xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision b4a58fbf640409a1e507d9f7b411c83a3f83a2f3)
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 		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
651 		if (error == EINTR || error == EIO) {
652 			if (NFS_ISV4(vp))
653 				(void) nfsrpc_close(vp, 0, ap->a_td);
654 			return (error);
655 		}
656 		NFSLOCKNODE(np);
657 		np->n_attrstamp = 0;
658 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
659 		if (vp->v_type == VDIR)
660 			np->n_direofoffset = 0;
661 		NFSUNLOCKNODE(np);
662 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
663 		if (error) {
664 			if (NFS_ISV4(vp))
665 				(void) nfsrpc_close(vp, 0, ap->a_td);
666 			return (error);
667 		}
668 		NFSLOCKNODE(np);
669 		np->n_mtime = vattr.va_mtime;
670 		if (NFS_ISV4(vp))
671 			np->n_change = vattr.va_filerev;
672 	} else {
673 		NFSUNLOCKNODE(np);
674 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
675 		if (error) {
676 			if (NFS_ISV4(vp))
677 				(void) nfsrpc_close(vp, 0, ap->a_td);
678 			return (error);
679 		}
680 		NFSLOCKNODE(np);
681 		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
682 		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
683 			if (vp->v_type == VDIR)
684 				np->n_direofoffset = 0;
685 			NFSUNLOCKNODE(np);
686 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
687 			if (error == EINTR || error == EIO) {
688 				if (NFS_ISV4(vp))
689 					(void) nfsrpc_close(vp, 0, ap->a_td);
690 				return (error);
691 			}
692 			NFSLOCKNODE(np);
693 			np->n_mtime = vattr.va_mtime;
694 			if (NFS_ISV4(vp))
695 				np->n_change = vattr.va_filerev;
696 		}
697 	}
698 
699 	/*
700 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
701 	 */
702 	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
703 	    (vp->v_type == VREG)) {
704 		if (np->n_directio_opens == 0) {
705 			NFSUNLOCKNODE(np);
706 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
707 			if (error) {
708 				if (NFS_ISV4(vp))
709 					(void) nfsrpc_close(vp, 0, ap->a_td);
710 				return (error);
711 			}
712 			NFSLOCKNODE(np);
713 			np->n_flag |= NNONCACHE;
714 		}
715 		np->n_directio_opens++;
716 	}
717 
718 	/* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
719 	if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
720 		np->n_flag |= NWRITEOPENED;
721 
722 	/*
723 	 * If this is an open for writing, capture a reference to the
724 	 * credentials, so they can be used by ncl_putpages(). Using
725 	 * these write credentials is preferable to the credentials of
726 	 * whatever thread happens to be doing the VOP_PUTPAGES() since
727 	 * the write RPCs are less likely to fail with EACCES.
728 	 */
729 	if ((fmode & FWRITE) != 0) {
730 		cred = np->n_writecred;
731 		np->n_writecred = crhold(ap->a_cred);
732 	} else
733 		cred = NULL;
734 	NFSUNLOCKNODE(np);
735 
736 	if (cred != NULL)
737 		crfree(cred);
738 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
739 
740 	/*
741 	 * If the text file has been mmap'd, flush any dirty pages to the
742 	 * buffer cache and then...
743 	 * Make sure all writes are pushed to the NFS server.  If this is not
744 	 * done, the modify time of the file can change while the text
745 	 * file is being executed.  This will cause the process that is
746 	 * executing the text file to be terminated.
747 	 */
748 	if (vp->v_writecount <= -1) {
749 		if ((obj = vp->v_object) != NULL &&
750 		    vm_object_mightbedirty(obj)) {
751 			VM_OBJECT_WLOCK(obj);
752 			vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
753 			VM_OBJECT_WUNLOCK(obj);
754 		}
755 
756 		/* Now, flush the buffer cache. */
757 		ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
758 
759 		/* And, finally, make sure that n_mtime is up to date. */
760 		np = VTONFS(vp);
761 		NFSLOCKNODE(np);
762 		np->n_mtime = np->n_vattr.na_mtime;
763 		NFSUNLOCKNODE(np);
764 	}
765 	return (0);
766 }
767 
768 /*
769  * nfs close vnode op
770  * What an NFS client should do upon close after writing is a debatable issue.
771  * Most NFS clients push delayed writes to the server upon close, basically for
772  * two reasons:
773  * 1 - So that any write errors may be reported back to the client process
774  *     doing the close system call. By far the two most likely errors are
775  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
776  * 2 - To put a worst case upper bound on cache inconsistency between
777  *     multiple clients for the file.
778  * There is also a consistency problem for Version 2 of the protocol w.r.t.
779  * not being able to tell if other clients are writing a file concurrently,
780  * since there is no way of knowing if the changed modify time in the reply
781  * is only due to the write for this client.
782  * (NFS Version 3 provides weak cache consistency data in the reply that
783  *  should be sufficient to detect and handle this case.)
784  *
785  * The current code does the following:
786  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
787  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
788  *                     or commit them (this satisfies 1 and 2 except for the
789  *                     case where the server crashes after this close but
790  *                     before the commit RPC, which is felt to be "good
791  *                     enough". Changing the last argument to ncl_flush() to
792  *                     a 1 would force a commit operation, if it is felt a
793  *                     commit is necessary now.
794  * for NFS Version 4 - flush the dirty buffers and commit them, if
795  *		       nfscl_mustflush() says this is necessary.
796  *                     It is necessary if there is no write delegation held,
797  *                     in order to satisfy open/close coherency.
798  *                     If the file isn't cached on local stable storage,
799  *                     it may be necessary in order to detect "out of space"
800  *                     errors from the server, if the write delegation
801  *                     issued by the server doesn't allow the file to grow.
802  */
803 /* ARGSUSED */
804 static int
805 nfs_close(struct vop_close_args *ap)
806 {
807 	struct vnode *vp = ap->a_vp;
808 	struct nfsnode *np = VTONFS(vp);
809 	struct nfsvattr nfsva;
810 	struct ucred *cred;
811 	int error = 0, ret, localcred = 0;
812 	int fmode = ap->a_fflag;
813 
814 	if (NFSCL_FORCEDISM(vp->v_mount))
815 		return (0);
816 	/*
817 	 * During shutdown, a_cred isn't valid, so just use root.
818 	 */
819 	if (ap->a_cred == NOCRED) {
820 		cred = newnfs_getcred();
821 		localcred = 1;
822 	} else {
823 		cred = ap->a_cred;
824 	}
825 	if (vp->v_type == VREG) {
826 	    /*
827 	     * Examine and clean dirty pages, regardless of NMODIFIED.
828 	     * This closes a major hole in close-to-open consistency.
829 	     * We want to push out all dirty pages (and buffers) on
830 	     * close, regardless of whether they were dirtied by
831 	     * mmap'ed writes or via write().
832 	     */
833 	    if (nfs_clean_pages_on_close && vp->v_object) {
834 		VM_OBJECT_WLOCK(vp->v_object);
835 		vm_object_page_clean(vp->v_object, 0, 0, 0);
836 		VM_OBJECT_WUNLOCK(vp->v_object);
837 	    }
838 	    NFSLOCKNODE(np);
839 	    if (np->n_flag & NMODIFIED) {
840 		NFSUNLOCKNODE(np);
841 		if (NFS_ISV3(vp)) {
842 		    /*
843 		     * Under NFSv3 we have dirty buffers to dispose of.  We
844 		     * must flush them to the NFS server.  We have the option
845 		     * of waiting all the way through the commit rpc or just
846 		     * waiting for the initial write.  The default is to only
847 		     * wait through the initial write so the data is in the
848 		     * server's cache, which is roughly similar to the state
849 		     * a standard disk subsystem leaves the file in on close().
850 		     *
851 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
852 		     * potential races with other processes, and certainly
853 		     * cannot clear it if we don't commit.
854 		     * These races occur when there is no longer the old
855 		     * traditional vnode locking implemented for Vnode Ops.
856 		     */
857 		    int cm = newnfs_commit_on_close ? 1 : 0;
858 		    error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
859 		    /* np->n_flag &= ~NMODIFIED; */
860 		} else if (NFS_ISV4(vp)) {
861 			if (nfscl_mustflush(vp) != 0) {
862 				int cm = newnfs_commit_on_close ? 1 : 0;
863 				error = ncl_flush(vp, MNT_WAIT, ap->a_td,
864 				    cm, 0);
865 				/*
866 				 * as above w.r.t races when clearing
867 				 * NMODIFIED.
868 				 * np->n_flag &= ~NMODIFIED;
869 				 */
870 			}
871 		} else {
872 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
873 		}
874 		NFSLOCKNODE(np);
875 	    }
876  	    /*
877  	     * Invalidate the attribute cache in all cases.
878  	     * An open is going to fetch fresh attrs any way, other procs
879  	     * on this node that have file open will be forced to do an
880  	     * otw attr fetch, but this is safe.
881 	     * --> A user found that their RPC count dropped by 20% when
882 	     *     this was commented out and I can't see any requirement
883 	     *     for it, so I've disabled it when negative lookups are
884 	     *     enabled. (What does this have to do with negative lookup
885 	     *     caching? Well nothing, except it was reported by the
886 	     *     same user that needed negative lookup caching and I wanted
887 	     *     there to be a way to disable it to see if it
888 	     *     is the cause of some caching/coherency issue that might
889 	     *     crop up.)
890  	     */
891 	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
892 		    np->n_attrstamp = 0;
893 		    KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
894 	    }
895 	    if (np->n_flag & NWRITEERR) {
896 		np->n_flag &= ~NWRITEERR;
897 		error = np->n_error;
898 	    }
899 	    NFSUNLOCKNODE(np);
900 	}
901 
902 	if (NFS_ISV4(vp)) {
903 		/*
904 		 * Get attributes so "change" is up to date.
905 		 */
906 		if (error == 0 && nfscl_mustflush(vp) != 0 &&
907 		    vp->v_type == VREG &&
908 		    (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
909 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
910 			    NULL);
911 			if (!ret) {
912 				np->n_change = nfsva.na_filerev;
913 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
914 				    NULL, 0, 0);
915 			}
916 		}
917 
918 		/*
919 		 * and do the close.
920 		 */
921 		ret = nfsrpc_close(vp, 0, ap->a_td);
922 		if (!error && ret)
923 			error = ret;
924 		if (error)
925 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
926 			    (gid_t)0);
927 	}
928 	if (newnfs_directio_enable)
929 		KASSERT((np->n_directio_asyncwr == 0),
930 			("nfs_close: dirty unflushed (%d) directio buffers\n",
931 			 np->n_directio_asyncwr));
932 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
933 		NFSLOCKNODE(np);
934 		KASSERT((np->n_directio_opens > 0),
935 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
936 		np->n_directio_opens--;
937 		if (np->n_directio_opens == 0)
938 			np->n_flag &= ~NNONCACHE;
939 		NFSUNLOCKNODE(np);
940 	}
941 	if (localcred)
942 		NFSFREECRED(cred);
943 	return (error);
944 }
945 
946 /*
947  * nfs getattr call from vfs.
948  */
949 static int
950 nfs_getattr(struct vop_getattr_args *ap)
951 {
952 	struct vnode *vp = ap->a_vp;
953 	struct thread *td = curthread;	/* XXX */
954 	struct nfsnode *np = VTONFS(vp);
955 	int error = 0;
956 	struct nfsvattr nfsva;
957 	struct vattr *vap = ap->a_vap;
958 	struct vattr vattr;
959 
960 	/*
961 	 * Update local times for special files.
962 	 */
963 	NFSLOCKNODE(np);
964 	if (np->n_flag & (NACC | NUPD))
965 		np->n_flag |= NCHG;
966 	NFSUNLOCKNODE(np);
967 	/*
968 	 * First look in the cache.
969 	 */
970 	if (ncl_getattrcache(vp, &vattr) == 0) {
971 		ncl_copy_vattr(vap, &vattr);
972 
973 		/*
974 		 * Get the local modify time for the case of a write
975 		 * delegation.
976 		 */
977 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
978 		return (0);
979 	}
980 
981 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
982 	    nfsaccess_cache_timeout > 0) {
983 		NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
984 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
985 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
986 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
987 			return (0);
988 		}
989 	}
990 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
991 	if (!error)
992 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
993 	if (!error) {
994 		/*
995 		 * Get the local modify time for the case of a write
996 		 * delegation.
997 		 */
998 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
999 	} else if (NFS_ISV4(vp)) {
1000 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1001 	}
1002 	return (error);
1003 }
1004 
1005 /*
1006  * nfs setattr call.
1007  */
1008 static int
1009 nfs_setattr(struct vop_setattr_args *ap)
1010 {
1011 	struct vnode *vp = ap->a_vp;
1012 	struct nfsnode *np = VTONFS(vp);
1013 	struct thread *td = curthread;	/* XXX */
1014 	struct vattr *vap = ap->a_vap;
1015 	int error = 0;
1016 	u_quad_t tsize;
1017 
1018 #ifndef nolint
1019 	tsize = (u_quad_t)0;
1020 #endif
1021 
1022 	/*
1023 	 * Setting of flags and marking of atimes are not supported.
1024 	 */
1025 	if (vap->va_flags != VNOVAL)
1026 		return (EOPNOTSUPP);
1027 
1028 	/*
1029 	 * Disallow write attempts if the filesystem is mounted read-only.
1030 	 */
1031   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
1032 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
1033 	    vap->va_mtime.tv_sec != VNOVAL ||
1034 	    vap->va_birthtime.tv_sec != VNOVAL ||
1035 	    vap->va_mode != (mode_t)VNOVAL) &&
1036 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
1037 		return (EROFS);
1038 	if (vap->va_size != VNOVAL) {
1039  		switch (vp->v_type) {
1040  		case VDIR:
1041  			return (EISDIR);
1042  		case VCHR:
1043  		case VBLK:
1044  		case VSOCK:
1045  		case VFIFO:
1046 			if (vap->va_mtime.tv_sec == VNOVAL &&
1047 			    vap->va_atime.tv_sec == VNOVAL &&
1048 			    vap->va_birthtime.tv_sec == VNOVAL &&
1049 			    vap->va_mode == (mode_t)VNOVAL &&
1050 			    vap->va_uid == (uid_t)VNOVAL &&
1051 			    vap->va_gid == (gid_t)VNOVAL)
1052 				return (0);
1053  			vap->va_size = VNOVAL;
1054  			break;
1055  		default:
1056 			/*
1057 			 * Disallow write attempts if the filesystem is
1058 			 * mounted read-only.
1059 			 */
1060 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
1061 				return (EROFS);
1062 			/*
1063 			 *  We run vnode_pager_setsize() early (why?),
1064 			 * we must set np->n_size now to avoid vinvalbuf
1065 			 * V_SAVE races that might setsize a lower
1066 			 * value.
1067 			 */
1068 			NFSLOCKNODE(np);
1069 			tsize = np->n_size;
1070 			NFSUNLOCKNODE(np);
1071 			error = ncl_meta_setsize(vp, td, vap->va_size);
1072 			NFSLOCKNODE(np);
1073  			if (np->n_flag & NMODIFIED) {
1074 			    tsize = np->n_size;
1075 			    NFSUNLOCKNODE(np);
1076 			    error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
1077 			        0 : V_SAVE, td, 1);
1078 			    if (error != 0) {
1079 				    vnode_pager_setsize(vp, tsize);
1080 				    return (error);
1081 			    }
1082 			    /*
1083 			     * Call nfscl_delegmodtime() to set the modify time
1084 			     * locally, as required.
1085 			     */
1086 			    nfscl_delegmodtime(vp);
1087  			} else
1088 			    NFSUNLOCKNODE(np);
1089 			/*
1090 			 * np->n_size has already been set to vap->va_size
1091 			 * in ncl_meta_setsize(). We must set it again since
1092 			 * nfs_loadattrcache() could be called through
1093 			 * ncl_meta_setsize() and could modify np->n_size.
1094 			 */
1095 			NFSLOCKNODE(np);
1096  			np->n_vattr.na_size = np->n_size = vap->va_size;
1097 			NFSUNLOCKNODE(np);
1098   		}
1099   	} else {
1100 		NFSLOCKNODE(np);
1101 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
1102 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
1103 			NFSUNLOCKNODE(np);
1104 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
1105 			if (error == EINTR || error == EIO)
1106 				return (error);
1107 		} else
1108 			NFSUNLOCKNODE(np);
1109 	}
1110 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
1111 	if (error && vap->va_size != VNOVAL) {
1112 		NFSLOCKNODE(np);
1113 		np->n_size = np->n_vattr.na_size = tsize;
1114 		vnode_pager_setsize(vp, tsize);
1115 		NFSUNLOCKNODE(np);
1116 	}
1117 	return (error);
1118 }
1119 
1120 /*
1121  * Do an nfs setattr rpc.
1122  */
1123 static int
1124 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1125     struct thread *td)
1126 {
1127 	struct nfsnode *np = VTONFS(vp);
1128 	int error, ret, attrflag, i;
1129 	struct nfsvattr nfsva;
1130 
1131 	if (NFS_ISV34(vp)) {
1132 		NFSLOCKNODE(np);
1133 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1134 			np->n_accesscache[i].stamp = 0;
1135 		np->n_flag |= NDELEGMOD;
1136 		NFSUNLOCKNODE(np);
1137 		KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1138 	}
1139 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1140 	    NULL);
1141 	if (attrflag) {
1142 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1143 		if (ret && !error)
1144 			error = ret;
1145 	}
1146 	if (error && NFS_ISV4(vp))
1147 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1148 	return (error);
1149 }
1150 
1151 /*
1152  * nfs lookup call, one step at a time...
1153  * First look in cache
1154  * If not found, unlock the directory nfsnode and do the rpc
1155  */
1156 static int
1157 nfs_lookup(struct vop_lookup_args *ap)
1158 {
1159 	struct componentname *cnp = ap->a_cnp;
1160 	struct vnode *dvp = ap->a_dvp;
1161 	struct vnode **vpp = ap->a_vpp;
1162 	struct mount *mp = dvp->v_mount;
1163 	int flags = cnp->cn_flags;
1164 	struct vnode *newvp;
1165 	struct nfsmount *nmp;
1166 	struct nfsnode *np, *newnp;
1167 	int error = 0, attrflag, dattrflag, ltype, ncticks;
1168 	struct thread *td = curthread;
1169 	struct nfsfh *nfhp;
1170 	struct nfsvattr dnfsva, nfsva;
1171 	struct vattr vattr;
1172 	struct timespec nctime;
1173 	uint32_t openmode;
1174 
1175 	*vpp = NULLVP;
1176 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1177 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1178 		return (EROFS);
1179 	if (dvp->v_type != VDIR)
1180 		return (ENOTDIR);
1181 	nmp = VFSTONFS(mp);
1182 	np = VTONFS(dvp);
1183 
1184 	/* For NFSv4, wait until any remove is done. */
1185 	NFSLOCKNODE(np);
1186 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1187 		np->n_flag |= NREMOVEWANT;
1188 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1189 	}
1190 	NFSUNLOCKNODE(np);
1191 
1192 	error = vn_dir_check_exec(dvp, cnp);
1193 	if (error != 0)
1194 		return (error);
1195 	error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1196 	if (error > 0 && error != ENOENT)
1197 		return (error);
1198 	if (error == -1) {
1199 		/*
1200 		 * Lookups of "." are special and always return the
1201 		 * current directory.  cache_lookup() already handles
1202 		 * associated locking bookkeeping, etc.
1203 		 */
1204 		if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1205 			/* XXX: Is this really correct? */
1206 			if (cnp->cn_nameiop != LOOKUP &&
1207 			    (flags & ISLASTCN))
1208 				cnp->cn_flags |= SAVENAME;
1209 			return (0);
1210 		}
1211 
1212 		/*
1213 		 * We only accept a positive hit in the cache if the
1214 		 * change time of the file matches our cached copy.
1215 		 * Otherwise, we discard the cache entry and fallback
1216 		 * to doing a lookup RPC.  We also only trust cache
1217 		 * entries for less than nm_nametimeo seconds.
1218 		 *
1219 		 * To better handle stale file handles and attributes,
1220 		 * clear the attribute cache of this node if it is a
1221 		 * leaf component, part of an open() call, and not
1222 		 * locally modified before fetching the attributes.
1223 		 * This should allow stale file handles to be detected
1224 		 * here where we can fall back to a LOOKUP RPC to
1225 		 * recover rather than having nfs_open() detect the
1226 		 * stale file handle and failing open(2) with ESTALE.
1227 		 */
1228 		newvp = *vpp;
1229 		newnp = VTONFS(newvp);
1230 		if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1231 		    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1232 		    !(newnp->n_flag & NMODIFIED)) {
1233 			NFSLOCKNODE(newnp);
1234 			newnp->n_attrstamp = 0;
1235 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1236 			NFSUNLOCKNODE(newnp);
1237 		}
1238 		if (nfscl_nodeleg(newvp, 0) == 0 ||
1239 		    ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1240 		    VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1241 		    timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1242 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1243 			if (cnp->cn_nameiop != LOOKUP &&
1244 			    (flags & ISLASTCN))
1245 				cnp->cn_flags |= SAVENAME;
1246 			return (0);
1247 		}
1248 		cache_purge(newvp);
1249 		if (dvp != newvp)
1250 			vput(newvp);
1251 		else
1252 			vrele(newvp);
1253 		*vpp = NULLVP;
1254 	} else if (error == ENOENT) {
1255 		if (VN_IS_DOOMED(dvp))
1256 			return (ENOENT);
1257 		/*
1258 		 * We only accept a negative hit in the cache if the
1259 		 * modification time of the parent directory matches
1260 		 * the cached copy in the name cache entry.
1261 		 * Otherwise, we discard all of the negative cache
1262 		 * entries for this directory.  We also only trust
1263 		 * negative cache entries for up to nm_negnametimeo
1264 		 * seconds.
1265 		 */
1266 		if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1267 		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1268 		    timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1269 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1270 			return (ENOENT);
1271 		}
1272 		cache_purge_negative(dvp);
1273 	}
1274 
1275 	/*
1276 	 * If this an NFSv4.1/4.2 mount using the "oneopenown" mount
1277 	 * option, it is possible to do the Open operation in the same
1278 	 * compound as Lookup, so long as delegations are not being
1279 	 * issued.  This saves doing a separate RPC for Open.
1280 	 */
1281 	openmode = 0;
1282 	NFSLOCKMNT(nmp);
1283 	if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) &&
1284 	    (nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 &&
1285 	    (!NFSMNT_RDONLY(mp) || (flags & OPENWRITE) == 0) &&
1286 	    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN)) {
1287 		if ((flags & OPENREAD) != 0)
1288 			openmode |= NFSV4OPEN_ACCESSREAD;
1289 		if ((flags & OPENWRITE) != 0)
1290 			openmode |= NFSV4OPEN_ACCESSWRITE;
1291 	}
1292 	NFSUNLOCKMNT(nmp);
1293 
1294 	newvp = NULLVP;
1295 	NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1296 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1297 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1298 	    NULL, openmode);
1299 	if (dattrflag)
1300 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1301 	if (error) {
1302 		if (newvp != NULLVP) {
1303 			vput(newvp);
1304 			*vpp = NULLVP;
1305 		}
1306 
1307 		if (error != ENOENT) {
1308 			if (NFS_ISV4(dvp))
1309 				error = nfscl_maperr(td, error, (uid_t)0,
1310 				    (gid_t)0);
1311 			return (error);
1312 		}
1313 
1314 		/* The requested file was not found. */
1315 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1316 		    (flags & ISLASTCN)) {
1317 			/*
1318 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1319 			 * VWRITE) here instead of just checking
1320 			 * MNT_RDONLY.
1321 			 */
1322 			if (mp->mnt_flag & MNT_RDONLY)
1323 				return (EROFS);
1324 			cnp->cn_flags |= SAVENAME;
1325 			return (EJUSTRETURN);
1326 		}
1327 
1328 		if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1329 			/*
1330 			 * Cache the modification time of the parent
1331 			 * directory from the post-op attributes in
1332 			 * the name cache entry.  The negative cache
1333 			 * entry will be ignored once the directory
1334 			 * has changed.  Don't bother adding the entry
1335 			 * if the directory has already changed.
1336 			 */
1337 			NFSLOCKNODE(np);
1338 			if (timespeccmp(&np->n_vattr.na_mtime,
1339 			    &dnfsva.na_mtime, ==)) {
1340 				NFSUNLOCKNODE(np);
1341 				cache_enter_time(dvp, NULL, cnp,
1342 				    &dnfsva.na_mtime, NULL);
1343 			} else
1344 				NFSUNLOCKNODE(np);
1345 		}
1346 		return (ENOENT);
1347 	}
1348 
1349 	/*
1350 	 * Handle RENAME case...
1351 	 */
1352 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1353 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1354 			free(nfhp, M_NFSFH);
1355 			return (EISDIR);
1356 		}
1357 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1358 		    LK_EXCLUSIVE);
1359 		if (error)
1360 			return (error);
1361 		newvp = NFSTOV(np);
1362 		if (attrflag)
1363 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1364 			    0, 1);
1365 		*vpp = newvp;
1366 		cnp->cn_flags |= SAVENAME;
1367 		return (0);
1368 	}
1369 
1370 	if (flags & ISDOTDOT) {
1371 		ltype = NFSVOPISLOCKED(dvp);
1372 		error = vfs_busy(mp, MBF_NOWAIT);
1373 		if (error != 0) {
1374 			vfs_ref(mp);
1375 			NFSVOPUNLOCK(dvp);
1376 			error = vfs_busy(mp, 0);
1377 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1378 			vfs_rel(mp);
1379 			if (error == 0 && VN_IS_DOOMED(dvp)) {
1380 				vfs_unbusy(mp);
1381 				error = ENOENT;
1382 			}
1383 			if (error != 0)
1384 				return (error);
1385 		}
1386 		NFSVOPUNLOCK(dvp);
1387 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1388 		    cnp->cn_lkflags);
1389 		if (error == 0)
1390 			newvp = NFSTOV(np);
1391 		vfs_unbusy(mp);
1392 		if (newvp != dvp)
1393 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1394 		if (VN_IS_DOOMED(dvp)) {
1395 			if (error == 0) {
1396 				if (newvp == dvp)
1397 					vrele(newvp);
1398 				else
1399 					vput(newvp);
1400 			}
1401 			error = ENOENT;
1402 		}
1403 		if (error != 0)
1404 			return (error);
1405 		if (attrflag)
1406 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1407 			    0, 1);
1408 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1409 		free(nfhp, M_NFSFH);
1410 		VREF(dvp);
1411 		newvp = dvp;
1412 		if (attrflag)
1413 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1414 			    0, 1);
1415 	} else {
1416 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1417 		    cnp->cn_lkflags);
1418 		if (error)
1419 			return (error);
1420 		newvp = NFSTOV(np);
1421 		if (attrflag)
1422 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1423 			    0, 1);
1424 		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1425 		    !(np->n_flag & NMODIFIED)) {
1426 			/*
1427 			 * Flush the attribute cache when opening a
1428 			 * leaf node to ensure that fresh attributes
1429 			 * are fetched in nfs_open() since we did not
1430 			 * fetch attributes from the LOOKUP reply.
1431 			 */
1432 			NFSLOCKNODE(np);
1433 			np->n_attrstamp = 0;
1434 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1435 			NFSUNLOCKNODE(np);
1436 		}
1437 	}
1438 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1439 		cnp->cn_flags |= SAVENAME;
1440 	if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp &&
1441 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1442 	    attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1443 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1444 		    newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1445 	*vpp = newvp;
1446 	return (0);
1447 }
1448 
1449 /*
1450  * nfs read call.
1451  * Just call ncl_bioread() to do the work.
1452  */
1453 static int
1454 nfs_read(struct vop_read_args *ap)
1455 {
1456 	struct vnode *vp = ap->a_vp;
1457 
1458 	switch (vp->v_type) {
1459 	case VREG:
1460 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1461 	case VDIR:
1462 		return (EISDIR);
1463 	default:
1464 		return (EOPNOTSUPP);
1465 	}
1466 }
1467 
1468 /*
1469  * nfs readlink call
1470  */
1471 static int
1472 nfs_readlink(struct vop_readlink_args *ap)
1473 {
1474 	struct vnode *vp = ap->a_vp;
1475 
1476 	if (vp->v_type != VLNK)
1477 		return (EINVAL);
1478 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1479 }
1480 
1481 /*
1482  * Do a readlink rpc.
1483  * Called by ncl_doio() from below the buffer cache.
1484  */
1485 int
1486 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1487 {
1488 	int error, ret, attrflag;
1489 	struct nfsvattr nfsva;
1490 
1491 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1492 	    &attrflag, NULL);
1493 	if (attrflag) {
1494 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1495 		if (ret && !error)
1496 			error = ret;
1497 	}
1498 	if (error && NFS_ISV4(vp))
1499 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1500 	return (error);
1501 }
1502 
1503 /*
1504  * nfs read rpc call
1505  * Ditto above
1506  */
1507 int
1508 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1509 {
1510 	int error, ret, attrflag;
1511 	struct nfsvattr nfsva;
1512 	struct nfsmount *nmp;
1513 
1514 	nmp = VFSTONFS(vp->v_mount);
1515 	error = EIO;
1516 	attrflag = 0;
1517 	if (NFSHASPNFS(nmp))
1518 		error = nfscl_doiods(vp, uiop, NULL, NULL,
1519 		    NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1520 	NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1521 	if (error != 0)
1522 		error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1523 		    &attrflag, NULL);
1524 	if (attrflag) {
1525 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1526 		if (ret && !error)
1527 			error = ret;
1528 	}
1529 	if (error && NFS_ISV4(vp))
1530 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1531 	return (error);
1532 }
1533 
1534 /*
1535  * nfs write call
1536  */
1537 int
1538 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1539     int *iomode, int *must_commit, int called_from_strategy)
1540 {
1541 	struct nfsvattr nfsva;
1542 	int error, attrflag, ret;
1543 	struct nfsmount *nmp;
1544 
1545 	nmp = VFSTONFS(vp->v_mount);
1546 	error = EIO;
1547 	attrflag = 0;
1548 	if (NFSHASPNFS(nmp))
1549 		error = nfscl_doiods(vp, uiop, iomode, must_commit,
1550 		    NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1551 	NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1552 	if (error != 0)
1553 		error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1554 		    uiop->uio_td, &nfsva, &attrflag, NULL,
1555 		    called_from_strategy);
1556 	if (attrflag) {
1557 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1558 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1559 			    1);
1560 		else
1561 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1562 			    1);
1563 		if (ret && !error)
1564 			error = ret;
1565 	}
1566 	if (DOINGASYNC(vp))
1567 		*iomode = NFSWRITE_FILESYNC;
1568 	if (error && NFS_ISV4(vp))
1569 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1570 	return (error);
1571 }
1572 
1573 /*
1574  * nfs mknod rpc
1575  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1576  * mode set to specify the file type and the size field for rdev.
1577  */
1578 static int
1579 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1580     struct vattr *vap)
1581 {
1582 	struct nfsvattr nfsva, dnfsva;
1583 	struct vnode *newvp = NULL;
1584 	struct nfsnode *np = NULL, *dnp;
1585 	struct nfsfh *nfhp;
1586 	struct vattr vattr;
1587 	int error = 0, attrflag, dattrflag;
1588 	u_int32_t rdev;
1589 
1590 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1591 		rdev = vap->va_rdev;
1592 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1593 		rdev = 0xffffffff;
1594 	else
1595 		return (EOPNOTSUPP);
1596 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1597 		return (error);
1598 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1599 	    rdev, vap->va_type, cnp->cn_cred, curthread, &dnfsva,
1600 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1601 	if (!error) {
1602 		if (!nfhp)
1603 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1604 			    cnp->cn_namelen, cnp->cn_cred, curthread,
1605 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1606 			    NULL, 0);
1607 		if (nfhp)
1608 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1609 			    curthread, &np, NULL, LK_EXCLUSIVE);
1610 	}
1611 	if (dattrflag)
1612 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1613 	if (!error) {
1614 		newvp = NFSTOV(np);
1615 		if (attrflag != 0) {
1616 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1617 			    0, 1);
1618 			if (error != 0)
1619 				vput(newvp);
1620 		}
1621 	}
1622 	if (!error) {
1623 		*vpp = newvp;
1624 	} else if (NFS_ISV4(dvp)) {
1625 		error = nfscl_maperr(curthread, error, vap->va_uid,
1626 		    vap->va_gid);
1627 	}
1628 	dnp = VTONFS(dvp);
1629 	NFSLOCKNODE(dnp);
1630 	dnp->n_flag |= NMODIFIED;
1631 	if (!dattrflag) {
1632 		dnp->n_attrstamp = 0;
1633 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1634 	}
1635 	NFSUNLOCKNODE(dnp);
1636 	return (error);
1637 }
1638 
1639 /*
1640  * nfs mknod vop
1641  * just call nfs_mknodrpc() to do the work.
1642  */
1643 /* ARGSUSED */
1644 static int
1645 nfs_mknod(struct vop_mknod_args *ap)
1646 {
1647 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1648 }
1649 
1650 static struct mtx nfs_cverf_mtx;
1651 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1652     MTX_DEF);
1653 
1654 static nfsquad_t
1655 nfs_get_cverf(void)
1656 {
1657 	static nfsquad_t cverf;
1658 	nfsquad_t ret;
1659 	static int cverf_initialized = 0;
1660 
1661 	mtx_lock(&nfs_cverf_mtx);
1662 	if (cverf_initialized == 0) {
1663 		cverf.lval[0] = arc4random();
1664 		cverf.lval[1] = arc4random();
1665 		cverf_initialized = 1;
1666 	} else
1667 		cverf.qval++;
1668 	ret = cverf;
1669 	mtx_unlock(&nfs_cverf_mtx);
1670 
1671 	return (ret);
1672 }
1673 
1674 /*
1675  * nfs file create call
1676  */
1677 static int
1678 nfs_create(struct vop_create_args *ap)
1679 {
1680 	struct vnode *dvp = ap->a_dvp;
1681 	struct vattr *vap = ap->a_vap;
1682 	struct componentname *cnp = ap->a_cnp;
1683 	struct nfsnode *np = NULL, *dnp;
1684 	struct vnode *newvp = NULL;
1685 	struct nfsmount *nmp;
1686 	struct nfsvattr dnfsva, nfsva;
1687 	struct nfsfh *nfhp;
1688 	nfsquad_t cverf;
1689 	int error = 0, attrflag, dattrflag, fmode = 0;
1690 	struct vattr vattr;
1691 
1692 	/*
1693 	 * Oops, not for me..
1694 	 */
1695 	if (vap->va_type == VSOCK)
1696 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1697 
1698 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1699 		return (error);
1700 	if (vap->va_vaflags & VA_EXCLUSIVE)
1701 		fmode |= O_EXCL;
1702 	dnp = VTONFS(dvp);
1703 	nmp = VFSTONFS(dvp->v_mount);
1704 again:
1705 	/* For NFSv4, wait until any remove is done. */
1706 	NFSLOCKNODE(dnp);
1707 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1708 		dnp->n_flag |= NREMOVEWANT;
1709 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1710 	}
1711 	NFSUNLOCKNODE(dnp);
1712 
1713 	cverf = nfs_get_cverf();
1714 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1715 	    vap, cverf, fmode, cnp->cn_cred, curthread, &dnfsva, &nfsva,
1716 	    &nfhp, &attrflag, &dattrflag, NULL);
1717 	if (!error) {
1718 		if (nfhp == NULL)
1719 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1720 			    cnp->cn_namelen, cnp->cn_cred, curthread,
1721 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1722 			    NULL, 0);
1723 		if (nfhp != NULL)
1724 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1725 			    curthread, &np, NULL, LK_EXCLUSIVE);
1726 	}
1727 	if (dattrflag)
1728 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1729 	if (!error) {
1730 		newvp = NFSTOV(np);
1731 		if (attrflag == 0)
1732 			error = nfsrpc_getattr(newvp, cnp->cn_cred,
1733 			    curthread, &nfsva, NULL);
1734 		if (error == 0)
1735 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1736 			    0, 1);
1737 	}
1738 	if (error) {
1739 		if (newvp != NULL) {
1740 			vput(newvp);
1741 			newvp = NULL;
1742 		}
1743 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1744 		    error == NFSERR_NOTSUPP) {
1745 			fmode &= ~O_EXCL;
1746 			goto again;
1747 		}
1748 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1749 		if (nfscl_checksattr(vap, &nfsva)) {
1750 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1751 			    curthread, &nfsva, &attrflag, NULL);
1752 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1753 			    vap->va_gid != (gid_t)VNOVAL)) {
1754 				/* try again without setting uid/gid */
1755 				vap->va_uid = (uid_t)VNOVAL;
1756 				vap->va_gid = (uid_t)VNOVAL;
1757 				error = nfsrpc_setattr(newvp, vap, NULL,
1758 				    cnp->cn_cred, curthread, &nfsva,
1759 				    &attrflag, NULL);
1760 			}
1761 			if (attrflag)
1762 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1763 				    NULL, 0, 1);
1764 			if (error != 0)
1765 				vput(newvp);
1766 		}
1767 	}
1768 	if (!error) {
1769 		if ((cnp->cn_flags & MAKEENTRY) && attrflag) {
1770 			if (dvp != newvp)
1771 				cache_enter_time(dvp, newvp, cnp,
1772 				    &nfsva.na_ctime, NULL);
1773 			else
1774 				printf("nfs_create: bogus NFS server returned "
1775 				    "the directory as the new file object\n");
1776 		}
1777 		*ap->a_vpp = newvp;
1778 	} else if (NFS_ISV4(dvp)) {
1779 		error = nfscl_maperr(curthread, error, vap->va_uid,
1780 		    vap->va_gid);
1781 	}
1782 	NFSLOCKNODE(dnp);
1783 	dnp->n_flag |= NMODIFIED;
1784 	if (!dattrflag) {
1785 		dnp->n_attrstamp = 0;
1786 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1787 	}
1788 	NFSUNLOCKNODE(dnp);
1789 	return (error);
1790 }
1791 
1792 /*
1793  * nfs file remove call
1794  * To try and make nfs semantics closer to ufs semantics, a file that has
1795  * other processes using the vnode is renamed instead of removed and then
1796  * removed later on the last close.
1797  * - If v_usecount > 1
1798  *	  If a rename is not already in the works
1799  *	     call nfs_sillyrename() to set it up
1800  *     else
1801  *	  do the remove rpc
1802  */
1803 static int
1804 nfs_remove(struct vop_remove_args *ap)
1805 {
1806 	struct vnode *vp = ap->a_vp;
1807 	struct vnode *dvp = ap->a_dvp;
1808 	struct componentname *cnp = ap->a_cnp;
1809 	struct nfsnode *np = VTONFS(vp);
1810 	int error = 0;
1811 	struct vattr vattr;
1812 
1813 	KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1814 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1815 	if (vp->v_type == VDIR)
1816 		error = EPERM;
1817 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1818 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1819 	    vattr.va_nlink > 1)) {
1820 		/*
1821 		 * Purge the name cache so that the chance of a lookup for
1822 		 * the name succeeding while the remove is in progress is
1823 		 * minimized. Without node locking it can still happen, such
1824 		 * that an I/O op returns ESTALE, but since you get this if
1825 		 * another host removes the file..
1826 		 */
1827 		cache_purge(vp);
1828 		/*
1829 		 * throw away biocache buffers, mainly to avoid
1830 		 * unnecessary delayed writes later.
1831 		 */
1832 		error = ncl_vinvalbuf(vp, 0, curthread, 1);
1833 		if (error != EINTR && error != EIO)
1834 			/* Do the rpc */
1835 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1836 			    cnp->cn_namelen, cnp->cn_cred, curthread);
1837 		/*
1838 		 * Kludge City: If the first reply to the remove rpc is lost..
1839 		 *   the reply to the retransmitted request will be ENOENT
1840 		 *   since the file was in fact removed
1841 		 *   Therefore, we cheat and return success.
1842 		 */
1843 		if (error == ENOENT)
1844 			error = 0;
1845 	} else if (!np->n_sillyrename)
1846 		error = nfs_sillyrename(dvp, vp, cnp);
1847 	NFSLOCKNODE(np);
1848 	np->n_attrstamp = 0;
1849 	NFSUNLOCKNODE(np);
1850 	KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1851 	return (error);
1852 }
1853 
1854 /*
1855  * nfs file remove rpc called from nfs_inactive
1856  */
1857 int
1858 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1859 {
1860 	/*
1861 	 * Make sure that the directory vnode is still valid.
1862 	 * XXX we should lock sp->s_dvp here.
1863 	 */
1864 	if (sp->s_dvp->v_type == VBAD)
1865 		return (0);
1866 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1867 	    sp->s_cred, NULL));
1868 }
1869 
1870 /*
1871  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1872  */
1873 static int
1874 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1875     int namelen, struct ucred *cred, struct thread *td)
1876 {
1877 	struct nfsvattr dnfsva;
1878 	struct nfsnode *dnp = VTONFS(dvp);
1879 	int error = 0, dattrflag;
1880 
1881 	NFSLOCKNODE(dnp);
1882 	dnp->n_flag |= NREMOVEINPROG;
1883 	NFSUNLOCKNODE(dnp);
1884 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1885 	    &dattrflag, NULL);
1886 	NFSLOCKNODE(dnp);
1887 	if ((dnp->n_flag & NREMOVEWANT)) {
1888 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1889 		NFSUNLOCKNODE(dnp);
1890 		wakeup((caddr_t)dnp);
1891 	} else {
1892 		dnp->n_flag &= ~NREMOVEINPROG;
1893 		NFSUNLOCKNODE(dnp);
1894 	}
1895 	if (dattrflag)
1896 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1897 	NFSLOCKNODE(dnp);
1898 	dnp->n_flag |= NMODIFIED;
1899 	if (!dattrflag) {
1900 		dnp->n_attrstamp = 0;
1901 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1902 	}
1903 	NFSUNLOCKNODE(dnp);
1904 	if (error && NFS_ISV4(dvp))
1905 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1906 	return (error);
1907 }
1908 
1909 /*
1910  * nfs file rename call
1911  */
1912 static int
1913 nfs_rename(struct vop_rename_args *ap)
1914 {
1915 	struct vnode *fvp = ap->a_fvp;
1916 	struct vnode *tvp = ap->a_tvp;
1917 	struct vnode *fdvp = ap->a_fdvp;
1918 	struct vnode *tdvp = ap->a_tdvp;
1919 	struct componentname *tcnp = ap->a_tcnp;
1920 	struct componentname *fcnp = ap->a_fcnp;
1921 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1922 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1923 	struct nfsv4node *newv4 = NULL;
1924 	int error;
1925 
1926 	KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1927 	    (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1928 	/* Check for cross-device rename */
1929 	if ((fvp->v_mount != tdvp->v_mount) ||
1930 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1931 		error = EXDEV;
1932 		goto out;
1933 	}
1934 
1935 	if (fvp == tvp) {
1936 		printf("nfs_rename: fvp == tvp (can't happen)\n");
1937 		error = 0;
1938 		goto out;
1939 	}
1940 	if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1941 		goto out;
1942 
1943 	/*
1944 	 * We have to flush B_DELWRI data prior to renaming
1945 	 * the file.  If we don't, the delayed-write buffers
1946 	 * can be flushed out later after the file has gone stale
1947 	 * under NFSV3.  NFSV2 does not have this problem because
1948 	 * ( as far as I can tell ) it flushes dirty buffers more
1949 	 * often.
1950 	 *
1951 	 * Skip the rename operation if the fsync fails, this can happen
1952 	 * due to the server's volume being full, when we pushed out data
1953 	 * that was written back to our cache earlier. Not checking for
1954 	 * this condition can result in potential (silent) data loss.
1955 	 */
1956 	error = VOP_FSYNC(fvp, MNT_WAIT, curthread);
1957 	NFSVOPUNLOCK(fvp);
1958 	if (!error && tvp)
1959 		error = VOP_FSYNC(tvp, MNT_WAIT, curthread);
1960 	if (error)
1961 		goto out;
1962 
1963 	/*
1964 	 * If the tvp exists and is in use, sillyrename it before doing the
1965 	 * rename of the new file over it.
1966 	 * XXX Can't sillyrename a directory.
1967 	 */
1968 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1969 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1970 		vput(tvp);
1971 		tvp = NULL;
1972 	}
1973 
1974 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1975 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1976 	    curthread);
1977 
1978 	if (error == 0 && NFS_ISV4(tdvp)) {
1979 		/*
1980 		 * For NFSv4, check to see if it is the same name and
1981 		 * replace the name, if it is different.
1982 		 */
1983 		newv4 = malloc(
1984 		    sizeof (struct nfsv4node) +
1985 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1986 		    M_NFSV4NODE, M_WAITOK);
1987 		NFSLOCKNODE(tdnp);
1988 		NFSLOCKNODE(fnp);
1989 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1990 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1991 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1992 		      tcnp->cn_namelen) ||
1993 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1994 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1995 			tdnp->n_fhp->nfh_len))) {
1996 #ifdef notdef
1997 { char nnn[100]; int nnnl;
1998 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1999 bcopy(tcnp->cn_nameptr, nnn, nnnl);
2000 nnn[nnnl] = '\0';
2001 printf("ren replace=%s\n",nnn);
2002 }
2003 #endif
2004 			free(fnp->n_v4, M_NFSV4NODE);
2005 			fnp->n_v4 = newv4;
2006 			newv4 = NULL;
2007 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
2008 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
2009 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
2010 			    tdnp->n_fhp->nfh_len);
2011 			NFSBCOPY(tcnp->cn_nameptr,
2012 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
2013 		}
2014 		NFSUNLOCKNODE(tdnp);
2015 		NFSUNLOCKNODE(fnp);
2016 		if (newv4 != NULL)
2017 			free(newv4, M_NFSV4NODE);
2018 	}
2019 
2020 	if (fvp->v_type == VDIR) {
2021 		if (tvp != NULL && tvp->v_type == VDIR)
2022 			cache_purge(tdvp);
2023 		cache_purge(fdvp);
2024 	}
2025 
2026 out:
2027 	if (tdvp == tvp)
2028 		vrele(tdvp);
2029 	else
2030 		vput(tdvp);
2031 	if (tvp)
2032 		vput(tvp);
2033 	vrele(fdvp);
2034 	vrele(fvp);
2035 	/*
2036 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2037 	 */
2038 	if (error == ENOENT)
2039 		error = 0;
2040 	return (error);
2041 }
2042 
2043 /*
2044  * nfs file rename rpc called from nfs_remove() above
2045  */
2046 static int
2047 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
2048     struct sillyrename *sp)
2049 {
2050 
2051 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
2052 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
2053 	    curthread));
2054 }
2055 
2056 /*
2057  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2058  */
2059 static int
2060 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
2061     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
2062     int tnamelen, struct ucred *cred, struct thread *td)
2063 {
2064 	struct nfsvattr fnfsva, tnfsva;
2065 	struct nfsnode *fdnp = VTONFS(fdvp);
2066 	struct nfsnode *tdnp = VTONFS(tdvp);
2067 	int error = 0, fattrflag, tattrflag;
2068 
2069 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
2070 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
2071 	    &tattrflag, NULL, NULL);
2072 	NFSLOCKNODE(fdnp);
2073 	fdnp->n_flag |= NMODIFIED;
2074 	if (fattrflag != 0) {
2075 		NFSUNLOCKNODE(fdnp);
2076 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
2077 	} else {
2078 		fdnp->n_attrstamp = 0;
2079 		NFSUNLOCKNODE(fdnp);
2080 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
2081 	}
2082 	NFSLOCKNODE(tdnp);
2083 	tdnp->n_flag |= NMODIFIED;
2084 	if (tattrflag != 0) {
2085 		NFSUNLOCKNODE(tdnp);
2086 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
2087 	} else {
2088 		tdnp->n_attrstamp = 0;
2089 		NFSUNLOCKNODE(tdnp);
2090 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2091 	}
2092 	if (error && NFS_ISV4(fdvp))
2093 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2094 	return (error);
2095 }
2096 
2097 /*
2098  * nfs hard link create call
2099  */
2100 static int
2101 nfs_link(struct vop_link_args *ap)
2102 {
2103 	struct vnode *vp = ap->a_vp;
2104 	struct vnode *tdvp = ap->a_tdvp;
2105 	struct componentname *cnp = ap->a_cnp;
2106 	struct nfsnode *np, *tdnp;
2107 	struct nfsvattr nfsva, dnfsva;
2108 	int error = 0, attrflag, dattrflag;
2109 
2110 	/*
2111 	 * Push all writes to the server, so that the attribute cache
2112 	 * doesn't get "out of sync" with the server.
2113 	 * XXX There should be a better way!
2114 	 */
2115 	VOP_FSYNC(vp, MNT_WAIT, curthread);
2116 
2117 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2118 	    cnp->cn_cred, curthread, &dnfsva, &nfsva, &attrflag,
2119 	    &dattrflag, NULL);
2120 	tdnp = VTONFS(tdvp);
2121 	NFSLOCKNODE(tdnp);
2122 	tdnp->n_flag |= NMODIFIED;
2123 	if (dattrflag != 0) {
2124 		NFSUNLOCKNODE(tdnp);
2125 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
2126 	} else {
2127 		tdnp->n_attrstamp = 0;
2128 		NFSUNLOCKNODE(tdnp);
2129 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2130 	}
2131 	if (attrflag)
2132 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2133 	else {
2134 		np = VTONFS(vp);
2135 		NFSLOCKNODE(np);
2136 		np->n_attrstamp = 0;
2137 		NFSUNLOCKNODE(np);
2138 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2139 	}
2140 	/*
2141 	 * If negative lookup caching is enabled, I might as well
2142 	 * add an entry for this node. Not necessary for correctness,
2143 	 * but if negative caching is enabled, then the system
2144 	 * must care about lookup caching hit rate, so...
2145 	 */
2146 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2147 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2148 		if (tdvp != vp)
2149 			cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2150 		else
2151 			printf("nfs_link: bogus NFS server returned "
2152 			    "the directory as the new link\n");
2153 	}
2154 	if (error && NFS_ISV4(vp))
2155 		error = nfscl_maperr(curthread, error, (uid_t)0,
2156 		    (gid_t)0);
2157 	return (error);
2158 }
2159 
2160 /*
2161  * nfs symbolic link create call
2162  */
2163 static int
2164 nfs_symlink(struct vop_symlink_args *ap)
2165 {
2166 	struct vnode *dvp = ap->a_dvp;
2167 	struct vattr *vap = ap->a_vap;
2168 	struct componentname *cnp = ap->a_cnp;
2169 	struct nfsvattr nfsva, dnfsva;
2170 	struct nfsfh *nfhp;
2171 	struct nfsnode *np = NULL, *dnp;
2172 	struct vnode *newvp = NULL;
2173 	int error = 0, attrflag, dattrflag, ret;
2174 
2175 	vap->va_type = VLNK;
2176 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2177 	    ap->a_target, vap, cnp->cn_cred, curthread, &dnfsva,
2178 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2179 	if (nfhp) {
2180 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread,
2181 		    &np, NULL, LK_EXCLUSIVE);
2182 		if (!ret)
2183 			newvp = NFSTOV(np);
2184 		else if (!error)
2185 			error = ret;
2186 	}
2187 	if (newvp != NULL) {
2188 		if (attrflag)
2189 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2190 			    0, 1);
2191 	} else if (!error) {
2192 		/*
2193 		 * If we do not have an error and we could not extract the
2194 		 * newvp from the response due to the request being NFSv2, we
2195 		 * have to do a lookup in order to obtain a newvp to return.
2196 		 */
2197 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2198 		    cnp->cn_cred, curthread, &np);
2199 		if (!error)
2200 			newvp = NFSTOV(np);
2201 	}
2202 	if (error) {
2203 		if (newvp)
2204 			vput(newvp);
2205 		if (NFS_ISV4(dvp))
2206 			error = nfscl_maperr(curthread, error,
2207 			    vap->va_uid, vap->va_gid);
2208 	} else {
2209 		*ap->a_vpp = newvp;
2210 	}
2211 
2212 	dnp = VTONFS(dvp);
2213 	NFSLOCKNODE(dnp);
2214 	dnp->n_flag |= NMODIFIED;
2215 	if (dattrflag != 0) {
2216 		NFSUNLOCKNODE(dnp);
2217 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2218 	} else {
2219 		dnp->n_attrstamp = 0;
2220 		NFSUNLOCKNODE(dnp);
2221 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2222 	}
2223 	/*
2224 	 * If negative lookup caching is enabled, I might as well
2225 	 * add an entry for this node. Not necessary for correctness,
2226 	 * but if negative caching is enabled, then the system
2227 	 * must care about lookup caching hit rate, so...
2228 	 */
2229 	if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2230 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2231 		if (dvp != newvp)
2232 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2233 			    NULL);
2234 		else
2235 			printf("nfs_symlink: bogus NFS server returned "
2236 			    "the directory as the new file object\n");
2237 	}
2238 	return (error);
2239 }
2240 
2241 /*
2242  * nfs make dir call
2243  */
2244 static int
2245 nfs_mkdir(struct vop_mkdir_args *ap)
2246 {
2247 	struct vnode *dvp = ap->a_dvp;
2248 	struct vattr *vap = ap->a_vap;
2249 	struct componentname *cnp = ap->a_cnp;
2250 	struct nfsnode *np = NULL, *dnp;
2251 	struct vnode *newvp = NULL;
2252 	struct vattr vattr;
2253 	struct nfsfh *nfhp;
2254 	struct nfsvattr nfsva, dnfsva;
2255 	int error = 0, attrflag, dattrflag, ret;
2256 
2257 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2258 		return (error);
2259 	vap->va_type = VDIR;
2260 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2261 	    vap, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp,
2262 	    &attrflag, &dattrflag, NULL);
2263 	dnp = VTONFS(dvp);
2264 	NFSLOCKNODE(dnp);
2265 	dnp->n_flag |= NMODIFIED;
2266 	if (dattrflag != 0) {
2267 		NFSUNLOCKNODE(dnp);
2268 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2269 	} else {
2270 		dnp->n_attrstamp = 0;
2271 		NFSUNLOCKNODE(dnp);
2272 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2273 	}
2274 	if (nfhp) {
2275 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread,
2276 		    &np, NULL, LK_EXCLUSIVE);
2277 		if (!ret) {
2278 			newvp = NFSTOV(np);
2279 			if (attrflag)
2280 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2281 				NULL, 0, 1);
2282 		} else if (!error)
2283 			error = ret;
2284 	}
2285 	if (!error && newvp == NULL) {
2286 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2287 		    cnp->cn_cred, curthread, &np);
2288 		if (!error) {
2289 			newvp = NFSTOV(np);
2290 			if (newvp->v_type != VDIR)
2291 				error = EEXIST;
2292 		}
2293 	}
2294 	if (error) {
2295 		if (newvp)
2296 			vput(newvp);
2297 		if (NFS_ISV4(dvp))
2298 			error = nfscl_maperr(curthread, error,
2299 			    vap->va_uid, vap->va_gid);
2300 	} else {
2301 		/*
2302 		 * If negative lookup caching is enabled, I might as well
2303 		 * add an entry for this node. Not necessary for correctness,
2304 		 * but if negative caching is enabled, then the system
2305 		 * must care about lookup caching hit rate, so...
2306 		 */
2307 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2308 		    (cnp->cn_flags & MAKEENTRY) &&
2309 		    attrflag != 0 && dattrflag != 0) {
2310 			if (dvp != newvp)
2311 				cache_enter_time(dvp, newvp, cnp,
2312 				    &nfsva.na_ctime, &dnfsva.na_ctime);
2313 			else
2314 				printf("nfs_mkdir: bogus NFS server returned "
2315 				    "the directory that the directory was "
2316 				    "created in as the new file object\n");
2317 		}
2318 		*ap->a_vpp = newvp;
2319 	}
2320 	return (error);
2321 }
2322 
2323 /*
2324  * nfs remove directory call
2325  */
2326 static int
2327 nfs_rmdir(struct vop_rmdir_args *ap)
2328 {
2329 	struct vnode *vp = ap->a_vp;
2330 	struct vnode *dvp = ap->a_dvp;
2331 	struct componentname *cnp = ap->a_cnp;
2332 	struct nfsnode *dnp;
2333 	struct nfsvattr dnfsva;
2334 	int error, dattrflag;
2335 
2336 	if (dvp == vp)
2337 		return (EINVAL);
2338 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2339 	    cnp->cn_cred, curthread, &dnfsva, &dattrflag, NULL);
2340 	dnp = VTONFS(dvp);
2341 	NFSLOCKNODE(dnp);
2342 	dnp->n_flag |= NMODIFIED;
2343 	if (dattrflag != 0) {
2344 		NFSUNLOCKNODE(dnp);
2345 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2346 	} else {
2347 		dnp->n_attrstamp = 0;
2348 		NFSUNLOCKNODE(dnp);
2349 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2350 	}
2351 
2352 	cache_purge(dvp);
2353 	cache_purge(vp);
2354 	if (error && NFS_ISV4(dvp))
2355 		error = nfscl_maperr(curthread, error, (uid_t)0,
2356 		    (gid_t)0);
2357 	/*
2358 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2359 	 */
2360 	if (error == ENOENT)
2361 		error = 0;
2362 	return (error);
2363 }
2364 
2365 /*
2366  * nfs readdir call
2367  */
2368 static int
2369 nfs_readdir(struct vop_readdir_args *ap)
2370 {
2371 	struct vnode *vp = ap->a_vp;
2372 	struct nfsnode *np = VTONFS(vp);
2373 	struct uio *uio = ap->a_uio;
2374 	ssize_t tresid, left;
2375 	int error = 0;
2376 	struct vattr vattr;
2377 
2378 	if (ap->a_eofflag != NULL)
2379 		*ap->a_eofflag = 0;
2380 	if (vp->v_type != VDIR)
2381 		return(EPERM);
2382 
2383 	/*
2384 	 * First, check for hit on the EOF offset cache
2385 	 */
2386 	NFSLOCKNODE(np);
2387 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2388 	    (np->n_flag & NMODIFIED) == 0) {
2389 		NFSUNLOCKNODE(np);
2390 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2391 			NFSLOCKNODE(np);
2392 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2393 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2394 				NFSUNLOCKNODE(np);
2395 				NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2396 				if (ap->a_eofflag != NULL)
2397 					*ap->a_eofflag = 1;
2398 				return (0);
2399 			} else
2400 				NFSUNLOCKNODE(np);
2401 		}
2402 	} else
2403 		NFSUNLOCKNODE(np);
2404 
2405 	/*
2406 	 * NFS always guarantees that directory entries don't straddle
2407 	 * DIRBLKSIZ boundaries.  As such, we need to limit the size
2408 	 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2409 	 * directory entry.
2410 	 */
2411 	left = uio->uio_resid % DIRBLKSIZ;
2412 	if (left == uio->uio_resid)
2413 		return (EINVAL);
2414 	uio->uio_resid -= left;
2415 
2416 	/*
2417 	 * Call ncl_bioread() to do the real work.
2418 	 */
2419 	tresid = uio->uio_resid;
2420 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2421 
2422 	if (!error && uio->uio_resid == tresid) {
2423 		NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2424 		if (ap->a_eofflag != NULL)
2425 			*ap->a_eofflag = 1;
2426 	}
2427 
2428 	/* Add the partial DIRBLKSIZ (left) back in. */
2429 	uio->uio_resid += left;
2430 	return (error);
2431 }
2432 
2433 /*
2434  * Readdir rpc call.
2435  * Called from below the buffer cache by ncl_doio().
2436  */
2437 int
2438 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2439     struct thread *td)
2440 {
2441 	struct nfsvattr nfsva;
2442 	nfsuint64 *cookiep, cookie;
2443 	struct nfsnode *dnp = VTONFS(vp);
2444 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2445 	int error = 0, eof, attrflag;
2446 
2447 	KASSERT(uiop->uio_iovcnt == 1 &&
2448 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2449 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2450 	    ("nfs readdirrpc bad uio"));
2451 
2452 	/*
2453 	 * If there is no cookie, assume directory was stale.
2454 	 */
2455 	ncl_dircookie_lock(dnp);
2456 	NFSUNLOCKNODE(dnp);
2457 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2458 	if (cookiep) {
2459 		cookie = *cookiep;
2460 		ncl_dircookie_unlock(dnp);
2461 	} else {
2462 		ncl_dircookie_unlock(dnp);
2463 		return (NFSERR_BAD_COOKIE);
2464 	}
2465 
2466 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2467 		(void)ncl_fsinfo(nmp, vp, cred, td);
2468 
2469 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2470 	    &attrflag, &eof, NULL);
2471 	if (attrflag)
2472 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2473 
2474 	if (!error) {
2475 		/*
2476 		 * We are now either at the end of the directory or have filled
2477 		 * the block.
2478 		 */
2479 		if (eof) {
2480 			NFSLOCKNODE(dnp);
2481 			dnp->n_direofoffset = uiop->uio_offset;
2482 			NFSUNLOCKNODE(dnp);
2483 		} else {
2484 			if (uiop->uio_resid > 0)
2485 				printf("EEK! readdirrpc resid > 0\n");
2486 			ncl_dircookie_lock(dnp);
2487 			NFSUNLOCKNODE(dnp);
2488 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2489 			*cookiep = cookie;
2490 			ncl_dircookie_unlock(dnp);
2491 		}
2492 	} else if (NFS_ISV4(vp)) {
2493 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2494 	}
2495 	return (error);
2496 }
2497 
2498 /*
2499  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2500  */
2501 int
2502 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2503     struct thread *td)
2504 {
2505 	struct nfsvattr nfsva;
2506 	nfsuint64 *cookiep, cookie;
2507 	struct nfsnode *dnp = VTONFS(vp);
2508 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2509 	int error = 0, attrflag, eof;
2510 
2511 	KASSERT(uiop->uio_iovcnt == 1 &&
2512 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2513 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2514 	    ("nfs readdirplusrpc bad uio"));
2515 
2516 	/*
2517 	 * If there is no cookie, assume directory was stale.
2518 	 */
2519 	ncl_dircookie_lock(dnp);
2520 	NFSUNLOCKNODE(dnp);
2521 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2522 	if (cookiep) {
2523 		cookie = *cookiep;
2524 		ncl_dircookie_unlock(dnp);
2525 	} else {
2526 		ncl_dircookie_unlock(dnp);
2527 		return (NFSERR_BAD_COOKIE);
2528 	}
2529 
2530 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2531 		(void)ncl_fsinfo(nmp, vp, cred, td);
2532 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2533 	    &attrflag, &eof, NULL);
2534 	if (attrflag)
2535 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2536 
2537 	if (!error) {
2538 		/*
2539 		 * We are now either at end of the directory or have filled the
2540 		 * the block.
2541 		 */
2542 		if (eof) {
2543 			NFSLOCKNODE(dnp);
2544 			dnp->n_direofoffset = uiop->uio_offset;
2545 			NFSUNLOCKNODE(dnp);
2546 		} else {
2547 			if (uiop->uio_resid > 0)
2548 				printf("EEK! readdirplusrpc resid > 0\n");
2549 			ncl_dircookie_lock(dnp);
2550 			NFSUNLOCKNODE(dnp);
2551 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2552 			*cookiep = cookie;
2553 			ncl_dircookie_unlock(dnp);
2554 		}
2555 	} else if (NFS_ISV4(vp)) {
2556 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2557 	}
2558 	return (error);
2559 }
2560 
2561 /*
2562  * Silly rename. To make the NFS filesystem that is stateless look a little
2563  * more like the "ufs" a remove of an active vnode is translated to a rename
2564  * to a funny looking filename that is removed by nfs_inactive on the
2565  * nfsnode. There is the potential for another process on a different client
2566  * to create the same funny name between the nfs_lookitup() fails and the
2567  * nfs_rename() completes, but...
2568  */
2569 static int
2570 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2571 {
2572 	struct sillyrename *sp;
2573 	struct nfsnode *np;
2574 	int error;
2575 	short pid;
2576 	unsigned int lticks;
2577 
2578 	cache_purge(dvp);
2579 	np = VTONFS(vp);
2580 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2581 	sp = malloc(sizeof (struct sillyrename),
2582 	    M_NEWNFSREQ, M_WAITOK);
2583 	sp->s_cred = crhold(cnp->cn_cred);
2584 	sp->s_dvp = dvp;
2585 	VREF(dvp);
2586 
2587 	/*
2588 	 * Fudge together a funny name.
2589 	 * Changing the format of the funny name to accommodate more
2590 	 * sillynames per directory.
2591 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2592 	 * CPU ticks since boot.
2593 	 */
2594 	pid = curthread->td_proc->p_pid;
2595 	lticks = (unsigned int)ticks;
2596 	for ( ; ; ) {
2597 		sp->s_namlen = sprintf(sp->s_name,
2598 				       ".nfs.%08x.%04x4.4", lticks,
2599 				       pid);
2600 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2601 				 curthread, NULL))
2602 			break;
2603 		lticks++;
2604 	}
2605 	error = nfs_renameit(dvp, vp, cnp, sp);
2606 	if (error)
2607 		goto bad;
2608 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2609 		curthread, &np);
2610 	np->n_sillyrename = sp;
2611 	return (0);
2612 bad:
2613 	vrele(sp->s_dvp);
2614 	crfree(sp->s_cred);
2615 	free(sp, M_NEWNFSREQ);
2616 	return (error);
2617 }
2618 
2619 /*
2620  * Look up a file name and optionally either update the file handle or
2621  * allocate an nfsnode, depending on the value of npp.
2622  * npp == NULL	--> just do the lookup
2623  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2624  *			handled too
2625  * *npp != NULL --> update the file handle in the vnode
2626  */
2627 static int
2628 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2629     struct thread *td, struct nfsnode **npp)
2630 {
2631 	struct vnode *newvp = NULL, *vp;
2632 	struct nfsnode *np, *dnp = VTONFS(dvp);
2633 	struct nfsfh *nfhp, *onfhp;
2634 	struct nfsvattr nfsva, dnfsva;
2635 	struct componentname cn;
2636 	int error = 0, attrflag, dattrflag;
2637 	u_int hash;
2638 
2639 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2640 	    &nfhp, &attrflag, &dattrflag, NULL, 0);
2641 	if (dattrflag)
2642 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2643 	if (npp && !error) {
2644 		if (*npp != NULL) {
2645 		    np = *npp;
2646 		    vp = NFSTOV(np);
2647 		    /*
2648 		     * For NFSv4, check to see if it is the same name and
2649 		     * replace the name, if it is different.
2650 		     */
2651 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2652 			(np->n_v4->n4_namelen != len ||
2653 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2654 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2655 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2656 			 dnp->n_fhp->nfh_len))) {
2657 #ifdef notdef
2658 { char nnn[100]; int nnnl;
2659 nnnl = (len < 100) ? len : 99;
2660 bcopy(name, nnn, nnnl);
2661 nnn[nnnl] = '\0';
2662 printf("replace=%s\n",nnn);
2663 }
2664 #endif
2665 			    free(np->n_v4, M_NFSV4NODE);
2666 			    np->n_v4 = malloc(
2667 				sizeof (struct nfsv4node) +
2668 				dnp->n_fhp->nfh_len + len - 1,
2669 				M_NFSV4NODE, M_WAITOK);
2670 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2671 			    np->n_v4->n4_namelen = len;
2672 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2673 				dnp->n_fhp->nfh_len);
2674 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2675 		    }
2676 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2677 			FNV1_32_INIT);
2678 		    onfhp = np->n_fhp;
2679 		    /*
2680 		     * Rehash node for new file handle.
2681 		     */
2682 		    vfs_hash_rehash(vp, hash);
2683 		    np->n_fhp = nfhp;
2684 		    if (onfhp != NULL)
2685 			free(onfhp, M_NFSFH);
2686 		    newvp = NFSTOV(np);
2687 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2688 		    free(nfhp, M_NFSFH);
2689 		    VREF(dvp);
2690 		    newvp = dvp;
2691 		} else {
2692 		    cn.cn_nameptr = name;
2693 		    cn.cn_namelen = len;
2694 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2695 			&np, NULL, LK_EXCLUSIVE);
2696 		    if (error)
2697 			return (error);
2698 		    newvp = NFSTOV(np);
2699 		}
2700 		if (!attrflag && *npp == NULL) {
2701 			if (newvp == dvp)
2702 				vrele(newvp);
2703 			else
2704 				vput(newvp);
2705 			return (ENOENT);
2706 		}
2707 		if (attrflag)
2708 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2709 			    0, 1);
2710 	}
2711 	if (npp && *npp == NULL) {
2712 		if (error) {
2713 			if (newvp) {
2714 				if (newvp == dvp)
2715 					vrele(newvp);
2716 				else
2717 					vput(newvp);
2718 			}
2719 		} else
2720 			*npp = np;
2721 	}
2722 	if (error && NFS_ISV4(dvp))
2723 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2724 	return (error);
2725 }
2726 
2727 /*
2728  * Nfs Version 3 and 4 commit rpc
2729  */
2730 int
2731 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2732    struct thread *td)
2733 {
2734 	struct nfsvattr nfsva;
2735 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2736 	struct nfsnode *np;
2737 	struct uio uio;
2738 	int error, attrflag;
2739 
2740 	np = VTONFS(vp);
2741 	error = EIO;
2742 	attrflag = 0;
2743 	if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2744 		uio.uio_offset = offset;
2745 		uio.uio_resid = cnt;
2746 		error = nfscl_doiods(vp, &uio, NULL, NULL,
2747 		    NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2748 		if (error != 0) {
2749 			NFSLOCKNODE(np);
2750 			np->n_flag &= ~NDSCOMMIT;
2751 			NFSUNLOCKNODE(np);
2752 		}
2753 	}
2754 	if (error != 0) {
2755 		mtx_lock(&nmp->nm_mtx);
2756 		if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2757 			mtx_unlock(&nmp->nm_mtx);
2758 			return (0);
2759 		}
2760 		mtx_unlock(&nmp->nm_mtx);
2761 		error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2762 		    &attrflag, NULL);
2763 	}
2764 	if (attrflag != 0)
2765 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2766 		    0, 1);
2767 	if (error != 0 && NFS_ISV4(vp))
2768 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2769 	return (error);
2770 }
2771 
2772 /*
2773  * Strategy routine.
2774  * For async requests when nfsiod(s) are running, queue the request by
2775  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2776  * request.
2777  */
2778 static int
2779 nfs_strategy(struct vop_strategy_args *ap)
2780 {
2781 	struct buf *bp;
2782 	struct vnode *vp;
2783 	struct ucred *cr;
2784 
2785 	bp = ap->a_bp;
2786 	vp = ap->a_vp;
2787 	KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2788 	KASSERT(!(bp->b_flags & B_DONE),
2789 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2790 
2791 	if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2792 		bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2793 		    DEV_BSIZE);
2794 	if (bp->b_iocmd == BIO_READ)
2795 		cr = bp->b_rcred;
2796 	else
2797 		cr = bp->b_wcred;
2798 
2799 	/*
2800 	 * If the op is asynchronous and an i/o daemon is waiting
2801 	 * queue the request, wake it up and wait for completion
2802 	 * otherwise just do it ourselves.
2803 	 */
2804 	if ((bp->b_flags & B_ASYNC) == 0 ||
2805 	    ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2806 		(void) ncl_doio(vp, bp, cr, curthread, 1);
2807 	return (0);
2808 }
2809 
2810 /*
2811  * fsync vnode op. Just call ncl_flush() with commit == 1.
2812  */
2813 /* ARGSUSED */
2814 static int
2815 nfs_fsync(struct vop_fsync_args *ap)
2816 {
2817 
2818 	if (ap->a_vp->v_type != VREG) {
2819 		/*
2820 		 * For NFS, metadata is changed synchronously on the server,
2821 		 * so there is nothing to flush. Also, ncl_flush() clears
2822 		 * the NMODIFIED flag and that shouldn't be done here for
2823 		 * directories.
2824 		 */
2825 		return (0);
2826 	}
2827 	return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2828 }
2829 
2830 /*
2831  * Flush all the blocks associated with a vnode.
2832  * 	Walk through the buffer pool and push any dirty pages
2833  *	associated with the vnode.
2834  * If the called_from_renewthread argument is TRUE, it has been called
2835  * from the NFSv4 renew thread and, as such, cannot block indefinitely
2836  * waiting for a buffer write to complete.
2837  */
2838 int
2839 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2840     int commit, int called_from_renewthread)
2841 {
2842 	struct nfsnode *np = VTONFS(vp);
2843 	struct buf *bp;
2844 	int i;
2845 	struct buf *nbp;
2846 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2847 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2848 	int passone = 1, trycnt = 0;
2849 	u_quad_t off, endoff, toff;
2850 	struct ucred* wcred = NULL;
2851 	struct buf **bvec = NULL;
2852 	struct bufobj *bo;
2853 #ifndef NFS_COMMITBVECSIZ
2854 #define	NFS_COMMITBVECSIZ	20
2855 #endif
2856 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2857 	u_int bvecsize = 0, bveccount;
2858 
2859 	if (called_from_renewthread != 0)
2860 		slptimeo = hz;
2861 	if (nmp->nm_flag & NFSMNT_INT)
2862 		slpflag = PCATCH;
2863 	if (!commit)
2864 		passone = 0;
2865 	bo = &vp->v_bufobj;
2866 	/*
2867 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2868 	 * server, but has not been committed to stable storage on the server
2869 	 * yet. On the first pass, the byte range is worked out and the commit
2870 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2871 	 * job.
2872 	 */
2873 again:
2874 	off = (u_quad_t)-1;
2875 	endoff = 0;
2876 	bvecpos = 0;
2877 	if (NFS_ISV34(vp) && commit) {
2878 		if (bvec != NULL && bvec != bvec_on_stack)
2879 			free(bvec, M_TEMP);
2880 		/*
2881 		 * Count up how many buffers waiting for a commit.
2882 		 */
2883 		bveccount = 0;
2884 		BO_LOCK(bo);
2885 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2886 			if (!BUF_ISLOCKED(bp) &&
2887 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2888 				== (B_DELWRI | B_NEEDCOMMIT))
2889 				bveccount++;
2890 		}
2891 		/*
2892 		 * Allocate space to remember the list of bufs to commit.  It is
2893 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2894 		 * If we can't get memory (for whatever reason), we will end up
2895 		 * committing the buffers one-by-one in the loop below.
2896 		 */
2897 		if (bveccount > NFS_COMMITBVECSIZ) {
2898 			/*
2899 			 * Release the vnode interlock to avoid a lock
2900 			 * order reversal.
2901 			 */
2902 			BO_UNLOCK(bo);
2903 			bvec = (struct buf **)
2904 				malloc(bveccount * sizeof(struct buf *),
2905 				       M_TEMP, M_NOWAIT);
2906 			BO_LOCK(bo);
2907 			if (bvec == NULL) {
2908 				bvec = bvec_on_stack;
2909 				bvecsize = NFS_COMMITBVECSIZ;
2910 			} else
2911 				bvecsize = bveccount;
2912 		} else {
2913 			bvec = bvec_on_stack;
2914 			bvecsize = NFS_COMMITBVECSIZ;
2915 		}
2916 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2917 			if (bvecpos >= bvecsize)
2918 				break;
2919 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2920 				nbp = TAILQ_NEXT(bp, b_bobufs);
2921 				continue;
2922 			}
2923 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2924 			    (B_DELWRI | B_NEEDCOMMIT)) {
2925 				BUF_UNLOCK(bp);
2926 				nbp = TAILQ_NEXT(bp, b_bobufs);
2927 				continue;
2928 			}
2929 			BO_UNLOCK(bo);
2930 			bremfree(bp);
2931 			/*
2932 			 * Work out if all buffers are using the same cred
2933 			 * so we can deal with them all with one commit.
2934 			 *
2935 			 * NOTE: we are not clearing B_DONE here, so we have
2936 			 * to do it later on in this routine if we intend to
2937 			 * initiate I/O on the bp.
2938 			 *
2939 			 * Note: to avoid loopback deadlocks, we do not
2940 			 * assign b_runningbufspace.
2941 			 */
2942 			if (wcred == NULL)
2943 				wcred = bp->b_wcred;
2944 			else if (wcred != bp->b_wcred)
2945 				wcred = NOCRED;
2946 			vfs_busy_pages(bp, 1);
2947 
2948 			BO_LOCK(bo);
2949 			/*
2950 			 * bp is protected by being locked, but nbp is not
2951 			 * and vfs_busy_pages() may sleep.  We have to
2952 			 * recalculate nbp.
2953 			 */
2954 			nbp = TAILQ_NEXT(bp, b_bobufs);
2955 
2956 			/*
2957 			 * A list of these buffers is kept so that the
2958 			 * second loop knows which buffers have actually
2959 			 * been committed. This is necessary, since there
2960 			 * may be a race between the commit rpc and new
2961 			 * uncommitted writes on the file.
2962 			 */
2963 			bvec[bvecpos++] = bp;
2964 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2965 				bp->b_dirtyoff;
2966 			if (toff < off)
2967 				off = toff;
2968 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2969 			if (toff > endoff)
2970 				endoff = toff;
2971 		}
2972 		BO_UNLOCK(bo);
2973 	}
2974 	if (bvecpos > 0) {
2975 		/*
2976 		 * Commit data on the server, as required.
2977 		 * If all bufs are using the same wcred, then use that with
2978 		 * one call for all of them, otherwise commit each one
2979 		 * separately.
2980 		 */
2981 		if (wcred != NOCRED)
2982 			retv = ncl_commit(vp, off, (int)(endoff - off),
2983 					  wcred, td);
2984 		else {
2985 			retv = 0;
2986 			for (i = 0; i < bvecpos; i++) {
2987 				off_t off, size;
2988 				bp = bvec[i];
2989 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2990 					bp->b_dirtyoff;
2991 				size = (u_quad_t)(bp->b_dirtyend
2992 						  - bp->b_dirtyoff);
2993 				retv = ncl_commit(vp, off, (int)size,
2994 						  bp->b_wcred, td);
2995 				if (retv) break;
2996 			}
2997 		}
2998 
2999 		if (retv == NFSERR_STALEWRITEVERF)
3000 			ncl_clearcommit(vp->v_mount);
3001 
3002 		/*
3003 		 * Now, either mark the blocks I/O done or mark the
3004 		 * blocks dirty, depending on whether the commit
3005 		 * succeeded.
3006 		 */
3007 		for (i = 0; i < bvecpos; i++) {
3008 			bp = bvec[i];
3009 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3010 			if (retv) {
3011 				/*
3012 				 * Error, leave B_DELWRI intact
3013 				 */
3014 				vfs_unbusy_pages(bp);
3015 				brelse(bp);
3016 			} else {
3017 				/*
3018 				 * Success, remove B_DELWRI ( bundirty() ).
3019 				 *
3020 				 * b_dirtyoff/b_dirtyend seem to be NFS
3021 				 * specific.  We should probably move that
3022 				 * into bundirty(). XXX
3023 				 */
3024 				bufobj_wref(bo);
3025 				bp->b_flags |= B_ASYNC;
3026 				bundirty(bp);
3027 				bp->b_flags &= ~B_DONE;
3028 				bp->b_ioflags &= ~BIO_ERROR;
3029 				bp->b_dirtyoff = bp->b_dirtyend = 0;
3030 				bufdone(bp);
3031 			}
3032 		}
3033 	}
3034 
3035 	/*
3036 	 * Start/do any write(s) that are required.
3037 	 */
3038 loop:
3039 	BO_LOCK(bo);
3040 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3041 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3042 			if (waitfor != MNT_WAIT || passone)
3043 				continue;
3044 
3045 			error = BUF_TIMELOCK(bp,
3046 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3047 			    BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
3048 			if (error == 0) {
3049 				BUF_UNLOCK(bp);
3050 				goto loop;
3051 			}
3052 			if (error == ENOLCK) {
3053 				error = 0;
3054 				goto loop;
3055 			}
3056 			if (called_from_renewthread != 0) {
3057 				/*
3058 				 * Return EIO so the flush will be retried
3059 				 * later.
3060 				 */
3061 				error = EIO;
3062 				goto done;
3063 			}
3064 			if (newnfs_sigintr(nmp, td)) {
3065 				error = EINTR;
3066 				goto done;
3067 			}
3068 			if (slpflag == PCATCH) {
3069 				slpflag = 0;
3070 				slptimeo = 2 * hz;
3071 			}
3072 			goto loop;
3073 		}
3074 		if ((bp->b_flags & B_DELWRI) == 0)
3075 			panic("nfs_fsync: not dirty");
3076 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3077 			BUF_UNLOCK(bp);
3078 			continue;
3079 		}
3080 		BO_UNLOCK(bo);
3081 		bremfree(bp);
3082 		bp->b_flags |= B_ASYNC;
3083 		bwrite(bp);
3084 		if (newnfs_sigintr(nmp, td)) {
3085 			error = EINTR;
3086 			goto done;
3087 		}
3088 		goto loop;
3089 	}
3090 	if (passone) {
3091 		passone = 0;
3092 		BO_UNLOCK(bo);
3093 		goto again;
3094 	}
3095 	if (waitfor == MNT_WAIT) {
3096 		while (bo->bo_numoutput) {
3097 			error = bufobj_wwait(bo, slpflag, slptimeo);
3098 			if (error) {
3099 			    BO_UNLOCK(bo);
3100 			    if (called_from_renewthread != 0) {
3101 				/*
3102 				 * Return EIO so that the flush will be
3103 				 * retried later.
3104 				 */
3105 				error = EIO;
3106 				goto done;
3107 			    }
3108 			    error = newnfs_sigintr(nmp, td);
3109 			    if (error)
3110 				goto done;
3111 			    if (slpflag == PCATCH) {
3112 				slpflag = 0;
3113 				slptimeo = 2 * hz;
3114 			    }
3115 			    BO_LOCK(bo);
3116 			}
3117 		}
3118 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
3119 			BO_UNLOCK(bo);
3120 			goto loop;
3121 		}
3122 		/*
3123 		 * Wait for all the async IO requests to drain
3124 		 */
3125 		BO_UNLOCK(bo);
3126 		NFSLOCKNODE(np);
3127 		while (np->n_directio_asyncwr > 0) {
3128 			np->n_flag |= NFSYNCWAIT;
3129 			error = newnfs_msleep(td, &np->n_directio_asyncwr,
3130 			    &np->n_mtx, slpflag | (PRIBIO + 1),
3131 			    "nfsfsync", 0);
3132 			if (error) {
3133 				if (newnfs_sigintr(nmp, td)) {
3134 					NFSUNLOCKNODE(np);
3135 					error = EINTR;
3136 					goto done;
3137 				}
3138 			}
3139 		}
3140 		NFSUNLOCKNODE(np);
3141 	} else
3142 		BO_UNLOCK(bo);
3143 	if (NFSHASPNFS(nmp)) {
3144 		nfscl_layoutcommit(vp, td);
3145 		/*
3146 		 * Invalidate the attribute cache, since writes to a DS
3147 		 * won't update the size attribute.
3148 		 */
3149 		NFSLOCKNODE(np);
3150 		np->n_attrstamp = 0;
3151 	} else
3152 		NFSLOCKNODE(np);
3153 	if (np->n_flag & NWRITEERR) {
3154 		error = np->n_error;
3155 		np->n_flag &= ~NWRITEERR;
3156 	}
3157   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
3158 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3159   		np->n_flag &= ~NMODIFIED;
3160 	NFSUNLOCKNODE(np);
3161 done:
3162 	if (bvec != NULL && bvec != bvec_on_stack)
3163 		free(bvec, M_TEMP);
3164 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
3165 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
3166 	    np->n_directio_asyncwr != 0)) {
3167 		if (trycnt++ < 5) {
3168 			/* try, try again... */
3169 			passone = 1;
3170 			wcred = NULL;
3171 			bvec = NULL;
3172 			bvecsize = 0;
3173 			goto again;
3174 		}
3175 		vn_printf(vp, "ncl_flush failed");
3176 		error = called_from_renewthread != 0 ? EIO : EBUSY;
3177 	}
3178 	return (error);
3179 }
3180 
3181 /*
3182  * NFS advisory byte-level locks.
3183  */
3184 static int
3185 nfs_advlock(struct vop_advlock_args *ap)
3186 {
3187 	struct vnode *vp = ap->a_vp;
3188 	struct ucred *cred;
3189 	struct nfsnode *np = VTONFS(ap->a_vp);
3190 	struct proc *p = (struct proc *)ap->a_id;
3191 	struct thread *td = curthread;	/* XXX */
3192 	struct vattr va;
3193 	int ret, error;
3194 	u_quad_t size;
3195 	struct nfsmount *nmp;
3196 
3197 	error = NFSVOPLOCK(vp, LK_SHARED);
3198 	if (error != 0)
3199 		return (EBADF);
3200 	if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3201 		if (vp->v_type != VREG) {
3202 			error = EINVAL;
3203 			goto out;
3204 		}
3205 		if ((ap->a_flags & F_POSIX) != 0)
3206 			cred = p->p_ucred;
3207 		else
3208 			cred = td->td_ucred;
3209 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
3210 		if (VN_IS_DOOMED(vp)) {
3211 			error = EBADF;
3212 			goto out;
3213 		}
3214 
3215 		/*
3216 		 * If this is unlocking a write locked region, flush and
3217 		 * commit them before unlocking. This is required by
3218 		 * RFC3530 Sec. 9.3.2.
3219 		 */
3220 		if (ap->a_op == F_UNLCK &&
3221 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3222 		    ap->a_flags))
3223 			(void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3224 
3225 		/*
3226 		 * Mark NFS node as might have acquired a lock.
3227 		 * This is separate from NHASBEENLOCKED, because it must
3228 		 * be done before the nfsrpc_advlock() call, which might
3229 		 * add a nfscllock structure to the client state.
3230 		 * It is used to check for the case where a nfscllock
3231 		 * state structure cannot exist for the file.
3232 		 * Only done for "oneopenown" NFSv4.1/4.2 mounts.
3233 		 */
3234 		nmp = VFSTONFS(vp->v_mount);
3235 		if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) {
3236 			NFSLOCKNODE(np);
3237 			np->n_flag |= NMIGHTBELOCKED;
3238 			NFSUNLOCKNODE(np);
3239 		}
3240 
3241 		/*
3242 		 * Loop around doing the lock op, while a blocking lock
3243 		 * must wait for the lock op to succeed.
3244 		 */
3245 		do {
3246 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3247 			    ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3248 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3249 			    ap->a_op == F_SETLK) {
3250 				NFSVOPUNLOCK(vp);
3251 				error = nfs_catnap(PZERO | PCATCH, ret,
3252 				    "ncladvl");
3253 				if (error)
3254 					return (EINTR);
3255 				NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3256 				if (VN_IS_DOOMED(vp)) {
3257 					error = EBADF;
3258 					goto out;
3259 				}
3260 			}
3261 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3262 		     ap->a_op == F_SETLK);
3263 		if (ret == NFSERR_DENIED) {
3264 			error = EAGAIN;
3265 			goto out;
3266 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3267 			error = ret;
3268 			goto out;
3269 		} else if (ret != 0) {
3270 			error = EACCES;
3271 			goto out;
3272 		}
3273 
3274 		/*
3275 		 * Now, if we just got a lock, invalidate data in the buffer
3276 		 * cache, as required, so that the coherency conforms with
3277 		 * RFC3530 Sec. 9.3.2.
3278 		 */
3279 		if (ap->a_op == F_SETLK) {
3280 			if ((np->n_flag & NMODIFIED) == 0) {
3281 				np->n_attrstamp = 0;
3282 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3283 				ret = VOP_GETATTR(vp, &va, cred);
3284 			}
3285 			if ((np->n_flag & NMODIFIED) || ret ||
3286 			    np->n_change != va.va_filerev) {
3287 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3288 				np->n_attrstamp = 0;
3289 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3290 				ret = VOP_GETATTR(vp, &va, cred);
3291 				if (!ret) {
3292 					np->n_mtime = va.va_mtime;
3293 					np->n_change = va.va_filerev;
3294 				}
3295 			}
3296 			/* Mark that a file lock has been acquired. */
3297 			NFSLOCKNODE(np);
3298 			np->n_flag |= NHASBEENLOCKED;
3299 			NFSUNLOCKNODE(np);
3300 		}
3301 	} else if (!NFS_ISV4(vp)) {
3302 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3303 			size = VTONFS(vp)->n_size;
3304 			NFSVOPUNLOCK(vp);
3305 			error = lf_advlock(ap, &(vp->v_lockf), size);
3306 		} else {
3307 			if (nfs_advlock_p != NULL)
3308 				error = nfs_advlock_p(ap);
3309 			else {
3310 				NFSVOPUNLOCK(vp);
3311 				error = ENOLCK;
3312 			}
3313 		}
3314 		if (error == 0 && ap->a_op == F_SETLK) {
3315 			error = NFSVOPLOCK(vp, LK_SHARED);
3316 			if (error == 0) {
3317 				/* Mark that a file lock has been acquired. */
3318 				NFSLOCKNODE(np);
3319 				np->n_flag |= NHASBEENLOCKED;
3320 				NFSUNLOCKNODE(np);
3321 				NFSVOPUNLOCK(vp);
3322 			}
3323 		}
3324 		return (error);
3325 	} else
3326 		error = EOPNOTSUPP;
3327 out:
3328 	NFSVOPUNLOCK(vp);
3329 	return (error);
3330 }
3331 
3332 /*
3333  * NFS advisory byte-level locks.
3334  */
3335 static int
3336 nfs_advlockasync(struct vop_advlockasync_args *ap)
3337 {
3338 	struct vnode *vp = ap->a_vp;
3339 	u_quad_t size;
3340 	int error;
3341 
3342 	if (NFS_ISV4(vp))
3343 		return (EOPNOTSUPP);
3344 	error = NFSVOPLOCK(vp, LK_SHARED);
3345 	if (error)
3346 		return (error);
3347 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3348 		size = VTONFS(vp)->n_size;
3349 		NFSVOPUNLOCK(vp);
3350 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
3351 	} else {
3352 		NFSVOPUNLOCK(vp);
3353 		error = EOPNOTSUPP;
3354 	}
3355 	return (error);
3356 }
3357 
3358 /*
3359  * Print out the contents of an nfsnode.
3360  */
3361 static int
3362 nfs_print(struct vop_print_args *ap)
3363 {
3364 	struct vnode *vp = ap->a_vp;
3365 	struct nfsnode *np = VTONFS(vp);
3366 
3367 	printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3368 	    (uintmax_t)np->n_vattr.na_fsid);
3369 	if (vp->v_type == VFIFO)
3370 		fifo_printinfo(vp);
3371 	printf("\n");
3372 	return (0);
3373 }
3374 
3375 /*
3376  * This is the "real" nfs::bwrite(struct buf*).
3377  * We set B_CACHE if this is a VMIO buffer.
3378  */
3379 int
3380 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3381 {
3382 	int oldflags, rtval;
3383 
3384 	if (bp->b_flags & B_INVAL) {
3385 		brelse(bp);
3386 		return (0);
3387 	}
3388 
3389 	oldflags = bp->b_flags;
3390 	bp->b_flags |= B_CACHE;
3391 
3392 	/*
3393 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3394 	 */
3395 	bundirty(bp);
3396 	bp->b_flags &= ~B_DONE;
3397 	bp->b_ioflags &= ~BIO_ERROR;
3398 	bp->b_iocmd = BIO_WRITE;
3399 
3400 	bufobj_wref(bp->b_bufobj);
3401 	curthread->td_ru.ru_oublock++;
3402 
3403 	/*
3404 	 * Note: to avoid loopback deadlocks, we do not
3405 	 * assign b_runningbufspace.
3406 	 */
3407 	vfs_busy_pages(bp, 1);
3408 
3409 	BUF_KERNPROC(bp);
3410 	bp->b_iooffset = dbtob(bp->b_blkno);
3411 	bstrategy(bp);
3412 
3413 	if ((oldflags & B_ASYNC) != 0)
3414 		return (0);
3415 
3416 	rtval = bufwait(bp);
3417 	if (oldflags & B_DELWRI)
3418 		reassignbuf(bp);
3419 	brelse(bp);
3420 	return (rtval);
3421 }
3422 
3423 /*
3424  * nfs special file access vnode op.
3425  * Essentially just get vattr and then imitate iaccess() since the device is
3426  * local to the client.
3427  */
3428 static int
3429 nfsspec_access(struct vop_access_args *ap)
3430 {
3431 	struct vattr *vap;
3432 	struct ucred *cred = ap->a_cred;
3433 	struct vnode *vp = ap->a_vp;
3434 	accmode_t accmode = ap->a_accmode;
3435 	struct vattr vattr;
3436 	int error;
3437 
3438 	/*
3439 	 * Disallow write attempts on filesystems mounted read-only;
3440 	 * unless the file is a socket, fifo, or a block or character
3441 	 * device resident on the filesystem.
3442 	 */
3443 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3444 		switch (vp->v_type) {
3445 		case VREG:
3446 		case VDIR:
3447 		case VLNK:
3448 			return (EROFS);
3449 		default:
3450 			break;
3451 		}
3452 	}
3453 	vap = &vattr;
3454 	error = VOP_GETATTR(vp, vap, cred);
3455 	if (error)
3456 		goto out;
3457 	error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3458 	    accmode, cred);
3459 out:
3460 	return error;
3461 }
3462 
3463 /*
3464  * Read wrapper for fifos.
3465  */
3466 static int
3467 nfsfifo_read(struct vop_read_args *ap)
3468 {
3469 	struct nfsnode *np = VTONFS(ap->a_vp);
3470 	int error;
3471 
3472 	/*
3473 	 * Set access flag.
3474 	 */
3475 	NFSLOCKNODE(np);
3476 	np->n_flag |= NACC;
3477 	vfs_timestamp(&np->n_atim);
3478 	NFSUNLOCKNODE(np);
3479 	error = fifo_specops.vop_read(ap);
3480 	return error;
3481 }
3482 
3483 /*
3484  * Write wrapper for fifos.
3485  */
3486 static int
3487 nfsfifo_write(struct vop_write_args *ap)
3488 {
3489 	struct nfsnode *np = VTONFS(ap->a_vp);
3490 
3491 	/*
3492 	 * Set update flag.
3493 	 */
3494 	NFSLOCKNODE(np);
3495 	np->n_flag |= NUPD;
3496 	vfs_timestamp(&np->n_mtim);
3497 	NFSUNLOCKNODE(np);
3498 	return(fifo_specops.vop_write(ap));
3499 }
3500 
3501 /*
3502  * Close wrapper for fifos.
3503  *
3504  * Update the times on the nfsnode then do fifo close.
3505  */
3506 static int
3507 nfsfifo_close(struct vop_close_args *ap)
3508 {
3509 	struct vnode *vp = ap->a_vp;
3510 	struct nfsnode *np = VTONFS(vp);
3511 	struct vattr vattr;
3512 	struct timespec ts;
3513 
3514 	NFSLOCKNODE(np);
3515 	if (np->n_flag & (NACC | NUPD)) {
3516 		vfs_timestamp(&ts);
3517 		if (np->n_flag & NACC)
3518 			np->n_atim = ts;
3519 		if (np->n_flag & NUPD)
3520 			np->n_mtim = ts;
3521 		np->n_flag |= NCHG;
3522 		if (vrefcnt(vp) == 1 &&
3523 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3524 			VATTR_NULL(&vattr);
3525 			if (np->n_flag & NACC)
3526 				vattr.va_atime = np->n_atim;
3527 			if (np->n_flag & NUPD)
3528 				vattr.va_mtime = np->n_mtim;
3529 			NFSUNLOCKNODE(np);
3530 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3531 			goto out;
3532 		}
3533 	}
3534 	NFSUNLOCKNODE(np);
3535 out:
3536 	return (fifo_specops.vop_close(ap));
3537 }
3538 
3539 /*
3540  * Just call ncl_writebp() with the force argument set to 1.
3541  *
3542  * NOTE: B_DONE may or may not be set in a_bp on call.
3543  */
3544 static int
3545 nfs_bwrite(struct buf *bp)
3546 {
3547 
3548 	return (ncl_writebp(bp, 1, curthread));
3549 }
3550 
3551 struct buf_ops buf_ops_newnfs = {
3552 	.bop_name	=	"buf_ops_nfs",
3553 	.bop_write	=	nfs_bwrite,
3554 	.bop_strategy	=	bufstrategy,
3555 	.bop_sync	=	bufsync,
3556 	.bop_bdflush	=	bufbdflush,
3557 };
3558 
3559 static int
3560 nfs_getacl(struct vop_getacl_args *ap)
3561 {
3562 	int error;
3563 
3564 	if (ap->a_type != ACL_TYPE_NFS4)
3565 		return (EOPNOTSUPP);
3566 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3567 	    NULL);
3568 	if (error > NFSERR_STALE) {
3569 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3570 		error = EPERM;
3571 	}
3572 	return (error);
3573 }
3574 
3575 static int
3576 nfs_setacl(struct vop_setacl_args *ap)
3577 {
3578 	int error;
3579 
3580 	if (ap->a_type != ACL_TYPE_NFS4)
3581 		return (EOPNOTSUPP);
3582 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3583 	    NULL);
3584 	if (error > NFSERR_STALE) {
3585 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3586 		error = EPERM;
3587 	}
3588 	return (error);
3589 }
3590 
3591 /*
3592  * VOP_ADVISE for NFS.
3593  * Just return 0 for any errors, since it is just a hint.
3594  */
3595 static int
3596 nfs_advise(struct vop_advise_args *ap)
3597 {
3598 	struct thread *td = curthread;
3599 	struct nfsmount *nmp;
3600 	uint64_t len;
3601 	int error;
3602 
3603 	/*
3604 	 * First do vop_stdadvise() to handle the buffer cache.
3605 	 */
3606 	error = vop_stdadvise(ap);
3607 	if (error != 0)
3608 		return (error);
3609 	if (ap->a_start < 0 || ap->a_end < 0)
3610 		return (0);
3611 	if (ap->a_end == OFF_MAX)
3612 		len = 0;
3613 	else if (ap->a_end < ap->a_start)
3614 		return (0);
3615 	else
3616 		len = ap->a_end - ap->a_start + 1;
3617 	nmp = VFSTONFS(ap->a_vp->v_mount);
3618 	mtx_lock(&nmp->nm_mtx);
3619 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3620 	    (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) ==
3621 	    0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) {
3622 		mtx_unlock(&nmp->nm_mtx);
3623 		return (0);
3624 	}
3625 	mtx_unlock(&nmp->nm_mtx);
3626 	error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice,
3627 	    td->td_ucred, td);
3628 	if (error == NFSERR_NOTSUPP) {
3629 		mtx_lock(&nmp->nm_mtx);
3630 		nmp->nm_privflag |= NFSMNTP_NOADVISE;
3631 		mtx_unlock(&nmp->nm_mtx);
3632 	}
3633 	return (0);
3634 }
3635 
3636 /*
3637  * nfs allocate call
3638  */
3639 static int
3640 nfs_allocate(struct vop_allocate_args *ap)
3641 {
3642 	struct vnode *vp = ap->a_vp;
3643 	struct thread *td = curthread;
3644 	struct nfsvattr nfsva;
3645 	struct nfsmount *nmp;
3646 	off_t alen;
3647 	int attrflag, error, ret;
3648 
3649 	attrflag = 0;
3650 	nmp = VFSTONFS(vp->v_mount);
3651 	mtx_lock(&nmp->nm_mtx);
3652 	if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3653 	    (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) {
3654 		mtx_unlock(&nmp->nm_mtx);
3655 		/*
3656 		 * Flush first to ensure that the allocate adds to the
3657 		 * file's allocation on the server.
3658 		 */
3659 		error = ncl_flush(vp, MNT_WAIT, td, 1, 0);
3660 		if (error == 0) {
3661 			alen = *ap->a_len;
3662 			if ((uint64_t)alen > nfs_maxalloclen)
3663 				alen = nfs_maxalloclen;
3664 			error = nfsrpc_allocate(vp, *ap->a_offset, alen,
3665 			    &nfsva, &attrflag, td->td_ucred, td, NULL);
3666 		}
3667 		if (error == 0) {
3668 			*ap->a_offset += alen;
3669 			*ap->a_len -= alen;
3670 		} else if (error == NFSERR_NOTSUPP) {
3671 			mtx_lock(&nmp->nm_mtx);
3672 			nmp->nm_privflag |= NFSMNTP_NOALLOCATE;
3673 			mtx_unlock(&nmp->nm_mtx);
3674 			error = EINVAL;
3675 		}
3676 	} else {
3677 		mtx_unlock(&nmp->nm_mtx);
3678 		error = EINVAL;
3679 	}
3680 	if (attrflag != 0) {
3681 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3682 		if (error == 0 && ret != 0)
3683 			error = ret;
3684 	}
3685 	if (error != 0)
3686 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3687 	return (error);
3688 }
3689 
3690 /*
3691  * nfs deallocate call
3692  */
3693 static int
3694 nfs_deallocate(struct vop_deallocate_args *ap)
3695 {
3696 	struct vnode *vp = ap->a_vp;
3697 	struct thread *td = curthread;
3698 	struct nfsvattr nfsva;
3699 	struct nfsmount *nmp;
3700 	struct nfsnode *np;
3701 	off_t tlen, mlen;
3702 	int attrflag, error, ret;
3703 	bool clipped;
3704 
3705 	error = 0;
3706 	attrflag = 0;
3707 	nmp = VFSTONFS(vp->v_mount);
3708 	np = VTONFS(vp);
3709 	mtx_lock(&nmp->nm_mtx);
3710 	if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3711 	    (nmp->nm_privflag & NFSMNTP_NODEALLOCATE) == 0) {
3712 		mtx_unlock(&nmp->nm_mtx);
3713 		tlen = omin(OFF_MAX - *ap->a_offset, *ap->a_len);
3714 		NFSCL_DEBUG(4, "dealloc: off=%jd len=%jd maxfilesize=%ju\n",
3715 		    (intmax_t)*ap->a_offset, (intmax_t)tlen,
3716 		    (uintmax_t)nmp->nm_maxfilesize);
3717 		if ((uint64_t)*ap->a_offset >= nmp->nm_maxfilesize) {
3718 			/* Avoid EFBIG error return from the NFSv4.2 server. */
3719 			*ap->a_len = 0;
3720 			return (0);
3721 		}
3722 		clipped = false;
3723 		if ((uint64_t)*ap->a_offset + tlen > nmp->nm_maxfilesize)
3724 			tlen = nmp->nm_maxfilesize - *ap->a_offset;
3725 		if ((uint64_t)*ap->a_offset < np->n_size) {
3726 			/* Limit the len to nfs_maxalloclen before EOF. */
3727 			mlen = omin((off_t)np->n_size - *ap->a_offset, tlen);
3728 			if ((uint64_t)mlen > nfs_maxalloclen) {
3729 				NFSCL_DEBUG(4, "dealloc: tlen maxalloclen\n");
3730 				tlen = nfs_maxalloclen;
3731 				clipped = true;
3732 			}
3733 		}
3734 		if (error == 0)
3735 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
3736 		if (error == 0) {
3737 			vnode_pager_purge_range(vp, *ap->a_offset,
3738 			    *ap->a_offset + tlen);
3739 			error = nfsrpc_deallocate(vp, *ap->a_offset, tlen,
3740 			    &nfsva, &attrflag, ap->a_cred, td, NULL);
3741 			NFSCL_DEBUG(4, "dealloc: rpc=%d\n", error);
3742 		}
3743 		if (error == 0) {
3744 			NFSCL_DEBUG(4, "dealloc: attrflag=%d na_size=%ju\n",
3745 			    attrflag, (uintmax_t)nfsva.na_size);
3746 			if (attrflag != 0) {
3747 				if ((uint64_t)*ap->a_offset < nfsva.na_size)
3748 					*ap->a_offset += omin((off_t)
3749 					    nfsva.na_size - *ap->a_offset,
3750 					    tlen);
3751 			}
3752 			if (clipped && tlen < *ap->a_len)
3753 				*ap->a_len -= tlen;
3754 			else
3755 				*ap->a_len = 0;
3756 		} else if (error == NFSERR_NOTSUPP) {
3757 			mtx_lock(&nmp->nm_mtx);
3758 			nmp->nm_privflag |= NFSMNTP_NODEALLOCATE;
3759 			mtx_unlock(&nmp->nm_mtx);
3760 		}
3761 	} else {
3762 		mtx_unlock(&nmp->nm_mtx);
3763 		error = EIO;
3764 	}
3765 	/*
3766 	 * If the NFS server cannot perform the Deallocate operation, just call
3767 	 * vop_stddeallocate() to perform it.
3768 	 */
3769 	if (error != 0 && error != NFSERR_FBIG && error != NFSERR_INVAL) {
3770 		error = vop_stddeallocate(ap);
3771 		NFSCL_DEBUG(4, "dealloc: stddeallocate=%d\n", error);
3772 	}
3773 	if (attrflag != 0) {
3774 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3775 		if (error == 0 && ret != 0)
3776 			error = ret;
3777 	}
3778 	if (error != 0)
3779 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3780 	return (error);
3781 }
3782 
3783 /*
3784  * nfs copy_file_range call
3785  */
3786 static int
3787 nfs_copy_file_range(struct vop_copy_file_range_args *ap)
3788 {
3789 	struct vnode *invp = ap->a_invp;
3790 	struct vnode *outvp = ap->a_outvp;
3791 	struct mount *mp;
3792 	struct nfsvattr innfsva, outnfsva;
3793 	struct vattr *vap;
3794 	struct uio io;
3795 	struct nfsmount *nmp;
3796 	size_t len, len2;
3797 	int error, inattrflag, outattrflag, ret, ret2;
3798 	off_t inoff, outoff;
3799 	bool consecutive, must_commit, tryoutcred;
3800 
3801 	ret = ret2 = 0;
3802 	nmp = VFSTONFS(invp->v_mount);
3803 	mtx_lock(&nmp->nm_mtx);
3804 	/* NFSv4.2 Copy is not permitted for infile == outfile. */
3805 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3806 	    (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0 || invp == outvp) {
3807 		mtx_unlock(&nmp->nm_mtx);
3808 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3809 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3810 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3811 		return (error);
3812 	}
3813 	mtx_unlock(&nmp->nm_mtx);
3814 
3815 	/* Lock both vnodes, avoiding risk of deadlock. */
3816 	do {
3817 		mp = NULL;
3818 		error = vn_start_write(outvp, &mp, V_WAIT);
3819 		if (error == 0) {
3820 			error = vn_lock(outvp, LK_EXCLUSIVE);
3821 			if (error == 0) {
3822 				error = vn_lock(invp, LK_SHARED | LK_NOWAIT);
3823 				if (error == 0)
3824 					break;
3825 				VOP_UNLOCK(outvp);
3826 				if (mp != NULL)
3827 					vn_finished_write(mp);
3828 				mp = NULL;
3829 				error = vn_lock(invp, LK_SHARED);
3830 				if (error == 0)
3831 					VOP_UNLOCK(invp);
3832 			}
3833 		}
3834 		if (mp != NULL)
3835 			vn_finished_write(mp);
3836 	} while (error == 0);
3837 	if (error != 0)
3838 		return (error);
3839 
3840 	/*
3841 	 * Do the vn_rlimit_fsize() check.  Should this be above the VOP layer?
3842 	 */
3843 	io.uio_offset = *ap->a_outoffp;
3844 	io.uio_resid = *ap->a_lenp;
3845 	error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd);
3846 
3847 	/*
3848 	 * Flush the input file so that the data is up to date before
3849 	 * the copy.  Flush writes for the output file so that they
3850 	 * do not overwrite the data copied to the output file by the Copy.
3851 	 * Set the commit argument for both flushes so that the data is on
3852 	 * stable storage before the Copy RPC.  This is done in case the
3853 	 * server reboots during the Copy and needs to be redone.
3854 	 */
3855 	if (error == 0)
3856 		error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0);
3857 	if (error == 0)
3858 		error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0);
3859 
3860 	/* Do the actual NFSv4.2 RPC. */
3861 	len = *ap->a_lenp;
3862 	mtx_lock(&nmp->nm_mtx);
3863 	if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0)
3864 		consecutive = true;
3865 	else
3866 		consecutive = false;
3867 	mtx_unlock(&nmp->nm_mtx);
3868 	inoff = *ap->a_inoffp;
3869 	outoff = *ap->a_outoffp;
3870 	tryoutcred = true;
3871 	must_commit = false;
3872 	if (error == 0) {
3873 		vap = &VTONFS(invp)->n_vattr.na_vattr;
3874 		error = VOP_GETATTR(invp, vap, ap->a_incred);
3875 		if (error == 0) {
3876 			/*
3877 			 * Clip "len" at va_size so that RFC compliant servers
3878 			 * will not reply NFSERR_INVAL.
3879 			 * Setting "len == 0" for the RPC would be preferred,
3880 			 * but some Linux servers do not support that.
3881 			 */
3882 			if (inoff >= vap->va_size)
3883 				*ap->a_lenp = len = 0;
3884 			else if (inoff + len > vap->va_size)
3885 				*ap->a_lenp = len = vap->va_size - inoff;
3886 		} else
3887 			error = 0;
3888 	}
3889 
3890 	/*
3891 	 * len will be set to 0 upon a successful Copy RPC.
3892 	 * As such, this only loops when the Copy RPC needs to be retried.
3893 	 */
3894 	while (len > 0 && error == 0) {
3895 		inattrflag = outattrflag = 0;
3896 		len2 = len;
3897 		if (tryoutcred)
3898 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3899 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3900 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3901 			    ap->a_outcred, consecutive, &must_commit);
3902 		else
3903 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3904 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3905 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3906 			    ap->a_incred, consecutive, &must_commit);
3907 		if (inattrflag != 0)
3908 			ret = nfscl_loadattrcache(&invp, &innfsva, NULL, NULL,
3909 			    0, 1);
3910 		if (outattrflag != 0)
3911 			ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL,
3912 			    NULL, 1, 1);
3913 		if (error == 0) {
3914 			if (consecutive == false) {
3915 				if (len2 == len) {
3916 					mtx_lock(&nmp->nm_mtx);
3917 					nmp->nm_privflag |=
3918 					    NFSMNTP_NOCONSECUTIVE;
3919 					mtx_unlock(&nmp->nm_mtx);
3920 				} else
3921 					error = NFSERR_OFFLOADNOREQS;
3922 			}
3923 			*ap->a_lenp = len2;
3924 			len = 0;
3925 			if (len2 > 0 && must_commit && error == 0)
3926 				error = ncl_commit(outvp, outoff, *ap->a_lenp,
3927 				    ap->a_outcred, curthread);
3928 			if (error == 0 && ret != 0)
3929 				error = ret;
3930 			if (error == 0 && ret2 != 0)
3931 				error = ret2;
3932 		} else if (error == NFSERR_OFFLOADNOREQS && consecutive) {
3933 			/*
3934 			 * Try consecutive == false, which is ok only if all
3935 			 * bytes are copied.
3936 			 * If only some bytes were copied when consecutive
3937 			 * is false, there is no way to know which bytes
3938 			 * still need to be written.
3939 			 */
3940 			consecutive = false;
3941 			error = 0;
3942 		} else if (error == NFSERR_ACCES && tryoutcred) {
3943 			/* Try again with incred. */
3944 			tryoutcred = false;
3945 			error = 0;
3946 		}
3947 		if (error == NFSERR_STALEWRITEVERF) {
3948 			/*
3949 			 * Server rebooted, so do it all again.
3950 			 */
3951 			*ap->a_inoffp = inoff;
3952 			*ap->a_outoffp = outoff;
3953 			len = *ap->a_lenp;
3954 			must_commit = false;
3955 			error = 0;
3956 		}
3957 	}
3958 	VOP_UNLOCK(invp);
3959 	VOP_UNLOCK(outvp);
3960 	if (mp != NULL)
3961 		vn_finished_write(mp);
3962 	if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS ||
3963 	    error == NFSERR_ACCES) {
3964 		/*
3965 		 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can
3966 		 * use a_incred for the read and a_outcred for the write, so
3967 		 * try this for NFSERR_ACCES failures for the Copy.
3968 		 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can
3969 		 * never succeed, so disable it.
3970 		 */
3971 		if (error != NFSERR_ACCES) {
3972 			/* Can never do Copy on this mount. */
3973 			mtx_lock(&nmp->nm_mtx);
3974 			nmp->nm_privflag |= NFSMNTP_NOCOPY;
3975 			mtx_unlock(&nmp->nm_mtx);
3976 		}
3977 		*ap->a_inoffp = inoff;
3978 		*ap->a_outoffp = outoff;
3979 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3980 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3981 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3982 	} else if (error != 0)
3983 		*ap->a_lenp = 0;
3984 
3985 	if (error != 0)
3986 		error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0);
3987 	return (error);
3988 }
3989 
3990 /*
3991  * nfs ioctl call
3992  */
3993 static int
3994 nfs_ioctl(struct vop_ioctl_args *ap)
3995 {
3996 	struct vnode *vp = ap->a_vp;
3997 	struct nfsvattr nfsva;
3998 	struct nfsmount *nmp;
3999 	int attrflag, content, error, ret;
4000 	bool eof = false;			/* shut up compiler. */
4001 
4002 	if (vp->v_type != VREG)
4003 		return (ENOTTY);
4004 	nmp = VFSTONFS(vp->v_mount);
4005 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) {
4006 		error = vop_stdioctl(ap);
4007 		return (error);
4008 	}
4009 
4010 	/* Do the actual NFSv4.2 RPC. */
4011 	switch (ap->a_command) {
4012 	case FIOSEEKDATA:
4013 		content = NFSV4CONTENT_DATA;
4014 		break;
4015 	case FIOSEEKHOLE:
4016 		content = NFSV4CONTENT_HOLE;
4017 		break;
4018 	default:
4019 		return (ENOTTY);
4020 	}
4021 
4022 	error = vn_lock(vp, LK_SHARED);
4023 	if (error != 0)
4024 		return (EBADF);
4025 	attrflag = 0;
4026 	if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size)
4027 		error = ENXIO;
4028 	else {
4029 		/*
4030 		 * Flush all writes, so that the server is up to date.
4031 		 * Although a Commit is not required, the commit argument
4032 		 * is set so that, for a pNFS File/Flexible File Layout
4033 		 * server, the LayoutCommit will be done to ensure the file
4034 		 * size is up to date on the Metadata Server.
4035 		 */
4036 		error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0);
4037 		if (error == 0)
4038 			error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof,
4039 			    content, ap->a_cred, &nfsva, &attrflag);
4040 		/* If at eof for FIOSEEKDATA, return ENXIO. */
4041 		if (eof && error == 0 && content == NFSV4CONTENT_DATA)
4042 			error = ENXIO;
4043 	}
4044 	if (attrflag != 0) {
4045 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4046 		if (error == 0 && ret != 0)
4047 			error = ret;
4048 	}
4049 	NFSVOPUNLOCK(vp);
4050 
4051 	if (error != 0)
4052 		error = ENXIO;
4053 	return (error);
4054 }
4055 
4056 /*
4057  * nfs getextattr call
4058  */
4059 static int
4060 nfs_getextattr(struct vop_getextattr_args *ap)
4061 {
4062 	struct vnode *vp = ap->a_vp;
4063 	struct nfsmount *nmp;
4064 	struct ucred *cred;
4065 	struct thread *td = ap->a_td;
4066 	struct nfsvattr nfsva;
4067 	ssize_t len;
4068 	int attrflag, error, ret;
4069 
4070 	nmp = VFSTONFS(vp->v_mount);
4071 	mtx_lock(&nmp->nm_mtx);
4072 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4073 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4074 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4075 		mtx_unlock(&nmp->nm_mtx);
4076 		return (EOPNOTSUPP);
4077 	}
4078 	mtx_unlock(&nmp->nm_mtx);
4079 
4080 	cred = ap->a_cred;
4081 	if (cred == NULL)
4082 		cred = td->td_ucred;
4083 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4084 	attrflag = 0;
4085 	error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva,
4086 	    &attrflag, cred, td);
4087 	if (attrflag != 0) {
4088 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4089 		if (error == 0 && ret != 0)
4090 			error = ret;
4091 	}
4092 	if (error == 0 && ap->a_size != NULL)
4093 		*ap->a_size = len;
4094 
4095 	switch (error) {
4096 	case NFSERR_NOTSUPP:
4097 	case NFSERR_OPILLEGAL:
4098 		mtx_lock(&nmp->nm_mtx);
4099 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4100 		mtx_unlock(&nmp->nm_mtx);
4101 		error = EOPNOTSUPP;
4102 		break;
4103 	case NFSERR_NOXATTR:
4104 	case NFSERR_XATTR2BIG:
4105 		error = ENOATTR;
4106 		break;
4107 	default:
4108 		error = nfscl_maperr(td, error, 0, 0);
4109 		break;
4110 	}
4111 	return (error);
4112 }
4113 
4114 /*
4115  * nfs setextattr call
4116  */
4117 static int
4118 nfs_setextattr(struct vop_setextattr_args *ap)
4119 {
4120 	struct vnode *vp = ap->a_vp;
4121 	struct nfsmount *nmp;
4122 	struct ucred *cred;
4123 	struct thread *td = ap->a_td;
4124 	struct nfsvattr nfsva;
4125 	int attrflag, error, ret;
4126 
4127 	nmp = VFSTONFS(vp->v_mount);
4128 	mtx_lock(&nmp->nm_mtx);
4129 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4130 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4131 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4132 		mtx_unlock(&nmp->nm_mtx);
4133 		return (EOPNOTSUPP);
4134 	}
4135 	mtx_unlock(&nmp->nm_mtx);
4136 
4137 	if (ap->a_uio->uio_resid < 0)
4138 		return (EINVAL);
4139 	cred = ap->a_cred;
4140 	if (cred == NULL)
4141 		cred = td->td_ucred;
4142 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4143 	attrflag = 0;
4144 	error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva,
4145 	    &attrflag, cred, td);
4146 	if (attrflag != 0) {
4147 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4148 		if (error == 0 && ret != 0)
4149 			error = ret;
4150 	}
4151 
4152 	switch (error) {
4153 	case NFSERR_NOTSUPP:
4154 	case NFSERR_OPILLEGAL:
4155 		mtx_lock(&nmp->nm_mtx);
4156 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4157 		mtx_unlock(&nmp->nm_mtx);
4158 		error = EOPNOTSUPP;
4159 		break;
4160 	case NFSERR_NOXATTR:
4161 	case NFSERR_XATTR2BIG:
4162 		error = ENOATTR;
4163 		break;
4164 	default:
4165 		error = nfscl_maperr(td, error, 0, 0);
4166 		break;
4167 	}
4168 	return (error);
4169 }
4170 
4171 /*
4172  * nfs listextattr call
4173  */
4174 static int
4175 nfs_listextattr(struct vop_listextattr_args *ap)
4176 {
4177 	struct vnode *vp = ap->a_vp;
4178 	struct nfsmount *nmp;
4179 	struct ucred *cred;
4180 	struct thread *td = ap->a_td;
4181 	struct nfsvattr nfsva;
4182 	size_t len, len2;
4183 	uint64_t cookie;
4184 	int attrflag, error, ret;
4185 	bool eof;
4186 
4187 	nmp = VFSTONFS(vp->v_mount);
4188 	mtx_lock(&nmp->nm_mtx);
4189 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4190 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4191 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4192 		mtx_unlock(&nmp->nm_mtx);
4193 		return (EOPNOTSUPP);
4194 	}
4195 	mtx_unlock(&nmp->nm_mtx);
4196 
4197 	cred = ap->a_cred;
4198 	if (cred == NULL)
4199 		cred = td->td_ucred;
4200 
4201 	/* Loop around doing List Extended Attribute RPCs. */
4202 	eof = false;
4203 	cookie = 0;
4204 	len2 = 0;
4205 	error = 0;
4206 	while (!eof && error == 0) {
4207 		len = nmp->nm_rsize;
4208 		attrflag = 0;
4209 		error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof,
4210 		    &nfsva, &attrflag, cred, td);
4211 		if (attrflag != 0) {
4212 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4213 			    1);
4214 			if (error == 0 && ret != 0)
4215 				error = ret;
4216 		}
4217 		if (error == 0) {
4218 			len2 += len;
4219 			if (len2 > SSIZE_MAX)
4220 				error = ENOATTR;
4221 		}
4222 	}
4223 	if (error == 0 && ap->a_size != NULL)
4224 		*ap->a_size = len2;
4225 
4226 	switch (error) {
4227 	case NFSERR_NOTSUPP:
4228 	case NFSERR_OPILLEGAL:
4229 		mtx_lock(&nmp->nm_mtx);
4230 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4231 		mtx_unlock(&nmp->nm_mtx);
4232 		error = EOPNOTSUPP;
4233 		break;
4234 	case NFSERR_NOXATTR:
4235 	case NFSERR_XATTR2BIG:
4236 		error = ENOATTR;
4237 		break;
4238 	default:
4239 		error = nfscl_maperr(td, error, 0, 0);
4240 		break;
4241 	}
4242 	return (error);
4243 }
4244 
4245 /*
4246  * nfs setextattr call
4247  */
4248 static int
4249 nfs_deleteextattr(struct vop_deleteextattr_args *ap)
4250 {
4251 	struct vnode *vp = ap->a_vp;
4252 	struct nfsmount *nmp;
4253 	struct nfsvattr nfsva;
4254 	int attrflag, error, ret;
4255 
4256 	nmp = VFSTONFS(vp->v_mount);
4257 	mtx_lock(&nmp->nm_mtx);
4258 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4259 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4260 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4261 		mtx_unlock(&nmp->nm_mtx);
4262 		return (EOPNOTSUPP);
4263 	}
4264 	mtx_unlock(&nmp->nm_mtx);
4265 
4266 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4267 	attrflag = 0;
4268 	error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred,
4269 	    ap->a_td);
4270 	if (attrflag != 0) {
4271 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4272 		if (error == 0 && ret != 0)
4273 			error = ret;
4274 	}
4275 
4276 	switch (error) {
4277 	case NFSERR_NOTSUPP:
4278 	case NFSERR_OPILLEGAL:
4279 		mtx_lock(&nmp->nm_mtx);
4280 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4281 		mtx_unlock(&nmp->nm_mtx);
4282 		error = EOPNOTSUPP;
4283 		break;
4284 	case NFSERR_NOXATTR:
4285 	case NFSERR_XATTR2BIG:
4286 		error = ENOATTR;
4287 		break;
4288 	default:
4289 		error = nfscl_maperr(ap->a_td, error, 0, 0);
4290 		break;
4291 	}
4292 	return (error);
4293 }
4294 
4295 /*
4296  * Return POSIX pathconf information applicable to nfs filesystems.
4297  */
4298 static int
4299 nfs_pathconf(struct vop_pathconf_args *ap)
4300 {
4301 	struct nfsv3_pathconf pc;
4302 	struct nfsvattr nfsva;
4303 	struct vnode *vp = ap->a_vp;
4304 	struct nfsmount *nmp;
4305 	struct thread *td = curthread;
4306 	off_t off;
4307 	bool eof;
4308 	int attrflag, error;
4309 
4310 	if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
4311 	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
4312 	    ap->a_name == _PC_NO_TRUNC)) ||
4313 	    (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
4314 		/*
4315 		 * Since only the above 4 a_names are returned by the NFSv3
4316 		 * Pathconf RPC, there is no point in doing it for others.
4317 		 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
4318 		 * be used for _PC_NFS4_ACL as well.
4319 		 */
4320 		error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
4321 		    &attrflag, NULL);
4322 		if (attrflag != 0)
4323 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4324 			    1);
4325 		if (error != 0)
4326 			return (error);
4327 	} else {
4328 		/*
4329 		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
4330 		 * just fake them.
4331 		 */
4332 		pc.pc_linkmax = NFS_LINK_MAX;
4333 		pc.pc_namemax = NFS_MAXNAMLEN;
4334 		pc.pc_notrunc = 1;
4335 		pc.pc_chownrestricted = 1;
4336 		pc.pc_caseinsensitive = 0;
4337 		pc.pc_casepreserving = 1;
4338 		error = 0;
4339 	}
4340 	switch (ap->a_name) {
4341 	case _PC_LINK_MAX:
4342 #ifdef _LP64
4343 		*ap->a_retval = pc.pc_linkmax;
4344 #else
4345 		*ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
4346 #endif
4347 		break;
4348 	case _PC_NAME_MAX:
4349 		*ap->a_retval = pc.pc_namemax;
4350 		break;
4351 	case _PC_PIPE_BUF:
4352 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
4353 			*ap->a_retval = PIPE_BUF;
4354 		else
4355 			error = EINVAL;
4356 		break;
4357 	case _PC_CHOWN_RESTRICTED:
4358 		*ap->a_retval = pc.pc_chownrestricted;
4359 		break;
4360 	case _PC_NO_TRUNC:
4361 		*ap->a_retval = pc.pc_notrunc;
4362 		break;
4363 	case _PC_ACL_NFS4:
4364 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
4365 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
4366 			*ap->a_retval = 1;
4367 		else
4368 			*ap->a_retval = 0;
4369 		break;
4370 	case _PC_ACL_PATH_MAX:
4371 		if (NFS_ISV4(vp))
4372 			*ap->a_retval = ACL_MAX_ENTRIES;
4373 		else
4374 			*ap->a_retval = 3;
4375 		break;
4376 	case _PC_PRIO_IO:
4377 		*ap->a_retval = 0;
4378 		break;
4379 	case _PC_SYNC_IO:
4380 		*ap->a_retval = 0;
4381 		break;
4382 	case _PC_ALLOC_SIZE_MIN:
4383 		*ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
4384 		break;
4385 	case _PC_FILESIZEBITS:
4386 		if (NFS_ISV34(vp))
4387 			*ap->a_retval = 64;
4388 		else
4389 			*ap->a_retval = 32;
4390 		break;
4391 	case _PC_REC_INCR_XFER_SIZE:
4392 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4393 		break;
4394 	case _PC_REC_MAX_XFER_SIZE:
4395 		*ap->a_retval = -1; /* means ``unlimited'' */
4396 		break;
4397 	case _PC_REC_MIN_XFER_SIZE:
4398 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4399 		break;
4400 	case _PC_REC_XFER_ALIGN:
4401 		*ap->a_retval = PAGE_SIZE;
4402 		break;
4403 	case _PC_SYMLINK_MAX:
4404 		*ap->a_retval = NFS_MAXPATHLEN;
4405 		break;
4406 	case _PC_MIN_HOLE_SIZE:
4407 		/* Only some NFSv4.2 servers support Seek for Holes. */
4408 		*ap->a_retval = 0;
4409 		nmp = VFSTONFS(vp->v_mount);
4410 		if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) {
4411 			/*
4412 			 * NFSv4.2 doesn't have an attribute for hole size,
4413 			 * so all we can do is see if the Seek operation is
4414 			 * supported and then use f_iosize as a "best guess".
4415 			 */
4416 			mtx_lock(&nmp->nm_mtx);
4417 			if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) {
4418 				mtx_unlock(&nmp->nm_mtx);
4419 				off = 0;
4420 				attrflag = 0;
4421 				error = nfsrpc_seek(vp, &off, &eof,
4422 				    NFSV4CONTENT_HOLE, td->td_ucred, &nfsva,
4423 				    &attrflag);
4424 				if (attrflag != 0)
4425 					nfscl_loadattrcache(&vp, &nfsva,
4426 					    NULL, NULL, 0, 1);
4427 				mtx_lock(&nmp->nm_mtx);
4428 				if (error == NFSERR_NOTSUPP)
4429 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED;
4430 				else
4431 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED |
4432 					    NFSMNTP_SEEK;
4433 				error = 0;
4434 			}
4435 			if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0)
4436 				*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4437 			mtx_unlock(&nmp->nm_mtx);
4438 		}
4439 		break;
4440 
4441 	default:
4442 		error = vop_stdpathconf(ap);
4443 		break;
4444 	}
4445 	return (error);
4446 }
4447