xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision 5ffd83dbcc34f10e07f6d3e968ae6365869615f4)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	from nfs_vnops.c	8.16 (Berkeley) 5/27/95
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 /*
41  * vnode op calls for Sun NFS version 2, 3 and 4
42  */
43 
44 #include "opt_inet.h"
45 
46 #include <sys/param.h>
47 #include <sys/kernel.h>
48 #include <sys/systm.h>
49 #include <sys/resourcevar.h>
50 #include <sys/proc.h>
51 #include <sys/mount.h>
52 #include <sys/bio.h>
53 #include <sys/buf.h>
54 #include <sys/extattr.h>
55 #include <sys/filio.h>
56 #include <sys/jail.h>
57 #include <sys/malloc.h>
58 #include <sys/mbuf.h>
59 #include <sys/namei.h>
60 #include <sys/socket.h>
61 #include <sys/vnode.h>
62 #include <sys/dirent.h>
63 #include <sys/fcntl.h>
64 #include <sys/lockf.h>
65 #include <sys/stat.h>
66 #include <sys/sysctl.h>
67 #include <sys/signalvar.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_object.h>
72 
73 #include <fs/nfs/nfsport.h>
74 #include <fs/nfsclient/nfsnode.h>
75 #include <fs/nfsclient/nfsmount.h>
76 #include <fs/nfsclient/nfs.h>
77 #include <fs/nfsclient/nfs_kdtrace.h>
78 
79 #include <net/if.h>
80 #include <netinet/in.h>
81 #include <netinet/in_var.h>
82 
83 #include <nfs/nfs_lock.h>
84 
85 #ifdef KDTRACE_HOOKS
86 #include <sys/dtrace_bsd.h>
87 
88 dtrace_nfsclient_accesscache_flush_probe_func_t
89 		dtrace_nfscl_accesscache_flush_done_probe;
90 uint32_t	nfscl_accesscache_flush_done_id;
91 
92 dtrace_nfsclient_accesscache_get_probe_func_t
93 		dtrace_nfscl_accesscache_get_hit_probe,
94 		dtrace_nfscl_accesscache_get_miss_probe;
95 uint32_t	nfscl_accesscache_get_hit_id;
96 uint32_t	nfscl_accesscache_get_miss_id;
97 
98 dtrace_nfsclient_accesscache_load_probe_func_t
99 		dtrace_nfscl_accesscache_load_done_probe;
100 uint32_t	nfscl_accesscache_load_done_id;
101 #endif /* !KDTRACE_HOOKS */
102 
103 /* Defs */
104 #define	TRUE	1
105 #define	FALSE	0
106 
107 extern struct nfsstatsv1 nfsstatsv1;
108 extern int nfsrv_useacl;
109 extern int nfscl_debuglevel;
110 MALLOC_DECLARE(M_NEWNFSREQ);
111 
112 static vop_read_t	nfsfifo_read;
113 static vop_write_t	nfsfifo_write;
114 static vop_close_t	nfsfifo_close;
115 static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
116 		    struct thread *);
117 static vop_lookup_t	nfs_lookup;
118 static vop_create_t	nfs_create;
119 static vop_mknod_t	nfs_mknod;
120 static vop_open_t	nfs_open;
121 static vop_pathconf_t	nfs_pathconf;
122 static vop_close_t	nfs_close;
123 static vop_access_t	nfs_access;
124 static vop_getattr_t	nfs_getattr;
125 static vop_setattr_t	nfs_setattr;
126 static vop_read_t	nfs_read;
127 static vop_fsync_t	nfs_fsync;
128 static vop_remove_t	nfs_remove;
129 static vop_link_t	nfs_link;
130 static vop_rename_t	nfs_rename;
131 static vop_mkdir_t	nfs_mkdir;
132 static vop_rmdir_t	nfs_rmdir;
133 static vop_symlink_t	nfs_symlink;
134 static vop_readdir_t	nfs_readdir;
135 static vop_strategy_t	nfs_strategy;
136 static	int	nfs_lookitup(struct vnode *, char *, int,
137 		    struct ucred *, struct thread *, struct nfsnode **);
138 static	int	nfs_sillyrename(struct vnode *, struct vnode *,
139 		    struct componentname *);
140 static vop_access_t	nfsspec_access;
141 static vop_readlink_t	nfs_readlink;
142 static vop_print_t	nfs_print;
143 static vop_advlock_t	nfs_advlock;
144 static vop_advlockasync_t nfs_advlockasync;
145 static vop_getacl_t nfs_getacl;
146 static vop_setacl_t nfs_setacl;
147 static vop_advise_t nfs_advise;
148 static vop_allocate_t nfs_allocate;
149 static vop_copy_file_range_t nfs_copy_file_range;
150 static vop_ioctl_t nfs_ioctl;
151 static vop_getextattr_t nfs_getextattr;
152 static vop_setextattr_t nfs_setextattr;
153 static vop_listextattr_t nfs_listextattr;
154 static vop_deleteextattr_t nfs_deleteextattr;
155 static vop_lock1_t	nfs_lock;
156 
157 /*
158  * Global vfs data structures for nfs
159  */
160 
161 static struct vop_vector newnfs_vnodeops_nosig = {
162 	.vop_default =		&default_vnodeops,
163 	.vop_access =		nfs_access,
164 	.vop_advlock =		nfs_advlock,
165 	.vop_advlockasync =	nfs_advlockasync,
166 	.vop_close =		nfs_close,
167 	.vop_create =		nfs_create,
168 	.vop_fsync =		nfs_fsync,
169 	.vop_getattr =		nfs_getattr,
170 	.vop_getpages =		ncl_getpages,
171 	.vop_putpages =		ncl_putpages,
172 	.vop_inactive =		ncl_inactive,
173 	.vop_link =		nfs_link,
174 	.vop_lock1 =		nfs_lock,
175 	.vop_lookup =		nfs_lookup,
176 	.vop_mkdir =		nfs_mkdir,
177 	.vop_mknod =		nfs_mknod,
178 	.vop_open =		nfs_open,
179 	.vop_pathconf =		nfs_pathconf,
180 	.vop_print =		nfs_print,
181 	.vop_read =		nfs_read,
182 	.vop_readdir =		nfs_readdir,
183 	.vop_readlink =		nfs_readlink,
184 	.vop_reclaim =		ncl_reclaim,
185 	.vop_remove =		nfs_remove,
186 	.vop_rename =		nfs_rename,
187 	.vop_rmdir =		nfs_rmdir,
188 	.vop_setattr =		nfs_setattr,
189 	.vop_strategy =		nfs_strategy,
190 	.vop_symlink =		nfs_symlink,
191 	.vop_write =		ncl_write,
192 	.vop_getacl =		nfs_getacl,
193 	.vop_setacl =		nfs_setacl,
194 	.vop_advise =		nfs_advise,
195 	.vop_allocate =		nfs_allocate,
196 	.vop_copy_file_range =	nfs_copy_file_range,
197 	.vop_ioctl =		nfs_ioctl,
198 	.vop_getextattr =	nfs_getextattr,
199 	.vop_setextattr =	nfs_setextattr,
200 	.vop_listextattr =	nfs_listextattr,
201 	.vop_deleteextattr =	nfs_deleteextattr,
202 };
203 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig);
204 
205 static int
206 nfs_vnodeops_bypass(struct vop_generic_args *a)
207 {
208 
209 	return (vop_sigdefer(&newnfs_vnodeops_nosig, a));
210 }
211 
212 struct vop_vector newnfs_vnodeops = {
213 	.vop_default =		&default_vnodeops,
214 	.vop_bypass =		nfs_vnodeops_bypass,
215 };
216 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops);
217 
218 static struct vop_vector newnfs_fifoops_nosig = {
219 	.vop_default =		&fifo_specops,
220 	.vop_access =		nfsspec_access,
221 	.vop_close =		nfsfifo_close,
222 	.vop_fsync =		nfs_fsync,
223 	.vop_getattr =		nfs_getattr,
224 	.vop_inactive =		ncl_inactive,
225 	.vop_pathconf =		nfs_pathconf,
226 	.vop_print =		nfs_print,
227 	.vop_read =		nfsfifo_read,
228 	.vop_reclaim =		ncl_reclaim,
229 	.vop_setattr =		nfs_setattr,
230 	.vop_write =		nfsfifo_write,
231 };
232 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig);
233 
234 static int
235 nfs_fifoops_bypass(struct vop_generic_args *a)
236 {
237 
238 	return (vop_sigdefer(&newnfs_fifoops_nosig, a));
239 }
240 
241 struct vop_vector newnfs_fifoops = {
242 	.vop_default =		&default_vnodeops,
243 	.vop_bypass =		nfs_fifoops_bypass,
244 };
245 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops);
246 
247 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
248     struct componentname *cnp, struct vattr *vap);
249 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
250     int namelen, struct ucred *cred, struct thread *td);
251 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
252     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
253     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
254 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
255     struct componentname *scnp, struct sillyrename *sp);
256 
257 /*
258  * Global variables
259  */
260 SYSCTL_DECL(_vfs_nfs);
261 
262 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
263 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
264 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
265 
266 static int	nfs_prime_access_cache = 0;
267 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
268 	   &nfs_prime_access_cache, 0,
269 	   "Prime NFS ACCESS cache when fetching attributes");
270 
271 static int	newnfs_commit_on_close = 0;
272 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
273     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
274 
275 static int	nfs_clean_pages_on_close = 1;
276 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
277 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
278 
279 int newnfs_directio_enable = 0;
280 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
281 	   &newnfs_directio_enable, 0, "Enable NFS directio");
282 
283 int nfs_keep_dirty_on_error;
284 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
285     &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
286 
287 /*
288  * This sysctl allows other processes to mmap a file that has been opened
289  * O_DIRECT by a process.  In general, having processes mmap the file while
290  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
291  * this by default to prevent DoS attacks - to prevent a malicious user from
292  * opening up files O_DIRECT preventing other users from mmap'ing these
293  * files.  "Protected" environments where stricter consistency guarantees are
294  * required can disable this knob.  The process that opened the file O_DIRECT
295  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
296  * meaningful.
297  */
298 int newnfs_directio_allow_mmap = 1;
299 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
300 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
301 
302 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
303 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
304 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
305 
306 /*
307  * SMP Locking Note :
308  * The list of locks after the description of the lock is the ordering
309  * of other locks acquired with the lock held.
310  * np->n_mtx : Protects the fields in the nfsnode.
311        VM Object Lock
312        VI_MTX (acquired indirectly)
313  * nmp->nm_mtx : Protects the fields in the nfsmount.
314        rep->r_mtx
315  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
316  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
317        nmp->nm_mtx
318        rep->r_mtx
319  * rep->r_mtx : Protects the fields in an nfsreq.
320  */
321 
322 static int
323 nfs_lock(struct vop_lock1_args *ap)
324 {
325 	struct vnode *vp;
326 	struct nfsnode *np;
327 	u_quad_t nsize;
328 	int error, lktype;
329 	bool onfault;
330 
331 	vp = ap->a_vp;
332 	lktype = ap->a_flags & LK_TYPE_MASK;
333 	error = VOP_LOCK1_APV(&default_vnodeops, ap);
334 	if (error != 0 || vp->v_op != &newnfs_vnodeops)
335 		return (error);
336 	np = VTONFS(vp);
337 	if (np == NULL)
338 		return (0);
339 	NFSLOCKNODE(np);
340 	if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED &&
341 	    lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE &&
342 	    lktype != LK_TRYUPGRADE)) {
343 		NFSUNLOCKNODE(np);
344 		return (0);
345 	}
346 	onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT &&
347 	    (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE &&
348 	    (lktype == LK_SHARED || lktype == LK_EXCLUSIVE);
349 	if (onfault && vp->v_vnlock->lk_recurse == 0) {
350 		/*
351 		 * Force retry in vm_fault(), to make the lock request
352 		 * sleepable, which allows us to piggy-back the
353 		 * sleepable call to vnode_pager_setsize().
354 		 */
355 		NFSUNLOCKNODE(np);
356 		VOP_UNLOCK(vp);
357 		return (EBUSY);
358 	}
359 	if ((ap->a_flags & LK_NOWAIT) != 0 ||
360 	    (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) {
361 		NFSUNLOCKNODE(np);
362 		return (0);
363 	}
364 	if (lktype == LK_SHARED) {
365 		NFSUNLOCKNODE(np);
366 		VOP_UNLOCK(vp);
367 		ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
368 		ap->a_flags |= LK_EXCLUSIVE;
369 		error = VOP_LOCK1_APV(&default_vnodeops, ap);
370 		if (error != 0 || vp->v_op != &newnfs_vnodeops)
371 			return (error);
372 		if (vp->v_data == NULL)
373 			goto downgrade;
374 		MPASS(vp->v_data == np);
375 		NFSLOCKNODE(np);
376 		if ((np->n_flag & NVNSETSZSKIP) == 0) {
377 			NFSUNLOCKNODE(np);
378 			goto downgrade;
379 		}
380 	}
381 	np->n_flag &= ~NVNSETSZSKIP;
382 	nsize = np->n_size;
383 	NFSUNLOCKNODE(np);
384 	vnode_pager_setsize(vp, nsize);
385 downgrade:
386 	if (lktype == LK_SHARED) {
387 		ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
388 		ap->a_flags |= LK_DOWNGRADE;
389 		(void)VOP_LOCK1_APV(&default_vnodeops, ap);
390 	}
391 	return (0);
392 }
393 
394 static int
395 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
396     struct ucred *cred, u_int32_t *retmode)
397 {
398 	int error = 0, attrflag, i, lrupos;
399 	u_int32_t rmode;
400 	struct nfsnode *np = VTONFS(vp);
401 	struct nfsvattr nfsva;
402 
403 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
404 	    &rmode, NULL);
405 	if (attrflag)
406 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
407 	if (!error) {
408 		lrupos = 0;
409 		NFSLOCKNODE(np);
410 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
411 			if (np->n_accesscache[i].uid == cred->cr_uid) {
412 				np->n_accesscache[i].mode = rmode;
413 				np->n_accesscache[i].stamp = time_second;
414 				break;
415 			}
416 			if (i > 0 && np->n_accesscache[i].stamp <
417 			    np->n_accesscache[lrupos].stamp)
418 				lrupos = i;
419 		}
420 		if (i == NFS_ACCESSCACHESIZE) {
421 			np->n_accesscache[lrupos].uid = cred->cr_uid;
422 			np->n_accesscache[lrupos].mode = rmode;
423 			np->n_accesscache[lrupos].stamp = time_second;
424 		}
425 		NFSUNLOCKNODE(np);
426 		if (retmode != NULL)
427 			*retmode = rmode;
428 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
429 	} else if (NFS_ISV4(vp)) {
430 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
431 	}
432 #ifdef KDTRACE_HOOKS
433 	if (error != 0)
434 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
435 		    error);
436 #endif
437 	return (error);
438 }
439 
440 /*
441  * nfs access vnode op.
442  * For nfs version 2, just return ok. File accesses may fail later.
443  * For nfs version 3, use the access rpc to check accessibility. If file modes
444  * are changed on the server, accesses might still fail later.
445  */
446 static int
447 nfs_access(struct vop_access_args *ap)
448 {
449 	struct vnode *vp = ap->a_vp;
450 	int error = 0, i, gotahit;
451 	u_int32_t mode, wmode, rmode;
452 	int v34 = NFS_ISV34(vp);
453 	struct nfsnode *np = VTONFS(vp);
454 
455 	/*
456 	 * Disallow write attempts on filesystems mounted read-only;
457 	 * unless the file is a socket, fifo, or a block or character
458 	 * device resident on the filesystem.
459 	 */
460 	if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
461 	    VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
462 	    VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
463 		switch (vp->v_type) {
464 		case VREG:
465 		case VDIR:
466 		case VLNK:
467 			return (EROFS);
468 		default:
469 			break;
470 		}
471 	}
472 	/*
473 	 * For nfs v3 or v4, check to see if we have done this recently, and if
474 	 * so return our cached result instead of making an ACCESS call.
475 	 * If not, do an access rpc, otherwise you are stuck emulating
476 	 * ufs_access() locally using the vattr. This may not be correct,
477 	 * since the server may apply other access criteria such as
478 	 * client uid-->server uid mapping that we do not know about.
479 	 */
480 	if (v34) {
481 		if (ap->a_accmode & VREAD)
482 			mode = NFSACCESS_READ;
483 		else
484 			mode = 0;
485 		if (vp->v_type != VDIR) {
486 			if (ap->a_accmode & VWRITE)
487 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
488 			if (ap->a_accmode & VAPPEND)
489 				mode |= NFSACCESS_EXTEND;
490 			if (ap->a_accmode & VEXEC)
491 				mode |= NFSACCESS_EXECUTE;
492 			if (ap->a_accmode & VDELETE)
493 				mode |= NFSACCESS_DELETE;
494 		} else {
495 			if (ap->a_accmode & VWRITE)
496 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
497 			if (ap->a_accmode & VAPPEND)
498 				mode |= NFSACCESS_EXTEND;
499 			if (ap->a_accmode & VEXEC)
500 				mode |= NFSACCESS_LOOKUP;
501 			if (ap->a_accmode & VDELETE)
502 				mode |= NFSACCESS_DELETE;
503 			if (ap->a_accmode & VDELETE_CHILD)
504 				mode |= NFSACCESS_MODIFY;
505 		}
506 		/* XXX safety belt, only make blanket request if caching */
507 		if (nfsaccess_cache_timeout > 0) {
508 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
509 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
510 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
511 		} else {
512 			wmode = mode;
513 		}
514 
515 		/*
516 		 * Does our cached result allow us to give a definite yes to
517 		 * this request?
518 		 */
519 		gotahit = 0;
520 		NFSLOCKNODE(np);
521 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
522 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
523 			    if (time_second < (np->n_accesscache[i].stamp
524 				+ nfsaccess_cache_timeout) &&
525 				(np->n_accesscache[i].mode & mode) == mode) {
526 				NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
527 				gotahit = 1;
528 			    }
529 			    break;
530 			}
531 		}
532 		NFSUNLOCKNODE(np);
533 #ifdef KDTRACE_HOOKS
534 		if (gotahit != 0)
535 			KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
536 			    ap->a_cred->cr_uid, mode);
537 		else
538 			KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
539 			    ap->a_cred->cr_uid, mode);
540 #endif
541 		if (gotahit == 0) {
542 			/*
543 			 * Either a no, or a don't know.  Go to the wire.
544 			 */
545 			NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
546 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
547 			    ap->a_cred, &rmode);
548 			if (!error &&
549 			    (rmode & mode) != mode)
550 				error = EACCES;
551 		}
552 		return (error);
553 	} else {
554 		if ((error = nfsspec_access(ap)) != 0) {
555 			return (error);
556 		}
557 		/*
558 		 * Attempt to prevent a mapped root from accessing a file
559 		 * which it shouldn't.  We try to read a byte from the file
560 		 * if the user is root and the file is not zero length.
561 		 * After calling nfsspec_access, we should have the correct
562 		 * file size cached.
563 		 */
564 		NFSLOCKNODE(np);
565 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
566 		    && VTONFS(vp)->n_size > 0) {
567 			struct iovec aiov;
568 			struct uio auio;
569 			char buf[1];
570 
571 			NFSUNLOCKNODE(np);
572 			aiov.iov_base = buf;
573 			aiov.iov_len = 1;
574 			auio.uio_iov = &aiov;
575 			auio.uio_iovcnt = 1;
576 			auio.uio_offset = 0;
577 			auio.uio_resid = 1;
578 			auio.uio_segflg = UIO_SYSSPACE;
579 			auio.uio_rw = UIO_READ;
580 			auio.uio_td = ap->a_td;
581 
582 			if (vp->v_type == VREG)
583 				error = ncl_readrpc(vp, &auio, ap->a_cred);
584 			else if (vp->v_type == VDIR) {
585 				char* bp;
586 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
587 				aiov.iov_base = bp;
588 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
589 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
590 				    ap->a_td);
591 				free(bp, M_TEMP);
592 			} else if (vp->v_type == VLNK)
593 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
594 			else
595 				error = EACCES;
596 		} else
597 			NFSUNLOCKNODE(np);
598 		return (error);
599 	}
600 }
601 
602 
603 /*
604  * nfs open vnode op
605  * Check to see if the type is ok
606  * and that deletion is not in progress.
607  * For paged in text files, you will need to flush the page cache
608  * if consistency is lost.
609  */
610 /* ARGSUSED */
611 static int
612 nfs_open(struct vop_open_args *ap)
613 {
614 	struct vnode *vp = ap->a_vp;
615 	struct nfsnode *np = VTONFS(vp);
616 	struct vattr vattr;
617 	int error;
618 	int fmode = ap->a_mode;
619 	struct ucred *cred;
620 	vm_object_t obj;
621 
622 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
623 		return (EOPNOTSUPP);
624 
625 	/*
626 	 * For NFSv4, we need to do the Open Op before cache validation,
627 	 * so that we conform to RFC3530 Sec. 9.3.1.
628 	 */
629 	if (NFS_ISV4(vp)) {
630 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
631 		if (error) {
632 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
633 			    (gid_t)0);
634 			return (error);
635 		}
636 	}
637 
638 	/*
639 	 * Now, if this Open will be doing reading, re-validate/flush the
640 	 * cache, so that Close/Open coherency is maintained.
641 	 */
642 	NFSLOCKNODE(np);
643 	if (np->n_flag & NMODIFIED) {
644 		NFSUNLOCKNODE(np);
645 		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
646 		if (error == EINTR || error == EIO) {
647 			if (NFS_ISV4(vp))
648 				(void) nfsrpc_close(vp, 0, ap->a_td);
649 			return (error);
650 		}
651 		NFSLOCKNODE(np);
652 		np->n_attrstamp = 0;
653 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
654 		if (vp->v_type == VDIR)
655 			np->n_direofoffset = 0;
656 		NFSUNLOCKNODE(np);
657 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
658 		if (error) {
659 			if (NFS_ISV4(vp))
660 				(void) nfsrpc_close(vp, 0, ap->a_td);
661 			return (error);
662 		}
663 		NFSLOCKNODE(np);
664 		np->n_mtime = vattr.va_mtime;
665 		if (NFS_ISV4(vp))
666 			np->n_change = vattr.va_filerev;
667 	} else {
668 		NFSUNLOCKNODE(np);
669 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
670 		if (error) {
671 			if (NFS_ISV4(vp))
672 				(void) nfsrpc_close(vp, 0, ap->a_td);
673 			return (error);
674 		}
675 		NFSLOCKNODE(np);
676 		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
677 		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
678 			if (vp->v_type == VDIR)
679 				np->n_direofoffset = 0;
680 			NFSUNLOCKNODE(np);
681 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
682 			if (error == EINTR || error == EIO) {
683 				if (NFS_ISV4(vp))
684 					(void) nfsrpc_close(vp, 0, ap->a_td);
685 				return (error);
686 			}
687 			NFSLOCKNODE(np);
688 			np->n_mtime = vattr.va_mtime;
689 			if (NFS_ISV4(vp))
690 				np->n_change = vattr.va_filerev;
691 		}
692 	}
693 
694 	/*
695 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
696 	 */
697 	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
698 	    (vp->v_type == VREG)) {
699 		if (np->n_directio_opens == 0) {
700 			NFSUNLOCKNODE(np);
701 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
702 			if (error) {
703 				if (NFS_ISV4(vp))
704 					(void) nfsrpc_close(vp, 0, ap->a_td);
705 				return (error);
706 			}
707 			NFSLOCKNODE(np);
708 			np->n_flag |= NNONCACHE;
709 		}
710 		np->n_directio_opens++;
711 	}
712 
713 	/* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
714 	if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
715 		np->n_flag |= NWRITEOPENED;
716 
717 	/*
718 	 * If this is an open for writing, capture a reference to the
719 	 * credentials, so they can be used by ncl_putpages(). Using
720 	 * these write credentials is preferable to the credentials of
721 	 * whatever thread happens to be doing the VOP_PUTPAGES() since
722 	 * the write RPCs are less likely to fail with EACCES.
723 	 */
724 	if ((fmode & FWRITE) != 0) {
725 		cred = np->n_writecred;
726 		np->n_writecred = crhold(ap->a_cred);
727 	} else
728 		cred = NULL;
729 	NFSUNLOCKNODE(np);
730 
731 	if (cred != NULL)
732 		crfree(cred);
733 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
734 
735 	/*
736 	 * If the text file has been mmap'd, flush any dirty pages to the
737 	 * buffer cache and then...
738 	 * Make sure all writes are pushed to the NFS server.  If this is not
739 	 * done, the modify time of the file can change while the text
740 	 * file is being executed.  This will cause the process that is
741 	 * executing the text file to be terminated.
742 	 */
743 	if (vp->v_writecount <= -1) {
744 		if ((obj = vp->v_object) != NULL &&
745 		    vm_object_mightbedirty(obj)) {
746 			VM_OBJECT_WLOCK(obj);
747 			vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
748 			VM_OBJECT_WUNLOCK(obj);
749 		}
750 
751 		/* Now, flush the buffer cache. */
752 		ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
753 
754 		/* And, finally, make sure that n_mtime is up to date. */
755 		np = VTONFS(vp);
756 		NFSLOCKNODE(np);
757 		np->n_mtime = np->n_vattr.na_mtime;
758 		NFSUNLOCKNODE(np);
759 	}
760 	return (0);
761 }
762 
763 /*
764  * nfs close vnode op
765  * What an NFS client should do upon close after writing is a debatable issue.
766  * Most NFS clients push delayed writes to the server upon close, basically for
767  * two reasons:
768  * 1 - So that any write errors may be reported back to the client process
769  *     doing the close system call. By far the two most likely errors are
770  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
771  * 2 - To put a worst case upper bound on cache inconsistency between
772  *     multiple clients for the file.
773  * There is also a consistency problem for Version 2 of the protocol w.r.t.
774  * not being able to tell if other clients are writing a file concurrently,
775  * since there is no way of knowing if the changed modify time in the reply
776  * is only due to the write for this client.
777  * (NFS Version 3 provides weak cache consistency data in the reply that
778  *  should be sufficient to detect and handle this case.)
779  *
780  * The current code does the following:
781  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
782  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
783  *                     or commit them (this satisfies 1 and 2 except for the
784  *                     case where the server crashes after this close but
785  *                     before the commit RPC, which is felt to be "good
786  *                     enough". Changing the last argument to ncl_flush() to
787  *                     a 1 would force a commit operation, if it is felt a
788  *                     commit is necessary now.
789  * for NFS Version 4 - flush the dirty buffers and commit them, if
790  *		       nfscl_mustflush() says this is necessary.
791  *                     It is necessary if there is no write delegation held,
792  *                     in order to satisfy open/close coherency.
793  *                     If the file isn't cached on local stable storage,
794  *                     it may be necessary in order to detect "out of space"
795  *                     errors from the server, if the write delegation
796  *                     issued by the server doesn't allow the file to grow.
797  */
798 /* ARGSUSED */
799 static int
800 nfs_close(struct vop_close_args *ap)
801 {
802 	struct vnode *vp = ap->a_vp;
803 	struct nfsnode *np = VTONFS(vp);
804 	struct nfsvattr nfsva;
805 	struct ucred *cred;
806 	int error = 0, ret, localcred = 0;
807 	int fmode = ap->a_fflag;
808 
809 	if (NFSCL_FORCEDISM(vp->v_mount))
810 		return (0);
811 	/*
812 	 * During shutdown, a_cred isn't valid, so just use root.
813 	 */
814 	if (ap->a_cred == NOCRED) {
815 		cred = newnfs_getcred();
816 		localcred = 1;
817 	} else {
818 		cred = ap->a_cred;
819 	}
820 	if (vp->v_type == VREG) {
821 	    /*
822 	     * Examine and clean dirty pages, regardless of NMODIFIED.
823 	     * This closes a major hole in close-to-open consistency.
824 	     * We want to push out all dirty pages (and buffers) on
825 	     * close, regardless of whether they were dirtied by
826 	     * mmap'ed writes or via write().
827 	     */
828 	    if (nfs_clean_pages_on_close && vp->v_object) {
829 		VM_OBJECT_WLOCK(vp->v_object);
830 		vm_object_page_clean(vp->v_object, 0, 0, 0);
831 		VM_OBJECT_WUNLOCK(vp->v_object);
832 	    }
833 	    NFSLOCKNODE(np);
834 	    if (np->n_flag & NMODIFIED) {
835 		NFSUNLOCKNODE(np);
836 		if (NFS_ISV3(vp)) {
837 		    /*
838 		     * Under NFSv3 we have dirty buffers to dispose of.  We
839 		     * must flush them to the NFS server.  We have the option
840 		     * of waiting all the way through the commit rpc or just
841 		     * waiting for the initial write.  The default is to only
842 		     * wait through the initial write so the data is in the
843 		     * server's cache, which is roughly similar to the state
844 		     * a standard disk subsystem leaves the file in on close().
845 		     *
846 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
847 		     * potential races with other processes, and certainly
848 		     * cannot clear it if we don't commit.
849 		     * These races occur when there is no longer the old
850 		     * traditional vnode locking implemented for Vnode Ops.
851 		     */
852 		    int cm = newnfs_commit_on_close ? 1 : 0;
853 		    error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
854 		    /* np->n_flag &= ~NMODIFIED; */
855 		} else if (NFS_ISV4(vp)) {
856 			if (nfscl_mustflush(vp) != 0) {
857 				int cm = newnfs_commit_on_close ? 1 : 0;
858 				error = ncl_flush(vp, MNT_WAIT, ap->a_td,
859 				    cm, 0);
860 				/*
861 				 * as above w.r.t races when clearing
862 				 * NMODIFIED.
863 				 * np->n_flag &= ~NMODIFIED;
864 				 */
865 			}
866 		} else {
867 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
868 		}
869 		NFSLOCKNODE(np);
870 	    }
871  	    /*
872  	     * Invalidate the attribute cache in all cases.
873  	     * An open is going to fetch fresh attrs any way, other procs
874  	     * on this node that have file open will be forced to do an
875  	     * otw attr fetch, but this is safe.
876 	     * --> A user found that their RPC count dropped by 20% when
877 	     *     this was commented out and I can't see any requirement
878 	     *     for it, so I've disabled it when negative lookups are
879 	     *     enabled. (What does this have to do with negative lookup
880 	     *     caching? Well nothing, except it was reported by the
881 	     *     same user that needed negative lookup caching and I wanted
882 	     *     there to be a way to disable it to see if it
883 	     *     is the cause of some caching/coherency issue that might
884 	     *     crop up.)
885  	     */
886 	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
887 		    np->n_attrstamp = 0;
888 		    KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
889 	    }
890 	    if (np->n_flag & NWRITEERR) {
891 		np->n_flag &= ~NWRITEERR;
892 		error = np->n_error;
893 	    }
894 	    NFSUNLOCKNODE(np);
895 	}
896 
897 	if (NFS_ISV4(vp)) {
898 		/*
899 		 * Get attributes so "change" is up to date.
900 		 */
901 		if (error == 0 && nfscl_mustflush(vp) != 0 &&
902 		    vp->v_type == VREG &&
903 		    (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
904 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
905 			    NULL);
906 			if (!ret) {
907 				np->n_change = nfsva.na_filerev;
908 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
909 				    NULL, 0, 0);
910 			}
911 		}
912 
913 		/*
914 		 * and do the close.
915 		 */
916 		ret = nfsrpc_close(vp, 0, ap->a_td);
917 		if (!error && ret)
918 			error = ret;
919 		if (error)
920 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
921 			    (gid_t)0);
922 	}
923 	if (newnfs_directio_enable)
924 		KASSERT((np->n_directio_asyncwr == 0),
925 			("nfs_close: dirty unflushed (%d) directio buffers\n",
926 			 np->n_directio_asyncwr));
927 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
928 		NFSLOCKNODE(np);
929 		KASSERT((np->n_directio_opens > 0),
930 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
931 		np->n_directio_opens--;
932 		if (np->n_directio_opens == 0)
933 			np->n_flag &= ~NNONCACHE;
934 		NFSUNLOCKNODE(np);
935 	}
936 	if (localcred)
937 		NFSFREECRED(cred);
938 	return (error);
939 }
940 
941 /*
942  * nfs getattr call from vfs.
943  */
944 static int
945 nfs_getattr(struct vop_getattr_args *ap)
946 {
947 	struct vnode *vp = ap->a_vp;
948 	struct thread *td = curthread;	/* XXX */
949 	struct nfsnode *np = VTONFS(vp);
950 	int error = 0;
951 	struct nfsvattr nfsva;
952 	struct vattr *vap = ap->a_vap;
953 	struct vattr vattr;
954 
955 	/*
956 	 * Update local times for special files.
957 	 */
958 	NFSLOCKNODE(np);
959 	if (np->n_flag & (NACC | NUPD))
960 		np->n_flag |= NCHG;
961 	NFSUNLOCKNODE(np);
962 	/*
963 	 * First look in the cache.
964 	 */
965 	if (ncl_getattrcache(vp, &vattr) == 0) {
966 		vap->va_type = vattr.va_type;
967 		vap->va_mode = vattr.va_mode;
968 		vap->va_nlink = vattr.va_nlink;
969 		vap->va_uid = vattr.va_uid;
970 		vap->va_gid = vattr.va_gid;
971 		vap->va_fsid = vattr.va_fsid;
972 		vap->va_fileid = vattr.va_fileid;
973 		vap->va_size = vattr.va_size;
974 		vap->va_blocksize = vattr.va_blocksize;
975 		vap->va_atime = vattr.va_atime;
976 		vap->va_mtime = vattr.va_mtime;
977 		vap->va_ctime = vattr.va_ctime;
978 		vap->va_gen = vattr.va_gen;
979 		vap->va_flags = vattr.va_flags;
980 		vap->va_rdev = vattr.va_rdev;
981 		vap->va_bytes = vattr.va_bytes;
982 		vap->va_filerev = vattr.va_filerev;
983 		/*
984 		 * Get the local modify time for the case of a write
985 		 * delegation.
986 		 */
987 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
988 		return (0);
989 	}
990 
991 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
992 	    nfsaccess_cache_timeout > 0) {
993 		NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
994 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
995 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
996 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
997 			return (0);
998 		}
999 	}
1000 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
1001 	if (!error)
1002 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
1003 	if (!error) {
1004 		/*
1005 		 * Get the local modify time for the case of a write
1006 		 * delegation.
1007 		 */
1008 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
1009 	} else if (NFS_ISV4(vp)) {
1010 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1011 	}
1012 	return (error);
1013 }
1014 
1015 /*
1016  * nfs setattr call.
1017  */
1018 static int
1019 nfs_setattr(struct vop_setattr_args *ap)
1020 {
1021 	struct vnode *vp = ap->a_vp;
1022 	struct nfsnode *np = VTONFS(vp);
1023 	struct thread *td = curthread;	/* XXX */
1024 	struct vattr *vap = ap->a_vap;
1025 	int error = 0;
1026 	u_quad_t tsize;
1027 
1028 #ifndef nolint
1029 	tsize = (u_quad_t)0;
1030 #endif
1031 
1032 	/*
1033 	 * Setting of flags and marking of atimes are not supported.
1034 	 */
1035 	if (vap->va_flags != VNOVAL)
1036 		return (EOPNOTSUPP);
1037 
1038 	/*
1039 	 * Disallow write attempts if the filesystem is mounted read-only.
1040 	 */
1041   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
1042 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
1043 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
1044 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
1045 		return (EROFS);
1046 	if (vap->va_size != VNOVAL) {
1047  		switch (vp->v_type) {
1048  		case VDIR:
1049  			return (EISDIR);
1050  		case VCHR:
1051  		case VBLK:
1052  		case VSOCK:
1053  		case VFIFO:
1054 			if (vap->va_mtime.tv_sec == VNOVAL &&
1055 			    vap->va_atime.tv_sec == VNOVAL &&
1056 			    vap->va_mode == (mode_t)VNOVAL &&
1057 			    vap->va_uid == (uid_t)VNOVAL &&
1058 			    vap->va_gid == (gid_t)VNOVAL)
1059 				return (0);
1060  			vap->va_size = VNOVAL;
1061  			break;
1062  		default:
1063 			/*
1064 			 * Disallow write attempts if the filesystem is
1065 			 * mounted read-only.
1066 			 */
1067 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
1068 				return (EROFS);
1069 			/*
1070 			 *  We run vnode_pager_setsize() early (why?),
1071 			 * we must set np->n_size now to avoid vinvalbuf
1072 			 * V_SAVE races that might setsize a lower
1073 			 * value.
1074 			 */
1075 			NFSLOCKNODE(np);
1076 			tsize = np->n_size;
1077 			NFSUNLOCKNODE(np);
1078 			error = ncl_meta_setsize(vp, td, vap->va_size);
1079 			NFSLOCKNODE(np);
1080  			if (np->n_flag & NMODIFIED) {
1081 			    tsize = np->n_size;
1082 			    NFSUNLOCKNODE(np);
1083 			    error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
1084 			        0 : V_SAVE, td, 1);
1085 			    if (error != 0) {
1086 				    vnode_pager_setsize(vp, tsize);
1087 				    return (error);
1088 			    }
1089 			    /*
1090 			     * Call nfscl_delegmodtime() to set the modify time
1091 			     * locally, as required.
1092 			     */
1093 			    nfscl_delegmodtime(vp);
1094  			} else
1095 			    NFSUNLOCKNODE(np);
1096 			/*
1097 			 * np->n_size has already been set to vap->va_size
1098 			 * in ncl_meta_setsize(). We must set it again since
1099 			 * nfs_loadattrcache() could be called through
1100 			 * ncl_meta_setsize() and could modify np->n_size.
1101 			 */
1102 			NFSLOCKNODE(np);
1103  			np->n_vattr.na_size = np->n_size = vap->va_size;
1104 			NFSUNLOCKNODE(np);
1105   		}
1106   	} else {
1107 		NFSLOCKNODE(np);
1108 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
1109 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
1110 			NFSUNLOCKNODE(np);
1111 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
1112 			if (error == EINTR || error == EIO)
1113 				return (error);
1114 		} else
1115 			NFSUNLOCKNODE(np);
1116 	}
1117 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
1118 	if (error && vap->va_size != VNOVAL) {
1119 		NFSLOCKNODE(np);
1120 		np->n_size = np->n_vattr.na_size = tsize;
1121 		vnode_pager_setsize(vp, tsize);
1122 		NFSUNLOCKNODE(np);
1123 	}
1124 	return (error);
1125 }
1126 
1127 /*
1128  * Do an nfs setattr rpc.
1129  */
1130 static int
1131 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1132     struct thread *td)
1133 {
1134 	struct nfsnode *np = VTONFS(vp);
1135 	int error, ret, attrflag, i;
1136 	struct nfsvattr nfsva;
1137 
1138 	if (NFS_ISV34(vp)) {
1139 		NFSLOCKNODE(np);
1140 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1141 			np->n_accesscache[i].stamp = 0;
1142 		np->n_flag |= NDELEGMOD;
1143 		NFSUNLOCKNODE(np);
1144 		KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1145 	}
1146 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1147 	    NULL);
1148 	if (attrflag) {
1149 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1150 		if (ret && !error)
1151 			error = ret;
1152 	}
1153 	if (error && NFS_ISV4(vp))
1154 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1155 	return (error);
1156 }
1157 
1158 /*
1159  * nfs lookup call, one step at a time...
1160  * First look in cache
1161  * If not found, unlock the directory nfsnode and do the rpc
1162  */
1163 static int
1164 nfs_lookup(struct vop_lookup_args *ap)
1165 {
1166 	struct componentname *cnp = ap->a_cnp;
1167 	struct vnode *dvp = ap->a_dvp;
1168 	struct vnode **vpp = ap->a_vpp;
1169 	struct mount *mp = dvp->v_mount;
1170 	int flags = cnp->cn_flags;
1171 	struct vnode *newvp;
1172 	struct nfsmount *nmp;
1173 	struct nfsnode *np, *newnp;
1174 	int error = 0, attrflag, dattrflag, ltype, ncticks;
1175 	struct thread *td = cnp->cn_thread;
1176 	struct nfsfh *nfhp;
1177 	struct nfsvattr dnfsva, nfsva;
1178 	struct vattr vattr;
1179 	struct timespec nctime;
1180 
1181 	*vpp = NULLVP;
1182 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1183 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1184 		return (EROFS);
1185 	if (dvp->v_type != VDIR)
1186 		return (ENOTDIR);
1187 	nmp = VFSTONFS(mp);
1188 	np = VTONFS(dvp);
1189 
1190 	/* For NFSv4, wait until any remove is done. */
1191 	NFSLOCKNODE(np);
1192 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1193 		np->n_flag |= NREMOVEWANT;
1194 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1195 	}
1196 	NFSUNLOCKNODE(np);
1197 
1198 	error = vn_dir_check_exec(dvp, cnp);
1199 	if (error != 0)
1200 		return (error);
1201 	error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1202 	if (error > 0 && error != ENOENT)
1203 		return (error);
1204 	if (error == -1) {
1205 		/*
1206 		 * Lookups of "." are special and always return the
1207 		 * current directory.  cache_lookup() already handles
1208 		 * associated locking bookkeeping, etc.
1209 		 */
1210 		if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1211 			/* XXX: Is this really correct? */
1212 			if (cnp->cn_nameiop != LOOKUP &&
1213 			    (flags & ISLASTCN))
1214 				cnp->cn_flags |= SAVENAME;
1215 			return (0);
1216 		}
1217 
1218 		/*
1219 		 * We only accept a positive hit in the cache if the
1220 		 * change time of the file matches our cached copy.
1221 		 * Otherwise, we discard the cache entry and fallback
1222 		 * to doing a lookup RPC.  We also only trust cache
1223 		 * entries for less than nm_nametimeo seconds.
1224 		 *
1225 		 * To better handle stale file handles and attributes,
1226 		 * clear the attribute cache of this node if it is a
1227 		 * leaf component, part of an open() call, and not
1228 		 * locally modified before fetching the attributes.
1229 		 * This should allow stale file handles to be detected
1230 		 * here where we can fall back to a LOOKUP RPC to
1231 		 * recover rather than having nfs_open() detect the
1232 		 * stale file handle and failing open(2) with ESTALE.
1233 		 */
1234 		newvp = *vpp;
1235 		newnp = VTONFS(newvp);
1236 		if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1237 		    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1238 		    !(newnp->n_flag & NMODIFIED)) {
1239 			NFSLOCKNODE(newnp);
1240 			newnp->n_attrstamp = 0;
1241 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1242 			NFSUNLOCKNODE(newnp);
1243 		}
1244 		if (nfscl_nodeleg(newvp, 0) == 0 ||
1245 		    ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1246 		    VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1247 		    timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1248 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1249 			if (cnp->cn_nameiop != LOOKUP &&
1250 			    (flags & ISLASTCN))
1251 				cnp->cn_flags |= SAVENAME;
1252 			return (0);
1253 		}
1254 		cache_purge(newvp);
1255 		if (dvp != newvp)
1256 			vput(newvp);
1257 		else
1258 			vrele(newvp);
1259 		*vpp = NULLVP;
1260 	} else if (error == ENOENT) {
1261 		if (VN_IS_DOOMED(dvp))
1262 			return (ENOENT);
1263 		/*
1264 		 * We only accept a negative hit in the cache if the
1265 		 * modification time of the parent directory matches
1266 		 * the cached copy in the name cache entry.
1267 		 * Otherwise, we discard all of the negative cache
1268 		 * entries for this directory.  We also only trust
1269 		 * negative cache entries for up to nm_negnametimeo
1270 		 * seconds.
1271 		 */
1272 		if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1273 		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1274 		    timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1275 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1276 			return (ENOENT);
1277 		}
1278 		cache_purge_negative(dvp);
1279 	}
1280 
1281 	newvp = NULLVP;
1282 	NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1283 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1284 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1285 	    NULL);
1286 	if (dattrflag)
1287 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1288 	if (error) {
1289 		if (newvp != NULLVP) {
1290 			vput(newvp);
1291 			*vpp = NULLVP;
1292 		}
1293 
1294 		if (error != ENOENT) {
1295 			if (NFS_ISV4(dvp))
1296 				error = nfscl_maperr(td, error, (uid_t)0,
1297 				    (gid_t)0);
1298 			return (error);
1299 		}
1300 
1301 		/* The requested file was not found. */
1302 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1303 		    (flags & ISLASTCN)) {
1304 			/*
1305 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1306 			 * VWRITE) here instead of just checking
1307 			 * MNT_RDONLY.
1308 			 */
1309 			if (mp->mnt_flag & MNT_RDONLY)
1310 				return (EROFS);
1311 			cnp->cn_flags |= SAVENAME;
1312 			return (EJUSTRETURN);
1313 		}
1314 
1315 		if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1316 			/*
1317 			 * Cache the modification time of the parent
1318 			 * directory from the post-op attributes in
1319 			 * the name cache entry.  The negative cache
1320 			 * entry will be ignored once the directory
1321 			 * has changed.  Don't bother adding the entry
1322 			 * if the directory has already changed.
1323 			 */
1324 			NFSLOCKNODE(np);
1325 			if (timespeccmp(&np->n_vattr.na_mtime,
1326 			    &dnfsva.na_mtime, ==)) {
1327 				NFSUNLOCKNODE(np);
1328 				cache_enter_time(dvp, NULL, cnp,
1329 				    &dnfsva.na_mtime, NULL);
1330 			} else
1331 				NFSUNLOCKNODE(np);
1332 		}
1333 		return (ENOENT);
1334 	}
1335 
1336 	/*
1337 	 * Handle RENAME case...
1338 	 */
1339 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1340 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1341 			free(nfhp, M_NFSFH);
1342 			return (EISDIR);
1343 		}
1344 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1345 		    LK_EXCLUSIVE);
1346 		if (error)
1347 			return (error);
1348 		newvp = NFSTOV(np);
1349 		if (attrflag)
1350 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1351 			    0, 1);
1352 		*vpp = newvp;
1353 		cnp->cn_flags |= SAVENAME;
1354 		return (0);
1355 	}
1356 
1357 	if (flags & ISDOTDOT) {
1358 		ltype = NFSVOPISLOCKED(dvp);
1359 		error = vfs_busy(mp, MBF_NOWAIT);
1360 		if (error != 0) {
1361 			vfs_ref(mp);
1362 			NFSVOPUNLOCK(dvp);
1363 			error = vfs_busy(mp, 0);
1364 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1365 			vfs_rel(mp);
1366 			if (error == 0 && VN_IS_DOOMED(dvp)) {
1367 				vfs_unbusy(mp);
1368 				error = ENOENT;
1369 			}
1370 			if (error != 0)
1371 				return (error);
1372 		}
1373 		NFSVOPUNLOCK(dvp);
1374 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1375 		    cnp->cn_lkflags);
1376 		if (error == 0)
1377 			newvp = NFSTOV(np);
1378 		vfs_unbusy(mp);
1379 		if (newvp != dvp)
1380 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1381 		if (VN_IS_DOOMED(dvp)) {
1382 			if (error == 0) {
1383 				if (newvp == dvp)
1384 					vrele(newvp);
1385 				else
1386 					vput(newvp);
1387 			}
1388 			error = ENOENT;
1389 		}
1390 		if (error != 0)
1391 			return (error);
1392 		if (attrflag)
1393 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1394 			    0, 1);
1395 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1396 		free(nfhp, M_NFSFH);
1397 		VREF(dvp);
1398 		newvp = dvp;
1399 		if (attrflag)
1400 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1401 			    0, 1);
1402 	} else {
1403 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1404 		    cnp->cn_lkflags);
1405 		if (error)
1406 			return (error);
1407 		newvp = NFSTOV(np);
1408 		if (attrflag)
1409 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1410 			    0, 1);
1411 		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1412 		    !(np->n_flag & NMODIFIED)) {
1413 			/*
1414 			 * Flush the attribute cache when opening a
1415 			 * leaf node to ensure that fresh attributes
1416 			 * are fetched in nfs_open() since we did not
1417 			 * fetch attributes from the LOOKUP reply.
1418 			 */
1419 			NFSLOCKNODE(np);
1420 			np->n_attrstamp = 0;
1421 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1422 			NFSUNLOCKNODE(np);
1423 		}
1424 	}
1425 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1426 		cnp->cn_flags |= SAVENAME;
1427 	if ((cnp->cn_flags & MAKEENTRY) &&
1428 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1429 	    attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1430 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1431 		    newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1432 	*vpp = newvp;
1433 	return (0);
1434 }
1435 
1436 /*
1437  * nfs read call.
1438  * Just call ncl_bioread() to do the work.
1439  */
1440 static int
1441 nfs_read(struct vop_read_args *ap)
1442 {
1443 	struct vnode *vp = ap->a_vp;
1444 
1445 	switch (vp->v_type) {
1446 	case VREG:
1447 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1448 	case VDIR:
1449 		return (EISDIR);
1450 	default:
1451 		return (EOPNOTSUPP);
1452 	}
1453 }
1454 
1455 /*
1456  * nfs readlink call
1457  */
1458 static int
1459 nfs_readlink(struct vop_readlink_args *ap)
1460 {
1461 	struct vnode *vp = ap->a_vp;
1462 
1463 	if (vp->v_type != VLNK)
1464 		return (EINVAL);
1465 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1466 }
1467 
1468 /*
1469  * Do a readlink rpc.
1470  * Called by ncl_doio() from below the buffer cache.
1471  */
1472 int
1473 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1474 {
1475 	int error, ret, attrflag;
1476 	struct nfsvattr nfsva;
1477 
1478 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1479 	    &attrflag, NULL);
1480 	if (attrflag) {
1481 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1482 		if (ret && !error)
1483 			error = ret;
1484 	}
1485 	if (error && NFS_ISV4(vp))
1486 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1487 	return (error);
1488 }
1489 
1490 /*
1491  * nfs read rpc call
1492  * Ditto above
1493  */
1494 int
1495 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1496 {
1497 	int error, ret, attrflag;
1498 	struct nfsvattr nfsva;
1499 	struct nfsmount *nmp;
1500 
1501 	nmp = VFSTONFS(vp->v_mount);
1502 	error = EIO;
1503 	attrflag = 0;
1504 	if (NFSHASPNFS(nmp))
1505 		error = nfscl_doiods(vp, uiop, NULL, NULL,
1506 		    NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1507 	NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1508 	if (error != 0)
1509 		error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1510 		    &attrflag, NULL);
1511 	if (attrflag) {
1512 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1513 		if (ret && !error)
1514 			error = ret;
1515 	}
1516 	if (error && NFS_ISV4(vp))
1517 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1518 	return (error);
1519 }
1520 
1521 /*
1522  * nfs write call
1523  */
1524 int
1525 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1526     int *iomode, int *must_commit, int called_from_strategy)
1527 {
1528 	struct nfsvattr nfsva;
1529 	int error, attrflag, ret;
1530 	struct nfsmount *nmp;
1531 
1532 	nmp = VFSTONFS(vp->v_mount);
1533 	error = EIO;
1534 	attrflag = 0;
1535 	if (NFSHASPNFS(nmp))
1536 		error = nfscl_doiods(vp, uiop, iomode, must_commit,
1537 		    NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1538 	NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1539 	if (error != 0)
1540 		error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1541 		    uiop->uio_td, &nfsva, &attrflag, NULL,
1542 		    called_from_strategy);
1543 	if (attrflag) {
1544 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1545 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1546 			    1);
1547 		else
1548 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1549 			    1);
1550 		if (ret && !error)
1551 			error = ret;
1552 	}
1553 	if (DOINGASYNC(vp))
1554 		*iomode = NFSWRITE_FILESYNC;
1555 	if (error && NFS_ISV4(vp))
1556 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1557 	return (error);
1558 }
1559 
1560 /*
1561  * nfs mknod rpc
1562  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1563  * mode set to specify the file type and the size field for rdev.
1564  */
1565 static int
1566 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1567     struct vattr *vap)
1568 {
1569 	struct nfsvattr nfsva, dnfsva;
1570 	struct vnode *newvp = NULL;
1571 	struct nfsnode *np = NULL, *dnp;
1572 	struct nfsfh *nfhp;
1573 	struct vattr vattr;
1574 	int error = 0, attrflag, dattrflag;
1575 	u_int32_t rdev;
1576 
1577 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1578 		rdev = vap->va_rdev;
1579 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1580 		rdev = 0xffffffff;
1581 	else
1582 		return (EOPNOTSUPP);
1583 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1584 		return (error);
1585 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1586 	    rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1587 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1588 	if (!error) {
1589 		if (!nfhp)
1590 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1591 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1592 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1593 			    NULL);
1594 		if (nfhp)
1595 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1596 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1597 	}
1598 	if (dattrflag)
1599 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1600 	if (!error) {
1601 		newvp = NFSTOV(np);
1602 		if (attrflag != 0) {
1603 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1604 			    0, 1);
1605 			if (error != 0)
1606 				vput(newvp);
1607 		}
1608 	}
1609 	if (!error) {
1610 		*vpp = newvp;
1611 	} else if (NFS_ISV4(dvp)) {
1612 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1613 		    vap->va_gid);
1614 	}
1615 	dnp = VTONFS(dvp);
1616 	NFSLOCKNODE(dnp);
1617 	dnp->n_flag |= NMODIFIED;
1618 	if (!dattrflag) {
1619 		dnp->n_attrstamp = 0;
1620 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1621 	}
1622 	NFSUNLOCKNODE(dnp);
1623 	return (error);
1624 }
1625 
1626 /*
1627  * nfs mknod vop
1628  * just call nfs_mknodrpc() to do the work.
1629  */
1630 /* ARGSUSED */
1631 static int
1632 nfs_mknod(struct vop_mknod_args *ap)
1633 {
1634 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1635 }
1636 
1637 static struct mtx nfs_cverf_mtx;
1638 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1639     MTX_DEF);
1640 
1641 static nfsquad_t
1642 nfs_get_cverf(void)
1643 {
1644 	static nfsquad_t cverf;
1645 	nfsquad_t ret;
1646 	static int cverf_initialized = 0;
1647 
1648 	mtx_lock(&nfs_cverf_mtx);
1649 	if (cverf_initialized == 0) {
1650 		cverf.lval[0] = arc4random();
1651 		cverf.lval[1] = arc4random();
1652 		cverf_initialized = 1;
1653 	} else
1654 		cverf.qval++;
1655 	ret = cverf;
1656 	mtx_unlock(&nfs_cverf_mtx);
1657 
1658 	return (ret);
1659 }
1660 
1661 /*
1662  * nfs file create call
1663  */
1664 static int
1665 nfs_create(struct vop_create_args *ap)
1666 {
1667 	struct vnode *dvp = ap->a_dvp;
1668 	struct vattr *vap = ap->a_vap;
1669 	struct componentname *cnp = ap->a_cnp;
1670 	struct nfsnode *np = NULL, *dnp;
1671 	struct vnode *newvp = NULL;
1672 	struct nfsmount *nmp;
1673 	struct nfsvattr dnfsva, nfsva;
1674 	struct nfsfh *nfhp;
1675 	nfsquad_t cverf;
1676 	int error = 0, attrflag, dattrflag, fmode = 0;
1677 	struct vattr vattr;
1678 
1679 	/*
1680 	 * Oops, not for me..
1681 	 */
1682 	if (vap->va_type == VSOCK)
1683 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1684 
1685 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1686 		return (error);
1687 	if (vap->va_vaflags & VA_EXCLUSIVE)
1688 		fmode |= O_EXCL;
1689 	dnp = VTONFS(dvp);
1690 	nmp = VFSTONFS(dvp->v_mount);
1691 again:
1692 	/* For NFSv4, wait until any remove is done. */
1693 	NFSLOCKNODE(dnp);
1694 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1695 		dnp->n_flag |= NREMOVEWANT;
1696 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1697 	}
1698 	NFSUNLOCKNODE(dnp);
1699 
1700 	cverf = nfs_get_cverf();
1701 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1702 	    vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1703 	    &nfhp, &attrflag, &dattrflag, NULL);
1704 	if (!error) {
1705 		if (nfhp == NULL)
1706 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1707 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1708 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1709 			    NULL);
1710 		if (nfhp != NULL)
1711 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1712 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1713 	}
1714 	if (dattrflag)
1715 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1716 	if (!error) {
1717 		newvp = NFSTOV(np);
1718 		if (attrflag == 0)
1719 			error = nfsrpc_getattr(newvp, cnp->cn_cred,
1720 			    cnp->cn_thread, &nfsva, NULL);
1721 		if (error == 0)
1722 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1723 			    0, 1);
1724 	}
1725 	if (error) {
1726 		if (newvp != NULL) {
1727 			vput(newvp);
1728 			newvp = NULL;
1729 		}
1730 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1731 		    error == NFSERR_NOTSUPP) {
1732 			fmode &= ~O_EXCL;
1733 			goto again;
1734 		}
1735 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1736 		if (nfscl_checksattr(vap, &nfsva)) {
1737 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1738 			    cnp->cn_thread, &nfsva, &attrflag, NULL);
1739 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1740 			    vap->va_gid != (gid_t)VNOVAL)) {
1741 				/* try again without setting uid/gid */
1742 				vap->va_uid = (uid_t)VNOVAL;
1743 				vap->va_gid = (uid_t)VNOVAL;
1744 				error = nfsrpc_setattr(newvp, vap, NULL,
1745 				    cnp->cn_cred, cnp->cn_thread, &nfsva,
1746 				    &attrflag, NULL);
1747 			}
1748 			if (attrflag)
1749 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1750 				    NULL, 0, 1);
1751 			if (error != 0)
1752 				vput(newvp);
1753 		}
1754 	}
1755 	if (!error) {
1756 		if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1757 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1758 			    NULL);
1759 		*ap->a_vpp = newvp;
1760 	} else if (NFS_ISV4(dvp)) {
1761 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1762 		    vap->va_gid);
1763 	}
1764 	NFSLOCKNODE(dnp);
1765 	dnp->n_flag |= NMODIFIED;
1766 	if (!dattrflag) {
1767 		dnp->n_attrstamp = 0;
1768 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1769 	}
1770 	NFSUNLOCKNODE(dnp);
1771 	return (error);
1772 }
1773 
1774 /*
1775  * nfs file remove call
1776  * To try and make nfs semantics closer to ufs semantics, a file that has
1777  * other processes using the vnode is renamed instead of removed and then
1778  * removed later on the last close.
1779  * - If v_usecount > 1
1780  *	  If a rename is not already in the works
1781  *	     call nfs_sillyrename() to set it up
1782  *     else
1783  *	  do the remove rpc
1784  */
1785 static int
1786 nfs_remove(struct vop_remove_args *ap)
1787 {
1788 	struct vnode *vp = ap->a_vp;
1789 	struct vnode *dvp = ap->a_dvp;
1790 	struct componentname *cnp = ap->a_cnp;
1791 	struct nfsnode *np = VTONFS(vp);
1792 	int error = 0;
1793 	struct vattr vattr;
1794 
1795 	KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1796 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1797 	if (vp->v_type == VDIR)
1798 		error = EPERM;
1799 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1800 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1801 	    vattr.va_nlink > 1)) {
1802 		/*
1803 		 * Purge the name cache so that the chance of a lookup for
1804 		 * the name succeeding while the remove is in progress is
1805 		 * minimized. Without node locking it can still happen, such
1806 		 * that an I/O op returns ESTALE, but since you get this if
1807 		 * another host removes the file..
1808 		 */
1809 		cache_purge(vp);
1810 		/*
1811 		 * throw away biocache buffers, mainly to avoid
1812 		 * unnecessary delayed writes later.
1813 		 */
1814 		error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1815 		if (error != EINTR && error != EIO)
1816 			/* Do the rpc */
1817 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1818 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1819 		/*
1820 		 * Kludge City: If the first reply to the remove rpc is lost..
1821 		 *   the reply to the retransmitted request will be ENOENT
1822 		 *   since the file was in fact removed
1823 		 *   Therefore, we cheat and return success.
1824 		 */
1825 		if (error == ENOENT)
1826 			error = 0;
1827 	} else if (!np->n_sillyrename)
1828 		error = nfs_sillyrename(dvp, vp, cnp);
1829 	NFSLOCKNODE(np);
1830 	np->n_attrstamp = 0;
1831 	NFSUNLOCKNODE(np);
1832 	KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1833 	return (error);
1834 }
1835 
1836 /*
1837  * nfs file remove rpc called from nfs_inactive
1838  */
1839 int
1840 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1841 {
1842 	/*
1843 	 * Make sure that the directory vnode is still valid.
1844 	 * XXX we should lock sp->s_dvp here.
1845 	 */
1846 	if (sp->s_dvp->v_type == VBAD)
1847 		return (0);
1848 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1849 	    sp->s_cred, NULL));
1850 }
1851 
1852 /*
1853  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1854  */
1855 static int
1856 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1857     int namelen, struct ucred *cred, struct thread *td)
1858 {
1859 	struct nfsvattr dnfsva;
1860 	struct nfsnode *dnp = VTONFS(dvp);
1861 	int error = 0, dattrflag;
1862 
1863 	NFSLOCKNODE(dnp);
1864 	dnp->n_flag |= NREMOVEINPROG;
1865 	NFSUNLOCKNODE(dnp);
1866 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1867 	    &dattrflag, NULL);
1868 	NFSLOCKNODE(dnp);
1869 	if ((dnp->n_flag & NREMOVEWANT)) {
1870 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1871 		NFSUNLOCKNODE(dnp);
1872 		wakeup((caddr_t)dnp);
1873 	} else {
1874 		dnp->n_flag &= ~NREMOVEINPROG;
1875 		NFSUNLOCKNODE(dnp);
1876 	}
1877 	if (dattrflag)
1878 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1879 	NFSLOCKNODE(dnp);
1880 	dnp->n_flag |= NMODIFIED;
1881 	if (!dattrflag) {
1882 		dnp->n_attrstamp = 0;
1883 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1884 	}
1885 	NFSUNLOCKNODE(dnp);
1886 	if (error && NFS_ISV4(dvp))
1887 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1888 	return (error);
1889 }
1890 
1891 /*
1892  * nfs file rename call
1893  */
1894 static int
1895 nfs_rename(struct vop_rename_args *ap)
1896 {
1897 	struct vnode *fvp = ap->a_fvp;
1898 	struct vnode *tvp = ap->a_tvp;
1899 	struct vnode *fdvp = ap->a_fdvp;
1900 	struct vnode *tdvp = ap->a_tdvp;
1901 	struct componentname *tcnp = ap->a_tcnp;
1902 	struct componentname *fcnp = ap->a_fcnp;
1903 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1904 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1905 	struct nfsv4node *newv4 = NULL;
1906 	int error;
1907 
1908 	KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1909 	    (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1910 	/* Check for cross-device rename */
1911 	if ((fvp->v_mount != tdvp->v_mount) ||
1912 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1913 		error = EXDEV;
1914 		goto out;
1915 	}
1916 
1917 	if (fvp == tvp) {
1918 		printf("nfs_rename: fvp == tvp (can't happen)\n");
1919 		error = 0;
1920 		goto out;
1921 	}
1922 	if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1923 		goto out;
1924 
1925 	/*
1926 	 * We have to flush B_DELWRI data prior to renaming
1927 	 * the file.  If we don't, the delayed-write buffers
1928 	 * can be flushed out later after the file has gone stale
1929 	 * under NFSV3.  NFSV2 does not have this problem because
1930 	 * ( as far as I can tell ) it flushes dirty buffers more
1931 	 * often.
1932 	 *
1933 	 * Skip the rename operation if the fsync fails, this can happen
1934 	 * due to the server's volume being full, when we pushed out data
1935 	 * that was written back to our cache earlier. Not checking for
1936 	 * this condition can result in potential (silent) data loss.
1937 	 */
1938 	error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1939 	NFSVOPUNLOCK(fvp);
1940 	if (!error && tvp)
1941 		error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1942 	if (error)
1943 		goto out;
1944 
1945 	/*
1946 	 * If the tvp exists and is in use, sillyrename it before doing the
1947 	 * rename of the new file over it.
1948 	 * XXX Can't sillyrename a directory.
1949 	 */
1950 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1951 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1952 		vput(tvp);
1953 		tvp = NULL;
1954 	}
1955 
1956 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1957 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1958 	    tcnp->cn_thread);
1959 
1960 	if (error == 0 && NFS_ISV4(tdvp)) {
1961 		/*
1962 		 * For NFSv4, check to see if it is the same name and
1963 		 * replace the name, if it is different.
1964 		 */
1965 		newv4 = malloc(
1966 		    sizeof (struct nfsv4node) +
1967 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1968 		    M_NFSV4NODE, M_WAITOK);
1969 		NFSLOCKNODE(tdnp);
1970 		NFSLOCKNODE(fnp);
1971 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1972 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1973 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1974 		      tcnp->cn_namelen) ||
1975 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1976 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1977 			tdnp->n_fhp->nfh_len))) {
1978 #ifdef notdef
1979 { char nnn[100]; int nnnl;
1980 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1981 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1982 nnn[nnnl] = '\0';
1983 printf("ren replace=%s\n",nnn);
1984 }
1985 #endif
1986 			free(fnp->n_v4, M_NFSV4NODE);
1987 			fnp->n_v4 = newv4;
1988 			newv4 = NULL;
1989 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1990 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1991 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1992 			    tdnp->n_fhp->nfh_len);
1993 			NFSBCOPY(tcnp->cn_nameptr,
1994 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1995 		}
1996 		NFSUNLOCKNODE(tdnp);
1997 		NFSUNLOCKNODE(fnp);
1998 		if (newv4 != NULL)
1999 			free(newv4, M_NFSV4NODE);
2000 	}
2001 
2002 	if (fvp->v_type == VDIR) {
2003 		if (tvp != NULL && tvp->v_type == VDIR)
2004 			cache_purge(tdvp);
2005 		cache_purge(fdvp);
2006 	}
2007 
2008 out:
2009 	if (tdvp == tvp)
2010 		vrele(tdvp);
2011 	else
2012 		vput(tdvp);
2013 	if (tvp)
2014 		vput(tvp);
2015 	vrele(fdvp);
2016 	vrele(fvp);
2017 	/*
2018 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2019 	 */
2020 	if (error == ENOENT)
2021 		error = 0;
2022 	return (error);
2023 }
2024 
2025 /*
2026  * nfs file rename rpc called from nfs_remove() above
2027  */
2028 static int
2029 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
2030     struct sillyrename *sp)
2031 {
2032 
2033 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
2034 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
2035 	    scnp->cn_thread));
2036 }
2037 
2038 /*
2039  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2040  */
2041 static int
2042 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
2043     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
2044     int tnamelen, struct ucred *cred, struct thread *td)
2045 {
2046 	struct nfsvattr fnfsva, tnfsva;
2047 	struct nfsnode *fdnp = VTONFS(fdvp);
2048 	struct nfsnode *tdnp = VTONFS(tdvp);
2049 	int error = 0, fattrflag, tattrflag;
2050 
2051 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
2052 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
2053 	    &tattrflag, NULL, NULL);
2054 	NFSLOCKNODE(fdnp);
2055 	fdnp->n_flag |= NMODIFIED;
2056 	if (fattrflag != 0) {
2057 		NFSUNLOCKNODE(fdnp);
2058 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
2059 	} else {
2060 		fdnp->n_attrstamp = 0;
2061 		NFSUNLOCKNODE(fdnp);
2062 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
2063 	}
2064 	NFSLOCKNODE(tdnp);
2065 	tdnp->n_flag |= NMODIFIED;
2066 	if (tattrflag != 0) {
2067 		NFSUNLOCKNODE(tdnp);
2068 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
2069 	} else {
2070 		tdnp->n_attrstamp = 0;
2071 		NFSUNLOCKNODE(tdnp);
2072 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2073 	}
2074 	if (error && NFS_ISV4(fdvp))
2075 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2076 	return (error);
2077 }
2078 
2079 /*
2080  * nfs hard link create call
2081  */
2082 static int
2083 nfs_link(struct vop_link_args *ap)
2084 {
2085 	struct vnode *vp = ap->a_vp;
2086 	struct vnode *tdvp = ap->a_tdvp;
2087 	struct componentname *cnp = ap->a_cnp;
2088 	struct nfsnode *np, *tdnp;
2089 	struct nfsvattr nfsva, dnfsva;
2090 	int error = 0, attrflag, dattrflag;
2091 
2092 	/*
2093 	 * Push all writes to the server, so that the attribute cache
2094 	 * doesn't get "out of sync" with the server.
2095 	 * XXX There should be a better way!
2096 	 */
2097 	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
2098 
2099 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2100 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
2101 	    &dattrflag, NULL);
2102 	tdnp = VTONFS(tdvp);
2103 	NFSLOCKNODE(tdnp);
2104 	tdnp->n_flag |= NMODIFIED;
2105 	if (dattrflag != 0) {
2106 		NFSUNLOCKNODE(tdnp);
2107 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
2108 	} else {
2109 		tdnp->n_attrstamp = 0;
2110 		NFSUNLOCKNODE(tdnp);
2111 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2112 	}
2113 	if (attrflag)
2114 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2115 	else {
2116 		np = VTONFS(vp);
2117 		NFSLOCKNODE(np);
2118 		np->n_attrstamp = 0;
2119 		NFSUNLOCKNODE(np);
2120 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2121 	}
2122 	/*
2123 	 * If negative lookup caching is enabled, I might as well
2124 	 * add an entry for this node. Not necessary for correctness,
2125 	 * but if negative caching is enabled, then the system
2126 	 * must care about lookup caching hit rate, so...
2127 	 */
2128 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2129 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2130 		cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2131 	}
2132 	if (error && NFS_ISV4(vp))
2133 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2134 		    (gid_t)0);
2135 	return (error);
2136 }
2137 
2138 /*
2139  * nfs symbolic link create call
2140  */
2141 static int
2142 nfs_symlink(struct vop_symlink_args *ap)
2143 {
2144 	struct vnode *dvp = ap->a_dvp;
2145 	struct vattr *vap = ap->a_vap;
2146 	struct componentname *cnp = ap->a_cnp;
2147 	struct nfsvattr nfsva, dnfsva;
2148 	struct nfsfh *nfhp;
2149 	struct nfsnode *np = NULL, *dnp;
2150 	struct vnode *newvp = NULL;
2151 	int error = 0, attrflag, dattrflag, ret;
2152 
2153 	vap->va_type = VLNK;
2154 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2155 	    ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2156 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2157 	if (nfhp) {
2158 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2159 		    &np, NULL, LK_EXCLUSIVE);
2160 		if (!ret)
2161 			newvp = NFSTOV(np);
2162 		else if (!error)
2163 			error = ret;
2164 	}
2165 	if (newvp != NULL) {
2166 		if (attrflag)
2167 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2168 			    0, 1);
2169 	} else if (!error) {
2170 		/*
2171 		 * If we do not have an error and we could not extract the
2172 		 * newvp from the response due to the request being NFSv2, we
2173 		 * have to do a lookup in order to obtain a newvp to return.
2174 		 */
2175 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2176 		    cnp->cn_cred, cnp->cn_thread, &np);
2177 		if (!error)
2178 			newvp = NFSTOV(np);
2179 	}
2180 	if (error) {
2181 		if (newvp)
2182 			vput(newvp);
2183 		if (NFS_ISV4(dvp))
2184 			error = nfscl_maperr(cnp->cn_thread, error,
2185 			    vap->va_uid, vap->va_gid);
2186 	} else {
2187 		*ap->a_vpp = newvp;
2188 	}
2189 
2190 	dnp = VTONFS(dvp);
2191 	NFSLOCKNODE(dnp);
2192 	dnp->n_flag |= NMODIFIED;
2193 	if (dattrflag != 0) {
2194 		NFSUNLOCKNODE(dnp);
2195 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2196 	} else {
2197 		dnp->n_attrstamp = 0;
2198 		NFSUNLOCKNODE(dnp);
2199 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2200 	}
2201 	/*
2202 	 * If negative lookup caching is enabled, I might as well
2203 	 * add an entry for this node. Not necessary for correctness,
2204 	 * but if negative caching is enabled, then the system
2205 	 * must care about lookup caching hit rate, so...
2206 	 */
2207 	if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2208 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2209 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2210 	}
2211 	return (error);
2212 }
2213 
2214 /*
2215  * nfs make dir call
2216  */
2217 static int
2218 nfs_mkdir(struct vop_mkdir_args *ap)
2219 {
2220 	struct vnode *dvp = ap->a_dvp;
2221 	struct vattr *vap = ap->a_vap;
2222 	struct componentname *cnp = ap->a_cnp;
2223 	struct nfsnode *np = NULL, *dnp;
2224 	struct vnode *newvp = NULL;
2225 	struct vattr vattr;
2226 	struct nfsfh *nfhp;
2227 	struct nfsvattr nfsva, dnfsva;
2228 	int error = 0, attrflag, dattrflag, ret;
2229 
2230 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2231 		return (error);
2232 	vap->va_type = VDIR;
2233 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2234 	    vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2235 	    &attrflag, &dattrflag, NULL);
2236 	dnp = VTONFS(dvp);
2237 	NFSLOCKNODE(dnp);
2238 	dnp->n_flag |= NMODIFIED;
2239 	if (dattrflag != 0) {
2240 		NFSUNLOCKNODE(dnp);
2241 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2242 	} else {
2243 		dnp->n_attrstamp = 0;
2244 		NFSUNLOCKNODE(dnp);
2245 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2246 	}
2247 	if (nfhp) {
2248 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2249 		    &np, NULL, LK_EXCLUSIVE);
2250 		if (!ret) {
2251 			newvp = NFSTOV(np);
2252 			if (attrflag)
2253 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2254 				NULL, 0, 1);
2255 		} else if (!error)
2256 			error = ret;
2257 	}
2258 	if (!error && newvp == NULL) {
2259 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2260 		    cnp->cn_cred, cnp->cn_thread, &np);
2261 		if (!error) {
2262 			newvp = NFSTOV(np);
2263 			if (newvp->v_type != VDIR)
2264 				error = EEXIST;
2265 		}
2266 	}
2267 	if (error) {
2268 		if (newvp)
2269 			vput(newvp);
2270 		if (NFS_ISV4(dvp))
2271 			error = nfscl_maperr(cnp->cn_thread, error,
2272 			    vap->va_uid, vap->va_gid);
2273 	} else {
2274 		/*
2275 		 * If negative lookup caching is enabled, I might as well
2276 		 * add an entry for this node. Not necessary for correctness,
2277 		 * but if negative caching is enabled, then the system
2278 		 * must care about lookup caching hit rate, so...
2279 		 */
2280 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2281 		    (cnp->cn_flags & MAKEENTRY) &&
2282 		    attrflag != 0 && dattrflag != 0)
2283 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2284 			    &dnfsva.na_ctime);
2285 		*ap->a_vpp = newvp;
2286 	}
2287 	return (error);
2288 }
2289 
2290 /*
2291  * nfs remove directory call
2292  */
2293 static int
2294 nfs_rmdir(struct vop_rmdir_args *ap)
2295 {
2296 	struct vnode *vp = ap->a_vp;
2297 	struct vnode *dvp = ap->a_dvp;
2298 	struct componentname *cnp = ap->a_cnp;
2299 	struct nfsnode *dnp;
2300 	struct nfsvattr dnfsva;
2301 	int error, dattrflag;
2302 
2303 	if (dvp == vp)
2304 		return (EINVAL);
2305 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2306 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2307 	dnp = VTONFS(dvp);
2308 	NFSLOCKNODE(dnp);
2309 	dnp->n_flag |= NMODIFIED;
2310 	if (dattrflag != 0) {
2311 		NFSUNLOCKNODE(dnp);
2312 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2313 	} else {
2314 		dnp->n_attrstamp = 0;
2315 		NFSUNLOCKNODE(dnp);
2316 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2317 	}
2318 
2319 	cache_purge(dvp);
2320 	cache_purge(vp);
2321 	if (error && NFS_ISV4(dvp))
2322 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2323 		    (gid_t)0);
2324 	/*
2325 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2326 	 */
2327 	if (error == ENOENT)
2328 		error = 0;
2329 	return (error);
2330 }
2331 
2332 /*
2333  * nfs readdir call
2334  */
2335 static int
2336 nfs_readdir(struct vop_readdir_args *ap)
2337 {
2338 	struct vnode *vp = ap->a_vp;
2339 	struct nfsnode *np = VTONFS(vp);
2340 	struct uio *uio = ap->a_uio;
2341 	ssize_t tresid, left;
2342 	int error = 0;
2343 	struct vattr vattr;
2344 
2345 	if (ap->a_eofflag != NULL)
2346 		*ap->a_eofflag = 0;
2347 	if (vp->v_type != VDIR)
2348 		return(EPERM);
2349 
2350 	/*
2351 	 * First, check for hit on the EOF offset cache
2352 	 */
2353 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2354 	    (np->n_flag & NMODIFIED) == 0) {
2355 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2356 			NFSLOCKNODE(np);
2357 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2358 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2359 				NFSUNLOCKNODE(np);
2360 				NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2361 				if (ap->a_eofflag != NULL)
2362 					*ap->a_eofflag = 1;
2363 				return (0);
2364 			} else
2365 				NFSUNLOCKNODE(np);
2366 		}
2367 	}
2368 
2369 	/*
2370 	 * NFS always guarantees that directory entries don't straddle
2371 	 * DIRBLKSIZ boundaries.  As such, we need to limit the size
2372 	 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2373 	 * directory entry.
2374 	 */
2375 	left = uio->uio_resid % DIRBLKSIZ;
2376 	if (left == uio->uio_resid)
2377 		return (EINVAL);
2378 	uio->uio_resid -= left;
2379 
2380 	/*
2381 	 * Call ncl_bioread() to do the real work.
2382 	 */
2383 	tresid = uio->uio_resid;
2384 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2385 
2386 	if (!error && uio->uio_resid == tresid) {
2387 		NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2388 		if (ap->a_eofflag != NULL)
2389 			*ap->a_eofflag = 1;
2390 	}
2391 
2392 	/* Add the partial DIRBLKSIZ (left) back in. */
2393 	uio->uio_resid += left;
2394 	return (error);
2395 }
2396 
2397 /*
2398  * Readdir rpc call.
2399  * Called from below the buffer cache by ncl_doio().
2400  */
2401 int
2402 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2403     struct thread *td)
2404 {
2405 	struct nfsvattr nfsva;
2406 	nfsuint64 *cookiep, cookie;
2407 	struct nfsnode *dnp = VTONFS(vp);
2408 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2409 	int error = 0, eof, attrflag;
2410 
2411 	KASSERT(uiop->uio_iovcnt == 1 &&
2412 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2413 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2414 	    ("nfs readdirrpc bad uio"));
2415 
2416 	/*
2417 	 * If there is no cookie, assume directory was stale.
2418 	 */
2419 	ncl_dircookie_lock(dnp);
2420 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2421 	if (cookiep) {
2422 		cookie = *cookiep;
2423 		ncl_dircookie_unlock(dnp);
2424 	} else {
2425 		ncl_dircookie_unlock(dnp);
2426 		return (NFSERR_BAD_COOKIE);
2427 	}
2428 
2429 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2430 		(void)ncl_fsinfo(nmp, vp, cred, td);
2431 
2432 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2433 	    &attrflag, &eof, NULL);
2434 	if (attrflag)
2435 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2436 
2437 	if (!error) {
2438 		/*
2439 		 * We are now either at the end of the directory or have filled
2440 		 * the block.
2441 		 */
2442 		if (eof)
2443 			dnp->n_direofoffset = uiop->uio_offset;
2444 		else {
2445 			if (uiop->uio_resid > 0)
2446 				printf("EEK! readdirrpc resid > 0\n");
2447 			ncl_dircookie_lock(dnp);
2448 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2449 			*cookiep = cookie;
2450 			ncl_dircookie_unlock(dnp);
2451 		}
2452 	} else if (NFS_ISV4(vp)) {
2453 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2454 	}
2455 	return (error);
2456 }
2457 
2458 /*
2459  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2460  */
2461 int
2462 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2463     struct thread *td)
2464 {
2465 	struct nfsvattr nfsva;
2466 	nfsuint64 *cookiep, cookie;
2467 	struct nfsnode *dnp = VTONFS(vp);
2468 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2469 	int error = 0, attrflag, eof;
2470 
2471 	KASSERT(uiop->uio_iovcnt == 1 &&
2472 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2473 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2474 	    ("nfs readdirplusrpc bad uio"));
2475 
2476 	/*
2477 	 * If there is no cookie, assume directory was stale.
2478 	 */
2479 	ncl_dircookie_lock(dnp);
2480 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2481 	if (cookiep) {
2482 		cookie = *cookiep;
2483 		ncl_dircookie_unlock(dnp);
2484 	} else {
2485 		ncl_dircookie_unlock(dnp);
2486 		return (NFSERR_BAD_COOKIE);
2487 	}
2488 
2489 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2490 		(void)ncl_fsinfo(nmp, vp, cred, td);
2491 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2492 	    &attrflag, &eof, NULL);
2493 	if (attrflag)
2494 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2495 
2496 	if (!error) {
2497 		/*
2498 		 * We are now either at end of the directory or have filled the
2499 		 * the block.
2500 		 */
2501 		if (eof)
2502 			dnp->n_direofoffset = uiop->uio_offset;
2503 		else {
2504 			if (uiop->uio_resid > 0)
2505 				printf("EEK! readdirplusrpc resid > 0\n");
2506 			ncl_dircookie_lock(dnp);
2507 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2508 			*cookiep = cookie;
2509 			ncl_dircookie_unlock(dnp);
2510 		}
2511 	} else if (NFS_ISV4(vp)) {
2512 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2513 	}
2514 	return (error);
2515 }
2516 
2517 /*
2518  * Silly rename. To make the NFS filesystem that is stateless look a little
2519  * more like the "ufs" a remove of an active vnode is translated to a rename
2520  * to a funny looking filename that is removed by nfs_inactive on the
2521  * nfsnode. There is the potential for another process on a different client
2522  * to create the same funny name between the nfs_lookitup() fails and the
2523  * nfs_rename() completes, but...
2524  */
2525 static int
2526 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2527 {
2528 	struct sillyrename *sp;
2529 	struct nfsnode *np;
2530 	int error;
2531 	short pid;
2532 	unsigned int lticks;
2533 
2534 	cache_purge(dvp);
2535 	np = VTONFS(vp);
2536 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2537 	sp = malloc(sizeof (struct sillyrename),
2538 	    M_NEWNFSREQ, M_WAITOK);
2539 	sp->s_cred = crhold(cnp->cn_cred);
2540 	sp->s_dvp = dvp;
2541 	VREF(dvp);
2542 
2543 	/*
2544 	 * Fudge together a funny name.
2545 	 * Changing the format of the funny name to accommodate more
2546 	 * sillynames per directory.
2547 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2548 	 * CPU ticks since boot.
2549 	 */
2550 	pid = cnp->cn_thread->td_proc->p_pid;
2551 	lticks = (unsigned int)ticks;
2552 	for ( ; ; ) {
2553 		sp->s_namlen = sprintf(sp->s_name,
2554 				       ".nfs.%08x.%04x4.4", lticks,
2555 				       pid);
2556 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2557 				 cnp->cn_thread, NULL))
2558 			break;
2559 		lticks++;
2560 	}
2561 	error = nfs_renameit(dvp, vp, cnp, sp);
2562 	if (error)
2563 		goto bad;
2564 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2565 		cnp->cn_thread, &np);
2566 	np->n_sillyrename = sp;
2567 	return (0);
2568 bad:
2569 	vrele(sp->s_dvp);
2570 	crfree(sp->s_cred);
2571 	free(sp, M_NEWNFSREQ);
2572 	return (error);
2573 }
2574 
2575 /*
2576  * Look up a file name and optionally either update the file handle or
2577  * allocate an nfsnode, depending on the value of npp.
2578  * npp == NULL	--> just do the lookup
2579  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2580  *			handled too
2581  * *npp != NULL --> update the file handle in the vnode
2582  */
2583 static int
2584 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2585     struct thread *td, struct nfsnode **npp)
2586 {
2587 	struct vnode *newvp = NULL, *vp;
2588 	struct nfsnode *np, *dnp = VTONFS(dvp);
2589 	struct nfsfh *nfhp, *onfhp;
2590 	struct nfsvattr nfsva, dnfsva;
2591 	struct componentname cn;
2592 	int error = 0, attrflag, dattrflag;
2593 	u_int hash;
2594 
2595 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2596 	    &nfhp, &attrflag, &dattrflag, NULL);
2597 	if (dattrflag)
2598 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2599 	if (npp && !error) {
2600 		if (*npp != NULL) {
2601 		    np = *npp;
2602 		    vp = NFSTOV(np);
2603 		    /*
2604 		     * For NFSv4, check to see if it is the same name and
2605 		     * replace the name, if it is different.
2606 		     */
2607 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2608 			(np->n_v4->n4_namelen != len ||
2609 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2610 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2611 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2612 			 dnp->n_fhp->nfh_len))) {
2613 #ifdef notdef
2614 { char nnn[100]; int nnnl;
2615 nnnl = (len < 100) ? len : 99;
2616 bcopy(name, nnn, nnnl);
2617 nnn[nnnl] = '\0';
2618 printf("replace=%s\n",nnn);
2619 }
2620 #endif
2621 			    free(np->n_v4, M_NFSV4NODE);
2622 			    np->n_v4 = malloc(
2623 				sizeof (struct nfsv4node) +
2624 				dnp->n_fhp->nfh_len + len - 1,
2625 				M_NFSV4NODE, M_WAITOK);
2626 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2627 			    np->n_v4->n4_namelen = len;
2628 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2629 				dnp->n_fhp->nfh_len);
2630 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2631 		    }
2632 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2633 			FNV1_32_INIT);
2634 		    onfhp = np->n_fhp;
2635 		    /*
2636 		     * Rehash node for new file handle.
2637 		     */
2638 		    vfs_hash_rehash(vp, hash);
2639 		    np->n_fhp = nfhp;
2640 		    if (onfhp != NULL)
2641 			free(onfhp, M_NFSFH);
2642 		    newvp = NFSTOV(np);
2643 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2644 		    free(nfhp, M_NFSFH);
2645 		    VREF(dvp);
2646 		    newvp = dvp;
2647 		} else {
2648 		    cn.cn_nameptr = name;
2649 		    cn.cn_namelen = len;
2650 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2651 			&np, NULL, LK_EXCLUSIVE);
2652 		    if (error)
2653 			return (error);
2654 		    newvp = NFSTOV(np);
2655 		}
2656 		if (!attrflag && *npp == NULL) {
2657 			if (newvp == dvp)
2658 				vrele(newvp);
2659 			else
2660 				vput(newvp);
2661 			return (ENOENT);
2662 		}
2663 		if (attrflag)
2664 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2665 			    0, 1);
2666 	}
2667 	if (npp && *npp == NULL) {
2668 		if (error) {
2669 			if (newvp) {
2670 				if (newvp == dvp)
2671 					vrele(newvp);
2672 				else
2673 					vput(newvp);
2674 			}
2675 		} else
2676 			*npp = np;
2677 	}
2678 	if (error && NFS_ISV4(dvp))
2679 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2680 	return (error);
2681 }
2682 
2683 /*
2684  * Nfs Version 3 and 4 commit rpc
2685  */
2686 int
2687 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2688    struct thread *td)
2689 {
2690 	struct nfsvattr nfsva;
2691 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2692 	struct nfsnode *np;
2693 	struct uio uio;
2694 	int error, attrflag;
2695 
2696 	np = VTONFS(vp);
2697 	error = EIO;
2698 	attrflag = 0;
2699 	if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2700 		uio.uio_offset = offset;
2701 		uio.uio_resid = cnt;
2702 		error = nfscl_doiods(vp, &uio, NULL, NULL,
2703 		    NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2704 		if (error != 0) {
2705 			NFSLOCKNODE(np);
2706 			np->n_flag &= ~NDSCOMMIT;
2707 			NFSUNLOCKNODE(np);
2708 		}
2709 	}
2710 	if (error != 0) {
2711 		mtx_lock(&nmp->nm_mtx);
2712 		if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2713 			mtx_unlock(&nmp->nm_mtx);
2714 			return (0);
2715 		}
2716 		mtx_unlock(&nmp->nm_mtx);
2717 		error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2718 		    &attrflag, NULL);
2719 	}
2720 	if (attrflag != 0)
2721 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2722 		    0, 1);
2723 	if (error != 0 && NFS_ISV4(vp))
2724 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2725 	return (error);
2726 }
2727 
2728 /*
2729  * Strategy routine.
2730  * For async requests when nfsiod(s) are running, queue the request by
2731  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2732  * request.
2733  */
2734 static int
2735 nfs_strategy(struct vop_strategy_args *ap)
2736 {
2737 	struct buf *bp;
2738 	struct vnode *vp;
2739 	struct ucred *cr;
2740 
2741 	bp = ap->a_bp;
2742 	vp = ap->a_vp;
2743 	KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2744 	KASSERT(!(bp->b_flags & B_DONE),
2745 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2746 
2747 	if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2748 		bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2749 		    DEV_BSIZE);
2750 	if (bp->b_iocmd == BIO_READ)
2751 		cr = bp->b_rcred;
2752 	else
2753 		cr = bp->b_wcred;
2754 
2755 	/*
2756 	 * If the op is asynchronous and an i/o daemon is waiting
2757 	 * queue the request, wake it up and wait for completion
2758 	 * otherwise just do it ourselves.
2759 	 */
2760 	if ((bp->b_flags & B_ASYNC) == 0 ||
2761 	    ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2762 		(void) ncl_doio(vp, bp, cr, curthread, 1);
2763 	return (0);
2764 }
2765 
2766 /*
2767  * fsync vnode op. Just call ncl_flush() with commit == 1.
2768  */
2769 /* ARGSUSED */
2770 static int
2771 nfs_fsync(struct vop_fsync_args *ap)
2772 {
2773 
2774 	if (ap->a_vp->v_type != VREG) {
2775 		/*
2776 		 * For NFS, metadata is changed synchronously on the server,
2777 		 * so there is nothing to flush. Also, ncl_flush() clears
2778 		 * the NMODIFIED flag and that shouldn't be done here for
2779 		 * directories.
2780 		 */
2781 		return (0);
2782 	}
2783 	return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2784 }
2785 
2786 /*
2787  * Flush all the blocks associated with a vnode.
2788  * 	Walk through the buffer pool and push any dirty pages
2789  *	associated with the vnode.
2790  * If the called_from_renewthread argument is TRUE, it has been called
2791  * from the NFSv4 renew thread and, as such, cannot block indefinitely
2792  * waiting for a buffer write to complete.
2793  */
2794 int
2795 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2796     int commit, int called_from_renewthread)
2797 {
2798 	struct nfsnode *np = VTONFS(vp);
2799 	struct buf *bp;
2800 	int i;
2801 	struct buf *nbp;
2802 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2803 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2804 	int passone = 1, trycnt = 0;
2805 	u_quad_t off, endoff, toff;
2806 	struct ucred* wcred = NULL;
2807 	struct buf **bvec = NULL;
2808 	struct bufobj *bo;
2809 #ifndef NFS_COMMITBVECSIZ
2810 #define	NFS_COMMITBVECSIZ	20
2811 #endif
2812 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2813 	u_int bvecsize = 0, bveccount;
2814 
2815 	if (called_from_renewthread != 0)
2816 		slptimeo = hz;
2817 	if (nmp->nm_flag & NFSMNT_INT)
2818 		slpflag = PCATCH;
2819 	if (!commit)
2820 		passone = 0;
2821 	bo = &vp->v_bufobj;
2822 	/*
2823 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2824 	 * server, but has not been committed to stable storage on the server
2825 	 * yet. On the first pass, the byte range is worked out and the commit
2826 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2827 	 * job.
2828 	 */
2829 again:
2830 	off = (u_quad_t)-1;
2831 	endoff = 0;
2832 	bvecpos = 0;
2833 	if (NFS_ISV34(vp) && commit) {
2834 		if (bvec != NULL && bvec != bvec_on_stack)
2835 			free(bvec, M_TEMP);
2836 		/*
2837 		 * Count up how many buffers waiting for a commit.
2838 		 */
2839 		bveccount = 0;
2840 		BO_LOCK(bo);
2841 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2842 			if (!BUF_ISLOCKED(bp) &&
2843 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2844 				== (B_DELWRI | B_NEEDCOMMIT))
2845 				bveccount++;
2846 		}
2847 		/*
2848 		 * Allocate space to remember the list of bufs to commit.  It is
2849 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2850 		 * If we can't get memory (for whatever reason), we will end up
2851 		 * committing the buffers one-by-one in the loop below.
2852 		 */
2853 		if (bveccount > NFS_COMMITBVECSIZ) {
2854 			/*
2855 			 * Release the vnode interlock to avoid a lock
2856 			 * order reversal.
2857 			 */
2858 			BO_UNLOCK(bo);
2859 			bvec = (struct buf **)
2860 				malloc(bveccount * sizeof(struct buf *),
2861 				       M_TEMP, M_NOWAIT);
2862 			BO_LOCK(bo);
2863 			if (bvec == NULL) {
2864 				bvec = bvec_on_stack;
2865 				bvecsize = NFS_COMMITBVECSIZ;
2866 			} else
2867 				bvecsize = bveccount;
2868 		} else {
2869 			bvec = bvec_on_stack;
2870 			bvecsize = NFS_COMMITBVECSIZ;
2871 		}
2872 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2873 			if (bvecpos >= bvecsize)
2874 				break;
2875 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2876 				nbp = TAILQ_NEXT(bp, b_bobufs);
2877 				continue;
2878 			}
2879 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2880 			    (B_DELWRI | B_NEEDCOMMIT)) {
2881 				BUF_UNLOCK(bp);
2882 				nbp = TAILQ_NEXT(bp, b_bobufs);
2883 				continue;
2884 			}
2885 			BO_UNLOCK(bo);
2886 			bremfree(bp);
2887 			/*
2888 			 * Work out if all buffers are using the same cred
2889 			 * so we can deal with them all with one commit.
2890 			 *
2891 			 * NOTE: we are not clearing B_DONE here, so we have
2892 			 * to do it later on in this routine if we intend to
2893 			 * initiate I/O on the bp.
2894 			 *
2895 			 * Note: to avoid loopback deadlocks, we do not
2896 			 * assign b_runningbufspace.
2897 			 */
2898 			if (wcred == NULL)
2899 				wcred = bp->b_wcred;
2900 			else if (wcred != bp->b_wcred)
2901 				wcred = NOCRED;
2902 			vfs_busy_pages(bp, 1);
2903 
2904 			BO_LOCK(bo);
2905 			/*
2906 			 * bp is protected by being locked, but nbp is not
2907 			 * and vfs_busy_pages() may sleep.  We have to
2908 			 * recalculate nbp.
2909 			 */
2910 			nbp = TAILQ_NEXT(bp, b_bobufs);
2911 
2912 			/*
2913 			 * A list of these buffers is kept so that the
2914 			 * second loop knows which buffers have actually
2915 			 * been committed. This is necessary, since there
2916 			 * may be a race between the commit rpc and new
2917 			 * uncommitted writes on the file.
2918 			 */
2919 			bvec[bvecpos++] = bp;
2920 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2921 				bp->b_dirtyoff;
2922 			if (toff < off)
2923 				off = toff;
2924 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2925 			if (toff > endoff)
2926 				endoff = toff;
2927 		}
2928 		BO_UNLOCK(bo);
2929 	}
2930 	if (bvecpos > 0) {
2931 		/*
2932 		 * Commit data on the server, as required.
2933 		 * If all bufs are using the same wcred, then use that with
2934 		 * one call for all of them, otherwise commit each one
2935 		 * separately.
2936 		 */
2937 		if (wcred != NOCRED)
2938 			retv = ncl_commit(vp, off, (int)(endoff - off),
2939 					  wcred, td);
2940 		else {
2941 			retv = 0;
2942 			for (i = 0; i < bvecpos; i++) {
2943 				off_t off, size;
2944 				bp = bvec[i];
2945 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2946 					bp->b_dirtyoff;
2947 				size = (u_quad_t)(bp->b_dirtyend
2948 						  - bp->b_dirtyoff);
2949 				retv = ncl_commit(vp, off, (int)size,
2950 						  bp->b_wcred, td);
2951 				if (retv) break;
2952 			}
2953 		}
2954 
2955 		if (retv == NFSERR_STALEWRITEVERF)
2956 			ncl_clearcommit(vp->v_mount);
2957 
2958 		/*
2959 		 * Now, either mark the blocks I/O done or mark the
2960 		 * blocks dirty, depending on whether the commit
2961 		 * succeeded.
2962 		 */
2963 		for (i = 0; i < bvecpos; i++) {
2964 			bp = bvec[i];
2965 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2966 			if (retv) {
2967 				/*
2968 				 * Error, leave B_DELWRI intact
2969 				 */
2970 				vfs_unbusy_pages(bp);
2971 				brelse(bp);
2972 			} else {
2973 				/*
2974 				 * Success, remove B_DELWRI ( bundirty() ).
2975 				 *
2976 				 * b_dirtyoff/b_dirtyend seem to be NFS
2977 				 * specific.  We should probably move that
2978 				 * into bundirty(). XXX
2979 				 */
2980 				bufobj_wref(bo);
2981 				bp->b_flags |= B_ASYNC;
2982 				bundirty(bp);
2983 				bp->b_flags &= ~B_DONE;
2984 				bp->b_ioflags &= ~BIO_ERROR;
2985 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2986 				bufdone(bp);
2987 			}
2988 		}
2989 	}
2990 
2991 	/*
2992 	 * Start/do any write(s) that are required.
2993 	 */
2994 loop:
2995 	BO_LOCK(bo);
2996 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2997 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2998 			if (waitfor != MNT_WAIT || passone)
2999 				continue;
3000 
3001 			error = BUF_TIMELOCK(bp,
3002 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3003 			    BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
3004 			if (error == 0) {
3005 				BUF_UNLOCK(bp);
3006 				goto loop;
3007 			}
3008 			if (error == ENOLCK) {
3009 				error = 0;
3010 				goto loop;
3011 			}
3012 			if (called_from_renewthread != 0) {
3013 				/*
3014 				 * Return EIO so the flush will be retried
3015 				 * later.
3016 				 */
3017 				error = EIO;
3018 				goto done;
3019 			}
3020 			if (newnfs_sigintr(nmp, td)) {
3021 				error = EINTR;
3022 				goto done;
3023 			}
3024 			if (slpflag == PCATCH) {
3025 				slpflag = 0;
3026 				slptimeo = 2 * hz;
3027 			}
3028 			goto loop;
3029 		}
3030 		if ((bp->b_flags & B_DELWRI) == 0)
3031 			panic("nfs_fsync: not dirty");
3032 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3033 			BUF_UNLOCK(bp);
3034 			continue;
3035 		}
3036 		BO_UNLOCK(bo);
3037 		bremfree(bp);
3038 		bp->b_flags |= B_ASYNC;
3039 		bwrite(bp);
3040 		if (newnfs_sigintr(nmp, td)) {
3041 			error = EINTR;
3042 			goto done;
3043 		}
3044 		goto loop;
3045 	}
3046 	if (passone) {
3047 		passone = 0;
3048 		BO_UNLOCK(bo);
3049 		goto again;
3050 	}
3051 	if (waitfor == MNT_WAIT) {
3052 		while (bo->bo_numoutput) {
3053 			error = bufobj_wwait(bo, slpflag, slptimeo);
3054 			if (error) {
3055 			    BO_UNLOCK(bo);
3056 			    if (called_from_renewthread != 0) {
3057 				/*
3058 				 * Return EIO so that the flush will be
3059 				 * retried later.
3060 				 */
3061 				error = EIO;
3062 				goto done;
3063 			    }
3064 			    error = newnfs_sigintr(nmp, td);
3065 			    if (error)
3066 				goto done;
3067 			    if (slpflag == PCATCH) {
3068 				slpflag = 0;
3069 				slptimeo = 2 * hz;
3070 			    }
3071 			    BO_LOCK(bo);
3072 			}
3073 		}
3074 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
3075 			BO_UNLOCK(bo);
3076 			goto loop;
3077 		}
3078 		/*
3079 		 * Wait for all the async IO requests to drain
3080 		 */
3081 		BO_UNLOCK(bo);
3082 		NFSLOCKNODE(np);
3083 		while (np->n_directio_asyncwr > 0) {
3084 			np->n_flag |= NFSYNCWAIT;
3085 			error = newnfs_msleep(td, &np->n_directio_asyncwr,
3086 			    &np->n_mtx, slpflag | (PRIBIO + 1),
3087 			    "nfsfsync", 0);
3088 			if (error) {
3089 				if (newnfs_sigintr(nmp, td)) {
3090 					NFSUNLOCKNODE(np);
3091 					error = EINTR;
3092 					goto done;
3093 				}
3094 			}
3095 		}
3096 		NFSUNLOCKNODE(np);
3097 	} else
3098 		BO_UNLOCK(bo);
3099 	if (NFSHASPNFS(nmp)) {
3100 		nfscl_layoutcommit(vp, td);
3101 		/*
3102 		 * Invalidate the attribute cache, since writes to a DS
3103 		 * won't update the size attribute.
3104 		 */
3105 		NFSLOCKNODE(np);
3106 		np->n_attrstamp = 0;
3107 	} else
3108 		NFSLOCKNODE(np);
3109 	if (np->n_flag & NWRITEERR) {
3110 		error = np->n_error;
3111 		np->n_flag &= ~NWRITEERR;
3112 	}
3113   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
3114 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3115   		np->n_flag &= ~NMODIFIED;
3116 	NFSUNLOCKNODE(np);
3117 done:
3118 	if (bvec != NULL && bvec != bvec_on_stack)
3119 		free(bvec, M_TEMP);
3120 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
3121 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
3122 	    np->n_directio_asyncwr != 0)) {
3123 		if (trycnt++ < 5) {
3124 			/* try, try again... */
3125 			passone = 1;
3126 			wcred = NULL;
3127 			bvec = NULL;
3128 			bvecsize = 0;
3129 			goto again;
3130 		}
3131 		vn_printf(vp, "ncl_flush failed");
3132 		error = called_from_renewthread != 0 ? EIO : EBUSY;
3133 	}
3134 	return (error);
3135 }
3136 
3137 /*
3138  * NFS advisory byte-level locks.
3139  */
3140 static int
3141 nfs_advlock(struct vop_advlock_args *ap)
3142 {
3143 	struct vnode *vp = ap->a_vp;
3144 	struct ucred *cred;
3145 	struct nfsnode *np = VTONFS(ap->a_vp);
3146 	struct proc *p = (struct proc *)ap->a_id;
3147 	struct thread *td = curthread;	/* XXX */
3148 	struct vattr va;
3149 	int ret, error;
3150 	u_quad_t size;
3151 
3152 	error = NFSVOPLOCK(vp, LK_SHARED);
3153 	if (error != 0)
3154 		return (EBADF);
3155 	if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3156 		if (vp->v_type != VREG) {
3157 			error = EINVAL;
3158 			goto out;
3159 		}
3160 		if ((ap->a_flags & F_POSIX) != 0)
3161 			cred = p->p_ucred;
3162 		else
3163 			cred = td->td_ucred;
3164 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
3165 		if (VN_IS_DOOMED(vp)) {
3166 			error = EBADF;
3167 			goto out;
3168 		}
3169 
3170 		/*
3171 		 * If this is unlocking a write locked region, flush and
3172 		 * commit them before unlocking. This is required by
3173 		 * RFC3530 Sec. 9.3.2.
3174 		 */
3175 		if (ap->a_op == F_UNLCK &&
3176 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3177 		    ap->a_flags))
3178 			(void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3179 
3180 		/*
3181 		 * Loop around doing the lock op, while a blocking lock
3182 		 * must wait for the lock op to succeed.
3183 		 */
3184 		do {
3185 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3186 			    ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3187 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3188 			    ap->a_op == F_SETLK) {
3189 				NFSVOPUNLOCK(vp);
3190 				error = nfs_catnap(PZERO | PCATCH, ret,
3191 				    "ncladvl");
3192 				if (error)
3193 					return (EINTR);
3194 				NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3195 				if (VN_IS_DOOMED(vp)) {
3196 					error = EBADF;
3197 					goto out;
3198 				}
3199 			}
3200 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3201 		     ap->a_op == F_SETLK);
3202 		if (ret == NFSERR_DENIED) {
3203 			error = EAGAIN;
3204 			goto out;
3205 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3206 			error = ret;
3207 			goto out;
3208 		} else if (ret != 0) {
3209 			error = EACCES;
3210 			goto out;
3211 		}
3212 
3213 		/*
3214 		 * Now, if we just got a lock, invalidate data in the buffer
3215 		 * cache, as required, so that the coherency conforms with
3216 		 * RFC3530 Sec. 9.3.2.
3217 		 */
3218 		if (ap->a_op == F_SETLK) {
3219 			if ((np->n_flag & NMODIFIED) == 0) {
3220 				np->n_attrstamp = 0;
3221 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3222 				ret = VOP_GETATTR(vp, &va, cred);
3223 			}
3224 			if ((np->n_flag & NMODIFIED) || ret ||
3225 			    np->n_change != va.va_filerev) {
3226 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3227 				np->n_attrstamp = 0;
3228 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3229 				ret = VOP_GETATTR(vp, &va, cred);
3230 				if (!ret) {
3231 					np->n_mtime = va.va_mtime;
3232 					np->n_change = va.va_filerev;
3233 				}
3234 			}
3235 			/* Mark that a file lock has been acquired. */
3236 			NFSLOCKNODE(np);
3237 			np->n_flag |= NHASBEENLOCKED;
3238 			NFSUNLOCKNODE(np);
3239 		}
3240 	} else if (!NFS_ISV4(vp)) {
3241 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3242 			size = VTONFS(vp)->n_size;
3243 			NFSVOPUNLOCK(vp);
3244 			error = lf_advlock(ap, &(vp->v_lockf), size);
3245 		} else {
3246 			if (nfs_advlock_p != NULL)
3247 				error = nfs_advlock_p(ap);
3248 			else {
3249 				NFSVOPUNLOCK(vp);
3250 				error = ENOLCK;
3251 			}
3252 		}
3253 		if (error == 0 && ap->a_op == F_SETLK) {
3254 			error = NFSVOPLOCK(vp, LK_SHARED);
3255 			if (error == 0) {
3256 				/* Mark that a file lock has been acquired. */
3257 				NFSLOCKNODE(np);
3258 				np->n_flag |= NHASBEENLOCKED;
3259 				NFSUNLOCKNODE(np);
3260 				NFSVOPUNLOCK(vp);
3261 			}
3262 		}
3263 		return (error);
3264 	} else
3265 		error = EOPNOTSUPP;
3266 out:
3267 	NFSVOPUNLOCK(vp);
3268 	return (error);
3269 }
3270 
3271 /*
3272  * NFS advisory byte-level locks.
3273  */
3274 static int
3275 nfs_advlockasync(struct vop_advlockasync_args *ap)
3276 {
3277 	struct vnode *vp = ap->a_vp;
3278 	u_quad_t size;
3279 	int error;
3280 
3281 	if (NFS_ISV4(vp))
3282 		return (EOPNOTSUPP);
3283 	error = NFSVOPLOCK(vp, LK_SHARED);
3284 	if (error)
3285 		return (error);
3286 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3287 		size = VTONFS(vp)->n_size;
3288 		NFSVOPUNLOCK(vp);
3289 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
3290 	} else {
3291 		NFSVOPUNLOCK(vp);
3292 		error = EOPNOTSUPP;
3293 	}
3294 	return (error);
3295 }
3296 
3297 /*
3298  * Print out the contents of an nfsnode.
3299  */
3300 static int
3301 nfs_print(struct vop_print_args *ap)
3302 {
3303 	struct vnode *vp = ap->a_vp;
3304 	struct nfsnode *np = VTONFS(vp);
3305 
3306 	printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3307 	    (uintmax_t)np->n_vattr.na_fsid);
3308 	if (vp->v_type == VFIFO)
3309 		fifo_printinfo(vp);
3310 	printf("\n");
3311 	return (0);
3312 }
3313 
3314 /*
3315  * This is the "real" nfs::bwrite(struct buf*).
3316  * We set B_CACHE if this is a VMIO buffer.
3317  */
3318 int
3319 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3320 {
3321 	int oldflags, rtval;
3322 
3323 	if (bp->b_flags & B_INVAL) {
3324 		brelse(bp);
3325 		return (0);
3326 	}
3327 
3328 	oldflags = bp->b_flags;
3329 	bp->b_flags |= B_CACHE;
3330 
3331 	/*
3332 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3333 	 */
3334 	bundirty(bp);
3335 	bp->b_flags &= ~B_DONE;
3336 	bp->b_ioflags &= ~BIO_ERROR;
3337 	bp->b_iocmd = BIO_WRITE;
3338 
3339 	bufobj_wref(bp->b_bufobj);
3340 	curthread->td_ru.ru_oublock++;
3341 
3342 	/*
3343 	 * Note: to avoid loopback deadlocks, we do not
3344 	 * assign b_runningbufspace.
3345 	 */
3346 	vfs_busy_pages(bp, 1);
3347 
3348 	BUF_KERNPROC(bp);
3349 	bp->b_iooffset = dbtob(bp->b_blkno);
3350 	bstrategy(bp);
3351 
3352 	if ((oldflags & B_ASYNC) != 0)
3353 		return (0);
3354 
3355 	rtval = bufwait(bp);
3356 	if (oldflags & B_DELWRI)
3357 		reassignbuf(bp);
3358 	brelse(bp);
3359 	return (rtval);
3360 }
3361 
3362 /*
3363  * nfs special file access vnode op.
3364  * Essentially just get vattr and then imitate iaccess() since the device is
3365  * local to the client.
3366  */
3367 static int
3368 nfsspec_access(struct vop_access_args *ap)
3369 {
3370 	struct vattr *vap;
3371 	struct ucred *cred = ap->a_cred;
3372 	struct vnode *vp = ap->a_vp;
3373 	accmode_t accmode = ap->a_accmode;
3374 	struct vattr vattr;
3375 	int error;
3376 
3377 	/*
3378 	 * Disallow write attempts on filesystems mounted read-only;
3379 	 * unless the file is a socket, fifo, or a block or character
3380 	 * device resident on the filesystem.
3381 	 */
3382 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3383 		switch (vp->v_type) {
3384 		case VREG:
3385 		case VDIR:
3386 		case VLNK:
3387 			return (EROFS);
3388 		default:
3389 			break;
3390 		}
3391 	}
3392 	vap = &vattr;
3393 	error = VOP_GETATTR(vp, vap, cred);
3394 	if (error)
3395 		goto out;
3396 	error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3397 	    accmode, cred, NULL);
3398 out:
3399 	return error;
3400 }
3401 
3402 /*
3403  * Read wrapper for fifos.
3404  */
3405 static int
3406 nfsfifo_read(struct vop_read_args *ap)
3407 {
3408 	struct nfsnode *np = VTONFS(ap->a_vp);
3409 	int error;
3410 
3411 	/*
3412 	 * Set access flag.
3413 	 */
3414 	NFSLOCKNODE(np);
3415 	np->n_flag |= NACC;
3416 	vfs_timestamp(&np->n_atim);
3417 	NFSUNLOCKNODE(np);
3418 	error = fifo_specops.vop_read(ap);
3419 	return error;
3420 }
3421 
3422 /*
3423  * Write wrapper for fifos.
3424  */
3425 static int
3426 nfsfifo_write(struct vop_write_args *ap)
3427 {
3428 	struct nfsnode *np = VTONFS(ap->a_vp);
3429 
3430 	/*
3431 	 * Set update flag.
3432 	 */
3433 	NFSLOCKNODE(np);
3434 	np->n_flag |= NUPD;
3435 	vfs_timestamp(&np->n_mtim);
3436 	NFSUNLOCKNODE(np);
3437 	return(fifo_specops.vop_write(ap));
3438 }
3439 
3440 /*
3441  * Close wrapper for fifos.
3442  *
3443  * Update the times on the nfsnode then do fifo close.
3444  */
3445 static int
3446 nfsfifo_close(struct vop_close_args *ap)
3447 {
3448 	struct vnode *vp = ap->a_vp;
3449 	struct nfsnode *np = VTONFS(vp);
3450 	struct vattr vattr;
3451 	struct timespec ts;
3452 
3453 	NFSLOCKNODE(np);
3454 	if (np->n_flag & (NACC | NUPD)) {
3455 		vfs_timestamp(&ts);
3456 		if (np->n_flag & NACC)
3457 			np->n_atim = ts;
3458 		if (np->n_flag & NUPD)
3459 			np->n_mtim = ts;
3460 		np->n_flag |= NCHG;
3461 		if (vrefcnt(vp) == 1 &&
3462 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3463 			VATTR_NULL(&vattr);
3464 			if (np->n_flag & NACC)
3465 				vattr.va_atime = np->n_atim;
3466 			if (np->n_flag & NUPD)
3467 				vattr.va_mtime = np->n_mtim;
3468 			NFSUNLOCKNODE(np);
3469 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3470 			goto out;
3471 		}
3472 	}
3473 	NFSUNLOCKNODE(np);
3474 out:
3475 	return (fifo_specops.vop_close(ap));
3476 }
3477 
3478 /*
3479  * Just call ncl_writebp() with the force argument set to 1.
3480  *
3481  * NOTE: B_DONE may or may not be set in a_bp on call.
3482  */
3483 static int
3484 nfs_bwrite(struct buf *bp)
3485 {
3486 
3487 	return (ncl_writebp(bp, 1, curthread));
3488 }
3489 
3490 struct buf_ops buf_ops_newnfs = {
3491 	.bop_name	=	"buf_ops_nfs",
3492 	.bop_write	=	nfs_bwrite,
3493 	.bop_strategy	=	bufstrategy,
3494 	.bop_sync	=	bufsync,
3495 	.bop_bdflush	=	bufbdflush,
3496 };
3497 
3498 static int
3499 nfs_getacl(struct vop_getacl_args *ap)
3500 {
3501 	int error;
3502 
3503 	if (ap->a_type != ACL_TYPE_NFS4)
3504 		return (EOPNOTSUPP);
3505 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3506 	    NULL);
3507 	if (error > NFSERR_STALE) {
3508 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3509 		error = EPERM;
3510 	}
3511 	return (error);
3512 }
3513 
3514 static int
3515 nfs_setacl(struct vop_setacl_args *ap)
3516 {
3517 	int error;
3518 
3519 	if (ap->a_type != ACL_TYPE_NFS4)
3520 		return (EOPNOTSUPP);
3521 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3522 	    NULL);
3523 	if (error > NFSERR_STALE) {
3524 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3525 		error = EPERM;
3526 	}
3527 	return (error);
3528 }
3529 
3530 /*
3531  * VOP_ADVISE for NFS.
3532  * Just return 0 for any errors, since it is just a hint.
3533  */
3534 static int
3535 nfs_advise(struct vop_advise_args *ap)
3536 {
3537 	struct thread *td = curthread;
3538 	struct nfsmount *nmp;
3539 	uint64_t len;
3540 	int error;
3541 
3542 	/*
3543 	 * First do vop_stdadvise() to handle the buffer cache.
3544 	 */
3545 	error = vop_stdadvise(ap);
3546 	if (error != 0)
3547 		return (error);
3548 	if (ap->a_start < 0 || ap->a_end < 0)
3549 		return (0);
3550 	if (ap->a_end == OFF_MAX)
3551 		len = 0;
3552 	else if (ap->a_end < ap->a_start)
3553 		return (0);
3554 	else
3555 		len = ap->a_end - ap->a_start + 1;
3556 	nmp = VFSTONFS(ap->a_vp->v_mount);
3557 	mtx_lock(&nmp->nm_mtx);
3558 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3559 	    (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) ==
3560 	    0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) {
3561 		mtx_unlock(&nmp->nm_mtx);
3562 		return (0);
3563 	}
3564 	mtx_unlock(&nmp->nm_mtx);
3565 	error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice,
3566 	    td->td_ucred, td);
3567 	if (error == NFSERR_NOTSUPP) {
3568 		mtx_lock(&nmp->nm_mtx);
3569 		nmp->nm_privflag |= NFSMNTP_NOADVISE;
3570 		mtx_unlock(&nmp->nm_mtx);
3571 	}
3572 	return (0);
3573 }
3574 
3575 /*
3576  * nfs allocate call
3577  */
3578 static int
3579 nfs_allocate(struct vop_allocate_args *ap)
3580 {
3581 	struct vnode *vp = ap->a_vp;
3582 	struct thread *td = curthread;
3583 	struct nfsvattr nfsva;
3584 	struct nfsmount *nmp;
3585 	int attrflag, error, ret;
3586 
3587 	attrflag = 0;
3588 	nmp = VFSTONFS(vp->v_mount);
3589 	mtx_lock(&nmp->nm_mtx);
3590 	if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3591 	    (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) {
3592 		mtx_unlock(&nmp->nm_mtx);
3593 		/*
3594 		 * Flush first to ensure that the allocate adds to the
3595 		 * file's allocation on the server.
3596 		 */
3597 		error = ncl_flush(vp, MNT_WAIT, td, 1, 0);
3598 		if (error == 0)
3599 			error = nfsrpc_allocate(vp, *ap->a_offset, *ap->a_len,
3600 			    &nfsva, &attrflag, td->td_ucred, td, NULL);
3601 		if (error == 0) {
3602 			*ap->a_offset += *ap->a_len;
3603 			*ap->a_len = 0;
3604 		} else if (error == NFSERR_NOTSUPP) {
3605 			mtx_lock(&nmp->nm_mtx);
3606 			nmp->nm_privflag |= NFSMNTP_NOALLOCATE;
3607 			mtx_unlock(&nmp->nm_mtx);
3608 		}
3609 	} else {
3610 		mtx_unlock(&nmp->nm_mtx);
3611 		error = EIO;
3612 	}
3613 	/*
3614 	 * If the NFS server cannot perform the Allocate operation, just call
3615 	 * vop_stdallocate() to perform it.
3616 	 */
3617 	if (error != 0)
3618 		error = vop_stdallocate(ap);
3619 	if (attrflag != 0) {
3620 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3621 		if (error == 0 && ret != 0)
3622 			error = ret;
3623 	}
3624 	if (error != 0)
3625 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3626 	return (error);
3627 }
3628 
3629 /*
3630  * nfs copy_file_range call
3631  */
3632 static int
3633 nfs_copy_file_range(struct vop_copy_file_range_args *ap)
3634 {
3635 	struct vnode *invp = ap->a_invp;
3636 	struct vnode *outvp = ap->a_outvp;
3637 	struct mount *mp;
3638 	struct nfsvattr innfsva, outnfsva;
3639 	struct vattr *vap;
3640 	struct uio io;
3641 	struct nfsmount *nmp;
3642 	size_t len, len2, copiedlen;
3643 	int error, inattrflag, outattrflag, ret, ret2;
3644 	off_t inoff, outoff;
3645 	bool consecutive, must_commit, tryoutcred;
3646 
3647 	ret = ret2 = 0;
3648 	nmp = VFSTONFS(invp->v_mount);
3649 	mtx_lock(&nmp->nm_mtx);
3650 	/* NFSv4.2 Copy is not permitted for infile == outfile. */
3651 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3652 	    (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0 || invp == outvp) {
3653 		mtx_unlock(&nmp->nm_mtx);
3654 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3655 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3656 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3657 		return (error);
3658 	}
3659 	mtx_unlock(&nmp->nm_mtx);
3660 
3661 	/* Lock both vnodes, avoiding risk of deadlock. */
3662 	do {
3663 		mp = NULL;
3664 		error = vn_start_write(outvp, &mp, V_WAIT);
3665 		if (error == 0) {
3666 			error = vn_lock(outvp, LK_EXCLUSIVE);
3667 			if (error == 0) {
3668 				error = vn_lock(invp, LK_SHARED | LK_NOWAIT);
3669 				if (error == 0)
3670 					break;
3671 				VOP_UNLOCK(outvp);
3672 				if (mp != NULL)
3673 					vn_finished_write(mp);
3674 				mp = NULL;
3675 				error = vn_lock(invp, LK_SHARED);
3676 				if (error == 0)
3677 					VOP_UNLOCK(invp);
3678 			}
3679 		}
3680 		if (mp != NULL)
3681 			vn_finished_write(mp);
3682 	} while (error == 0);
3683 	if (error != 0)
3684 		return (error);
3685 
3686 	/*
3687 	 * Do the vn_rlimit_fsize() check.  Should this be above the VOP layer?
3688 	 */
3689 	io.uio_offset = *ap->a_outoffp;
3690 	io.uio_resid = *ap->a_lenp;
3691 	error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd);
3692 
3693 	/*
3694 	 * Flush the input file so that the data is up to date before
3695 	 * the copy.  Flush writes for the output file so that they
3696 	 * do not overwrite the data copied to the output file by the Copy.
3697 	 * Set the commit argument for both flushes so that the data is on
3698 	 * stable storage before the Copy RPC.  This is done in case the
3699 	 * server reboots during the Copy and needs to be redone.
3700 	 */
3701 	if (error == 0)
3702 		error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0);
3703 	if (error == 0)
3704 		error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0);
3705 
3706 	/* Do the actual NFSv4.2 RPC. */
3707 	len = *ap->a_lenp;
3708 	mtx_lock(&nmp->nm_mtx);
3709 	if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0)
3710 		consecutive = true;
3711 	else
3712 		consecutive = false;
3713 	mtx_unlock(&nmp->nm_mtx);
3714 	inoff = *ap->a_inoffp;
3715 	outoff = *ap->a_outoffp;
3716 	tryoutcred = true;
3717 	must_commit = false;
3718 	if (error == 0) {
3719 		vap = &VTONFS(invp)->n_vattr.na_vattr;
3720 		error = VOP_GETATTR(invp, vap, ap->a_incred);
3721 		if (error == 0) {
3722 			/*
3723 			 * Clip "len" at va_size so that RFC compliant servers
3724 			 * will not reply NFSERR_INVAL.
3725 			 * Setting "len == 0" for the RPC would be preferred,
3726 			 * but some Linux servers do not support that.
3727 			 */
3728 			if (inoff >= vap->va_size)
3729 				*ap->a_lenp = len = 0;
3730 			else if (inoff + len > vap->va_size)
3731 				*ap->a_lenp = len = vap->va_size - inoff;
3732 		} else
3733 			error = 0;
3734 	}
3735 	copiedlen = 0;
3736 	while (len > 0 && error == 0) {
3737 		inattrflag = outattrflag = 0;
3738 		len2 = len;
3739 		if (tryoutcred)
3740 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3741 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3742 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3743 			    ap->a_outcred, consecutive, &must_commit);
3744 		else
3745 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3746 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3747 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3748 			    ap->a_incred, consecutive, &must_commit);
3749 		if (inattrflag != 0)
3750 			ret = nfscl_loadattrcache(&invp, &innfsva, NULL, NULL,
3751 			    0, 1);
3752 		if (outattrflag != 0)
3753 			ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL,
3754 			    NULL, 1, 1);
3755 		if (error == 0) {
3756 			if (consecutive == false) {
3757 				if (len2 == len) {
3758 					mtx_lock(&nmp->nm_mtx);
3759 					nmp->nm_privflag |=
3760 					    NFSMNTP_NOCONSECUTIVE;
3761 					mtx_unlock(&nmp->nm_mtx);
3762 				} else
3763 					error = NFSERR_OFFLOADNOREQS;
3764 			}
3765 			/*
3766 			 * If the Copy returns a length == 0, it hit the
3767 			 * EOF on the input file.
3768 			 */
3769 			if (len2 == 0) {
3770 				*ap->a_lenp = copiedlen;
3771 				len = 0;
3772 			} else {
3773 				len -= len2;
3774 				copiedlen += len2;
3775 			}
3776 			if (len == 0 && must_commit && error == 0)
3777 				error = ncl_commit(outvp, outoff, *ap->a_lenp,
3778 				    ap->a_outcred, curthread);
3779 			if (error == 0 && ret != 0)
3780 				error = ret;
3781 			if (error == 0 && ret2 != 0)
3782 				error = ret2;
3783 		} else if (error == NFSERR_OFFLOADNOREQS && consecutive) {
3784 			/*
3785 			 * Try consecutive == false, which is ok only if all
3786 			 * bytes are copied.
3787 			 */
3788 			consecutive = false;
3789 			error = 0;
3790 		} else if (error == NFSERR_ACCES && tryoutcred) {
3791 			/* Try again with incred. */
3792 			tryoutcred = false;
3793 			error = 0;
3794 		}
3795 		if (error == NFSERR_STALEWRITEVERF) {
3796 			/*
3797 			 * Server rebooted, so do it all again.
3798 			 */
3799 			*ap->a_inoffp = inoff;
3800 			*ap->a_outoffp = outoff;
3801 			len = *ap->a_lenp;
3802 			must_commit = false;
3803 			error = 0;
3804 		}
3805 	}
3806 	VOP_UNLOCK(invp);
3807 	VOP_UNLOCK(outvp);
3808 	if (mp != NULL)
3809 		vn_finished_write(mp);
3810 	if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS ||
3811 	    error == NFSERR_ACCES) {
3812 		/*
3813 		 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can
3814 		 * use a_incred for the read and a_outcred for the write, so
3815 		 * try this for NFSERR_ACCES failures for the Copy.
3816 		 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can
3817 		 * never succeed, so disable it.
3818 		 */
3819 		if (error != NFSERR_ACCES) {
3820 			/* Can never do Copy on this mount. */
3821 			mtx_lock(&nmp->nm_mtx);
3822 			nmp->nm_privflag |= NFSMNTP_NOCOPY;
3823 			mtx_unlock(&nmp->nm_mtx);
3824 		}
3825 		*ap->a_inoffp = inoff;
3826 		*ap->a_outoffp = outoff;
3827 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3828 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3829 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3830 	} else if (error != 0)
3831 		*ap->a_lenp = 0;
3832 
3833 	if (error != 0)
3834 		error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0);
3835 	return (error);
3836 }
3837 
3838 /*
3839  * nfs ioctl call
3840  */
3841 static int
3842 nfs_ioctl(struct vop_ioctl_args *ap)
3843 {
3844 	struct vnode *vp = ap->a_vp;
3845 	struct nfsvattr nfsva;
3846 	struct nfsmount *nmp;
3847 	int attrflag, content, error, ret;
3848 	bool eof = false;			/* shut up compiler. */
3849 
3850 	if (vp->v_type != VREG)
3851 		return (ENOTTY);
3852 	nmp = VFSTONFS(vp->v_mount);
3853 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) {
3854 		error = vop_stdioctl(ap);
3855 		return (error);
3856 	}
3857 
3858 	/* Do the actual NFSv4.2 RPC. */
3859 	switch (ap->a_command) {
3860 	case FIOSEEKDATA:
3861 		content = NFSV4CONTENT_DATA;
3862 		break;
3863 	case FIOSEEKHOLE:
3864 		content = NFSV4CONTENT_HOLE;
3865 		break;
3866 	default:
3867 		return (ENOTTY);
3868 	}
3869 
3870 	error = vn_lock(vp, LK_SHARED);
3871 	if (error != 0)
3872 		return (EBADF);
3873 	attrflag = 0;
3874 	if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size)
3875 		error = ENXIO;
3876 	else {
3877 		/*
3878 		 * Flush all writes, so that the server is up to date.
3879 		 * Although a Commit is not required, the commit argument
3880 		 * is set so that, for a pNFS File/Flexible File Layout
3881 		 * server, the LayoutCommit will be done to ensure the file
3882 		 * size is up to date on the Metadata Server.
3883 		 */
3884 		error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0);
3885 		if (error == 0)
3886 			error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof,
3887 			    content, ap->a_cred, &nfsva, &attrflag);
3888 		/* If at eof for FIOSEEKDATA, return ENXIO. */
3889 		if (eof && error == 0 && content == NFSV4CONTENT_DATA)
3890 			error = ENXIO;
3891 	}
3892 	if (attrflag != 0) {
3893 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3894 		if (error == 0 && ret != 0)
3895 			error = ret;
3896 	}
3897 	NFSVOPUNLOCK(vp);
3898 
3899 	if (error != 0)
3900 		error = ENXIO;
3901 	return (error);
3902 }
3903 
3904 /*
3905  * nfs getextattr call
3906  */
3907 static int
3908 nfs_getextattr(struct vop_getextattr_args *ap)
3909 {
3910 	struct vnode *vp = ap->a_vp;
3911 	struct nfsmount *nmp;
3912 	struct ucred *cred;
3913 	struct thread *td = ap->a_td;
3914 	struct nfsvattr nfsva;
3915 	ssize_t len;
3916 	int attrflag, error, ret;
3917 
3918 	nmp = VFSTONFS(vp->v_mount);
3919 	mtx_lock(&nmp->nm_mtx);
3920 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3921 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
3922 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
3923 		mtx_unlock(&nmp->nm_mtx);
3924 		return (EOPNOTSUPP);
3925 	}
3926 	mtx_unlock(&nmp->nm_mtx);
3927 
3928 	cred = ap->a_cred;
3929 	if (cred == NULL)
3930 		cred = td->td_ucred;
3931 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
3932 	attrflag = 0;
3933 	error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva,
3934 	    &attrflag, cred, td);
3935 	if (attrflag != 0) {
3936 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3937 		if (error == 0 && ret != 0)
3938 			error = ret;
3939 	}
3940 	if (error == 0 && ap->a_size != NULL)
3941 		*ap->a_size = len;
3942 
3943 	switch (error) {
3944 	case NFSERR_NOTSUPP:
3945 	case NFSERR_OPILLEGAL:
3946 		mtx_lock(&nmp->nm_mtx);
3947 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
3948 		mtx_unlock(&nmp->nm_mtx);
3949 		error = EOPNOTSUPP;
3950 		break;
3951 	case NFSERR_NOXATTR:
3952 	case NFSERR_XATTR2BIG:
3953 		error = ENOATTR;
3954 		break;
3955 	default:
3956 		error = nfscl_maperr(td, error, 0, 0);
3957 		break;
3958 	}
3959 	return (error);
3960 }
3961 
3962 /*
3963  * nfs setextattr call
3964  */
3965 static int
3966 nfs_setextattr(struct vop_setextattr_args *ap)
3967 {
3968 	struct vnode *vp = ap->a_vp;
3969 	struct nfsmount *nmp;
3970 	struct ucred *cred;
3971 	struct thread *td = ap->a_td;
3972 	struct nfsvattr nfsva;
3973 	int attrflag, error, ret;
3974 
3975 	nmp = VFSTONFS(vp->v_mount);
3976 	mtx_lock(&nmp->nm_mtx);
3977 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3978 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
3979 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
3980 		mtx_unlock(&nmp->nm_mtx);
3981 		return (EOPNOTSUPP);
3982 	}
3983 	mtx_unlock(&nmp->nm_mtx);
3984 
3985 	if (ap->a_uio->uio_resid < 0)
3986 		return (EINVAL);
3987 	cred = ap->a_cred;
3988 	if (cred == NULL)
3989 		cred = td->td_ucred;
3990 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
3991 	attrflag = 0;
3992 	error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva,
3993 	    &attrflag, cred, td);
3994 	if (attrflag != 0) {
3995 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3996 		if (error == 0 && ret != 0)
3997 			error = ret;
3998 	}
3999 
4000 	switch (error) {
4001 	case NFSERR_NOTSUPP:
4002 	case NFSERR_OPILLEGAL:
4003 		mtx_lock(&nmp->nm_mtx);
4004 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4005 		mtx_unlock(&nmp->nm_mtx);
4006 		error = EOPNOTSUPP;
4007 		break;
4008 	case NFSERR_NOXATTR:
4009 	case NFSERR_XATTR2BIG:
4010 		error = ENOATTR;
4011 		break;
4012 	default:
4013 		error = nfscl_maperr(td, error, 0, 0);
4014 		break;
4015 	}
4016 	return (error);
4017 }
4018 
4019 /*
4020  * nfs listextattr call
4021  */
4022 static int
4023 nfs_listextattr(struct vop_listextattr_args *ap)
4024 {
4025 	struct vnode *vp = ap->a_vp;
4026 	struct nfsmount *nmp;
4027 	struct ucred *cred;
4028 	struct thread *td = ap->a_td;
4029 	struct nfsvattr nfsva;
4030 	size_t len, len2;
4031 	uint64_t cookie;
4032 	int attrflag, error, ret;
4033 	bool eof;
4034 
4035 	nmp = VFSTONFS(vp->v_mount);
4036 	mtx_lock(&nmp->nm_mtx);
4037 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4038 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4039 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4040 		mtx_unlock(&nmp->nm_mtx);
4041 		return (EOPNOTSUPP);
4042 	}
4043 	mtx_unlock(&nmp->nm_mtx);
4044 
4045 	cred = ap->a_cred;
4046 	if (cred == NULL)
4047 		cred = td->td_ucred;
4048 
4049 	/* Loop around doing List Extended Attribute RPCs. */
4050 	eof = false;
4051 	cookie = 0;
4052 	len2 = 0;
4053 	error = 0;
4054 	while (!eof && error == 0) {
4055 		len = nmp->nm_rsize;
4056 		attrflag = 0;
4057 		error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof,
4058 		    &nfsva, &attrflag, cred, td);
4059 		if (attrflag != 0) {
4060 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4061 			    1);
4062 			if (error == 0 && ret != 0)
4063 				error = ret;
4064 		}
4065 		if (error == 0) {
4066 			len2 += len;
4067 			if (len2 > SSIZE_MAX)
4068 				error = ENOATTR;
4069 		}
4070 	}
4071 	if (error == 0 && ap->a_size != NULL)
4072 		*ap->a_size = len2;
4073 
4074 	switch (error) {
4075 	case NFSERR_NOTSUPP:
4076 	case NFSERR_OPILLEGAL:
4077 		mtx_lock(&nmp->nm_mtx);
4078 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4079 		mtx_unlock(&nmp->nm_mtx);
4080 		error = EOPNOTSUPP;
4081 		break;
4082 	case NFSERR_NOXATTR:
4083 	case NFSERR_XATTR2BIG:
4084 		error = ENOATTR;
4085 		break;
4086 	default:
4087 		error = nfscl_maperr(td, error, 0, 0);
4088 		break;
4089 	}
4090 	return (error);
4091 }
4092 
4093 /*
4094  * nfs setextattr call
4095  */
4096 static int
4097 nfs_deleteextattr(struct vop_deleteextattr_args *ap)
4098 {
4099 	struct vnode *vp = ap->a_vp;
4100 	struct nfsmount *nmp;
4101 	struct nfsvattr nfsva;
4102 	int attrflag, error, ret;
4103 
4104 	nmp = VFSTONFS(vp->v_mount);
4105 	mtx_lock(&nmp->nm_mtx);
4106 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4107 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4108 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4109 		mtx_unlock(&nmp->nm_mtx);
4110 		return (EOPNOTSUPP);
4111 	}
4112 	mtx_unlock(&nmp->nm_mtx);
4113 
4114 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4115 	attrflag = 0;
4116 	error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred,
4117 	    ap->a_td);
4118 	if (attrflag != 0) {
4119 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4120 		if (error == 0 && ret != 0)
4121 			error = ret;
4122 	}
4123 
4124 	switch (error) {
4125 	case NFSERR_NOTSUPP:
4126 	case NFSERR_OPILLEGAL:
4127 		mtx_lock(&nmp->nm_mtx);
4128 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4129 		mtx_unlock(&nmp->nm_mtx);
4130 		error = EOPNOTSUPP;
4131 		break;
4132 	case NFSERR_NOXATTR:
4133 	case NFSERR_XATTR2BIG:
4134 		error = ENOATTR;
4135 		break;
4136 	default:
4137 		error = nfscl_maperr(ap->a_td, error, 0, 0);
4138 		break;
4139 	}
4140 	return (error);
4141 }
4142 
4143 /*
4144  * Return POSIX pathconf information applicable to nfs filesystems.
4145  */
4146 static int
4147 nfs_pathconf(struct vop_pathconf_args *ap)
4148 {
4149 	struct nfsv3_pathconf pc;
4150 	struct nfsvattr nfsva;
4151 	struct vnode *vp = ap->a_vp;
4152 	struct nfsmount *nmp;
4153 	struct thread *td = curthread;
4154 	off_t off;
4155 	bool eof;
4156 	int attrflag, error;
4157 
4158 	if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
4159 	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
4160 	    ap->a_name == _PC_NO_TRUNC)) ||
4161 	    (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
4162 		/*
4163 		 * Since only the above 4 a_names are returned by the NFSv3
4164 		 * Pathconf RPC, there is no point in doing it for others.
4165 		 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
4166 		 * be used for _PC_NFS4_ACL as well.
4167 		 */
4168 		error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
4169 		    &attrflag, NULL);
4170 		if (attrflag != 0)
4171 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4172 			    1);
4173 		if (error != 0)
4174 			return (error);
4175 	} else {
4176 		/*
4177 		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
4178 		 * just fake them.
4179 		 */
4180 		pc.pc_linkmax = NFS_LINK_MAX;
4181 		pc.pc_namemax = NFS_MAXNAMLEN;
4182 		pc.pc_notrunc = 1;
4183 		pc.pc_chownrestricted = 1;
4184 		pc.pc_caseinsensitive = 0;
4185 		pc.pc_casepreserving = 1;
4186 		error = 0;
4187 	}
4188 	switch (ap->a_name) {
4189 	case _PC_LINK_MAX:
4190 #ifdef _LP64
4191 		*ap->a_retval = pc.pc_linkmax;
4192 #else
4193 		*ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
4194 #endif
4195 		break;
4196 	case _PC_NAME_MAX:
4197 		*ap->a_retval = pc.pc_namemax;
4198 		break;
4199 	case _PC_PIPE_BUF:
4200 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
4201 			*ap->a_retval = PIPE_BUF;
4202 		else
4203 			error = EINVAL;
4204 		break;
4205 	case _PC_CHOWN_RESTRICTED:
4206 		*ap->a_retval = pc.pc_chownrestricted;
4207 		break;
4208 	case _PC_NO_TRUNC:
4209 		*ap->a_retval = pc.pc_notrunc;
4210 		break;
4211 	case _PC_ACL_NFS4:
4212 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
4213 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
4214 			*ap->a_retval = 1;
4215 		else
4216 			*ap->a_retval = 0;
4217 		break;
4218 	case _PC_ACL_PATH_MAX:
4219 		if (NFS_ISV4(vp))
4220 			*ap->a_retval = ACL_MAX_ENTRIES;
4221 		else
4222 			*ap->a_retval = 3;
4223 		break;
4224 	case _PC_PRIO_IO:
4225 		*ap->a_retval = 0;
4226 		break;
4227 	case _PC_SYNC_IO:
4228 		*ap->a_retval = 0;
4229 		break;
4230 	case _PC_ALLOC_SIZE_MIN:
4231 		*ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
4232 		break;
4233 	case _PC_FILESIZEBITS:
4234 		if (NFS_ISV34(vp))
4235 			*ap->a_retval = 64;
4236 		else
4237 			*ap->a_retval = 32;
4238 		break;
4239 	case _PC_REC_INCR_XFER_SIZE:
4240 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4241 		break;
4242 	case _PC_REC_MAX_XFER_SIZE:
4243 		*ap->a_retval = -1; /* means ``unlimited'' */
4244 		break;
4245 	case _PC_REC_MIN_XFER_SIZE:
4246 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4247 		break;
4248 	case _PC_REC_XFER_ALIGN:
4249 		*ap->a_retval = PAGE_SIZE;
4250 		break;
4251 	case _PC_SYMLINK_MAX:
4252 		*ap->a_retval = NFS_MAXPATHLEN;
4253 		break;
4254 	case _PC_MIN_HOLE_SIZE:
4255 		/* Only some NFSv4.2 servers support Seek for Holes. */
4256 		*ap->a_retval = 0;
4257 		nmp = VFSTONFS(vp->v_mount);
4258 		if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) {
4259 			/*
4260 			 * NFSv4.2 doesn't have an attribute for hole size,
4261 			 * so all we can do is see if the Seek operation is
4262 			 * supported and then use f_iosize as a "best guess".
4263 			 */
4264 			mtx_lock(&nmp->nm_mtx);
4265 			if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) {
4266 				mtx_unlock(&nmp->nm_mtx);
4267 				off = 0;
4268 				attrflag = 0;
4269 				error = nfsrpc_seek(vp, &off, &eof,
4270 				    NFSV4CONTENT_HOLE, td->td_ucred, &nfsva,
4271 				    &attrflag);
4272 				if (attrflag != 0)
4273 					nfscl_loadattrcache(&vp, &nfsva,
4274 					    NULL, NULL, 0, 1);
4275 				mtx_lock(&nmp->nm_mtx);
4276 				if (error == NFSERR_NOTSUPP)
4277 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED;
4278 				else
4279 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED |
4280 					    NFSMNTP_SEEK;
4281 				error = 0;
4282 			}
4283 			if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0)
4284 				*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4285 			mtx_unlock(&nmp->nm_mtx);
4286 		}
4287 		break;
4288 
4289 	default:
4290 		error = vop_stdpathconf(ap);
4291 		break;
4292 	}
4293 	return (error);
4294 }
4295 
4296