xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision 23c30549affbe0343c01fd78998248956a246252)
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_copy_file_range_t nfs_copy_file_range;
150 static vop_ioctl_t nfs_ioctl;
151 static vop_getextattr_t nfs_getextattr;
152 static vop_setextattr_t nfs_setextattr;
153 static vop_listextattr_t nfs_listextattr;
154 static vop_deleteextattr_t nfs_deleteextattr;
155 static vop_lock1_t	nfs_lock;
156 
157 /*
158  * Global vfs data structures for nfs
159  */
160 
161 static struct vop_vector newnfs_vnodeops_nosig = {
162 	.vop_default =		&default_vnodeops,
163 	.vop_access =		nfs_access,
164 	.vop_advlock =		nfs_advlock,
165 	.vop_advlockasync =	nfs_advlockasync,
166 	.vop_close =		nfs_close,
167 	.vop_create =		nfs_create,
168 	.vop_fsync =		nfs_fsync,
169 	.vop_getattr =		nfs_getattr,
170 	.vop_getpages =		ncl_getpages,
171 	.vop_putpages =		ncl_putpages,
172 	.vop_inactive =		ncl_inactive,
173 	.vop_link =		nfs_link,
174 	.vop_lock1 =		nfs_lock,
175 	.vop_lookup =		nfs_lookup,
176 	.vop_mkdir =		nfs_mkdir,
177 	.vop_mknod =		nfs_mknod,
178 	.vop_open =		nfs_open,
179 	.vop_pathconf =		nfs_pathconf,
180 	.vop_print =		nfs_print,
181 	.vop_read =		nfs_read,
182 	.vop_readdir =		nfs_readdir,
183 	.vop_readlink =		nfs_readlink,
184 	.vop_reclaim =		ncl_reclaim,
185 	.vop_remove =		nfs_remove,
186 	.vop_rename =		nfs_rename,
187 	.vop_rmdir =		nfs_rmdir,
188 	.vop_setattr =		nfs_setattr,
189 	.vop_strategy =		nfs_strategy,
190 	.vop_symlink =		nfs_symlink,
191 	.vop_write =		ncl_write,
192 	.vop_getacl =		nfs_getacl,
193 	.vop_setacl =		nfs_setacl,
194 	.vop_advise =		nfs_advise,
195 	.vop_allocate =		nfs_allocate,
196 	.vop_copy_file_range =	nfs_copy_file_range,
197 	.vop_ioctl =		nfs_ioctl,
198 	.vop_getextattr =	nfs_getextattr,
199 	.vop_setextattr =	nfs_setextattr,
200 	.vop_listextattr =	nfs_listextattr,
201 	.vop_deleteextattr =	nfs_deleteextattr,
202 };
203 
204 static int
205 nfs_vnodeops_bypass(struct vop_generic_args *a)
206 {
207 
208 	return (vop_sigdefer(&newnfs_vnodeops_nosig, a));
209 }
210 
211 struct vop_vector newnfs_vnodeops = {
212 	.vop_default =		&default_vnodeops,
213 	.vop_bypass =		nfs_vnodeops_bypass,
214 };
215 
216 static struct vop_vector newnfs_fifoops_nosig = {
217 	.vop_default =		&fifo_specops,
218 	.vop_access =		nfsspec_access,
219 	.vop_close =		nfsfifo_close,
220 	.vop_fsync =		nfs_fsync,
221 	.vop_getattr =		nfs_getattr,
222 	.vop_inactive =		ncl_inactive,
223 	.vop_pathconf =		nfs_pathconf,
224 	.vop_print =		nfs_print,
225 	.vop_read =		nfsfifo_read,
226 	.vop_reclaim =		ncl_reclaim,
227 	.vop_setattr =		nfs_setattr,
228 	.vop_write =		nfsfifo_write,
229 };
230 
231 static int
232 nfs_fifoops_bypass(struct vop_generic_args *a)
233 {
234 
235 	return (vop_sigdefer(&newnfs_fifoops_nosig, a));
236 }
237 
238 struct vop_vector newnfs_fifoops = {
239 	.vop_default =		&default_vnodeops,
240 	.vop_bypass =		nfs_fifoops_bypass,
241 };
242 
243 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
244     struct componentname *cnp, struct vattr *vap);
245 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
246     int namelen, struct ucred *cred, struct thread *td);
247 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
248     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
249     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
250 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
251     struct componentname *scnp, struct sillyrename *sp);
252 
253 /*
254  * Global variables
255  */
256 SYSCTL_DECL(_vfs_nfs);
257 
258 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
259 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
260 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
261 
262 static int	nfs_prime_access_cache = 0;
263 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
264 	   &nfs_prime_access_cache, 0,
265 	   "Prime NFS ACCESS cache when fetching attributes");
266 
267 static int	newnfs_commit_on_close = 0;
268 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
269     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
270 
271 static int	nfs_clean_pages_on_close = 1;
272 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
273 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
274 
275 int newnfs_directio_enable = 0;
276 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
277 	   &newnfs_directio_enable, 0, "Enable NFS directio");
278 
279 int nfs_keep_dirty_on_error;
280 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
281     &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
282 
283 /*
284  * This sysctl allows other processes to mmap a file that has been opened
285  * O_DIRECT by a process.  In general, having processes mmap the file while
286  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
287  * this by default to prevent DoS attacks - to prevent a malicious user from
288  * opening up files O_DIRECT preventing other users from mmap'ing these
289  * files.  "Protected" environments where stricter consistency guarantees are
290  * required can disable this knob.  The process that opened the file O_DIRECT
291  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
292  * meaningful.
293  */
294 int newnfs_directio_allow_mmap = 1;
295 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
296 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
297 
298 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
299 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
300 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
301 
302 /*
303  * SMP Locking Note :
304  * The list of locks after the description of the lock is the ordering
305  * of other locks acquired with the lock held.
306  * np->n_mtx : Protects the fields in the nfsnode.
307        VM Object Lock
308        VI_MTX (acquired indirectly)
309  * nmp->nm_mtx : Protects the fields in the nfsmount.
310        rep->r_mtx
311  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
312  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
313        nmp->nm_mtx
314        rep->r_mtx
315  * rep->r_mtx : Protects the fields in an nfsreq.
316  */
317 
318 static int
319 nfs_lock(struct vop_lock1_args *ap)
320 {
321 	struct vnode *vp;
322 	struct nfsnode *np;
323 	u_quad_t nsize;
324 	int error, lktype;
325 	bool onfault;
326 
327 	vp = ap->a_vp;
328 	lktype = ap->a_flags & LK_TYPE_MASK;
329 	error = VOP_LOCK1_APV(&default_vnodeops, ap);
330 	if (error != 0 || vp->v_op != &newnfs_vnodeops)
331 		return (error);
332 	np = VTONFS(vp);
333 	if (np == NULL)
334 		return (0);
335 	NFSLOCKNODE(np);
336 	if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED &&
337 	    lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE &&
338 	    lktype != LK_TRYUPGRADE)) {
339 		NFSUNLOCKNODE(np);
340 		return (0);
341 	}
342 	onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT &&
343 	    (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE &&
344 	    (lktype == LK_SHARED || lktype == LK_EXCLUSIVE);
345 	if (onfault && vp->v_vnlock->lk_recurse == 0) {
346 		/*
347 		 * Force retry in vm_fault(), to make the lock request
348 		 * sleepable, which allows us to piggy-back the
349 		 * sleepable call to vnode_pager_setsize().
350 		 */
351 		NFSUNLOCKNODE(np);
352 		VOP_UNLOCK(vp, 0);
353 		return (EBUSY);
354 	}
355 	if ((ap->a_flags & LK_NOWAIT) != 0 ||
356 	    (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) {
357 		NFSUNLOCKNODE(np);
358 		return (0);
359 	}
360 	if (lktype == LK_SHARED) {
361 		NFSUNLOCKNODE(np);
362 		VOP_UNLOCK(vp, 0);
363 		ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
364 		ap->a_flags |= LK_EXCLUSIVE;
365 		error = VOP_LOCK1_APV(&default_vnodeops, ap);
366 		if (error != 0 || vp->v_op != &newnfs_vnodeops)
367 			return (error);
368 		if (vp->v_data == NULL)
369 			goto downgrade;
370 		MPASS(vp->v_data == np);
371 		NFSLOCKNODE(np);
372 		if ((np->n_flag & NVNSETSZSKIP) == 0) {
373 			NFSUNLOCKNODE(np);
374 			goto downgrade;
375 		}
376 	}
377 	np->n_flag &= ~NVNSETSZSKIP;
378 	nsize = np->n_size;
379 	NFSUNLOCKNODE(np);
380 	vnode_pager_setsize(vp, nsize);
381 downgrade:
382 	if (lktype == LK_SHARED) {
383 		ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
384 		ap->a_flags |= LK_DOWNGRADE;
385 		(void)VOP_LOCK1_APV(&default_vnodeops, ap);
386 	}
387 	return (0);
388 }
389 
390 static int
391 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
392     struct ucred *cred, u_int32_t *retmode)
393 {
394 	int error = 0, attrflag, i, lrupos;
395 	u_int32_t rmode;
396 	struct nfsnode *np = VTONFS(vp);
397 	struct nfsvattr nfsva;
398 
399 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
400 	    &rmode, NULL);
401 	if (attrflag)
402 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
403 	if (!error) {
404 		lrupos = 0;
405 		NFSLOCKNODE(np);
406 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
407 			if (np->n_accesscache[i].uid == cred->cr_uid) {
408 				np->n_accesscache[i].mode = rmode;
409 				np->n_accesscache[i].stamp = time_second;
410 				break;
411 			}
412 			if (i > 0 && np->n_accesscache[i].stamp <
413 			    np->n_accesscache[lrupos].stamp)
414 				lrupos = i;
415 		}
416 		if (i == NFS_ACCESSCACHESIZE) {
417 			np->n_accesscache[lrupos].uid = cred->cr_uid;
418 			np->n_accesscache[lrupos].mode = rmode;
419 			np->n_accesscache[lrupos].stamp = time_second;
420 		}
421 		NFSUNLOCKNODE(np);
422 		if (retmode != NULL)
423 			*retmode = rmode;
424 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
425 	} else if (NFS_ISV4(vp)) {
426 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
427 	}
428 #ifdef KDTRACE_HOOKS
429 	if (error != 0)
430 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
431 		    error);
432 #endif
433 	return (error);
434 }
435 
436 /*
437  * nfs access vnode op.
438  * For nfs version 2, just return ok. File accesses may fail later.
439  * For nfs version 3, use the access rpc to check accessibility. If file modes
440  * are changed on the server, accesses might still fail later.
441  */
442 static int
443 nfs_access(struct vop_access_args *ap)
444 {
445 	struct vnode *vp = ap->a_vp;
446 	int error = 0, i, gotahit;
447 	u_int32_t mode, wmode, rmode;
448 	int v34 = NFS_ISV34(vp);
449 	struct nfsnode *np = VTONFS(vp);
450 
451 	/*
452 	 * Disallow write attempts on filesystems mounted read-only;
453 	 * unless the file is a socket, fifo, or a block or character
454 	 * device resident on the filesystem.
455 	 */
456 	if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
457 	    VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
458 	    VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
459 		switch (vp->v_type) {
460 		case VREG:
461 		case VDIR:
462 		case VLNK:
463 			return (EROFS);
464 		default:
465 			break;
466 		}
467 	}
468 	/*
469 	 * For nfs v3 or v4, check to see if we have done this recently, and if
470 	 * so return our cached result instead of making an ACCESS call.
471 	 * If not, do an access rpc, otherwise you are stuck emulating
472 	 * ufs_access() locally using the vattr. This may not be correct,
473 	 * since the server may apply other access criteria such as
474 	 * client uid-->server uid mapping that we do not know about.
475 	 */
476 	if (v34) {
477 		if (ap->a_accmode & VREAD)
478 			mode = NFSACCESS_READ;
479 		else
480 			mode = 0;
481 		if (vp->v_type != VDIR) {
482 			if (ap->a_accmode & VWRITE)
483 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
484 			if (ap->a_accmode & VAPPEND)
485 				mode |= NFSACCESS_EXTEND;
486 			if (ap->a_accmode & VEXEC)
487 				mode |= NFSACCESS_EXECUTE;
488 			if (ap->a_accmode & VDELETE)
489 				mode |= NFSACCESS_DELETE;
490 		} else {
491 			if (ap->a_accmode & VWRITE)
492 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
493 			if (ap->a_accmode & VAPPEND)
494 				mode |= NFSACCESS_EXTEND;
495 			if (ap->a_accmode & VEXEC)
496 				mode |= NFSACCESS_LOOKUP;
497 			if (ap->a_accmode & VDELETE)
498 				mode |= NFSACCESS_DELETE;
499 			if (ap->a_accmode & VDELETE_CHILD)
500 				mode |= NFSACCESS_MODIFY;
501 		}
502 		/* XXX safety belt, only make blanket request if caching */
503 		if (nfsaccess_cache_timeout > 0) {
504 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
505 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
506 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
507 		} else {
508 			wmode = mode;
509 		}
510 
511 		/*
512 		 * Does our cached result allow us to give a definite yes to
513 		 * this request?
514 		 */
515 		gotahit = 0;
516 		NFSLOCKNODE(np);
517 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
518 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
519 			    if (time_second < (np->n_accesscache[i].stamp
520 				+ nfsaccess_cache_timeout) &&
521 				(np->n_accesscache[i].mode & mode) == mode) {
522 				NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
523 				gotahit = 1;
524 			    }
525 			    break;
526 			}
527 		}
528 		NFSUNLOCKNODE(np);
529 #ifdef KDTRACE_HOOKS
530 		if (gotahit != 0)
531 			KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
532 			    ap->a_cred->cr_uid, mode);
533 		else
534 			KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
535 			    ap->a_cred->cr_uid, mode);
536 #endif
537 		if (gotahit == 0) {
538 			/*
539 			 * Either a no, or a don't know.  Go to the wire.
540 			 */
541 			NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
542 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
543 			    ap->a_cred, &rmode);
544 			if (!error &&
545 			    (rmode & mode) != mode)
546 				error = EACCES;
547 		}
548 		return (error);
549 	} else {
550 		if ((error = nfsspec_access(ap)) != 0) {
551 			return (error);
552 		}
553 		/*
554 		 * Attempt to prevent a mapped root from accessing a file
555 		 * which it shouldn't.  We try to read a byte from the file
556 		 * if the user is root and the file is not zero length.
557 		 * After calling nfsspec_access, we should have the correct
558 		 * file size cached.
559 		 */
560 		NFSLOCKNODE(np);
561 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
562 		    && VTONFS(vp)->n_size > 0) {
563 			struct iovec aiov;
564 			struct uio auio;
565 			char buf[1];
566 
567 			NFSUNLOCKNODE(np);
568 			aiov.iov_base = buf;
569 			aiov.iov_len = 1;
570 			auio.uio_iov = &aiov;
571 			auio.uio_iovcnt = 1;
572 			auio.uio_offset = 0;
573 			auio.uio_resid = 1;
574 			auio.uio_segflg = UIO_SYSSPACE;
575 			auio.uio_rw = UIO_READ;
576 			auio.uio_td = ap->a_td;
577 
578 			if (vp->v_type == VREG)
579 				error = ncl_readrpc(vp, &auio, ap->a_cred);
580 			else if (vp->v_type == VDIR) {
581 				char* bp;
582 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
583 				aiov.iov_base = bp;
584 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
585 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
586 				    ap->a_td);
587 				free(bp, M_TEMP);
588 			} else if (vp->v_type == VLNK)
589 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
590 			else
591 				error = EACCES;
592 		} else
593 			NFSUNLOCKNODE(np);
594 		return (error);
595 	}
596 }
597 
598 
599 /*
600  * nfs open vnode op
601  * Check to see if the type is ok
602  * and that deletion is not in progress.
603  * For paged in text files, you will need to flush the page cache
604  * if consistency is lost.
605  */
606 /* ARGSUSED */
607 static int
608 nfs_open(struct vop_open_args *ap)
609 {
610 	struct vnode *vp = ap->a_vp;
611 	struct nfsnode *np = VTONFS(vp);
612 	struct vattr vattr;
613 	int error;
614 	int fmode = ap->a_mode;
615 	struct ucred *cred;
616 	vm_object_t obj;
617 
618 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
619 		return (EOPNOTSUPP);
620 
621 	/*
622 	 * For NFSv4, we need to do the Open Op before cache validation,
623 	 * so that we conform to RFC3530 Sec. 9.3.1.
624 	 */
625 	if (NFS_ISV4(vp)) {
626 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
627 		if (error) {
628 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
629 			    (gid_t)0);
630 			return (error);
631 		}
632 	}
633 
634 	/*
635 	 * Now, if this Open will be doing reading, re-validate/flush the
636 	 * cache, so that Close/Open coherency is maintained.
637 	 */
638 	NFSLOCKNODE(np);
639 	if (np->n_flag & NMODIFIED) {
640 		NFSUNLOCKNODE(np);
641 		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
642 		if (error == EINTR || error == EIO) {
643 			if (NFS_ISV4(vp))
644 				(void) nfsrpc_close(vp, 0, ap->a_td);
645 			return (error);
646 		}
647 		NFSLOCKNODE(np);
648 		np->n_attrstamp = 0;
649 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
650 		if (vp->v_type == VDIR)
651 			np->n_direofoffset = 0;
652 		NFSUNLOCKNODE(np);
653 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
654 		if (error) {
655 			if (NFS_ISV4(vp))
656 				(void) nfsrpc_close(vp, 0, ap->a_td);
657 			return (error);
658 		}
659 		NFSLOCKNODE(np);
660 		np->n_mtime = vattr.va_mtime;
661 		if (NFS_ISV4(vp))
662 			np->n_change = vattr.va_filerev;
663 	} else {
664 		NFSUNLOCKNODE(np);
665 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
666 		if (error) {
667 			if (NFS_ISV4(vp))
668 				(void) nfsrpc_close(vp, 0, ap->a_td);
669 			return (error);
670 		}
671 		NFSLOCKNODE(np);
672 		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
673 		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
674 			if (vp->v_type == VDIR)
675 				np->n_direofoffset = 0;
676 			NFSUNLOCKNODE(np);
677 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
678 			if (error == EINTR || error == EIO) {
679 				if (NFS_ISV4(vp))
680 					(void) nfsrpc_close(vp, 0, ap->a_td);
681 				return (error);
682 			}
683 			NFSLOCKNODE(np);
684 			np->n_mtime = vattr.va_mtime;
685 			if (NFS_ISV4(vp))
686 				np->n_change = vattr.va_filerev;
687 		}
688 	}
689 
690 	/*
691 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
692 	 */
693 	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
694 	    (vp->v_type == VREG)) {
695 		if (np->n_directio_opens == 0) {
696 			NFSUNLOCKNODE(np);
697 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
698 			if (error) {
699 				if (NFS_ISV4(vp))
700 					(void) nfsrpc_close(vp, 0, ap->a_td);
701 				return (error);
702 			}
703 			NFSLOCKNODE(np);
704 			np->n_flag |= NNONCACHE;
705 		}
706 		np->n_directio_opens++;
707 	}
708 
709 	/* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
710 	if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
711 		np->n_flag |= NWRITEOPENED;
712 
713 	/*
714 	 * If this is an open for writing, capture a reference to the
715 	 * credentials, so they can be used by ncl_putpages(). Using
716 	 * these write credentials is preferable to the credentials of
717 	 * whatever thread happens to be doing the VOP_PUTPAGES() since
718 	 * the write RPCs are less likely to fail with EACCES.
719 	 */
720 	if ((fmode & FWRITE) != 0) {
721 		cred = np->n_writecred;
722 		np->n_writecred = crhold(ap->a_cred);
723 	} else
724 		cred = NULL;
725 	NFSUNLOCKNODE(np);
726 
727 	if (cred != NULL)
728 		crfree(cred);
729 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
730 
731 	/*
732 	 * If the text file has been mmap'd, flush any dirty pages to the
733 	 * buffer cache and then...
734 	 * Make sure all writes are pushed to the NFS server.  If this is not
735 	 * done, the modify time of the file can change while the text
736 	 * file is being executed.  This will cause the process that is
737 	 * executing the text file to be terminated.
738 	 */
739 	if (vp->v_writecount <= -1) {
740 		if ((obj = vp->v_object) != NULL &&
741 		    vm_object_mightbedirty(obj)) {
742 			VM_OBJECT_WLOCK(obj);
743 			vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
744 			VM_OBJECT_WUNLOCK(obj);
745 		}
746 
747 		/* Now, flush the buffer cache. */
748 		ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
749 
750 		/* And, finally, make sure that n_mtime is up to date. */
751 		np = VTONFS(vp);
752 		NFSLOCKNODE(np);
753 		np->n_mtime = np->n_vattr.na_mtime;
754 		NFSUNLOCKNODE(np);
755 	}
756 	return (0);
757 }
758 
759 /*
760  * nfs close vnode op
761  * What an NFS client should do upon close after writing is a debatable issue.
762  * Most NFS clients push delayed writes to the server upon close, basically for
763  * two reasons:
764  * 1 - So that any write errors may be reported back to the client process
765  *     doing the close system call. By far the two most likely errors are
766  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
767  * 2 - To put a worst case upper bound on cache inconsistency between
768  *     multiple clients for the file.
769  * There is also a consistency problem for Version 2 of the protocol w.r.t.
770  * not being able to tell if other clients are writing a file concurrently,
771  * since there is no way of knowing if the changed modify time in the reply
772  * is only due to the write for this client.
773  * (NFS Version 3 provides weak cache consistency data in the reply that
774  *  should be sufficient to detect and handle this case.)
775  *
776  * The current code does the following:
777  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
778  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
779  *                     or commit them (this satisfies 1 and 2 except for the
780  *                     case where the server crashes after this close but
781  *                     before the commit RPC, which is felt to be "good
782  *                     enough". Changing the last argument to ncl_flush() to
783  *                     a 1 would force a commit operation, if it is felt a
784  *                     commit is necessary now.
785  * for NFS Version 4 - flush the dirty buffers and commit them, if
786  *		       nfscl_mustflush() says this is necessary.
787  *                     It is necessary if there is no write delegation held,
788  *                     in order to satisfy open/close coherency.
789  *                     If the file isn't cached on local stable storage,
790  *                     it may be necessary in order to detect "out of space"
791  *                     errors from the server, if the write delegation
792  *                     issued by the server doesn't allow the file to grow.
793  */
794 /* ARGSUSED */
795 static int
796 nfs_close(struct vop_close_args *ap)
797 {
798 	struct vnode *vp = ap->a_vp;
799 	struct nfsnode *np = VTONFS(vp);
800 	struct nfsvattr nfsva;
801 	struct ucred *cred;
802 	int error = 0, ret, localcred = 0;
803 	int fmode = ap->a_fflag;
804 
805 	if (NFSCL_FORCEDISM(vp->v_mount))
806 		return (0);
807 	/*
808 	 * During shutdown, a_cred isn't valid, so just use root.
809 	 */
810 	if (ap->a_cred == NOCRED) {
811 		cred = newnfs_getcred();
812 		localcred = 1;
813 	} else {
814 		cred = ap->a_cred;
815 	}
816 	if (vp->v_type == VREG) {
817 	    /*
818 	     * Examine and clean dirty pages, regardless of NMODIFIED.
819 	     * This closes a major hole in close-to-open consistency.
820 	     * We want to push out all dirty pages (and buffers) on
821 	     * close, regardless of whether they were dirtied by
822 	     * mmap'ed writes or via write().
823 	     */
824 	    if (nfs_clean_pages_on_close && vp->v_object) {
825 		VM_OBJECT_WLOCK(vp->v_object);
826 		vm_object_page_clean(vp->v_object, 0, 0, 0);
827 		VM_OBJECT_WUNLOCK(vp->v_object);
828 	    }
829 	    NFSLOCKNODE(np);
830 	    if (np->n_flag & NMODIFIED) {
831 		NFSUNLOCKNODE(np);
832 		if (NFS_ISV3(vp)) {
833 		    /*
834 		     * Under NFSv3 we have dirty buffers to dispose of.  We
835 		     * must flush them to the NFS server.  We have the option
836 		     * of waiting all the way through the commit rpc or just
837 		     * waiting for the initial write.  The default is to only
838 		     * wait through the initial write so the data is in the
839 		     * server's cache, which is roughly similar to the state
840 		     * a standard disk subsystem leaves the file in on close().
841 		     *
842 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
843 		     * potential races with other processes, and certainly
844 		     * cannot clear it if we don't commit.
845 		     * These races occur when there is no longer the old
846 		     * traditional vnode locking implemented for Vnode Ops.
847 		     */
848 		    int cm = newnfs_commit_on_close ? 1 : 0;
849 		    error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
850 		    /* np->n_flag &= ~NMODIFIED; */
851 		} else if (NFS_ISV4(vp)) {
852 			if (nfscl_mustflush(vp) != 0) {
853 				int cm = newnfs_commit_on_close ? 1 : 0;
854 				error = ncl_flush(vp, MNT_WAIT, ap->a_td,
855 				    cm, 0);
856 				/*
857 				 * as above w.r.t races when clearing
858 				 * NMODIFIED.
859 				 * np->n_flag &= ~NMODIFIED;
860 				 */
861 			}
862 		} else {
863 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
864 		}
865 		NFSLOCKNODE(np);
866 	    }
867  	    /*
868  	     * Invalidate the attribute cache in all cases.
869  	     * An open is going to fetch fresh attrs any way, other procs
870  	     * on this node that have file open will be forced to do an
871  	     * otw attr fetch, but this is safe.
872 	     * --> A user found that their RPC count dropped by 20% when
873 	     *     this was commented out and I can't see any requirement
874 	     *     for it, so I've disabled it when negative lookups are
875 	     *     enabled. (What does this have to do with negative lookup
876 	     *     caching? Well nothing, except it was reported by the
877 	     *     same user that needed negative lookup caching and I wanted
878 	     *     there to be a way to disable it to see if it
879 	     *     is the cause of some caching/coherency issue that might
880 	     *     crop up.)
881  	     */
882 	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
883 		    np->n_attrstamp = 0;
884 		    KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
885 	    }
886 	    if (np->n_flag & NWRITEERR) {
887 		np->n_flag &= ~NWRITEERR;
888 		error = np->n_error;
889 	    }
890 	    NFSUNLOCKNODE(np);
891 	}
892 
893 	if (NFS_ISV4(vp)) {
894 		/*
895 		 * Get attributes so "change" is up to date.
896 		 */
897 		if (error == 0 && nfscl_mustflush(vp) != 0 &&
898 		    vp->v_type == VREG &&
899 		    (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
900 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
901 			    NULL);
902 			if (!ret) {
903 				np->n_change = nfsva.na_filerev;
904 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
905 				    NULL, 0, 0);
906 			}
907 		}
908 
909 		/*
910 		 * and do the close.
911 		 */
912 		ret = nfsrpc_close(vp, 0, ap->a_td);
913 		if (!error && ret)
914 			error = ret;
915 		if (error)
916 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
917 			    (gid_t)0);
918 	}
919 	if (newnfs_directio_enable)
920 		KASSERT((np->n_directio_asyncwr == 0),
921 			("nfs_close: dirty unflushed (%d) directio buffers\n",
922 			 np->n_directio_asyncwr));
923 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
924 		NFSLOCKNODE(np);
925 		KASSERT((np->n_directio_opens > 0),
926 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
927 		np->n_directio_opens--;
928 		if (np->n_directio_opens == 0)
929 			np->n_flag &= ~NNONCACHE;
930 		NFSUNLOCKNODE(np);
931 	}
932 	if (localcred)
933 		NFSFREECRED(cred);
934 	return (error);
935 }
936 
937 /*
938  * nfs getattr call from vfs.
939  */
940 static int
941 nfs_getattr(struct vop_getattr_args *ap)
942 {
943 	struct vnode *vp = ap->a_vp;
944 	struct thread *td = curthread;	/* XXX */
945 	struct nfsnode *np = VTONFS(vp);
946 	int error = 0;
947 	struct nfsvattr nfsva;
948 	struct vattr *vap = ap->a_vap;
949 	struct vattr vattr;
950 
951 	/*
952 	 * Update local times for special files.
953 	 */
954 	NFSLOCKNODE(np);
955 	if (np->n_flag & (NACC | NUPD))
956 		np->n_flag |= NCHG;
957 	NFSUNLOCKNODE(np);
958 	/*
959 	 * First look in the cache.
960 	 */
961 	if (ncl_getattrcache(vp, &vattr) == 0) {
962 		vap->va_type = vattr.va_type;
963 		vap->va_mode = vattr.va_mode;
964 		vap->va_nlink = vattr.va_nlink;
965 		vap->va_uid = vattr.va_uid;
966 		vap->va_gid = vattr.va_gid;
967 		vap->va_fsid = vattr.va_fsid;
968 		vap->va_fileid = vattr.va_fileid;
969 		vap->va_size = vattr.va_size;
970 		vap->va_blocksize = vattr.va_blocksize;
971 		vap->va_atime = vattr.va_atime;
972 		vap->va_mtime = vattr.va_mtime;
973 		vap->va_ctime = vattr.va_ctime;
974 		vap->va_gen = vattr.va_gen;
975 		vap->va_flags = vattr.va_flags;
976 		vap->va_rdev = vattr.va_rdev;
977 		vap->va_bytes = vattr.va_bytes;
978 		vap->va_filerev = vattr.va_filerev;
979 		/*
980 		 * Get the local modify time for the case of a write
981 		 * delegation.
982 		 */
983 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
984 		return (0);
985 	}
986 
987 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
988 	    nfsaccess_cache_timeout > 0) {
989 		NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
990 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
991 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
992 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
993 			return (0);
994 		}
995 	}
996 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
997 	if (!error)
998 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
999 	if (!error) {
1000 		/*
1001 		 * Get the local modify time for the case of a write
1002 		 * delegation.
1003 		 */
1004 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
1005 	} else if (NFS_ISV4(vp)) {
1006 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1007 	}
1008 	return (error);
1009 }
1010 
1011 /*
1012  * nfs setattr call.
1013  */
1014 static int
1015 nfs_setattr(struct vop_setattr_args *ap)
1016 {
1017 	struct vnode *vp = ap->a_vp;
1018 	struct nfsnode *np = VTONFS(vp);
1019 	struct thread *td = curthread;	/* XXX */
1020 	struct vattr *vap = ap->a_vap;
1021 	int error = 0;
1022 	u_quad_t tsize;
1023 
1024 #ifndef nolint
1025 	tsize = (u_quad_t)0;
1026 #endif
1027 
1028 	/*
1029 	 * Setting of flags and marking of atimes are not supported.
1030 	 */
1031 	if (vap->va_flags != VNOVAL)
1032 		return (EOPNOTSUPP);
1033 
1034 	/*
1035 	 * Disallow write attempts if the filesystem is mounted read-only.
1036 	 */
1037   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
1038 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
1039 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
1040 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
1041 		return (EROFS);
1042 	if (vap->va_size != VNOVAL) {
1043  		switch (vp->v_type) {
1044  		case VDIR:
1045  			return (EISDIR);
1046  		case VCHR:
1047  		case VBLK:
1048  		case VSOCK:
1049  		case VFIFO:
1050 			if (vap->va_mtime.tv_sec == VNOVAL &&
1051 			    vap->va_atime.tv_sec == VNOVAL &&
1052 			    vap->va_mode == (mode_t)VNOVAL &&
1053 			    vap->va_uid == (uid_t)VNOVAL &&
1054 			    vap->va_gid == (gid_t)VNOVAL)
1055 				return (0);
1056  			vap->va_size = VNOVAL;
1057  			break;
1058  		default:
1059 			/*
1060 			 * Disallow write attempts if the filesystem is
1061 			 * mounted read-only.
1062 			 */
1063 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
1064 				return (EROFS);
1065 			/*
1066 			 *  We run vnode_pager_setsize() early (why?),
1067 			 * we must set np->n_size now to avoid vinvalbuf
1068 			 * V_SAVE races that might setsize a lower
1069 			 * value.
1070 			 */
1071 			NFSLOCKNODE(np);
1072 			tsize = np->n_size;
1073 			NFSUNLOCKNODE(np);
1074 			error = ncl_meta_setsize(vp, td, vap->va_size);
1075 			NFSLOCKNODE(np);
1076  			if (np->n_flag & NMODIFIED) {
1077 			    tsize = np->n_size;
1078 			    NFSUNLOCKNODE(np);
1079 			    error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
1080 			        0 : V_SAVE, td, 1);
1081 			    if (error != 0) {
1082 				    vnode_pager_setsize(vp, tsize);
1083 				    return (error);
1084 			    }
1085 			    /*
1086 			     * Call nfscl_delegmodtime() to set the modify time
1087 			     * locally, as required.
1088 			     */
1089 			    nfscl_delegmodtime(vp);
1090  			} else
1091 			    NFSUNLOCKNODE(np);
1092 			/*
1093 			 * np->n_size has already been set to vap->va_size
1094 			 * in ncl_meta_setsize(). We must set it again since
1095 			 * nfs_loadattrcache() could be called through
1096 			 * ncl_meta_setsize() and could modify np->n_size.
1097 			 */
1098 			NFSLOCKNODE(np);
1099  			np->n_vattr.na_size = np->n_size = vap->va_size;
1100 			NFSUNLOCKNODE(np);
1101   		}
1102   	} else {
1103 		NFSLOCKNODE(np);
1104 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
1105 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
1106 			NFSUNLOCKNODE(np);
1107 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
1108 			if (error == EINTR || error == EIO)
1109 				return (error);
1110 		} else
1111 			NFSUNLOCKNODE(np);
1112 	}
1113 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
1114 	if (error && vap->va_size != VNOVAL) {
1115 		NFSLOCKNODE(np);
1116 		np->n_size = np->n_vattr.na_size = tsize;
1117 		vnode_pager_setsize(vp, tsize);
1118 		NFSUNLOCKNODE(np);
1119 	}
1120 	return (error);
1121 }
1122 
1123 /*
1124  * Do an nfs setattr rpc.
1125  */
1126 static int
1127 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1128     struct thread *td)
1129 {
1130 	struct nfsnode *np = VTONFS(vp);
1131 	int error, ret, attrflag, i;
1132 	struct nfsvattr nfsva;
1133 
1134 	if (NFS_ISV34(vp)) {
1135 		NFSLOCKNODE(np);
1136 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1137 			np->n_accesscache[i].stamp = 0;
1138 		np->n_flag |= NDELEGMOD;
1139 		NFSUNLOCKNODE(np);
1140 		KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1141 	}
1142 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1143 	    NULL);
1144 	if (attrflag) {
1145 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1146 		if (ret && !error)
1147 			error = ret;
1148 	}
1149 	if (error && NFS_ISV4(vp))
1150 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1151 	return (error);
1152 }
1153 
1154 /*
1155  * nfs lookup call, one step at a time...
1156  * First look in cache
1157  * If not found, unlock the directory nfsnode and do the rpc
1158  */
1159 static int
1160 nfs_lookup(struct vop_lookup_args *ap)
1161 {
1162 	struct componentname *cnp = ap->a_cnp;
1163 	struct vnode *dvp = ap->a_dvp;
1164 	struct vnode **vpp = ap->a_vpp;
1165 	struct mount *mp = dvp->v_mount;
1166 	int flags = cnp->cn_flags;
1167 	struct vnode *newvp;
1168 	struct nfsmount *nmp;
1169 	struct nfsnode *np, *newnp;
1170 	int error = 0, attrflag, dattrflag, ltype, ncticks;
1171 	struct thread *td = cnp->cn_thread;
1172 	struct nfsfh *nfhp;
1173 	struct nfsvattr dnfsva, nfsva;
1174 	struct vattr vattr;
1175 	struct timespec nctime;
1176 
1177 	*vpp = NULLVP;
1178 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1179 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1180 		return (EROFS);
1181 	if (dvp->v_type != VDIR)
1182 		return (ENOTDIR);
1183 	nmp = VFSTONFS(mp);
1184 	np = VTONFS(dvp);
1185 
1186 	/* For NFSv4, wait until any remove is done. */
1187 	NFSLOCKNODE(np);
1188 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1189 		np->n_flag |= NREMOVEWANT;
1190 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1191 	}
1192 	NFSUNLOCKNODE(np);
1193 
1194 	if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1195 		return (error);
1196 	error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1197 	if (error > 0 && error != ENOENT)
1198 		return (error);
1199 	if (error == -1) {
1200 		/*
1201 		 * Lookups of "." are special and always return the
1202 		 * current directory.  cache_lookup() already handles
1203 		 * associated locking bookkeeping, etc.
1204 		 */
1205 		if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1206 			/* XXX: Is this really correct? */
1207 			if (cnp->cn_nameiop != LOOKUP &&
1208 			    (flags & ISLASTCN))
1209 				cnp->cn_flags |= SAVENAME;
1210 			return (0);
1211 		}
1212 
1213 		/*
1214 		 * We only accept a positive hit in the cache if the
1215 		 * change time of the file matches our cached copy.
1216 		 * Otherwise, we discard the cache entry and fallback
1217 		 * to doing a lookup RPC.  We also only trust cache
1218 		 * entries for less than nm_nametimeo seconds.
1219 		 *
1220 		 * To better handle stale file handles and attributes,
1221 		 * clear the attribute cache of this node if it is a
1222 		 * leaf component, part of an open() call, and not
1223 		 * locally modified before fetching the attributes.
1224 		 * This should allow stale file handles to be detected
1225 		 * here where we can fall back to a LOOKUP RPC to
1226 		 * recover rather than having nfs_open() detect the
1227 		 * stale file handle and failing open(2) with ESTALE.
1228 		 */
1229 		newvp = *vpp;
1230 		newnp = VTONFS(newvp);
1231 		if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1232 		    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1233 		    !(newnp->n_flag & NMODIFIED)) {
1234 			NFSLOCKNODE(newnp);
1235 			newnp->n_attrstamp = 0;
1236 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1237 			NFSUNLOCKNODE(newnp);
1238 		}
1239 		if (nfscl_nodeleg(newvp, 0) == 0 ||
1240 		    ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1241 		    VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1242 		    timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1243 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1244 			if (cnp->cn_nameiop != LOOKUP &&
1245 			    (flags & ISLASTCN))
1246 				cnp->cn_flags |= SAVENAME;
1247 			return (0);
1248 		}
1249 		cache_purge(newvp);
1250 		if (dvp != newvp)
1251 			vput(newvp);
1252 		else
1253 			vrele(newvp);
1254 		*vpp = NULLVP;
1255 	} else if (error == ENOENT) {
1256 		if (VN_IS_DOOMED(dvp))
1257 			return (ENOENT);
1258 		/*
1259 		 * We only accept a negative hit in the cache if the
1260 		 * modification time of the parent directory matches
1261 		 * the cached copy in the name cache entry.
1262 		 * Otherwise, we discard all of the negative cache
1263 		 * entries for this directory.  We also only trust
1264 		 * negative cache entries for up to nm_negnametimeo
1265 		 * seconds.
1266 		 */
1267 		if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1268 		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1269 		    timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1270 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1271 			return (ENOENT);
1272 		}
1273 		cache_purge_negative(dvp);
1274 	}
1275 
1276 	newvp = NULLVP;
1277 	NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1278 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1279 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1280 	    NULL);
1281 	if (dattrflag)
1282 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1283 	if (error) {
1284 		if (newvp != NULLVP) {
1285 			vput(newvp);
1286 			*vpp = NULLVP;
1287 		}
1288 
1289 		if (error != ENOENT) {
1290 			if (NFS_ISV4(dvp))
1291 				error = nfscl_maperr(td, error, (uid_t)0,
1292 				    (gid_t)0);
1293 			return (error);
1294 		}
1295 
1296 		/* The requested file was not found. */
1297 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1298 		    (flags & ISLASTCN)) {
1299 			/*
1300 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1301 			 * VWRITE) here instead of just checking
1302 			 * MNT_RDONLY.
1303 			 */
1304 			if (mp->mnt_flag & MNT_RDONLY)
1305 				return (EROFS);
1306 			cnp->cn_flags |= SAVENAME;
1307 			return (EJUSTRETURN);
1308 		}
1309 
1310 		if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1311 			/*
1312 			 * Cache the modification time of the parent
1313 			 * directory from the post-op attributes in
1314 			 * the name cache entry.  The negative cache
1315 			 * entry will be ignored once the directory
1316 			 * has changed.  Don't bother adding the entry
1317 			 * if the directory has already changed.
1318 			 */
1319 			NFSLOCKNODE(np);
1320 			if (timespeccmp(&np->n_vattr.na_mtime,
1321 			    &dnfsva.na_mtime, ==)) {
1322 				NFSUNLOCKNODE(np);
1323 				cache_enter_time(dvp, NULL, cnp,
1324 				    &dnfsva.na_mtime, NULL);
1325 			} else
1326 				NFSUNLOCKNODE(np);
1327 		}
1328 		return (ENOENT);
1329 	}
1330 
1331 	/*
1332 	 * Handle RENAME case...
1333 	 */
1334 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1335 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1336 			free(nfhp, M_NFSFH);
1337 			return (EISDIR);
1338 		}
1339 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1340 		    LK_EXCLUSIVE);
1341 		if (error)
1342 			return (error);
1343 		newvp = NFSTOV(np);
1344 		if (attrflag)
1345 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1346 			    0, 1);
1347 		*vpp = newvp;
1348 		cnp->cn_flags |= SAVENAME;
1349 		return (0);
1350 	}
1351 
1352 	if (flags & ISDOTDOT) {
1353 		ltype = NFSVOPISLOCKED(dvp);
1354 		error = vfs_busy(mp, MBF_NOWAIT);
1355 		if (error != 0) {
1356 			vfs_ref(mp);
1357 			NFSVOPUNLOCK(dvp, 0);
1358 			error = vfs_busy(mp, 0);
1359 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1360 			vfs_rel(mp);
1361 			if (error == 0 && VN_IS_DOOMED(dvp)) {
1362 				vfs_unbusy(mp);
1363 				error = ENOENT;
1364 			}
1365 			if (error != 0)
1366 				return (error);
1367 		}
1368 		NFSVOPUNLOCK(dvp, 0);
1369 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1370 		    cnp->cn_lkflags);
1371 		if (error == 0)
1372 			newvp = NFSTOV(np);
1373 		vfs_unbusy(mp);
1374 		if (newvp != dvp)
1375 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1376 		if (VN_IS_DOOMED(dvp)) {
1377 			if (error == 0) {
1378 				if (newvp == dvp)
1379 					vrele(newvp);
1380 				else
1381 					vput(newvp);
1382 			}
1383 			error = ENOENT;
1384 		}
1385 		if (error != 0)
1386 			return (error);
1387 		if (attrflag)
1388 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1389 			    0, 1);
1390 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1391 		free(nfhp, M_NFSFH);
1392 		VREF(dvp);
1393 		newvp = dvp;
1394 		if (attrflag)
1395 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1396 			    0, 1);
1397 	} else {
1398 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1399 		    cnp->cn_lkflags);
1400 		if (error)
1401 			return (error);
1402 		newvp = NFSTOV(np);
1403 		if (attrflag)
1404 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1405 			    0, 1);
1406 		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1407 		    !(np->n_flag & NMODIFIED)) {
1408 			/*
1409 			 * Flush the attribute cache when opening a
1410 			 * leaf node to ensure that fresh attributes
1411 			 * are fetched in nfs_open() since we did not
1412 			 * fetch attributes from the LOOKUP reply.
1413 			 */
1414 			NFSLOCKNODE(np);
1415 			np->n_attrstamp = 0;
1416 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1417 			NFSUNLOCKNODE(np);
1418 		}
1419 	}
1420 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1421 		cnp->cn_flags |= SAVENAME;
1422 	if ((cnp->cn_flags & MAKEENTRY) &&
1423 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1424 	    attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1425 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1426 		    newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1427 	*vpp = newvp;
1428 	return (0);
1429 }
1430 
1431 /*
1432  * nfs read call.
1433  * Just call ncl_bioread() to do the work.
1434  */
1435 static int
1436 nfs_read(struct vop_read_args *ap)
1437 {
1438 	struct vnode *vp = ap->a_vp;
1439 
1440 	switch (vp->v_type) {
1441 	case VREG:
1442 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1443 	case VDIR:
1444 		return (EISDIR);
1445 	default:
1446 		return (EOPNOTSUPP);
1447 	}
1448 }
1449 
1450 /*
1451  * nfs readlink call
1452  */
1453 static int
1454 nfs_readlink(struct vop_readlink_args *ap)
1455 {
1456 	struct vnode *vp = ap->a_vp;
1457 
1458 	if (vp->v_type != VLNK)
1459 		return (EINVAL);
1460 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1461 }
1462 
1463 /*
1464  * Do a readlink rpc.
1465  * Called by ncl_doio() from below the buffer cache.
1466  */
1467 int
1468 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1469 {
1470 	int error, ret, attrflag;
1471 	struct nfsvattr nfsva;
1472 
1473 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1474 	    &attrflag, NULL);
1475 	if (attrflag) {
1476 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1477 		if (ret && !error)
1478 			error = ret;
1479 	}
1480 	if (error && NFS_ISV4(vp))
1481 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1482 	return (error);
1483 }
1484 
1485 /*
1486  * nfs read rpc call
1487  * Ditto above
1488  */
1489 int
1490 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1491 {
1492 	int error, ret, attrflag;
1493 	struct nfsvattr nfsva;
1494 	struct nfsmount *nmp;
1495 
1496 	nmp = VFSTONFS(vnode_mount(vp));
1497 	error = EIO;
1498 	attrflag = 0;
1499 	if (NFSHASPNFS(nmp))
1500 		error = nfscl_doiods(vp, uiop, NULL, NULL,
1501 		    NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1502 	NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1503 	if (error != 0)
1504 		error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1505 		    &attrflag, NULL);
1506 	if (attrflag) {
1507 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1508 		if (ret && !error)
1509 			error = ret;
1510 	}
1511 	if (error && NFS_ISV4(vp))
1512 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1513 	return (error);
1514 }
1515 
1516 /*
1517  * nfs write call
1518  */
1519 int
1520 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1521     int *iomode, int *must_commit, int called_from_strategy)
1522 {
1523 	struct nfsvattr nfsva;
1524 	int error, attrflag, ret;
1525 	struct nfsmount *nmp;
1526 
1527 	nmp = VFSTONFS(vnode_mount(vp));
1528 	error = EIO;
1529 	attrflag = 0;
1530 	if (NFSHASPNFS(nmp))
1531 		error = nfscl_doiods(vp, uiop, iomode, must_commit,
1532 		    NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1533 	NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1534 	if (error != 0)
1535 		error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1536 		    uiop->uio_td, &nfsva, &attrflag, NULL,
1537 		    called_from_strategy);
1538 	if (attrflag) {
1539 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1540 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1541 			    1);
1542 		else
1543 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1544 			    1);
1545 		if (ret && !error)
1546 			error = ret;
1547 	}
1548 	if (DOINGASYNC(vp))
1549 		*iomode = NFSWRITE_FILESYNC;
1550 	if (error && NFS_ISV4(vp))
1551 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1552 	return (error);
1553 }
1554 
1555 /*
1556  * nfs mknod rpc
1557  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1558  * mode set to specify the file type and the size field for rdev.
1559  */
1560 static int
1561 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1562     struct vattr *vap)
1563 {
1564 	struct nfsvattr nfsva, dnfsva;
1565 	struct vnode *newvp = NULL;
1566 	struct nfsnode *np = NULL, *dnp;
1567 	struct nfsfh *nfhp;
1568 	struct vattr vattr;
1569 	int error = 0, attrflag, dattrflag;
1570 	u_int32_t rdev;
1571 
1572 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1573 		rdev = vap->va_rdev;
1574 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1575 		rdev = 0xffffffff;
1576 	else
1577 		return (EOPNOTSUPP);
1578 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1579 		return (error);
1580 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1581 	    rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1582 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1583 	if (!error) {
1584 		if (!nfhp)
1585 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1586 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1587 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1588 			    NULL);
1589 		if (nfhp)
1590 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1591 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1592 	}
1593 	if (dattrflag)
1594 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1595 	if (!error) {
1596 		newvp = NFSTOV(np);
1597 		if (attrflag != 0) {
1598 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1599 			    0, 1);
1600 			if (error != 0)
1601 				vput(newvp);
1602 		}
1603 	}
1604 	if (!error) {
1605 		*vpp = newvp;
1606 	} else if (NFS_ISV4(dvp)) {
1607 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1608 		    vap->va_gid);
1609 	}
1610 	dnp = VTONFS(dvp);
1611 	NFSLOCKNODE(dnp);
1612 	dnp->n_flag |= NMODIFIED;
1613 	if (!dattrflag) {
1614 		dnp->n_attrstamp = 0;
1615 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1616 	}
1617 	NFSUNLOCKNODE(dnp);
1618 	return (error);
1619 }
1620 
1621 /*
1622  * nfs mknod vop
1623  * just call nfs_mknodrpc() to do the work.
1624  */
1625 /* ARGSUSED */
1626 static int
1627 nfs_mknod(struct vop_mknod_args *ap)
1628 {
1629 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1630 }
1631 
1632 static struct mtx nfs_cverf_mtx;
1633 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1634     MTX_DEF);
1635 
1636 static nfsquad_t
1637 nfs_get_cverf(void)
1638 {
1639 	static nfsquad_t cverf;
1640 	nfsquad_t ret;
1641 	static int cverf_initialized = 0;
1642 
1643 	mtx_lock(&nfs_cverf_mtx);
1644 	if (cverf_initialized == 0) {
1645 		cverf.lval[0] = arc4random();
1646 		cverf.lval[1] = arc4random();
1647 		cverf_initialized = 1;
1648 	} else
1649 		cverf.qval++;
1650 	ret = cverf;
1651 	mtx_unlock(&nfs_cverf_mtx);
1652 
1653 	return (ret);
1654 }
1655 
1656 /*
1657  * nfs file create call
1658  */
1659 static int
1660 nfs_create(struct vop_create_args *ap)
1661 {
1662 	struct vnode *dvp = ap->a_dvp;
1663 	struct vattr *vap = ap->a_vap;
1664 	struct componentname *cnp = ap->a_cnp;
1665 	struct nfsnode *np = NULL, *dnp;
1666 	struct vnode *newvp = NULL;
1667 	struct nfsmount *nmp;
1668 	struct nfsvattr dnfsva, nfsva;
1669 	struct nfsfh *nfhp;
1670 	nfsquad_t cverf;
1671 	int error = 0, attrflag, dattrflag, fmode = 0;
1672 	struct vattr vattr;
1673 
1674 	/*
1675 	 * Oops, not for me..
1676 	 */
1677 	if (vap->va_type == VSOCK)
1678 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1679 
1680 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1681 		return (error);
1682 	if (vap->va_vaflags & VA_EXCLUSIVE)
1683 		fmode |= O_EXCL;
1684 	dnp = VTONFS(dvp);
1685 	nmp = VFSTONFS(vnode_mount(dvp));
1686 again:
1687 	/* For NFSv4, wait until any remove is done. */
1688 	NFSLOCKNODE(dnp);
1689 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1690 		dnp->n_flag |= NREMOVEWANT;
1691 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1692 	}
1693 	NFSUNLOCKNODE(dnp);
1694 
1695 	cverf = nfs_get_cverf();
1696 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1697 	    vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1698 	    &nfhp, &attrflag, &dattrflag, NULL);
1699 	if (!error) {
1700 		if (nfhp == NULL)
1701 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1702 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1703 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1704 			    NULL);
1705 		if (nfhp != NULL)
1706 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1707 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1708 	}
1709 	if (dattrflag)
1710 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1711 	if (!error) {
1712 		newvp = NFSTOV(np);
1713 		if (attrflag == 0)
1714 			error = nfsrpc_getattr(newvp, cnp->cn_cred,
1715 			    cnp->cn_thread, &nfsva, NULL);
1716 		if (error == 0)
1717 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1718 			    0, 1);
1719 	}
1720 	if (error) {
1721 		if (newvp != NULL) {
1722 			vput(newvp);
1723 			newvp = NULL;
1724 		}
1725 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1726 		    error == NFSERR_NOTSUPP) {
1727 			fmode &= ~O_EXCL;
1728 			goto again;
1729 		}
1730 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1731 		if (nfscl_checksattr(vap, &nfsva)) {
1732 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1733 			    cnp->cn_thread, &nfsva, &attrflag, NULL);
1734 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1735 			    vap->va_gid != (gid_t)VNOVAL)) {
1736 				/* try again without setting uid/gid */
1737 				vap->va_uid = (uid_t)VNOVAL;
1738 				vap->va_gid = (uid_t)VNOVAL;
1739 				error = nfsrpc_setattr(newvp, vap, NULL,
1740 				    cnp->cn_cred, cnp->cn_thread, &nfsva,
1741 				    &attrflag, NULL);
1742 			}
1743 			if (attrflag)
1744 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1745 				    NULL, 0, 1);
1746 			if (error != 0)
1747 				vput(newvp);
1748 		}
1749 	}
1750 	if (!error) {
1751 		if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1752 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1753 			    NULL);
1754 		*ap->a_vpp = newvp;
1755 	} else if (NFS_ISV4(dvp)) {
1756 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1757 		    vap->va_gid);
1758 	}
1759 	NFSLOCKNODE(dnp);
1760 	dnp->n_flag |= NMODIFIED;
1761 	if (!dattrflag) {
1762 		dnp->n_attrstamp = 0;
1763 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1764 	}
1765 	NFSUNLOCKNODE(dnp);
1766 	return (error);
1767 }
1768 
1769 /*
1770  * nfs file remove call
1771  * To try and make nfs semantics closer to ufs semantics, a file that has
1772  * other processes using the vnode is renamed instead of removed and then
1773  * removed later on the last close.
1774  * - If v_usecount > 1
1775  *	  If a rename is not already in the works
1776  *	     call nfs_sillyrename() to set it up
1777  *     else
1778  *	  do the remove rpc
1779  */
1780 static int
1781 nfs_remove(struct vop_remove_args *ap)
1782 {
1783 	struct vnode *vp = ap->a_vp;
1784 	struct vnode *dvp = ap->a_dvp;
1785 	struct componentname *cnp = ap->a_cnp;
1786 	struct nfsnode *np = VTONFS(vp);
1787 	int error = 0;
1788 	struct vattr vattr;
1789 
1790 	KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1791 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1792 	if (vp->v_type == VDIR)
1793 		error = EPERM;
1794 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1795 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1796 	    vattr.va_nlink > 1)) {
1797 		/*
1798 		 * Purge the name cache so that the chance of a lookup for
1799 		 * the name succeeding while the remove is in progress is
1800 		 * minimized. Without node locking it can still happen, such
1801 		 * that an I/O op returns ESTALE, but since you get this if
1802 		 * another host removes the file..
1803 		 */
1804 		cache_purge(vp);
1805 		/*
1806 		 * throw away biocache buffers, mainly to avoid
1807 		 * unnecessary delayed writes later.
1808 		 */
1809 		error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1810 		if (error != EINTR && error != EIO)
1811 			/* Do the rpc */
1812 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1813 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1814 		/*
1815 		 * Kludge City: If the first reply to the remove rpc is lost..
1816 		 *   the reply to the retransmitted request will be ENOENT
1817 		 *   since the file was in fact removed
1818 		 *   Therefore, we cheat and return success.
1819 		 */
1820 		if (error == ENOENT)
1821 			error = 0;
1822 	} else if (!np->n_sillyrename)
1823 		error = nfs_sillyrename(dvp, vp, cnp);
1824 	NFSLOCKNODE(np);
1825 	np->n_attrstamp = 0;
1826 	NFSUNLOCKNODE(np);
1827 	KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1828 	return (error);
1829 }
1830 
1831 /*
1832  * nfs file remove rpc called from nfs_inactive
1833  */
1834 int
1835 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1836 {
1837 	/*
1838 	 * Make sure that the directory vnode is still valid.
1839 	 * XXX we should lock sp->s_dvp here.
1840 	 */
1841 	if (sp->s_dvp->v_type == VBAD)
1842 		return (0);
1843 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1844 	    sp->s_cred, NULL));
1845 }
1846 
1847 /*
1848  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1849  */
1850 static int
1851 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1852     int namelen, struct ucred *cred, struct thread *td)
1853 {
1854 	struct nfsvattr dnfsva;
1855 	struct nfsnode *dnp = VTONFS(dvp);
1856 	int error = 0, dattrflag;
1857 
1858 	NFSLOCKNODE(dnp);
1859 	dnp->n_flag |= NREMOVEINPROG;
1860 	NFSUNLOCKNODE(dnp);
1861 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1862 	    &dattrflag, NULL);
1863 	NFSLOCKNODE(dnp);
1864 	if ((dnp->n_flag & NREMOVEWANT)) {
1865 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1866 		NFSUNLOCKNODE(dnp);
1867 		wakeup((caddr_t)dnp);
1868 	} else {
1869 		dnp->n_flag &= ~NREMOVEINPROG;
1870 		NFSUNLOCKNODE(dnp);
1871 	}
1872 	if (dattrflag)
1873 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1874 	NFSLOCKNODE(dnp);
1875 	dnp->n_flag |= NMODIFIED;
1876 	if (!dattrflag) {
1877 		dnp->n_attrstamp = 0;
1878 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1879 	}
1880 	NFSUNLOCKNODE(dnp);
1881 	if (error && NFS_ISV4(dvp))
1882 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1883 	return (error);
1884 }
1885 
1886 /*
1887  * nfs file rename call
1888  */
1889 static int
1890 nfs_rename(struct vop_rename_args *ap)
1891 {
1892 	struct vnode *fvp = ap->a_fvp;
1893 	struct vnode *tvp = ap->a_tvp;
1894 	struct vnode *fdvp = ap->a_fdvp;
1895 	struct vnode *tdvp = ap->a_tdvp;
1896 	struct componentname *tcnp = ap->a_tcnp;
1897 	struct componentname *fcnp = ap->a_fcnp;
1898 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1899 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1900 	struct nfsv4node *newv4 = NULL;
1901 	int error;
1902 
1903 	KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1904 	    (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1905 	/* Check for cross-device rename */
1906 	if ((fvp->v_mount != tdvp->v_mount) ||
1907 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1908 		error = EXDEV;
1909 		goto out;
1910 	}
1911 
1912 	if (fvp == tvp) {
1913 		printf("nfs_rename: fvp == tvp (can't happen)\n");
1914 		error = 0;
1915 		goto out;
1916 	}
1917 	if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1918 		goto out;
1919 
1920 	/*
1921 	 * We have to flush B_DELWRI data prior to renaming
1922 	 * the file.  If we don't, the delayed-write buffers
1923 	 * can be flushed out later after the file has gone stale
1924 	 * under NFSV3.  NFSV2 does not have this problem because
1925 	 * ( as far as I can tell ) it flushes dirty buffers more
1926 	 * often.
1927 	 *
1928 	 * Skip the rename operation if the fsync fails, this can happen
1929 	 * due to the server's volume being full, when we pushed out data
1930 	 * that was written back to our cache earlier. Not checking for
1931 	 * this condition can result in potential (silent) data loss.
1932 	 */
1933 	error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1934 	NFSVOPUNLOCK(fvp, 0);
1935 	if (!error && tvp)
1936 		error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1937 	if (error)
1938 		goto out;
1939 
1940 	/*
1941 	 * If the tvp exists and is in use, sillyrename it before doing the
1942 	 * rename of the new file over it.
1943 	 * XXX Can't sillyrename a directory.
1944 	 */
1945 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1946 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1947 		vput(tvp);
1948 		tvp = NULL;
1949 	}
1950 
1951 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1952 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1953 	    tcnp->cn_thread);
1954 
1955 	if (error == 0 && NFS_ISV4(tdvp)) {
1956 		/*
1957 		 * For NFSv4, check to see if it is the same name and
1958 		 * replace the name, if it is different.
1959 		 */
1960 		newv4 = malloc(
1961 		    sizeof (struct nfsv4node) +
1962 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1963 		    M_NFSV4NODE, M_WAITOK);
1964 		NFSLOCKNODE(tdnp);
1965 		NFSLOCKNODE(fnp);
1966 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1967 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1968 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1969 		      tcnp->cn_namelen) ||
1970 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1971 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1972 			tdnp->n_fhp->nfh_len))) {
1973 #ifdef notdef
1974 { char nnn[100]; int nnnl;
1975 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1976 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1977 nnn[nnnl] = '\0';
1978 printf("ren replace=%s\n",nnn);
1979 }
1980 #endif
1981 			free(fnp->n_v4, M_NFSV4NODE);
1982 			fnp->n_v4 = newv4;
1983 			newv4 = NULL;
1984 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1985 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1986 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1987 			    tdnp->n_fhp->nfh_len);
1988 			NFSBCOPY(tcnp->cn_nameptr,
1989 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1990 		}
1991 		NFSUNLOCKNODE(tdnp);
1992 		NFSUNLOCKNODE(fnp);
1993 		if (newv4 != NULL)
1994 			free(newv4, M_NFSV4NODE);
1995 	}
1996 
1997 	if (fvp->v_type == VDIR) {
1998 		if (tvp != NULL && tvp->v_type == VDIR)
1999 			cache_purge(tdvp);
2000 		cache_purge(fdvp);
2001 	}
2002 
2003 out:
2004 	if (tdvp == tvp)
2005 		vrele(tdvp);
2006 	else
2007 		vput(tdvp);
2008 	if (tvp)
2009 		vput(tvp);
2010 	vrele(fdvp);
2011 	vrele(fvp);
2012 	/*
2013 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2014 	 */
2015 	if (error == ENOENT)
2016 		error = 0;
2017 	return (error);
2018 }
2019 
2020 /*
2021  * nfs file rename rpc called from nfs_remove() above
2022  */
2023 static int
2024 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
2025     struct sillyrename *sp)
2026 {
2027 
2028 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
2029 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
2030 	    scnp->cn_thread));
2031 }
2032 
2033 /*
2034  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2035  */
2036 static int
2037 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
2038     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
2039     int tnamelen, struct ucred *cred, struct thread *td)
2040 {
2041 	struct nfsvattr fnfsva, tnfsva;
2042 	struct nfsnode *fdnp = VTONFS(fdvp);
2043 	struct nfsnode *tdnp = VTONFS(tdvp);
2044 	int error = 0, fattrflag, tattrflag;
2045 
2046 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
2047 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
2048 	    &tattrflag, NULL, NULL);
2049 	NFSLOCKNODE(fdnp);
2050 	fdnp->n_flag |= NMODIFIED;
2051 	if (fattrflag != 0) {
2052 		NFSUNLOCKNODE(fdnp);
2053 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
2054 	} else {
2055 		fdnp->n_attrstamp = 0;
2056 		NFSUNLOCKNODE(fdnp);
2057 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
2058 	}
2059 	NFSLOCKNODE(tdnp);
2060 	tdnp->n_flag |= NMODIFIED;
2061 	if (tattrflag != 0) {
2062 		NFSUNLOCKNODE(tdnp);
2063 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
2064 	} else {
2065 		tdnp->n_attrstamp = 0;
2066 		NFSUNLOCKNODE(tdnp);
2067 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2068 	}
2069 	if (error && NFS_ISV4(fdvp))
2070 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2071 	return (error);
2072 }
2073 
2074 /*
2075  * nfs hard link create call
2076  */
2077 static int
2078 nfs_link(struct vop_link_args *ap)
2079 {
2080 	struct vnode *vp = ap->a_vp;
2081 	struct vnode *tdvp = ap->a_tdvp;
2082 	struct componentname *cnp = ap->a_cnp;
2083 	struct nfsnode *np, *tdnp;
2084 	struct nfsvattr nfsva, dnfsva;
2085 	int error = 0, attrflag, dattrflag;
2086 
2087 	/*
2088 	 * Push all writes to the server, so that the attribute cache
2089 	 * doesn't get "out of sync" with the server.
2090 	 * XXX There should be a better way!
2091 	 */
2092 	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
2093 
2094 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2095 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
2096 	    &dattrflag, NULL);
2097 	tdnp = VTONFS(tdvp);
2098 	NFSLOCKNODE(tdnp);
2099 	tdnp->n_flag |= NMODIFIED;
2100 	if (dattrflag != 0) {
2101 		NFSUNLOCKNODE(tdnp);
2102 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
2103 	} else {
2104 		tdnp->n_attrstamp = 0;
2105 		NFSUNLOCKNODE(tdnp);
2106 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2107 	}
2108 	if (attrflag)
2109 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2110 	else {
2111 		np = VTONFS(vp);
2112 		NFSLOCKNODE(np);
2113 		np->n_attrstamp = 0;
2114 		NFSUNLOCKNODE(np);
2115 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2116 	}
2117 	/*
2118 	 * If negative lookup caching is enabled, I might as well
2119 	 * add an entry for this node. Not necessary for correctness,
2120 	 * but if negative caching is enabled, then the system
2121 	 * must care about lookup caching hit rate, so...
2122 	 */
2123 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2124 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2125 		cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2126 	}
2127 	if (error && NFS_ISV4(vp))
2128 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2129 		    (gid_t)0);
2130 	return (error);
2131 }
2132 
2133 /*
2134  * nfs symbolic link create call
2135  */
2136 static int
2137 nfs_symlink(struct vop_symlink_args *ap)
2138 {
2139 	struct vnode *dvp = ap->a_dvp;
2140 	struct vattr *vap = ap->a_vap;
2141 	struct componentname *cnp = ap->a_cnp;
2142 	struct nfsvattr nfsva, dnfsva;
2143 	struct nfsfh *nfhp;
2144 	struct nfsnode *np = NULL, *dnp;
2145 	struct vnode *newvp = NULL;
2146 	int error = 0, attrflag, dattrflag, ret;
2147 
2148 	vap->va_type = VLNK;
2149 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2150 	    ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2151 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2152 	if (nfhp) {
2153 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2154 		    &np, NULL, LK_EXCLUSIVE);
2155 		if (!ret)
2156 			newvp = NFSTOV(np);
2157 		else if (!error)
2158 			error = ret;
2159 	}
2160 	if (newvp != NULL) {
2161 		if (attrflag)
2162 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2163 			    0, 1);
2164 	} else if (!error) {
2165 		/*
2166 		 * If we do not have an error and we could not extract the
2167 		 * newvp from the response due to the request being NFSv2, we
2168 		 * have to do a lookup in order to obtain a newvp to return.
2169 		 */
2170 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2171 		    cnp->cn_cred, cnp->cn_thread, &np);
2172 		if (!error)
2173 			newvp = NFSTOV(np);
2174 	}
2175 	if (error) {
2176 		if (newvp)
2177 			vput(newvp);
2178 		if (NFS_ISV4(dvp))
2179 			error = nfscl_maperr(cnp->cn_thread, error,
2180 			    vap->va_uid, vap->va_gid);
2181 	} else {
2182 		*ap->a_vpp = newvp;
2183 	}
2184 
2185 	dnp = VTONFS(dvp);
2186 	NFSLOCKNODE(dnp);
2187 	dnp->n_flag |= NMODIFIED;
2188 	if (dattrflag != 0) {
2189 		NFSUNLOCKNODE(dnp);
2190 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2191 	} else {
2192 		dnp->n_attrstamp = 0;
2193 		NFSUNLOCKNODE(dnp);
2194 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2195 	}
2196 	/*
2197 	 * If negative lookup caching is enabled, I might as well
2198 	 * add an entry for this node. Not necessary for correctness,
2199 	 * but if negative caching is enabled, then the system
2200 	 * must care about lookup caching hit rate, so...
2201 	 */
2202 	if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2203 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2204 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2205 	}
2206 	return (error);
2207 }
2208 
2209 /*
2210  * nfs make dir call
2211  */
2212 static int
2213 nfs_mkdir(struct vop_mkdir_args *ap)
2214 {
2215 	struct vnode *dvp = ap->a_dvp;
2216 	struct vattr *vap = ap->a_vap;
2217 	struct componentname *cnp = ap->a_cnp;
2218 	struct nfsnode *np = NULL, *dnp;
2219 	struct vnode *newvp = NULL;
2220 	struct vattr vattr;
2221 	struct nfsfh *nfhp;
2222 	struct nfsvattr nfsva, dnfsva;
2223 	int error = 0, attrflag, dattrflag, ret;
2224 
2225 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2226 		return (error);
2227 	vap->va_type = VDIR;
2228 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2229 	    vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2230 	    &attrflag, &dattrflag, NULL);
2231 	dnp = VTONFS(dvp);
2232 	NFSLOCKNODE(dnp);
2233 	dnp->n_flag |= NMODIFIED;
2234 	if (dattrflag != 0) {
2235 		NFSUNLOCKNODE(dnp);
2236 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2237 	} else {
2238 		dnp->n_attrstamp = 0;
2239 		NFSUNLOCKNODE(dnp);
2240 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2241 	}
2242 	if (nfhp) {
2243 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2244 		    &np, NULL, LK_EXCLUSIVE);
2245 		if (!ret) {
2246 			newvp = NFSTOV(np);
2247 			if (attrflag)
2248 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2249 				NULL, 0, 1);
2250 		} else if (!error)
2251 			error = ret;
2252 	}
2253 	if (!error && newvp == NULL) {
2254 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2255 		    cnp->cn_cred, cnp->cn_thread, &np);
2256 		if (!error) {
2257 			newvp = NFSTOV(np);
2258 			if (newvp->v_type != VDIR)
2259 				error = EEXIST;
2260 		}
2261 	}
2262 	if (error) {
2263 		if (newvp)
2264 			vput(newvp);
2265 		if (NFS_ISV4(dvp))
2266 			error = nfscl_maperr(cnp->cn_thread, error,
2267 			    vap->va_uid, vap->va_gid);
2268 	} else {
2269 		/*
2270 		 * If negative lookup caching is enabled, I might as well
2271 		 * add an entry for this node. Not necessary for correctness,
2272 		 * but if negative caching is enabled, then the system
2273 		 * must care about lookup caching hit rate, so...
2274 		 */
2275 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2276 		    (cnp->cn_flags & MAKEENTRY) &&
2277 		    attrflag != 0 && dattrflag != 0)
2278 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2279 			    &dnfsva.na_ctime);
2280 		*ap->a_vpp = newvp;
2281 	}
2282 	return (error);
2283 }
2284 
2285 /*
2286  * nfs remove directory call
2287  */
2288 static int
2289 nfs_rmdir(struct vop_rmdir_args *ap)
2290 {
2291 	struct vnode *vp = ap->a_vp;
2292 	struct vnode *dvp = ap->a_dvp;
2293 	struct componentname *cnp = ap->a_cnp;
2294 	struct nfsnode *dnp;
2295 	struct nfsvattr dnfsva;
2296 	int error, dattrflag;
2297 
2298 	if (dvp == vp)
2299 		return (EINVAL);
2300 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2301 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2302 	dnp = VTONFS(dvp);
2303 	NFSLOCKNODE(dnp);
2304 	dnp->n_flag |= NMODIFIED;
2305 	if (dattrflag != 0) {
2306 		NFSUNLOCKNODE(dnp);
2307 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2308 	} else {
2309 		dnp->n_attrstamp = 0;
2310 		NFSUNLOCKNODE(dnp);
2311 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2312 	}
2313 
2314 	cache_purge(dvp);
2315 	cache_purge(vp);
2316 	if (error && NFS_ISV4(dvp))
2317 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2318 		    (gid_t)0);
2319 	/*
2320 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2321 	 */
2322 	if (error == ENOENT)
2323 		error = 0;
2324 	return (error);
2325 }
2326 
2327 /*
2328  * nfs readdir call
2329  */
2330 static int
2331 nfs_readdir(struct vop_readdir_args *ap)
2332 {
2333 	struct vnode *vp = ap->a_vp;
2334 	struct nfsnode *np = VTONFS(vp);
2335 	struct uio *uio = ap->a_uio;
2336 	ssize_t tresid, left;
2337 	int error = 0;
2338 	struct vattr vattr;
2339 
2340 	if (ap->a_eofflag != NULL)
2341 		*ap->a_eofflag = 0;
2342 	if (vp->v_type != VDIR)
2343 		return(EPERM);
2344 
2345 	/*
2346 	 * First, check for hit on the EOF offset cache
2347 	 */
2348 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2349 	    (np->n_flag & NMODIFIED) == 0) {
2350 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2351 			NFSLOCKNODE(np);
2352 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2353 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2354 				NFSUNLOCKNODE(np);
2355 				NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2356 				if (ap->a_eofflag != NULL)
2357 					*ap->a_eofflag = 1;
2358 				return (0);
2359 			} else
2360 				NFSUNLOCKNODE(np);
2361 		}
2362 	}
2363 
2364 	/*
2365 	 * NFS always guarantees that directory entries don't straddle
2366 	 * DIRBLKSIZ boundaries.  As such, we need to limit the size
2367 	 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2368 	 * directory entry.
2369 	 */
2370 	left = uio->uio_resid % DIRBLKSIZ;
2371 	if (left == uio->uio_resid)
2372 		return (EINVAL);
2373 	uio->uio_resid -= left;
2374 
2375 	/*
2376 	 * Call ncl_bioread() to do the real work.
2377 	 */
2378 	tresid = uio->uio_resid;
2379 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2380 
2381 	if (!error && uio->uio_resid == tresid) {
2382 		NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2383 		if (ap->a_eofflag != NULL)
2384 			*ap->a_eofflag = 1;
2385 	}
2386 
2387 	/* Add the partial DIRBLKSIZ (left) back in. */
2388 	uio->uio_resid += left;
2389 	return (error);
2390 }
2391 
2392 /*
2393  * Readdir rpc call.
2394  * Called from below the buffer cache by ncl_doio().
2395  */
2396 int
2397 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2398     struct thread *td)
2399 {
2400 	struct nfsvattr nfsva;
2401 	nfsuint64 *cookiep, cookie;
2402 	struct nfsnode *dnp = VTONFS(vp);
2403 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2404 	int error = 0, eof, attrflag;
2405 
2406 	KASSERT(uiop->uio_iovcnt == 1 &&
2407 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2408 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2409 	    ("nfs readdirrpc bad uio"));
2410 
2411 	/*
2412 	 * If there is no cookie, assume directory was stale.
2413 	 */
2414 	ncl_dircookie_lock(dnp);
2415 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2416 	if (cookiep) {
2417 		cookie = *cookiep;
2418 		ncl_dircookie_unlock(dnp);
2419 	} else {
2420 		ncl_dircookie_unlock(dnp);
2421 		return (NFSERR_BAD_COOKIE);
2422 	}
2423 
2424 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2425 		(void)ncl_fsinfo(nmp, vp, cred, td);
2426 
2427 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2428 	    &attrflag, &eof, NULL);
2429 	if (attrflag)
2430 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2431 
2432 	if (!error) {
2433 		/*
2434 		 * We are now either at the end of the directory or have filled
2435 		 * the block.
2436 		 */
2437 		if (eof)
2438 			dnp->n_direofoffset = uiop->uio_offset;
2439 		else {
2440 			if (uiop->uio_resid > 0)
2441 				printf("EEK! readdirrpc resid > 0\n");
2442 			ncl_dircookie_lock(dnp);
2443 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2444 			*cookiep = cookie;
2445 			ncl_dircookie_unlock(dnp);
2446 		}
2447 	} else if (NFS_ISV4(vp)) {
2448 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2449 	}
2450 	return (error);
2451 }
2452 
2453 /*
2454  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2455  */
2456 int
2457 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2458     struct thread *td)
2459 {
2460 	struct nfsvattr nfsva;
2461 	nfsuint64 *cookiep, cookie;
2462 	struct nfsnode *dnp = VTONFS(vp);
2463 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2464 	int error = 0, attrflag, eof;
2465 
2466 	KASSERT(uiop->uio_iovcnt == 1 &&
2467 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2468 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2469 	    ("nfs readdirplusrpc bad uio"));
2470 
2471 	/*
2472 	 * If there is no cookie, assume directory was stale.
2473 	 */
2474 	ncl_dircookie_lock(dnp);
2475 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2476 	if (cookiep) {
2477 		cookie = *cookiep;
2478 		ncl_dircookie_unlock(dnp);
2479 	} else {
2480 		ncl_dircookie_unlock(dnp);
2481 		return (NFSERR_BAD_COOKIE);
2482 	}
2483 
2484 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2485 		(void)ncl_fsinfo(nmp, vp, cred, td);
2486 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2487 	    &attrflag, &eof, NULL);
2488 	if (attrflag)
2489 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2490 
2491 	if (!error) {
2492 		/*
2493 		 * We are now either at end of the directory or have filled the
2494 		 * the block.
2495 		 */
2496 		if (eof)
2497 			dnp->n_direofoffset = uiop->uio_offset;
2498 		else {
2499 			if (uiop->uio_resid > 0)
2500 				printf("EEK! readdirplusrpc resid > 0\n");
2501 			ncl_dircookie_lock(dnp);
2502 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2503 			*cookiep = cookie;
2504 			ncl_dircookie_unlock(dnp);
2505 		}
2506 	} else if (NFS_ISV4(vp)) {
2507 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2508 	}
2509 	return (error);
2510 }
2511 
2512 /*
2513  * Silly rename. To make the NFS filesystem that is stateless look a little
2514  * more like the "ufs" a remove of an active vnode is translated to a rename
2515  * to a funny looking filename that is removed by nfs_inactive on the
2516  * nfsnode. There is the potential for another process on a different client
2517  * to create the same funny name between the nfs_lookitup() fails and the
2518  * nfs_rename() completes, but...
2519  */
2520 static int
2521 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2522 {
2523 	struct sillyrename *sp;
2524 	struct nfsnode *np;
2525 	int error;
2526 	short pid;
2527 	unsigned int lticks;
2528 
2529 	cache_purge(dvp);
2530 	np = VTONFS(vp);
2531 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2532 	sp = malloc(sizeof (struct sillyrename),
2533 	    M_NEWNFSREQ, M_WAITOK);
2534 	sp->s_cred = crhold(cnp->cn_cred);
2535 	sp->s_dvp = dvp;
2536 	VREF(dvp);
2537 
2538 	/*
2539 	 * Fudge together a funny name.
2540 	 * Changing the format of the funny name to accommodate more
2541 	 * sillynames per directory.
2542 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2543 	 * CPU ticks since boot.
2544 	 */
2545 	pid = cnp->cn_thread->td_proc->p_pid;
2546 	lticks = (unsigned int)ticks;
2547 	for ( ; ; ) {
2548 		sp->s_namlen = sprintf(sp->s_name,
2549 				       ".nfs.%08x.%04x4.4", lticks,
2550 				       pid);
2551 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2552 				 cnp->cn_thread, NULL))
2553 			break;
2554 		lticks++;
2555 	}
2556 	error = nfs_renameit(dvp, vp, cnp, sp);
2557 	if (error)
2558 		goto bad;
2559 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2560 		cnp->cn_thread, &np);
2561 	np->n_sillyrename = sp;
2562 	return (0);
2563 bad:
2564 	vrele(sp->s_dvp);
2565 	crfree(sp->s_cred);
2566 	free(sp, M_NEWNFSREQ);
2567 	return (error);
2568 }
2569 
2570 /*
2571  * Look up a file name and optionally either update the file handle or
2572  * allocate an nfsnode, depending on the value of npp.
2573  * npp == NULL	--> just do the lookup
2574  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2575  *			handled too
2576  * *npp != NULL --> update the file handle in the vnode
2577  */
2578 static int
2579 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2580     struct thread *td, struct nfsnode **npp)
2581 {
2582 	struct vnode *newvp = NULL, *vp;
2583 	struct nfsnode *np, *dnp = VTONFS(dvp);
2584 	struct nfsfh *nfhp, *onfhp;
2585 	struct nfsvattr nfsva, dnfsva;
2586 	struct componentname cn;
2587 	int error = 0, attrflag, dattrflag;
2588 	u_int hash;
2589 
2590 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2591 	    &nfhp, &attrflag, &dattrflag, NULL);
2592 	if (dattrflag)
2593 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2594 	if (npp && !error) {
2595 		if (*npp != NULL) {
2596 		    np = *npp;
2597 		    vp = NFSTOV(np);
2598 		    /*
2599 		     * For NFSv4, check to see if it is the same name and
2600 		     * replace the name, if it is different.
2601 		     */
2602 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2603 			(np->n_v4->n4_namelen != len ||
2604 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2605 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2606 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2607 			 dnp->n_fhp->nfh_len))) {
2608 #ifdef notdef
2609 { char nnn[100]; int nnnl;
2610 nnnl = (len < 100) ? len : 99;
2611 bcopy(name, nnn, nnnl);
2612 nnn[nnnl] = '\0';
2613 printf("replace=%s\n",nnn);
2614 }
2615 #endif
2616 			    free(np->n_v4, M_NFSV4NODE);
2617 			    np->n_v4 = malloc(
2618 				sizeof (struct nfsv4node) +
2619 				dnp->n_fhp->nfh_len + len - 1,
2620 				M_NFSV4NODE, M_WAITOK);
2621 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2622 			    np->n_v4->n4_namelen = len;
2623 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2624 				dnp->n_fhp->nfh_len);
2625 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2626 		    }
2627 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2628 			FNV1_32_INIT);
2629 		    onfhp = np->n_fhp;
2630 		    /*
2631 		     * Rehash node for new file handle.
2632 		     */
2633 		    vfs_hash_rehash(vp, hash);
2634 		    np->n_fhp = nfhp;
2635 		    if (onfhp != NULL)
2636 			free(onfhp, M_NFSFH);
2637 		    newvp = NFSTOV(np);
2638 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2639 		    free(nfhp, M_NFSFH);
2640 		    VREF(dvp);
2641 		    newvp = dvp;
2642 		} else {
2643 		    cn.cn_nameptr = name;
2644 		    cn.cn_namelen = len;
2645 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2646 			&np, NULL, LK_EXCLUSIVE);
2647 		    if (error)
2648 			return (error);
2649 		    newvp = NFSTOV(np);
2650 		}
2651 		if (!attrflag && *npp == NULL) {
2652 			if (newvp == dvp)
2653 				vrele(newvp);
2654 			else
2655 				vput(newvp);
2656 			return (ENOENT);
2657 		}
2658 		if (attrflag)
2659 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2660 			    0, 1);
2661 	}
2662 	if (npp && *npp == NULL) {
2663 		if (error) {
2664 			if (newvp) {
2665 				if (newvp == dvp)
2666 					vrele(newvp);
2667 				else
2668 					vput(newvp);
2669 			}
2670 		} else
2671 			*npp = np;
2672 	}
2673 	if (error && NFS_ISV4(dvp))
2674 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2675 	return (error);
2676 }
2677 
2678 /*
2679  * Nfs Version 3 and 4 commit rpc
2680  */
2681 int
2682 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2683    struct thread *td)
2684 {
2685 	struct nfsvattr nfsva;
2686 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2687 	struct nfsnode *np;
2688 	struct uio uio;
2689 	int error, attrflag;
2690 
2691 	np = VTONFS(vp);
2692 	error = EIO;
2693 	attrflag = 0;
2694 	if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2695 		uio.uio_offset = offset;
2696 		uio.uio_resid = cnt;
2697 		error = nfscl_doiods(vp, &uio, NULL, NULL,
2698 		    NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2699 		if (error != 0) {
2700 			NFSLOCKNODE(np);
2701 			np->n_flag &= ~NDSCOMMIT;
2702 			NFSUNLOCKNODE(np);
2703 		}
2704 	}
2705 	if (error != 0) {
2706 		mtx_lock(&nmp->nm_mtx);
2707 		if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2708 			mtx_unlock(&nmp->nm_mtx);
2709 			return (0);
2710 		}
2711 		mtx_unlock(&nmp->nm_mtx);
2712 		error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2713 		    &attrflag, NULL);
2714 	}
2715 	if (attrflag != 0)
2716 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2717 		    0, 1);
2718 	if (error != 0 && NFS_ISV4(vp))
2719 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2720 	return (error);
2721 }
2722 
2723 /*
2724  * Strategy routine.
2725  * For async requests when nfsiod(s) are running, queue the request by
2726  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2727  * request.
2728  */
2729 static int
2730 nfs_strategy(struct vop_strategy_args *ap)
2731 {
2732 	struct buf *bp;
2733 	struct vnode *vp;
2734 	struct ucred *cr;
2735 
2736 	bp = ap->a_bp;
2737 	vp = ap->a_vp;
2738 	KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2739 	KASSERT(!(bp->b_flags & B_DONE),
2740 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2741 
2742 	if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2743 		bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2744 		    DEV_BSIZE);
2745 	if (bp->b_iocmd == BIO_READ)
2746 		cr = bp->b_rcred;
2747 	else
2748 		cr = bp->b_wcred;
2749 
2750 	/*
2751 	 * If the op is asynchronous and an i/o daemon is waiting
2752 	 * queue the request, wake it up and wait for completion
2753 	 * otherwise just do it ourselves.
2754 	 */
2755 	if ((bp->b_flags & B_ASYNC) == 0 ||
2756 	    ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2757 		(void) ncl_doio(vp, bp, cr, curthread, 1);
2758 	return (0);
2759 }
2760 
2761 /*
2762  * fsync vnode op. Just call ncl_flush() with commit == 1.
2763  */
2764 /* ARGSUSED */
2765 static int
2766 nfs_fsync(struct vop_fsync_args *ap)
2767 {
2768 
2769 	if (ap->a_vp->v_type != VREG) {
2770 		/*
2771 		 * For NFS, metadata is changed synchronously on the server,
2772 		 * so there is nothing to flush. Also, ncl_flush() clears
2773 		 * the NMODIFIED flag and that shouldn't be done here for
2774 		 * directories.
2775 		 */
2776 		return (0);
2777 	}
2778 	return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2779 }
2780 
2781 /*
2782  * Flush all the blocks associated with a vnode.
2783  * 	Walk through the buffer pool and push any dirty pages
2784  *	associated with the vnode.
2785  * If the called_from_renewthread argument is TRUE, it has been called
2786  * from the NFSv4 renew thread and, as such, cannot block indefinitely
2787  * waiting for a buffer write to complete.
2788  */
2789 int
2790 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2791     int commit, int called_from_renewthread)
2792 {
2793 	struct nfsnode *np = VTONFS(vp);
2794 	struct buf *bp;
2795 	int i;
2796 	struct buf *nbp;
2797 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2798 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2799 	int passone = 1, trycnt = 0;
2800 	u_quad_t off, endoff, toff;
2801 	struct ucred* wcred = NULL;
2802 	struct buf **bvec = NULL;
2803 	struct bufobj *bo;
2804 #ifndef NFS_COMMITBVECSIZ
2805 #define	NFS_COMMITBVECSIZ	20
2806 #endif
2807 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2808 	u_int bvecsize = 0, bveccount;
2809 
2810 	if (called_from_renewthread != 0)
2811 		slptimeo = hz;
2812 	if (nmp->nm_flag & NFSMNT_INT)
2813 		slpflag = PCATCH;
2814 	if (!commit)
2815 		passone = 0;
2816 	bo = &vp->v_bufobj;
2817 	/*
2818 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2819 	 * server, but has not been committed to stable storage on the server
2820 	 * yet. On the first pass, the byte range is worked out and the commit
2821 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2822 	 * job.
2823 	 */
2824 again:
2825 	off = (u_quad_t)-1;
2826 	endoff = 0;
2827 	bvecpos = 0;
2828 	if (NFS_ISV34(vp) && commit) {
2829 		if (bvec != NULL && bvec != bvec_on_stack)
2830 			free(bvec, M_TEMP);
2831 		/*
2832 		 * Count up how many buffers waiting for a commit.
2833 		 */
2834 		bveccount = 0;
2835 		BO_LOCK(bo);
2836 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2837 			if (!BUF_ISLOCKED(bp) &&
2838 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2839 				== (B_DELWRI | B_NEEDCOMMIT))
2840 				bveccount++;
2841 		}
2842 		/*
2843 		 * Allocate space to remember the list of bufs to commit.  It is
2844 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2845 		 * If we can't get memory (for whatever reason), we will end up
2846 		 * committing the buffers one-by-one in the loop below.
2847 		 */
2848 		if (bveccount > NFS_COMMITBVECSIZ) {
2849 			/*
2850 			 * Release the vnode interlock to avoid a lock
2851 			 * order reversal.
2852 			 */
2853 			BO_UNLOCK(bo);
2854 			bvec = (struct buf **)
2855 				malloc(bveccount * sizeof(struct buf *),
2856 				       M_TEMP, M_NOWAIT);
2857 			BO_LOCK(bo);
2858 			if (bvec == NULL) {
2859 				bvec = bvec_on_stack;
2860 				bvecsize = NFS_COMMITBVECSIZ;
2861 			} else
2862 				bvecsize = bveccount;
2863 		} else {
2864 			bvec = bvec_on_stack;
2865 			bvecsize = NFS_COMMITBVECSIZ;
2866 		}
2867 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2868 			if (bvecpos >= bvecsize)
2869 				break;
2870 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2871 				nbp = TAILQ_NEXT(bp, b_bobufs);
2872 				continue;
2873 			}
2874 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2875 			    (B_DELWRI | B_NEEDCOMMIT)) {
2876 				BUF_UNLOCK(bp);
2877 				nbp = TAILQ_NEXT(bp, b_bobufs);
2878 				continue;
2879 			}
2880 			BO_UNLOCK(bo);
2881 			bremfree(bp);
2882 			/*
2883 			 * Work out if all buffers are using the same cred
2884 			 * so we can deal with them all with one commit.
2885 			 *
2886 			 * NOTE: we are not clearing B_DONE here, so we have
2887 			 * to do it later on in this routine if we intend to
2888 			 * initiate I/O on the bp.
2889 			 *
2890 			 * Note: to avoid loopback deadlocks, we do not
2891 			 * assign b_runningbufspace.
2892 			 */
2893 			if (wcred == NULL)
2894 				wcred = bp->b_wcred;
2895 			else if (wcred != bp->b_wcred)
2896 				wcred = NOCRED;
2897 			vfs_busy_pages(bp, 1);
2898 
2899 			BO_LOCK(bo);
2900 			/*
2901 			 * bp is protected by being locked, but nbp is not
2902 			 * and vfs_busy_pages() may sleep.  We have to
2903 			 * recalculate nbp.
2904 			 */
2905 			nbp = TAILQ_NEXT(bp, b_bobufs);
2906 
2907 			/*
2908 			 * A list of these buffers is kept so that the
2909 			 * second loop knows which buffers have actually
2910 			 * been committed. This is necessary, since there
2911 			 * may be a race between the commit rpc and new
2912 			 * uncommitted writes on the file.
2913 			 */
2914 			bvec[bvecpos++] = bp;
2915 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2916 				bp->b_dirtyoff;
2917 			if (toff < off)
2918 				off = toff;
2919 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2920 			if (toff > endoff)
2921 				endoff = toff;
2922 		}
2923 		BO_UNLOCK(bo);
2924 	}
2925 	if (bvecpos > 0) {
2926 		/*
2927 		 * Commit data on the server, as required.
2928 		 * If all bufs are using the same wcred, then use that with
2929 		 * one call for all of them, otherwise commit each one
2930 		 * separately.
2931 		 */
2932 		if (wcred != NOCRED)
2933 			retv = ncl_commit(vp, off, (int)(endoff - off),
2934 					  wcred, td);
2935 		else {
2936 			retv = 0;
2937 			for (i = 0; i < bvecpos; i++) {
2938 				off_t off, size;
2939 				bp = bvec[i];
2940 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2941 					bp->b_dirtyoff;
2942 				size = (u_quad_t)(bp->b_dirtyend
2943 						  - bp->b_dirtyoff);
2944 				retv = ncl_commit(vp, off, (int)size,
2945 						  bp->b_wcred, td);
2946 				if (retv) break;
2947 			}
2948 		}
2949 
2950 		if (retv == NFSERR_STALEWRITEVERF)
2951 			ncl_clearcommit(vp->v_mount);
2952 
2953 		/*
2954 		 * Now, either mark the blocks I/O done or mark the
2955 		 * blocks dirty, depending on whether the commit
2956 		 * succeeded.
2957 		 */
2958 		for (i = 0; i < bvecpos; i++) {
2959 			bp = bvec[i];
2960 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2961 			if (retv) {
2962 				/*
2963 				 * Error, leave B_DELWRI intact
2964 				 */
2965 				vfs_unbusy_pages(bp);
2966 				brelse(bp);
2967 			} else {
2968 				/*
2969 				 * Success, remove B_DELWRI ( bundirty() ).
2970 				 *
2971 				 * b_dirtyoff/b_dirtyend seem to be NFS
2972 				 * specific.  We should probably move that
2973 				 * into bundirty(). XXX
2974 				 */
2975 				bufobj_wref(bo);
2976 				bp->b_flags |= B_ASYNC;
2977 				bundirty(bp);
2978 				bp->b_flags &= ~B_DONE;
2979 				bp->b_ioflags &= ~BIO_ERROR;
2980 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2981 				bufdone(bp);
2982 			}
2983 		}
2984 	}
2985 
2986 	/*
2987 	 * Start/do any write(s) that are required.
2988 	 */
2989 loop:
2990 	BO_LOCK(bo);
2991 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2992 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2993 			if (waitfor != MNT_WAIT || passone)
2994 				continue;
2995 
2996 			error = BUF_TIMELOCK(bp,
2997 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2998 			    BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
2999 			if (error == 0) {
3000 				BUF_UNLOCK(bp);
3001 				goto loop;
3002 			}
3003 			if (error == ENOLCK) {
3004 				error = 0;
3005 				goto loop;
3006 			}
3007 			if (called_from_renewthread != 0) {
3008 				/*
3009 				 * Return EIO so the flush will be retried
3010 				 * later.
3011 				 */
3012 				error = EIO;
3013 				goto done;
3014 			}
3015 			if (newnfs_sigintr(nmp, td)) {
3016 				error = EINTR;
3017 				goto done;
3018 			}
3019 			if (slpflag == PCATCH) {
3020 				slpflag = 0;
3021 				slptimeo = 2 * hz;
3022 			}
3023 			goto loop;
3024 		}
3025 		if ((bp->b_flags & B_DELWRI) == 0)
3026 			panic("nfs_fsync: not dirty");
3027 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3028 			BUF_UNLOCK(bp);
3029 			continue;
3030 		}
3031 		BO_UNLOCK(bo);
3032 		bremfree(bp);
3033 		bp->b_flags |= B_ASYNC;
3034 		bwrite(bp);
3035 		if (newnfs_sigintr(nmp, td)) {
3036 			error = EINTR;
3037 			goto done;
3038 		}
3039 		goto loop;
3040 	}
3041 	if (passone) {
3042 		passone = 0;
3043 		BO_UNLOCK(bo);
3044 		goto again;
3045 	}
3046 	if (waitfor == MNT_WAIT) {
3047 		while (bo->bo_numoutput) {
3048 			error = bufobj_wwait(bo, slpflag, slptimeo);
3049 			if (error) {
3050 			    BO_UNLOCK(bo);
3051 			    if (called_from_renewthread != 0) {
3052 				/*
3053 				 * Return EIO so that the flush will be
3054 				 * retried later.
3055 				 */
3056 				error = EIO;
3057 				goto done;
3058 			    }
3059 			    error = newnfs_sigintr(nmp, td);
3060 			    if (error)
3061 				goto done;
3062 			    if (slpflag == PCATCH) {
3063 				slpflag = 0;
3064 				slptimeo = 2 * hz;
3065 			    }
3066 			    BO_LOCK(bo);
3067 			}
3068 		}
3069 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
3070 			BO_UNLOCK(bo);
3071 			goto loop;
3072 		}
3073 		/*
3074 		 * Wait for all the async IO requests to drain
3075 		 */
3076 		BO_UNLOCK(bo);
3077 		NFSLOCKNODE(np);
3078 		while (np->n_directio_asyncwr > 0) {
3079 			np->n_flag |= NFSYNCWAIT;
3080 			error = newnfs_msleep(td, &np->n_directio_asyncwr,
3081 			    &np->n_mtx, slpflag | (PRIBIO + 1),
3082 			    "nfsfsync", 0);
3083 			if (error) {
3084 				if (newnfs_sigintr(nmp, td)) {
3085 					NFSUNLOCKNODE(np);
3086 					error = EINTR;
3087 					goto done;
3088 				}
3089 			}
3090 		}
3091 		NFSUNLOCKNODE(np);
3092 	} else
3093 		BO_UNLOCK(bo);
3094 	if (NFSHASPNFS(nmp)) {
3095 		nfscl_layoutcommit(vp, td);
3096 		/*
3097 		 * Invalidate the attribute cache, since writes to a DS
3098 		 * won't update the size attribute.
3099 		 */
3100 		NFSLOCKNODE(np);
3101 		np->n_attrstamp = 0;
3102 	} else
3103 		NFSLOCKNODE(np);
3104 	if (np->n_flag & NWRITEERR) {
3105 		error = np->n_error;
3106 		np->n_flag &= ~NWRITEERR;
3107 	}
3108   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
3109 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3110   		np->n_flag &= ~NMODIFIED;
3111 	NFSUNLOCKNODE(np);
3112 done:
3113 	if (bvec != NULL && bvec != bvec_on_stack)
3114 		free(bvec, M_TEMP);
3115 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
3116 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
3117 	    np->n_directio_asyncwr != 0)) {
3118 		if (trycnt++ < 5) {
3119 			/* try, try again... */
3120 			passone = 1;
3121 			wcred = NULL;
3122 			bvec = NULL;
3123 			bvecsize = 0;
3124 			goto again;
3125 		}
3126 		vn_printf(vp, "ncl_flush failed");
3127 		error = called_from_renewthread != 0 ? EIO : EBUSY;
3128 	}
3129 	return (error);
3130 }
3131 
3132 /*
3133  * NFS advisory byte-level locks.
3134  */
3135 static int
3136 nfs_advlock(struct vop_advlock_args *ap)
3137 {
3138 	struct vnode *vp = ap->a_vp;
3139 	struct ucred *cred;
3140 	struct nfsnode *np = VTONFS(ap->a_vp);
3141 	struct proc *p = (struct proc *)ap->a_id;
3142 	struct thread *td = curthread;	/* XXX */
3143 	struct vattr va;
3144 	int ret, error;
3145 	u_quad_t size;
3146 
3147 	error = NFSVOPLOCK(vp, LK_SHARED);
3148 	if (error != 0)
3149 		return (EBADF);
3150 	if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3151 		if (vp->v_type != VREG) {
3152 			error = EINVAL;
3153 			goto out;
3154 		}
3155 		if ((ap->a_flags & F_POSIX) != 0)
3156 			cred = p->p_ucred;
3157 		else
3158 			cred = td->td_ucred;
3159 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
3160 		if (VN_IS_DOOMED(vp)) {
3161 			error = EBADF;
3162 			goto out;
3163 		}
3164 
3165 		/*
3166 		 * If this is unlocking a write locked region, flush and
3167 		 * commit them before unlocking. This is required by
3168 		 * RFC3530 Sec. 9.3.2.
3169 		 */
3170 		if (ap->a_op == F_UNLCK &&
3171 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3172 		    ap->a_flags))
3173 			(void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3174 
3175 		/*
3176 		 * Loop around doing the lock op, while a blocking lock
3177 		 * must wait for the lock op to succeed.
3178 		 */
3179 		do {
3180 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3181 			    ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3182 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3183 			    ap->a_op == F_SETLK) {
3184 				NFSVOPUNLOCK(vp, 0);
3185 				error = nfs_catnap(PZERO | PCATCH, ret,
3186 				    "ncladvl");
3187 				if (error)
3188 					return (EINTR);
3189 				NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3190 				if (VN_IS_DOOMED(vp)) {
3191 					error = EBADF;
3192 					goto out;
3193 				}
3194 			}
3195 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3196 		     ap->a_op == F_SETLK);
3197 		if (ret == NFSERR_DENIED) {
3198 			error = EAGAIN;
3199 			goto out;
3200 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3201 			error = ret;
3202 			goto out;
3203 		} else if (ret != 0) {
3204 			error = EACCES;
3205 			goto out;
3206 		}
3207 
3208 		/*
3209 		 * Now, if we just got a lock, invalidate data in the buffer
3210 		 * cache, as required, so that the coherency conforms with
3211 		 * RFC3530 Sec. 9.3.2.
3212 		 */
3213 		if (ap->a_op == F_SETLK) {
3214 			if ((np->n_flag & NMODIFIED) == 0) {
3215 				np->n_attrstamp = 0;
3216 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3217 				ret = VOP_GETATTR(vp, &va, cred);
3218 			}
3219 			if ((np->n_flag & NMODIFIED) || ret ||
3220 			    np->n_change != va.va_filerev) {
3221 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3222 				np->n_attrstamp = 0;
3223 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3224 				ret = VOP_GETATTR(vp, &va, cred);
3225 				if (!ret) {
3226 					np->n_mtime = va.va_mtime;
3227 					np->n_change = va.va_filerev;
3228 				}
3229 			}
3230 			/* Mark that a file lock has been acquired. */
3231 			NFSLOCKNODE(np);
3232 			np->n_flag |= NHASBEENLOCKED;
3233 			NFSUNLOCKNODE(np);
3234 		}
3235 	} else if (!NFS_ISV4(vp)) {
3236 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3237 			size = VTONFS(vp)->n_size;
3238 			NFSVOPUNLOCK(vp, 0);
3239 			error = lf_advlock(ap, &(vp->v_lockf), size);
3240 		} else {
3241 			if (nfs_advlock_p != NULL)
3242 				error = nfs_advlock_p(ap);
3243 			else {
3244 				NFSVOPUNLOCK(vp, 0);
3245 				error = ENOLCK;
3246 			}
3247 		}
3248 		if (error == 0 && ap->a_op == F_SETLK) {
3249 			error = NFSVOPLOCK(vp, LK_SHARED);
3250 			if (error == 0) {
3251 				/* Mark that a file lock has been acquired. */
3252 				NFSLOCKNODE(np);
3253 				np->n_flag |= NHASBEENLOCKED;
3254 				NFSUNLOCKNODE(np);
3255 				NFSVOPUNLOCK(vp, 0);
3256 			}
3257 		}
3258 		return (error);
3259 	} else
3260 		error = EOPNOTSUPP;
3261 out:
3262 	NFSVOPUNLOCK(vp, 0);
3263 	return (error);
3264 }
3265 
3266 /*
3267  * NFS advisory byte-level locks.
3268  */
3269 static int
3270 nfs_advlockasync(struct vop_advlockasync_args *ap)
3271 {
3272 	struct vnode *vp = ap->a_vp;
3273 	u_quad_t size;
3274 	int error;
3275 
3276 	if (NFS_ISV4(vp))
3277 		return (EOPNOTSUPP);
3278 	error = NFSVOPLOCK(vp, LK_SHARED);
3279 	if (error)
3280 		return (error);
3281 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3282 		size = VTONFS(vp)->n_size;
3283 		NFSVOPUNLOCK(vp, 0);
3284 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
3285 	} else {
3286 		NFSVOPUNLOCK(vp, 0);
3287 		error = EOPNOTSUPP;
3288 	}
3289 	return (error);
3290 }
3291 
3292 /*
3293  * Print out the contents of an nfsnode.
3294  */
3295 static int
3296 nfs_print(struct vop_print_args *ap)
3297 {
3298 	struct vnode *vp = ap->a_vp;
3299 	struct nfsnode *np = VTONFS(vp);
3300 
3301 	printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3302 	    (uintmax_t)np->n_vattr.na_fsid);
3303 	if (vp->v_type == VFIFO)
3304 		fifo_printinfo(vp);
3305 	printf("\n");
3306 	return (0);
3307 }
3308 
3309 /*
3310  * This is the "real" nfs::bwrite(struct buf*).
3311  * We set B_CACHE if this is a VMIO buffer.
3312  */
3313 int
3314 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3315 {
3316 	int oldflags, rtval;
3317 
3318 	if (bp->b_flags & B_INVAL) {
3319 		brelse(bp);
3320 		return (0);
3321 	}
3322 
3323 	oldflags = bp->b_flags;
3324 	bp->b_flags |= B_CACHE;
3325 
3326 	/*
3327 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3328 	 */
3329 	bundirty(bp);
3330 	bp->b_flags &= ~B_DONE;
3331 	bp->b_ioflags &= ~BIO_ERROR;
3332 	bp->b_iocmd = BIO_WRITE;
3333 
3334 	bufobj_wref(bp->b_bufobj);
3335 	curthread->td_ru.ru_oublock++;
3336 
3337 	/*
3338 	 * Note: to avoid loopback deadlocks, we do not
3339 	 * assign b_runningbufspace.
3340 	 */
3341 	vfs_busy_pages(bp, 1);
3342 
3343 	BUF_KERNPROC(bp);
3344 	bp->b_iooffset = dbtob(bp->b_blkno);
3345 	bstrategy(bp);
3346 
3347 	if ((oldflags & B_ASYNC) != 0)
3348 		return (0);
3349 
3350 	rtval = bufwait(bp);
3351 	if (oldflags & B_DELWRI)
3352 		reassignbuf(bp);
3353 	brelse(bp);
3354 	return (rtval);
3355 }
3356 
3357 /*
3358  * nfs special file access vnode op.
3359  * Essentially just get vattr and then imitate iaccess() since the device is
3360  * local to the client.
3361  */
3362 static int
3363 nfsspec_access(struct vop_access_args *ap)
3364 {
3365 	struct vattr *vap;
3366 	struct ucred *cred = ap->a_cred;
3367 	struct vnode *vp = ap->a_vp;
3368 	accmode_t accmode = ap->a_accmode;
3369 	struct vattr vattr;
3370 	int error;
3371 
3372 	/*
3373 	 * Disallow write attempts on filesystems mounted read-only;
3374 	 * unless the file is a socket, fifo, or a block or character
3375 	 * device resident on the filesystem.
3376 	 */
3377 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3378 		switch (vp->v_type) {
3379 		case VREG:
3380 		case VDIR:
3381 		case VLNK:
3382 			return (EROFS);
3383 		default:
3384 			break;
3385 		}
3386 	}
3387 	vap = &vattr;
3388 	error = VOP_GETATTR(vp, vap, cred);
3389 	if (error)
3390 		goto out;
3391 	error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3392 	    accmode, cred, NULL);
3393 out:
3394 	return error;
3395 }
3396 
3397 /*
3398  * Read wrapper for fifos.
3399  */
3400 static int
3401 nfsfifo_read(struct vop_read_args *ap)
3402 {
3403 	struct nfsnode *np = VTONFS(ap->a_vp);
3404 	int error;
3405 
3406 	/*
3407 	 * Set access flag.
3408 	 */
3409 	NFSLOCKNODE(np);
3410 	np->n_flag |= NACC;
3411 	vfs_timestamp(&np->n_atim);
3412 	NFSUNLOCKNODE(np);
3413 	error = fifo_specops.vop_read(ap);
3414 	return error;
3415 }
3416 
3417 /*
3418  * Write wrapper for fifos.
3419  */
3420 static int
3421 nfsfifo_write(struct vop_write_args *ap)
3422 {
3423 	struct nfsnode *np = VTONFS(ap->a_vp);
3424 
3425 	/*
3426 	 * Set update flag.
3427 	 */
3428 	NFSLOCKNODE(np);
3429 	np->n_flag |= NUPD;
3430 	vfs_timestamp(&np->n_mtim);
3431 	NFSUNLOCKNODE(np);
3432 	return(fifo_specops.vop_write(ap));
3433 }
3434 
3435 /*
3436  * Close wrapper for fifos.
3437  *
3438  * Update the times on the nfsnode then do fifo close.
3439  */
3440 static int
3441 nfsfifo_close(struct vop_close_args *ap)
3442 {
3443 	struct vnode *vp = ap->a_vp;
3444 	struct nfsnode *np = VTONFS(vp);
3445 	struct vattr vattr;
3446 	struct timespec ts;
3447 
3448 	NFSLOCKNODE(np);
3449 	if (np->n_flag & (NACC | NUPD)) {
3450 		vfs_timestamp(&ts);
3451 		if (np->n_flag & NACC)
3452 			np->n_atim = ts;
3453 		if (np->n_flag & NUPD)
3454 			np->n_mtim = ts;
3455 		np->n_flag |= NCHG;
3456 		if (vrefcnt(vp) == 1 &&
3457 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3458 			VATTR_NULL(&vattr);
3459 			if (np->n_flag & NACC)
3460 				vattr.va_atime = np->n_atim;
3461 			if (np->n_flag & NUPD)
3462 				vattr.va_mtime = np->n_mtim;
3463 			NFSUNLOCKNODE(np);
3464 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3465 			goto out;
3466 		}
3467 	}
3468 	NFSUNLOCKNODE(np);
3469 out:
3470 	return (fifo_specops.vop_close(ap));
3471 }
3472 
3473 /*
3474  * Just call ncl_writebp() with the force argument set to 1.
3475  *
3476  * NOTE: B_DONE may or may not be set in a_bp on call.
3477  */
3478 static int
3479 nfs_bwrite(struct buf *bp)
3480 {
3481 
3482 	return (ncl_writebp(bp, 1, curthread));
3483 }
3484 
3485 struct buf_ops buf_ops_newnfs = {
3486 	.bop_name	=	"buf_ops_nfs",
3487 	.bop_write	=	nfs_bwrite,
3488 	.bop_strategy	=	bufstrategy,
3489 	.bop_sync	=	bufsync,
3490 	.bop_bdflush	=	bufbdflush,
3491 };
3492 
3493 static int
3494 nfs_getacl(struct vop_getacl_args *ap)
3495 {
3496 	int error;
3497 
3498 	if (ap->a_type != ACL_TYPE_NFS4)
3499 		return (EOPNOTSUPP);
3500 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3501 	    NULL);
3502 	if (error > NFSERR_STALE) {
3503 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3504 		error = EPERM;
3505 	}
3506 	return (error);
3507 }
3508 
3509 static int
3510 nfs_setacl(struct vop_setacl_args *ap)
3511 {
3512 	int error;
3513 
3514 	if (ap->a_type != ACL_TYPE_NFS4)
3515 		return (EOPNOTSUPP);
3516 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3517 	    NULL);
3518 	if (error > NFSERR_STALE) {
3519 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3520 		error = EPERM;
3521 	}
3522 	return (error);
3523 }
3524 
3525 /*
3526  * VOP_ADVISE for NFS.
3527  * Just return 0 for any errors, since it is just a hint.
3528  */
3529 static int
3530 nfs_advise(struct vop_advise_args *ap)
3531 {
3532 	struct thread *td = curthread;
3533 	struct nfsmount *nmp;
3534 	uint64_t len;
3535 	int error;
3536 
3537 	/*
3538 	 * First do vop_stdadvise() to handle the buffer cache.
3539 	 */
3540 	error = vop_stdadvise(ap);
3541 	if (error != 0)
3542 		return (error);
3543 	if (ap->a_start < 0 || ap->a_end < 0)
3544 		return (0);
3545 	if (ap->a_end == OFF_MAX)
3546 		len = 0;
3547 	else if (ap->a_end < ap->a_start)
3548 		return (0);
3549 	else
3550 		len = ap->a_end - ap->a_start + 1;
3551 	nmp = VFSTONFS(ap->a_vp->v_mount);
3552 	mtx_lock(&nmp->nm_mtx);
3553 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3554 	    (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) ==
3555 	    0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) {
3556 		mtx_unlock(&nmp->nm_mtx);
3557 		return (0);
3558 	}
3559 	mtx_unlock(&nmp->nm_mtx);
3560 	error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice,
3561 	    td->td_ucred, td);
3562 	if (error == NFSERR_NOTSUPP) {
3563 		mtx_lock(&nmp->nm_mtx);
3564 		nmp->nm_privflag |= NFSMNTP_NOADVISE;
3565 		mtx_unlock(&nmp->nm_mtx);
3566 	}
3567 	return (0);
3568 }
3569 
3570 /*
3571  * nfs allocate call
3572  */
3573 static int
3574 nfs_allocate(struct vop_allocate_args *ap)
3575 {
3576 	struct vnode *vp = ap->a_vp;
3577 	struct thread *td = curthread;
3578 	struct nfsvattr nfsva;
3579 	struct nfsmount *nmp;
3580 	int attrflag, error, ret;
3581 
3582 	attrflag = 0;
3583 	nmp = VFSTONFS(vp->v_mount);
3584 	mtx_lock(&nmp->nm_mtx);
3585 	if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3586 	    (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) {
3587 		mtx_unlock(&nmp->nm_mtx);
3588 		/*
3589 		 * Flush first to ensure that the allocate adds to the
3590 		 * file's allocation on the server.
3591 		 */
3592 		error = ncl_flush(vp, MNT_WAIT, td, 1, 0);
3593 		if (error == 0)
3594 			error = nfsrpc_allocate(vp, *ap->a_offset, *ap->a_len,
3595 			    &nfsva, &attrflag, td->td_ucred, td, NULL);
3596 		if (error == 0) {
3597 			*ap->a_offset += *ap->a_len;
3598 			*ap->a_len = 0;
3599 		} else if (error == NFSERR_NOTSUPP) {
3600 			mtx_lock(&nmp->nm_mtx);
3601 			nmp->nm_privflag |= NFSMNTP_NOALLOCATE;
3602 			mtx_unlock(&nmp->nm_mtx);
3603 		}
3604 	} else {
3605 		mtx_unlock(&nmp->nm_mtx);
3606 		error = EIO;
3607 	}
3608 	/*
3609 	 * If the NFS server cannot perform the Allocate operation, just call
3610 	 * vop_stdallocate() to perform it.
3611 	 */
3612 	if (error != 0)
3613 		error = vop_stdallocate(ap);
3614 	if (attrflag != 0) {
3615 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3616 		if (error == 0 && ret != 0)
3617 			error = ret;
3618 	}
3619 	if (error != 0)
3620 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3621 	return (error);
3622 }
3623 
3624 /*
3625  * nfs copy_file_range call
3626  */
3627 static int
3628 nfs_copy_file_range(struct vop_copy_file_range_args *ap)
3629 {
3630 	struct vnode *invp = ap->a_invp;
3631 	struct vnode *outvp = ap->a_outvp;
3632 	struct mount *mp;
3633 	struct nfsvattr innfsva, outnfsva;
3634 	struct vattr *vap;
3635 	struct uio io;
3636 	struct nfsmount *nmp;
3637 	size_t len, len2, copiedlen;
3638 	int error, inattrflag, outattrflag, ret, ret2;
3639 	off_t inoff, outoff;
3640 	bool consecutive, must_commit, tryoutcred;
3641 
3642 	ret = ret2 = 0;
3643 	nmp = VFSTONFS(invp->v_mount);
3644 	mtx_lock(&nmp->nm_mtx);
3645 	/* NFSv4.2 Copy is not permitted for infile == outfile. */
3646 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3647 	    (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0 || invp == outvp) {
3648 		mtx_unlock(&nmp->nm_mtx);
3649 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3650 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3651 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3652 		return (error);
3653 	}
3654 	mtx_unlock(&nmp->nm_mtx);
3655 
3656 	/* Lock both vnodes, avoiding risk of deadlock. */
3657 	do {
3658 		mp = NULL;
3659 		error = vn_start_write(outvp, &mp, V_WAIT);
3660 		if (error == 0) {
3661 			error = vn_lock(outvp, LK_EXCLUSIVE);
3662 			if (error == 0) {
3663 				error = vn_lock(invp, LK_SHARED | LK_NOWAIT);
3664 				if (error == 0)
3665 					break;
3666 				VOP_UNLOCK(outvp, 0);
3667 				if (mp != NULL)
3668 					vn_finished_write(mp);
3669 				mp = NULL;
3670 				error = vn_lock(invp, LK_SHARED);
3671 				if (error == 0)
3672 					VOP_UNLOCK(invp, 0);
3673 			}
3674 		}
3675 		if (mp != NULL)
3676 			vn_finished_write(mp);
3677 	} while (error == 0);
3678 	if (error != 0)
3679 		return (error);
3680 
3681 	/*
3682 	 * Do the vn_rlimit_fsize() check.  Should this be above the VOP layer?
3683 	 */
3684 	io.uio_offset = *ap->a_outoffp;
3685 	io.uio_resid = *ap->a_lenp;
3686 	error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd);
3687 
3688 	/*
3689 	 * Flush the input file so that the data is up to date before
3690 	 * the copy.  Flush writes for the output file so that they
3691 	 * do not overwrite the data copied to the output file by the Copy.
3692 	 * Set the commit argument for both flushes so that the data is on
3693 	 * stable storage before the Copy RPC.  This is done in case the
3694 	 * server reboots during the Copy and needs to be redone.
3695 	 */
3696 	if (error == 0)
3697 		error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0);
3698 	if (error == 0)
3699 		error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0);
3700 
3701 	/* Do the actual NFSv4.2 RPC. */
3702 	len = *ap->a_lenp;
3703 	mtx_lock(&nmp->nm_mtx);
3704 	if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0)
3705 		consecutive = true;
3706 	else
3707 		consecutive = false;
3708 	mtx_unlock(&nmp->nm_mtx);
3709 	inoff = *ap->a_inoffp;
3710 	outoff = *ap->a_outoffp;
3711 	tryoutcred = true;
3712 	must_commit = false;
3713 	if (error == 0) {
3714 		vap = &VTONFS(invp)->n_vattr.na_vattr;
3715 		error = VOP_GETATTR(invp, vap, ap->a_incred);
3716 		if (error == 0) {
3717 			/*
3718 			 * Clip "len" at va_size so that RFC compliant servers
3719 			 * will not reply NFSERR_INVAL.
3720 			 * Setting "len == 0" for the RPC would be preferred,
3721 			 * but some Linux servers do not support that.
3722 			 */
3723 			if (inoff >= vap->va_size)
3724 				*ap->a_lenp = len = 0;
3725 			else if (inoff + len > vap->va_size)
3726 				*ap->a_lenp = len = vap->va_size - inoff;
3727 		} else
3728 			error = 0;
3729 	}
3730 	copiedlen = 0;
3731 	while (len > 0 && error == 0) {
3732 		inattrflag = outattrflag = 0;
3733 		len2 = len;
3734 		if (tryoutcred)
3735 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3736 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3737 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3738 			    ap->a_outcred, consecutive, &must_commit);
3739 		else
3740 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3741 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3742 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3743 			    ap->a_incred, consecutive, &must_commit);
3744 		if (inattrflag != 0)
3745 			ret = nfscl_loadattrcache(&invp, &innfsva, NULL, NULL,
3746 			    0, 1);
3747 		if (outattrflag != 0)
3748 			ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL,
3749 			    NULL, 1, 1);
3750 		if (error == 0) {
3751 			if (consecutive == false) {
3752 				if (len2 == len) {
3753 					mtx_lock(&nmp->nm_mtx);
3754 					nmp->nm_privflag |=
3755 					    NFSMNTP_NOCONSECUTIVE;
3756 					mtx_unlock(&nmp->nm_mtx);
3757 				} else
3758 					error = NFSERR_OFFLOADNOREQS;
3759 			}
3760 			/*
3761 			 * If the Copy returns a length == 0, it hit the
3762 			 * EOF on the input file.
3763 			 */
3764 			if (len2 == 0) {
3765 				*ap->a_lenp = copiedlen;
3766 				len = 0;
3767 			} else {
3768 				len -= len2;
3769 				copiedlen += len2;
3770 			}
3771 			if (len == 0 && must_commit && error == 0)
3772 				error = ncl_commit(outvp, outoff, *ap->a_lenp,
3773 				    ap->a_outcred, curthread);
3774 			if (error == 0 && ret != 0)
3775 				error = ret;
3776 			if (error == 0 && ret2 != 0)
3777 				error = ret2;
3778 		} else if (error == NFSERR_OFFLOADNOREQS && consecutive) {
3779 			/*
3780 			 * Try consecutive == false, which is ok only if all
3781 			 * bytes are copied.
3782 			 */
3783 			consecutive = false;
3784 			error = 0;
3785 		} else if (error == NFSERR_ACCES && tryoutcred) {
3786 			/* Try again with incred. */
3787 			tryoutcred = false;
3788 			error = 0;
3789 		}
3790 		if (error == NFSERR_STALEWRITEVERF) {
3791 			/*
3792 			 * Server rebooted, so do it all again.
3793 			 */
3794 			*ap->a_inoffp = inoff;
3795 			*ap->a_outoffp = outoff;
3796 			len = *ap->a_lenp;
3797 			must_commit = false;
3798 			error = 0;
3799 		}
3800 	}
3801 	VOP_UNLOCK(invp, 0);
3802 	VOP_UNLOCK(outvp, 0);
3803 	if (mp != NULL)
3804 		vn_finished_write(mp);
3805 	if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS ||
3806 	    error == NFSERR_ACCES) {
3807 		/*
3808 		 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can
3809 		 * use a_incred for the read and a_outcred for the write, so
3810 		 * try this for NFSERR_ACCES failures for the Copy.
3811 		 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can
3812 		 * never succeed, so disable it.
3813 		 */
3814 		if (error != NFSERR_ACCES) {
3815 			/* Can never do Copy on this mount. */
3816 			mtx_lock(&nmp->nm_mtx);
3817 			nmp->nm_privflag |= NFSMNTP_NOCOPY;
3818 			mtx_unlock(&nmp->nm_mtx);
3819 		}
3820 		*ap->a_inoffp = inoff;
3821 		*ap->a_outoffp = outoff;
3822 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3823 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3824 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3825 	} else if (error != 0)
3826 		*ap->a_lenp = 0;
3827 
3828 	if (error != 0)
3829 		error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0);
3830 	return (error);
3831 }
3832 
3833 /*
3834  * nfs ioctl call
3835  */
3836 static int
3837 nfs_ioctl(struct vop_ioctl_args *ap)
3838 {
3839 	struct vnode *vp = ap->a_vp;
3840 	struct nfsvattr nfsva;
3841 	struct nfsmount *nmp;
3842 	int attrflag, content, error, ret;
3843 	bool eof = false;			/* shut up compiler. */
3844 
3845 	if (vp->v_type != VREG)
3846 		return (ENOTTY);
3847 	nmp = VFSTONFS(vp->v_mount);
3848 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) {
3849 		error = vop_stdioctl(ap);
3850 		return (error);
3851 	}
3852 
3853 	/* Do the actual NFSv4.2 RPC. */
3854 	switch (ap->a_command) {
3855 	case FIOSEEKDATA:
3856 		content = NFSV4CONTENT_DATA;
3857 		break;
3858 	case FIOSEEKHOLE:
3859 		content = NFSV4CONTENT_HOLE;
3860 		break;
3861 	default:
3862 		return (ENOTTY);
3863 	}
3864 
3865 	error = vn_lock(vp, LK_SHARED);
3866 	if (error != 0)
3867 		return (EBADF);
3868 	attrflag = 0;
3869 	if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size)
3870 		error = ENXIO;
3871 	else {
3872 		/*
3873 		 * Flush all writes, so that the server is up to date.
3874 		 * Although a Commit is not required, the commit argument
3875 		 * is set so that, for a pNFS File/Flexible File Layout
3876 		 * server, the LayoutCommit will be done to ensure the file
3877 		 * size is up to date on the Metadata Server.
3878 		 */
3879 		error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0);
3880 		if (error == 0)
3881 			error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof,
3882 			    content, ap->a_cred, &nfsva, &attrflag);
3883 		/* If at eof for FIOSEEKDATA, return ENXIO. */
3884 		if (eof && error == 0 && content == NFSV4CONTENT_DATA)
3885 			error = ENXIO;
3886 	}
3887 	if (attrflag != 0) {
3888 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3889 		if (error == 0 && ret != 0)
3890 			error = ret;
3891 	}
3892 	NFSVOPUNLOCK(vp, 0);
3893 
3894 	if (error != 0)
3895 		error = ENXIO;
3896 	return (error);
3897 }
3898 
3899 /*
3900  * nfs getextattr call
3901  */
3902 static int
3903 nfs_getextattr(struct vop_getextattr_args *ap)
3904 {
3905 	struct vnode *vp = ap->a_vp;
3906 	struct nfsmount *nmp;
3907 	struct ucred *cred;
3908 	struct thread *td = ap->a_td;
3909 	struct nfsvattr nfsva;
3910 	ssize_t len;
3911 	int attrflag, error, ret;
3912 
3913 	nmp = VFSTONFS(vp->v_mount);
3914 	mtx_lock(&nmp->nm_mtx);
3915 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3916 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
3917 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
3918 		mtx_unlock(&nmp->nm_mtx);
3919 		return (EOPNOTSUPP);
3920 	}
3921 	mtx_unlock(&nmp->nm_mtx);
3922 
3923 	cred = ap->a_cred;
3924 	if (cred == NULL)
3925 		cred = td->td_ucred;
3926 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
3927 	attrflag = 0;
3928 	error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva,
3929 	    &attrflag, cred, td);
3930 	if (attrflag != 0) {
3931 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3932 		if (error == 0 && ret != 0)
3933 			error = ret;
3934 	}
3935 	if (error == 0 && ap->a_size != NULL)
3936 		*ap->a_size = len;
3937 
3938 	switch (error) {
3939 	case NFSERR_NOTSUPP:
3940 	case NFSERR_OPILLEGAL:
3941 		mtx_lock(&nmp->nm_mtx);
3942 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
3943 		mtx_unlock(&nmp->nm_mtx);
3944 		error = EOPNOTSUPP;
3945 		break;
3946 	case NFSERR_NOXATTR:
3947 	case NFSERR_XATTR2BIG:
3948 		error = ENOATTR;
3949 		break;
3950 	default:
3951 		error = nfscl_maperr(td, error, 0, 0);
3952 		break;
3953 	}
3954 	return (error);
3955 }
3956 
3957 /*
3958  * nfs setextattr call
3959  */
3960 static int
3961 nfs_setextattr(struct vop_setextattr_args *ap)
3962 {
3963 	struct vnode *vp = ap->a_vp;
3964 	struct nfsmount *nmp;
3965 	struct ucred *cred;
3966 	struct thread *td = ap->a_td;
3967 	struct nfsvattr nfsva;
3968 	int attrflag, error, ret;
3969 
3970 	nmp = VFSTONFS(vp->v_mount);
3971 	mtx_lock(&nmp->nm_mtx);
3972 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3973 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
3974 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
3975 		mtx_unlock(&nmp->nm_mtx);
3976 		return (EOPNOTSUPP);
3977 	}
3978 	mtx_unlock(&nmp->nm_mtx);
3979 
3980 	if (ap->a_uio->uio_resid <= 0)
3981 		return (EINVAL);
3982 	cred = ap->a_cred;
3983 	if (cred == NULL)
3984 		cred = td->td_ucred;
3985 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
3986 	attrflag = 0;
3987 	error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva,
3988 	    &attrflag, cred, td);
3989 	if (attrflag != 0) {
3990 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3991 		if (error == 0 && ret != 0)
3992 			error = ret;
3993 	}
3994 
3995 	switch (error) {
3996 	case NFSERR_NOTSUPP:
3997 	case NFSERR_OPILLEGAL:
3998 		mtx_lock(&nmp->nm_mtx);
3999 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4000 		mtx_unlock(&nmp->nm_mtx);
4001 		error = EOPNOTSUPP;
4002 		break;
4003 	case NFSERR_NOXATTR:
4004 	case NFSERR_XATTR2BIG:
4005 		error = ENOATTR;
4006 		break;
4007 	default:
4008 		error = nfscl_maperr(td, error, 0, 0);
4009 		break;
4010 	}
4011 	return (error);
4012 }
4013 
4014 /*
4015  * nfs listextattr call
4016  */
4017 static int
4018 nfs_listextattr(struct vop_listextattr_args *ap)
4019 {
4020 	struct vnode *vp = ap->a_vp;
4021 	struct nfsmount *nmp;
4022 	struct ucred *cred;
4023 	struct thread *td = ap->a_td;
4024 	struct nfsvattr nfsva;
4025 	size_t len, len2;
4026 	uint64_t cookie;
4027 	int attrflag, error, ret;
4028 	bool eof;
4029 
4030 	nmp = VFSTONFS(vp->v_mount);
4031 	mtx_lock(&nmp->nm_mtx);
4032 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4033 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4034 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4035 		mtx_unlock(&nmp->nm_mtx);
4036 		return (EOPNOTSUPP);
4037 	}
4038 	mtx_unlock(&nmp->nm_mtx);
4039 
4040 	cred = ap->a_cred;
4041 	if (cred == NULL)
4042 		cred = td->td_ucred;
4043 
4044 	/* Loop around doing List Extended Attribute RPCs. */
4045 	eof = false;
4046 	cookie = 0;
4047 	len2 = 0;
4048 	error = 0;
4049 	while (!eof && error == 0) {
4050 		len = nmp->nm_rsize;
4051 		attrflag = 0;
4052 		error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof,
4053 		    &nfsva, &attrflag, cred, td);
4054 		if (attrflag != 0) {
4055 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4056 			    1);
4057 			if (error == 0 && ret != 0)
4058 				error = ret;
4059 		}
4060 		if (error == 0) {
4061 			len2 += len;
4062 			if (len2 > SSIZE_MAX)
4063 				error = ENOATTR;
4064 		}
4065 	}
4066 	if (error == 0 && ap->a_size != NULL)
4067 		*ap->a_size = len2;
4068 
4069 	switch (error) {
4070 	case NFSERR_NOTSUPP:
4071 	case NFSERR_OPILLEGAL:
4072 		mtx_lock(&nmp->nm_mtx);
4073 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4074 		mtx_unlock(&nmp->nm_mtx);
4075 		error = EOPNOTSUPP;
4076 		break;
4077 	case NFSERR_NOXATTR:
4078 	case NFSERR_XATTR2BIG:
4079 		error = ENOATTR;
4080 		break;
4081 	default:
4082 		error = nfscl_maperr(td, error, 0, 0);
4083 		break;
4084 	}
4085 	return (error);
4086 }
4087 
4088 /*
4089  * nfs setextattr call
4090  */
4091 static int
4092 nfs_deleteextattr(struct vop_deleteextattr_args *ap)
4093 {
4094 	struct vnode *vp = ap->a_vp;
4095 	struct nfsmount *nmp;
4096 	struct nfsvattr nfsva;
4097 	int attrflag, error, ret;
4098 
4099 	nmp = VFSTONFS(vp->v_mount);
4100 	mtx_lock(&nmp->nm_mtx);
4101 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4102 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4103 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4104 		mtx_unlock(&nmp->nm_mtx);
4105 		return (EOPNOTSUPP);
4106 	}
4107 	mtx_unlock(&nmp->nm_mtx);
4108 
4109 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4110 	attrflag = 0;
4111 	error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred,
4112 	    ap->a_td);
4113 	if (attrflag != 0) {
4114 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4115 		if (error == 0 && ret != 0)
4116 			error = ret;
4117 	}
4118 
4119 	switch (error) {
4120 	case NFSERR_NOTSUPP:
4121 	case NFSERR_OPILLEGAL:
4122 		mtx_lock(&nmp->nm_mtx);
4123 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4124 		mtx_unlock(&nmp->nm_mtx);
4125 		error = EOPNOTSUPP;
4126 		break;
4127 	case NFSERR_NOXATTR:
4128 	case NFSERR_XATTR2BIG:
4129 		error = ENOATTR;
4130 		break;
4131 	default:
4132 		error = nfscl_maperr(ap->a_td, error, 0, 0);
4133 		break;
4134 	}
4135 	return (error);
4136 }
4137 
4138 /*
4139  * Return POSIX pathconf information applicable to nfs filesystems.
4140  */
4141 static int
4142 nfs_pathconf(struct vop_pathconf_args *ap)
4143 {
4144 	struct nfsv3_pathconf pc;
4145 	struct nfsvattr nfsva;
4146 	struct vnode *vp = ap->a_vp;
4147 	struct nfsmount *nmp;
4148 	struct thread *td = curthread;
4149 	off_t off;
4150 	bool eof;
4151 	int attrflag, error;
4152 
4153 	if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
4154 	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
4155 	    ap->a_name == _PC_NO_TRUNC)) ||
4156 	    (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
4157 		/*
4158 		 * Since only the above 4 a_names are returned by the NFSv3
4159 		 * Pathconf RPC, there is no point in doing it for others.
4160 		 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
4161 		 * be used for _PC_NFS4_ACL as well.
4162 		 */
4163 		error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
4164 		    &attrflag, NULL);
4165 		if (attrflag != 0)
4166 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4167 			    1);
4168 		if (error != 0)
4169 			return (error);
4170 	} else {
4171 		/*
4172 		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
4173 		 * just fake them.
4174 		 */
4175 		pc.pc_linkmax = NFS_LINK_MAX;
4176 		pc.pc_namemax = NFS_MAXNAMLEN;
4177 		pc.pc_notrunc = 1;
4178 		pc.pc_chownrestricted = 1;
4179 		pc.pc_caseinsensitive = 0;
4180 		pc.pc_casepreserving = 1;
4181 		error = 0;
4182 	}
4183 	switch (ap->a_name) {
4184 	case _PC_LINK_MAX:
4185 #ifdef _LP64
4186 		*ap->a_retval = pc.pc_linkmax;
4187 #else
4188 		*ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
4189 #endif
4190 		break;
4191 	case _PC_NAME_MAX:
4192 		*ap->a_retval = pc.pc_namemax;
4193 		break;
4194 	case _PC_PIPE_BUF:
4195 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
4196 			*ap->a_retval = PIPE_BUF;
4197 		else
4198 			error = EINVAL;
4199 		break;
4200 	case _PC_CHOWN_RESTRICTED:
4201 		*ap->a_retval = pc.pc_chownrestricted;
4202 		break;
4203 	case _PC_NO_TRUNC:
4204 		*ap->a_retval = pc.pc_notrunc;
4205 		break;
4206 	case _PC_ACL_NFS4:
4207 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
4208 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
4209 			*ap->a_retval = 1;
4210 		else
4211 			*ap->a_retval = 0;
4212 		break;
4213 	case _PC_ACL_PATH_MAX:
4214 		if (NFS_ISV4(vp))
4215 			*ap->a_retval = ACL_MAX_ENTRIES;
4216 		else
4217 			*ap->a_retval = 3;
4218 		break;
4219 	case _PC_PRIO_IO:
4220 		*ap->a_retval = 0;
4221 		break;
4222 	case _PC_SYNC_IO:
4223 		*ap->a_retval = 0;
4224 		break;
4225 	case _PC_ALLOC_SIZE_MIN:
4226 		*ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
4227 		break;
4228 	case _PC_FILESIZEBITS:
4229 		if (NFS_ISV34(vp))
4230 			*ap->a_retval = 64;
4231 		else
4232 			*ap->a_retval = 32;
4233 		break;
4234 	case _PC_REC_INCR_XFER_SIZE:
4235 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4236 		break;
4237 	case _PC_REC_MAX_XFER_SIZE:
4238 		*ap->a_retval = -1; /* means ``unlimited'' */
4239 		break;
4240 	case _PC_REC_MIN_XFER_SIZE:
4241 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4242 		break;
4243 	case _PC_REC_XFER_ALIGN:
4244 		*ap->a_retval = PAGE_SIZE;
4245 		break;
4246 	case _PC_SYMLINK_MAX:
4247 		*ap->a_retval = NFS_MAXPATHLEN;
4248 		break;
4249 	case _PC_MIN_HOLE_SIZE:
4250 		/* Only some NFSv4.2 servers support Seek for Holes. */
4251 		*ap->a_retval = 0;
4252 		nmp = VFSTONFS(vp->v_mount);
4253 		if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) {
4254 			/*
4255 			 * NFSv4.2 doesn't have an attribute for hole size,
4256 			 * so all we can do is see if the Seek operation is
4257 			 * supported and then use f_iosize as a "best guess".
4258 			 */
4259 			mtx_lock(&nmp->nm_mtx);
4260 			if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) {
4261 				mtx_unlock(&nmp->nm_mtx);
4262 				off = 0;
4263 				attrflag = 0;
4264 				error = nfsrpc_seek(vp, &off, &eof,
4265 				    NFSV4CONTENT_HOLE, td->td_ucred, &nfsva,
4266 				    &attrflag);
4267 				if (attrflag != 0)
4268 					nfscl_loadattrcache(&vp, &nfsva,
4269 					    NULL, NULL, 0, 1);
4270 				mtx_lock(&nmp->nm_mtx);
4271 				if (error == NFSERR_NOTSUPP)
4272 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED;
4273 				else
4274 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED |
4275 					    NFSMNTP_SEEK;
4276 				error = 0;
4277 			}
4278 			if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0)
4279 				*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4280 			mtx_unlock(&nmp->nm_mtx);
4281 		}
4282 		break;
4283 
4284 	default:
4285 		error = vop_stdpathconf(ap);
4286 		break;
4287 	}
4288 	return (error);
4289 }
4290 
4291