xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision 572680712c317b81d66475203ac1b9d6bbeca5d5)
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 /*
39  * vnode op calls for Sun NFS version 2, 3 and 4
40  */
41 
42 #include "opt_inet.h"
43 
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/resourcevar.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/extattr.h>
53 #include <sys/filio.h>
54 #include <sys/jail.h>
55 #include <sys/malloc.h>
56 #include <sys/mbuf.h>
57 #include <sys/namei.h>
58 #include <sys/socket.h>
59 #include <sys/vnode.h>
60 #include <sys/dirent.h>
61 #include <sys/fcntl.h>
62 #include <sys/lockf.h>
63 #include <sys/stat.h>
64 #include <sys/sysctl.h>
65 #include <sys/signalvar.h>
66 
67 #include <vm/vm.h>
68 #include <vm/vm_extern.h>
69 #include <vm/vm_object.h>
70 #include <vm/vnode_pager.h>
71 
72 #include <fs/nfs/nfsport.h>
73 #include <fs/nfsclient/nfsnode.h>
74 #include <fs/nfsclient/nfsmount.h>
75 #include <fs/nfsclient/nfs.h>
76 #include <fs/nfsclient/nfs_kdtrace.h>
77 
78 #include <net/if.h>
79 #include <netinet/in.h>
80 #include <netinet/in_var.h>
81 
82 #include <nfs/nfs_lock.h>
83 
84 #ifdef KDTRACE_HOOKS
85 #include <sys/dtrace_bsd.h>
86 
87 dtrace_nfsclient_accesscache_flush_probe_func_t
88 		dtrace_nfscl_accesscache_flush_done_probe;
89 uint32_t	nfscl_accesscache_flush_done_id;
90 
91 dtrace_nfsclient_accesscache_get_probe_func_t
92 		dtrace_nfscl_accesscache_get_hit_probe,
93 		dtrace_nfscl_accesscache_get_miss_probe;
94 uint32_t	nfscl_accesscache_get_hit_id;
95 uint32_t	nfscl_accesscache_get_miss_id;
96 
97 dtrace_nfsclient_accesscache_load_probe_func_t
98 		dtrace_nfscl_accesscache_load_done_probe;
99 uint32_t	nfscl_accesscache_load_done_id;
100 #endif /* !KDTRACE_HOOKS */
101 
102 /* Defs */
103 #define	TRUE	1
104 #define	FALSE	0
105 
106 extern struct nfsstatsv1 nfsstatsv1;
107 extern int nfsrv_useacl;
108 extern int nfscl_debuglevel;
109 NFSCLSTATEMUTEX;
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 int	nfs_get_namedattrdir(struct vnode *, struct componentname *,
118 	    struct vnode **);
119 static vop_lookup_t	nfs_lookup;
120 static vop_create_t	nfs_create;
121 static vop_mknod_t	nfs_mknod;
122 static vop_open_t	nfs_open;
123 static vop_pathconf_t	nfs_pathconf;
124 static vop_close_t	nfs_close;
125 static vop_access_t	nfs_access;
126 static vop_getattr_t	nfs_getattr;
127 static vop_setattr_t	nfs_setattr;
128 static vop_read_t	nfs_read;
129 static vop_fsync_t	nfs_fsync;
130 static vop_remove_t	nfs_remove;
131 static vop_link_t	nfs_link;
132 static vop_rename_t	nfs_rename;
133 static vop_mkdir_t	nfs_mkdir;
134 static vop_rmdir_t	nfs_rmdir;
135 static vop_symlink_t	nfs_symlink;
136 static vop_readdir_t	nfs_readdir;
137 static vop_strategy_t	nfs_strategy;
138 static	int	nfs_lookitup(struct vnode *, char *, int,
139 		    struct ucred *, struct thread *, struct nfsnode **);
140 static	int	nfs_sillyrename(struct vnode *, struct vnode *,
141 		    struct componentname *);
142 static vop_access_t	nfsspec_access;
143 static vop_readlink_t	nfs_readlink;
144 static vop_print_t	nfs_print;
145 static vop_advlock_t	nfs_advlock;
146 static vop_advlockasync_t nfs_advlockasync;
147 static vop_getacl_t nfs_getacl;
148 static vop_setacl_t nfs_setacl;
149 static vop_advise_t nfs_advise;
150 static vop_allocate_t nfs_allocate;
151 static vop_deallocate_t nfs_deallocate;
152 static vop_copy_file_range_t nfs_copy_file_range;
153 static vop_ioctl_t nfs_ioctl;
154 static vop_getextattr_t nfs_getextattr;
155 static vop_setextattr_t nfs_setextattr;
156 static vop_listextattr_t nfs_listextattr;
157 static vop_deleteextattr_t nfs_deleteextattr;
158 static vop_delayed_setsize_t	nfs_delayed_setsize;
159 
160 /*
161  * Global vfs data structures for nfs
162  */
163 
164 static struct vop_vector newnfs_vnodeops_nosig = {
165 	.vop_default =		&default_vnodeops,
166 	.vop_access =		nfs_access,
167 	.vop_advlock =		nfs_advlock,
168 	.vop_advlockasync =	nfs_advlockasync,
169 	.vop_close =		nfs_close,
170 	.vop_create =		nfs_create,
171 	.vop_delayed_setsize =	nfs_delayed_setsize,
172 	.vop_fsync =		nfs_fsync,
173 	.vop_getattr =		nfs_getattr,
174 	.vop_getpages =		ncl_getpages,
175 	.vop_putpages =		ncl_putpages,
176 	.vop_inactive =		ncl_inactive,
177 	.vop_link =		nfs_link,
178 	.vop_lookup =		nfs_lookup,
179 	.vop_mkdir =		nfs_mkdir,
180 	.vop_mknod =		nfs_mknod,
181 	.vop_open =		nfs_open,
182 	.vop_pathconf =		nfs_pathconf,
183 	.vop_print =		nfs_print,
184 	.vop_read =		nfs_read,
185 	.vop_readdir =		nfs_readdir,
186 	.vop_readlink =		nfs_readlink,
187 	.vop_reclaim =		ncl_reclaim,
188 	.vop_remove =		nfs_remove,
189 	.vop_rename =		nfs_rename,
190 	.vop_rmdir =		nfs_rmdir,
191 	.vop_setattr =		nfs_setattr,
192 	.vop_strategy =		nfs_strategy,
193 	.vop_symlink =		nfs_symlink,
194 	.vop_write =		ncl_write,
195 	.vop_getacl =		nfs_getacl,
196 	.vop_setacl =		nfs_setacl,
197 	.vop_advise =		nfs_advise,
198 	.vop_allocate =		nfs_allocate,
199 	.vop_deallocate =	nfs_deallocate,
200 	.vop_copy_file_range =	nfs_copy_file_range,
201 	.vop_ioctl =		nfs_ioctl,
202 	.vop_getextattr =	nfs_getextattr,
203 	.vop_setextattr =	nfs_setextattr,
204 	.vop_listextattr =	nfs_listextattr,
205 	.vop_deleteextattr =	nfs_deleteextattr,
206 };
207 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig);
208 
209 static int
210 nfs_vnodeops_bypass(struct vop_generic_args *a)
211 {
212 
213 	return (vop_sigdefer(&newnfs_vnodeops_nosig, a));
214 }
215 
216 struct vop_vector newnfs_vnodeops = {
217 	.vop_default =		&default_vnodeops,
218 	.vop_bypass =		nfs_vnodeops_bypass,
219 };
220 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops);
221 
222 static struct vop_vector newnfs_fifoops_nosig = {
223 	.vop_default =		&fifo_specops,
224 	.vop_access =		nfsspec_access,
225 	.vop_close =		nfsfifo_close,
226 	.vop_fsync =		nfs_fsync,
227 	.vop_getattr =		nfs_getattr,
228 	.vop_inactive =		ncl_inactive,
229 	.vop_pathconf =		nfs_pathconf,
230 	.vop_print =		nfs_print,
231 	.vop_read =		nfsfifo_read,
232 	.vop_reclaim =		ncl_reclaim,
233 	.vop_setattr =		nfs_setattr,
234 	.vop_write =		nfsfifo_write,
235 };
236 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig);
237 
238 static int
239 nfs_fifoops_bypass(struct vop_generic_args *a)
240 {
241 
242 	return (vop_sigdefer(&newnfs_fifoops_nosig, a));
243 }
244 
245 struct vop_vector newnfs_fifoops = {
246 	.vop_default =		&default_vnodeops,
247 	.vop_bypass =		nfs_fifoops_bypass,
248 };
249 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops);
250 
251 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
252     struct componentname *cnp, struct vattr *vap);
253 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
254     int namelen, struct ucred *cred, struct thread *td, bool silly);
255 static void nfs_removestatus(struct vnode *vp, nfsremove_status file_status,
256     bool silly, struct thread *td);
257 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
258     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
259     char *tnameptr, int tnamelen, bool silly, struct ucred *cred,
260     struct thread *td);
261 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
262     struct componentname *scnp, struct sillyrename *sp);
263 
264 /*
265  * Global variables
266  */
267 SYSCTL_DECL(_vfs_nfs);
268 
269 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
270 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
271 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
272 
273 static int	nfs_prime_access_cache = 0;
274 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
275 	   &nfs_prime_access_cache, 0,
276 	   "Prime NFS ACCESS cache when fetching attributes");
277 
278 static int	newnfs_commit_on_close = 0;
279 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
280     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
281 
282 static int	nfs_clean_pages_on_close = 1;
283 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
284 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
285 
286 int newnfs_directio_enable = 0;
287 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
288 	   &newnfs_directio_enable, 0, "Enable NFS directio");
289 
290 int nfs_keep_dirty_on_error;
291 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
292     &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
293 
294 /*
295  * This sysctl allows other processes to mmap a file that has been opened
296  * O_DIRECT by a process.  In general, having processes mmap the file while
297  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
298  * this by default to prevent DoS attacks - to prevent a malicious user from
299  * opening up files O_DIRECT preventing other users from mmap'ing these
300  * files.  "Protected" environments where stricter consistency guarantees are
301  * required can disable this knob.  The process that opened the file O_DIRECT
302  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
303  * meaningful.
304  */
305 int newnfs_directio_allow_mmap = 1;
306 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
307 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
308 
309 static uint64_t	nfs_maxalloclen = 64 * 1024 * 1024;
310 SYSCTL_U64(_vfs_nfs, OID_AUTO, maxalloclen, CTLFLAG_RW,
311 	   &nfs_maxalloclen, 0, "NFS max allocate/deallocate length");
312 
313 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
314 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
315 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
316 
317 /*
318  * SMP Locking Note :
319  * The list of locks after the description of the lock is the ordering
320  * of other locks acquired with the lock held.
321  * np->n_mtx : Protects the fields in the nfsnode.
322        VM Object Lock
323        VI_MTX (acquired indirectly)
324  * nmp->nm_mtx : Protects the fields in the nfsmount.
325        rep->r_mtx
326  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
327  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
328        nmp->nm_mtx
329        rep->r_mtx
330  * rep->r_mtx : Protects the fields in an nfsreq.
331  */
332 
333 static int
334 nfs_delayed_setsize(struct vop_delayed_setsize_args *ap)
335 {
336 	struct vnode *vp;
337 	struct nfsnode *np;
338 	u_quad_t nsize;
339 
340 	vp = ap->a_vp;
341 	np = VTONFS(vp);
342 	if (np != NULL) {
343 		NFSLOCKNODE(np);
344 		nsize = np->n_size;
345 		NFSUNLOCKNODE(np);
346 		vnode_pager_setsize(vp, nsize);
347 	}
348 	return (0);
349 }
350 
351 static int
352 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
353     struct ucred *cred, u_int32_t *retmode)
354 {
355 	int error = 0, attrflag, i, lrupos;
356 	u_int32_t rmode;
357 	struct nfsnode *np = VTONFS(vp);
358 	struct nfsvattr nfsva;
359 
360 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
361 	    &rmode);
362 	if (attrflag)
363 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
364 	if (!error) {
365 		lrupos = 0;
366 		NFSLOCKNODE(np);
367 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
368 			if (np->n_accesscache[i].uid == cred->cr_uid) {
369 				np->n_accesscache[i].mode = rmode;
370 				np->n_accesscache[i].stamp = time_second;
371 				break;
372 			}
373 			if (i > 0 && np->n_accesscache[i].stamp <
374 			    np->n_accesscache[lrupos].stamp)
375 				lrupos = i;
376 		}
377 		if (i == NFS_ACCESSCACHESIZE) {
378 			np->n_accesscache[lrupos].uid = cred->cr_uid;
379 			np->n_accesscache[lrupos].mode = rmode;
380 			np->n_accesscache[lrupos].stamp = time_second;
381 		}
382 		NFSUNLOCKNODE(np);
383 		if (retmode != NULL)
384 			*retmode = rmode;
385 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
386 	} else if (NFS_ISV4(vp)) {
387 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
388 	}
389 #ifdef KDTRACE_HOOKS
390 	if (error != 0)
391 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
392 		    error);
393 #endif
394 	return (error);
395 }
396 
397 /*
398  * nfs access vnode op.
399  * For nfs version 2, just return ok. File accesses may fail later.
400  * For nfs version 3, use the access rpc to check accessibility. If file modes
401  * are changed on the server, accesses might still fail later.
402  */
403 static int
404 nfs_access(struct vop_access_args *ap)
405 {
406 	struct vnode *vp = ap->a_vp;
407 	int error = 0, i, gotahit;
408 	u_int32_t mode, wmode, rmode;
409 	int v34 = NFS_ISV34(vp);
410 	struct nfsnode *np = VTONFS(vp);
411 
412 	/*
413 	 * Disallow write attempts on filesystems mounted read-only;
414 	 * unless the file is a socket, fifo, or a block or character
415 	 * device resident on the filesystem.
416 	 */
417 	if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
418 	    VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
419 	    VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
420 		switch (vp->v_type) {
421 		case VREG:
422 		case VDIR:
423 		case VLNK:
424 			return (EROFS);
425 		default:
426 			break;
427 		}
428 	}
429 
430 	/*
431 	 * For NFSv4, check for a delegation with an Allow ACE, to see
432 	 * if that permits access.
433 	 */
434 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) != 0) {
435 		error = nfscl_delegacecheck(vp, ap->a_accmode, ap->a_cred);
436 		if (error == 0)
437 			return (error);
438 		error = 0;
439 	}
440 
441 	/*
442 	 * For nfs v3 or v4, check to see if we have done this recently, and if
443 	 * so return our cached result instead of making an ACCESS call.
444 	 * If not, do an access rpc, otherwise you are stuck emulating
445 	 * ufs_access() locally using the vattr. This may not be correct,
446 	 * since the server may apply other access criteria such as
447 	 * client uid-->server uid mapping that we do not know about.
448 	 */
449 	if (v34) {
450 		if (ap->a_accmode & VREAD)
451 			mode = NFSACCESS_READ;
452 		else
453 			mode = 0;
454 		if (vp->v_type != VDIR) {
455 			if (ap->a_accmode & VWRITE)
456 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
457 			if (ap->a_accmode & VAPPEND)
458 				mode |= NFSACCESS_EXTEND;
459 			if (ap->a_accmode & VEXEC)
460 				mode |= NFSACCESS_EXECUTE;
461 			if (ap->a_accmode & VDELETE)
462 				mode |= NFSACCESS_DELETE;
463 		} else {
464 			if (ap->a_accmode & VWRITE)
465 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
466 			if (ap->a_accmode & VAPPEND)
467 				mode |= NFSACCESS_EXTEND;
468 			if (ap->a_accmode & VEXEC)
469 				mode |= NFSACCESS_LOOKUP;
470 			if (ap->a_accmode & VDELETE)
471 				mode |= NFSACCESS_DELETE;
472 			if (ap->a_accmode & VDELETE_CHILD)
473 				mode |= NFSACCESS_MODIFY;
474 		}
475 		/* XXX safety belt, only make blanket request if caching */
476 		if (nfsaccess_cache_timeout > 0) {
477 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
478 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
479 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
480 		} else {
481 			wmode = mode;
482 		}
483 
484 		/*
485 		 * Does our cached result allow us to give a definite yes to
486 		 * this request?
487 		 */
488 		gotahit = 0;
489 		NFSLOCKNODE(np);
490 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
491 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
492 			    if (time_second < (np->n_accesscache[i].stamp
493 				+ nfsaccess_cache_timeout) &&
494 				(np->n_accesscache[i].mode & mode) == mode) {
495 				NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
496 				gotahit = 1;
497 			    }
498 			    break;
499 			}
500 		}
501 		NFSUNLOCKNODE(np);
502 #ifdef KDTRACE_HOOKS
503 		if (gotahit != 0)
504 			KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
505 			    ap->a_cred->cr_uid, mode);
506 		else
507 			KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
508 			    ap->a_cred->cr_uid, mode);
509 #endif
510 		if (gotahit == 0) {
511 			/*
512 			 * Either a no, or a don't know.  Go to the wire.
513 			 */
514 			NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
515 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
516 			    ap->a_cred, &rmode);
517 			if (!error &&
518 			    (rmode & mode) != mode)
519 				error = EACCES;
520 		}
521 		return (error);
522 	} else {
523 		if ((error = nfsspec_access(ap)) != 0) {
524 			return (error);
525 		}
526 		/*
527 		 * Attempt to prevent a mapped root from accessing a file
528 		 * which it shouldn't.  We try to read a byte from the file
529 		 * if the user is root and the file is not zero length.
530 		 * After calling nfsspec_access, we should have the correct
531 		 * file size cached.
532 		 */
533 		NFSLOCKNODE(np);
534 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
535 		    && VTONFS(vp)->n_size > 0) {
536 			struct iovec aiov;
537 			struct uio auio;
538 			char buf[1];
539 
540 			NFSUNLOCKNODE(np);
541 			aiov.iov_base = buf;
542 			aiov.iov_len = 1;
543 			auio.uio_iov = &aiov;
544 			auio.uio_iovcnt = 1;
545 			auio.uio_offset = 0;
546 			auio.uio_resid = 1;
547 			auio.uio_segflg = UIO_SYSSPACE;
548 			auio.uio_rw = UIO_READ;
549 			auio.uio_td = ap->a_td;
550 
551 			if (vp->v_type == VREG)
552 				error = ncl_readrpc(vp, &auio, ap->a_cred);
553 			else if (vp->v_type == VDIR) {
554 				char* bp;
555 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
556 				aiov.iov_base = bp;
557 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
558 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
559 				    ap->a_td);
560 				free(bp, M_TEMP);
561 			} else if (vp->v_type == VLNK)
562 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
563 			else
564 				error = EACCES;
565 		} else
566 			NFSUNLOCKNODE(np);
567 		return (error);
568 	}
569 }
570 
571 /*
572  * nfs open vnode op
573  * Check to see if the type is ok
574  * and that deletion is not in progress.
575  * For paged in text files, you will need to flush the page cache
576  * if consistency is lost.
577  */
578 /* ARGSUSED */
579 static int
580 nfs_open(struct vop_open_args *ap)
581 {
582 	struct vnode *vp = ap->a_vp;
583 	struct nfsnode *np = VTONFS(vp);
584 	struct vattr vattr;
585 	int error;
586 	int fmode = ap->a_mode;
587 	struct ucred *cred;
588 	vm_object_t obj;
589 
590 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
591 		return (EOPNOTSUPP);
592 
593 	/*
594 	 * For NFSv4, we need to do the Open Op before cache validation,
595 	 * so that we conform to RFC3530 Sec. 9.3.1.
596 	 */
597 	if (NFS_ISV4(vp)) {
598 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
599 		if (error) {
600 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
601 			    (gid_t)0);
602 			return (error);
603 		}
604 	}
605 
606 	/*
607 	 * Now, if this Open will be doing reading, re-validate/flush the
608 	 * cache, so that Close/Open coherency is maintained.
609 	 */
610 	NFSLOCKNODE(np);
611 	if (np->n_flag & NMODIFIED) {
612 		NFSUNLOCKNODE(np);
613 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
614 			NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
615 			if (VN_IS_DOOMED(vp))
616 				return (EBADF);
617 		}
618 		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
619 		if (error == EINTR || error == EIO) {
620 			if (NFS_ISV4(vp))
621 				(void) nfsrpc_close(vp, 0, ap->a_td);
622 			return (error);
623 		}
624 		NFSLOCKNODE(np);
625 		np->n_attrstamp = 0;
626 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
627 		if (vp->v_type == VDIR)
628 			np->n_direofoffset = 0;
629 		NFSUNLOCKNODE(np);
630 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
631 		if (error) {
632 			if (NFS_ISV4(vp))
633 				(void) nfsrpc_close(vp, 0, ap->a_td);
634 			return (error);
635 		}
636 		NFSLOCKNODE(np);
637 		np->n_mtime = vattr.va_mtime;
638 		if (NFS_ISV4(vp))
639 			np->n_change = vattr.va_filerev;
640 	} else {
641 		NFSUNLOCKNODE(np);
642 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
643 		if (error) {
644 			if (NFS_ISV4(vp))
645 				(void) nfsrpc_close(vp, 0, ap->a_td);
646 			return (error);
647 		}
648 		NFSLOCKNODE(np);
649 		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
650 		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
651 			if (vp->v_type == VDIR)
652 				np->n_direofoffset = 0;
653 			NFSUNLOCKNODE(np);
654 			if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
655 				NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
656 				if (VN_IS_DOOMED(vp))
657 					return (EBADF);
658 			}
659 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
660 			if (error == EINTR || error == EIO) {
661 				if (NFS_ISV4(vp))
662 					(void) nfsrpc_close(vp, 0, ap->a_td);
663 				return (error);
664 			}
665 			NFSLOCKNODE(np);
666 			np->n_mtime = vattr.va_mtime;
667 			if (NFS_ISV4(vp))
668 				np->n_change = vattr.va_filerev;
669 		}
670 	}
671 
672 	/*
673 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
674 	 */
675 	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
676 	    (vp->v_type == VREG)) {
677 		if (np->n_directio_opens == 0) {
678 			NFSUNLOCKNODE(np);
679 			if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
680 				NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
681 				if (VN_IS_DOOMED(vp))
682 					return (EBADF);
683 			}
684 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
685 			if (error) {
686 				if (NFS_ISV4(vp))
687 					(void) nfsrpc_close(vp, 0, ap->a_td);
688 				return (error);
689 			}
690 			NFSLOCKNODE(np);
691 			np->n_flag |= NNONCACHE;
692 		}
693 		np->n_directio_opens++;
694 	}
695 
696 	/* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
697 	if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
698 		np->n_flag |= NWRITEOPENED;
699 
700 	/*
701 	 * If this is an open for writing, capture a reference to the
702 	 * credentials, so they can be used by ncl_putpages(). Using
703 	 * these write credentials is preferable to the credentials of
704 	 * whatever thread happens to be doing the VOP_PUTPAGES() since
705 	 * the write RPCs are less likely to fail with EACCES.
706 	 */
707 	if ((fmode & FWRITE) != 0) {
708 		cred = np->n_writecred;
709 		np->n_writecred = crhold(ap->a_cred);
710 	} else
711 		cred = NULL;
712 	NFSUNLOCKNODE(np);
713 
714 	if (cred != NULL)
715 		crfree(cred);
716 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
717 
718 	/*
719 	 * If the text file has been mmap'd, flush any dirty pages to the
720 	 * buffer cache and then...
721 	 * Make sure all writes are pushed to the NFS server.  If this is not
722 	 * done, the modify time of the file can change while the text
723 	 * file is being executed.  This will cause the process that is
724 	 * executing the text file to be terminated.
725 	 */
726 	if (vp->v_writecount <= -1) {
727 		if ((obj = vp->v_object) != NULL &&
728 		    vm_object_mightbedirty(obj)) {
729 			if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
730 				NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
731 				if (VN_IS_DOOMED(vp))
732 					return (EBADF);
733 			}
734 			vnode_pager_clean_sync(vp);
735 		}
736 
737 		/* Now, flush the buffer cache. */
738 		ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
739 
740 		/* And, finally, make sure that n_mtime is up to date. */
741 		np = VTONFS(vp);
742 		NFSLOCKNODE(np);
743 		np->n_mtime = np->n_vattr.na_mtime;
744 		NFSUNLOCKNODE(np);
745 	}
746 	return (0);
747 }
748 
749 /*
750  * nfs close vnode op
751  * What an NFS client should do upon close after writing is a debatable issue.
752  * Most NFS clients push delayed writes to the server upon close, basically for
753  * two reasons:
754  * 1 - So that any write errors may be reported back to the client process
755  *     doing the close system call. By far the two most likely errors are
756  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
757  * 2 - To put a worst case upper bound on cache inconsistency between
758  *     multiple clients for the file.
759  * There is also a consistency problem for Version 2 of the protocol w.r.t.
760  * not being able to tell if other clients are writing a file concurrently,
761  * since there is no way of knowing if the changed modify time in the reply
762  * is only due to the write for this client.
763  * (NFS Version 3 provides weak cache consistency data in the reply that
764  *  should be sufficient to detect and handle this case.)
765  *
766  * The current code does the following:
767  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
768  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
769  *                     or commit them (this satisfies 1 and 2 except for the
770  *                     case where the server crashes after this close but
771  *                     before the commit RPC, which is felt to be "good
772  *                     enough". Changing the last argument to ncl_flush() to
773  *                     a 1 would force a commit operation, if it is felt a
774  *                     commit is necessary now.
775  * for NFS Version 4 - flush the dirty buffers and commit them, if
776  *		       nfscl_mustflush() says this is necessary.
777  *                     It is necessary if there is no write delegation held,
778  *                     in order to satisfy open/close coherency.
779  *                     If the file isn't cached on local stable storage,
780  *                     it may be necessary in order to detect "out of space"
781  *                     errors from the server, if the write delegation
782  *                     issued by the server doesn't allow the file to grow.
783  */
784 /* ARGSUSED */
785 static int
786 nfs_close(struct vop_close_args *ap)
787 {
788 	struct vnode *vp = ap->a_vp;
789 	struct nfsnode *np = VTONFS(vp);
790 	struct nfsvattr nfsva;
791 	struct ucred *cred;
792 	int error = 0, ret, localcred = 0;
793 	int fmode = ap->a_fflag;
794 	struct nfsmount *nmp;
795 
796 	if (NFSCL_FORCEDISM(vp->v_mount))
797 		return (0);
798 	nmp = VFSTONFS(vp->v_mount);
799 	/*
800 	 * During shutdown, a_cred isn't valid, so just use root.
801 	 */
802 	if (ap->a_cred == NOCRED) {
803 		cred = newnfs_getcred();
804 		localcred = 1;
805 	} else {
806 		cred = ap->a_cred;
807 	}
808 	if (vp->v_type == VREG) {
809 	    /*
810 	     * Examine and clean dirty pages, regardless of NMODIFIED.
811 	     * This closes a major hole in close-to-open consistency.
812 	     * We want to push out all dirty pages (and buffers) on
813 	     * close, regardless of whether they were dirtied by
814 	     * mmap'ed writes or via write().
815 	     */
816 	    if (nfs_clean_pages_on_close && vp->v_object) {
817 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
818 			NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
819 			if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK)
820 				return (EBADF);
821 		}
822 		vnode_pager_clean_async(vp);
823 	    }
824 	    NFSLOCKNODE(np);
825 	    if (np->n_flag & NMODIFIED) {
826 		NFSUNLOCKNODE(np);
827 		if (NFS_ISV3(vp)) {
828 		    /*
829 		     * Under NFSv3 we have dirty buffers to dispose of.  We
830 		     * must flush them to the NFS server.  We have the option
831 		     * of waiting all the way through the commit rpc or just
832 		     * waiting for the initial write.  The default is to only
833 		     * wait through the initial write so the data is in the
834 		     * server's cache, which is roughly similar to the state
835 		     * a standard disk subsystem leaves the file in on close().
836 		     *
837 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
838 		     * potential races with other processes, and certainly
839 		     * cannot clear it if we don't commit.
840 		     * These races occur when there is no longer the old
841 		     * traditional vnode locking implemented for Vnode Ops.
842 		     */
843 		    int cm = newnfs_commit_on_close ? 1 : 0;
844 		    if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
845 			    NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
846 			    if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK)
847 				    return (EBADF);
848 		    }
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 (!NFSHASNFSV4N(nmp) ||
853 			    (nmp->nm_flag & NFSMNT_NOCTO) == 0 ||
854 			    nfscl_mustflush(vp) != 0) {
855 				int cm = newnfs_commit_on_close ? 1 : 0;
856 				if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
857 					NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
858 					if (VN_IS_DOOMED(vp) && ap->a_fflag !=
859 					    FNONBLOCK)
860 						return (EBADF);
861 				}
862 				error = ncl_flush(vp, MNT_WAIT, ap->a_td,
863 				    cm, 0);
864 				/*
865 				 * as above w.r.t races when clearing
866 				 * NMODIFIED.
867 				 * np->n_flag &= ~NMODIFIED;
868 				 */
869 			}
870 		} else {
871 			if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
872 				NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
873 				if (VN_IS_DOOMED(vp) && ap->a_fflag !=
874 				    FNONBLOCK)
875 					return (EBADF);
876 			}
877 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
878 		}
879 		NFSLOCKNODE(np);
880 	    }
881  	    /*
882  	     * Invalidate the attribute cache in all cases.
883  	     * An open is going to fetch fresh attrs any way, other procs
884  	     * on this node that have file open will be forced to do an
885  	     * otw attr fetch, but this is safe.
886 	     * --> A user found that their RPC count dropped by 20% when
887 	     *     this was commented out and I can't see any requirement
888 	     *     for it, so I've disabled it when negative lookups are
889 	     *     enabled. (What does this have to do with negative lookup
890 	     *     caching? Well nothing, except it was reported by the
891 	     *     same user that needed negative lookup caching and I wanted
892 	     *     there to be a way to disable it to see if it
893 	     *     is the cause of some caching/coherency issue that might
894 	     *     crop up.)
895  	     */
896 	    if (nmp->nm_negnametimeo == 0) {
897 		    np->n_attrstamp = 0;
898 		    KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
899 	    }
900 	    if (np->n_flag & NWRITEERR) {
901 		np->n_flag &= ~NWRITEERR;
902 		error = np->n_error;
903 	    }
904 	    NFSUNLOCKNODE(np);
905 	}
906 
907 	if (NFS_ISV4(vp)) {
908 		/*
909 		 * Get attributes so "change" is up to date.
910 		 */
911 		if (error == 0 && nfscl_nodeleg(vp, 0) != 0 &&
912 		    vp->v_type == VREG &&
913 		    (nmp->nm_flag & NFSMNT_NOCTO) == 0) {
914 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva);
915 			if (!ret) {
916 				np->n_change = nfsva.na_filerev;
917 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
918 				    0, 0);
919 			}
920 		}
921 
922 		/*
923 		 * and do the close.
924 		 */
925 		ret = nfsrpc_close(vp, 0, ap->a_td);
926 		if (!error && ret)
927 			error = ret;
928 		if (error)
929 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
930 			    (gid_t)0);
931 	}
932 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
933 		NFSLOCKNODE(np);
934 		KASSERT((np->n_directio_opens > 0),
935 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
936 		np->n_directio_opens--;
937 		if (np->n_directio_opens == 0)
938 			np->n_flag &= ~NNONCACHE;
939 		NFSUNLOCKNODE(np);
940 	}
941 	if (localcred)
942 		NFSFREECRED(cred);
943 	return (error);
944 }
945 
946 /*
947  * nfs getattr call from vfs.
948  */
949 static int
950 nfs_getattr(struct vop_getattr_args *ap)
951 {
952 	struct vnode *vp = ap->a_vp;
953 	struct thread *td = curthread;	/* XXX */
954 	struct nfsnode *np = VTONFS(vp);
955 	int error = 0;
956 	struct nfsvattr nfsva;
957 	struct vattr *vap = ap->a_vap;
958 	struct vattr vattr;
959 	struct nfsmount *nmp;
960 
961 	nmp = VFSTONFS(vp->v_mount);
962 	/*
963 	 * Update local times for special files.
964 	 */
965 	NFSLOCKNODE(np);
966 	if (np->n_flag & (NACC | NUPD))
967 		np->n_flag |= NCHG;
968 	NFSUNLOCKNODE(np);
969 	/*
970 	 * First look in the cache.
971 	 * For "syskrb5" mounts, nm_fhsize might still be zero and
972 	 * cached attributes should be ignored.
973 	 */
974 	if (nmp->nm_fhsize > 0 && ncl_getattrcache(vp, &vattr) == 0) {
975 		ncl_copy_vattr(vp, vap, &vattr);
976 
977 		/*
978 		 * Get the local modify time for the case of a write
979 		 * delegation.
980 		 */
981 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
982 		return (0);
983 	}
984 
985 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
986 	    nfsaccess_cache_timeout > 0) {
987 		NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
988 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
989 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
990 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
991 			return (0);
992 		}
993 	}
994 
995 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva);
996 	if (error == 0)
997 		error = nfscl_loadattrcache(&vp, &nfsva, vap, 0, 0);
998 	if (!error) {
999 		/*
1000 		 * Get the local modify time for the case of a write
1001 		 * delegation.
1002 		 */
1003 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
1004 	} else if (NFS_ISV4(vp)) {
1005 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1006 	}
1007 	return (error);
1008 }
1009 
1010 /*
1011  * nfs setattr call.
1012  */
1013 static int
1014 nfs_setattr(struct vop_setattr_args *ap)
1015 {
1016 	struct vnode *vp = ap->a_vp;
1017 	struct nfsnode *np = VTONFS(vp);
1018 	struct thread *td = curthread;	/* XXX */
1019 	struct vattr *vap = ap->a_vap;
1020 	int error = 0;
1021 	u_quad_t tsize;
1022 	struct timespec ts;
1023 	struct nfsmount *nmp;
1024 
1025 #ifndef nolint
1026 	tsize = (u_quad_t)0;
1027 #endif
1028 
1029 	/*
1030 	 * Only setting of UF_ARCHIVE, UF_HIDDEN and UF_SYSTEM are supported and
1031 	 * only for NFSv4 servers that support them.
1032 	 */
1033 	nmp = VFSTONFS(vp->v_mount);
1034 	if (vap->va_flags != VNOVAL && (!NFSHASNFSV4(nmp) ||
1035 	    (vap->va_flags & ~(UF_ARCHIVE | UF_HIDDEN | UF_SYSTEM)) != 0 ||
1036 	     !NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, NFSATTRBIT_ARCHIVE) ||
1037 	     !NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, NFSATTRBIT_HIDDEN) ||
1038 	     !NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr, NFSATTRBIT_SYSTEM)))
1039 		return (EOPNOTSUPP);
1040 
1041 	/*
1042 	 * Disallow write attempts if the filesystem is mounted read-only.
1043 	 */
1044 	if ((vap->va_flags != (u_long)VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
1045 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
1046 	    vap->va_mtime.tv_sec != VNOVAL ||
1047 	    vap->va_birthtime.tv_sec != VNOVAL ||
1048 	    vap->va_mode != (mode_t)VNOVAL) &&
1049 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
1050 		return (EROFS);
1051 	if (vap->va_size != VNOVAL) {
1052  		switch (vp->v_type) {
1053  		case VDIR:
1054  			return (EISDIR);
1055  		case VCHR:
1056  		case VBLK:
1057  		case VSOCK:
1058  		case VFIFO:
1059 			if (vap->va_mtime.tv_sec == VNOVAL &&
1060 			    vap->va_atime.tv_sec == VNOVAL &&
1061 			    vap->va_birthtime.tv_sec == VNOVAL &&
1062 			    vap->va_mode == (mode_t)VNOVAL &&
1063 			    vap->va_uid == (uid_t)VNOVAL &&
1064 			    vap->va_gid == (gid_t)VNOVAL)
1065 				return (0);
1066  			vap->va_size = VNOVAL;
1067  			break;
1068  		default:
1069 			/*
1070 			 * Disallow write attempts if the filesystem is
1071 			 * mounted read-only.
1072 			 */
1073 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
1074 				return (EROFS);
1075 			/*
1076 			 *  We run vnode_pager_setsize() early (why?),
1077 			 * we must set np->n_size now to avoid vinvalbuf
1078 			 * V_SAVE races that might setsize a lower
1079 			 * value.
1080 			 */
1081 			NFSLOCKNODE(np);
1082 			tsize = np->n_size;
1083 			NFSUNLOCKNODE(np);
1084 			error = ncl_meta_setsize(vp, td, vap->va_size);
1085 			NFSLOCKNODE(np);
1086  			if (np->n_flag & NMODIFIED) {
1087 			    tsize = np->n_size;
1088 			    NFSUNLOCKNODE(np);
1089 			    error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
1090 			        0 : V_SAVE, td, 1);
1091 			    if (error != 0) {
1092 				    vnode_pager_setsize(vp, tsize);
1093 				    return (error);
1094 			    }
1095 			    /*
1096 			     * Call nfscl_delegmodtime() to set the modify time
1097 			     * locally, as required.
1098 			     */
1099 			    nfscl_delegmodtime(vp, NULL);
1100  			} else
1101 			    NFSUNLOCKNODE(np);
1102 			/*
1103 			 * np->n_size has already been set to vap->va_size
1104 			 * in ncl_meta_setsize(). We must set it again since
1105 			 * nfs_loadattrcache() could be called through
1106 			 * ncl_meta_setsize() and could modify np->n_size.
1107 			 */
1108 			NFSLOCKNODE(np);
1109  			np->n_vattr.na_size = np->n_size = vap->va_size;
1110 			NFSUNLOCKNODE(np);
1111   		}
1112   	} else {
1113 		NFSLOCKNODE(np);
1114 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
1115 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
1116 			NFSUNLOCKNODE(np);
1117 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
1118 			if (error == EINTR || error == EIO)
1119 				return (error);
1120 		} else
1121 			NFSUNLOCKNODE(np);
1122 	}
1123 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
1124 	if (vap->va_size != VNOVAL) {
1125 		if (error == 0) {
1126 			nanouptime(&ts);
1127 			NFSLOCKNODE(np);
1128 			np->n_localmodtime = ts;
1129 			NFSUNLOCKNODE(np);
1130 		} else {
1131 			NFSLOCKNODE(np);
1132 			np->n_size = np->n_vattr.na_size = tsize;
1133 			vnode_pager_setsize(vp, tsize);
1134 			NFSUNLOCKNODE(np);
1135 		}
1136 	}
1137 	if (vap->va_mtime.tv_sec != VNOVAL && error == 0)
1138 		nfscl_delegmodtime(vp, &vap->va_mtime);
1139 	return (error);
1140 }
1141 
1142 /*
1143  * Do an nfs setattr rpc.
1144  */
1145 static int
1146 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1147     struct thread *td)
1148 {
1149 	struct nfsnode *np = VTONFS(vp);
1150 	int error, ret, attrflag, i;
1151 	struct nfsvattr nfsva;
1152 
1153 	if (NFS_ISV34(vp)) {
1154 		NFSLOCKNODE(np);
1155 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1156 			np->n_accesscache[i].stamp = 0;
1157 		np->n_flag |= NDELEGMOD;
1158 		NFSUNLOCKNODE(np);
1159 		KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1160 	}
1161 	error = nfsrpc_setattr(vp, vap, NULL, 0, cred, td, &nfsva, &attrflag);
1162 	if (attrflag) {
1163 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
1164 		if (ret && !error)
1165 			error = ret;
1166 	}
1167 	if (error && NFS_ISV4(vp))
1168 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1169 	return (error);
1170 }
1171 
1172 /*
1173  * Get a named attribute directory for the vnode.
1174  */
1175 static int
1176 nfs_get_namedattrdir(struct vnode *vp, struct componentname *cnp,
1177     struct vnode **vpp)
1178 {
1179 	struct nfsfh *nfhp;
1180 	struct nfsnode *np;
1181 	struct vnode *newvp;
1182 	struct nfsvattr nfsva;
1183 	int attrflag, error;
1184 
1185 	attrflag = 0;
1186 	*vpp = NULL;
1187 	np = VTONFS(vp);
1188 	error = nfsrpc_openattr(VFSTONFS(vp->v_mount), vp, np->n_fhp->nfh_fh,
1189 	    np->n_fhp->nfh_len, (cnp->cn_flags & CREATENAMED),
1190 	    cnp->cn_cred, curthread, &nfsva, &nfhp, &attrflag);
1191 	if (error == NFSERR_NOTSUPP)
1192 		error = ENOATTR;
1193 	if (error == 0)
1194 		error = nfscl_nget(vp->v_mount, vp, nfhp, cnp, curthread, &np,
1195 		    cnp->cn_lkflags);
1196 	if (error != 0)
1197 		return (error);
1198 	newvp = NFSTOV(np);
1199 	vn_irflag_set_cond(newvp, VIRF_NAMEDDIR);
1200 	if (attrflag != 0)
1201 		(void)nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1202 	*vpp = newvp;
1203 	return (0);
1204 }
1205 
1206 /*
1207  * nfs lookup call, one step at a time...
1208  * First look in cache
1209  * If not found, unlock the directory nfsnode and do the rpc
1210  */
1211 static int
1212 nfs_lookup(struct vop_lookup_args *ap)
1213 {
1214 	struct componentname *cnp = ap->a_cnp;
1215 	struct vnode *dvp = ap->a_dvp;
1216 	struct vnode **vpp = ap->a_vpp;
1217 	struct mount *mp = dvp->v_mount;
1218 	uint64_t flags = cnp->cn_flags;
1219 	struct vnode *newvp;
1220 	struct nfsmount *nmp;
1221 	struct nfsnode *np, *newnp;
1222 	int error = 0, attrflag, dattrflag, ltype, ncticks;
1223 	struct thread *td = curthread;
1224 	struct nfsfh *nfhp;
1225 	struct nfsvattr dnfsva, nfsva;
1226 	struct vattr vattr;
1227 	struct timespec nctime, ts;
1228 	uint32_t openmode;
1229 	bool is_nameddir, needs_nameddir, opennamed;
1230 
1231 	dattrflag = 0;
1232 	*vpp = NULL;
1233 	nmp = VFSTONFS(mp);
1234 	opennamed = (flags & (OPENNAMED | ISLASTCN)) == (OPENNAMED | ISLASTCN);
1235 	if (opennamed && (!NFSHASNFSV4(nmp) || !NFSHASNFSV4N(nmp)))
1236 		return (ENOATTR);
1237 	is_nameddir = (vn_irflag_read(dvp) & VIRF_NAMEDDIR) != 0;
1238 	if ((is_nameddir && (flags & ISLASTCN) == 0 && (cnp->cn_namelen > 1 ||
1239 	    *cnp->cn_nameptr != '.')) ||
1240 	    (opennamed && !is_nameddir && (flags & ISDOTDOT) != 0))
1241 		return (ENOATTR);
1242 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1243 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1244 		return (EROFS);
1245 	np = VTONFS(dvp);
1246 
1247 	needs_nameddir = false;
1248 	if (opennamed || is_nameddir) {
1249 		cnp->cn_flags &= ~MAKEENTRY;
1250 		if (!is_nameddir)
1251 			needs_nameddir = true;
1252 	}
1253 
1254 	/*
1255 	 * If the named attribute directory is needed, acquire it now.
1256 	 */
1257 	newvp = NULL;
1258 	if (needs_nameddir) {
1259 		KASSERT(np->n_v4 == NULL, ("nfs_lookup: O_NAMEDATTR when"
1260 		    " n_v4 not NULL"));
1261 		error = nfs_get_namedattrdir(dvp, cnp, &newvp);
1262 		if (error != 0)
1263 			goto handle_error;
1264 		if (cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.') {
1265 			*vpp = newvp;
1266 			return (0);
1267 		}
1268 		dvp = newvp;
1269 		np = VTONFS(dvp);
1270 		newvp = NULL;
1271 	} else if (opennamed && cnp->cn_namelen == 1 &&
1272 	    *cnp->cn_nameptr == '.') {
1273 		vref(dvp);
1274 		*vpp = dvp;
1275 		return (0);
1276 	}
1277 
1278 	if (dvp->v_type != VDIR)
1279 		return (ENOTDIR);
1280 
1281 	/* For NFSv4, wait until any remove is done. */
1282 	NFSLOCKNODE(np);
1283 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1284 		np->n_flag |= NREMOVEWANT;
1285 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1286 	}
1287 	NFSUNLOCKNODE(np);
1288 
1289 	error = vn_dir_check_exec(dvp, cnp);
1290 	if (error != 0)
1291 		return (error);
1292 
1293 	if (!opennamed && !is_nameddir) {
1294 		error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1295 		if (error > 0 && error != ENOENT)
1296 			return (error);
1297 		if (error == -1) {
1298 			/*
1299 			 * Lookups of "." are special and always return the
1300 			 * current directory.  cache_lookup() already handles
1301 			 * associated locking bookkeeping, etc.
1302 			 */
1303 			if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1304 				return (0);
1305 			}
1306 
1307 			/*
1308 			 * We only accept a positive hit in the cache if the
1309 			 * change time of the file matches our cached copy.
1310 			 * Otherwise, we discard the cache entry and fallback
1311 			 * to doing a lookup RPC.  We also only trust cache
1312 			 * entries for less than nm_nametimeo seconds.
1313 			 *
1314 			 * To better handle stale file handles and attributes,
1315 			 * clear the attribute cache of this node if it is a
1316 			 * leaf component, part of an open() call, and not
1317 			 * locally modified before fetching the attributes.
1318 			 * This should allow stale file handles to be detected
1319 			 * here where we can fall back to a LOOKUP RPC to
1320 			 * recover rather than having nfs_open() detect the
1321 			 * stale file handle and failing open(2) with ESTALE.
1322 			 */
1323 			newvp = *vpp;
1324 			newnp = VTONFS(newvp);
1325 			if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1326 			    (flags & (ISLASTCN | ISOPEN)) ==
1327 			     (ISLASTCN | ISOPEN) &&
1328 			    !(newnp->n_flag & NMODIFIED)) {
1329 				NFSLOCKNODE(newnp);
1330 				newnp->n_attrstamp = 0;
1331 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1332 				NFSUNLOCKNODE(newnp);
1333 			}
1334 			if (nfscl_nodeleg(newvp, 0) == 0 ||
1335 			    ((u_int)(ticks - ncticks) <
1336 			    (nmp->nm_nametimeo * hz) &&
1337 			    VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1338 			    timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1339 				NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1340 				return (0);
1341 			}
1342 			cache_purge(newvp);
1343 			if (dvp != newvp)
1344 				vput(newvp);
1345 			else
1346 				vrele(newvp);
1347 			*vpp = NULL;
1348 		} else if (error == ENOENT) {
1349 			if (VN_IS_DOOMED(dvp))
1350 				return (ENOENT);
1351 			/*
1352 			 * We only accept a negative hit in the cache if the
1353 			 * modification time of the parent directory matches
1354 			 * the cached copy in the name cache entry.
1355 			 * Otherwise, we discard all of the negative cache
1356 			 * entries for this directory.  We also only trust
1357 			 * negative cache entries for up to nm_negnametimeo
1358 			 * seconds.
1359 			 */
1360 			if ((u_int)(ticks - ncticks) <
1361 			    (nmp->nm_negnametimeo * hz) &&
1362 			    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1363 			    timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1364 				NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1365 				return (ENOENT);
1366 			}
1367 			cache_purge_negative(dvp);
1368 		}
1369 	}
1370 
1371 	openmode = 0;
1372 #if 0
1373 	/*
1374 	 * The use of LookupOpen breaks some builds.  It is disabled
1375 	 * until that is fixed.
1376 	 */
1377 	/*
1378 	 * If this an NFSv4.1/4.2 mount using the "oneopenown" mount
1379 	 * option, it is possible to do the Open operation in the same
1380 	 * compound as Lookup, so long as delegations are not being
1381 	 * issued.  This saves doing a separate RPC for Open.
1382 	 * For pnfs, do not do this, since the Open+LayoutGet will
1383 	 * be needed as a separate RPC.
1384 	 */
1385 	NFSLOCKMNT(nmp);
1386 	if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && !NFSHASPNFS(nmp) &&
1387 	    (nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 &&
1388 	    (!NFSMNT_RDONLY(mp) || (flags & OPENWRITE) == 0) &&
1389 	    (flags & (ISLASTCN | ISOPEN | OPENNAMED))) == (ISLASTCN | ISOPEN)) {
1390 		if ((flags & OPENREAD) != 0)
1391 			openmode |= NFSV4OPEN_ACCESSREAD;
1392 		if ((flags & OPENWRITE) != 0)
1393 			openmode |= NFSV4OPEN_ACCESSWRITE;
1394 	}
1395 	NFSUNLOCKMNT(nmp);
1396 #endif
1397 
1398 	newvp = NULL;
1399 	NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1400 	nanouptime(&ts);
1401 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1402 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1403 	    openmode);
1404 	if (dattrflag)
1405 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
1406 	if (needs_nameddir) {
1407 		vput(dvp);
1408 		dvp = ap->a_dvp;
1409 	}
1410 handle_error:
1411 	if (error) {
1412 		if (newvp != NULL) {
1413 			vput(newvp);
1414 			*vpp = NULL;
1415 		}
1416 
1417 		if (error != ENOENT) {
1418 			if (NFS_ISV4(dvp))
1419 				error = nfscl_maperr(td, error,
1420 				    (uid_t)0, (gid_t)0);
1421 			return (error);
1422 		}
1423 
1424 		/* The requested file was not found. */
1425 		if ((cnp->cn_nameiop == CREATE ||
1426 		     cnp->cn_nameiop == RENAME) &&
1427 		    (flags & ISLASTCN)) {
1428 			/*
1429 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1430 			 * VWRITE) here instead of just checking
1431 			 * MNT_RDONLY.
1432 			 */
1433 			if (mp->mnt_flag & MNT_RDONLY)
1434 				return (EROFS);
1435 			return (EJUSTRETURN);
1436 		}
1437 
1438 		if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag &&
1439 		    !NFSHASCASEINSENSITIVE(nmp)) {
1440 			/*
1441 			 * Cache the modification time of the parent
1442 			 * directory from the post-op attributes in
1443 			 * the name cache entry.  The negative cache
1444 			 * entry will be ignored once the directory
1445 			 * has changed.  Don't bother adding the entry
1446 			 * if the directory has already changed.
1447 			 */
1448 			NFSLOCKNODE(np);
1449 			if (timespeccmp(&np->n_vattr.na_mtime,
1450 			    &dnfsva.na_mtime, ==)) {
1451 				NFSUNLOCKNODE(np);
1452 				cache_enter_time(dvp, NULL, cnp,
1453 				    &dnfsva.na_mtime, NULL);
1454 			} else
1455 				NFSUNLOCKNODE(np);
1456 		}
1457 		return (ENOENT);
1458 	}
1459 
1460 	/*
1461 	 * Handle RENAME case...
1462 	 */
1463 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1464 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1465 			free(nfhp, M_NFSFH);
1466 			return (EISDIR);
1467 		}
1468 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np,
1469 		    LK_EXCLUSIVE);
1470 		if (error)
1471 			return (error);
1472 		newvp = NFSTOV(np);
1473 		/*
1474 		 * If n_localmodtime >= time before RPC, then
1475 		 * a file modification operation, such as
1476 		 * VOP_SETATTR() of size, has occurred while
1477 		 * the Lookup RPC and acquisition of the vnode
1478 		 * happened.  As such, the attributes might
1479 		 * be stale, with possibly an incorrect size.
1480 		 */
1481 		NFSLOCKNODE(np);
1482 		if (timespecisset(&np->n_localmodtime) &&
1483 		    timespeccmp(&np->n_localmodtime, &ts, >=)) {
1484 			NFSCL_DEBUG(4, "nfs_lookup: rename localmod "
1485 			    "stale attributes\n");
1486 			attrflag = 0;
1487 		}
1488 		NFSUNLOCKNODE(np);
1489 		if (attrflag)
1490 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1491 			    0, 1);
1492 		*vpp = newvp;
1493 		return (0);
1494 	}
1495 
1496 	if (flags & ISDOTDOT) {
1497 		ltype = NFSVOPISLOCKED(dvp);
1498 		error = vfs_busy(mp, MBF_NOWAIT);
1499 		if (error != 0) {
1500 			vfs_ref(mp);
1501 			NFSVOPUNLOCK(dvp);
1502 			error = vfs_busy(mp, 0);
1503 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1504 			vfs_rel(mp);
1505 			if (error == 0 && VN_IS_DOOMED(dvp)) {
1506 				vfs_unbusy(mp);
1507 				error = ENOENT;
1508 			}
1509 			if (error != 0)
1510 				return (error);
1511 		}
1512 		NFSVOPUNLOCK(dvp);
1513 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np,
1514 		    cnp->cn_lkflags);
1515 		if (error == 0)
1516 			newvp = NFSTOV(np);
1517 		vfs_unbusy(mp);
1518 		if (newvp != dvp)
1519 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1520 		if (VN_IS_DOOMED(dvp)) {
1521 			if (error == 0) {
1522 				if (newvp == dvp)
1523 					vrele(newvp);
1524 				else
1525 					vput(newvp);
1526 			}
1527 			error = ENOENT;
1528 		}
1529 		if (error != 0)
1530 			return (error);
1531 		if (attrflag)
1532 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1533 			    0, 1);
1534 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1535 		free(nfhp, M_NFSFH);
1536 		vref(dvp);
1537 		newvp = dvp;
1538 		if (attrflag)
1539 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1540 			    0, 1);
1541 	} else {
1542 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np,
1543 		    cnp->cn_lkflags);
1544 		if (error)
1545 			return (error);
1546 		newvp = NFSTOV(np);
1547 		if (opennamed)
1548 			vn_irflag_set_cond(newvp, VIRF_NAMEDATTR);
1549 		/*
1550 		 * If n_localmodtime >= time before RPC, then
1551 		 * a file modification operation, such as
1552 		 * VOP_SETATTR() of size, has occurred while
1553 		 * the Lookup RPC and acquisition of the vnode
1554 		 * happened.  As such, the attributes might
1555 		 * be stale, with possibly an incorrect size.
1556 		 */
1557 		NFSLOCKNODE(np);
1558 		if (timespecisset(&np->n_localmodtime) &&
1559 		    timespeccmp(&np->n_localmodtime, &ts, >=)) {
1560 			NFSCL_DEBUG(4, "nfs_lookup: localmod "
1561 			    "stale attributes\n");
1562 			attrflag = 0;
1563 		}
1564 		NFSUNLOCKNODE(np);
1565 		if (attrflag)
1566 			(void)nfscl_loadattrcache(&newvp, &nfsva, NULL,
1567 			    0, 1);
1568 		else if ((flags & (ISLASTCN | ISOPEN)) ==
1569 		    (ISLASTCN | ISOPEN) &&
1570 		    !(np->n_flag & NMODIFIED)) {
1571 			/*
1572 			 * Flush the attribute cache when opening a
1573 			 * leaf node to ensure that fresh attributes
1574 			 * are fetched in nfs_open() since we did not
1575 			 * fetch attributes from the LOOKUP reply.
1576 			 */
1577 			NFSLOCKNODE(np);
1578 			np->n_attrstamp = 0;
1579 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1580 			NFSUNLOCKNODE(np);
1581 		}
1582 	}
1583 	if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp &&
1584 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1585 	    attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1586 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1587 		    newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1588 	*vpp = newvp;
1589 	return (0);
1590 }
1591 
1592 /*
1593  * nfs read call.
1594  * Just call ncl_bioread() to do the work.
1595  */
1596 static int
1597 nfs_read(struct vop_read_args *ap)
1598 {
1599 	struct vnode *vp = ap->a_vp;
1600 
1601 	switch (vp->v_type) {
1602 	case VREG:
1603 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1604 	case VDIR:
1605 		return (EISDIR);
1606 	default:
1607 		return (EOPNOTSUPP);
1608 	}
1609 }
1610 
1611 /*
1612  * nfs readlink call
1613  */
1614 static int
1615 nfs_readlink(struct vop_readlink_args *ap)
1616 {
1617 	struct vnode *vp = ap->a_vp;
1618 
1619 	if (vp->v_type != VLNK)
1620 		return (EINVAL);
1621 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1622 }
1623 
1624 /*
1625  * Do a readlink rpc.
1626  * Called by ncl_doio() from below the buffer cache.
1627  */
1628 int
1629 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1630 {
1631 	int error, ret, attrflag;
1632 	struct nfsvattr nfsva;
1633 
1634 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1635 	    &attrflag);
1636 	if (attrflag) {
1637 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
1638 		if (ret && !error)
1639 			error = ret;
1640 	}
1641 	if (error && NFS_ISV4(vp))
1642 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1643 	return (error);
1644 }
1645 
1646 /*
1647  * nfs read rpc call
1648  * Ditto above
1649  */
1650 int
1651 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1652 {
1653 	int error, ret, attrflag;
1654 	struct nfsvattr nfsva;
1655 	struct nfsmount *nmp;
1656 
1657 	nmp = VFSTONFS(vp->v_mount);
1658 	error = EIO;
1659 	attrflag = 0;
1660 	if (NFSHASPNFS(nmp))
1661 		error = nfscl_doiods(vp, uiop, NULL, NULL,
1662 		    NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1663 	NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1664 	if (error != 0 && error != EFAULT)
1665 		error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1666 		    &attrflag);
1667 	if (attrflag) {
1668 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
1669 		if (ret && !error)
1670 			error = ret;
1671 	}
1672 	if (error && NFS_ISV4(vp))
1673 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1674 	return (error);
1675 }
1676 
1677 /*
1678  * nfs write call
1679  */
1680 int
1681 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1682     int *iomode, int *must_commit, int called_from_strategy, int ioflag)
1683 {
1684 	struct nfsvattr nfsva;
1685 	int error, attrflag, ret;
1686 	struct nfsmount *nmp;
1687 
1688 	nmp = VFSTONFS(vp->v_mount);
1689 	error = EIO;
1690 	attrflag = 0;
1691 	if (NFSHASPNFS(nmp))
1692 		error = nfscl_doiods(vp, uiop, iomode, must_commit,
1693 		    NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1694 	NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1695 	if (error != 0 && error != EFAULT)
1696 		error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1697 		    uiop->uio_td, &nfsva, &attrflag, called_from_strategy,
1698 		    ioflag);
1699 	if (attrflag) {
1700 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1701 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 1, 1);
1702 		else
1703 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
1704 		if (ret && !error)
1705 			error = ret;
1706 	}
1707 	if (DOINGASYNC(vp))
1708 		*iomode = NFSWRITE_FILESYNC;
1709 	if (error && NFS_ISV4(vp))
1710 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1711 	return (error);
1712 }
1713 
1714 /*
1715  * nfs mknod rpc
1716  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1717  * mode set to specify the file type and the size field for rdev.
1718  */
1719 static int
1720 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1721     struct vattr *vap)
1722 {
1723 	struct nfsvattr nfsva, dnfsva;
1724 	struct vnode *newvp = NULL;
1725 	struct nfsnode *np = NULL, *dnp;
1726 	struct nfsfh *nfhp;
1727 	struct vattr vattr;
1728 	int error = 0, attrflag, dattrflag;
1729 	u_int32_t rdev;
1730 
1731 	if (VATTR_ISDEV(vap))
1732 		rdev = vap->va_rdev;
1733 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1734 		rdev = 0xffffffff;
1735 	else
1736 		return (EOPNOTSUPP);
1737 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1738 		return (error);
1739 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1740 	    rdev, vap->va_type, cnp->cn_cred, curthread, &dnfsva,
1741 	    &nfsva, &nfhp, &attrflag, &dattrflag);
1742 	if (!error) {
1743 		if (!nfhp)
1744 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1745 			    cnp->cn_namelen, cnp->cn_cred, curthread,
1746 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0);
1747 		if (nfhp)
1748 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1749 			    curthread, &np, LK_EXCLUSIVE);
1750 	}
1751 	if (dattrflag)
1752 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
1753 	if (!error) {
1754 		newvp = NFSTOV(np);
1755 		if (attrflag != 0) {
1756 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1757 			if (error != 0)
1758 				vput(newvp);
1759 		}
1760 	}
1761 	if (!error) {
1762 		*vpp = newvp;
1763 	} else if (NFS_ISV4(dvp)) {
1764 		error = nfscl_maperr(curthread, error, vap->va_uid,
1765 		    vap->va_gid);
1766 	}
1767 	dnp = VTONFS(dvp);
1768 	NFSLOCKNODE(dnp);
1769 	dnp->n_flag |= NMODIFIED;
1770 	if (!dattrflag) {
1771 		dnp->n_attrstamp = 0;
1772 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1773 	}
1774 	NFSUNLOCKNODE(dnp);
1775 	return (error);
1776 }
1777 
1778 /*
1779  * nfs mknod vop
1780  * just call nfs_mknodrpc() to do the work.
1781  */
1782 /* ARGSUSED */
1783 static int
1784 nfs_mknod(struct vop_mknod_args *ap)
1785 {
1786 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1787 }
1788 
1789 static struct mtx nfs_cverf_mtx;
1790 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1791     MTX_DEF);
1792 
1793 static nfsquad_t
1794 nfs_get_cverf(void)
1795 {
1796 	static nfsquad_t cverf;
1797 	nfsquad_t ret;
1798 	static int cverf_initialized = 0;
1799 
1800 	mtx_lock(&nfs_cverf_mtx);
1801 	if (cverf_initialized == 0) {
1802 		cverf.lval[0] = arc4random();
1803 		cverf.lval[1] = arc4random();
1804 		cverf_initialized = 1;
1805 	} else
1806 		cverf.qval++;
1807 	ret = cverf;
1808 	mtx_unlock(&nfs_cverf_mtx);
1809 
1810 	return (ret);
1811 }
1812 
1813 /*
1814  * nfs file create call
1815  */
1816 static int
1817 nfs_create(struct vop_create_args *ap)
1818 {
1819 	struct vnode *dvp = ap->a_dvp;
1820 	struct vattr *vap = ap->a_vap;
1821 	struct componentname *cnp = ap->a_cnp;
1822 	struct nfsnode *np = NULL, *dnp;
1823 	struct vnode *newvp = NULL;
1824 	struct nfsmount *nmp;
1825 	struct nfsvattr dnfsva, nfsva;
1826 	struct nfsfh *nfhp;
1827 	nfsquad_t cverf;
1828 	int error = 0, attrflag, dattrflag, fmode = 0;
1829 	struct vattr vattr;
1830 	bool is_nameddir, needs_nameddir, opennamed;
1831 
1832 	/*
1833 	 * Oops, not for me..
1834 	 */
1835 	if (vap->va_type == VSOCK)
1836 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1837 
1838 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1839 		return (error);
1840 	if (vap->va_vaflags & VA_EXCLUSIVE)
1841 		fmode |= O_EXCL;
1842 	dnp = VTONFS(dvp);
1843 	nmp = VFSTONFS(dvp->v_mount);
1844 	needs_nameddir = false;
1845 	if (NFSHASNFSV4(nmp) && NFSHASNFSV4N(nmp)) {
1846 		opennamed = (cnp->cn_flags & (OPENNAMED | ISLASTCN)) ==
1847 		    (OPENNAMED | ISLASTCN);
1848 		is_nameddir = (vn_irflag_read(dvp) & VIRF_NAMEDDIR) != 0;
1849 		if (opennamed || is_nameddir) {
1850 			cnp->cn_flags &= ~MAKEENTRY;
1851 			if (!is_nameddir)
1852 				needs_nameddir = true;
1853 		}
1854 	}
1855 
1856 	/*
1857 	 * If the named attribute directory is needed, acquire it now.
1858 	 */
1859 	if (needs_nameddir) {
1860 		KASSERT(dnp->n_v4 == NULL, ("nfs_create: O_NAMEDATTR when"
1861 		    " n_v4 not NULL"));
1862 		error = nfs_get_namedattrdir(dvp, cnp, &newvp);
1863 		if (error != 0)
1864 			return (error);
1865 		dvp = newvp;
1866 		dnp = VTONFS(dvp);
1867 		newvp = NULL;
1868 	}
1869 
1870 again:
1871 	/* For NFSv4, wait until any remove is done. */
1872 	NFSLOCKNODE(dnp);
1873 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1874 		dnp->n_flag |= NREMOVEWANT;
1875 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1876 	}
1877 	NFSUNLOCKNODE(dnp);
1878 
1879 	cverf = nfs_get_cverf();
1880 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1881 	    vap, cverf, fmode, cnp->cn_cred, curthread, &dnfsva, &nfsva,
1882 	    &nfhp, &attrflag, &dattrflag);
1883 	if (!error) {
1884 		if (nfhp == NULL)
1885 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1886 			    cnp->cn_namelen, cnp->cn_cred, curthread,
1887 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0);
1888 		if (nfhp != NULL)
1889 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1890 			    curthread, &np, LK_EXCLUSIVE);
1891 	}
1892 	if (dattrflag)
1893 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
1894 	if (!error) {
1895 		newvp = NFSTOV(np);
1896 		if (attrflag == 0)
1897 			error = nfsrpc_getattr(newvp, cnp->cn_cred, curthread,
1898 			    &nfsva);
1899 		if (error == 0)
1900 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1901 	}
1902 	if (error) {
1903 		if (newvp != NULL) {
1904 			vput(newvp);
1905 			newvp = NULL;
1906 		}
1907 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1908 		    error == NFSERR_NOTSUPP) {
1909 			fmode &= ~O_EXCL;
1910 			goto again;
1911 		}
1912 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1913 		if (nfscl_checksattr(vap, &nfsva)) {
1914 			error = nfsrpc_setattr(newvp, vap, NULL, 0,
1915 			    cnp->cn_cred, curthread, &nfsva, &attrflag);
1916 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1917 			    vap->va_gid != (gid_t)VNOVAL)) {
1918 				/* try again without setting uid/gid */
1919 				vap->va_uid = (uid_t)VNOVAL;
1920 				vap->va_gid = (uid_t)VNOVAL;
1921 				error = nfsrpc_setattr(newvp, vap, NULL, 0,
1922 				    cnp->cn_cred, curthread, &nfsva, &attrflag);
1923 			}
1924 			if (attrflag)
1925 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1926 				    0, 1);
1927 			if (error != 0)
1928 				vput(newvp);
1929 		}
1930 	}
1931 	if (!error) {
1932 		if ((cnp->cn_flags & MAKEENTRY) && attrflag) {
1933 			if (dvp != newvp)
1934 				cache_enter_time(dvp, newvp, cnp,
1935 				    &nfsva.na_ctime, NULL);
1936 			else
1937 				printf("nfs_create: bogus NFS server returned "
1938 				    "the directory as the new file object\n");
1939 		}
1940 		*ap->a_vpp = newvp;
1941 	} else if (NFS_ISV4(dvp)) {
1942 		error = nfscl_maperr(curthread, error, vap->va_uid,
1943 		    vap->va_gid);
1944 	}
1945 	NFSLOCKNODE(dnp);
1946 	dnp->n_flag |= NMODIFIED;
1947 	if (!dattrflag) {
1948 		dnp->n_attrstamp = 0;
1949 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1950 	}
1951 	NFSUNLOCKNODE(dnp);
1952 	if (needs_nameddir)
1953 		vput(dvp);
1954 	return (error);
1955 }
1956 
1957 /*
1958  * nfs file remove call
1959  * To try and make nfs semantics closer to ufs semantics, a file that has
1960  * other processes using the vnode is renamed instead of removed and then
1961  * removed later on the last close.
1962  * - If v_usecount > 1
1963  *	  If a rename is not already in the works
1964  *	     call nfs_sillyrename() to set it up
1965  *     else
1966  *	  do the remove rpc
1967  */
1968 static int
1969 nfs_remove(struct vop_remove_args *ap)
1970 {
1971 	struct vnode *vp = ap->a_vp;
1972 	struct vnode *dvp = ap->a_dvp;
1973 	struct componentname *cnp = ap->a_cnp;
1974 	struct nfsnode *np = VTONFS(vp);
1975 	int error = 0;
1976 	struct vattr vattr;
1977 	struct nfsmount *nmp;
1978 
1979 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1980 	if (vp->v_type == VDIR)
1981 		error = EPERM;
1982 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1983 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1984 	    vattr.va_nlink > 1)) {
1985 		nmp = VFSTONFS(vp->v_mount);
1986 		/*
1987 		 * Purge the name cache so that the chance of a lookup for
1988 		 * the name succeeding while the remove is in progress is
1989 		 * minimized. Without node locking it can still happen, such
1990 		 * that an I/O op returns ESTALE, but since you get this if
1991 		 * another host removes the file..
1992 		 */
1993 		cache_purge(vp);
1994 		/*
1995 		 * throw away biocache buffers, mainly to avoid
1996 		 * unnecessary delayed writes later.
1997 		 * Flushing here would be more correct for the case
1998 		 * where nfs_close() did not do a flush.  However, it
1999 		 * could be a large performance hit for some servers
2000 		 * and only matters when the file name being removed is
2001 		 * one of multiple hard links.
2002 		 */
2003 		if (!NFSHASNFSV4(nmp) || !NFSHASNFSV4N(nmp) ||
2004 		    (nmp->nm_flag & NFSMNT_NOCTO) == 0)
2005 			error = ncl_vinvalbuf(vp, 0, curthread, 1);
2006 		if (error != EINTR && error != EIO)
2007 			/* Do the rpc */
2008 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
2009 			    cnp->cn_namelen, cnp->cn_cred, curthread, false);
2010 		/*
2011 		 * Kludge City: If the first reply to the remove rpc is lost..
2012 		 *   the reply to the retransmitted request will be ENOENT
2013 		 *   since the file was in fact removed
2014 		 *   Therefore, we cheat and return success.
2015 		 */
2016 		if (error == ENOENT)
2017 			error = 0;
2018 	} else if (!np->n_sillyrename)
2019 		error = nfs_sillyrename(dvp, vp, cnp);
2020 	NFSLOCKNODE(np);
2021 	np->n_attrstamp = 0;
2022 	NFSUNLOCKNODE(np);
2023 	KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2024 	return (error);
2025 }
2026 
2027 /*
2028  * nfs file remove rpc called from nfs_inactive
2029  */
2030 int
2031 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
2032 {
2033 	/*
2034 	 * Make sure that the directory vnode is still valid.
2035 	 * XXX we should lock sp->s_dvp here.
2036 	 */
2037 	if (sp->s_dvp->v_type == VBAD)
2038 		return (0);
2039 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
2040 	    sp->s_cred, NULL, true));
2041 }
2042 
2043 /*
2044  * Handle the nfsremove_status reply from the RPC function.
2045  */
2046 static void
2047 nfs_removestatus(struct vnode *vp, nfsremove_status file_status,
2048     bool silly, struct thread *td)
2049 {
2050 
2051 	switch (file_status) {
2052 	case NLINK_ZERO:
2053 		/* Get rid of any delegation. */
2054 		nfscl_delegreturnvp(vp, false, td);
2055 		/* FALLTHROUGH */
2056 	case DELETED:
2057 		/* Throw away buffer cache blocks. */
2058 		(void)ncl_vinvalbuf(vp, 0, td, 1);
2059 		break;
2060 	case VALID:
2061 		/* Nothing to do, delegation is still ok. */
2062 		break;
2063 	default:
2064 		break;
2065 	}
2066 }
2067 
2068 /*
2069  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
2070  */
2071 static int
2072 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
2073     int namelen, struct ucred *cred, struct thread *td, bool silly)
2074 {
2075 	struct nfsvattr dnfsva, nfsva;
2076 	struct nfsnode *dnp = VTONFS(dvp);
2077 	struct nfsmount *nmp;
2078 	int attrflag, error = 0, dattrflag;
2079 	nfsremove_status file_status;
2080 
2081 	nmp = VFSTONFS(dvp->v_mount);
2082 	NFSLOCKNODE(dnp);
2083 	dnp->n_flag |= NREMOVEINPROG;
2084 	NFSUNLOCKNODE(dnp);
2085 	error = nfsrpc_remove(dvp, name, namelen, vp, &nfsva, &attrflag,
2086 	    &file_status, &dnfsva, &dattrflag, cred, td);
2087 	NFSLOCKNODE(dnp);
2088 	if ((dnp->n_flag & NREMOVEWANT)) {
2089 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
2090 		NFSUNLOCKNODE(dnp);
2091 		wakeup((caddr_t)dnp);
2092 	} else {
2093 		dnp->n_flag &= ~NREMOVEINPROG;
2094 		NFSUNLOCKNODE(dnp);
2095 	}
2096 
2097 	if (NFSHASNFSV4(nmp) && NFSHASNFSV4N(nmp)) {
2098 		if (file_status != DELETED && attrflag != 0)
2099 			(void)nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
2100 		if ((nmp->nm_flag & NFSMNT_NOCTO) != 0)
2101 			nfs_removestatus(vp, file_status, silly, td);
2102 	}
2103 
2104 	if (dattrflag != 0)
2105 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
2106 	NFSLOCKNODE(dnp);
2107 	dnp->n_flag |= NMODIFIED;
2108 	if (dattrflag == 0) {
2109 		dnp->n_attrstamp = 0;
2110 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2111 	}
2112 	NFSUNLOCKNODE(dnp);
2113 	if (error && NFS_ISV4(dvp))
2114 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2115 	return (error);
2116 }
2117 
2118 /*
2119  * nfs file rename call
2120  */
2121 static int
2122 nfs_rename(struct vop_rename_args *ap)
2123 {
2124 	struct vnode *fvp = ap->a_fvp;
2125 	struct vnode *tvp = ap->a_tvp;
2126 	struct vnode *fdvp = ap->a_fdvp;
2127 	struct vnode *tdvp = ap->a_tdvp;
2128 	struct componentname *tcnp = ap->a_tcnp;
2129 	struct componentname *fcnp = ap->a_fcnp;
2130 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
2131 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
2132 	struct nfsv4node *newv4 = NULL;
2133 	struct nfsmount *nmp;
2134 	int error;
2135 
2136 	/* Check for cross-device rename */
2137 	if ((fvp->v_mount != tdvp->v_mount) ||
2138 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
2139 		error = EXDEV;
2140 		goto out;
2141 	}
2142 
2143 	if (ap->a_flags != 0) {
2144 		error = EOPNOTSUPP;
2145 		goto out;
2146 	}
2147 
2148 	nmp = VFSTONFS(fvp->v_mount);
2149 
2150 	if (fvp == tvp) {
2151 		printf("nfs_rename: fvp == tvp (can't happen)\n");
2152 		error = 0;
2153 		goto out;
2154 	}
2155 	if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
2156 		goto out;
2157 
2158 	/*
2159 	 * For case insensitive file systems, there may be multiple
2160 	 * names cached for the one name being rename'd, so purge
2161 	 * all names from the cache.
2162 	 */
2163 	if (NFSHASCASEINSENSITIVE(nmp))
2164 		cache_purge(fvp);
2165 
2166 	/*
2167 	 * We have to flush B_DELWRI data prior to renaming
2168 	 * the file.  If we don't, the delayed-write buffers
2169 	 * can be flushed out later after the file has gone stale
2170 	 * under NFSV3.  NFSV2 does not have this problem because
2171 	 * ( as far as I can tell ) it flushes dirty buffers more
2172 	 * often.
2173 	 *
2174 	 * Skip the rename operation if the fsync fails, this can happen
2175 	 * due to the server's volume being full, when we pushed out data
2176 	 * that was written back to our cache earlier. Not checking for
2177 	 * this condition can result in potential (silent) data loss.
2178 	 */
2179 	if ((nmp->nm_flag & NFSMNT_NOCTO) == 0 || !NFSHASNFSV4(nmp) ||
2180 	    !NFSHASNFSV4N(nmp) || nfscl_mustflush(fvp) != 0)
2181 		error = VOP_FSYNC(fvp, MNT_WAIT, curthread);
2182 
2183 	NFSVOPUNLOCK(fvp);
2184 	if (error == 0 && tvp != NULL && ((nmp->nm_flag & NFSMNT_NOCTO) == 0 ||
2185 	    !NFSHASNFSV4(nmp) || !NFSHASNFSV4N(nmp) ||
2186 	    nfscl_mustflush(tvp) != 0))
2187 		error = VOP_FSYNC(tvp, MNT_WAIT, curthread);
2188 	if (error != 0)
2189 		goto out;
2190 
2191 	/*
2192 	 * If the tvp exists and is in use, sillyrename it before doing the
2193 	 * rename of the new file over it.
2194 	 * XXX Can't sillyrename a directory.
2195 	 */
2196 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
2197 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
2198 		vput(tvp);
2199 		tvp = NULL;
2200 	}
2201 
2202 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
2203 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, false, tcnp->cn_cred,
2204 	    curthread);
2205 
2206 	if (error == 0 && NFS_ISV4(tdvp)) {
2207 		/*
2208 		 * For NFSv4, check to see if it is the same name and
2209 		 * replace the name, if it is different.
2210 		 */
2211 		newv4 = malloc(
2212 		    sizeof (struct nfsv4node) +
2213 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
2214 		    M_NFSV4NODE, M_WAITOK);
2215 		NFSLOCKNODE(tdnp);
2216 		NFSLOCKNODE(fnp);
2217 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
2218 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
2219 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
2220 		      tcnp->cn_namelen) ||
2221 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
2222 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
2223 			tdnp->n_fhp->nfh_len))) {
2224 			free(fnp->n_v4, M_NFSV4NODE);
2225 			fnp->n_v4 = newv4;
2226 			newv4 = NULL;
2227 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
2228 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
2229 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
2230 			    tdnp->n_fhp->nfh_len);
2231 			NFSBCOPY(tcnp->cn_nameptr,
2232 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
2233 		}
2234 		NFSUNLOCKNODE(tdnp);
2235 		NFSUNLOCKNODE(fnp);
2236 		if (newv4 != NULL)
2237 			free(newv4, M_NFSV4NODE);
2238 	}
2239 
2240 	if (fvp->v_type == VDIR) {
2241 		if (tvp != NULL && tvp->v_type == VDIR)
2242 			cache_purge(tdvp);
2243 		cache_purge(fdvp);
2244 	}
2245 
2246 out:
2247 	if (tdvp == tvp)
2248 		vrele(tdvp);
2249 	else
2250 		vput(tdvp);
2251 	if (tvp)
2252 		vput(tvp);
2253 	vrele(fdvp);
2254 	vrele(fvp);
2255 	/*
2256 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2257 	 */
2258 	if (error == ENOENT)
2259 		error = 0;
2260 	return (error);
2261 }
2262 
2263 /*
2264  * nfs file rename rpc called from nfs_remove() above
2265  */
2266 static int
2267 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
2268     struct sillyrename *sp)
2269 {
2270 
2271 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
2272 	    sdvp, NULL, sp->s_name, sp->s_namlen, true, scnp->cn_cred,
2273 	    curthread));
2274 }
2275 
2276 /*
2277  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2278  */
2279 static int
2280 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
2281     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
2282     int tnamelen, bool silly, struct ucred *cred, struct thread *td)
2283 {
2284 	struct nfsvattr fnfsva, tnfsva, tvpnfsva;
2285 	struct nfsnode *fdnp = VTONFS(fdvp);
2286 	struct nfsnode *tdnp = VTONFS(tdvp);
2287 	struct nfsmount *nmp;
2288 	int error = 0, fattrflag, tattrflag, tvpattrflag;
2289 	nfsremove_status tvp_status;
2290 
2291 	nmp = VFSTONFS(fdvp->v_mount);
2292 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
2293 	    tnameptr, tnamelen, &tvp_status, &fnfsva, &tnfsva, &fattrflag,
2294 	    &tattrflag, &tvpnfsva, &tvpattrflag, cred, td);
2295 	NFSLOCKNODE(fdnp);
2296 	fdnp->n_flag |= NMODIFIED;
2297 	if (fattrflag != 0) {
2298 		NFSUNLOCKNODE(fdnp);
2299 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, 0, 1);
2300 	} else {
2301 		fdnp->n_attrstamp = 0;
2302 		NFSUNLOCKNODE(fdnp);
2303 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
2304 	}
2305 	NFSLOCKNODE(tdnp);
2306 	tdnp->n_flag |= NMODIFIED;
2307 	if (tattrflag != 0) {
2308 		NFSUNLOCKNODE(tdnp);
2309 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, 0, 1);
2310 	} else {
2311 		tdnp->n_attrstamp = 0;
2312 		NFSUNLOCKNODE(tdnp);
2313 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2314 	}
2315 
2316 	if (tvp != NULL) {
2317 		if (NFSHASNFSV4(nmp) && NFSHASNFSV4N(nmp) &&
2318 		    (nmp->nm_flag & NFSMNT_NOCTO) != 0)
2319 			nfs_removestatus(tvp, tvp_status, silly, td);
2320 		if (!silly && tvpattrflag != 0)
2321 			(void)nfscl_loadattrcache(&tvp, &tvpnfsva, NULL, 0, 1);
2322 	}
2323 
2324 	if (error && NFS_ISV4(fdvp))
2325 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2326 	return (error);
2327 }
2328 
2329 /*
2330  * nfs hard link create call
2331  */
2332 static int
2333 nfs_link(struct vop_link_args *ap)
2334 {
2335 	struct vnode *vp = ap->a_vp;
2336 	struct vnode *tdvp = ap->a_tdvp;
2337 	struct componentname *cnp = ap->a_cnp;
2338 	struct nfsnode *np, *tdnp;
2339 	struct nfsvattr nfsva, dnfsva;
2340 	int error = 0, attrflag, dattrflag;
2341 
2342 	/*
2343 	 * Push all writes to the server, so that the attribute cache
2344 	 * doesn't get "out of sync" with the server.
2345 	 * XXX There should be a better way!
2346 	 */
2347 #ifdef notnow
2348 	VOP_FSYNC(vp, MNT_WAIT, curthread);
2349 #endif
2350 
2351 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2352 	    cnp->cn_cred, curthread, &dnfsva, &nfsva, &attrflag, &dattrflag);
2353 	tdnp = VTONFS(tdvp);
2354 	NFSLOCKNODE(tdnp);
2355 	tdnp->n_flag |= NMODIFIED;
2356 	if (dattrflag != 0) {
2357 		NFSUNLOCKNODE(tdnp);
2358 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, 0, 1);
2359 	} else {
2360 		tdnp->n_attrstamp = 0;
2361 		NFSUNLOCKNODE(tdnp);
2362 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2363 	}
2364 	if (attrflag)
2365 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
2366 	else {
2367 		np = VTONFS(vp);
2368 		NFSLOCKNODE(np);
2369 		np->n_attrstamp = 0;
2370 		NFSUNLOCKNODE(np);
2371 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2372 	}
2373 	/*
2374 	 * If negative lookup caching is enabled, I might as well
2375 	 * add an entry for this node. Not necessary for correctness,
2376 	 * but if negative caching is enabled, then the system
2377 	 * must care about lookup caching hit rate, so...
2378 	 */
2379 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2380 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2381 		if (tdvp != vp)
2382 			cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2383 		else
2384 			printf("nfs_link: bogus NFS server returned "
2385 			    "the directory as the new link\n");
2386 	}
2387 	if (error && NFS_ISV4(vp))
2388 		error = nfscl_maperr(curthread, error, (uid_t)0,
2389 		    (gid_t)0);
2390 	return (error);
2391 }
2392 
2393 /*
2394  * nfs symbolic link create call
2395  */
2396 static int
2397 nfs_symlink(struct vop_symlink_args *ap)
2398 {
2399 	struct vnode *dvp = ap->a_dvp;
2400 	struct vattr *vap = ap->a_vap;
2401 	struct componentname *cnp = ap->a_cnp;
2402 	struct nfsvattr nfsva, dnfsva;
2403 	struct nfsfh *nfhp;
2404 	struct nfsnode *np = NULL, *dnp;
2405 	struct vnode *newvp = NULL;
2406 	int error = 0, attrflag, dattrflag, ret;
2407 
2408 	vap->va_type = VLNK;
2409 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2410 	    ap->a_target, vap, cnp->cn_cred, curthread, &dnfsva,
2411 	    &nfsva, &nfhp, &attrflag, &dattrflag);
2412 	if (nfhp) {
2413 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread,
2414 		    &np, LK_EXCLUSIVE);
2415 		if (!ret)
2416 			newvp = NFSTOV(np);
2417 		else if (!error)
2418 			error = ret;
2419 	}
2420 	if (newvp != NULL) {
2421 		if (attrflag)
2422 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
2423 	} else if (!error) {
2424 		/*
2425 		 * If we do not have an error and we could not extract the
2426 		 * newvp from the response due to the request being NFSv2, we
2427 		 * have to do a lookup in order to obtain a newvp to return.
2428 		 */
2429 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2430 		    cnp->cn_cred, curthread, &np);
2431 		if (!error)
2432 			newvp = NFSTOV(np);
2433 	}
2434 	if (error) {
2435 		if (newvp)
2436 			vput(newvp);
2437 		if (NFS_ISV4(dvp))
2438 			error = nfscl_maperr(curthread, error,
2439 			    vap->va_uid, vap->va_gid);
2440 	} else {
2441 		*ap->a_vpp = newvp;
2442 	}
2443 
2444 	dnp = VTONFS(dvp);
2445 	NFSLOCKNODE(dnp);
2446 	dnp->n_flag |= NMODIFIED;
2447 	if (dattrflag != 0) {
2448 		NFSUNLOCKNODE(dnp);
2449 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
2450 	} else {
2451 		dnp->n_attrstamp = 0;
2452 		NFSUNLOCKNODE(dnp);
2453 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2454 	}
2455 	/*
2456 	 * If negative lookup caching is enabled, I might as well
2457 	 * add an entry for this node. Not necessary for correctness,
2458 	 * but if negative caching is enabled, then the system
2459 	 * must care about lookup caching hit rate, so...
2460 	 */
2461 	if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2462 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2463 		if (dvp != newvp)
2464 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2465 			    NULL);
2466 		else
2467 			printf("nfs_symlink: bogus NFS server returned "
2468 			    "the directory as the new file object\n");
2469 	}
2470 	return (error);
2471 }
2472 
2473 /*
2474  * nfs make dir call
2475  */
2476 static int
2477 nfs_mkdir(struct vop_mkdir_args *ap)
2478 {
2479 	struct vnode *dvp = ap->a_dvp;
2480 	struct vattr *vap = ap->a_vap;
2481 	struct componentname *cnp = ap->a_cnp;
2482 	struct nfsnode *np = NULL, *dnp;
2483 	struct vnode *newvp = NULL;
2484 	struct vattr vattr;
2485 	struct nfsfh *nfhp;
2486 	struct nfsvattr nfsva, dnfsva;
2487 	int error = 0, attrflag, dattrflag, ret;
2488 
2489 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2490 		return (error);
2491 	vap->va_type = VDIR;
2492 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2493 	    vap, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp,
2494 	    &attrflag, &dattrflag);
2495 	dnp = VTONFS(dvp);
2496 	NFSLOCKNODE(dnp);
2497 	dnp->n_flag |= NMODIFIED;
2498 	if (dattrflag != 0) {
2499 		NFSUNLOCKNODE(dnp);
2500 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
2501 	} else {
2502 		dnp->n_attrstamp = 0;
2503 		NFSUNLOCKNODE(dnp);
2504 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2505 	}
2506 	if (nfhp) {
2507 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread,
2508 		    &np, LK_EXCLUSIVE);
2509 		if (!ret) {
2510 			newvp = NFSTOV(np);
2511 			if (attrflag)
2512 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2513 				0, 1);
2514 		} else if (!error)
2515 			error = ret;
2516 	}
2517 	if (!error && newvp == NULL) {
2518 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2519 		    cnp->cn_cred, curthread, &np);
2520 		if (!error) {
2521 			newvp = NFSTOV(np);
2522 			if (newvp->v_type != VDIR)
2523 				error = EEXIST;
2524 		}
2525 	}
2526 	if (error) {
2527 		if (newvp)
2528 			vput(newvp);
2529 		if (NFS_ISV4(dvp))
2530 			error = nfscl_maperr(curthread, error,
2531 			    vap->va_uid, vap->va_gid);
2532 	} else {
2533 		/*
2534 		 * If negative lookup caching is enabled, I might as well
2535 		 * add an entry for this node. Not necessary for correctness,
2536 		 * but if negative caching is enabled, then the system
2537 		 * must care about lookup caching hit rate, so...
2538 		 */
2539 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2540 		    (cnp->cn_flags & MAKEENTRY) &&
2541 		    attrflag != 0 && dattrflag != 0) {
2542 			if (dvp != newvp)
2543 				cache_enter_time(dvp, newvp, cnp,
2544 				    &nfsva.na_ctime, &dnfsva.na_ctime);
2545 			else
2546 				printf("nfs_mkdir: bogus NFS server returned "
2547 				    "the directory that the directory was "
2548 				    "created in as the new file object\n");
2549 		}
2550 		*ap->a_vpp = newvp;
2551 	}
2552 	return (error);
2553 }
2554 
2555 /*
2556  * nfs remove directory call
2557  */
2558 static int
2559 nfs_rmdir(struct vop_rmdir_args *ap)
2560 {
2561 	struct vnode *vp = ap->a_vp;
2562 	struct vnode *dvp = ap->a_dvp;
2563 	struct componentname *cnp = ap->a_cnp;
2564 	struct nfsnode *dnp;
2565 	struct nfsvattr dnfsva;
2566 	int error, dattrflag;
2567 
2568 	if (dvp == vp)
2569 		return (EINVAL);
2570 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2571 	    cnp->cn_cred, curthread, &dnfsva, &dattrflag);
2572 	dnp = VTONFS(dvp);
2573 	NFSLOCKNODE(dnp);
2574 	dnp->n_flag |= NMODIFIED;
2575 	if (dattrflag != 0) {
2576 		NFSUNLOCKNODE(dnp);
2577 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
2578 	} else {
2579 		dnp->n_attrstamp = 0;
2580 		NFSUNLOCKNODE(dnp);
2581 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2582 	}
2583 
2584 	cache_purge(dvp);
2585 	cache_purge(vp);
2586 	if (error && NFS_ISV4(dvp))
2587 		error = nfscl_maperr(curthread, error, (uid_t)0,
2588 		    (gid_t)0);
2589 	/*
2590 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2591 	 */
2592 	if (error == ENOENT)
2593 		error = 0;
2594 	return (error);
2595 }
2596 
2597 /*
2598  * nfs readdir call
2599  */
2600 static int
2601 nfs_readdir(struct vop_readdir_args *ap)
2602 {
2603 	struct vnode *vp = ap->a_vp;
2604 	struct nfsnode *np = VTONFS(vp);
2605 	struct uio *uio = ap->a_uio;
2606 	ssize_t tresid, left;
2607 	int error = 0;
2608 	struct vattr vattr;
2609 
2610 	if (ap->a_eofflag != NULL)
2611 		*ap->a_eofflag = 0;
2612 	if (vp->v_type != VDIR)
2613 		return(EPERM);
2614 
2615 	/*
2616 	 * First, check for hit on the EOF offset cache
2617 	 */
2618 	NFSLOCKNODE(np);
2619 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2620 	    (np->n_flag & NMODIFIED) == 0) {
2621 		NFSUNLOCKNODE(np);
2622 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2623 			NFSLOCKNODE(np);
2624 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2625 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2626 				NFSUNLOCKNODE(np);
2627 				NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2628 				if (ap->a_eofflag != NULL)
2629 					*ap->a_eofflag = 1;
2630 				return (0);
2631 			} else
2632 				NFSUNLOCKNODE(np);
2633 		}
2634 	} else
2635 		NFSUNLOCKNODE(np);
2636 
2637 	/*
2638 	 * NFS always guarantees that directory entries don't straddle
2639 	 * DIRBLKSIZ boundaries.  As such, we need to limit the size
2640 	 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2641 	 * directory entry.
2642 	 */
2643 	left = uio->uio_resid % DIRBLKSIZ;
2644 	if (left == uio->uio_resid)
2645 		return (EINVAL);
2646 	uio->uio_resid -= left;
2647 
2648 	/*
2649 	 * For readdirplus, if starting to read the directory,
2650 	 * purge the name cache, since it will be reloaded by
2651 	 * this directory read.
2652 	 * This removes potentially stale name cache entries.
2653 	 */
2654 	if (uio->uio_offset == 0 &&
2655 	    (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_RDIRPLUS) != 0)
2656 		cache_purge(vp);
2657 
2658 	/*
2659 	 * Call ncl_bioread() to do the real work.
2660 	 */
2661 	tresid = uio->uio_resid;
2662 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2663 
2664 	if (!error && uio->uio_resid == tresid) {
2665 		NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2666 		if (ap->a_eofflag != NULL)
2667 			*ap->a_eofflag = 1;
2668 	}
2669 
2670 	/* Add the partial DIRBLKSIZ (left) back in. */
2671 	uio->uio_resid += left;
2672 	return (error);
2673 }
2674 
2675 /*
2676  * Readdir rpc call.
2677  * Called from below the buffer cache by ncl_doio().
2678  */
2679 int
2680 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2681     struct thread *td)
2682 {
2683 	struct nfsvattr nfsva;
2684 	nfsuint64 *cookiep, cookie;
2685 	struct nfsnode *dnp = VTONFS(vp);
2686 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2687 	int error = 0, eof, attrflag;
2688 
2689 	KASSERT(uiop->uio_iovcnt == 1 &&
2690 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2691 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2692 	    ("nfs readdirrpc bad uio"));
2693 
2694 	/*
2695 	 * If there is no cookie, assume directory was stale.
2696 	 */
2697 	ncl_dircookie_lock(dnp);
2698 	NFSUNLOCKNODE(dnp);
2699 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2700 	if (cookiep) {
2701 		cookie = *cookiep;
2702 		ncl_dircookie_unlock(dnp);
2703 	} else {
2704 		ncl_dircookie_unlock(dnp);
2705 		return (NFSERR_BAD_COOKIE);
2706 	}
2707 
2708 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2709 		(void)ncl_fsinfo(nmp, vp, cred, td);
2710 
2711 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2712 	    &attrflag, &eof);
2713 	if (attrflag)
2714 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
2715 
2716 	if (!error) {
2717 		/*
2718 		 * We are now either at the end of the directory or have filled
2719 		 * the block.
2720 		 */
2721 		if (eof) {
2722 			NFSLOCKNODE(dnp);
2723 			dnp->n_direofoffset = uiop->uio_offset;
2724 			NFSUNLOCKNODE(dnp);
2725 		} else {
2726 			if (uiop->uio_resid > 0)
2727 				printf("EEK! readdirrpc resid > 0\n");
2728 			ncl_dircookie_lock(dnp);
2729 			NFSUNLOCKNODE(dnp);
2730 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2731 			*cookiep = cookie;
2732 			ncl_dircookie_unlock(dnp);
2733 		}
2734 	} else if (NFS_ISV4(vp)) {
2735 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2736 	}
2737 	return (error);
2738 }
2739 
2740 /*
2741  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2742  */
2743 int
2744 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2745     struct thread *td)
2746 {
2747 	struct nfsvattr nfsva;
2748 	nfsuint64 *cookiep, cookie;
2749 	struct nfsnode *dnp = VTONFS(vp);
2750 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2751 	int error = 0, attrflag, eof;
2752 
2753 	KASSERT(uiop->uio_iovcnt == 1 &&
2754 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2755 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2756 	    ("nfs readdirplusrpc bad uio"));
2757 
2758 	/*
2759 	 * If there is no cookie, assume directory was stale.
2760 	 */
2761 	ncl_dircookie_lock(dnp);
2762 	NFSUNLOCKNODE(dnp);
2763 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2764 	if (cookiep) {
2765 		cookie = *cookiep;
2766 		ncl_dircookie_unlock(dnp);
2767 	} else {
2768 		ncl_dircookie_unlock(dnp);
2769 		return (NFSERR_BAD_COOKIE);
2770 	}
2771 
2772 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2773 		(void)ncl_fsinfo(nmp, vp, cred, td);
2774 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2775 	    &attrflag, &eof);
2776 	if (attrflag)
2777 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
2778 
2779 	if (!error) {
2780 		/*
2781 		 * We are now either at end of the directory or have filled the
2782 		 * the block.
2783 		 */
2784 		if (eof) {
2785 			NFSLOCKNODE(dnp);
2786 			dnp->n_direofoffset = uiop->uio_offset;
2787 			NFSUNLOCKNODE(dnp);
2788 		} else {
2789 			if (uiop->uio_resid > 0)
2790 				printf("EEK! readdirplusrpc resid > 0\n");
2791 			ncl_dircookie_lock(dnp);
2792 			NFSUNLOCKNODE(dnp);
2793 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2794 			*cookiep = cookie;
2795 			ncl_dircookie_unlock(dnp);
2796 		}
2797 	} else if (NFS_ISV4(vp)) {
2798 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2799 	}
2800 	return (error);
2801 }
2802 
2803 /*
2804  * Silly rename. To make the NFS filesystem that is stateless look a little
2805  * more like the "ufs" a remove of an active vnode is translated to a rename
2806  * to a funny looking filename that is removed by nfs_inactive on the
2807  * nfsnode. There is the potential for another process on a different client
2808  * to create the same funny name between the nfs_lookitup() fails and the
2809  * nfs_rename() completes, but...
2810  */
2811 static int
2812 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2813 {
2814 	struct sillyrename *sp;
2815 	struct nfsnode *np;
2816 	int error;
2817 	short pid;
2818 	unsigned int lticks;
2819 
2820 	cache_purge(dvp);
2821 	np = VTONFS(vp);
2822 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2823 	sp = malloc(sizeof (struct sillyrename),
2824 	    M_NEWNFSREQ, M_WAITOK);
2825 	sp->s_cred = crhold(cnp->cn_cred);
2826 	sp->s_dvp = dvp;
2827 	vref(dvp);
2828 
2829 	/*
2830 	 * Fudge together a funny name.
2831 	 * Changing the format of the funny name to accommodate more
2832 	 * sillynames per directory.
2833 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2834 	 * CPU ticks since boot.
2835 	 */
2836 	pid = curthread->td_proc->p_pid;
2837 	lticks = (unsigned int)ticks;
2838 	for ( ; ; ) {
2839 		sp->s_namlen = sprintf(sp->s_name,
2840 				       ".nfs.%08x.%04x4.4", lticks,
2841 				       pid);
2842 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2843 				 curthread, NULL))
2844 			break;
2845 		lticks++;
2846 	}
2847 	error = nfs_renameit(dvp, vp, cnp, sp);
2848 	if (error)
2849 		goto bad;
2850 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2851 		curthread, &np);
2852 	np->n_sillyrename = sp;
2853 	return (0);
2854 bad:
2855 	vrele(sp->s_dvp);
2856 	crfree(sp->s_cred);
2857 	free(sp, M_NEWNFSREQ);
2858 	return (error);
2859 }
2860 
2861 /*
2862  * Look up a file name and optionally either update the file handle or
2863  * allocate an nfsnode, depending on the value of npp.
2864  * npp == NULL	--> just do the lookup
2865  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2866  *			handled too
2867  * *npp != NULL --> update the file handle in the vnode
2868  */
2869 static int
2870 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2871     struct thread *td, struct nfsnode **npp)
2872 {
2873 	struct vnode *newvp = NULL, *vp;
2874 	struct nfsnode *np, *dnp = VTONFS(dvp);
2875 	struct nfsfh *nfhp, *onfhp;
2876 	struct nfsvattr nfsva, dnfsva;
2877 	struct componentname cn;
2878 	int error = 0, attrflag, dattrflag;
2879 	u_int hash;
2880 	struct timespec ts;
2881 
2882 	nanouptime(&ts);
2883 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2884 	    &nfhp, &attrflag, &dattrflag, 0);
2885 	if (dattrflag)
2886 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
2887 	if (npp && !error) {
2888 		if (*npp != NULL) {
2889 		    np = *npp;
2890 		    vp = NFSTOV(np);
2891 		    /*
2892 		     * For NFSv4, check to see if it is the same name and
2893 		     * replace the name, if it is different.
2894 		     */
2895 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2896 			(np->n_v4->n4_namelen != len ||
2897 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2898 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2899 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2900 			 dnp->n_fhp->nfh_len))) {
2901 			    free(np->n_v4, M_NFSV4NODE);
2902 			    np->n_v4 = malloc(
2903 				sizeof (struct nfsv4node) +
2904 				dnp->n_fhp->nfh_len + len - 1,
2905 				M_NFSV4NODE, M_WAITOK);
2906 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2907 			    np->n_v4->n4_namelen = len;
2908 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2909 				dnp->n_fhp->nfh_len);
2910 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2911 		    }
2912 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2913 			FNV1_32_INIT);
2914 		    onfhp = np->n_fhp;
2915 		    /*
2916 		     * Rehash node for new file handle.
2917 		     */
2918 		    vfs_hash_rehash(vp, hash);
2919 		    np->n_fhp = nfhp;
2920 		    if (onfhp != NULL)
2921 			free(onfhp, M_NFSFH);
2922 		    newvp = NFSTOV(np);
2923 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2924 		    free(nfhp, M_NFSFH);
2925 		    vref(dvp);
2926 		    newvp = dvp;
2927 		} else {
2928 		    cn.cn_nameptr = name;
2929 		    cn.cn_namelen = len;
2930 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2931 			&np, LK_EXCLUSIVE);
2932 		    if (error)
2933 			return (error);
2934 		    newvp = NFSTOV(np);
2935 		    /*
2936 		     * If n_localmodtime >= time before RPC, then
2937 		     * a file modification operation, such as
2938 		     * VOP_SETATTR() of size, has occurred while
2939 		     * the Lookup RPC and acquisition of the vnode
2940 		     * happened.  As such, the attributes might
2941 		     * be stale, with possibly an incorrect size.
2942 		     */
2943 		    NFSLOCKNODE(np);
2944 		    if (timespecisset(&np->n_localmodtime) &&
2945 			timespeccmp(&np->n_localmodtime, &ts, >=)) {
2946 			NFSCL_DEBUG(4, "nfs_lookitup: localmod "
2947 			    "stale attributes\n");
2948 			attrflag = 0;
2949 		    }
2950 		    NFSUNLOCKNODE(np);
2951 		}
2952 		if (!attrflag && *npp == NULL) {
2953 			if (newvp == dvp)
2954 				vrele(newvp);
2955 			else
2956 				vput(newvp);
2957 			return (ENOENT);
2958 		}
2959 		if (attrflag)
2960 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
2961 	}
2962 	if (npp && *npp == NULL) {
2963 		if (error) {
2964 			if (newvp) {
2965 				if (newvp == dvp)
2966 					vrele(newvp);
2967 				else
2968 					vput(newvp);
2969 			}
2970 		} else
2971 			*npp = np;
2972 	}
2973 	if (error && NFS_ISV4(dvp))
2974 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2975 	return (error);
2976 }
2977 
2978 /*
2979  * Nfs Version 3 and 4 commit rpc
2980  */
2981 int
2982 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2983    struct thread *td)
2984 {
2985 	struct nfsvattr nfsva;
2986 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2987 	struct nfsnode *np;
2988 	struct uio uio;
2989 	int error, attrflag;
2990 
2991 	np = VTONFS(vp);
2992 	error = EIO;
2993 	attrflag = 0;
2994 	if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2995 		uio.uio_offset = offset;
2996 		uio.uio_resid = cnt;
2997 		error = nfscl_doiods(vp, &uio, NULL, NULL,
2998 		    NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2999 		if (error != 0) {
3000 			NFSLOCKNODE(np);
3001 			np->n_flag &= ~NDSCOMMIT;
3002 			NFSUNLOCKNODE(np);
3003 		}
3004 	}
3005 	if (error != 0) {
3006 		mtx_lock(&nmp->nm_mtx);
3007 		if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
3008 			mtx_unlock(&nmp->nm_mtx);
3009 			return (0);
3010 		}
3011 		mtx_unlock(&nmp->nm_mtx);
3012 		error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
3013 		    &attrflag);
3014 	}
3015 	if (attrflag != 0)
3016 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
3017 	if (error != 0 && NFS_ISV4(vp))
3018 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3019 	return (error);
3020 }
3021 
3022 /*
3023  * Strategy routine.
3024  * For async requests when nfsiod(s) are running, queue the request by
3025  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
3026  * request.
3027  */
3028 static int
3029 nfs_strategy(struct vop_strategy_args *ap)
3030 {
3031 	struct buf *bp;
3032 	struct vnode *vp;
3033 	struct ucred *cr;
3034 
3035 	bp = ap->a_bp;
3036 	vp = ap->a_vp;
3037 	KASSERT(bp->b_vp == vp, ("missing b_getvp"));
3038 	KASSERT(!(bp->b_flags & B_DONE),
3039 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
3040 
3041 	if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
3042 		bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
3043 		    DEV_BSIZE);
3044 	if (bp->b_iocmd == BIO_READ)
3045 		cr = bp->b_rcred;
3046 	else
3047 		cr = bp->b_wcred;
3048 
3049 	/*
3050 	 * If the op is asynchronous and an i/o daemon is waiting
3051 	 * queue the request, wake it up and wait for completion
3052 	 * otherwise just do it ourselves.
3053 	 */
3054 	if ((bp->b_flags & B_ASYNC) == 0 ||
3055 	    ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
3056 		(void) ncl_doio(vp, bp, cr, curthread, 1);
3057 	return (0);
3058 }
3059 
3060 /*
3061  * fsync vnode op. Just call ncl_flush() with commit == 1.
3062  */
3063 /* ARGSUSED */
3064 static int
3065 nfs_fsync(struct vop_fsync_args *ap)
3066 {
3067 
3068 	if (ap->a_vp->v_type != VREG) {
3069 		/*
3070 		 * For NFS, metadata is changed synchronously on the server,
3071 		 * so there is nothing to flush. Also, ncl_flush() clears
3072 		 * the NMODIFIED flag and that shouldn't be done here for
3073 		 * directories.
3074 		 */
3075 		return (0);
3076 	}
3077 	return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
3078 }
3079 
3080 /*
3081  * Flush all the blocks associated with a vnode.
3082  * 	Walk through the buffer pool and push any dirty pages
3083  *	associated with the vnode.
3084  * If the called_from_renewthread argument is TRUE, it has been called
3085  * from the NFSv4 renew thread and, as such, cannot block indefinitely
3086  * waiting for a buffer write to complete.
3087  */
3088 int
3089 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
3090     int commit, int called_from_renewthread)
3091 {
3092 	struct nfsnode *np = VTONFS(vp);
3093 	struct buf *bp;
3094 	int i;
3095 	struct buf *nbp;
3096 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
3097 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
3098 	int passone = 1, trycnt = 0;
3099 	u_quad_t off, endoff, toff;
3100 	struct ucred* wcred = NULL;
3101 	struct buf **bvec = NULL;
3102 	struct bufobj *bo;
3103 #ifndef NFS_COMMITBVECSIZ
3104 #define	NFS_COMMITBVECSIZ	20
3105 #endif
3106 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
3107 	u_int bvecsize = 0, bveccount;
3108 	struct timespec ts;
3109 
3110 	if (called_from_renewthread != 0)
3111 		slptimeo = hz;
3112 	if (nmp->nm_flag & NFSMNT_INT)
3113 		slpflag = PCATCH;
3114 	if (!commit)
3115 		passone = 0;
3116 	bo = &vp->v_bufobj;
3117 	/*
3118 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
3119 	 * server, but has not been committed to stable storage on the server
3120 	 * yet. On the first pass, the byte range is worked out and the commit
3121 	 * rpc is done. On the second pass, bwrite() is called to do the
3122 	 * job.
3123 	 */
3124 again:
3125 	off = (u_quad_t)-1;
3126 	endoff = 0;
3127 	bvecpos = 0;
3128 	if (NFS_ISV34(vp) && commit) {
3129 		if (bvec != NULL && bvec != bvec_on_stack)
3130 			free(bvec, M_TEMP);
3131 		/*
3132 		 * Count up how many buffers waiting for a commit.
3133 		 */
3134 		bveccount = 0;
3135 		BO_LOCK(bo);
3136 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3137 			if (!BUF_ISLOCKED(bp) &&
3138 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
3139 				== (B_DELWRI | B_NEEDCOMMIT))
3140 				bveccount++;
3141 		}
3142 		/*
3143 		 * Allocate space to remember the list of bufs to commit.  It is
3144 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
3145 		 * If we can't get memory (for whatever reason), we will end up
3146 		 * committing the buffers one-by-one in the loop below.
3147 		 */
3148 		if (bveccount > NFS_COMMITBVECSIZ) {
3149 			/*
3150 			 * Release the vnode interlock to avoid a lock
3151 			 * order reversal.
3152 			 */
3153 			BO_UNLOCK(bo);
3154 			bvec = (struct buf **)
3155 				malloc(bveccount * sizeof(struct buf *),
3156 				       M_TEMP, M_NOWAIT);
3157 			BO_LOCK(bo);
3158 			if (bvec == NULL) {
3159 				bvec = bvec_on_stack;
3160 				bvecsize = NFS_COMMITBVECSIZ;
3161 			} else
3162 				bvecsize = bveccount;
3163 		} else {
3164 			bvec = bvec_on_stack;
3165 			bvecsize = NFS_COMMITBVECSIZ;
3166 		}
3167 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3168 			if (bvecpos >= bvecsize)
3169 				break;
3170 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3171 				nbp = TAILQ_NEXT(bp, b_bobufs);
3172 				continue;
3173 			}
3174 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3175 			    (B_DELWRI | B_NEEDCOMMIT)) {
3176 				BUF_UNLOCK(bp);
3177 				nbp = TAILQ_NEXT(bp, b_bobufs);
3178 				continue;
3179 			}
3180 			BO_UNLOCK(bo);
3181 			bremfree(bp);
3182 			/*
3183 			 * Work out if all buffers are using the same cred
3184 			 * so we can deal with them all with one commit.
3185 			 *
3186 			 * NOTE: we are not clearing B_DONE here, so we have
3187 			 * to do it later on in this routine if we intend to
3188 			 * initiate I/O on the bp.
3189 			 *
3190 			 * Note: to avoid loopback deadlocks, we do not
3191 			 * assign b_runningbufspace.
3192 			 */
3193 			if (wcred == NULL)
3194 				wcred = bp->b_wcred;
3195 			else if (wcred != bp->b_wcred)
3196 				wcred = NOCRED;
3197 			vfs_busy_pages(bp, 0);
3198 
3199 			BO_LOCK(bo);
3200 			/*
3201 			 * bp is protected by being locked, but nbp is not
3202 			 * and vfs_busy_pages() may sleep.  We have to
3203 			 * recalculate nbp.
3204 			 */
3205 			nbp = TAILQ_NEXT(bp, b_bobufs);
3206 
3207 			/*
3208 			 * A list of these buffers is kept so that the
3209 			 * second loop knows which buffers have actually
3210 			 * been committed. This is necessary, since there
3211 			 * may be a race between the commit rpc and new
3212 			 * uncommitted writes on the file.
3213 			 */
3214 			bvec[bvecpos++] = bp;
3215 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3216 				bp->b_dirtyoff;
3217 			if (toff < off)
3218 				off = toff;
3219 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3220 			if (toff > endoff)
3221 				endoff = toff;
3222 		}
3223 		BO_UNLOCK(bo);
3224 	}
3225 	if (bvecpos > 0) {
3226 		/*
3227 		 * Commit data on the server, as required.
3228 		 * If all bufs are using the same wcred, then use that with
3229 		 * one call for all of them, otherwise commit each one
3230 		 * separately.
3231 		 */
3232 		if (wcred != NOCRED)
3233 			retv = ncl_commit(vp, off, (int)(endoff - off),
3234 					  wcred, td);
3235 		else {
3236 			retv = 0;
3237 			for (i = 0; i < bvecpos; i++) {
3238 				off_t off, size;
3239 				bp = bvec[i];
3240 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3241 					bp->b_dirtyoff;
3242 				size = (u_quad_t)(bp->b_dirtyend
3243 						  - bp->b_dirtyoff);
3244 				retv = ncl_commit(vp, off, (int)size,
3245 						  bp->b_wcred, td);
3246 				if (retv) break;
3247 			}
3248 		}
3249 
3250 		if (retv == NFSERR_STALEWRITEVERF)
3251 			ncl_clearcommit(vp->v_mount);
3252 
3253 		/*
3254 		 * Now, either mark the blocks I/O done or mark the
3255 		 * blocks dirty, depending on whether the commit
3256 		 * succeeded.
3257 		 */
3258 		for (i = 0; i < bvecpos; i++) {
3259 			bp = bvec[i];
3260 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3261 			if (!NFSCL_FORCEDISM(vp->v_mount) && retv) {
3262 				/*
3263 				 * Error, leave B_DELWRI intact
3264 				 */
3265 				vfs_unbusy_pages(bp);
3266 				brelse(bp);
3267 			} else {
3268 				/*
3269 				 * Success, remove B_DELWRI ( bundirty() ).
3270 				 *
3271 				 * b_dirtyoff/b_dirtyend seem to be NFS
3272 				 * specific.  We should probably move that
3273 				 * into bundirty(). XXX
3274 				 */
3275 				bufobj_wref(bo);
3276 				bp->b_flags |= B_ASYNC;
3277 				bundirty(bp);
3278 				bp->b_flags &= ~B_DONE;
3279 				bp->b_ioflags &= ~BIO_ERROR;
3280 				bp->b_dirtyoff = bp->b_dirtyend = 0;
3281 				bufdone(bp);
3282 			}
3283 		}
3284 	}
3285 
3286 	/*
3287 	 * Start/do any write(s) that are required.
3288 	 */
3289 loop:
3290 	BO_LOCK(bo);
3291 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3292 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3293 			if (waitfor != MNT_WAIT || passone)
3294 				continue;
3295 
3296 			error = BUF_TIMELOCK(bp,
3297 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3298 			    BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
3299 			if (error == 0) {
3300 				BUF_UNLOCK(bp);
3301 				goto loop;
3302 			}
3303 			if (error == ENOLCK) {
3304 				error = 0;
3305 				goto loop;
3306 			}
3307 			if (called_from_renewthread != 0) {
3308 				/*
3309 				 * Return EIO so the flush will be retried
3310 				 * later.
3311 				 */
3312 				error = EIO;
3313 				goto done;
3314 			}
3315 			if (newnfs_sigintr(nmp, td)) {
3316 				error = EINTR;
3317 				goto done;
3318 			}
3319 			if (slpflag == PCATCH) {
3320 				slpflag = 0;
3321 				slptimeo = 2 * hz;
3322 			}
3323 			goto loop;
3324 		}
3325 		if ((bp->b_flags & B_DELWRI) == 0)
3326 			panic("nfs_fsync: not dirty");
3327 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3328 			BUF_UNLOCK(bp);
3329 			continue;
3330 		}
3331 		BO_UNLOCK(bo);
3332 		bremfree(bp);
3333 		bp->b_flags |= B_ASYNC;
3334 		bwrite(bp);
3335 		if (newnfs_sigintr(nmp, td)) {
3336 			error = EINTR;
3337 			goto done;
3338 		}
3339 		goto loop;
3340 	}
3341 	if (passone) {
3342 		passone = 0;
3343 		BO_UNLOCK(bo);
3344 		goto again;
3345 	}
3346 	if (waitfor == MNT_WAIT) {
3347 		while (bo->bo_numoutput) {
3348 			error = bufobj_wwait(bo, slpflag, slptimeo);
3349 			if (error) {
3350 			    BO_UNLOCK(bo);
3351 			    if (called_from_renewthread != 0) {
3352 				/*
3353 				 * Return EIO so that the flush will be
3354 				 * retried later.
3355 				 */
3356 				error = EIO;
3357 				goto done;
3358 			    }
3359 			    error = newnfs_sigintr(nmp, td);
3360 			    if (error)
3361 				goto done;
3362 			    if (slpflag == PCATCH) {
3363 				slpflag = 0;
3364 				slptimeo = 2 * hz;
3365 			    }
3366 			    BO_LOCK(bo);
3367 			}
3368 		}
3369 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
3370 			BO_UNLOCK(bo);
3371 			goto loop;
3372 		}
3373 		/*
3374 		 * Wait for all the async IO requests to drain
3375 		 */
3376 		BO_UNLOCK(bo);
3377 	} else
3378 		BO_UNLOCK(bo);
3379 	if (NFSHASPNFS(nmp)) {
3380 		nfscl_layoutcommit(vp, td);
3381 		/*
3382 		 * Invalidate the attribute cache, since writes to a DS
3383 		 * won't update the size attribute.
3384 		 */
3385 		NFSLOCKNODE(np);
3386 		np->n_attrstamp = 0;
3387 	} else
3388 		NFSLOCKNODE(np);
3389 	if (np->n_flag & NWRITEERR) {
3390 		error = np->n_error;
3391 		np->n_flag &= ~NWRITEERR;
3392 	}
3393   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
3394 	    bo->bo_numoutput == 0)
3395   		np->n_flag &= ~NMODIFIED;
3396 	NFSUNLOCKNODE(np);
3397 done:
3398 	if (bvec != NULL && bvec != bvec_on_stack)
3399 		free(bvec, M_TEMP);
3400 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
3401 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0)) {
3402 		if (trycnt++ < 5) {
3403 			/* try, try again... */
3404 			passone = 1;
3405 			wcred = NULL;
3406 			bvec = NULL;
3407 			bvecsize = 0;
3408 			goto again;
3409 		}
3410 		vn_printf(vp, "ncl_flush failed");
3411 		error = called_from_renewthread != 0 ? EIO : EBUSY;
3412 	}
3413 	if (error == 0) {
3414 		nanouptime(&ts);
3415 		NFSLOCKNODE(np);
3416 		np->n_localmodtime = ts;
3417 		NFSUNLOCKNODE(np);
3418 	}
3419 	return (error);
3420 }
3421 
3422 /*
3423  * NFS advisory byte-level locks.
3424  */
3425 static int
3426 nfs_advlock(struct vop_advlock_args *ap)
3427 {
3428 	struct vnode *vp = ap->a_vp;
3429 	struct ucred *cred;
3430 	struct nfsnode *np = VTONFS(ap->a_vp);
3431 	struct proc *p = (struct proc *)ap->a_id;
3432 	struct thread *td = curthread;	/* XXX */
3433 	struct vattr va;
3434 	int ret, error;
3435 	u_quad_t size;
3436 	struct nfsmount *nmp;
3437 
3438 	error = NFSVOPLOCK(vp, LK_EXCLUSIVE);
3439 	if (error != 0)
3440 		return (EBADF);
3441 	nmp = VFSTONFS(vp->v_mount);
3442 	if (!NFS_ISV4(vp) || (nmp->nm_flag & NFSMNT_NOLOCKD) != 0) {
3443 		if ((nmp->nm_flag & NFSMNT_NOLOCKD) != 0) {
3444 			size = np->n_size;
3445 			NFSVOPUNLOCK(vp);
3446 			error = lf_advlock(ap, &(vp->v_lockf), size);
3447 		} else {
3448 			if (nfs_advlock_p != NULL)
3449 				error = nfs_advlock_p(ap);
3450 			else {
3451 				NFSVOPUNLOCK(vp);
3452 				error = ENOLCK;
3453 			}
3454 		}
3455 		if (error == 0 && ap->a_op == F_SETLK) {
3456 			error = NFSVOPLOCK(vp, LK_SHARED);
3457 			if (error == 0) {
3458 				/* Mark that a file lock has been acquired. */
3459 				NFSLOCKNODE(np);
3460 				np->n_flag |= NHASBEENLOCKED;
3461 				NFSUNLOCKNODE(np);
3462 				NFSVOPUNLOCK(vp);
3463 			}
3464 		}
3465 		return (error);
3466 	} else if ((ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3467 		if (vp->v_type != VREG) {
3468 			error = EINVAL;
3469 			goto out;
3470 		}
3471 		if ((ap->a_flags & F_POSIX) != 0)
3472 			cred = p->p_ucred;
3473 		else
3474 			cred = td->td_ucred;
3475 
3476 		/*
3477 		 * If this is unlocking a write locked region, flush and
3478 		 * commit them before unlocking. This is required by
3479 		 * RFC3530 Sec. 9.3.2.
3480 		 */
3481 		if (ap->a_op == F_UNLCK &&
3482 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3483 		    ap->a_flags))
3484 			(void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3485 
3486 		/*
3487 		 * Mark NFS node as might have acquired a lock.
3488 		 * This is separate from NHASBEENLOCKED, because it must
3489 		 * be done before the nfsrpc_advlock() call, which might
3490 		 * add a nfscllock structure to the client state.
3491 		 * It is used to check for the case where a nfscllock
3492 		 * state structure cannot exist for the file.
3493 		 * Only done for "oneopenown" NFSv4.1/4.2 mounts.
3494 		 */
3495 		if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) {
3496 			NFSLOCKNODE(np);
3497 			np->n_flag |= NMIGHTBELOCKED;
3498 			NFSUNLOCKNODE(np);
3499 		}
3500 
3501 		/*
3502 		 * Loop around doing the lock op, while a blocking lock
3503 		 * must wait for the lock op to succeed.
3504 		 */
3505 		do {
3506 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3507 			    ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3508 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3509 			    ap->a_op == F_SETLK) {
3510 				NFSVOPUNLOCK(vp);
3511 				error = nfs_catnap(PZERO | PCATCH, ret,
3512 				    "ncladvl");
3513 				if (error)
3514 					return (EINTR);
3515 				NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3516 				if (VN_IS_DOOMED(vp)) {
3517 					error = EBADF;
3518 					goto out;
3519 				}
3520 			}
3521 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3522 		     ap->a_op == F_SETLK);
3523 		if (ret == NFSERR_DENIED) {
3524 			error = EAGAIN;
3525 			goto out;
3526 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3527 			error = ret;
3528 			goto out;
3529 		} else if (ret != 0) {
3530 			error = EACCES;
3531 			goto out;
3532 		}
3533 
3534 		/*
3535 		 * Now, if we just got a lock, invalidate data in the buffer
3536 		 * cache, as required, so that the coherency conforms with
3537 		 * RFC3530 Sec. 9.3.2.
3538 		 */
3539 		if (ap->a_op == F_SETLK) {
3540 			if ((np->n_flag & NMODIFIED) == 0) {
3541 				np->n_attrstamp = 0;
3542 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3543 				ret = VOP_GETATTR(vp, &va, cred);
3544 			}
3545 			if ((np->n_flag & NMODIFIED) || ret ||
3546 			    np->n_change != va.va_filerev) {
3547 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3548 				np->n_attrstamp = 0;
3549 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3550 				ret = VOP_GETATTR(vp, &va, cred);
3551 				if (!ret) {
3552 					np->n_mtime = va.va_mtime;
3553 					np->n_change = va.va_filerev;
3554 				}
3555 			}
3556 			/* Mark that a file lock has been acquired. */
3557 			NFSLOCKNODE(np);
3558 			np->n_flag |= NHASBEENLOCKED;
3559 			NFSUNLOCKNODE(np);
3560 		}
3561 	} else
3562 		error = EOPNOTSUPP;
3563 out:
3564 	NFSVOPUNLOCK(vp);
3565 	return (error);
3566 }
3567 
3568 /*
3569  * NFS advisory byte-level locks.
3570  */
3571 static int
3572 nfs_advlockasync(struct vop_advlockasync_args *ap)
3573 {
3574 	struct vnode *vp = ap->a_vp;
3575 	u_quad_t size;
3576 	int error;
3577 
3578 	error = NFSVOPLOCK(vp, LK_SHARED);
3579 	if (error)
3580 		return (error);
3581 	if (NFS_ISV4(vp)) {
3582 		NFSVOPUNLOCK(vp);
3583 		return (EOPNOTSUPP);
3584 	}
3585 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3586 		size = VTONFS(vp)->n_size;
3587 		NFSVOPUNLOCK(vp);
3588 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
3589 	} else {
3590 		NFSVOPUNLOCK(vp);
3591 		error = EOPNOTSUPP;
3592 	}
3593 	return (error);
3594 }
3595 
3596 /*
3597  * Print out the contents of an nfsnode.
3598  */
3599 static int
3600 nfs_print(struct vop_print_args *ap)
3601 {
3602 	struct vnode *vp = ap->a_vp;
3603 	struct nfsnode *np = VTONFS(vp);
3604 
3605 	printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3606 	    (uintmax_t)np->n_vattr.na_fsid);
3607 	if (vp->v_type == VFIFO)
3608 		fifo_printinfo(vp);
3609 	printf("\n");
3610 	return (0);
3611 }
3612 
3613 /*
3614  * nfs special file access vnode op.
3615  * Essentially just get vattr and then imitate iaccess() since the device is
3616  * local to the client.
3617  */
3618 static int
3619 nfsspec_access(struct vop_access_args *ap)
3620 {
3621 	struct vattr *vap;
3622 	struct ucred *cred = ap->a_cred;
3623 	struct vnode *vp = ap->a_vp;
3624 	accmode_t accmode = ap->a_accmode;
3625 	struct vattr vattr;
3626 	int error;
3627 
3628 	/*
3629 	 * Disallow write attempts on filesystems mounted read-only;
3630 	 * unless the file is a socket, fifo, or a block or character
3631 	 * device resident on the filesystem.
3632 	 */
3633 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3634 		switch (vp->v_type) {
3635 		case VREG:
3636 		case VDIR:
3637 		case VLNK:
3638 			return (EROFS);
3639 		default:
3640 			break;
3641 		}
3642 	}
3643 	vap = &vattr;
3644 	error = VOP_GETATTR(vp, vap, cred);
3645 	if (error)
3646 		goto out;
3647 	error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3648 	    accmode, cred);
3649 out:
3650 	return error;
3651 }
3652 
3653 /*
3654  * Read wrapper for fifos.
3655  */
3656 static int
3657 nfsfifo_read(struct vop_read_args *ap)
3658 {
3659 	struct nfsnode *np = VTONFS(ap->a_vp);
3660 	int error;
3661 
3662 	/*
3663 	 * Set access flag.
3664 	 */
3665 	NFSLOCKNODE(np);
3666 	np->n_flag |= NACC;
3667 	vfs_timestamp(&np->n_atim);
3668 	NFSUNLOCKNODE(np);
3669 	error = fifo_specops.vop_read(ap);
3670 	return error;
3671 }
3672 
3673 /*
3674  * Write wrapper for fifos.
3675  */
3676 static int
3677 nfsfifo_write(struct vop_write_args *ap)
3678 {
3679 	struct nfsnode *np = VTONFS(ap->a_vp);
3680 
3681 	/*
3682 	 * Set update flag.
3683 	 */
3684 	NFSLOCKNODE(np);
3685 	np->n_flag |= NUPD;
3686 	vfs_timestamp(&np->n_mtim);
3687 	NFSUNLOCKNODE(np);
3688 	return(fifo_specops.vop_write(ap));
3689 }
3690 
3691 /*
3692  * Close wrapper for fifos.
3693  *
3694  * Update the times on the nfsnode then do fifo close.
3695  */
3696 static int
3697 nfsfifo_close(struct vop_close_args *ap)
3698 {
3699 	struct vnode *vp = ap->a_vp;
3700 	struct nfsnode *np = VTONFS(vp);
3701 	struct vattr vattr;
3702 	struct timespec ts;
3703 
3704 	NFSLOCKNODE(np);
3705 	if (np->n_flag & (NACC | NUPD)) {
3706 		vfs_timestamp(&ts);
3707 		if (np->n_flag & NACC)
3708 			np->n_atim = ts;
3709 		if (np->n_flag & NUPD)
3710 			np->n_mtim = ts;
3711 		np->n_flag |= NCHG;
3712 		if (vrefcnt(vp) == 1 &&
3713 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3714 			VATTR_NULL(&vattr);
3715 			if (np->n_flag & NACC)
3716 				vattr.va_atime = np->n_atim;
3717 			if (np->n_flag & NUPD)
3718 				vattr.va_mtime = np->n_mtim;
3719 			NFSUNLOCKNODE(np);
3720 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3721 			goto out;
3722 		}
3723 	}
3724 	NFSUNLOCKNODE(np);
3725 out:
3726 	return (fifo_specops.vop_close(ap));
3727 }
3728 
3729 static int
3730 nfs_getacl(struct vop_getacl_args *ap)
3731 {
3732 	int error;
3733 
3734 	if (ap->a_type != ACL_TYPE_NFS4 && ap->a_type != ACL_TYPE_ACCESS &&
3735 	    ap->a_type != ACL_TYPE_DEFAULT)
3736 		return (EOPNOTSUPP);
3737 	if (ap->a_type == ACL_TYPE_DEFAULT && ap->a_vp->v_type != VDIR)
3738 		return (EINVAL);
3739 	error = nfsrpc_getacl(ap->a_vp, ap->a_type, ap->a_cred, ap->a_td,
3740 	    ap->a_aclp);
3741 	if (error == 0 && ap->a_aclp->acl_cnt == 0 &&
3742 	    ap->a_type != ACL_TYPE_DEFAULT)
3743 		return (EOPNOTSUPP);
3744 	if (error > NFSERR_STALE) {
3745 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3746 		error = EPERM;
3747 	}
3748 	return (error);
3749 }
3750 
3751 static int
3752 nfs_setacl(struct vop_setacl_args *ap)
3753 {
3754 	int error;
3755 
3756 	if (ap->a_type != ACL_TYPE_NFS4 && ap->a_type != ACL_TYPE_ACCESS &&
3757 	    ap->a_type != ACL_TYPE_DEFAULT)
3758 		return (EOPNOTSUPP);
3759 	if (ap->a_aclp == NULL) {
3760 		if (ap->a_type != ACL_TYPE_DEFAULT)
3761 			return (EINVAL);
3762 		if (ap->a_vp->v_type != VDIR)
3763 			return (ENOTDIR);
3764 	}
3765 	error = nfsrpc_setacl(ap->a_vp, ap->a_type, ap->a_cred, ap->a_td,
3766 	    ap->a_aclp);
3767 	if (error > NFSERR_STALE) {
3768 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3769 		error = EPERM;
3770 	}
3771 	return (error);
3772 }
3773 
3774 /*
3775  * VOP_ADVISE for NFS.
3776  * Just return 0 for any errors, since it is just a hint.
3777  */
3778 static int
3779 nfs_advise(struct vop_advise_args *ap)
3780 {
3781 	struct thread *td = curthread;
3782 	struct nfsmount *nmp;
3783 	uint64_t len;
3784 	int error;
3785 
3786 	/*
3787 	 * First do vop_stdadvise() to handle the buffer cache.
3788 	 */
3789 	error = vop_stdadvise(ap);
3790 	if (error != 0)
3791 		return (error);
3792 	if (ap->a_start < 0 || ap->a_end < 0)
3793 		return (0);
3794 	if (ap->a_end == OFF_MAX)
3795 		len = 0;
3796 	else if (ap->a_end < ap->a_start)
3797 		return (0);
3798 	else
3799 		len = ap->a_end - ap->a_start + 1;
3800 	nmp = VFSTONFS(ap->a_vp->v_mount);
3801 	mtx_lock(&nmp->nm_mtx);
3802 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3803 	    (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) ==
3804 	    0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) {
3805 		mtx_unlock(&nmp->nm_mtx);
3806 		return (0);
3807 	}
3808 	mtx_unlock(&nmp->nm_mtx);
3809 	error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice,
3810 	    td->td_ucred, td);
3811 	if (error == NFSERR_NOTSUPP) {
3812 		mtx_lock(&nmp->nm_mtx);
3813 		nmp->nm_privflag |= NFSMNTP_NOADVISE;
3814 		mtx_unlock(&nmp->nm_mtx);
3815 	}
3816 	return (0);
3817 }
3818 
3819 /*
3820  * nfs allocate call
3821  */
3822 static int
3823 nfs_allocate(struct vop_allocate_args *ap)
3824 {
3825 	struct vnode *vp = ap->a_vp;
3826 	struct thread *td = curthread;
3827 	struct nfsvattr nfsva;
3828 	struct nfsmount *nmp;
3829 	struct nfsnode *np;
3830 	off_t alen;
3831 	int attrflag, error, ret;
3832 	struct timespec ts;
3833 	struct uio io;
3834 
3835 	attrflag = 0;
3836 	nmp = VFSTONFS(vp->v_mount);
3837 	np = VTONFS(vp);
3838 	mtx_lock(&nmp->nm_mtx);
3839 	if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3840 	    (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) {
3841 		mtx_unlock(&nmp->nm_mtx);
3842 		alen = *ap->a_len;
3843 		if ((uint64_t)alen > nfs_maxalloclen)
3844 			alen = nfs_maxalloclen;
3845 
3846 		/* Check the file size limit. */
3847 		io.uio_offset = *ap->a_offset;
3848 		io.uio_resid = alen;
3849 		error = vn_rlimit_fsize(vp, &io, td);
3850 
3851 		/*
3852 		 * Flush first to ensure that the allocate adds to the
3853 		 * file's allocation on the server.
3854 		 */
3855 		if (error == 0) {
3856 			vnode_pager_clean_sync(vp);
3857 			error = ncl_flush(vp, MNT_WAIT, td, 1, 0);
3858 		}
3859 		if (error == 0)
3860 			error = nfsrpc_allocate(vp, *ap->a_offset, alen,
3861 			    &nfsva, &attrflag, ap->a_cred, td);
3862 		if (error == 0) {
3863 			*ap->a_offset += alen;
3864 			*ap->a_len -= alen;
3865 			nanouptime(&ts);
3866 			NFSLOCKNODE(np);
3867 			np->n_localmodtime = ts;
3868 			NFSUNLOCKNODE(np);
3869 		} else if (error == NFSERR_NOTSUPP) {
3870 			mtx_lock(&nmp->nm_mtx);
3871 			nmp->nm_privflag |= NFSMNTP_NOALLOCATE;
3872 			mtx_unlock(&nmp->nm_mtx);
3873 			error = EOPNOTSUPP;
3874 		}
3875 	} else {
3876 		/*
3877 		 * Pre-v4.2 NFS server that doesn't support it, or a newer
3878 		 * NFS server that has indicated that it doesn't support it.
3879 		 */
3880 		mtx_unlock(&nmp->nm_mtx);
3881 		error = EOPNOTSUPP;
3882 	}
3883 	if (attrflag != 0) {
3884 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
3885 		if (error == 0 && ret != 0)
3886 			error = ret;
3887 	}
3888 	if (error != 0)
3889 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3890 	return (error);
3891 }
3892 
3893 /*
3894  * nfs deallocate call
3895  */
3896 static int
3897 nfs_deallocate(struct vop_deallocate_args *ap)
3898 {
3899 	struct vnode *vp = ap->a_vp;
3900 	struct thread *td = curthread;
3901 	struct nfsvattr nfsva;
3902 	struct nfsmount *nmp;
3903 	struct nfsnode *np;
3904 	off_t tlen, mlen;
3905 	int attrflag, error, ret;
3906 	bool clipped;
3907 	struct timespec ts;
3908 
3909 	error = 0;
3910 	attrflag = 0;
3911 	nmp = VFSTONFS(vp->v_mount);
3912 	np = VTONFS(vp);
3913 	mtx_lock(&nmp->nm_mtx);
3914 	if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3915 	    (nmp->nm_privflag & NFSMNTP_NODEALLOCATE) == 0) {
3916 		mtx_unlock(&nmp->nm_mtx);
3917 		tlen = omin(OFF_MAX - *ap->a_offset, *ap->a_len);
3918 		NFSCL_DEBUG(4, "dealloc: off=%jd len=%jd maxfilesize=%ju\n",
3919 		    (intmax_t)*ap->a_offset, (intmax_t)tlen,
3920 		    (uintmax_t)nmp->nm_maxfilesize);
3921 		if ((uint64_t)*ap->a_offset >= nmp->nm_maxfilesize) {
3922 			/* Avoid EFBIG error return from the NFSv4.2 server. */
3923 			*ap->a_len = 0;
3924 			return (0);
3925 		}
3926 		clipped = false;
3927 		if ((uint64_t)*ap->a_offset + tlen > nmp->nm_maxfilesize)
3928 			tlen = nmp->nm_maxfilesize - *ap->a_offset;
3929 		if ((uint64_t)*ap->a_offset < np->n_size) {
3930 			/* Limit the len to nfs_maxalloclen before EOF. */
3931 			mlen = omin((off_t)np->n_size - *ap->a_offset, tlen);
3932 			if ((uint64_t)mlen > nfs_maxalloclen) {
3933 				NFSCL_DEBUG(4, "dealloc: tlen maxalloclen\n");
3934 				tlen = nfs_maxalloclen;
3935 				clipped = true;
3936 			}
3937 		}
3938 		if (error == 0)
3939 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
3940 		if (error == 0) {
3941 			vnode_pager_purge_range(vp, *ap->a_offset,
3942 			    *ap->a_offset + tlen);
3943 			error = nfsrpc_deallocate(vp, *ap->a_offset, tlen,
3944 			    &nfsva, &attrflag, ap->a_cred, td);
3945 			NFSCL_DEBUG(4, "dealloc: rpc=%d\n", error);
3946 		}
3947 		if (error == 0) {
3948 			NFSCL_DEBUG(4, "dealloc: attrflag=%d na_size=%ju\n",
3949 			    attrflag, (uintmax_t)nfsva.na_size);
3950 			nanouptime(&ts);
3951 			NFSLOCKNODE(np);
3952 			np->n_localmodtime = ts;
3953 			NFSUNLOCKNODE(np);
3954 			if (attrflag != 0) {
3955 				if ((uint64_t)*ap->a_offset < nfsva.na_size)
3956 					*ap->a_offset += omin((off_t)
3957 					    nfsva.na_size - *ap->a_offset,
3958 					    tlen);
3959 			}
3960 			if (clipped && tlen < *ap->a_len)
3961 				*ap->a_len -= tlen;
3962 			else
3963 				*ap->a_len = 0;
3964 		} else if (error == NFSERR_NOTSUPP) {
3965 			mtx_lock(&nmp->nm_mtx);
3966 			nmp->nm_privflag |= NFSMNTP_NODEALLOCATE;
3967 			mtx_unlock(&nmp->nm_mtx);
3968 		}
3969 	} else {
3970 		mtx_unlock(&nmp->nm_mtx);
3971 		error = EIO;
3972 	}
3973 	/*
3974 	 * If the NFS server cannot perform the Deallocate operation, just call
3975 	 * vop_stddeallocate() to perform it.
3976 	 */
3977 	if (error != 0 && error != NFSERR_FBIG && error != NFSERR_INVAL) {
3978 		error = vop_stddeallocate(ap);
3979 		NFSCL_DEBUG(4, "dealloc: stddeallocate=%d\n", error);
3980 	}
3981 	if (attrflag != 0) {
3982 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
3983 		if (error == 0 && ret != 0)
3984 			error = ret;
3985 	}
3986 	if (error != 0)
3987 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3988 	return (error);
3989 }
3990 
3991 /*
3992  * nfs copy_file_range call
3993  */
3994 static int
3995 nfs_copy_file_range(struct vop_copy_file_range_args *ap)
3996 {
3997 	struct vnode *invp = ap->a_invp;
3998 	struct vnode *outvp = ap->a_outvp;
3999 	struct mount *mp;
4000 	vm_object_t invp_obj;
4001 	struct nfsvattr innfsva, outnfsva;
4002 	struct vattr va, *vap;
4003 	struct uio io;
4004 	struct nfsmount *nmp;
4005 	struct nfsnode *np;
4006 	size_t len, len2;
4007 	ssize_t r;
4008 	int error, inattrflag, outattrflag, ret, ret2, invp_lock;
4009 	off_t inoff, outoff;
4010 	bool consecutive, must_commit, onevp, toeof, tryclone, tryoutcred;
4011 	bool mustclone;
4012 
4013 	/*
4014 	 * NFSv4.2 Copy is not permitted for infile == outfile.
4015 	 * The NFSv4.2 Clone operation does work on non-overlapping
4016 	 * byte ranges in the same file, but only if offsets
4017 	 * (and len if not to EOF) are aligned properly.
4018 	 * TODO: copy_file_range() between multiple NFS mountpoints
4019 	 * --> This is not possible now, since each mount appears to
4020 	 *     the NFSv4.n server as a separate client.
4021 	 */
4022 	if ((invp == outvp && (ap->a_flags & COPY_FILE_RANGE_CLONE) == 0) ||
4023 	    (invp != outvp && invp->v_mount != outvp->v_mount)) {
4024 generic_copy:
4025 		return (ENOSYS);
4026 	}
4027 	if (invp == outvp) {
4028 		onevp = true;
4029 		invp_lock = LK_EXCLUSIVE;
4030 	} else {
4031 		onevp = false;
4032 		invp_lock = LK_SHARED;
4033 	}
4034 	mustclone = false;
4035 	if (onevp || (ap->a_flags & COPY_FILE_RANGE_CLONE) != 0)
4036 		mustclone = true;
4037 relock:
4038 	inoff = *ap->a_inoffp;
4039 	outoff = *ap->a_outoffp;
4040 
4041 	/* Lock vnode(s), avoiding risk of deadlock. */
4042 	do {
4043 		mp = NULL;
4044 		error = vn_start_write(outvp, &mp, V_WAIT);
4045 		if (error == 0) {
4046 			error = vn_lock(outvp, LK_EXCLUSIVE);
4047 			if (error == 0) {
4048 				if (onevp)
4049 					break;
4050 				error = vn_lock(invp, invp_lock | LK_NOWAIT);
4051 				if (error == 0)
4052 					break;
4053 				VOP_UNLOCK(outvp);
4054 				if (mp != NULL)
4055 					vn_finished_write(mp);
4056 				mp = NULL;
4057 				error = vn_lock(invp, invp_lock);
4058 				if (error == 0)
4059 					VOP_UNLOCK(invp);
4060 			}
4061 		}
4062 		if (mp != NULL)
4063 			vn_finished_write(mp);
4064 	} while (error == 0);
4065 	if (error != 0)
4066 		return (error);
4067 
4068 	/*
4069 	 * More reasons to avoid nfs copy/clone: not NFSv4.2, explicitly
4070 	 * disabled or requires cloning and unable to clone.
4071 	 * Only clone if the clone_blksize attribute is supported
4072 	 * and the clone_blksize is greater than 0.
4073 	 * Alignment of offsets and length will be checked later.
4074 	 */
4075 	nmp = VFSTONFS(invp->v_mount);
4076 	np = VTONFS(invp);
4077 	mtx_lock(&nmp->nm_mtx);
4078 	if ((nmp->nm_privflag & NFSMNTP_NOCOPY) != 0)
4079 		mustclone = true;
4080 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4081 	    (mustclone && (!NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr,
4082 	     NFSATTRBIT_CLONEBLKSIZE) || nmp->nm_cloneblksize == 0))) {
4083 		mtx_unlock(&nmp->nm_mtx);
4084 		VOP_UNLOCK(invp);
4085 		if (!onevp)
4086 			VOP_UNLOCK(outvp);	/* For onevp, same as invp. */
4087 		if (mp != NULL)
4088 			vn_finished_write(mp);
4089 		goto generic_copy;
4090 	}
4091 	mtx_unlock(&nmp->nm_mtx);
4092 
4093 	/*
4094 	 * Do the vn_rlimit_fsize() check.  Should this be above the VOP layer?
4095 	 */
4096 	io.uio_offset = *ap->a_outoffp;
4097 	io.uio_resid = *ap->a_lenp;
4098 	error = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd);
4099 	*ap->a_lenp = io.uio_resid;
4100 	/*
4101 	 * No need to call vn_rlimit_fsizex_res before return, since the uio is
4102 	 * local.
4103 	 */
4104 
4105 	/*
4106 	 * Flush the input file so that the data is up to date before
4107 	 * the copy.  Flush writes for the output file so that they
4108 	 * do not overwrite the data copied to the output file by the Copy.
4109 	 * Set the commit argument for both flushes so that the data is on
4110 	 * stable storage before the Copy RPC.  This is done in case the
4111 	 * server reboots during the Copy and needs to be redone.
4112 	 */
4113 	if (error == 0) {
4114 		invp_obj = invp->v_object;
4115 		if (invp_obj != NULL && vm_object_mightbedirty(invp_obj)) {
4116 			if (invp_lock != LK_EXCLUSIVE) {
4117 				KASSERT(!onevp, ("nfs_copy_file_range: "
4118 				    "invp_lock LK_SHARED for onevp"));
4119 				invp_lock = LK_EXCLUSIVE;
4120 				VOP_UNLOCK(invp);
4121 				VOP_UNLOCK(outvp);
4122 				if (mp != NULL)
4123 					vn_finished_write(mp);
4124 				goto relock;
4125 			}
4126 			vnode_pager_clean_sync(invp);
4127 		}
4128 		error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0);
4129 	}
4130 	if (error == 0)
4131 		error = ncl_vinvalbuf(outvp, V_SAVE, curthread, 0);
4132 
4133 	/* Do the actual NFSv4.2 RPC. */
4134 	ret = ret2 = 0;
4135 	len = *ap->a_lenp;
4136 	mtx_lock(&nmp->nm_mtx);
4137 	if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0)
4138 		consecutive = true;
4139 	else
4140 		consecutive = false;
4141 	mtx_unlock(&nmp->nm_mtx);
4142 	tryoutcred = true;
4143 	must_commit = false;
4144 	toeof = false;
4145 
4146 	if (error == 0) {
4147 		vap = &VTONFS(invp)->n_vattr.na_vattr;
4148 		error = VOP_GETATTR(invp, vap, ap->a_incred);
4149 		if (error == 0) {
4150 			/*
4151 			 * Clip "len" at va_size so that RFC compliant servers
4152 			 * will not reply NFSERR_INVAL.
4153 			 * Setting "len == 0" for the RPC would be preferred,
4154 			 * but some Linux servers do not support that.
4155 			 * If the len is being set to 0, do a Setattr RPC to
4156 			 * set the server's atime.  This behaviour was the
4157 			 * preferred one for the FreeBSD "collective".
4158 			 */
4159 			if (inoff >= vap->va_size) {
4160 				*ap->a_lenp = len = 0;
4161 				if ((nmp->nm_mountp->mnt_flag & MNT_NOATIME) ==
4162 				    0) {
4163 					VATTR_NULL(&va);
4164 					va.va_atime.tv_sec = 0;
4165 					va.va_atime.tv_nsec = 0;
4166 					va.va_vaflags = VA_UTIMES_NULL;
4167 					inattrflag = 0;
4168 					error = nfsrpc_setattr(invp, &va, NULL,
4169 					    0, ap->a_incred, curthread,
4170 					    &innfsva, &inattrflag);
4171 					if (inattrflag != 0)
4172 						ret = nfscl_loadattrcache(&invp,
4173 						    &innfsva, NULL, 0, 1);
4174 					if (error == 0 && ret != 0)
4175 						error = ret;
4176 				}
4177 			} else if (inoff + len >= vap->va_size) {
4178 				toeof = true;
4179 				*ap->a_lenp = len = vap->va_size - inoff;
4180 			}
4181 		} else
4182 			error = 0;
4183 	}
4184 
4185 	/*
4186 	 * For cloning, the offsets must be clone blksize aligned and
4187 	 * the len must be blksize aligned unless it goes to EOF on
4188 	 * the input file.
4189 	 */
4190 	tryclone = false;
4191 	if (len > 0) {
4192 		if (error == 0 && NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr,
4193 		    NFSATTRBIT_CLONEBLKSIZE) && nmp->nm_cloneblksize != 0 &&
4194 		    (inoff % nmp->nm_cloneblksize) == 0 &&
4195 		    (outoff % nmp->nm_cloneblksize) == 0 &&
4196 		    (toeof || (len % nmp->nm_cloneblksize) == 0))
4197 			tryclone = true;
4198 		else if (mustclone)
4199 			error = ENOSYS;
4200 	}
4201 
4202 	/*
4203 	 * len will be set to 0 upon a successful Copy RPC.
4204 	 * As such, this only loops when the Copy/Clone RPC needs to be retried.
4205 	 */
4206 	while (len > 0 && error == 0) {
4207 		inattrflag = outattrflag = 0;
4208 		len2 = len;
4209 		if (tryclone) {
4210 			if (tryoutcred)
4211 				error = nfsrpc_clone(invp, ap->a_inoffp, outvp,
4212 				    ap->a_outoffp, &len2, toeof, &inattrflag,
4213 				    &innfsva, &outattrflag, &outnfsva,
4214 				    ap->a_outcred);
4215 			else
4216 				error = nfsrpc_clone(invp, ap->a_inoffp, outvp,
4217 				    ap->a_outoffp, &len2, toeof, &inattrflag,
4218 				    &innfsva, &outattrflag, &outnfsva,
4219 				    ap->a_incred);
4220 		} else {
4221 			if (tryoutcred)
4222 				error = nfsrpc_copy_file_range(invp,
4223 				    ap->a_inoffp, outvp, ap->a_outoffp, &len2,
4224 				    ap->a_flags, &inattrflag, &innfsva,
4225 				    &outattrflag, &outnfsva,
4226 				    ap->a_outcred, consecutive, &must_commit);
4227 			else
4228 				error = nfsrpc_copy_file_range(invp,
4229 				    ap->a_inoffp, outvp, ap->a_outoffp, &len2,
4230 				    ap->a_flags, &inattrflag, &innfsva,
4231 				    &outattrflag, &outnfsva,
4232 				    ap->a_incred, consecutive, &must_commit);
4233 		}
4234 		if (inattrflag != 0)
4235 			ret = nfscl_loadattrcache(&invp, &innfsva, NULL, 0, 1);
4236 		if (outattrflag != 0)
4237 			ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL,
4238 			    1, 1);
4239 		if (error == 0) {
4240 			if (consecutive == false) {
4241 				if (len2 == len) {
4242 					mtx_lock(&nmp->nm_mtx);
4243 					nmp->nm_privflag |=
4244 					    NFSMNTP_NOCONSECUTIVE;
4245 					mtx_unlock(&nmp->nm_mtx);
4246 				} else
4247 					error = NFSERR_OFFLOADNOREQS;
4248 			}
4249 			*ap->a_lenp = len2;
4250 			len = 0;
4251 			if (len2 > 0 && must_commit && error == 0)
4252 				error = ncl_commit(outvp, outoff, *ap->a_lenp,
4253 				    ap->a_outcred, curthread);
4254 			if (error == 0 && ret != 0)
4255 				error = ret;
4256 			if (error == 0 && ret2 != 0)
4257 				error = ret2;
4258 		} else if (error == NFSERR_OFFLOADNOREQS && consecutive) {
4259 			/*
4260 			 * Try consecutive == false, which is ok only if all
4261 			 * bytes are copied.
4262 			 * If only some bytes were copied when consecutive
4263 			 * is false, there is no way to know which bytes
4264 			 * still need to be written.
4265 			 */
4266 			consecutive = false;
4267 			error = 0;
4268 		} else if (error == NFSERR_ACCES && tryoutcred) {
4269 			/* Try again with incred. */
4270 			tryoutcred = false;
4271 			error = 0;
4272 		} else if (tryclone && error != 0) {
4273 			if (mustclone) {
4274 				error = ENOSYS;
4275 			} else {
4276 				tryclone = false;
4277 				error = 0;
4278 			}
4279 		}
4280 		if (error == NFSERR_STALEWRITEVERF) {
4281 			/*
4282 			 * Server rebooted, so do it all again.
4283 			 */
4284 			*ap->a_inoffp = inoff;
4285 			*ap->a_outoffp = outoff;
4286 			len = *ap->a_lenp;
4287 			must_commit = false;
4288 			error = 0;
4289 		}
4290 	}
4291 	VOP_UNLOCK(invp);
4292 	if (!onevp)
4293 		VOP_UNLOCK(outvp);	/* For onevp, same as invp. */
4294 	if (mp != NULL)
4295 		vn_finished_write(mp);
4296 	if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS ||
4297 	    error == NFSERR_ACCES || error == ENOSYS) {
4298 		/*
4299 		 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can
4300 		 * use a_incred for the read and a_outcred for the write, so
4301 		 * try this for NFSERR_ACCES failures for the Copy.
4302 		 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can
4303 		 * never succeed, so disable it.
4304 		 */
4305 		if (error != NFSERR_ACCES && error != ENOSYS) {
4306 			/* Can never do Copy on this mount. */
4307 			mtx_lock(&nmp->nm_mtx);
4308 			nmp->nm_privflag |= NFSMNTP_NOCOPY;
4309 			mtx_unlock(&nmp->nm_mtx);
4310 		}
4311 		*ap->a_inoffp = inoff;
4312 		*ap->a_outoffp = outoff;
4313 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
4314 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
4315 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
4316 	} else if (error != 0)
4317 		*ap->a_lenp = 0;
4318 
4319 	if (error != 0)
4320 		error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0);
4321 	return (error);
4322 }
4323 
4324 /*
4325  * nfs ioctl call
4326  */
4327 static int
4328 nfs_ioctl(struct vop_ioctl_args *ap)
4329 {
4330 	struct vnode *vp = ap->a_vp;
4331 	struct nfsvattr nfsva;
4332 	struct nfsmount *nmp;
4333 	int attrflag, content, error, ret;
4334 	bool eof = false;			/* shut up compiler. */
4335 
4336 	/* Do the actual NFSv4.2 RPC. */
4337 	switch (ap->a_command) {
4338 	case FIOSEEKDATA:
4339 		content = NFSV4CONTENT_DATA;
4340 		break;
4341 	case FIOSEEKHOLE:
4342 		content = NFSV4CONTENT_HOLE;
4343 		break;
4344 	default:
4345 		return (ENOTTY);
4346 	}
4347 
4348 	error = vn_lock(vp, LK_EXCLUSIVE);
4349 	if (error != 0)
4350 		return (EBADF);
4351 
4352 	if (vp->v_type != VREG) {
4353 		VOP_UNLOCK(vp);
4354 		return (ENOTTY);
4355 	}
4356 	nmp = VFSTONFS(vp->v_mount);
4357 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) {
4358 		VOP_UNLOCK(vp);
4359 		error = vop_stdioctl(ap);
4360 		return (error);
4361 	}
4362 
4363 	attrflag = 0;
4364 	if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size)
4365 		error = ENXIO;
4366 	else {
4367 		/*
4368 		 * Flush all writes, so that the server is up to date.
4369 		 * Although a Commit is not required, the commit argument
4370 		 * is set so that, for a pNFS File/Flexible File Layout
4371 		 * server, the LayoutCommit will be done to ensure the file
4372 		 * size is up to date on the Metadata Server.
4373 		 */
4374 
4375 		vnode_pager_clean_sync(vp);
4376 		error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0);
4377 		if (error == 0)
4378 			error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof,
4379 			    content, ap->a_cred, &nfsva, &attrflag);
4380 		/* If at eof for FIOSEEKDATA, return ENXIO. */
4381 		if (eof && error == 0 && content == NFSV4CONTENT_DATA)
4382 			error = ENXIO;
4383 	}
4384 	if (attrflag != 0) {
4385 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4386 		if (error == 0 && ret != 0)
4387 			error = ret;
4388 	}
4389 	NFSVOPUNLOCK(vp);
4390 
4391 	if (error != 0)
4392 		error = ENXIO;
4393 	return (error);
4394 }
4395 
4396 /*
4397  * nfs getextattr call
4398  */
4399 static int
4400 nfs_getextattr(struct vop_getextattr_args *ap)
4401 {
4402 	struct vnode *vp = ap->a_vp;
4403 	struct nfsmount *nmp;
4404 	struct ucred *cred;
4405 	struct thread *td = ap->a_td;
4406 	struct nfsvattr nfsva;
4407 	ssize_t len;
4408 	int attrflag, error, ret;
4409 
4410 	nmp = VFSTONFS(vp->v_mount);
4411 	mtx_lock(&nmp->nm_mtx);
4412 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4413 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4414 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4415 		mtx_unlock(&nmp->nm_mtx);
4416 		return (EOPNOTSUPP);
4417 	}
4418 	mtx_unlock(&nmp->nm_mtx);
4419 
4420 	cred = ap->a_cred;
4421 	if (cred == NULL)
4422 		cred = td->td_ucred;
4423 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4424 	attrflag = 0;
4425 	error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva,
4426 	    &attrflag, cred, td);
4427 	if (attrflag != 0) {
4428 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4429 		if (error == 0 && ret != 0)
4430 			error = ret;
4431 	}
4432 	if (error == 0 && ap->a_size != NULL)
4433 		*ap->a_size = len;
4434 
4435 	switch (error) {
4436 	case NFSERR_NOTSUPP:
4437 	case NFSERR_OPILLEGAL:
4438 		mtx_lock(&nmp->nm_mtx);
4439 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4440 		mtx_unlock(&nmp->nm_mtx);
4441 		error = EOPNOTSUPP;
4442 		break;
4443 	case NFSERR_NOXATTR:
4444 	case NFSERR_XATTR2BIG:
4445 		error = ENOATTR;
4446 		break;
4447 	default:
4448 		error = nfscl_maperr(td, error, 0, 0);
4449 		break;
4450 	}
4451 	return (error);
4452 }
4453 
4454 /*
4455  * nfs setextattr call
4456  */
4457 static int
4458 nfs_setextattr(struct vop_setextattr_args *ap)
4459 {
4460 	struct vnode *vp = ap->a_vp;
4461 	struct nfsmount *nmp;
4462 	struct ucred *cred;
4463 	struct thread *td = ap->a_td;
4464 	struct nfsvattr nfsva;
4465 	int attrflag, error, ret;
4466 
4467 	nmp = VFSTONFS(vp->v_mount);
4468 	mtx_lock(&nmp->nm_mtx);
4469 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4470 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4471 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4472 		mtx_unlock(&nmp->nm_mtx);
4473 		return (EOPNOTSUPP);
4474 	}
4475 	mtx_unlock(&nmp->nm_mtx);
4476 
4477 	if (ap->a_uio->uio_resid < 0)
4478 		return (EINVAL);
4479 	cred = ap->a_cred;
4480 	if (cred == NULL)
4481 		cred = td->td_ucred;
4482 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4483 	attrflag = 0;
4484 	error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva,
4485 	    &attrflag, cred, td);
4486 	if (attrflag != 0) {
4487 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4488 		if (error == 0 && ret != 0)
4489 			error = ret;
4490 	}
4491 
4492 	switch (error) {
4493 	case NFSERR_NOTSUPP:
4494 	case NFSERR_OPILLEGAL:
4495 		mtx_lock(&nmp->nm_mtx);
4496 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4497 		mtx_unlock(&nmp->nm_mtx);
4498 		error = EOPNOTSUPP;
4499 		break;
4500 	case NFSERR_NOXATTR:
4501 	case NFSERR_XATTR2BIG:
4502 		error = ENOATTR;
4503 		break;
4504 	default:
4505 		error = nfscl_maperr(td, error, 0, 0);
4506 		break;
4507 	}
4508 	return (error);
4509 }
4510 
4511 /*
4512  * nfs listextattr call
4513  */
4514 static int
4515 nfs_listextattr(struct vop_listextattr_args *ap)
4516 {
4517 	struct vnode *vp = ap->a_vp;
4518 	struct nfsmount *nmp;
4519 	struct ucred *cred;
4520 	struct thread *td = ap->a_td;
4521 	struct nfsvattr nfsva;
4522 	size_t len, len2;
4523 	uint64_t cookie;
4524 	int attrflag, error, ret;
4525 	bool eof;
4526 
4527 	nmp = VFSTONFS(vp->v_mount);
4528 	mtx_lock(&nmp->nm_mtx);
4529 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4530 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4531 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4532 		mtx_unlock(&nmp->nm_mtx);
4533 		return (EOPNOTSUPP);
4534 	}
4535 	mtx_unlock(&nmp->nm_mtx);
4536 
4537 	cred = ap->a_cred;
4538 	if (cred == NULL)
4539 		cred = td->td_ucred;
4540 
4541 	/* Loop around doing List Extended Attribute RPCs. */
4542 	eof = false;
4543 	cookie = 0;
4544 	len2 = 0;
4545 	error = 0;
4546 	while (!eof && error == 0) {
4547 		len = nmp->nm_rsize;
4548 		attrflag = 0;
4549 		error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof,
4550 		    &nfsva, &attrflag, cred, td);
4551 		if (attrflag != 0) {
4552 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4553 			if (error == 0 && ret != 0)
4554 				error = ret;
4555 		}
4556 		if (error == 0) {
4557 			len2 += len;
4558 			if (len2 > SSIZE_MAX)
4559 				error = ENOATTR;
4560 		}
4561 	}
4562 	if (error == 0 && ap->a_size != NULL)
4563 		*ap->a_size = len2;
4564 
4565 	switch (error) {
4566 	case NFSERR_NOTSUPP:
4567 	case NFSERR_OPILLEGAL:
4568 		mtx_lock(&nmp->nm_mtx);
4569 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4570 		mtx_unlock(&nmp->nm_mtx);
4571 		error = EOPNOTSUPP;
4572 		break;
4573 	case NFSERR_NOXATTR:
4574 	case NFSERR_XATTR2BIG:
4575 		error = ENOATTR;
4576 		break;
4577 	default:
4578 		error = nfscl_maperr(td, error, 0, 0);
4579 		break;
4580 	}
4581 	return (error);
4582 }
4583 
4584 /*
4585  * nfs setextattr call
4586  */
4587 static int
4588 nfs_deleteextattr(struct vop_deleteextattr_args *ap)
4589 {
4590 	struct vnode *vp = ap->a_vp;
4591 	struct nfsmount *nmp;
4592 	struct nfsvattr nfsva;
4593 	int attrflag, error, ret;
4594 
4595 	nmp = VFSTONFS(vp->v_mount);
4596 	mtx_lock(&nmp->nm_mtx);
4597 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4598 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4599 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4600 		mtx_unlock(&nmp->nm_mtx);
4601 		return (EOPNOTSUPP);
4602 	}
4603 	mtx_unlock(&nmp->nm_mtx);
4604 
4605 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4606 	attrflag = 0;
4607 	error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred,
4608 	    ap->a_td);
4609 	if (attrflag != 0) {
4610 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4611 		if (error == 0 && ret != 0)
4612 			error = ret;
4613 	}
4614 
4615 	switch (error) {
4616 	case NFSERR_NOTSUPP:
4617 	case NFSERR_OPILLEGAL:
4618 		mtx_lock(&nmp->nm_mtx);
4619 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4620 		mtx_unlock(&nmp->nm_mtx);
4621 		error = EOPNOTSUPP;
4622 		break;
4623 	case NFSERR_NOXATTR:
4624 	case NFSERR_XATTR2BIG:
4625 		error = ENOATTR;
4626 		break;
4627 	default:
4628 		error = nfscl_maperr(ap->a_td, error, 0, 0);
4629 		break;
4630 	}
4631 	return (error);
4632 }
4633 
4634 /*
4635  * Return POSIX pathconf information applicable to nfs filesystems.
4636  */
4637 static int
4638 nfs_pathconf(struct vop_pathconf_args *ap)
4639 {
4640 	struct nfsv3_pathconf pc;
4641 	struct nfsvattr nfsva;
4642 	struct vnode *vp = ap->a_vp;
4643 	struct nfsmount *nmp;
4644 	struct thread *td = curthread;
4645 	off_t off;
4646 	uint32_t clone_blksize;
4647 	bool eof, has_namedattr, named_enabled;
4648 	int attrflag, error;
4649 	struct nfsnode *np;
4650 	uint32_t trueform;
4651 
4652 	nmp = VFSTONFS(vp->v_mount);
4653 	np = VTONFS(vp);
4654 	named_enabled = false;
4655 	has_namedattr = false;
4656 	clone_blksize = 0;
4657 	if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
4658 	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
4659 	    ap->a_name == _PC_NO_TRUNC ||
4660 	    ap->a_name == _PC_CASE_INSENSITIVE)) ||
4661 	    (NFS_ISV4(vp) && (ap->a_name == _PC_ACL_NFS4 ||
4662 	     ap->a_name == _PC_HAS_NAMEDATTR ||
4663 	     ap->a_name == _PC_CLONE_BLKSIZE ||
4664 	     ap->a_name == _PC_ACL_EXTENDED))) {
4665 		/*
4666 		 * Since only the above 5 a_names are returned by the NFSv3
4667 		 * Pathconf RPC, there is no point in doing it for others.
4668 		 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
4669 		 * be used for _PC_ACL_NFS4, _PC_HAS_NAMEDATTR,
4670 		 * and _PC_ACL_EXTENDED as well.
4671 		 */
4672 		trueform = UINT32_MAX;
4673 		error = nfsrpc_pathconf(vp, &pc, &has_namedattr, &clone_blksize,
4674 		    td->td_ucred, td, &nfsva, &attrflag, &trueform);
4675 		if (attrflag != 0)
4676 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4677 		if (error != 0)
4678 			return (error);
4679 	} else if (NFS_ISV4(vp) && ap->a_name == _PC_NAMEDATTR_ENABLED &&
4680 	    (np->n_flag & NNAMEDNOTSUPP) == 0) {
4681 		struct nfsfh *nfhp;
4682 
4683 		error = nfsrpc_openattr(nmp, vp, np->n_fhp->nfh_fh,
4684 		    np->n_fhp->nfh_len, false, td->td_ucred, td, &nfsva, &nfhp,
4685 		    &attrflag);
4686 		named_enabled = true;
4687 		if (error == 0) {
4688 			free(nfhp, M_NFSFH);
4689 		} else if (error == NFSERR_NOTSUPP) {
4690 			named_enabled = false;
4691 			NFSLOCKNODE(np);
4692 			np->n_flag |= NNAMEDNOTSUPP;
4693 			NFSUNLOCKNODE(np);
4694 		}
4695 		error = 0;
4696 	} else {
4697 		/*
4698 		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
4699 		 * just fake them.
4700 		 */
4701 		pc.pc_linkmax = NFS_LINK_MAX;
4702 		pc.pc_namemax = NFS_MAXNAMLEN;
4703 		pc.pc_notrunc = 1;
4704 		pc.pc_chownrestricted = 1;
4705 		pc.pc_caseinsensitive = 0;
4706 		pc.pc_casepreserving = 1;
4707 		error = 0;
4708 	}
4709 	switch (ap->a_name) {
4710 	case _PC_LINK_MAX:
4711 #ifdef _LP64
4712 		*ap->a_retval = pc.pc_linkmax;
4713 #else
4714 		*ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
4715 #endif
4716 		break;
4717 	case _PC_NAME_MAX:
4718 		*ap->a_retval = pc.pc_namemax;
4719 		break;
4720 	case _PC_PIPE_BUF:
4721 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
4722 			*ap->a_retval = PIPE_BUF;
4723 		else
4724 			error = EINVAL;
4725 		break;
4726 	case _PC_CHOWN_RESTRICTED:
4727 		*ap->a_retval = pc.pc_chownrestricted;
4728 		break;
4729 	case _PC_NO_TRUNC:
4730 		*ap->a_retval = pc.pc_notrunc;
4731 		break;
4732 	case _PC_ACL_NFS4:
4733 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
4734 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL) &&
4735 		    (trueform == NFSV4_ACL_MODEL_NFS4 ||
4736 		     trueform == UINT32_MAX))
4737 			*ap->a_retval = 1;
4738 		else
4739 			*ap->a_retval = 0;
4740 		break;
4741 	case _PC_ACL_EXTENDED:
4742 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
4743 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr,
4744 		    NFSATTRBIT_POSIXACCESSACL) &&
4745 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr,
4746 		    NFSATTRBIT_POSIXDEFAULTACL) &&
4747 		    trueform == NFSV4_ACL_MODEL_POSIX_DRAFT)
4748 			*ap->a_retval = 1;
4749 		else
4750 			*ap->a_retval = 0;
4751 		break;
4752 	case _PC_ACL_PATH_MAX:
4753 		if (NFS_ISV4(vp))
4754 			*ap->a_retval = ACL_MAX_ENTRIES;
4755 		else
4756 			*ap->a_retval = 3;
4757 		break;
4758 	case _PC_PRIO_IO:
4759 		*ap->a_retval = 0;
4760 		break;
4761 	case _PC_SYNC_IO:
4762 		*ap->a_retval = 0;
4763 		break;
4764 	case _PC_ALLOC_SIZE_MIN:
4765 		*ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
4766 		break;
4767 	case _PC_FILESIZEBITS:
4768 		if (NFS_ISV34(vp))
4769 			*ap->a_retval = 64;
4770 		else
4771 			*ap->a_retval = 32;
4772 		break;
4773 	case _PC_REC_INCR_XFER_SIZE:
4774 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4775 		break;
4776 	case _PC_REC_MAX_XFER_SIZE:
4777 		*ap->a_retval = -1; /* means ``unlimited'' */
4778 		break;
4779 	case _PC_REC_MIN_XFER_SIZE:
4780 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4781 		break;
4782 	case _PC_REC_XFER_ALIGN:
4783 		*ap->a_retval = PAGE_SIZE;
4784 		break;
4785 	case _PC_SYMLINK_MAX:
4786 		*ap->a_retval = NFS_MAXPATHLEN;
4787 		break;
4788 	case _PC_MIN_HOLE_SIZE:
4789 		/* Only some NFSv4.2 servers support Seek for Holes. */
4790 		*ap->a_retval = 0;
4791 		if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) {
4792 			/*
4793 			 * NFSv4.2 doesn't have an attribute for hole size,
4794 			 * so all we can do is see if the Seek operation is
4795 			 * supported and then use f_iosize as a "best guess".
4796 			 */
4797 			mtx_lock(&nmp->nm_mtx);
4798 			if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) {
4799 				mtx_unlock(&nmp->nm_mtx);
4800 				off = 0;
4801 				attrflag = 0;
4802 				error = nfsrpc_seek(vp, &off, &eof,
4803 				    NFSV4CONTENT_HOLE, td->td_ucred, &nfsva,
4804 				    &attrflag);
4805 				if (attrflag != 0)
4806 					(void) nfscl_loadattrcache(&vp, &nfsva,
4807 					    NULL, 0, 1);
4808 				mtx_lock(&nmp->nm_mtx);
4809 				if (error == NFSERR_NOTSUPP)
4810 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED;
4811 				else
4812 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED |
4813 					    NFSMNTP_SEEK;
4814 				error = 0;
4815 			}
4816 			if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0)
4817 				*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4818 			mtx_unlock(&nmp->nm_mtx);
4819 		}
4820 		break;
4821 	case _PC_NAMEDATTR_ENABLED:
4822 		if (named_enabled)
4823 			*ap->a_retval = 1;
4824 		else
4825 			*ap->a_retval = 0;
4826 		break;
4827 	case _PC_HAS_NAMEDATTR:
4828 		if (has_namedattr)
4829 			*ap->a_retval = 1;
4830 		else
4831 			*ap->a_retval = 0;
4832 		break;
4833 	case _PC_HAS_HIDDENSYSTEM:
4834 		if (NFS_ISV4(vp) && NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr,
4835 		    NFSATTRBIT_ARCHIVE) &&
4836 		    NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr,
4837 		    NFSATTRBIT_HIDDEN) &&
4838 		    NFSISSET_ATTRBIT(&np->n_vattr.na_suppattr,
4839 		    NFSATTRBIT_SYSTEM))
4840 			*ap->a_retval = 1;
4841 		else
4842 			*ap->a_retval = 0;
4843 		break;
4844 	case _PC_CLONE_BLKSIZE:
4845 		*ap->a_retval = clone_blksize;
4846 		break;
4847 	case _PC_CASE_INSENSITIVE:
4848 		*ap->a_retval = pc.pc_caseinsensitive;
4849 		break;
4850 
4851 	default:
4852 		error = vop_stdpathconf(ap);
4853 		break;
4854 	}
4855 	return (error);
4856 }
4857