vnode *, char *, int,
137 		    struct ucred *, struct thread *, struct nfsnode **);
138 static	int	nfs_sillyrename(struct vnode *, struct vnode *,
139 		    struct componentname *);
140 static vop_access_t	nfsspec_access;
141 static vop_readlink_t	nfs_readlink;
142 static vop_print_t	nfs_print;
143 static vop_advlock_t	nfs_advlock;
144 static vop_advlockasync_t nfs_advlockasync;
145 static vop_getacl_t nfs_getacl;
146 static vop_setacl_t nfs_setacl;
147 static vop_advise_t nfs_advise;
148 static vop_allocate_t nfs_allocate;
149 static vop_deallocate_t nfs_deallocate;
150 static vop_copy_file_range_t nfs_copy_file_range;
151 static vop_ioctl_t nfs_ioctl;
152 static vop_getextattr_t nfs_getextattr;
153 static vop_setextattr_t nfs_setextattr;
154 static vop_listextattr_t nfs_listextattr;
155 static vop_deleteextattr_t nfs_deleteextattr;
156 static vop_lock1_t	nfs_lock;
157 
158 /*
159  * Global vfs data structures for nfs
160  */
161 
162 static struct vop_vector newnfs_vnodeops_nosig = {
163 	.vop_default =		&default_vnodeops,
164 	.vop_access =		nfs_access,
165 	.vop_advlock =		nfs_advlock,
166 	.vop_advlockasync =	nfs_advlockasync,
167 	.vop_close =		nfs_close,
168 	.vop_create =		nfs_create,
169 	.vop_fsync =		nfs_fsync,
170 	.vop_getattr =		nfs_getattr,
171 	.vop_getpages =		ncl_getpages,
172 	.vop_putpages =		ncl_putpages,
173 	.vop_inactive =		ncl_inactive,
174 	.vop_link =		nfs_link,
175 	.vop_lock1 =		nfs_lock,
176 	.vop_lookup =		nfs_lookup,
177 	.vop_mkdir =		nfs_mkdir,
178 	.vop_mknod =		nfs_mknod,
179 	.vop_open =		nfs_open,
180 	.vop_pathconf =		nfs_pathconf,
181 	.vop_print =		nfs_print,
182 	.vop_read =		nfs_read,
183 	.vop_readdir =		nfs_readdir,
184 	.vop_readlink =		nfs_readlink,
185 	.vop_reclaim =		ncl_reclaim,
186 	.vop_remove =		nfs_remove,
187 	.vop_rename =		nfs_rename,
188 	.vop_rmdir =		nfs_rmdir,
189 	.vop_setattr =		nfs_setattr,
190 	.vop_strategy =		nfs_strategy,
191 	.vop_symlink =		nfs_symlink,
192 	.vop_write =		ncl_write,
193 	.vop_getacl =		nfs_getacl,
194 	.vop_setacl =		nfs_setacl,
195 	.vop_advise =		nfs_advise,
196 	.vop_allocate =		nfs_allocate,
197 	.vop_deallocate =	nfs_deallocate,
198 	.vop_copy_file_range =	nfs_copy_file_range,
199 	.vop_ioctl =		nfs_ioctl,
200 	.vop_getextattr =	nfs_getextattr,
201 	.vop_setextattr =	nfs_setextattr,
202 	.vop_listextattr =	nfs_listextattr,
203 	.vop_deleteextattr =	nfs_deleteextattr,
204 };
205 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig);
206 
207 static int
208 nfs_vnodeops_bypass(struct vop_generic_args *a)
209 {
210 
211 	return (vop_sigdefer(&newnfs_vnodeops_nosig, a));
212 }
213 
214 struct vop_vector newnfs_vnodeops = {
215 	.vop_default =		&default_vnodeops,
216 	.vop_bypass =		nfs_vnodeops_bypass,
217 };
218 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops);
219 
220 static struct vop_vector newnfs_fifoops_nosig = {
221 	.vop_default =		&fifo_specops,
222 	.vop_access =		nfsspec_access,
223 	.vop_close =		nfsfifo_close,
224 	.vop_fsync =		nfs_fsync,
225 	.vop_getattr =		nfs_getattr,
226 	.vop_inactive =		ncl_inactive,
227 	.vop_pathconf =		nfs_pathconf,
228 	.vop_print =		nfs_print,
229 	.vop_read =		nfsfifo_read,
230 	.vop_reclaim =		ncl_reclaim,
231 	.vop_setattr =		nfs_setattr,
232 	.vop_write =		nfsfifo_write,
233 };
234 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig);
235 
236 static int
237 nfs_fifoops_bypass(struct vop_generic_args *a)
238 {
239 
240 	return (vop_sigdefer(&newnfs_fifoops_nosig, a));
241 }
242 
243 struct vop_vector newnfs_fifoops = {
244 	.vop_default =		&default_vnodeops,
245 	.vop_bypass =		nfs_fifoops_bypass,
246 };
247 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops);
248 
249 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
250     struct componentname *cnp, struct vattr *vap);
251 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
252     int namelen, struct ucred *cred, struct thread *td);
253 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
254     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
255     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
256 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
257     struct componentname *scnp, struct sillyrename *sp);
258 
259 /*
260  * Global variables
261  */
262 SYSCTL_DECL(_vfs_nfs);
263 
264 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
265 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
266 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
267 
268 static int	nfs_prime_access_cache = 0;
269 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
270 	   &nfs_prime_access_cache, 0,
271 	   "Prime NFS ACCESS cache when fetching attributes");
272 
273 static int	newnfs_commit_on_close = 0;
274 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
275     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
276 
277 static int	nfs_clean_pages_on_close = 1;
278 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
279 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
280 
281 int newnfs_directio_enable = 0;
282 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
283 	   &newnfs_directio_enable, 0, "Enable NFS directio");
284 
285 int nfs_keep_dirty_on_error;
286 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
287     &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
288 
289 /*
290  * This sysctl allows other processes to mmap a file that has been opened
291  * O_DIRECT by a process.  In general, having processes mmap the file while
292  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
293  * this by default to prevent DoS attacks - to prevent a malicious user from
294  * opening up files O_DIRECT preventing other users from mmap'ing these
295  * files.  "Protected" environments where stricter consistency guarantees are
296  * required can disable this knob.  The process that opened the file O_DIRECT
297  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
298  * meaningful.
299  */
300 int newnfs_directio_allow_mmap = 1;
301 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
302 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
303 
304 static uint64_t	nfs_maxalloclen = 64 * 1024 * 1024;
305 SYSCTL_U64(_vfs_nfs, OID_AUTO, maxalloclen, CTLFLAG_RW,
306 	   &nfs_maxalloclen, 0, "NFS max allocate/deallocate length");
307 
308 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
309 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
310 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
311 
312 /*
313  * SMP Locking Note :
314  * The list of locks after the description of the lock is the ordering
315  * of other locks acquired with the lock held.
316  * np->n_mtx : Protects the fields in the nfsnode.
317        VM Object Lock
318        VI_MTX (acquired indirectly)
319  * nmp->nm_mtx : Protects the fields in the nfsmount.
320        rep->r_mtx
321  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
322  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
323        nmp->nm_mtx
324        rep->r_mtx
325  * rep->r_mtx : Protects the fields in an nfsreq.
326  */
327 
328 static int
329 nfs_lock(struct vop_lock1_args *ap)
330 {
331 	struct vnode *vp;
332 	struct nfsnode *np;
333 	u_quad_t nsize;
334 	int error, lktype;
335 	bool onfault;
336 
337 	vp = ap->a_vp;
338 	lktype = ap->a_flags & LK_TYPE_MASK;
339 	error = VOP_LOCK1_APV(&default_vnodeops, ap);
340 	if (error != 0 || vp->v_op != &newnfs_vnodeops)
341 		return (error);
342 	np = VTONFS(vp);
343 	if (np == NULL)
344 		return (0);
345 	NFSLOCKNODE(np);
346 	if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED &&
347 	    lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE &&
348 	    lktype != LK_TRYUPGRADE)) {
349 		NFSUNLOCKNODE(np);
350 		return (0);
351 	}
352 	onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT &&
353 	    (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE &&
354 	    (lktype == LK_SHARED || lktype == LK_EXCLUSIVE);
355 	if (onfault && vp->v_vnlock->lk_recurse == 0) {
356 		/*
357 		 * Force retry in vm_fault(), to make the lock request
358 		 * sleepable, which allows us to piggy-back the
359 		 * sleepable call to vnode_pager_setsize().
360 		 */
361 		NFSUNLOCKNODE(np);
362 		VOP_UNLOCK(vp);
363 		return (EBUSY);
364 	}
365 	if ((ap->a_flags & LK_NOWAIT) != 0 ||
366 	    (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) {
367 		NFSUNLOCKNODE(np);
368 		return (0);
369 	}
370 	if (lktype == LK_SHARED) {
371 		NFSUNLOCKNODE(np);
372 		VOP_UNLOCK(vp);
373 		ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
374 		ap->a_flags |= LK_EXCLUSIVE;
375 		error = VOP_LOCK1_APV(&default_vnodeops, ap);
376 		if (error != 0 || vp->v_op != &newnfs_vnodeops)
377 			return (error);
378 		if (vp->v_data == NULL)
379 			goto downgrade;
380 		MPASS(vp->v_data == np);
381 		NFSLOCKNODE(np);
382 		if ((np->n_flag & NVNSETSZSKIP) == 0) {
383 			NFSUNLOCKNODE(np);
384 			goto downgrade;
385 		}
386 	}
387 	np->n_flag &= ~NVNSETSZSKIP;
388 	nsize = np->n_size;
389 	NFSUNLOCKNODE(np);
390 	vnode_pager_setsize(vp, nsize);
391 downgrade:
392 	if (lktype == LK_SHARED) {
393 		ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
394 		ap->a_flags |= LK_DOWNGRADE;
395 		(void)VOP_LOCK1_APV(&default_vnodeops, ap);
396 	}
397 	return (0);
398 }
399 
400 static int
401 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
402     struct ucred *cred, u_int32_t *retmode)
403 {
404 	int error = 0, attrflag, i, lrupos;
405 	u_int32_t rmode;
406 	struct nfsnode *np = VTONFS(vp);
407 	struct nfsvattr nfsva;
408 
409 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
410 	    &rmode, NULL);
411 	if (attrflag)
412 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
413 	if (!error) {
414 		lrupos = 0;
415 		NFSLOCKNODE(np);
416 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
417 			if (np->n_accesscache[i].uid == cred->cr_uid) {
418 				np->n_accesscache[i].mode = rmode;
419 				np->n_accesscache[i].stamp = time_second;
420 				break;
421 			}
422 			if (i > 0 && np->n_accesscache[i].stamp <
423 			    np->n_accesscache[lrupos].stamp)
424 				lrupos = i;
425 		}
426 		if (i == NFS_ACCESSCACHESIZE) {
427 			np->n_accesscache[lrupos].uid = cred->cr_uid;
428 			np->n_accesscache[lrupos].mode = rmode;
429 			np->n_accesscache[lrupos].stamp = time_second;
430 		}
431 		NFSUNLOCKNODE(np);
432 		if (retmode != NULL)
433 			*retmode = rmode;
434 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
435 	} else if (NFS_ISV4(vp)) {
436 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
437 	}
438 #ifdef KDTRACE_HOOKS
439 	if (error != 0)
440 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
441 		    error);
442 #endif
443 	return (error);
444 }
445 
446 /*
447  * nfs access vnode op.
448  * For nfs version 2, just return ok. File accesses may fail later.
449  * For nfs version 3, use the access rpc to check accessibility. If file modes
450  * are changed on the server, accesses might still fail later.
451  */
452 static int
453 nfs_access(struct vop_access_args *ap)
454 {
455 	struct vnode *vp = ap->a_vp;
456 	int error = 0, i, gotahit;
457 	u_int32_t mode, wmode, rmode;
458 	int v34 = NFS_ISV34(vp);
459 	struct nfsnode *np = VTONFS(vp);
460 
461 	/*
462 	 * Disallow write attempts on filesystems mounted read-only;
463 	 * unless the file is a socket, fifo, or a block or character
464 	 * device resident on the filesystem.
465 	 */
466 	if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
467 	    VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
468 	    VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
469 		switch (vp->v_type) {
470 		case VREG:
471 		case VDIR:
472 		case VLNK:
473 			return (EROFS);
474 		default:
475 			break;
476 		}
477 	}
478 	/*
479 	 * For nfs v3 or v4, check to see if we have done this recently, and if
480 	 * so return our cached result instead of making an ACCESS call.
481 	 * If not, do an access rpc, otherwise you are stuck emulating
482 	 * ufs_access() locally using the vattr. This may not be correct,
483 	 * since the server may apply other access criteria such as
484 	 * client uid-->server uid mapping that we do not know about.
485 	 */
486 	if (v34) {
487 		if (ap->a_accmode & VREAD)
488 			mode = NFSACCESS_READ;
489 		else
490 			mode = 0;
491 		if (vp->v_type != VDIR) {
492 			if (ap->a_accmode & VWRITE)
493 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
494 			if (ap->a_accmode & VAPPEND)
495 				mode |= NFSACCESS_EXTEND;
496 			if (ap->a_accmode & VEXEC)
497 				mode |= NFSACCESS_EXECUTE;
498 			if (ap->a_accmode & VDELETE)
499 				mode |= NFSACCESS_DELETE;
500 		} else {
501 			if (ap->a_accmode & VWRITE)
502 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
503 			if (ap->a_accmode & VAPPEND)
504 				mode |= NFSACCESS_EXTEND;
505 			if (ap->a_accmode & VEXEC)
506 				mode |= NFSACCESS_LOOKUP;
507 			if (ap->a_accmode & VDELETE)
508 				mode |= NFSACCESS_DELETE;
509 			if (ap->a_accmode & VDELETE_CHILD)
510 				mode |= NFSACCESS_MODIFY;
511 		}
512 		/* XXX safety belt, only make blanket request if caching */
513 		if (nfsaccess_cache_timeout > 0) {
514 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
515 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
516 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
517 		} else {
518 			wmode = mode;
519 		}
520 
521 		/*
522 		 * Does our cached result allow us to give a definite yes to
523 		 * this request?
524 		 */
525 		gotahit = 0;
526 		NFSLOCKNODE(np);
527 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
528 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
529 			    if (time_second < (np->n_accesscache[i].stamp
530 				+ nfsaccess_cache_timeout) &&
531 				(np->n_accesscache[i].mode & mode) == mode) {
532 				NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
533 				gotahit = 1;
534 			    }
535 			    break;
536 			}
537 		}
538 		NFSUNLOCKNODE(np);
539 #ifdef KDTRACE_HOOKS
540 		if (gotahit != 0)
541 			KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
542 			    ap->a_cred->cr_uid, mode);
543 		else
544 			KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
545 			    ap->a_cred->cr_uid, mode);
546 #endif
547 		if (gotahit == 0) {
548 			/*
549 			 * Either a no, or a don't know.  Go to the wire.
550 			 */
551 			NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
552 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
553 			    ap->a_cred, &rmode);
554 			if (!error &&
555 			    (rmode & mode) != mode)
556 				error = EACCES;
557 		}
558 		return (error);
559 	} else {
560 		if ((error = nfsspec_access(ap)) != 0) {
561 			return (error);
562 		}
563 		/*
564 		 * Attempt to prevent a mapped root from accessing a file
565 		 * which it shouldn't.  We try to read a byte from the file
566 		 * if the user is root and the file is not zero length.
567 		 * After calling nfsspec_access, we should have the correct
568 		 * file size cached.
569 		 */
570 		NFSLOCKNODE(np);
571 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
572 		    && VTONFS(vp)->n_size > 0) {
573 			struct iovec aiov;
574 			struct uio auio;
575 			char buf[1];
576 
577 			NFSUNLOCKNODE(np);
578 			aiov.iov_base = buf;
579 			aiov.iov_len = 1;
580 			auio.uio_iov = &aiov;
581 			auio.uio_iovcnt = 1;
582 			auio.uio_offset = 0;
583 			auio.uio_resid = 1;
584 			auio.uio_segflg = UIO_SYSSPACE;
585 			auio.uio_rw = UIO_READ;
586 			auio.uio_td = ap->a_td;
587 
588 			if (vp->v_type == VREG)
589 				error = ncl_readrpc(vp, &auio, ap->a_cred);
590 			else if (vp->v_type == VDIR) {
591 				char* bp;
592 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
593 				aiov.iov_base = bp;
594 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
595 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
596 				    ap->a_td);
597 				free(bp, M_TEMP);
598 			} else if (vp->v_type == VLNK)
599 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
600 			else
601 				error = EACCES;
602 		} else
603 			NFSUNLOCKNODE(np);
604 		return (error);
605 	}
606 }
607 
608 /*
609  * nfs open vnode op
610  * Check to see if the type is ok
611  * and that deletion is not in progress.
612  * For paged in text files, you will need to flush the page cache
613  * if consistency is lost.
614  */
615 /* ARGSUSED */
616 static int
617 nfs_open(struct vop_open_args *ap)
618 {
619 	struct vnode *vp = ap->a_vp;
620 	struct nfsnode *np = VTONFS(vp);
621 	struct vattr vattr;
622 	int error;
623 	int fmode = ap->a_mode;
624 	struct ucred *cred;
625 	vm_object_t obj;
626 
627 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
628 		return (EOPNOTSUPP);
629 
630 	/*
631 	 * For NFSv4, we need to do the Open Op before cache validation,
632 	 * so that we conform to RFC3530 Sec. 9.3.1.
633 	 */
634 	if (NFS_ISV4(vp)) {
635 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
636 		if (error) {
637 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
638 			    (gid_t)0);
639 			return (error);
640 		}
641 	}
642 
643 	/*
644 	 * Now, if this Open will be doing reading, re-validate/flush the
645 	 * cache, so that Close/Open coherency is maintained.
646 	 */
647 	NFSLOCKNODE(np);
648 	if (np->n_flag & NMODIFIED) {
649 		NFSUNLOCKNODE(np);
650 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
651 			NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
652 			if (VN_IS_DOOMED(vp))
653 				return (EBADF);
654 		}
655 		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
656 		if (error == EINTR || error == EIO) {
657 			if (NFS_ISV4(vp))
658 				(void) nfsrpc_close(vp, 0, ap->a_td);
659 			return (error);
660 		}
661 		NFSLOCKNODE(np);
662 		np->n_attrstamp = 0;
663 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
664 		if (vp->v_type == VDIR)
665 			np->n_direofoffset = 0;
666 		NFSUNLOCKNODE(np);
667 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
668 		if (error) {
669 			if (NFS_ISV4(vp))
670 				(void) nfsrpc_close(vp, 0, ap->a_td);
671 			return (error);
672 		}
673 		NFSLOCKNODE(np);
674 		np->n_mtime = vattr.va_mtime;
675 		if (NFS_ISV4(vp))
676 			np->n_change = vattr.va_filerev;
677 	} else {
678 		NFSUNLOCKNODE(np);
679 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
680 		if (error) {
681 			if (NFS_ISV4(vp))
682 				(void) nfsrpc_close(vp, 0, ap->a_td);
683 			return (error);
684 		}
685 		NFSLOCKNODE(np);
686 		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
687 		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
688 			if (vp->v_type == VDIR)
689 				np->n_direofoffset = 0;
690 			NFSUNLOCKNODE(np);
691 			if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
692 				NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
693 				if (VN_IS_DOOMED(vp))
694 					return (EBADF);
695 			}
696 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
697 			if (error == EINTR || error == EIO) {
698 				if (NFS_ISV4(vp))
699 					(void) nfsrpc_close(vp, 0, ap->a_td);
700 				return (error);
701 			}
702 			NFSLOCKNODE(np);
703 			np->n_mtime = vattr.va_mtime;
704 			if (NFS_ISV4(vp))
705 				np->n_change = vattr.va_filerev;
706 		}
707 	}
708 
709 	/*
710 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
711 	 */
712 	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
713 	    (vp->v_type == VREG)) {
714 		if (np->n_directio_opens == 0) {
715 			NFSUNLOCKNODE(np);
716 			if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
717 				NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
718 				if (VN_IS_DOOMED(vp))
719 					return (EBADF);
720 			}
721 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
722 			if (error) {
723 				if (NFS_ISV4(vp))
724 					(void) nfsrpc_close(vp, 0, ap->a_td);
725 				return (error);
726 			}
727 			NFSLOCKNODE(np);
728 			np->n_flag |= NNONCACHE;
729 		}
730 		np->n_directio_opens++;
731 	}
732 
733 	/* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
734 	if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
735 		np->n_flag |= NWRITEOPENED;
736 
737 	/*
738 	 * If this is an open for writing, capture a reference to the
739 	 * credentials, so they can be used by ncl_putpages(). Using
740 	 * these write credentials is preferable to the credentials of
741 	 * whatever thread happens to be doing the VOP_PUTPAGES() since
742 	 * the write RPCs are less likely to fail with EACCES.
743 	 */
744 	if ((fmode & FWRITE) != 0) {
745 		cred = np->n_writecred;
746 		np->n_writecred = crhold(ap->a_cred);
747 	} else
748 		cred = NULL;
749 	NFSUNLOCKNODE(np);
750 
751 	if (cred != NULL)
752 		crfree(cred);
753 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
754 
755 	/*
756 	 * If the text file has been mmap'd, flush any dirty pages to the
757 	 * buffer cache and then...
758 	 * Make sure all writes are pushed to the NFS server.  If this is not
759 	 * done, the modify time of the file can change while the text
760 	 * file is being executed.  This will cause the process that is
761 	 * executing the text file to be terminated.
762 	 */
763 	if (vp->v_writecount <= -1) {
764 		if ((obj = vp->v_object) != NULL &&
765 		    vm_object_mightbedirty(obj)) {
766 			if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
767 				NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
768 				if (VN_IS_DOOMED(vp))
769 					return (EBADF);
770 			}
771 			VM_OBJECT_WLOCK(obj);
772 			vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
773 			VM_OBJECT_WUNLOCK(obj);
774 		}
775 
776 		/* Now, flush the buffer cache. */
777 		ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
778 
779 		/* And, finally, make sure that n_mtime is up to date. */
780 		np = VTONFS(vp);
781 		NFSLOCKNODE(np);
782 		np->n_mtime = np->n_vattr.na_mtime;
783 		NFSUNLOCKNODE(np);
784 	}
785 	return (0);
786 }
787 
788 /*
789  * nfs close vnode op
790  * What an NFS client should do upon close after writing is a debatable issue.
791  * Most NFS clients push delayed writes to the server upon close, basically for
792  * two reasons:
793  * 1 - So that any write errors may be reported back to the client process
794  *     doing the close system call. By far the two most likely errors are
795  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
796  * 2 - To put a worst case upper bound on cache inconsistency between
797  *     multiple clients for the file.
798  * There is also a consistency problem for Version 2 of the protocol w.r.t.
799  * not being able to tell if other clients are writing a file concurrently,
800  * since there is no way of knowing if the changed modify time in the reply
801  * is only due to the write for this client.
802  * (NFS Version 3 provides weak cache consistency data in the reply that
803  *  should be sufficient to detect and handle this case.)
804  *
805  * The current code does the following:
806  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
807  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
808  *                     or commit them (this satisfies 1 and 2 except for the
809  *                     case where the server crashes after this close but
810  *                     before the commit RPC, which is felt to be "good
811  *                     enough". Changing the last argument to ncl_flush() to
812  *                     a 1 would force a commit operation, if it is felt a
813  *                     commit is necessary now.
814  * for NFS Version 4 - flush the dirty buffers and commit them, if
815  *		       nfscl_mustflush() says this is necessary.
816  *                     It is necessary if there is no write delegation held,
817  *                     in order to satisfy open/close coherency.
818  *                     If the file isn't cached on local stable storage,
819  *                     it may be necessary in order to detect "out of space"
820  *                     errors from the server, if the write delegation
821  *                     issued by the server doesn't allow the file to grow.
822  */
823 /* ARGSUSED */
824 static int
825 nfs_close(struct vop_close_args *ap)
826 {
827 	struct vnode *vp = ap->a_vp;
828 	struct nfsnode *np = VTONFS(vp);
829 	struct nfsvattr nfsva;
830 	struct ucred *cred;
831 	int error = 0, ret, localcred = 0;
832 	int fmode = ap->a_fflag;
833 
834 	if (NFSCL_FORCEDISM(vp->v_mount))
835 		return (0);
836 	/*
837 	 * During shutdown, a_cred isn't valid, so just use root.
838 	 */
839 	if (ap->a_cred == NOCRED) {
840 		cred = newnfs_getcred();
841 		localcred = 1;
842 	} else {
843 		cred = ap->a_cred;
844 	}
845 	if (vp->v_type == VREG) {
846 	    /*
847 	     * Examine and clean dirty pages, regardless of NMODIFIED.
848 	     * This closes a major hole in close-to-open consistency.
849 	     * We want to push out all dirty pages (and buffers) on
850 	     * close, regardless of whether they were dirtied by
851 	     * mmap'ed writes or via write().
852 	     */
853 	    if (nfs_clean_pages_on_close && vp->v_object) {
854 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
855 			NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
856 			if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK)
857 				return (EBADF);
858 		}
859 		VM_OBJECT_WLOCK(vp->v_object);
860 		vm_object_page_clean(vp->v_object, 0, 0, 0);
861 		VM_OBJECT_WUNLOCK(vp->v_object);
862 	    }
863 	    NFSLOCKNODE(np);
864 	    if (np->n_flag & NMODIFIED) {
865 		NFSUNLOCKNODE(np);
866 		if (NFS_ISV3(vp)) {
867 		    /*
868 		     * Under NFSv3 we have dirty buffers to dispose of.  We
869 		     * must flush them to the NFS server.  We have the option
870 		     * of waiting all the way through the commit rpc or just
871 		     * waiting for the initial write.  The default is to only
872 		     * wait through the initial write so the data is in the
873 		     * server's cache, which is roughly similar to the state
874 		     * a standard disk subsystem leaves the file in on close().
875 		     *
876 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
877 		     * potential races with other processes, and certainly
878 		     * cannot clear it if we don't commit.
879 		     * These races occur when there is no longer the old
880 		     * traditional vnode locking implemented for Vnode Ops.
881 		     */
882 		    int cm = newnfs_commit_on_close ? 1 : 0;
883 		    if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
884 			    NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
885 			    if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK)
886 				    return (EBADF);
887 		    }
888 		    error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
889 		    /* np->n_flag &= ~NMODIFIED; */
890 		} else if (NFS_ISV4(vp)) {
891 			if (nfscl_mustflush(vp) != 0) {
892 				int cm = newnfs_commit_on_close ? 1 : 0;
893 				if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
894 					NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
895 					if (VN_IS_DOOMED(vp) && ap->a_fflag !=
896 					    FNONBLOCK)
897 						return (EBADF);
898 				}
899 				error = ncl_flush(vp, MNT_WAIT, ap->a_td,
900 				    cm, 0);
901 				/*
902 				 * as above w.r.t races when clearing
903 				 * NMODIFIED.
904 				 * np->n_flag &= ~NMODIFIED;
905 				 */
906 			}
907 		} else {
908 			if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
909 				NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
910 				if (VN_IS_DOOMED(vp) && ap->a_fflag !=
911 				    FNONBLOCK)
912 					return (EBADF);
913 			}
914 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
915 		}
916 		NFSLOCKNODE(np);
917 	    }
918  	    /*
919  	     * Invalidate the attribute cache in all cases.
920  	     * An open is going to fetch fresh attrs any way, other procs
921  	     * on this node that have file open will be forced to do an
922  	     * otw attr fetch, but this is safe.
923 	     * --> A user found that their RPC count dropped by 20% when
924 	     *     this was commented out and I can't see any requirement
925 	     *     for it, so I've disabled it when negative lookups are
926 	     *     enabled. (What does this have to do with negative lookup
927 	     *     caching? Well nothing, except it was reported by the
928 	     *     same user that needed negative lookup caching and I wanted
929 	     *     there to be a way to disable it to see if it
930 	     *     is the cause of some caching/coherency issue that might
931 	     *     crop up.)
932  	     */
933 	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
934 		    np->n_attrstamp = 0;
935 		    KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
936 	    }
937 	    if (np->n_flag & NWRITEERR) {
938 		np->n_flag &= ~NWRITEERR;
939 		error = np->n_error;
940 	    }
941 	    NFSUNLOCKNODE(np);
942 	}
943 
944 	if (NFS_ISV4(vp)) {
945 		/*
946 		 * Get attributes so "change" is up to date.
947 		 */
948 		if (error == 0 && nfscl_mustflush(vp) != 0 &&
949 		    vp->v_type == VREG &&
950 		    (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
951 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
952 			    NULL);
953 			if (!ret) {
954 				np->n_change = nfsva.na_filerev;
955 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
956 				    NULL, 0, 0);
957 			}
958 		}
959 
960 		/*
961 		 * and do the close.
962 		 */
963 		ret = nfsrpc_close(vp, 0, ap->a_td);
964 		if (!error && ret)
965 			error = ret;
966 		if (error)
967 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
968 			    (gid_t)0);
969 	}
970 	if (newnfs_directio_enable)
971 		KASSERT((np->n_directio_asyncwr == 0),
972 			("nfs_close: dirty unflushed (%d) directio buffers\n",
973 			 np->n_directio_asyncwr));
974 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
975 		NFSLOCKNODE(np);
976 		KASSERT((np->n_directio_opens > 0),
977 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
978 		np->n_directio_opens--;
979 		if (np->n_directio_opens == 0)
980 			np->n_flag &= ~NNONCACHE;
981 		NFSUNLOCKNODE(np);
982 	}
983 	if (localcred)
984 		NFSFREECRED(cred);
985 	return (error);
986 }
987 
988 /*
989  * nfs getattr call from vfs.
990  */
991 static int
992 nfs_getattr(struct vop_getattr_args *ap)
993 {
994 	struct vnode *vp = ap->a_vp;
995 	struct thread *td = curthread;	/* XXX */
996 	struct nfsnode *np = VTONFS(vp);
997 	int error = 0;
998 	struct nfsvattr nfsva;
999 	struct vattr *vap = ap->a_vap;
1000 	struct vattr vattr;
1001 
1002 	/*
1003 	 * Update local times for special files.
1004 	 */
1005 	NFSLOCKNODE(np);
1006 	if (np->n_flag & (NACC | NUPD))
1007 		np->n_flag |= NCHG;
1008 	NFSUNLOCKNODE(np);
1009 	/*
1010 	 * First look in the cache.
1011 	 */
1012 	if (ncl_getattrcache(vp, &vattr) == 0) {
1013 		ncl_copy_vattr(vap, &vattr);
1014 
1015 		/*
1016 		 * Get the local modify time for the case of a write
1017 		 * delegation.
1018 		 */
1019 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
1020 		return (0);
1021 	}
1022 
1023 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
1024 	    nfsaccess_cache_timeout > 0) {
1025 		NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
1026 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
1027 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
1028 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
1029 			return (0);
1030 		}
1031 	}
1032 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
1033 	if (!error)
1034 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
1035 	if (!error) {
1036 		/*
1037 		 * Get the local modify time for the case of a write
1038 		 * delegation.
1039 		 */
1040 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
1041 	} else if (NFS_ISV4(vp)) {
1042 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1043 	}
1044 	return (error);
1045 }
1046 
1047 /*
1048  * nfs setattr call.
1049  */
1050 static int
1051 nfs_setattr(struct vop_setattr_args *ap)
1052 {
1053 	struct vnode *vp = ap->a_vp;
1054 	struct nfsnode *np = VTONFS(vp);
1055 	struct thread *td = curthread;	/* XXX */
1056 	struct vattr *vap = ap->a_vap;
1057 	int error = 0;
1058 	u_quad_t tsize;
1059 	struct timespec ts;
1060 
1061 #ifndef nolint
1062 	tsize = (u_quad_t)0;
1063 #endif
1064 
1065 	/*
1066 	 * Setting of flags and marking of atimes are not supported.
1067 	 */
1068 	if (vap->va_flags != VNOVAL)
1069 		return (EOPNOTSUPP);
1070 
1071 	/*
1072 	 * Disallow write attempts if the filesystem is mounted read-only.
1073 	 */
1074   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
1075 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
1076 	    vap->va_mtime.tv_sec != VNOVAL ||
1077 	    vap->va_birthtime.tv_sec != VNOVAL ||
1078 	    vap->va_mode != (mode_t)VNOVAL) &&
1079 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
1080 		return (EROFS);
1081 	if (vap->va_size != VNOVAL) {
1082  		switch (vp->v_type) {
1083  		case VDIR:
1084  			return (EISDIR);
1085  		case VCHR:
1086  		case VBLK:
1087  		case VSOCK:
1088  		case VFIFO:
1089 			if (vap->va_mtime.tv_sec == VNOVAL &&
1090 			    vap->va_atime.tv_sec == VNOVAL &&
1091 			    vap->va_birthtime.tv_sec == VNOVAL &&
1092 			    vap->va_mode == (mode_t)VNOVAL &&
1093 			    vap->va_uid == (uid_t)VNOVAL &&
1094 			    vap->va_gid == (gid_t)VNOVAL)
1095 				return (0);
1096  			vap->va_size = VNOVAL;
1097  			break;
1098  		default:
1099 			/*
1100 			 * Disallow write attempts if the filesystem is
1101 			 * mounted read-only.
1102 			 */
1103 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
1104 				return (EROFS);
1105 			/*
1106 			 *  We run vnode_pager_setsize() early (why?),
1107 			 * we must set np->n_size now to avoid vinvalbuf
1108 			 * V_SAVE races that might setsize a lower
1109 			 * value.
1110 			 */
1111 			NFSLOCKNODE(np);
1112 			tsize = np->n_size;
1113 			NFSUNLOCKNODE(np);
1114 			error = ncl_meta_setsize(vp, td, vap->va_size);
1115 			NFSLOCKNODE(np);
1116  			if (np->n_flag & NMODIFIED) {
1117 			    tsize = np->n_size;
1118 			    NFSUNLOCKNODE(np);
1119 			    error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
1120 			        0 : V_SAVE, td, 1);
1121 			    if (error != 0) {
1122 				    vnode_pager_setsize(vp, tsize);
1123 				    return (error);
1124 			    }
1125 			    /*
1126 			     * Call nfscl_delegmodtime() to set the modify time
1127 			     * locally, as required.
1128 			     */
1129 			    nfscl_delegmodtime(vp);
1130  			} else
1131 			    NFSUNLOCKNODE(np);
1132 			/*
1133 			 * np->n_size has already been set to vap->va_size
1134 			 * in ncl_meta_setsize(). We must set it again since
1135 			 * nfs_loadattrcache() could be called through
1136 			 * ncl_meta_setsize() and could modify np->n_size.
1137 			 */
1138 			NFSLOCKNODE(np);
1139  			np->n_vattr.na_size = np->n_size = vap->va_size;
1140 			NFSUNLOCKNODE(np);
1141   		}
1142   	} else {
1143 		NFSLOCKNODE(np);
1144 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
1145 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
1146 			NFSUNLOCKNODE(np);
1147 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
1148 			if (error == EINTR || error == EIO)
1149 				return (error);
1150 		} else
1151 			NFSUNLOCKNODE(np);
1152 	}
1153 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
1154 	if (vap->va_size != VNOVAL) {
1155 		if (error == 0) {
1156 			nanouptime(&ts);
1157 			NFSLOCKNODE(np);
1158 			np->n_localmodtime = ts;
1159 			NFSUNLOCKNODE(np);
1160 		} else {
1161 			NFSLOCKNODE(np);
1162 			np->n_size = np->n_vattr.na_size = tsize;
1163 			vnode_pager_setsize(vp, tsize);
1164 			NFSUNLOCKNODE(np);
1165 		}
1166 	}
1167 	return (error);
1168 }
1169 
1170 /*
1171  * Do an nfs setattr rpc.
1172  */
1173 static int
1174 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1175     struct thread *td)
1176 {
1177 	struct nfsnode *np = VTONFS(vp);
1178 	int error, ret, attrflag, i;
1179 	struct nfsvattr nfsva;
1180 
1181 	if (NFS_ISV34(vp)) {
1182 		NFSLOCKNODE(np);
1183 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1184 			np->n_accesscache[i].stamp = 0;
1185 		np->n_flag |= NDELEGMOD;
1186 		NFSUNLOCKNODE(np);
1187 		KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1188 	}
1189 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1190 	    NULL);
1191 	if (attrflag) {
1192 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1193 		if (ret && !error)
1194 			error = ret;
1195 	}
1196 	if (error && NFS_ISV4(vp))
1197 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1198 	return (error);
1199 }
1200 
1201 /*
1202  * nfs lookup call, one step at a time...
1203  * First look in cache
1204  * If not found, unlock the directory nfsnode and do the rpc
1205  */
1206 static int
1207 nfs_lookup(struct vop_lookup_args *ap)
1208 {
1209 	struct componentname *cnp = ap->a_cnp;
1210 	struct vnode *dvp = ap->a_dvp;
1211 	struct vnode **vpp = ap->a_vpp;
1212 	struct mount *mp = dvp->v_mount;
1213 	int flags = cnp->cn_flags;
1214 	struct vnode *newvp;
1215 	struct nfsmount *nmp;
1216 	struct nfsnode *np, *newnp;
1217 	int error = 0, attrflag, dattrflag, ltype, ncticks;
1218 	struct thread *td = curthread;
1219 	struct nfsfh *nfhp;
1220 	struct nfsvattr dnfsva, nfsva;
1221 	struct vattr vattr;
1222 	struct timespec nctime, ts;
1223 	uint32_t openmode;
1224 
1225 	*vpp = NULLVP;
1226 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1227 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1228 		return (EROFS);
1229 	if (dvp->v_type != VDIR)
1230 		return (ENOTDIR);
1231 	nmp = VFSTONFS(mp);
1232 	np = VTONFS(dvp);
1233 
1234 	/* For NFSv4, wait until any remove is done. */
1235 	NFSLOCKNODE(np);
1236 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1237 		np->n_flag |= NREMOVEWANT;
1238 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1239 	}
1240 	NFSUNLOCKNODE(np);
1241 
1242 	error = vn_dir_check_exec(dvp, cnp);
1243 	if (error != 0)
1244 		return (error);
1245 	error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1246 	if (error > 0 && error != ENOENT)
1247 		return (error);
1248 	if (error == -1) {
1249 		/*
1250 		 * Lookups of "." are special and always return the
1251 		 * current directory.  cache_lookup() already handles
1252 		 * associated locking bookkeeping, etc.
1253 		 */
1254 		if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1255 			/* XXX: Is this really correct? */
1256 			if (cnp->cn_nameiop != LOOKUP &&
1257 			    (flags & ISLASTCN))
1258 				cnp->cn_flags |= SAVENAME;
1259 			return (0);
1260 		}
1261 
1262 		/*
1263 		 * We only accept a positive hit in the cache if the
1264 		 * change time of the file matches our cached copy.
1265 		 * Otherwise, we discard the cache entry and fallback
1266 		 * to doing a lookup RPC.  We also only trust cache
1267 		 * entries for less than nm_nametimeo seconds.
1268 		 *
1269 		 * To better handle stale file handles and attributes,
1270 		 * clear the attribute cache of this node if it is a
1271 		 * leaf component, part of an open() call, and not
1272 		 * locally modified before fetching the attributes.
1273 		 * This should allow stale file handles to be detected
1274 		 * here where we can fall back to a LOOKUP RPC to
1275 		 * recover rather than having nfs_open() detect the
1276 		 * stale file handle and failing open(2) with ESTALE.
1277 		 */
1278 		newvp = *vpp;
1279 		newnp = VTONFS(newvp);
1280 		if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1281 		    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1282 		    !(newnp->n_flag & NMODIFIED)) {
1283 			NFSLOCKNODE(newnp);
1284 			newnp->n_attrstamp = 0;
1285 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1286 			NFSUNLOCKNODE(newnp);
1287 		}
1288 		if (nfscl_nodeleg(newvp, 0) == 0 ||
1289 		    ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1290 		    VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1291 		    timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1292 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1293 			if (cnp->cn_nameiop != LOOKUP &&
1294 			    (flags & ISLASTCN))
1295 				cnp->cn_flags |= SAVENAME;
1296 			return (0);
1297 		}
1298 		cache_purge(newvp);
1299 		if (dvp != newvp)
1300 			vput(newvp);
1301 		else
1302 			vrele(newvp);
1303 		*vpp = NULLVP;
1304 	} else if (error == ENOENT) {
1305 		if (VN_IS_DOOMED(dvp))
1306 			return (ENOENT);
1307 		/*
1308 		 * We only accept a negative hit in the cache if the
1309 		 * modification time of the parent directory matches
1310 		 * the cached copy in the name cache entry.
1311 		 * Otherwise, we discard all of the negative cache
1312 		 * entries for this directory.  We also only trust
1313 		 * negative cache entries for up to nm_negnametimeo
1314 		 * seconds.
1315 		 */
1316 		if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1317 		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1318 		    timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1319 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1320 			return (ENOENT);
1321 		}
1322 		cache_purge_negative(dvp);
1323 	}
1324 
1325 	openmode = 0;
1326 #if 0
1327 	/*
1328 	 * The use of LookupOpen breaks some builds.  It is disabled
1329 	 * until that is fixed.
1330 	 */
1331 	/*
1332 	 * If this an NFSv4.1/4.2 mount using the "oneopenown" mount
1333 	 * option, it is possible to do the Open operation in the same
1334 	 * compound as Lookup, so long as delegations are not being
1335 	 * issued.  This saves doing a separate RPC for Open.
1336 	 * For pnfs, do not do this, since the Open+LayoutGet will
1337 	 * be needed as a separate RPC.
1338 	 */
1339 	NFSLOCKMNT(nmp);
1340 	if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && !NFSHASPNFS(nmp) &&
1341 	    (nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 &&
1342 	    (!NFSMNT_RDONLY(mp) || (flags & OPENWRITE) == 0) &&
1343 	    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN)) {
1344 		if ((flags & OPENREAD) != 0)
1345 			openmode |= NFSV4OPEN_ACCESSREAD;
1346 		if ((flags & OPENWRITE) != 0)
1347 			openmode |= NFSV4OPEN_ACCESSWRITE;
1348 	}
1349 	NFSUNLOCKMNT(nmp);
1350 #endif
1351 
1352 	newvp = NULLVP;
1353 	NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1354 	nanouptime(&ts);
1355 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1356 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1357 	    NULL, openmode);
1358 	if (dattrflag)
1359 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1360 	if (error) {
1361 		if (newvp != NULLVP) {
1362 			vput(newvp);
1363 			*vpp = NULLVP;
1364 		}
1365 
1366 		if (error != ENOENT) {
1367 			if (NFS_ISV4(dvp))
1368 				error = nfscl_maperr(td, error, (uid_t)0,
1369 				    (gid_t)0);
1370 			return (error);
1371 		}
1372 
1373 		/* The requested file was not found. */
1374 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1375 		    (flags & ISLASTCN)) {
1376 			/*
1377 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1378 			 * VWRITE) here instead of just checking
1379 			 * MNT_RDONLY.
1380 			 */
1381 			if (mp->mnt_flag & MNT_RDONLY)
1382 				return (EROFS);
1383 			cnp->cn_flags |= SAVENAME;
1384 			return (EJUSTRETURN);
1385 		}
1386 
1387 		if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1388 			/*
1389 			 * Cache the modification time of the parent
1390 			 * directory from the post-op attributes in
1391 			 * the name cache entry.  The negative cache
1392 			 * entry will be ignored once the directory
1393 			 * has changed.  Don't bother adding the entry
1394 			 * if the directory has already changed.
1395 			 */
1396 			NFSLOCKNODE(np);
1397 			if (timespeccmp(&np->n_vattr.na_mtime,
1398 			    &dnfsva.na_mtime, ==)) {
1399 				NFSUNLOCKNODE(np);
1400 				cache_enter_time(dvp, NULL, cnp,
1401 				    &dnfsva.na_mtime, NULL);
1402 			} else
1403 				NFSUNLOCKNODE(np);
1404 		}
1405 		return (ENOENT);
1406 	}
1407 
1408 	/*
1409 	 * Handle RENAME case...
1410 	 */
1411 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1412 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1413 			free(nfhp, M_NFSFH);
1414 			return (EISDIR);
1415 		}
1416 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1417 		    LK_EXCLUSIVE);
1418 		if (error)
1419 			return (error);
1420 		newvp = NFSTOV(np);
1421 		/*
1422 		 * If n_localmodtime >= time before RPC, then
1423 		 * a file modification operation, such as
1424 		 * VOP_SETATTR() of size, has occurred while
1425 		 * the Lookup RPC and acquisition of the vnode
1426 		 * happened.  As such, the attributes might
1427 		 * be stale, with possibly an incorrect size.
1428 		 */
1429 		NFSLOCKNODE(np);
1430 		if (timespecisset(&np->n_localmodtime) &&
1431 		    timespeccmp(&np->n_localmodtime, &ts, >=)) {
1432 			NFSCL_DEBUG(4, "nfs_lookup: rename localmod "
1433 			    "stale attributes\n");
1434 			attrflag = 0;
1435 		}
1436 		NFSUNLOCKNODE(np);
1437 		if (attrflag)
1438 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1439 			    0, 1);
1440 		*vpp = newvp;
1441 		cnp->cn_flags |= SAVENAME;
1442 		return (0);
1443 	}
1444 
1445 	if (flags & ISDOTDOT) {
1446 		ltype = NFSVOPISLOCKED(dvp);
1447 		error = vfs_busy(mp, MBF_NOWAIT);
1448 		if (error != 0) {
1449 			vfs_ref(mp);
1450 			NFSVOPUNLOCK(dvp);
1451 			error = vfs_busy(mp, 0);
1452 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1453 			vfs_rel(mp);
1454 			if (error == 0 && VN_IS_DOOMED(dvp)) {
1455 				vfs_unbusy(mp);
1456 				error = ENOENT;
1457 			}
1458 			if (error != 0)
1459 				return (error);
1460 		}
1461 		NFSVOPUNLOCK(dvp);
1462 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1463 		    cnp->cn_lkflags);
1464 		if (error == 0)
1465 			newvp = NFSTOV(np);
1466 		vfs_unbusy(mp);
1467 		if (newvp != dvp)
1468 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1469 		if (VN_IS_DOOMED(dvp)) {
1470 			if (error == 0) {
1471 				if (newvp == dvp)
1472 					vrele(newvp);
1473 				else
1474 					vput(newvp);
1475 			}
1476 			error = ENOENT;
1477 		}
1478 		if (error != 0)
1479 			return (error);
1480 		if (attrflag)
1481 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1482 			    0, 1);
1483 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1484 		free(nfhp, M_NFSFH);
1485 		VREF(dvp);
1486 		newvp = dvp;
1487 		if (attrflag)
1488 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1489 			    0, 1);
1490 	} else {
1491 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1492 		    cnp->cn_lkflags);
1493 		if (error)
1494 			return (error);
1495 		newvp = NFSTOV(np);
1496 		/*
1497 		 * If n_localmodtime >= time before RPC, then
1498 		 * a file modification operation, such as
1499 		 * VOP_SETATTR() of size, has occurred while
1500 		 * the Lookup RPC and acquisition of the vnode
1501 		 * happened.  As such, the attributes might
1502 		 * be stale, with possibly an incorrect size.
1503 		 */
1504 		NFSLOCKNODE(np);
1505 		if (timespecisset(&np->n_localmodtime) &&
1506 		    timespeccmp(&np->n_localmodtime, &ts, >=)) {
1507 			NFSCL_DEBUG(4, "nfs_lookup: localmod "
1508 			    "stale attributes\n");
1509 			attrflag = 0;
1510 		}
1511 		NFSUNLOCKNODE(np);
1512 		if (attrflag)
1513 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1514 			    0, 1);
1515 		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1516 		    !(np->n_flag & NMODIFIED)) {
1517 			/*
1518 			 * Flush the attribute cache when opening a
1519 			 * leaf node to ensure that fresh attributes
1520 			 * are fetched in nfs_open() since we did not
1521 			 * fetch attributes from the LOOKUP reply.
1522 			 */
1523 			NFSLOCKNODE(np);
1524 			np->n_attrstamp = 0;
1525 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1526 			NFSUNLOCKNODE(np);
1527 		}
1528 	}
1529 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1530 		cnp->cn_flags |= SAVENAME;
1531 	if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp &&
1532 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1533 	    attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1534 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1535 		    newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1536 	*vpp = newvp;
1537 	return (0);
1538 }
1539 
1540 /*
1541  * nfs read call.
1542  * Just call ncl_bioread() to do the work.
1543  */
1544 static int
1545 nfs_read(struct vop_read_args *ap)
1546 {
1547 	struct vnode *vp = ap->a_vp;
1548 
1549 	switch (vp->v_type) {
1550 	case VREG:
1551 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1552 	case VDIR:
1553 		return (EISDIR);
1554 	default:
1555 		return (EOPNOTSUPP);
1556 	}
1557 }
1558 
1559 /*
1560  * nfs readlink call
1561  */
1562 static int
1563 nfs_readlink(struct vop_readlink_args *ap)
1564 {
1565 	struct vnode *vp = ap->a_vp;
1566 
1567 	if (vp->v_type != VLNK)
1568 		return (EINVAL);
1569 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1570 }
1571 
1572 /*
1573  * Do a readlink rpc.
1574  * Called by ncl_doio() from below the buffer cache.
1575  */
1576 int
1577 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1578 {
1579 	int error, ret, attrflag;
1580 	struct nfsvattr nfsva;
1581 
1582 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1583 	    &attrflag, NULL);
1584 	if (attrflag) {
1585 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1586 		if (ret && !error)
1587 			error = ret;
1588 	}
1589 	if (error && NFS_ISV4(vp))
1590 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1591 	return (error);
1592 }
1593 
1594 /*
1595  * nfs read rpc call
1596  * Ditto above
1597  */
1598 int
1599 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1600 {
1601 	int error, ret, attrflag;
1602 	struct nfsvattr nfsva;
1603 	struct nfsmount *nmp;
1604 
1605 	nmp = VFSTONFS(vp->v_mount);
1606 	error = EIO;
1607 	attrflag = 0;
1608 	if (NFSHASPNFS(nmp))
1609 		error = nfscl_doiods(vp, uiop, NULL, NULL,
1610 		    NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1611 	NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1612 	if (error != 0)
1613 		error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1614 		    &attrflag, NULL);
1615 	if (attrflag) {
1616 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1617 		if (ret && !error)
1618 			error = ret;
1619 	}
1620 	if (error && NFS_ISV4(vp))
1621 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1622 	return (error);
1623 }
1624 
1625 /*
1626  * nfs write call
1627  */
1628 int
1629 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1630     int *iomode, int *must_commit, int called_from_strategy)
1631 {
1632 	struct nfsvattr nfsva;
1633 	int error, attrflag, ret;
1634 	struct nfsmount *nmp;
1635 
1636 	nmp = VFSTONFS(vp->v_mount);
1637 	error = EIO;
1638 	attrflag = 0;
1639 	if (NFSHASPNFS(nmp))
1640 		error = nfscl_doiods(vp, uiop, iomode, must_commit,
1641 		    NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1642 	NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1643 	if (error != 0)
1644 		error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1645 		    uiop->uio_td, &nfsva, &attrflag, NULL,
1646 		    called_from_strategy);
1647 	if (attrflag) {
1648 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1649 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1650 			    1);
1651 		else
1652 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1653 			    1);
1654 		if (ret && !error)
1655 			error = ret;
1656 	}
1657 	if (DOINGASYNC(vp))
1658 		*iomode = NFSWRITE_FILESYNC;
1659 	if (error && NFS_ISV4(vp))
1660 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1661 	return (error);
1662 }
1663 
1664 /*
1665  * nfs mknod rpc
1666  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1667  * mode set to specify the file type and the size field for rdev.
1668  */
1669 static int
1670 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1671     struct vattr *vap)
1672 {
1673 	struct nfsvattr nfsva, dnfsva;
1674 	struct vnode *newvp = NULL;
1675 	struct nfsnode *np = NULL, *dnp;
1676 	struct nfsfh *nfhp;
1677 	struct vattr vattr;
1678 	int error = 0, attrflag, dattrflag;
1679 	u_int32_t rdev;
1680 
1681 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1682 		rdev = vap->va_rdev;
1683 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1684 		rdev = 0xffffffff;
1685 	else
1686 		return (EOPNOTSUPP);
1687 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1688 		return (error);
1689 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1690 	    rdev, vap->va_type, cnp->cn_cred, curthread, &dnfsva,
1691 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1692 	if (!error) {
1693 		if (!nfhp)
1694 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1695 			    cnp->cn_namelen, cnp->cn_cred, curthread,
1696 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1697 			    NULL, 0);
1698 		if (nfhp)
1699 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1700 			    curthread, &np, NULL, LK_EXCLUSIVE);
1701 	}
1702 	if (dattrflag)
1703 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1704 	if (!error) {
1705 		newvp = NFSTOV(np);
1706 		if (attrflag != 0) {
1707 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1708 			    0, 1);
1709 			if (error != 0)
1710 				vput(newvp);
1711 		}
1712 	}
1713 	if (!error) {
1714 		*vpp = newvp;
1715 	} else if (NFS_ISV4(dvp)) {
1716 		error = nfscl_maperr(curthread, error, vap->va_uid,
1717 		    vap->va_gid);
1718 	}
1719 	dnp = VTONFS(dvp);
1720 	NFSLOCKNODE(dnp);
1721 	dnp->n_flag |= NMODIFIED;
1722 	if (!dattrflag) {
1723 		dnp->n_attrstamp = 0;
1724 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1725 	}
1726 	NFSUNLOCKNODE(dnp);
1727 	return (error);
1728 }
1729 
1730 /*
1731  * nfs mknod vop
1732  * just call nfs_mknodrpc() to do the work.
1733  */
1734 /* ARGSUSED */
1735 static int
1736 nfs_mknod(struct vop_mknod_args *ap)
1737 {
1738 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1739 }
1740 
1741 static struct mtx nfs_cverf_mtx;
1742 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1743     MTX_DEF);
1744 
1745 static nfsquad_t
1746 nfs_get_cverf(void)
1747 {
1748 	static nfsquad_t cverf;
1749 	nfsquad_t ret;
1750 	static int cverf_initialized = 0;
1751 
1752 	mtx_lock(&nfs_cverf_mtx);
1753 	if (cverf_initialized == 0) {
1754 		cverf.lval[0] = arc4random();
1755 		cverf.lval[1] = arc4random();
1756 		cverf_initialized = 1;
1757 	} else
1758 		cverf.qval++;
1759 	ret = cverf;
1760 	mtx_unlock(&nfs_cverf_mtx);
1761 
1762 	return (ret);
1763 }
1764 
1765 /*
1766  * nfs file create call
1767  */
1768 static int
1769 nfs_create(struct vop_create_args *ap)
1770 {
1771 	struct vnode *dvp = ap->a_dvp;
1772 	struct vattr *vap = ap->a_vap;
1773 	struct componentname *cnp = ap->a_cnp;
1774 	struct nfsnode *np = NULL, *dnp;
1775 	struct vnode *newvp = NULL;
1776 	struct nfsmount *nmp;
1777 	struct nfsvattr dnfsva, nfsva;
1778 	struct nfsfh *nfhp;
1779 	nfsquad_t cverf;
1780 	int error = 0, attrflag, dattrflag, fmode = 0;
1781 	struct vattr vattr;
1782 
1783 	/*
1784 	 * Oops, not for me..
1785 	 */
1786 	if (vap->va_type == VSOCK)
1787 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1788 
1789 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1790 		return (error);
1791 	if (vap->va_vaflags & VA_EXCLUSIVE)
1792 		fmode |= O_EXCL;
1793 	dnp = VTONFS(dvp);
1794 	nmp = VFSTONFS(dvp->v_mount);
1795 again:
1796 	/* For NFSv4, wait until any remove is done. */
1797 	NFSLOCKNODE(dnp);
1798 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1799 		dnp->n_flag |= NREMOVEWANT;
1800 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1801 	}
1802 	NFSUNLOCKNODE(dnp);
1803 
1804 	cverf = nfs_get_cverf();
1805 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1806 	    vap, cverf, fmode, cnp->cn_cred, curthread, &dnfsva, &nfsva,
1807 	    &nfhp, &attrflag, &dattrflag, NULL);
1808 	if (!error) {
1809 		if (nfhp == NULL)
1810 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1811 			    cnp->cn_namelen, cnp->cn_cred, curthread,
1812 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1813 			    NULL, 0);
1814 		if (nfhp != NULL)
1815 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1816 			    curthread, &np, NULL, LK_EXCLUSIVE);
1817 	}
1818 	if (dattrflag)
1819 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1820 	if (!error) {
1821 		newvp = NFSTOV(np);
1822 		if (attrflag == 0)
1823 			error = nfsrpc_getattr(newvp, cnp->cn_cred,
1824 			    curthread, &nfsva, NULL);
1825 		if (error == 0)
1826 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1827 			    0, 1);
1828 	}
1829 	if (error) {
1830 		if (newvp != NULL) {
1831 			vput(newvp);
1832 			newvp = NULL;
1833 		}
1834 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1835 		    error == NFSERR_NOTSUPP) {
1836 			fmode &= ~O_EXCL;
1837 			goto again;
1838 		}
1839 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1840 		if (nfscl_checksattr(vap, &nfsva)) {
1841 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1842 			    curthread, &nfsva, &attrflag, NULL);
1843 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1844 			    vap->va_gid != (gid_t)VNOVAL)) {
1845 				/* try again without setting uid/gid */
1846 				vap->va_uid = (uid_t)VNOVAL;
1847 				vap->va_gid = (uid_t)VNOVAL;
1848 				error = nfsrpc_setattr(newvp, vap, NULL,
1849 				    cnp->cn_cred, curthread, &nfsva,
1850 				    &attrflag, NULL);
1851 			}
1852 			if (attrflag)
1853 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1854 				    NULL, 0, 1);
1855 			if (error != 0)
1856 				vput(newvp);
1857 		}
1858 	}
1859 	if (!error) {
1860 		if ((cnp->cn_flags & MAKEENTRY) && attrflag) {
1861 			if (dvp != newvp)
1862 				cache_enter_time(dvp, newvp, cnp,
1863 				    &nfsva.na_ctime, NULL);
1864 			else
1865 				printf("nfs_create: bogus NFS server returned "
1866 				    "the directory as the new file object\n");
1867 		}
1868 		*ap->a_vpp = newvp;
1869 	} else if (NFS_ISV4(dvp)) {
1870 		error = nfscl_maperr(curthread, error, vap->va_uid,
1871 		    vap->va_gid);
1872 	}
1873 	NFSLOCKNODE(dnp);
1874 	dnp->n_flag |= NMODIFIED;
1875 	if (!dattrflag) {
1876 		dnp->n_attrstamp = 0;
1877 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1878 	}
1879 	NFSUNLOCKNODE(dnp);
1880 	return (error);
1881 }
1882 
1883 /*
1884  * nfs file remove call
1885  * To try and make nfs semantics closer to ufs semantics, a file that has
1886  * other processes using the vnode is renamed instead of removed and then
1887  * removed later on the last close.
1888  * - If v_usecount > 1
1889  *	  If a rename is not already in the works
1890  *	     call nfs_sillyrename() to set it up
1891  *     else
1892  *	  do the remove rpc
1893  */
1894 static int
1895 nfs_remove(struct vop_remove_args *ap)
1896 {
1897 	struct vnode *vp = ap->a_vp;
1898 	struct vnode *dvp = ap->a_dvp;
1899 	struct componentname *cnp = ap->a_cnp;
1900 	struct nfsnode *np = VTONFS(vp);
1901 	int error = 0;
1902 	struct vattr vattr;
1903 
1904 	KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1905 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1906 	if (vp->v_type == VDIR)
1907 		error = EPERM;
1908 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1909 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1910 	    vattr.va_nlink > 1)) {
1911 		/*
1912 		 * Purge the name cache so that the chance of a lookup for
1913 		 * the name succeeding while the remove is in progress is
1914 		 * minimized. Without node locking it can still happen, such
1915 		 * that an I/O op returns ESTALE, but since you get this if
1916 		 * another host removes the file..
1917 		 */
1918 		cache_purge(vp);
1919 		/*
1920 		 * throw away biocache buffers, mainly to avoid
1921 		 * unnecessary delayed writes later.
1922 		 */
1923 		error = ncl_vinvalbuf(vp, 0, curthread, 1);
1924 		if (error != EINTR && error != EIO)
1925 			/* Do the rpc */
1926 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1927 			    cnp->cn_namelen, cnp->cn_cred, curthread);
1928 		/*
1929 		 * Kludge City: If the first reply to the remove rpc is lost..
1930 		 *   the reply to the retransmitted request will be ENOENT
1931 		 *   since the file was in fact removed
1932 		 *   Therefore, we cheat and return success.
1933 		 */
1934 		if (error == ENOENT)
1935 			error = 0;
1936 	} else if (!np->n_sillyrename)
1937 		error = nfs_sillyrename(dvp, vp, cnp);
1938 	NFSLOCKNODE(np);
1939 	np->n_attrstamp = 0;
1940 	NFSUNLOCKNODE(np);
1941 	KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1942 	return (error);
1943 }
1944 
1945 /*
1946  * nfs file remove rpc called from nfs_inactive
1947  */
1948 int
1949 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1950 {
1951 	/*
1952 	 * Make sure that the directory vnode is still valid.
1953 	 * XXX we should lock sp->s_dvp here.
1954 	 */
1955 	if (sp->s_dvp->v_type == VBAD)
1956 		return (0);
1957 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1958 	    sp->s_cred, NULL));
1959 }
1960 
1961 /*
1962  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1963  */
1964 static int
1965 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1966     int namelen, struct ucred *cred, struct thread *td)
1967 {
1968 	struct nfsvattr dnfsva;
1969 	struct nfsnode *dnp = VTONFS(dvp);
1970 	int error = 0, dattrflag;
1971 
1972 	NFSLOCKNODE(dnp);
1973 	dnp->n_flag |= NREMOVEINPROG;
1974 	NFSUNLOCKNODE(dnp);
1975 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1976 	    &dattrflag, NULL);
1977 	NFSLOCKNODE(dnp);
1978 	if ((dnp->n_flag & NREMOVEWANT)) {
1979 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1980 		NFSUNLOCKNODE(dnp);
1981 		wakeup((caddr_t)dnp);
1982 	} else {
1983 		dnp->n_flag &= ~NREMOVEINPROG;
1984 		NFSUNLOCKNODE(dnp);
1985 	}
1986 	if (dattrflag)
1987 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1988 	NFSLOCKNODE(dnp);
1989 	dnp->n_flag |= NMODIFIED;
1990 	if (!dattrflag) {
1991 		dnp->n_attrstamp = 0;
1992 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1993 	}
1994 	NFSUNLOCKNODE(dnp);
1995 	if (error && NFS_ISV4(dvp))
1996 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1997 	return (error);
1998 }
1999 
2000 /*
2001  * nfs file rename call
2002  */
2003 static int
2004 nfs_rename(struct vop_rename_args *ap)
2005 {
2006 	struct vnode *fvp = ap->a_fvp;
2007 	struct vnode *tvp = ap->a_tvp;
2008 	struct vnode *fdvp = ap->a_fdvp;
2009 	struct vnode *tdvp = ap->a_tdvp;
2010 	struct componentname *tcnp = ap->a_tcnp;
2011 	struct componentname *fcnp = ap->a_fcnp;
2012 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
2013 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
2014 	struct nfsv4node *newv4 = NULL;
2015 	int error;
2016 
2017 	KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
2018 	    (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
2019 	/* Check for cross-device rename */
2020 	if ((fvp->v_mount != tdvp->v_mount) ||
2021 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
2022 		error = EXDEV;
2023 		goto out;
2024 	}
2025 
2026 	if (fvp == tvp) {
2027 		printf("nfs_rename: fvp == tvp (can't happen)\n");
2028 		error = 0;
2029 		goto out;
2030 	}
2031 	if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
2032 		goto out;
2033 
2034 	/*
2035 	 * We have to flush B_DELWRI data prior to renaming
2036 	 * the file.  If we don't, the delayed-write buffers
2037 	 * can be flushed out later after the file has gone stale
2038 	 * under NFSV3.  NFSV2 does not have this problem because
2039 	 * ( as far as I can tell ) it flushes dirty buffers more
2040 	 * often.
2041 	 *
2042 	 * Skip the rename operation if the fsync fails, this can happen
2043 	 * due to the server's volume being full, when we pushed out data
2044 	 * that was written back to our cache earlier. Not checking for
2045 	 * this condition can result in potential (silent) data loss.
2046 	 */
2047 	error = VOP_FSYNC(fvp, MNT_WAIT, curthread);
2048 	NFSVOPUNLOCK(fvp);
2049 	if (!error && tvp)
2050 		error = VOP_FSYNC(tvp, MNT_WAIT, curthread);
2051 	if (error)
2052 		goto out;
2053 
2054 	/*
2055 	 * If the tvp exists and is in use, sillyrename it before doing the
2056 	 * rename of the new file over it.
2057 	 * XXX Can't sillyrename a directory.
2058 	 */
2059 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
2060 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
2061 		vput(tvp);
2062 		tvp = NULL;
2063 	}
2064 
2065 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
2066 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
2067 	    curthread);
2068 
2069 	if (error == 0 && NFS_ISV4(tdvp)) {
2070 		/*
2071 		 * For NFSv4, check to see if it is the same name and
2072 		 * replace the name, if it is different.
2073 		 */
2074 		newv4 = malloc(
2075 		    sizeof (struct nfsv4node) +
2076 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
2077 		    M_NFSV4NODE, M_WAITOK);
2078 		NFSLOCKNODE(tdnp);
2079 		NFSLOCKNODE(fnp);
2080 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
2081 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
2082 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
2083 		      tcnp->cn_namelen) ||
2084 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
2085 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
2086 			tdnp->n_fhp->nfh_len))) {
2087 #ifdef notdef
2088 { char nnn[100]; int nnnl;
2089 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
2090 bcopy(tcnp->cn_nameptr, nnn, nnnl);
2091 nnn[nnnl] = '\0';
2092 printf("ren replace=%s\n",nnn);
2093 }
2094 #endif
2095 			free(fnp->n_v4, M_NFSV4NODE);
2096 			fnp->n_v4 = newv4;
2097 			newv4 = NULL;
2098 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
2099 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
2100 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
2101 			    tdnp->n_fhp->nfh_len);
2102 			NFSBCOPY(tcnp->cn_nameptr,
2103 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
2104 		}
2105 		NFSUNLOCKNODE(tdnp);
2106 		NFSUNLOCKNODE(fnp);
2107 		if (newv4 != NULL)
2108 			free(newv4, M_NFSV4NODE);
2109 	}
2110 
2111 	if (fvp->v_type == VDIR) {
2112 		if (tvp != NULL && tvp->v_type == VDIR)
2113 			cache_purge(tdvp);
2114 		cache_purge(fdvp);
2115 	}
2116 
2117 out:
2118 	if (tdvp == tvp)
2119 		vrele(tdvp);
2120 	else
2121 		vput(tdvp);
2122 	if (tvp)
2123 		vput(tvp);
2124 	vrele(fdvp);
2125 	vrele(fvp);
2126 	/*
2127 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2128 	 */
2129 	if (error == ENOENT)
2130 		error = 0;
2131 	return (error);
2132 }
2133 
2134 /*
2135  * nfs file rename rpc called from nfs_remove() above
2136  */
2137 static int
2138 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
2139     struct sillyrename *sp)
2140 {
2141 
2142 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
2143 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
2144 	    curthread));
2145 }
2146 
2147 /*
2148  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2149  */
2150 static int
2151 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
2152     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
2153     int tnamelen, struct ucred *cred, struct thread *td)
2154 {
2155 	struct nfsvattr fnfsva, tnfsva;
2156 	struct nfsnode *fdnp = VTONFS(fdvp);
2157 	struct nfsnode *tdnp = VTONFS(tdvp);
2158 	int error = 0, fattrflag, tattrflag;
2159 
2160 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
2161 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
2162 	    &tattrflag, NULL, NULL);
2163 	NFSLOCKNODE(fdnp);
2164 	fdnp->n_flag |= NMODIFIED;
2165 	if (fattrflag != 0) {
2166 		NFSUNLOCKNODE(fdnp);
2167 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
2168 	} else {
2169 		fdnp->n_attrstamp = 0;
2170 		NFSUNLOCKNODE(fdnp);
2171 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
2172 	}
2173 	NFSLOCKNODE(tdnp);
2174 	tdnp->n_flag |= NMODIFIED;
2175 	if (tattrflag != 0) {
2176 		NFSUNLOCKNODE(tdnp);
2177 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
2178 	} else {
2179 		tdnp->n_attrstamp = 0;
2180 		NFSUNLOCKNODE(tdnp);
2181 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2182 	}
2183 	if (error && NFS_ISV4(fdvp))
2184 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2185 	return (error);
2186 }
2187 
2188 /*
2189  * nfs hard link create call
2190  */
2191 static int
2192 nfs_link(struct vop_link_args *ap)
2193 {
2194 	struct vnode *vp = ap->a_vp;
2195 	struct vnode *tdvp = ap->a_tdvp;
2196 	struct componentname *cnp = ap->a_cnp;
2197 	struct nfsnode *np, *tdnp;
2198 	struct nfsvattr nfsva, dnfsva;
2199 	int error = 0, attrflag, dattrflag;
2200 
2201 	/*
2202 	 * Push all writes to the server, so that the attribute cache
2203 	 * doesn't get "out of sync" with the server.
2204 	 * XXX There should be a better way!
2205 	 */
2206 	VOP_FSYNC(vp, MNT_WAIT, curthread);
2207 
2208 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2209 	    cnp->cn_cred, curthread, &dnfsva, &nfsva, &attrflag,
2210 	    &dattrflag, NULL);
2211 	tdnp = VTONFS(tdvp);
2212 	NFSLOCKNODE(tdnp);
2213 	tdnp->n_flag |= NMODIFIED;
2214 	if (dattrflag != 0) {
2215 		NFSUNLOCKNODE(tdnp);
2216 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
2217 	} else {
2218 		tdnp->n_attrstamp = 0;
2219 		NFSUNLOCKNODE(tdnp);
2220 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2221 	}
2222 	if (attrflag)
2223 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2224 	else {
2225 		np = VTONFS(vp);
2226 		NFSLOCKNODE(np);
2227 		np->n_attrstamp = 0;
2228 		NFSUNLOCKNODE(np);
2229 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2230 	}
2231 	/*
2232 	 * If negative lookup caching is enabled, I might as well
2233 	 * add an entry for this node. Not necessary for correctness,
2234 	 * but if negative caching is enabled, then the system
2235 	 * must care about lookup caching hit rate, so...
2236 	 */
2237 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2238 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2239 		if (tdvp != vp)
2240 			cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2241 		else
2242 			printf("nfs_link: bogus NFS server returned "
2243 			    "the directory as the new link\n");
2244 	}
2245 	if (error && NFS_ISV4(vp))
2246 		error = nfscl_maperr(curthread, error, (uid_t)0,
2247 		    (gid_t)0);
2248 	return (error);
2249 }
2250 
2251 /*
2252  * nfs symbolic link create call
2253  */
2254 static int
2255 nfs_symlink(struct vop_symlink_args *ap)
2256 {
2257 	struct vnode *dvp = ap->a_dvp;
2258 	struct vattr *vap = ap->a_vap;
2259 	struct componentname *cnp = ap->a_cnp;
2260 	struct nfsvattr nfsva, dnfsva;
2261 	struct nfsfh *nfhp;
2262 	struct nfsnode *np = NULL, *dnp;
2263 	struct vnode *newvp = NULL;
2264 	int error = 0, attrflag, dattrflag, ret;
2265 
2266 	vap->va_type = VLNK;
2267 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2268 	    ap->a_target, vap, cnp->cn_cred, curthread, &dnfsva,
2269 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2270 	if (nfhp) {
2271 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread,
2272 		    &np, NULL, LK_EXCLUSIVE);
2273 		if (!ret)
2274 			newvp = NFSTOV(np);
2275 		else if (!error)
2276 			error = ret;
2277 	}
2278 	if (newvp != NULL) {
2279 		if (attrflag)
2280 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2281 			    0, 1);
2282 	} else if (!error) {
2283 		/*
2284 		 * If we do not have an error and we could not extract the
2285 		 * newvp from the response due to the request being NFSv2, we
2286 		 * have to do a lookup in order to obtain a newvp to return.
2287 		 */
2288 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2289 		    cnp->cn_cred, curthread, &np);
2290 		if (!error)
2291 			newvp = NFSTOV(np);
2292 	}
2293 	if (error) {
2294 		if (newvp)
2295 			vput(newvp);
2296 		if (NFS_ISV4(dvp))
2297 			error = nfscl_maperr(curthread, error,
2298 			    vap->va_uid, vap->va_gid);
2299 	} else {
2300 		*ap->a_vpp = newvp;
2301 	}
2302 
2303 	dnp = VTONFS(dvp);
2304 	NFSLOCKNODE(dnp);
2305 	dnp->n_flag |= NMODIFIED;
2306 	if (dattrflag != 0) {
2307 		NFSUNLOCKNODE(dnp);
2308 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2309 	} else {
2310 		dnp->n_attrstamp = 0;
2311 		NFSUNLOCKNODE(dnp);
2312 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2313 	}
2314 	/*
2315 	 * If negative lookup caching is enabled, I might as well
2316 	 * add an entry for this node. Not necessary for correctness,
2317 	 * but if negative caching is enabled, then the system
2318 	 * must care about lookup caching hit rate, so...
2319 	 */
2320 	if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2321 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2322 		if (dvp != newvp)
2323 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2324 			    NULL);
2325 		else
2326 			printf("nfs_symlink: bogus NFS server returned "
2327 			    "the directory as the new file object\n");
2328 	}
2329 	return (error);
2330 }
2331 
2332 /*
2333  * nfs make dir call
2334  */
2335 static int
2336 nfs_mkdir(struct vop_mkdir_args *ap)
2337 {
2338 	struct vnode *dvp = ap->a_dvp;
2339 	struct vattr *vap = ap->a_vap;
2340 	struct componentname *cnp = ap->a_cnp;
2341 	struct nfsnode *np = NULL, *dnp;
2342 	struct vnode *newvp = NULL;
2343 	struct vattr vattr;
2344 	struct nfsfh *nfhp;
2345 	struct nfsvattr nfsva, dnfsva;
2346 	int error = 0, attrflag, dattrflag, ret;
2347 
2348 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2349 		return (error);
2350 	vap->va_type = VDIR;
2351 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2352 	    vap, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp,
2353 	    &attrflag, &dattrflag, NULL);
2354 	dnp = VTONFS(dvp);
2355 	NFSLOCKNODE(dnp);
2356 	dnp->n_flag |= NMODIFIED;
2357 	if (dattrflag != 0) {
2358 		NFSUNLOCKNODE(dnp);
2359 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2360 	} else {
2361 		dnp->n_attrstamp = 0;
2362 		NFSUNLOCKNODE(dnp);
2363 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2364 	}
2365 	if (nfhp) {
2366 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread,
2367 		    &np, NULL, LK_EXCLUSIVE);
2368 		if (!ret) {
2369 			newvp = NFSTOV(np);
2370 			if (attrflag)
2371 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2372 				NULL, 0, 1);
2373 		} else if (!error)
2374 			error = ret;
2375 	}
2376 	if (!error && newvp == NULL) {
2377 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2378 		    cnp->cn_cred, curthread, &np);
2379 		if (!error) {
2380 			newvp = NFSTOV(np);
2381 			if (newvp->v_type != VDIR)
2382 				error = EEXIST;
2383 		}
2384 	}
2385 	if (error) {
2386 		if (newvp)
2387 			vput(newvp);
2388 		if (NFS_ISV4(dvp))
2389 			error = nfscl_maperr(curthread, error,
2390 			    vap->va_uid, vap->va_gid);
2391 	} else {
2392 		/*
2393 		 * If negative lookup caching is enabled, I might as well
2394 		 * add an entry for this node. Not necessary for correctness,
2395 		 * but if negative caching is enabled, then the system
2396 		 * must care about lookup caching hit rate, so...
2397 		 */
2398 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2399 		    (cnp->cn_flags & MAKEENTRY) &&
2400 		    attrflag != 0 && dattrflag != 0) {
2401 			if (dvp != newvp)
2402 				cache_enter_time(dvp, newvp, cnp,
2403 				    &nfsva.na_ctime, &dnfsva.na_ctime);
2404 			else
2405 				printf("nfs_mkdir: bogus NFS server returned "
2406 				    "the directory that the directory was "
2407 				    "created in as the new file object\n");
2408 		}
2409 		*ap->a_vpp = newvp;
2410 	}
2411 	return (error);
2412 }
2413 
2414 /*
2415  * nfs remove directory call
2416  */
2417 static int
2418 nfs_rmdir(struct vop_rmdir_args *ap)
2419 {
2420 	struct vnode *vp = ap->a_vp;
2421 	struct vnode *dvp = ap->a_dvp;
2422 	struct componentname *cnp = ap->a_cnp;
2423 	struct nfsnode *dnp;
2424 	struct nfsvattr dnfsva;
2425 	int error, dattrflag;
2426 
2427 	if (dvp == vp)
2428 		return (EINVAL);
2429 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2430 	    cnp->cn_cred, curthread, &dnfsva, &dattrflag, NULL);
2431 	dnp = VTONFS(dvp);
2432 	NFSLOCKNODE(dnp);
2433 	dnp->n_flag |= NMODIFIED;
2434 	if (dattrflag != 0) {
2435 		NFSUNLOCKNODE(dnp);
2436 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2437 	} else {
2438 		dnp->n_attrstamp = 0;
2439 		NFSUNLOCKNODE(dnp);
2440 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2441 	}
2442 
2443 	cache_purge(dvp);
2444 	cache_purge(vp);
2445 	if (error && NFS_ISV4(dvp))
2446 		error = nfscl_maperr(curthread, error, (uid_t)0,
2447 		    (gid_t)0);
2448 	/*
2449 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2450 	 */
2451 	if (error == ENOENT)
2452 		error = 0;
2453 	return (error);
2454 }
2455 
2456 /*
2457  * nfs readdir call
2458  */
2459 static int
2460 nfs_readdir(struct vop_readdir_args *ap)
2461 {
2462 	struct vnode *vp = ap->a_vp;
2463 	struct nfsnode *np = VTONFS(vp);
2464 	struct uio *uio = ap->a_uio;
2465 	ssize_t tresid, left;
2466 	int error = 0;
2467 	struct vattr vattr;
2468 
2469 	if (ap->a_eofflag != NULL)
2470 		*ap->a_eofflag = 0;
2471 	if (vp->v_type != VDIR)
2472 		return(EPERM);
2473 
2474 	/*
2475 	 * First, check for hit on the EOF offset cache
2476 	 */
2477 	NFSLOCKNODE(np);
2478 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2479 	    (np->n_flag & NMODIFIED) == 0) {
2480 		NFSUNLOCKNODE(np);
2481 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2482 			NFSLOCKNODE(np);
2483 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2484 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2485 				NFSUNLOCKNODE(np);
2486 				NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2487 				if (ap->a_eofflag != NULL)
2488 					*ap->a_eofflag = 1;
2489 				return (0);
2490 			} else
2491 				NFSUNLOCKNODE(np);
2492 		}
2493 	} else
2494 		NFSUNLOCKNODE(np);
2495 
2496 	/*
2497 	 * NFS always guarantees that directory entries don't straddle
2498 	 * DIRBLKSIZ boundaries.  As such, we need to limit the size
2499 	 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2500 	 * directory entry.
2501 	 */
2502 	left = uio->uio_resid % DIRBLKSIZ;
2503 	if (left == uio->uio_resid)
2504 		return (EINVAL);
2505 	uio->uio_resid -= left;
2506 
2507 	/*
2508 	 * Call ncl_bioread() to do the real work.
2509 	 */
2510 	tresid = uio->uio_resid;
2511 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2512 
2513 	if (!error && uio->uio_resid == tresid) {
2514 		NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2515 		if (ap->a_eofflag != NULL)
2516 			*ap->a_eofflag = 1;
2517 	}
2518 
2519 	/* Add the partial DIRBLKSIZ (left) back in. */
2520 	uio->uio_resid += left;
2521 	return (error);
2522 }
2523 
2524 /*
2525  * Readdir rpc call.
2526  * Called from below the buffer cache by ncl_doio().
2527  */
2528 int
2529 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2530     struct thread *td)
2531 {
2532 	struct nfsvattr nfsva;
2533 	nfsuint64 *cookiep, cookie;
2534 	struct nfsnode *dnp = VTONFS(vp);
2535 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2536 	int error = 0, eof, attrflag;
2537 
2538 	KASSERT(uiop->uio_iovcnt == 1 &&
2539 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2540 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2541 	    ("nfs readdirrpc bad uio"));
2542 
2543 	/*
2544 	 * If there is no cookie, assume directory was stale.
2545 	 */
2546 	ncl_dircookie_lock(dnp);
2547 	NFSUNLOCKNODE(dnp);
2548 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2549 	if (cookiep) {
2550 		cookie = *cookiep;
2551 		ncl_dircookie_unlock(dnp);
2552 	} else {
2553 		ncl_dircookie_unlock(dnp);
2554 		return (NFSERR_BAD_COOKIE);
2555 	}
2556 
2557 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2558 		(void)ncl_fsinfo(nmp, vp, cred, td);
2559 
2560 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2561 	    &attrflag, &eof, NULL);
2562 	if (attrflag)
2563 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2564 
2565 	if (!error) {
2566 		/*
2567 		 * We are now either at the end of the directory or have filled
2568 		 * the block.
2569 		 */
2570 		if (eof) {
2571 			NFSLOCKNODE(dnp);
2572 			dnp->n_direofoffset = uiop->uio_offset;
2573 			NFSUNLOCKNODE(dnp);
2574 		} else {
2575 			if (uiop->uio_resid > 0)
2576 				printf("EEK! readdirrpc resid > 0\n");
2577 			ncl_dircookie_lock(dnp);
2578 			NFSUNLOCKNODE(dnp);
2579 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2580 			*cookiep = cookie;
2581 			ncl_dircookie_unlock(dnp);
2582 		}
2583 	} else if (NFS_ISV4(vp)) {
2584 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2585 	}
2586 	return (error);
2587 }
2588 
2589 /*
2590  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2591  */
2592 int
2593 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2594     struct thread *td)
2595 {
2596 	struct nfsvattr nfsva;
2597 	nfsuint64 *cookiep, cookie;
2598 	struct nfsnode *dnp = VTONFS(vp);
2599 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2600 	int error = 0, attrflag, eof;
2601 
2602 	KASSERT(uiop->uio_iovcnt == 1 &&
2603 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2604 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2605 	    ("nfs readdirplusrpc bad uio"));
2606 
2607 	/*
2608 	 * If there is no cookie, assume directory was stale.
2609 	 */
2610 	ncl_dircookie_lock(dnp);
2611 	NFSUNLOCKNODE(dnp);
2612 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2613 	if (cookiep) {
2614 		cookie = *cookiep;
2615 		ncl_dircookie_unlock(dnp);
2616 	} else {
2617 		ncl_dircookie_unlock(dnp);
2618 		return (NFSERR_BAD_COOKIE);
2619 	}
2620 
2621 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2622 		(void)ncl_fsinfo(nmp, vp, cred, td);
2623 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2624 	    &attrflag, &eof, NULL);
2625 	if (attrflag)
2626 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2627 
2628 	if (!error) {
2629 		/*
2630 		 * We are now either at end of the directory or have filled the
2631 		 * the block.
2632 		 */
2633 		if (eof) {
2634 			NFSLOCKNODE(dnp);
2635 			dnp->n_direofoffset = uiop->uio_offset;
2636 			NFSUNLOCKNODE(dnp);
2637 		} else {
2638 			if (uiop->uio_resid > 0)
2639 				printf("EEK! readdirplusrpc resid > 0\n");
2640 			ncl_dircookie_lock(dnp);
2641 			NFSUNLOCKNODE(dnp);
2642 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2643 			*cookiep = cookie;
2644 			ncl_dircookie_unlock(dnp);
2645 		}
2646 	} else if (NFS_ISV4(vp)) {
2647 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2648 	}
2649 	return (error);
2650 }
2651 
2652 /*
2653  * Silly rename. To make the NFS filesystem that is stateless look a little
2654  * more like the "ufs" a remove of an active vnode is translated to a rename
2655  * to a funny looking filename that is removed by nfs_inactive on the
2656  * nfsnode. There is the potential for another process on a different client
2657  * to create the same funny name between the nfs_lookitup() fails and the
2658  * nfs_rename() completes, but...
2659  */
2660 static int
2661 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2662 {
2663 	struct sillyrename *sp;
2664 	struct nfsnode *np;
2665 	int error;
2666 	short pid;
2667 	unsigned int lticks;
2668 
2669 	cache_purge(dvp);
2670 	np = VTONFS(vp);
2671 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2672 	sp = malloc(sizeof (struct sillyrename),
2673 	    M_NEWNFSREQ, M_WAITOK);
2674 	sp->s_cred = crhold(cnp->cn_cred);
2675 	sp->s_dvp = dvp;
2676 	VREF(dvp);
2677 
2678 	/*
2679 	 * Fudge together a funny name.
2680 	 * Changing the format of the funny name to accommodate more
2681 	 * sillynames per directory.
2682 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2683 	 * CPU ticks since boot.
2684 	 */
2685 	pid = curthread->td_proc->p_pid;
2686 	lticks = (unsigned int)ticks;
2687 	for ( ; ; ) {
2688 		sp->s_namlen = sprintf(sp->s_name,
2689 				       ".nfs.%08x.%04x4.4", lticks,
2690 				       pid);
2691 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2692 				 curthread, NULL))
2693 			break;
2694 		lticks++;
2695 	}
2696 	error = nfs_renameit(dvp, vp, cnp, sp);
2697 	if (error)
2698 		goto bad;
2699 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2700 		curthread, &np);
2701 	np->n_sillyrename = sp;
2702 	return (0);
2703 bad:
2704 	vrele(sp->s_dvp);
2705 	crfree(sp->s_cred);
2706 	free(sp, M_NEWNFSREQ);
2707 	return (error);
2708 }
2709 
2710 /*
2711  * Look up a file name and optionally either update the file handle or
2712  * allocate an nfsnode, depending on the value of npp.
2713  * npp == NULL	--> just do the lookup
2714  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2715  *			handled too
2716  * *npp != NULL --> update the file handle in the vnode
2717  */
2718 static int
2719 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2720     struct thread *td, struct nfsnode **npp)
2721 {
2722 	struct vnode *newvp = NULL, *vp;
2723 	struct nfsnode *np, *dnp = VTONFS(dvp);
2724 	struct nfsfh *nfhp, *onfhp;
2725 	struct nfsvattr nfsva, dnfsva;
2726 	struct componentname cn;
2727 	int error = 0, attrflag, dattrflag;
2728 	u_int hash;
2729 	struct timespec ts;
2730 
2731 	nanouptime(&ts);
2732 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2733 	    &nfhp, &attrflag, &dattrflag, NULL, 0);
2734 	if (dattrflag)
2735 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2736 	if (npp && !error) {
2737 		if (*npp != NULL) {
2738 		    np = *npp;
2739 		    vp = NFSTOV(np);
2740 		    /*
2741 		     * For NFSv4, check to see if it is the same name and
2742 		     * replace the name, if it is different.
2743 		     */
2744 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2745 			(np->n_v4->n4_namelen != len ||
2746 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2747 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2748 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2749 			 dnp->n_fhp->nfh_len))) {
2750 #ifdef notdef
2751 { char nnn[100]; int nnnl;
2752 nnnl = (len < 100) ? len : 99;
2753 bcopy(name, nnn, nnnl);
2754 nnn[nnnl] = '\0';
2755 printf("replace=%s\n",nnn);
2756 }
2757 #endif
2758 			    free(np->n_v4, M_NFSV4NODE);
2759 			    np->n_v4 = malloc(
2760 				sizeof (struct nfsv4node) +
2761 				dnp->n_fhp->nfh_len + len - 1,
2762 				M_NFSV4NODE, M_WAITOK);
2763 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2764 			    np->n_v4->n4_namelen = len;
2765 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2766 				dnp->n_fhp->nfh_len);
2767 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2768 		    }
2769 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2770 			FNV1_32_INIT);
2771 		    onfhp = np->n_fhp;
2772 		    /*
2773 		     * Rehash node for new file handle.
2774 		     */
2775 		    vfs_hash_rehash(vp, hash);
2776 		    np->n_fhp = nfhp;
2777 		    if (onfhp != NULL)
2778 			free(onfhp, M_NFSFH);
2779 		    newvp = NFSTOV(np);
2780 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2781 		    free(nfhp, M_NFSFH);
2782 		    VREF(dvp);
2783 		    newvp = dvp;
2784 		} else {
2785 		    cn.cn_nameptr = name;
2786 		    cn.cn_namelen = len;
2787 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2788 			&np, NULL, LK_EXCLUSIVE);
2789 		    if (error)
2790 			return (error);
2791 		    newvp = NFSTOV(np);
2792 		    /*
2793 		     * If n_localmodtime >= time before RPC, then
2794 		     * a file modification operation, such as
2795 		     * VOP_SETATTR() of size, has occurred while
2796 		     * the Lookup RPC and acquisition of the vnode
2797 		     * happened.  As such, the attributes might
2798 		     * be stale, with possibly an incorrect size.
2799 		     */
2800 		    NFSLOCKNODE(np);
2801 		    if (timespecisset(&np->n_localmodtime) &&
2802 			timespeccmp(&np->n_localmodtime, &ts, >=)) {
2803 			NFSCL_DEBUG(4, "nfs_lookitup: localmod "
2804 			    "stale attributes\n");
2805 			attrflag = 0;
2806 		    }
2807 		    NFSUNLOCKNODE(np);
2808 		}
2809 		if (!attrflag && *npp == NULL) {
2810 			if (newvp == dvp)
2811 				vrele(newvp);
2812 			else
2813 				vput(newvp);
2814 			return (ENOENT);
2815 		}
2816 		if (attrflag)
2817 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2818 			    0, 1);
2819 	}
2820 	if (npp && *npp == NULL) {
2821 		if (error) {
2822 			if (newvp) {
2823 				if (newvp == dvp)
2824 					vrele(newvp);
2825 				else
2826 					vput(newvp);
2827 			}
2828 		} else
2829 			*npp = np;
2830 	}
2831 	if (error && NFS_ISV4(dvp))
2832 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2833 	return (error);
2834 }
2835 
2836 /*
2837  * Nfs Version 3 and 4 commit rpc
2838  */
2839 int
2840 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2841    struct thread *td)
2842 {
2843 	struct nfsvattr nfsva;
2844 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2845 	struct nfsnode *np;
2846 	struct uio uio;
2847 	int error, attrflag;
2848 
2849 	np = VTONFS(vp);
2850 	error = EIO;
2851 	attrflag = 0;
2852 	if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2853 		uio.uio_offset = offset;
2854 		uio.uio_resid = cnt;
2855 		error = nfscl_doiods(vp, &uio, NULL, NULL,
2856 		    NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2857 		if (error != 0) {
2858 			NFSLOCKNODE(np);
2859 			np->n_flag &= ~NDSCOMMIT;
2860 			NFSUNLOCKNODE(np);
2861 		}
2862 	}
2863 	if (error != 0) {
2864 		mtx_lock(&nmp->nm_mtx);
2865 		if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2866 			mtx_unlock(&nmp->nm_mtx);
2867 			return (0);
2868 		}
2869 		mtx_unlock(&nmp->nm_mtx);
2870 		error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2871 		    &attrflag, NULL);
2872 	}
2873 	if (attrflag != 0)
2874 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2875 		    0, 1);
2876 	if (error != 0 && NFS_ISV4(vp))
2877 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2878 	return (error);
2879 }
2880 
2881 /*
2882  * Strategy routine.
2883  * For async requests when nfsiod(s) are running, queue the request by
2884  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2885  * request.
2886  */
2887 static int
2888 nfs_strategy(struct vop_strategy_args *ap)
2889 {
2890 	struct buf *bp;
2891 	struct vnode *vp;
2892 	struct ucred *cr;
2893 
2894 	bp = ap->a_bp;
2895 	vp = ap->a_vp;
2896 	KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2897 	KASSERT(!(bp->b_flags & B_DONE),
2898 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2899 
2900 	if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2901 		bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2902 		    DEV_BSIZE);
2903 	if (bp->b_iocmd == BIO_READ)
2904 		cr = bp->b_rcred;
2905 	else
2906 		cr = bp->b_wcred;
2907 
2908 	/*
2909 	 * If the op is asynchronous and an i/o daemon is waiting
2910 	 * queue the request, wake it up and wait for completion
2911 	 * otherwise just do it ourselves.
2912 	 */
2913 	if ((bp->b_flags & B_ASYNC) == 0 ||
2914 	    ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2915 		(void) ncl_doio(vp, bp, cr, curthread, 1);
2916 	return (0);
2917 }
2918 
2919 /*
2920  * fsync vnode op. Just call ncl_flush() with commit == 1.
2921  */
2922 /* ARGSUSED */
2923 static int
2924 nfs_fsync(struct vop_fsync_args *ap)
2925 {
2926 
2927 	if (ap->a_vp->v_type != VREG) {
2928 		/*
2929 		 * For NFS, metadata is changed synchronously on the server,
2930 		 * so there is nothing to flush. Also, ncl_flush() clears
2931 		 * the NMODIFIED flag and that shouldn't be done here for
2932 		 * directories.
2933 		 */
2934 		return (0);
2935 	}
2936 	return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2937 }
2938 
2939 /*
2940  * Flush all the blocks associated with a vnode.
2941  * 	Walk through the buffer pool and push any dirty pages
2942  *	associated with the vnode.
2943  * If the called_from_renewthread argument is TRUE, it has been called
2944  * from the NFSv4 renew thread and, as such, cannot block indefinitely
2945  * waiting for a buffer write to complete.
2946  */
2947 int
2948 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2949     int commit, int called_from_renewthread)
2950 {
2951 	struct nfsnode *np = VTONFS(vp);
2952 	struct buf *bp;
2953 	int i;
2954 	struct buf *nbp;
2955 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2956 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2957 	int passone = 1, trycnt = 0;
2958 	u_quad_t off, endoff, toff;
2959 	struct ucred* wcred = NULL;
2960 	struct buf **bvec = NULL;
2961 	struct bufobj *bo;
2962 #ifndef NFS_COMMITBVECSIZ
2963 #define	NFS_COMMITBVECSIZ	20
2964 #endif
2965 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2966 	u_int bvecsize = 0, bveccount;
2967 	struct timespec ts;
2968 
2969 	if (called_from_renewthread != 0)
2970 		slptimeo = hz;
2971 	if (nmp->nm_flag & NFSMNT_INT)
2972 		slpflag = PCATCH;
2973 	if (!commit)
2974 		passone = 0;
2975 	bo = &vp->v_bufobj;
2976 	/*
2977 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2978 	 * server, but has not been committed to stable storage on the server
2979 	 * yet. On the first pass, the byte range is worked out and the commit
2980 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2981 	 * job.
2982 	 */
2983 again:
2984 	off = (u_quad_t)-1;
2985 	endoff = 0;
2986 	bvecpos = 0;
2987 	if (NFS_ISV34(vp) && commit) {
2988 		if (bvec != NULL && bvec != bvec_on_stack)
2989 			free(bvec, M_TEMP);
2990 		/*
2991 		 * Count up how many buffers waiting for a commit.
2992 		 */
2993 		bveccount = 0;
2994 		BO_LOCK(bo);
2995 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2996 			if (!BUF_ISLOCKED(bp) &&
2997 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2998 				== (B_DELWRI | B_NEEDCOMMIT))
2999 				bveccount++;
3000 		}
3001 		/*
3002 		 * Allocate space to remember the list of bufs to commit.  It is
3003 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
3004 		 * If we can't get memory (for whatever reason), we will end up
3005 		 * committing the buffers one-by-one in the loop below.
3006 		 */
3007 		if (bveccount > NFS_COMMITBVECSIZ) {
3008 			/*
3009 			 * Release the vnode interlock to avoid a lock
3010 			 * order reversal.
3011 			 */
3012 			BO_UNLOCK(bo);
3013 			bvec = (struct buf **)
3014 				malloc(bveccount * sizeof(struct buf *),
3015 				       M_TEMP, M_NOWAIT);
3016 			BO_LOCK(bo);
3017 			if (bvec == NULL) {
3018 				bvec = bvec_on_stack;
3019 				bvecsize = NFS_COMMITBVECSIZ;
3020 			} else
3021 				bvecsize = bveccount;
3022 		} else {
3023 			bvec = bvec_on_stack;
3024 			bvecsize = NFS_COMMITBVECSIZ;
3025 		}
3026 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3027 			if (bvecpos >= bvecsize)
3028 				break;
3029 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3030 				nbp = TAILQ_NEXT(bp, b_bobufs);
3031 				continue;
3032 			}
3033 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3034 			    (B_DELWRI | B_NEEDCOMMIT)) {
3035 				BUF_UNLOCK(bp);
3036 				nbp = TAILQ_NEXT(bp, b_bobufs);
3037 				continue;
3038 			}
3039 			BO_UNLOCK(bo);
3040 			bremfree(bp);
3041 			/*
3042 			 * Work out if all buffers are using the same cred
3043 			 * so we can deal with them all with one commit.
3044 			 *
3045 			 * NOTE: we are not clearing B_DONE here, so we have
3046 			 * to do it later on in this routine if we intend to
3047 			 * initiate I/O on the bp.
3048 			 *
3049 			 * Note: to avoid loopback deadlocks, we do not
3050 			 * assign b_runningbufspace.
3051 			 */
3052 			if (wcred == NULL)
3053 				wcred = bp->b_wcred;
3054 			else if (wcred != bp->b_wcred)
3055 				wcred = NOCRED;
3056 			vfs_busy_pages(bp, 1);
3057 
3058 			BO_LOCK(bo);
3059 			/*
3060 			 * bp is protected by being locked, but nbp is not
3061 			 * and vfs_busy_pages() may sleep.  We have to
3062 			 * recalculate nbp.
3063 			 */
3064 			nbp = TAILQ_NEXT(bp, b_bobufs);
3065 
3066 			/*
3067 			 * A list of these buffers is kept so that the
3068 			 * second loop knows which buffers have actually
3069 			 * been committed. This is necessary, since there
3070 			 * may be a race between the commit rpc and new
3071 			 * uncommitted writes on the file.
3072 			 */
3073 			bvec[bvecpos++] = bp;
3074 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3075 				bp->b_dirtyoff;
3076 			if (toff < off)
3077 				off = toff;
3078 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3079 			if (toff > endoff)
3080 				endoff = toff;
3081 		}
3082 		BO_UNLOCK(bo);
3083 	}
3084 	if (bvecpos > 0) {
3085 		/*
3086 		 * Commit data on the server, as required.
3087 		 * If all bufs are using the same wcred, then use that with
3088 		 * one call for all of them, otherwise commit each one
3089 		 * separately.
3090 		 */
3091 		if (wcred != NOCRED)
3092 			retv = ncl_commit(vp, off, (int)(endoff - off),
3093 					  wcred, td);
3094 		else {
3095 			retv = 0;
3096 			for (i = 0; i < bvecpos; i++) {
3097 				off_t off, size;
3098 				bp = bvec[i];
3099 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3100 					bp->b_dirtyoff;
3101 				size = (u_quad_t)(bp->b_dirtyend
3102 						  - bp->b_dirtyoff);
3103 				retv = ncl_commit(vp, off, (int)size,
3104 						  bp->b_wcred, td);
3105 				if (retv) break;
3106 			}
3107 		}
3108 
3109 		if (retv == NFSERR_STALEWRITEVERF)
3110 			ncl_clearcommit(vp->v_mount);
3111 
3112 		/*
3113 		 * Now, either mark the blocks I/O done or mark the
3114 		 * blocks dirty, depending on whether the commit
3115 		 * succeeded.
3116 		 */
3117 		for (i = 0; i < bvecpos; i++) {
3118 			bp = bvec[i];
3119 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3120 			if (!NFSCL_FORCEDISM(vp->v_mount) && retv) {
3121 				/*
3122 				 * Error, leave B_DELWRI intact
3123 				 */
3124 				vfs_unbusy_pages(bp);
3125 				brelse(bp);
3126 			} else {
3127 				/*
3128 				 * Success, remove B_DELWRI ( bundirty() ).
3129 				 *
3130 				 * b_dirtyoff/b_dirtyend seem to be NFS
3131 				 * specific.  We should probably move that
3132 				 * into bundirty(). XXX
3133 				 */
3134 				bufobj_wref(bo);
3135 				bp->b_flags |= B_ASYNC;
3136 				bundirty(bp);
3137 				bp->b_flags &= ~B_DONE;
3138 				bp->b_ioflags &= ~BIO_ERROR;
3139 				bp->b_dirtyoff = bp->b_dirtyend = 0;
3140 				bufdone(bp);
3141 			}
3142 		}
3143 	}
3144 
3145 	/*
3146 	 * Start/do any write(s) that are required.
3147 	 */
3148 loop:
3149 	BO_LOCK(bo);
3150 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3151 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3152 			if (waitfor != MNT_WAIT || passone)
3153 				continue;
3154 
3155 			error = BUF_TIMELOCK(bp,
3156 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3157 			    BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
3158 			if (error == 0) {
3159 				BUF_UNLOCK(bp);
3160 				goto loop;
3161 			}
3162 			if (error == ENOLCK) {
3163 				error = 0;
3164 				goto loop;
3165 			}
3166 			if (called_from_renewthread != 0) {
3167 				/*
3168 				 * Return EIO so the flush will be retried
3169 				 * later.
3170 				 */
3171 				error = EIO;
3172 				goto done;
3173 			}
3174 			if (newnfs_sigintr(nmp, td)) {
3175 				error = EINTR;
3176 				goto done;
3177 			}
3178 			if (slpflag == PCATCH) {
3179 				slpflag = 0;
3180 				slptimeo = 2 * hz;
3181 			}
3182 			goto loop;
3183 		}
3184 		if ((bp->b_flags & B_DELWRI) == 0)
3185 			panic("nfs_fsync: not dirty");
3186 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3187 			BUF_UNLOCK(bp);
3188 			continue;
3189 		}
3190 		BO_UNLOCK(bo);
3191 		bremfree(bp);
3192 		bp->b_flags |= B_ASYNC;
3193 		bwrite(bp);
3194 		if (newnfs_sigintr(nmp, td)) {
3195 			error = EINTR;
3196 			goto done;
3197 		}
3198 		goto loop;
3199 	}
3200 	if (passone) {
3201 		passone = 0;
3202 		BO_UNLOCK(bo);
3203 		goto again;
3204 	}
3205 	if (waitfor == MNT_WAIT) {
3206 		while (bo->bo_numoutput) {
3207 			error = bufobj_wwait(bo, slpflag, slptimeo);
3208 			if (error) {
3209 			    BO_UNLOCK(bo);
3210 			    if (called_from_renewthread != 0) {
3211 				/*
3212 				 * Return EIO so that the flush will be
3213 				 * retried later.
3214 				 */
3215 				error = EIO;
3216 				goto done;
3217 			    }
3218 			    error = newnfs_sigintr(nmp, td);
3219 			    if (error)
3220 				goto done;
3221 			    if (slpflag == PCATCH) {
3222 				slpflag = 0;
3223 				slptimeo = 2 * hz;
3224 			    }
3225 			    BO_LOCK(bo);
3226 			}
3227 		}
3228 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
3229 			BO_UNLOCK(bo);
3230 			goto loop;
3231 		}
3232 		/*
3233 		 * Wait for all the async IO requests to drain
3234 		 */
3235 		BO_UNLOCK(bo);
3236 		NFSLOCKNODE(np);
3237 		while (np->n_directio_asyncwr > 0) {
3238 			np->n_flag |= NFSYNCWAIT;
3239 			error = newnfs_msleep(td, &np->n_directio_asyncwr,
3240 			    &np->n_mtx, slpflag | (PRIBIO + 1),
3241 			    "nfsfsync", 0);
3242 			if (error) {
3243 				if (newnfs_sigintr(nmp, td)) {
3244 					NFSUNLOCKNODE(np);
3245 					error = EINTR;
3246 					goto done;
3247 				}
3248 			}
3249 		}
3250 		NFSUNLOCKNODE(np);
3251 	} else
3252 		BO_UNLOCK(bo);
3253 	if (NFSHASPNFS(nmp)) {
3254 		nfscl_layoutcommit(vp, td);
3255 		/*
3256 		 * Invalidate the attribute cache, since writes to a DS
3257 		 * won't update the size attribute.
3258 		 */
3259 		NFSLOCKNODE(np);
3260 		np->n_attrstamp = 0;
3261 	} else
3262 		NFSLOCKNODE(np);
3263 	if (np->n_flag & NWRITEERR) {
3264 		error = np->n_error;
3265 		np->n_flag &= ~NWRITEERR;
3266 	}
3267   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
3268 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3269   		np->n_flag &= ~NMODIFIED;
3270 	NFSUNLOCKNODE(np);
3271 done:
3272 	if (bvec != NULL && bvec != bvec_on_stack)
3273 		free(bvec, M_TEMP);
3274 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
3275 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
3276 	    np->n_directio_asyncwr != 0)) {
3277 		if (trycnt++ < 5) {
3278 			/* try, try again... */
3279 			passone = 1;
3280 			wcred = NULL;
3281 			bvec = NULL;
3282 			bvecsize = 0;
3283 			goto again;
3284 		}
3285 		vn_printf(vp, "ncl_flush failed");
3286 		error = called_from_renewthread != 0 ? EIO : EBUSY;
3287 	}
3288 	if (error == 0) {
3289 		nanouptime(&ts);
3290 		NFSLOCKNODE(np);
3291 		np->n_localmodtime = ts;
3292 		NFSUNLOCKNODE(np);
3293 	}
3294 	return (error);
3295 }
3296 
3297 /*
3298  * NFS advisory byte-level locks.
3299  */
3300 static int
3301 nfs_advlock(struct vop_advlock_args *ap)
3302 {
3303 	struct vnode *vp = ap->a_vp;
3304 	struct ucred *cred;
3305 	struct nfsnode *np = VTONFS(ap->a_vp);
3306 	struct proc *p = (struct proc *)ap->a_id;
3307 	struct thread *td = curthread;	/* XXX */
3308 	struct vattr va;
3309 	int ret, error;
3310 	u_quad_t size;
3311 	struct nfsmount *nmp;
3312 
3313 	error = NFSVOPLOCK(vp, LK_SHARED);
3314 	if (error != 0)
3315 		return (EBADF);
3316 	if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3317 		if (vp->v_type != VREG) {
3318 			error = EINVAL;
3319 			goto out;
3320 		}
3321 		if ((ap->a_flags & F_POSIX) != 0)
3322 			cred = p->p_ucred;
3323 		else
3324 			cred = td->td_ucred;
3325 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
3326 		if (VN_IS_DOOMED(vp)) {
3327 			error = EBADF;
3328 			goto out;
3329 		}
3330 
3331 		/*
3332 		 * If this is unlocking a write locked region, flush and
3333 		 * commit them before unlocking. This is required by
3334 		 * RFC3530 Sec. 9.3.2.
3335 		 */
3336 		if (ap->a_op == F_UNLCK &&
3337 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3338 		    ap->a_flags))
3339 			(void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3340 
3341 		/*
3342 		 * Mark NFS node as might have acquired a lock.
3343 		 * This is separate from NHASBEENLOCKED, because it must
3344 		 * be done before the nfsrpc_advlock() call, which might
3345 		 * add a nfscllock structure to the client state.
3346 		 * It is used to check for the case where a nfscllock
3347 		 * state structure cannot exist for the file.
3348 		 * Only done for "oneopenown" NFSv4.1/4.2 mounts.
3349 		 */
3350 		nmp = VFSTONFS(vp->v_mount);
3351 		if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) {
3352 			NFSLOCKNODE(np);
3353 			np->n_flag |= NMIGHTBELOCKED;
3354 			NFSUNLOCKNODE(np);
3355 		}
3356 
3357 		/*
3358 		 * Loop around doing the lock op, while a blocking lock
3359 		 * must wait for the lock op to succeed.
3360 		 */
3361 		do {
3362 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3363 			    ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3364 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3365 			    ap->a_op == F_SETLK) {
3366 				NFSVOPUNLOCK(vp);
3367 				error = nfs_catnap(PZERO | PCATCH, ret,
3368 				    "ncladvl");
3369 				if (error)
3370 					return (EINTR);
3371 				NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3372 				if (VN_IS_DOOMED(vp)) {
3373 					error = EBADF;
3374 					goto out;
3375 				}
3376 			}
3377 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3378 		     ap->a_op == F_SETLK);
3379 		if (ret == NFSERR_DENIED) {
3380 			error = EAGAIN;
3381 			goto out;
3382 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3383 			error = ret;
3384 			goto out;
3385 		} else if (ret != 0) {
3386 			error = EACCES;
3387 			goto out;
3388 		}
3389 
3390 		/*
3391 		 * Now, if we just got a lock, invalidate data in the buffer
3392 		 * cache, as required, so that the coherency conforms with
3393 		 * RFC3530 Sec. 9.3.2.
3394 		 */
3395 		if (ap->a_op == F_SETLK) {
3396 			if ((np->n_flag & NMODIFIED) == 0) {
3397 				np->n_attrstamp = 0;
3398 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3399 				ret = VOP_GETATTR(vp, &va, cred);
3400 			}
3401 			if ((np->n_flag & NMODIFIED) || ret ||
3402 			    np->n_change != va.va_filerev) {
3403 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3404 				np->n_attrstamp = 0;
3405 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3406 				ret = VOP_GETATTR(vp, &va, cred);
3407 				if (!ret) {
3408 					np->n_mtime = va.va_mtime;
3409 					np->n_change = va.va_filerev;
3410 				}
3411 			}
3412 			/* Mark that a file lock has been acquired. */
3413 			NFSLOCKNODE(np);
3414 			np->n_flag |= NHASBEENLOCKED;
3415 			NFSUNLOCKNODE(np);
3416 		}
3417 	} else if (!NFS_ISV4(vp)) {
3418 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3419 			size = VTONFS(vp)->n_size;
3420 			NFSVOPUNLOCK(vp);
3421 			error = lf_advlock(ap, &(vp->v_lockf), size);
3422 		} else {
3423 			if (nfs_advlock_p != NULL)
3424 				error = nfs_advlock_p(ap);
3425 			else {
3426 				NFSVOPUNLOCK(vp);
3427 				error = ENOLCK;
3428 			}
3429 		}
3430 		if (error == 0 && ap->a_op == F_SETLK) {
3431 			error = NFSVOPLOCK(vp, LK_SHARED);
3432 			if (error == 0) {
3433 				/* Mark that a file lock has been acquired. */
3434 				NFSLOCKNODE(np);
3435 				np->n_flag |= NHASBEENLOCKED;
3436 				NFSUNLOCKNODE(np);
3437 				NFSVOPUNLOCK(vp);
3438 			}
3439 		}
3440 		return (error);
3441 	} else
3442 		error = EOPNOTSUPP;
3443 out:
3444 	NFSVOPUNLOCK(vp);
3445 	return (error);
3446 }
3447 
3448 /*
3449  * NFS advisory byte-level locks.
3450  */
3451 static int
3452 nfs_advlockasync(struct vop_advlockasync_args *ap)
3453 {
3454 	struct vnode *vp = ap->a_vp;
3455 	u_quad_t size;
3456 	int error;
3457 
3458 	if (NFS_ISV4(vp))
3459 		return (EOPNOTSUPP);
3460 	error = NFSVOPLOCK(vp, LK_SHARED);
3461 	if (error)
3462 		return (error);
3463 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3464 		size = VTONFS(vp)->n_size;
3465 		NFSVOPUNLOCK(vp);
3466 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
3467 	} else {
3468 		NFSVOPUNLOCK(vp);
3469 		error = EOPNOTSUPP;
3470 	}
3471 	return (error);
3472 }
3473 
3474 /*
3475  * Print out the contents of an nfsnode.
3476  */
3477 static int
3478 nfs_print(struct vop_print_args *ap)
3479 {
3480 	struct vnode *vp = ap->a_vp;
3481 	struct nfsnode *np = VTONFS(vp);
3482 
3483 	printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3484 	    (uintmax_t)np->n_vattr.na_fsid);
3485 	if (vp->v_type == VFIFO)
3486 		fifo_printinfo(vp);
3487 	printf("\n");
3488 	return (0);
3489 }
3490 
3491 /*
3492  * This is the "real" nfs::bwrite(struct buf*).
3493  * We set B_CACHE if this is a VMIO buffer.
3494  */
3495 int
3496 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3497 {
3498 	int oldflags, rtval;
3499 
3500 	if (bp->b_flags & B_INVAL) {
3501 		brelse(bp);
3502 		return (0);
3503 	}
3504 
3505 	oldflags = bp->b_flags;
3506 	bp->b_flags |= B_CACHE;
3507 
3508 	/*
3509 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3510 	 */
3511 	bundirty(bp);
3512 	bp->b_flags &= ~B_DONE;
3513 	bp->b_ioflags &= ~BIO_ERROR;
3514 	bp->b_iocmd = BIO_WRITE;
3515 
3516 	bufobj_wref(bp->b_bufobj);
3517 	curthread->td_ru.ru_oublock++;
3518 
3519 	/*
3520 	 * Note: to avoid loopback deadlocks, we do not
3521 	 * assign b_runningbufspace.
3522 	 */
3523 	vfs_busy_pages(bp, 1);
3524 
3525 	BUF_KERNPROC(bp);
3526 	bp->b_iooffset = dbtob(bp->b_blkno);
3527 	bstrategy(bp);
3528 
3529 	if ((oldflags & B_ASYNC) != 0)
3530 		return (0);
3531 
3532 	rtval = bufwait(bp);
3533 	if (oldflags & B_DELWRI)
3534 		reassignbuf(bp);
3535 	brelse(bp);
3536 	return (rtval);
3537 }
3538 
3539 /*
3540  * nfs special file access vnode op.
3541  * Essentially just get vattr and then imitate iaccess() since the device is
3542  * local to the client.
3543  */
3544 static int
3545 nfsspec_access(struct vop_access_args *ap)
3546 {
3547 	struct vattr *vap;
3548 	struct ucred *cred = ap->a_cred;
3549 	struct vnode *vp = ap->a_vp;
3550 	accmode_t accmode = ap->a_accmode;
3551 	struct vattr vattr;
3552 	int error;
3553 
3554 	/*
3555 	 * Disallow write attempts on filesystems mounted read-only;
3556 	 * unless the file is a socket, fifo, or a block or character
3557 	 * device resident on the filesystem.
3558 	 */
3559 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3560 		switch (vp->v_type) {
3561 		case VREG:
3562 		case VDIR:
3563 		case VLNK:
3564 			return (EROFS);
3565 		default:
3566 			break;
3567 		}
3568 	}
3569 	vap = &vattr;
3570 	error = VOP_GETATTR(vp, vap, cred);
3571 	if (error)
3572 		goto out;
3573 	error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3574 	    accmode, cred);
3575 out:
3576 	return error;
3577 }
3578 
3579 /*
3580  * Read wrapper for fifos.
3581  */
3582 static int
3583 nfsfifo_read(struct vop_read_args *ap)
3584 {
3585 	struct nfsnode *np = VTONFS(ap->a_vp);
3586 	int error;
3587 
3588 	/*
3589 	 * Set access flag.
3590 	 */
3591 	NFSLOCKNODE(np);
3592 	np->n_flag |= NACC;
3593 	vfs_timestamp(&np->n_atim);
3594 	NFSUNLOCKNODE(np);
3595 	error = fifo_specops.vop_read(ap);
3596 	return error;
3597 }
3598 
3599 /*
3600  * Write wrapper for fifos.
3601  */
3602 static int
3603 nfsfifo_write(struct vop_write_args *ap)
3604 {
3605 	struct nfsnode *np = VTONFS(ap->a_vp);
3606 
3607 	/*
3608 	 * Set update flag.
3609 	 */
3610 	NFSLOCKNODE(np);
3611 	np->n_flag |= NUPD;
3612 	vfs_timestamp(&np->n_mtim);
3613 	NFSUNLOCKNODE(np);
3614 	return(fifo_specops.vop_write(ap));
3615 }
3616 
3617 /*
3618  * Close wrapper for fifos.
3619  *
3620  * Update the times on the nfsnode then do fifo close.
3621  */
3622 static int
3623 nfsfifo_close(struct vop_close_args *ap)
3624 {
3625 	struct vnode *vp = ap->a_vp;
3626 	struct nfsnode *np = VTONFS(vp);
3627 	struct vattr vattr;
3628 	struct timespec ts;
3629 
3630 	NFSLOCKNODE(np);
3631 	if (np->n_flag & (NACC | NUPD)) {
3632 		vfs_timestamp(&ts);
3633 		if (np->n_flag & NACC)
3634 			np->n_atim = ts;
3635 		if (np->n_flag & NUPD)
3636 			np->n_mtim = ts;
3637 		np->n_flag |= NCHG;
3638 		if (vrefcnt(vp) == 1 &&
3639 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3640 			VATTR_NULL(&vattr);
3641 			if (np->n_flag & NACC)
3642 				vattr.va_atime = np->n_atim;
3643 			if (np->n_flag & NUPD)
3644 				vattr.va_mtime = np->n_mtim;
3645 			NFSUNLOCKNODE(np);
3646 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3647 			goto out;
3648 		}
3649 	}
3650 	NFSUNLOCKNODE(np);
3651 out:
3652 	return (fifo_specops.vop_close(ap));
3653 }
3654 
3655 /*
3656  * Just call ncl_writebp() with the force argument set to 1.
3657  *
3658  * NOTE: B_DONE may or may not be set in a_bp on call.
3659  */
3660 static int
3661 nfs_bwrite(struct buf *bp)
3662 {
3663 
3664 	return (ncl_writebp(bp, 1, curthread));
3665 }
3666 
3667 struct buf_ops buf_ops_newnfs = {
3668 	.bop_name	=	"buf_ops_nfs",
3669 	.bop_write	=	nfs_bwrite,
3670 	.bop_strategy	=	bufstrategy,
3671 	.bop_sync	=	bufsync,
3672 	.bop_bdflush	=	bufbdflush,
3673 };
3674 
3675 static int
3676 nfs_getacl(struct vop_getacl_args *ap)
3677 {
3678 	int error;
3679 
3680 	if (ap->a_type != ACL_TYPE_NFS4)
3681 		return (EOPNOTSUPP);
3682 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3683 	    NULL);
3684 	if (error > NFSERR_STALE) {
3685 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3686 		error = EPERM;
3687 	}
3688 	return (error);
3689 }
3690 
3691 static int
3692 nfs_setacl(struct vop_setacl_args *ap)
3693 {
3694 	int error;
3695 
3696 	if (ap->a_type != ACL_TYPE_NFS4)
3697 		return (EOPNOTSUPP);
3698 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3699 	    NULL);
3700 	if (error > NFSERR_STALE) {
3701 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3702 		error = EPERM;
3703 	}
3704 	return (error);
3705 }
3706 
3707 /*
3708  * VOP_ADVISE for NFS.
3709  * Just return 0 for any errors, since it is just a hint.
3710  */
3711 static int
3712 nfs_advise(struct vop_advise_args *ap)
3713 {
3714 	struct thread *td = curthread;
3715 	struct nfsmount *nmp;
3716 	uint64_t len;
3717 	int error;
3718 
3719 	/*
3720 	 * First do vop_stdadvise() to handle the buffer cache.
3721 	 */
3722 	error = vop_stdadvise(ap);
3723 	if (error != 0)
3724 		return (error);
3725 	if (ap->a_start < 0 || ap->a_end < 0)
3726 		return (0);
3727 	if (ap->a_end == OFF_MAX)
3728 		len = 0;
3729 	else if (ap->a_end < ap->a_start)
3730 		return (0);
3731 	else
3732 		len = ap->a_end - ap->a_start + 1;
3733 	nmp = VFSTONFS(ap->a_vp->v_mount);
3734 	mtx_lock(&nmp->nm_mtx);
3735 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3736 	    (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) ==
3737 	    0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) {
3738 		mtx_unlock(&nmp->nm_mtx);
3739 		return (0);
3740 	}
3741 	mtx_unlock(&nmp->nm_mtx);
3742 	error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice,
3743 	    td->td_ucred, td);
3744 	if (error == NFSERR_NOTSUPP) {
3745 		mtx_lock(&nmp->nm_mtx);
3746 		nmp->nm_privflag |= NFSMNTP_NOADVISE;
3747 		mtx_unlock(&nmp->nm_mtx);
3748 	}
3749 	return (0);
3750 }
3751 
3752 /*
3753  * nfs allocate call
3754  */
3755 static int
3756 nfs_allocate(struct vop_allocate_args *ap)
3757 {
3758 	struct vnode *vp = ap->a_vp;
3759 	struct thread *td = curthread;
3760 	struct nfsvattr nfsva;
3761 	struct nfsmount *nmp;
3762 	struct nfsnode *np;
3763 	off_t alen;
3764 	int attrflag, error, ret;
3765 	struct timespec ts;
3766 	struct uio io;
3767 
3768 	attrflag = 0;
3769 	nmp = VFSTONFS(vp->v_mount);
3770 	np = VTONFS(vp);
3771 	mtx_lock(&nmp->nm_mtx);
3772 	if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3773 	    (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) {
3774 		mtx_unlock(&nmp->nm_mtx);
3775 		alen = *ap->a_len;
3776 		if ((uint64_t)alen > nfs_maxalloclen)
3777 			alen = nfs_maxalloclen;
3778 
3779 		/* Check the file size limit. */
3780 		io.uio_offset = *ap->a_offset;
3781 		io.uio_resid = alen;
3782 		error = vn_rlimit_fsize(vp, &io, td);
3783 
3784 		/*
3785 		 * Flush first to ensure that the allocate adds to the
3786 		 * file's allocation on the server.
3787 		 */
3788 		if (error == 0)
3789 			error = ncl_flush(vp, MNT_WAIT, td, 1, 0);
3790 		if (error == 0)
3791 			error = nfsrpc_allocate(vp, *ap->a_offset, alen,
3792 			    &nfsva, &attrflag, ap->a_cred, td, NULL);
3793 		if (error == 0) {
3794 			*ap->a_offset += alen;
3795 			*ap->a_len -= alen;
3796 			nanouptime(&ts);
3797 			NFSLOCKNODE(np);
3798 			np->n_localmodtime = ts;
3799 			NFSUNLOCKNODE(np);
3800 		} else if (error == NFSERR_NOTSUPP) {
3801 			mtx_lock(&nmp->nm_mtx);
3802 			nmp->nm_privflag |= NFSMNTP_NOALLOCATE;
3803 			mtx_unlock(&nmp->nm_mtx);
3804 			error = EINVAL;
3805 		}
3806 	} else {
3807 		mtx_unlock(&nmp->nm_mtx);
3808 		error = EINVAL;
3809 	}
3810 	if (attrflag != 0) {
3811 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3812 		if (error == 0 && ret != 0)
3813 			error = ret;
3814 	}
3815 	if (error != 0)
3816 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3817 	return (error);
3818 }
3819 
3820 /*
3821  * nfs deallocate call
3822  */
3823 static int
3824 nfs_deallocate(struct vop_deallocate_args *ap)
3825 {
3826 	struct vnode *vp = ap->a_vp;
3827 	struct thread *td = curthread;
3828 	struct nfsvattr nfsva;
3829 	struct nfsmount *nmp;
3830 	struct nfsnode *np;
3831 	off_t tlen, mlen;
3832 	int attrflag, error, ret;
3833 	bool clipped;
3834 	struct timespec ts;
3835 
3836 	error = 0;
3837 	attrflag = 0;
3838 	nmp = VFSTONFS(vp->v_mount);
3839 	np = VTONFS(vp);
3840 	mtx_lock(&nmp->nm_mtx);
3841 	if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3842 	    (nmp->nm_privflag & NFSMNTP_NODEALLOCATE) == 0) {
3843 		mtx_unlock(&nmp->nm_mtx);
3844 		tlen = omin(OFF_MAX - *ap->a_offset, *ap->a_len);
3845 		NFSCL_DEBUG(4, "dealloc: off=%jd len=%jd maxfilesize=%ju\n",
3846 		    (intmax_t)*ap->a_offset, (intmax_t)tlen,
3847 		    (uintmax_t)nmp->nm_maxfilesize);
3848 		if ((uint64_t)*ap->a_offset >= nmp->nm_maxfilesize) {
3849 			/* Avoid EFBIG error return from the NFSv4.2 server. */
3850 			*ap->a_len = 0;
3851 			return (0);
3852 		}
3853 		clipped = false;
3854 		if ((uint64_t)*ap->a_offset + tlen > nmp->nm_maxfilesize)
3855 			tlen = nmp->nm_maxfilesize - *ap->a_offset;
3856 		if ((uint64_t)*ap->a_offset < np->n_size) {
3857 			/* Limit the len to nfs_maxalloclen before EOF. */
3858 			mlen = omin((off_t)np->n_size - *ap->a_offset, tlen);
3859 			if ((uint64_t)mlen > nfs_maxalloclen) {
3860 				NFSCL_DEBUG(4, "dealloc: tlen maxalloclen\n");
3861 				tlen = nfs_maxalloclen;
3862 				clipped = true;
3863 			}
3864 		}
3865 		if (error == 0)
3866 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
3867 		if (error == 0) {
3868 			vnode_pager_purge_range(vp, *ap->a_offset,
3869 			    *ap->a_offset + tlen);
3870 			error = nfsrpc_deallocate(vp, *ap->a_offset, tlen,
3871 			    &nfsva, &attrflag, ap->a_cred, td, NULL);
3872 			NFSCL_DEBUG(4, "dealloc: rpc=%d\n", error);
3873 		}
3874 		if (error == 0) {
3875 			NFSCL_DEBUG(4, "dealloc: attrflag=%d na_size=%ju\n",
3876 			    attrflag, (uintmax_t)nfsva.na_size);
3877 			nanouptime(&ts);
3878 			NFSLOCKNODE(np);
3879 			np->n_localmodtime = ts;
3880 			NFSUNLOCKNODE(np);
3881 			if (attrflag != 0) {
3882 				if ((uint64_t)*ap->a_offset < nfsva.na_size)
3883 					*ap->a_offset += omin((off_t)
3884 					    nfsva.na_size - *ap->a_offset,
3885 					    tlen);
3886 			}
3887 			if (clipped && tlen < *ap->a_len)
3888 				*ap->a_len -= tlen;
3889 			else
3890 				*ap->a_len = 0;
3891 		} else if (error == NFSERR_NOTSUPP) {
3892 			mtx_lock(&nmp->nm_mtx);
3893 			nmp->nm_privflag |= NFSMNTP_NODEALLOCATE;
3894 			mtx_unlock(&nmp->nm_mtx);
3895 		}
3896 	} else {
3897 		mtx_unlock(&nmp->nm_mtx);
3898 		error = EIO;
3899 	}
3900 	/*
3901 	 * If the NFS server cannot perform the Deallocate operation, just call
3902 	 * vop_stddeallocate() to perform it.
3903 	 */
3904 	if (error != 0 && error != NFSERR_FBIG && error != NFSERR_INVAL) {
3905 		error = vop_stddeallocate(ap);
3906 		NFSCL_DEBUG(4, "dealloc: stddeallocate=%d\n", error);
3907 	}
3908 	if (attrflag != 0) {
3909 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3910 		if (error == 0 && ret != 0)
3911 			error = ret;
3912 	}
3913 	if (error != 0)
3914 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3915 	return (error);
3916 }
3917 
3918 /*
3919  * nfs copy_file_range call
3920  */
3921 static int
3922 nfs_copy_file_range(struct vop_copy_file_range_args *ap)
3923 {
3924 	struct vnode *invp = ap->a_invp;
3925 	struct vnode *outvp = ap->a_outvp;
3926 	struct mount *mp;
3927 	struct nfsvattr innfsva, outnfsva;
3928 	struct vattr *vap;
3929 	struct uio io;
3930 	struct nfsmount *nmp;
3931 	size_t len, len2;
3932 	int error, inattrflag, outattrflag, ret, ret2;
3933 	off_t inoff, outoff;
3934 	bool consecutive, must_commit, tryoutcred;
3935 
3936 	ret = ret2 = 0;
3937 	nmp = VFSTONFS(invp->v_mount);
3938 	mtx_lock(&nmp->nm_mtx);
3939 	/* NFSv4.2 Copy is not permitted for infile == outfile. */
3940 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3941 	    (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0 || invp == outvp) {
3942 		mtx_unlock(&nmp->nm_mtx);
3943 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3944 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3945 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3946 		return (error);
3947 	}
3948 	mtx_unlock(&nmp->nm_mtx);
3949 
3950 	/* Lock both vnodes, avoiding risk of deadlock. */
3951 	do {
3952 		mp = NULL;
3953 		error = vn_start_write(outvp, &mp, V_WAIT);
3954 		if (error == 0) {
3955 			error = vn_lock(outvp, LK_EXCLUSIVE);
3956 			if (error == 0) {
3957 				error = vn_lock(invp, LK_SHARED | LK_NOWAIT);
3958 				if (error == 0)
3959 					break;
3960 				VOP_UNLOCK(outvp);
3961 				if (mp != NULL)
3962 					vn_finished_write(mp);
3963 				mp = NULL;
3964 				error = vn_lock(invp, LK_SHARED);
3965 				if (error == 0)
3966 					VOP_UNLOCK(invp);
3967 			}
3968 		}
3969 		if (mp != NULL)
3970 			vn_finished_write(mp);
3971 	} while (error == 0);
3972 	if (error != 0)
3973 		return (error);
3974 
3975 	/*
3976 	 * Do the vn_rlimit_fsize() check.  Should this be above the VOP layer?
3977 	 */
3978 	io.uio_offset = *ap->a_outoffp;
3979 	io.uio_resid = *ap->a_lenp;
3980 	error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd);
3981 
3982 	/*
3983 	 * Flush the input file so that the data is up to date before
3984 	 * the copy.  Flush writes for the output file so that they
3985 	 * do not overwrite the data copied to the output file by the Copy.
3986 	 * Set the commit argument for both flushes so that the data is on
3987 	 * stable storage before the Copy RPC.  This is done in case the
3988 	 * server reboots during the Copy and needs to be redone.
3989 	 */
3990 	if (error == 0)
3991 		error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0);
3992 	if (error == 0)
3993 		error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0);
3994 
3995 	/* Do the actual NFSv4.2 RPC. */
3996 	len = *ap->a_lenp;
3997 	mtx_lock(&nmp->nm_mtx);
3998 	if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0)
3999 		consecutive = true;
4000 	else
4001 		consecutive = false;
4002 	mtx_unlock(&nmp->nm_mtx);
4003 	inoff = *ap->a_inoffp;
4004 	outoff = *ap->a_outoffp;
4005 	tryoutcred = true;
4006 	must_commit = false;
4007 	if (error == 0) {
4008 		vap = &VTONFS(invp)->n_vattr.na_vattr;
4009 		error = VOP_GETATTR(invp, vap, ap->a_incred);
4010 		if (error == 0) {
4011 			/*
4012 			 * Clip "len" at va_size so that RFC compliant servers
4013 			 * will not reply NFSERR_INVAL.
4014 			 * Setting "len == 0" for the RPC would be preferred,
4015 			 * but some Linux servers do not support that.
4016 			 */
4017 			if (inoff >= vap->va_size)
4018 				*ap->a_lenp = len = 0;
4019 			else if (inoff + len > vap->va_size)
4020 				*ap->a_lenp = len = vap->va_size - inoff;
4021 		} else
4022 			error = 0;
4023 	}
4024 
4025 	/*
4026 	 * len will be set to 0 upon a successful Copy RPC.
4027 	 * As such, this only loops when the Copy RPC needs to be retried.
4028 	 */
4029 	while (len > 0 && error == 0) {
4030 		inattrflag = outattrflag = 0;
4031 		len2 = len;
4032 		if (tryoutcred)
4033 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
4034 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
4035 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
4036 			    ap->a_outcred, consecutive, &must_commit);
4037 		else
4038 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
4039 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
4040 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
4041 			    ap->a_incred, consecutive, &must_commit);
4042 		if (inattrflag != 0)
4043 			ret = nfscl_loadattrcache(&invp, &innfsva, NULL, NULL,
4044 			    0, 1);
4045 		if (outattrflag != 0)
4046 			ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL,
4047 			    NULL, 1, 1);
4048 		if (error == 0) {
4049 			if (consecutive == false) {
4050 				if (len2 == len) {
4051 					mtx_lock(&nmp->nm_mtx);
4052 					nmp->nm_privflag |=
4053 					    NFSMNTP_NOCONSECUTIVE;
4054 					mtx_unlock(&nmp->nm_mtx);
4055 				} else
4056 					error = NFSERR_OFFLOADNOREQS;
4057 			}
4058 			*ap->a_lenp = len2;
4059 			len = 0;
4060 			if (len2 > 0 && must_commit && error == 0)
4061 				error = ncl_commit(outvp, outoff, *ap->a_lenp,
4062 				    ap->a_outcred, curthread);
4063 			if (error == 0 && ret != 0)
4064 				error = ret;
4065 			if (error == 0 && ret2 != 0)
4066 				error = ret2;
4067 		} else if (error == NFSERR_OFFLOADNOREQS && consecutive) {
4068 			/*
4069 			 * Try consecutive == false, which is ok only if all
4070 			 * bytes are copied.
4071 			 * If only some bytes were copied when consecutive
4072 			 * is false, there is no way to know which bytes
4073 			 * still need to be written.
4074 			 */
4075 			consecutive = false;
4076 			error = 0;
4077 		} else if (error == NFSERR_ACCES && tryoutcred) {
4078 			/* Try again with incred. */
4079 			tryoutcred = false;
4080 			error = 0;
4081 		}
4082 		if (error == NFSERR_STALEWRITEVERF) {
4083 			/*
4084 			 * Server rebooted, so do it all again.
4085 			 */
4086 			*ap->a_inoffp = inoff;
4087 			*ap->a_outoffp = outoff;
4088 			len = *ap->a_lenp;
4089 			must_commit = false;
4090 			error = 0;
4091 		}
4092 	}
4093 	VOP_UNLOCK(invp);
4094 	VOP_UNLOCK(outvp);
4095 	if (mp != NULL)
4096 		vn_finished_write(mp);
4097 	if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS ||
4098 	    error == NFSERR_ACCES) {
4099 		/*
4100 		 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can
4101 		 * use a_incred for the read and a_outcred for the write, so
4102 		 * try this for NFSERR_ACCES failures for the Copy.
4103 		 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can
4104 		 * never succeed, so disable it.
4105 		 */
4106 		if (error != NFSERR_ACCES) {
4107 			/* Can never do Copy on this mount. */
4108 			mtx_lock(&nmp->nm_mtx);
4109 			nmp->nm_privflag |= NFSMNTP_NOCOPY;
4110 			mtx_unlock(&nmp->nm_mtx);
4111 		}
4112 		*ap->a_inoffp = inoff;
4113 		*ap->a_outoffp = outoff;
4114 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
4115 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
4116 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
4117 	} else if (error != 0)
4118 		*ap->a_lenp = 0;
4119 
4120 	if (error != 0)
4121 		error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0);
4122 	return (error);
4123 }
4124 
4125 /*
4126  * nfs ioctl call
4127  */
4128 static int
4129 nfs_ioctl(struct vop_ioctl_args *ap)
4130 {
4131 	struct vnode *vp = ap->a_vp;
4132 	struct nfsvattr nfsva;
4133 	struct nfsmount *nmp;
4134 	int attrflag, content, error, ret;
4135 	bool eof = false;			/* shut up compiler. */
4136 
4137 	if (vp->v_type != VREG)
4138 		return (ENOTTY);
4139 	nmp = VFSTONFS(vp->v_mount);
4140 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) {
4141 		error = vop_stdioctl(ap);
4142 		return (error);
4143 	}
4144 
4145 	/* Do the actual NFSv4.2 RPC. */
4146 	switch (ap->a_command) {
4147 	case FIOSEEKDATA:
4148 		content = NFSV4CONTENT_DATA;
4149 		break;
4150 	case FIOSEEKHOLE:
4151 		content = NFSV4CONTENT_HOLE;
4152 		break;
4153 	default:
4154 		return (ENOTTY);
4155 	}
4156 
4157 	error = vn_lock(vp, LK_SHARED);
4158 	if (error != 0)
4159 		return (EBADF);
4160 	attrflag = 0;
4161 	if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size)
4162 		error = ENXIO;
4163 	else {
4164 		/*
4165 		 * Flush all writes, so that the server is up to date.
4166 		 * Although a Commit is not required, the commit argument
4167 		 * is set so that, for a pNFS File/Flexible File Layout
4168 		 * server, the LayoutCommit will be done to ensure the file
4169 		 * size is up to date on the Metadata Server.
4170 		 */
4171 		error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0);
4172 		if (error == 0)
4173 			error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof,
4174 			    content, ap->a_cred, &nfsva, &attrflag);
4175 		/* If at eof for FIOSEEKDATA, return ENXIO. */
4176 		if (eof && error == 0 && content == NFSV4CONTENT_DATA)
4177 			error = ENXIO;
4178 	}
4179 	if (attrflag != 0) {
4180 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4181 		if (error == 0 && ret != 0)
4182 			error = ret;
4183 	}
4184 	NFSVOPUNLOCK(vp);
4185 
4186 	if (error != 0)
4187 		error = ENXIO;
4188 	return (error);
4189 }
4190 
4191 /*
4192  * nfs getextattr call
4193  */
4194 static int
4195 nfs_getextattr(struct vop_getextattr_args *ap)
4196 {
4197 	struct vnode *vp = ap->a_vp;
4198 	struct nfsmount *nmp;
4199 	struct ucred *cred;
4200 	struct thread *td = ap->a_td;
4201 	struct nfsvattr nfsva;
4202 	ssize_t len;
4203 	int attrflag, error, ret;
4204 
4205 	nmp = VFSTONFS(vp->v_mount);
4206 	mtx_lock(&nmp->nm_mtx);
4207 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4208 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4209 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4210 		mtx_unlock(&nmp->nm_mtx);
4211 		return (EOPNOTSUPP);
4212 	}
4213 	mtx_unlock(&nmp->nm_mtx);
4214 
4215 	cred = ap->a_cred;
4216 	if (cred == NULL)
4217 		cred = td->td_ucred;
4218 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4219 	attrflag = 0;
4220 	error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva,
4221 	    &attrflag, cred, td);
4222 	if (attrflag != 0) {
4223 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4224 		if (error == 0 && ret != 0)
4225 			error = ret;
4226 	}
4227 	if (error == 0 && ap->a_size != NULL)
4228 		*ap->a_size = len;
4229 
4230 	switch (error) {
4231 	case NFSERR_NOTSUPP:
4232 	case NFSERR_OPILLEGAL:
4233 		mtx_lock(&nmp->nm_mtx);
4234 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4235 		mtx_unlock(&nmp->nm_mtx);
4236 		error = EOPNOTSUPP;
4237 		break;
4238 	case NFSERR_NOXATTR:
4239 	case NFSERR_XATTR2BIG:
4240 		error = ENOATTR;
4241 		break;
4242 	default:
4243 		error = nfscl_maperr(td, error, 0, 0);
4244 		break;
4245 	}
4246 	return (error);
4247 }
4248 
4249 /*
4250  * nfs setextattr call
4251  */
4252 static int
4253 nfs_setextattr(struct vop_setextattr_args *ap)
4254 {
4255 	struct vnode *vp = ap->a_vp;
4256 	struct nfsmount *nmp;
4257 	struct ucred *cred;
4258 	struct thread *td = ap->a_td;
4259 	struct nfsvattr nfsva;
4260 	int attrflag, error, ret;
4261 
4262 	nmp = VFSTONFS(vp->v_mount);
4263 	mtx_lock(&nmp->nm_mtx);
4264 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4265 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4266 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4267 		mtx_unlock(&nmp->nm_mtx);
4268 		return (EOPNOTSUPP);
4269 	}
4270 	mtx_unlock(&nmp->nm_mtx);
4271 
4272 	if (ap->a_uio->uio_resid < 0)
4273 		return (EINVAL);
4274 	cred = ap->a_cred;
4275 	if (cred == NULL)
4276 		cred = td->td_ucred;
4277 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4278 	attrflag = 0;
4279 	error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva,
4280 	    &attrflag, cred, td);
4281 	if (attrflag != 0) {
4282 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4283 		if (error == 0 && ret != 0)
4284 			error = ret;
4285 	}
4286 
4287 	switch (error) {
4288 	case NFSERR_NOTSUPP:
4289 	case NFSERR_OPILLEGAL:
4290 		mtx_lock(&nmp->nm_mtx);
4291 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4292 		mtx_unlock(&nmp->nm_mtx);
4293 		error = EOPNOTSUPP;
4294 		break;
4295 	case NFSERR_NOXATTR:
4296 	case NFSERR_XATTR2BIG:
4297 		error = ENOATTR;
4298 		break;
4299 	default:
4300 		error = nfscl_maperr(td, error, 0, 0);
4301 		break;
4302 	}
4303 	return (error);
4304 }
4305 
4306 /*
4307  * nfs listextattr call
4308  */
4309 static int
4310 nfs_listextattr(struct vop_listextattr_args *ap)
4311 {
4312 	struct vnode *vp = ap->a_vp;
4313 	struct nfsmount *nmp;
4314 	struct ucred *cred;
4315 	struct thread *td = ap->a_td;
4316 	struct nfsvattr nfsva;
4317 	size_t len, len2;
4318 	uint64_t cookie;
4319 	int attrflag, error, ret;
4320 	bool eof;
4321 
4322 	nmp = VFSTONFS(vp->v_mount);
4323 	mtx_lock(&nmp->nm_mtx);
4324 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4325 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4326 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4327 		mtx_unlock(&nmp->nm_mtx);
4328 		return (EOPNOTSUPP);
4329 	}
4330 	mtx_unlock(&nmp->nm_mtx);
4331 
4332 	cred = ap->a_cred;
4333 	if (cred == NULL)
4334 		cred = td->td_ucred;
4335 
4336 	/* Loop around doing List Extended Attribute RPCs. */
4337 	eof = false;
4338 	cookie = 0;
4339 	len2 = 0;
4340 	error = 0;
4341 	while (!eof && error == 0) {
4342 		len = nmp->nm_rsize;
4343 		attrflag = 0;
4344 		error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof,
4345 		    &nfsva, &attrflag, cred, td);
4346 		if (attrflag != 0) {
4347 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4348 			    1);
4349 			if (error == 0 && ret != 0)
4350 				error = ret;
4351 		}
4352 		if (error == 0) {
4353 			len2 += len;
4354 			if (len2 > SSIZE_MAX)
4355 				error = ENOATTR;
4356 		}
4357 	}
4358 	if (error == 0 && ap->a_size != NULL)
4359 		*ap->a_size = len2;
4360 
4361 	switch (error) {
4362 	case NFSERR_NOTSUPP:
4363 	case NFSERR_OPILLEGAL:
4364 		mtx_lock(&nmp->nm_mtx);
4365 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4366 		mtx_unlock(&nmp->nm_mtx);
4367 		error = EOPNOTSUPP;
4368 		break;
4369 	case NFSERR_NOXATTR:
4370 	case NFSERR_XATTR2BIG:
4371 		error = ENOATTR;
4372 		break;
4373 	default:
4374 		error = nfscl_maperr(td, error, 0, 0);
4375 		break;
4376 	}
4377 	return (error);
4378 }
4379 
4380 /*
4381  * nfs setextattr call
4382  */
4383 static int
4384 nfs_deleteextattr(struct vop_deleteextattr_args *ap)
4385 {
4386 	struct vnode *vp = ap->a_vp;
4387 	struct nfsmount *nmp;
4388 	struct nfsvattr nfsva;
4389 	int attrflag, error, ret;
4390 
4391 	nmp = VFSTONFS(vp->v_mount);
4392 	mtx_lock(&nmp->nm_mtx);
4393 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4394 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4395 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4396 		mtx_unlock(&nmp->nm_mtx);
4397 		return (EOPNOTSUPP);
4398 	}
4399 	mtx_unlock(&nmp->nm_mtx);
4400 
4401 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4402 	attrflag = 0;
4403 	error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred,
4404 	    ap->a_td);
4405 	if (attrflag != 0) {
4406 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4407 		if (error == 0 && ret != 0)
4408 			error = ret;
4409 	}
4410 
4411 	switch (error) {
4412 	case NFSERR_NOTSUPP:
4413 	case NFSERR_OPILLEGAL:
4414 		mtx_lock(&nmp->nm_mtx);
4415 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4416 		mtx_unlock(&nmp->nm_mtx);
4417 		error = EOPNOTSUPP;
4418 		break;
4419 	case NFSERR_NOXATTR:
4420 	case NFSERR_XATTR2BIG:
4421 		error = ENOATTR;
4422 		break;
4423 	default:
4424 		error = nfscl_maperr(ap->a_td, error, 0, 0);
4425 		break;
4426 	}
4427 	return (error);
4428 }
4429 
4430 /*
4431  * Return POSIX pathconf information applicable to nfs filesystems.
4432  */
4433 static int
4434 nfs_pathconf(struct vop_pathconf_args *ap)
4435 {
4436 	struct nfsv3_pathconf pc;
4437 	struct nfsvattr nfsva;
4438 	struct vnode *vp = ap->a_vp;
4439 	struct nfsmount *nmp;
4440 	struct thread *td = curthread;
4441 	off_t off;
4442 	bool eof;
4443 	int attrflag, error;
4444 
4445 	if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
4446 	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
4447 	    ap->a_name == _PC_NO_TRUNC)) ||
4448 	    (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
4449 		/*
4450 		 * Since only the above 4 a_names are returned by the NFSv3
4451 		 * Pathconf RPC, there is no point in doing it for others.
4452 		 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
4453 		 * be used for _PC_NFS4_ACL as well.
4454 		 */
4455 		error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
4456 		    &attrflag, NULL);
4457 		if (attrflag != 0)
4458 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4459 			    1);
4460 		if (error != 0)
4461 			return (error);
4462 	} else {
4463 		/*
4464 		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
4465 		 * just fake them.
4466 		 */
4467 		pc.pc_linkmax = NFS_LINK_MAX;
4468 		pc.pc_namemax = NFS_MAXNAMLEN;
4469 		pc.pc_notrunc = 1;
4470 		pc.pc_chownrestricted = 1;
4471 		pc.pc_caseinsensitive = 0;
4472 		pc.pc_casepreserving = 1;
4473 		error = 0;
4474 	}
4475 	switch (ap->a_name) {
4476 	case _PC_LINK_MAX:
4477 #ifdef _LP64
4478 		*ap->a_retval = pc.pc_linkmax;
4479 #else
4480 		*ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
4481 #endif
4482 		break;
4483 	case _PC_NAME_MAX:
4484 		*ap->a_retval = pc.pc_namemax;
4485 		break;
4486 	case _PC_PIPE_BUF:
4487 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
4488 			*ap->a_retval = PIPE_BUF;
4489 		else
4490 			error = EINVAL;
4491 		break;
4492 	case _PC_CHOWN_RESTRICTED:
4493 		*ap->a_retval = pc.pc_chownrestricted;
4494 		break;
4495 	case _PC_NO_TRUNC:
4496 		*ap->a_retval = pc.pc_notrunc;
4497 		break;
4498 	case _PC_ACL_NFS4:
4499 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
4500 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
4501 			*ap->a_retval = 1;
4502 		else
4503 			*ap->a_retval = 0;
4504 		break;
4505 	case _PC_ACL_PATH_MAX:
4506 		if (NFS_ISV4(vp))
4507 			*ap->a_retval = ACL_MAX_ENTRIES;
4508 		else
4509 			*ap->a_retval = 3;
4510 		break;
4511 	case _PC_PRIO_IO:
4512 		*ap->a_retval = 0;
4513 		break;
4514 	case _PC_SYNC_IO:
4515 		*ap->a_retval = 0;
4516 		break;
4517 	case _PC_ALLOC_SIZE_MIN:
4518 		*ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
4519 		break;
4520 	case _PC_FILESIZEBITS:
4521 		if (NFS_ISV34(vp))
4522 			*ap->a_retval = 64;
4523 		else
4524 			*ap->a_retval = 32;
4525 		break;
4526 	case _PC_REC_INCR_XFER_SIZE:
4527 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4528 		break;
4529 	case _PC_REC_MAX_XFER_SIZE:
4530 		*ap->a_retval = -1; /* means ``unlimited'' */
4531 		break;
4532 	case _PC_REC_MIN_XFER_SIZE:
4533 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4534 		break;
4535 	case _PC_REC_XFER_ALIGN:
4536 		*ap->a_retval = PAGE_SIZE;
4537 		break;
4538 	case _PC_SYMLINK_MAX:
4539 		*ap->a_retval = NFS_MAXPATHLEN;
4540 		break;
4541 	case _PC_MIN_HOLE_SIZE:
4542 		/* Only some NFSv4.2 servers support Seek for Holes. */
4543 		*ap->a_retval = 0;
4544 		nmp = VFSTONFS(vp->v_mount);
4545 		if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) {
4546 			/*
4547 			 * NFSv4.2 doesn't have an attribute for hole size,
4548 			 * so all we can do is see if the Seek operation is
4549 			 * supported and then use f_iosize as a "best guess".
4550 			 */
4551 			mtx_lock(&nmp->nm_mtx);
4552 			if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) {
4553 				mtx_unlock(&nmp->nm_mtx);
4554 				off = 0;
4555 				attrflag = 0;
4556 				error = nfsrpc_seek(vp, &off, &eof,
4557 				    NFSV4CONTENT_HOLE, td->td_ucred, &nfsva,
4558 				    &attrflag);
4559 				if (attrflag != 0)
4560 					nfscl_loadattrcache(&vp, &nfsva,
4561 					    NULL, NULL, 0, 1);
4562 				mtx_lock(&nmp->nm_mtx);
4563 				if (error == NFSERR_NOTSUPP)
4564 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED;
4565 				else
4566 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED |
4567 					    NFSMNTP_SEEK;
4568 				error = 0;
4569 			}
4570 			if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0)
4571 				*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4572 			mtx_unlock(&nmp->nm_mtx);
4573 		}
4574 		break;
4575 
4576 	default:
4577 		error = vop_stdpathconf(ap);
4578 		break;
4579 	}
4580 	return (error);
4581 }
4582