xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision e9ac41698b2f322d55ccf9da50a3596edb2c1800)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	from nfs_vnops.c	8.16 (Berkeley) 5/27/95
35  */
36 
37 #include <sys/cdefs.h>
38 /*
39  * vnode op calls for Sun NFS version 2, 3 and 4
40  */
41 
42 #include "opt_inet.h"
43 
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/resourcevar.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/extattr.h>
53 #include <sys/filio.h>
54 #include <sys/jail.h>
55 #include <sys/malloc.h>
56 #include <sys/mbuf.h>
57 #include <sys/namei.h>
58 #include <sys/socket.h>
59 #include <sys/vnode.h>
60 #include <sys/dirent.h>
61 #include <sys/fcntl.h>
62 #include <sys/lockf.h>
63 #include <sys/stat.h>
64 #include <sys/sysctl.h>
65 #include <sys/signalvar.h>
66 
67 #include <vm/vm.h>
68 #include <vm/vm_extern.h>
69 #include <vm/vm_object.h>
70 #include <vm/vnode_pager.h>
71 
72 #include <fs/nfs/nfsport.h>
73 #include <fs/nfsclient/nfsnode.h>
74 #include <fs/nfsclient/nfsmount.h>
75 #include <fs/nfsclient/nfs.h>
76 #include <fs/nfsclient/nfs_kdtrace.h>
77 
78 #include <net/if.h>
79 #include <netinet/in.h>
80 #include <netinet/in_var.h>
81 
82 #include <nfs/nfs_lock.h>
83 
84 #ifdef KDTRACE_HOOKS
85 #include <sys/dtrace_bsd.h>
86 
87 dtrace_nfsclient_accesscache_flush_probe_func_t
88 		dtrace_nfscl_accesscache_flush_done_probe;
89 uint32_t	nfscl_accesscache_flush_done_id;
90 
91 dtrace_nfsclient_accesscache_get_probe_func_t
92 		dtrace_nfscl_accesscache_get_hit_probe,
93 		dtrace_nfscl_accesscache_get_miss_probe;
94 uint32_t	nfscl_accesscache_get_hit_id;
95 uint32_t	nfscl_accesscache_get_miss_id;
96 
97 dtrace_nfsclient_accesscache_load_probe_func_t
98 		dtrace_nfscl_accesscache_load_done_probe;
99 uint32_t	nfscl_accesscache_load_done_id;
100 #endif /* !KDTRACE_HOOKS */
101 
102 /* Defs */
103 #define	TRUE	1
104 #define	FALSE	0
105 
106 extern struct nfsstatsv1 nfsstatsv1;
107 extern int nfsrv_useacl;
108 extern int nfscl_debuglevel;
109 MALLOC_DECLARE(M_NEWNFSREQ);
110 
111 static vop_read_t	nfsfifo_read;
112 static vop_write_t	nfsfifo_write;
113 static vop_close_t	nfsfifo_close;
114 static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
115 		    struct thread *);
116 static vop_lookup_t	nfs_lookup;
117 static vop_create_t	nfs_create;
118 static vop_mknod_t	nfs_mknod;
119 static vop_open_t	nfs_open;
120 static vop_pathconf_t	nfs_pathconf;
121 static vop_close_t	nfs_close;
122 static vop_access_t	nfs_access;
123 static vop_getattr_t	nfs_getattr;
124 static vop_setattr_t	nfs_setattr;
125 static vop_read_t	nfs_read;
126 static vop_fsync_t	nfs_fsync;
127 static vop_remove_t	nfs_remove;
128 static vop_link_t	nfs_link;
129 static vop_rename_t	nfs_rename;
130 static vop_mkdir_t	nfs_mkdir;
131 static vop_rmdir_t	nfs_rmdir;
132 static vop_symlink_t	nfs_symlink;
133 static vop_readdir_t	nfs_readdir;
134 static vop_strategy_t	nfs_strategy;
135 static	int	nfs_lookitup(struct vnode *, char *, int,
136 		    struct ucred *, struct thread *, struct nfsnode **);
137 static	int	nfs_sillyrename(struct vnode *, struct vnode *,
138 		    struct componentname *);
139 static vop_access_t	nfsspec_access;
140 static vop_readlink_t	nfs_readlink;
141 static vop_print_t	nfs_print;
142 static vop_advlock_t	nfs_advlock;
143 static vop_advlockasync_t nfs_advlockasync;
144 static vop_getacl_t nfs_getacl;
145 static vop_setacl_t nfs_setacl;
146 static vop_advise_t nfs_advise;
147 static vop_allocate_t nfs_allocate;
148 static vop_deallocate_t nfs_deallocate;
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_deallocate =	nfs_deallocate,
197 	.vop_copy_file_range =	nfs_copy_file_range,
198 	.vop_ioctl =		nfs_ioctl,
199 	.vop_getextattr =	nfs_getextattr,
200 	.vop_setextattr =	nfs_setextattr,
201 	.vop_listextattr =	nfs_listextattr,
202 	.vop_deleteextattr =	nfs_deleteextattr,
203 };
204 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig);
205 
206 static int
207 nfs_vnodeops_bypass(struct vop_generic_args *a)
208 {
209 
210 	return (vop_sigdefer(&newnfs_vnodeops_nosig, a));
211 }
212 
213 struct vop_vector newnfs_vnodeops = {
214 	.vop_default =		&default_vnodeops,
215 	.vop_bypass =		nfs_vnodeops_bypass,
216 };
217 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops);
218 
219 static struct vop_vector newnfs_fifoops_nosig = {
220 	.vop_default =		&fifo_specops,
221 	.vop_access =		nfsspec_access,
222 	.vop_close =		nfsfifo_close,
223 	.vop_fsync =		nfs_fsync,
224 	.vop_getattr =		nfs_getattr,
225 	.vop_inactive =		ncl_inactive,
226 	.vop_pathconf =		nfs_pathconf,
227 	.vop_print =		nfs_print,
228 	.vop_read =		nfsfifo_read,
229 	.vop_reclaim =		ncl_reclaim,
230 	.vop_setattr =		nfs_setattr,
231 	.vop_write =		nfsfifo_write,
232 };
233 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig);
234 
235 static int
236 nfs_fifoops_bypass(struct vop_generic_args *a)
237 {
238 
239 	return (vop_sigdefer(&newnfs_fifoops_nosig, a));
240 }
241 
242 struct vop_vector newnfs_fifoops = {
243 	.vop_default =		&default_vnodeops,
244 	.vop_bypass =		nfs_fifoops_bypass,
245 };
246 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops);
247 
248 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
249     struct componentname *cnp, struct vattr *vap);
250 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
251     int namelen, struct ucred *cred, struct thread *td);
252 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
253     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
254     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
255 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
256     struct componentname *scnp, struct sillyrename *sp);
257 
258 /*
259  * Global variables
260  */
261 SYSCTL_DECL(_vfs_nfs);
262 
263 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
264 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
265 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
266 
267 static int	nfs_prime_access_cache = 0;
268 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
269 	   &nfs_prime_access_cache, 0,
270 	   "Prime NFS ACCESS cache when fetching attributes");
271 
272 static int	newnfs_commit_on_close = 0;
273 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
274     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
275 
276 static int	nfs_clean_pages_on_close = 1;
277 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
278 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
279 
280 int newnfs_directio_enable = 0;
281 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
282 	   &newnfs_directio_enable, 0, "Enable NFS directio");
283 
284 int nfs_keep_dirty_on_error;
285 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
286     &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
287 
288 /*
289  * This sysctl allows other processes to mmap a file that has been opened
290  * O_DIRECT by a process.  In general, having processes mmap the file while
291  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
292  * this by default to prevent DoS attacks - to prevent a malicious user from
293  * opening up files O_DIRECT preventing other users from mmap'ing these
294  * files.  "Protected" environments where stricter consistency guarantees are
295  * required can disable this knob.  The process that opened the file O_DIRECT
296  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
297  * meaningful.
298  */
299 int newnfs_directio_allow_mmap = 1;
300 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
301 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
302 
303 static uint64_t	nfs_maxalloclen = 64 * 1024 * 1024;
304 SYSCTL_U64(_vfs_nfs, OID_AUTO, maxalloclen, CTLFLAG_RW,
305 	   &nfs_maxalloclen, 0, "NFS max allocate/deallocate length");
306 
307 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
308 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
309 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
310 
311 /*
312  * SMP Locking Note :
313  * The list of locks after the description of the lock is the ordering
314  * of other locks acquired with the lock held.
315  * np->n_mtx : Protects the fields in the nfsnode.
316        VM Object Lock
317        VI_MTX (acquired indirectly)
318  * nmp->nm_mtx : Protects the fields in the nfsmount.
319        rep->r_mtx
320  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
321  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
322        nmp->nm_mtx
323        rep->r_mtx
324  * rep->r_mtx : Protects the fields in an nfsreq.
325  */
326 
327 static int
328 nfs_lock(struct vop_lock1_args *ap)
329 {
330 	struct vnode *vp;
331 	struct nfsnode *np;
332 	u_quad_t nsize;
333 	int error, lktype;
334 	bool onfault;
335 
336 	vp = ap->a_vp;
337 	lktype = ap->a_flags & LK_TYPE_MASK;
338 	error = VOP_LOCK1_APV(&default_vnodeops, ap);
339 	if (error != 0 || vp->v_op != &newnfs_vnodeops)
340 		return (error);
341 	np = VTONFS(vp);
342 	if (np == NULL)
343 		return (0);
344 	NFSLOCKNODE(np);
345 	if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED &&
346 	    lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE &&
347 	    lktype != LK_TRYUPGRADE)) {
348 		NFSUNLOCKNODE(np);
349 		return (0);
350 	}
351 	onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT &&
352 	    (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE &&
353 	    (lktype == LK_SHARED || lktype == LK_EXCLUSIVE);
354 	if (onfault && vp->v_vnlock->lk_recurse == 0) {
355 		/*
356 		 * Force retry in vm_fault(), to make the lock request
357 		 * sleepable, which allows us to piggy-back the
358 		 * sleepable call to vnode_pager_setsize().
359 		 */
360 		NFSUNLOCKNODE(np);
361 		VOP_UNLOCK(vp);
362 		return (EBUSY);
363 	}
364 	if ((ap->a_flags & LK_NOWAIT) != 0 ||
365 	    (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) {
366 		NFSUNLOCKNODE(np);
367 		return (0);
368 	}
369 	if (lktype == LK_SHARED) {
370 		NFSUNLOCKNODE(np);
371 		VOP_UNLOCK(vp);
372 		ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
373 		ap->a_flags |= LK_EXCLUSIVE;
374 		error = VOP_LOCK1_APV(&default_vnodeops, ap);
375 		if (error != 0 || vp->v_op != &newnfs_vnodeops)
376 			return (error);
377 		if (vp->v_data == NULL)
378 			goto downgrade;
379 		MPASS(vp->v_data == np);
380 		NFSLOCKNODE(np);
381 		if ((np->n_flag & NVNSETSZSKIP) == 0) {
382 			NFSUNLOCKNODE(np);
383 			goto downgrade;
384 		}
385 	}
386 	np->n_flag &= ~NVNSETSZSKIP;
387 	nsize = np->n_size;
388 	NFSUNLOCKNODE(np);
389 	vnode_pager_setsize(vp, nsize);
390 downgrade:
391 	if (lktype == LK_SHARED) {
392 		ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
393 		ap->a_flags |= LK_DOWNGRADE;
394 		(void)VOP_LOCK1_APV(&default_vnodeops, ap);
395 	}
396 	return (0);
397 }
398 
399 static int
400 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
401     struct ucred *cred, u_int32_t *retmode)
402 {
403 	int error = 0, attrflag, i, lrupos;
404 	u_int32_t rmode;
405 	struct nfsnode *np = VTONFS(vp);
406 	struct nfsvattr nfsva;
407 
408 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
409 	    &rmode);
410 	if (attrflag)
411 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
412 	if (!error) {
413 		lrupos = 0;
414 		NFSLOCKNODE(np);
415 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
416 			if (np->n_accesscache[i].uid == cred->cr_uid) {
417 				np->n_accesscache[i].mode = rmode;
418 				np->n_accesscache[i].stamp = time_second;
419 				break;
420 			}
421 			if (i > 0 && np->n_accesscache[i].stamp <
422 			    np->n_accesscache[lrupos].stamp)
423 				lrupos = i;
424 		}
425 		if (i == NFS_ACCESSCACHESIZE) {
426 			np->n_accesscache[lrupos].uid = cred->cr_uid;
427 			np->n_accesscache[lrupos].mode = rmode;
428 			np->n_accesscache[lrupos].stamp = time_second;
429 		}
430 		NFSUNLOCKNODE(np);
431 		if (retmode != NULL)
432 			*retmode = rmode;
433 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
434 	} else if (NFS_ISV4(vp)) {
435 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
436 	}
437 #ifdef KDTRACE_HOOKS
438 	if (error != 0)
439 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
440 		    error);
441 #endif
442 	return (error);
443 }
444 
445 /*
446  * nfs access vnode op.
447  * For nfs version 2, just return ok. File accesses may fail later.
448  * For nfs version 3, use the access rpc to check accessibility. If file modes
449  * are changed on the server, accesses might still fail later.
450  */
451 static int
452 nfs_access(struct vop_access_args *ap)
453 {
454 	struct vnode *vp = ap->a_vp;
455 	int error = 0, i, gotahit;
456 	u_int32_t mode, wmode, rmode;
457 	int v34 = NFS_ISV34(vp);
458 	struct nfsnode *np = VTONFS(vp);
459 
460 	/*
461 	 * Disallow write attempts on filesystems mounted read-only;
462 	 * unless the file is a socket, fifo, or a block or character
463 	 * device resident on the filesystem.
464 	 */
465 	if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
466 	    VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
467 	    VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
468 		switch (vp->v_type) {
469 		case VREG:
470 		case VDIR:
471 		case VLNK:
472 			return (EROFS);
473 		default:
474 			break;
475 		}
476 	}
477 	/*
478 	 * For nfs v3 or v4, check to see if we have done this recently, and if
479 	 * so return our cached result instead of making an ACCESS call.
480 	 * If not, do an access rpc, otherwise you are stuck emulating
481 	 * ufs_access() locally using the vattr. This may not be correct,
482 	 * since the server may apply other access criteria such as
483 	 * client uid-->server uid mapping that we do not know about.
484 	 */
485 	if (v34) {
486 		if (ap->a_accmode & VREAD)
487 			mode = NFSACCESS_READ;
488 		else
489 			mode = 0;
490 		if (vp->v_type != VDIR) {
491 			if (ap->a_accmode & VWRITE)
492 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
493 			if (ap->a_accmode & VAPPEND)
494 				mode |= NFSACCESS_EXTEND;
495 			if (ap->a_accmode & VEXEC)
496 				mode |= NFSACCESS_EXECUTE;
497 			if (ap->a_accmode & VDELETE)
498 				mode |= NFSACCESS_DELETE;
499 		} else {
500 			if (ap->a_accmode & VWRITE)
501 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
502 			if (ap->a_accmode & VAPPEND)
503 				mode |= NFSACCESS_EXTEND;
504 			if (ap->a_accmode & VEXEC)
505 				mode |= NFSACCESS_LOOKUP;
506 			if (ap->a_accmode & VDELETE)
507 				mode |= NFSACCESS_DELETE;
508 			if (ap->a_accmode & VDELETE_CHILD)
509 				mode |= NFSACCESS_MODIFY;
510 		}
511 		/* XXX safety belt, only make blanket request if caching */
512 		if (nfsaccess_cache_timeout > 0) {
513 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
514 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
515 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
516 		} else {
517 			wmode = mode;
518 		}
519 
520 		/*
521 		 * Does our cached result allow us to give a definite yes to
522 		 * this request?
523 		 */
524 		gotahit = 0;
525 		NFSLOCKNODE(np);
526 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
527 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
528 			    if (time_second < (np->n_accesscache[i].stamp
529 				+ nfsaccess_cache_timeout) &&
530 				(np->n_accesscache[i].mode & mode) == mode) {
531 				NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
532 				gotahit = 1;
533 			    }
534 			    break;
535 			}
536 		}
537 		NFSUNLOCKNODE(np);
538 #ifdef KDTRACE_HOOKS
539 		if (gotahit != 0)
540 			KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
541 			    ap->a_cred->cr_uid, mode);
542 		else
543 			KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
544 			    ap->a_cred->cr_uid, mode);
545 #endif
546 		if (gotahit == 0) {
547 			/*
548 			 * Either a no, or a don't know.  Go to the wire.
549 			 */
550 			NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
551 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
552 			    ap->a_cred, &rmode);
553 			if (!error &&
554 			    (rmode & mode) != mode)
555 				error = EACCES;
556 		}
557 		return (error);
558 	} else {
559 		if ((error = nfsspec_access(ap)) != 0) {
560 			return (error);
561 		}
562 		/*
563 		 * Attempt to prevent a mapped root from accessing a file
564 		 * which it shouldn't.  We try to read a byte from the file
565 		 * if the user is root and the file is not zero length.
566 		 * After calling nfsspec_access, we should have the correct
567 		 * file size cached.
568 		 */
569 		NFSLOCKNODE(np);
570 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
571 		    && VTONFS(vp)->n_size > 0) {
572 			struct iovec aiov;
573 			struct uio auio;
574 			char buf[1];
575 
576 			NFSUNLOCKNODE(np);
577 			aiov.iov_base = buf;
578 			aiov.iov_len = 1;
579 			auio.uio_iov = &aiov;
580 			auio.uio_iovcnt = 1;
581 			auio.uio_offset = 0;
582 			auio.uio_resid = 1;
583 			auio.uio_segflg = UIO_SYSSPACE;
584 			auio.uio_rw = UIO_READ;
585 			auio.uio_td = ap->a_td;
586 
587 			if (vp->v_type == VREG)
588 				error = ncl_readrpc(vp, &auio, ap->a_cred);
589 			else if (vp->v_type == VDIR) {
590 				char* bp;
591 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
592 				aiov.iov_base = bp;
593 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
594 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
595 				    ap->a_td);
596 				free(bp, M_TEMP);
597 			} else if (vp->v_type == VLNK)
598 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
599 			else
600 				error = EACCES;
601 		} else
602 			NFSUNLOCKNODE(np);
603 		return (error);
604 	}
605 }
606 
607 /*
608  * nfs open vnode op
609  * Check to see if the type is ok
610  * and that deletion is not in progress.
611  * For paged in text files, you will need to flush the page cache
612  * if consistency is lost.
613  */
614 /* ARGSUSED */
615 static int
616 nfs_open(struct vop_open_args *ap)
617 {
618 	struct vnode *vp = ap->a_vp;
619 	struct nfsnode *np = VTONFS(vp);
620 	struct vattr vattr;
621 	int error;
622 	int fmode = ap->a_mode;
623 	struct ucred *cred;
624 	vm_object_t obj;
625 
626 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
627 		return (EOPNOTSUPP);
628 
629 	/*
630 	 * For NFSv4, we need to do the Open Op before cache validation,
631 	 * so that we conform to RFC3530 Sec. 9.3.1.
632 	 */
633 	if (NFS_ISV4(vp)) {
634 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
635 		if (error) {
636 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
637 			    (gid_t)0);
638 			return (error);
639 		}
640 	}
641 
642 	/*
643 	 * Now, if this Open will be doing reading, re-validate/flush the
644 	 * cache, so that Close/Open coherency is maintained.
645 	 */
646 	NFSLOCKNODE(np);
647 	if (np->n_flag & NMODIFIED) {
648 		NFSUNLOCKNODE(np);
649 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
650 			NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
651 			if (VN_IS_DOOMED(vp))
652 				return (EBADF);
653 		}
654 		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
655 		if (error == EINTR || error == EIO) {
656 			if (NFS_ISV4(vp))
657 				(void) nfsrpc_close(vp, 0, ap->a_td);
658 			return (error);
659 		}
660 		NFSLOCKNODE(np);
661 		np->n_attrstamp = 0;
662 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
663 		if (vp->v_type == VDIR)
664 			np->n_direofoffset = 0;
665 		NFSUNLOCKNODE(np);
666 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
667 		if (error) {
668 			if (NFS_ISV4(vp))
669 				(void) nfsrpc_close(vp, 0, ap->a_td);
670 			return (error);
671 		}
672 		NFSLOCKNODE(np);
673 		np->n_mtime = vattr.va_mtime;
674 		if (NFS_ISV4(vp))
675 			np->n_change = vattr.va_filerev;
676 	} else {
677 		NFSUNLOCKNODE(np);
678 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
679 		if (error) {
680 			if (NFS_ISV4(vp))
681 				(void) nfsrpc_close(vp, 0, ap->a_td);
682 			return (error);
683 		}
684 		NFSLOCKNODE(np);
685 		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
686 		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
687 			if (vp->v_type == VDIR)
688 				np->n_direofoffset = 0;
689 			NFSUNLOCKNODE(np);
690 			if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
691 				NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
692 				if (VN_IS_DOOMED(vp))
693 					return (EBADF);
694 			}
695 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
696 			if (error == EINTR || error == EIO) {
697 				if (NFS_ISV4(vp))
698 					(void) nfsrpc_close(vp, 0, ap->a_td);
699 				return (error);
700 			}
701 			NFSLOCKNODE(np);
702 			np->n_mtime = vattr.va_mtime;
703 			if (NFS_ISV4(vp))
704 				np->n_change = vattr.va_filerev;
705 		}
706 	}
707 
708 	/*
709 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
710 	 */
711 	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
712 	    (vp->v_type == VREG)) {
713 		if (np->n_directio_opens == 0) {
714 			NFSUNLOCKNODE(np);
715 			if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
716 				NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
717 				if (VN_IS_DOOMED(vp))
718 					return (EBADF);
719 			}
720 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
721 			if (error) {
722 				if (NFS_ISV4(vp))
723 					(void) nfsrpc_close(vp, 0, ap->a_td);
724 				return (error);
725 			}
726 			NFSLOCKNODE(np);
727 			np->n_flag |= NNONCACHE;
728 		}
729 		np->n_directio_opens++;
730 	}
731 
732 	/* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
733 	if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
734 		np->n_flag |= NWRITEOPENED;
735 
736 	/*
737 	 * If this is an open for writing, capture a reference to the
738 	 * credentials, so they can be used by ncl_putpages(). Using
739 	 * these write credentials is preferable to the credentials of
740 	 * whatever thread happens to be doing the VOP_PUTPAGES() since
741 	 * the write RPCs are less likely to fail with EACCES.
742 	 */
743 	if ((fmode & FWRITE) != 0) {
744 		cred = np->n_writecred;
745 		np->n_writecred = crhold(ap->a_cred);
746 	} else
747 		cred = NULL;
748 	NFSUNLOCKNODE(np);
749 
750 	if (cred != NULL)
751 		crfree(cred);
752 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
753 
754 	/*
755 	 * If the text file has been mmap'd, flush any dirty pages to the
756 	 * buffer cache and then...
757 	 * Make sure all writes are pushed to the NFS server.  If this is not
758 	 * done, the modify time of the file can change while the text
759 	 * file is being executed.  This will cause the process that is
760 	 * executing the text file to be terminated.
761 	 */
762 	if (vp->v_writecount <= -1) {
763 		if ((obj = vp->v_object) != NULL &&
764 		    vm_object_mightbedirty(obj)) {
765 			if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
766 				NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
767 				if (VN_IS_DOOMED(vp))
768 					return (EBADF);
769 			}
770 			vnode_pager_clean_sync(vp);
771 		}
772 
773 		/* Now, flush the buffer cache. */
774 		ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
775 
776 		/* And, finally, make sure that n_mtime is up to date. */
777 		np = VTONFS(vp);
778 		NFSLOCKNODE(np);
779 		np->n_mtime = np->n_vattr.na_mtime;
780 		NFSUNLOCKNODE(np);
781 	}
782 	return (0);
783 }
784 
785 /*
786  * nfs close vnode op
787  * What an NFS client should do upon close after writing is a debatable issue.
788  * Most NFS clients push delayed writes to the server upon close, basically for
789  * two reasons:
790  * 1 - So that any write errors may be reported back to the client process
791  *     doing the close system call. By far the two most likely errors are
792  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
793  * 2 - To put a worst case upper bound on cache inconsistency between
794  *     multiple clients for the file.
795  * There is also a consistency problem for Version 2 of the protocol w.r.t.
796  * not being able to tell if other clients are writing a file concurrently,
797  * since there is no way of knowing if the changed modify time in the reply
798  * is only due to the write for this client.
799  * (NFS Version 3 provides weak cache consistency data in the reply that
800  *  should be sufficient to detect and handle this case.)
801  *
802  * The current code does the following:
803  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
804  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
805  *                     or commit them (this satisfies 1 and 2 except for the
806  *                     case where the server crashes after this close but
807  *                     before the commit RPC, which is felt to be "good
808  *                     enough". Changing the last argument to ncl_flush() to
809  *                     a 1 would force a commit operation, if it is felt a
810  *                     commit is necessary now.
811  * for NFS Version 4 - flush the dirty buffers and commit them, if
812  *		       nfscl_mustflush() says this is necessary.
813  *                     It is necessary if there is no write delegation held,
814  *                     in order to satisfy open/close coherency.
815  *                     If the file isn't cached on local stable storage,
816  *                     it may be necessary in order to detect "out of space"
817  *                     errors from the server, if the write delegation
818  *                     issued by the server doesn't allow the file to grow.
819  */
820 /* ARGSUSED */
821 static int
822 nfs_close(struct vop_close_args *ap)
823 {
824 	struct vnode *vp = ap->a_vp;
825 	struct nfsnode *np = VTONFS(vp);
826 	struct nfsvattr nfsva;
827 	struct ucred *cred;
828 	int error = 0, ret, localcred = 0;
829 	int fmode = ap->a_fflag;
830 
831 	if (NFSCL_FORCEDISM(vp->v_mount))
832 		return (0);
833 	/*
834 	 * During shutdown, a_cred isn't valid, so just use root.
835 	 */
836 	if (ap->a_cred == NOCRED) {
837 		cred = newnfs_getcred();
838 		localcred = 1;
839 	} else {
840 		cred = ap->a_cred;
841 	}
842 	if (vp->v_type == VREG) {
843 	    /*
844 	     * Examine and clean dirty pages, regardless of NMODIFIED.
845 	     * This closes a major hole in close-to-open consistency.
846 	     * We want to push out all dirty pages (and buffers) on
847 	     * close, regardless of whether they were dirtied by
848 	     * mmap'ed writes or via write().
849 	     */
850 	    if (nfs_clean_pages_on_close && vp->v_object) {
851 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
852 			NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
853 			if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK)
854 				return (EBADF);
855 		}
856 		vnode_pager_clean_async(vp);
857 	    }
858 	    NFSLOCKNODE(np);
859 	    if (np->n_flag & NMODIFIED) {
860 		NFSUNLOCKNODE(np);
861 		if (NFS_ISV3(vp)) {
862 		    /*
863 		     * Under NFSv3 we have dirty buffers to dispose of.  We
864 		     * must flush them to the NFS server.  We have the option
865 		     * of waiting all the way through the commit rpc or just
866 		     * waiting for the initial write.  The default is to only
867 		     * wait through the initial write so the data is in the
868 		     * server's cache, which is roughly similar to the state
869 		     * a standard disk subsystem leaves the file in on close().
870 		     *
871 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
872 		     * potential races with other processes, and certainly
873 		     * cannot clear it if we don't commit.
874 		     * These races occur when there is no longer the old
875 		     * traditional vnode locking implemented for Vnode Ops.
876 		     */
877 		    int cm = newnfs_commit_on_close ? 1 : 0;
878 		    if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
879 			    NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
880 			    if (VN_IS_DOOMED(vp) && ap->a_fflag != FNONBLOCK)
881 				    return (EBADF);
882 		    }
883 		    error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
884 		    /* np->n_flag &= ~NMODIFIED; */
885 		} else if (NFS_ISV4(vp)) {
886 			if (nfscl_mustflush(vp) != 0) {
887 				int cm = newnfs_commit_on_close ? 1 : 0;
888 				if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
889 					NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
890 					if (VN_IS_DOOMED(vp) && ap->a_fflag !=
891 					    FNONBLOCK)
892 						return (EBADF);
893 				}
894 				error = ncl_flush(vp, MNT_WAIT, ap->a_td,
895 				    cm, 0);
896 				/*
897 				 * as above w.r.t races when clearing
898 				 * NMODIFIED.
899 				 * np->n_flag &= ~NMODIFIED;
900 				 */
901 			}
902 		} else {
903 			if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
904 				NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
905 				if (VN_IS_DOOMED(vp) && ap->a_fflag !=
906 				    FNONBLOCK)
907 					return (EBADF);
908 			}
909 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
910 		}
911 		NFSLOCKNODE(np);
912 	    }
913  	    /*
914  	     * Invalidate the attribute cache in all cases.
915  	     * An open is going to fetch fresh attrs any way, other procs
916  	     * on this node that have file open will be forced to do an
917  	     * otw attr fetch, but this is safe.
918 	     * --> A user found that their RPC count dropped by 20% when
919 	     *     this was commented out and I can't see any requirement
920 	     *     for it, so I've disabled it when negative lookups are
921 	     *     enabled. (What does this have to do with negative lookup
922 	     *     caching? Well nothing, except it was reported by the
923 	     *     same user that needed negative lookup caching and I wanted
924 	     *     there to be a way to disable it to see if it
925 	     *     is the cause of some caching/coherency issue that might
926 	     *     crop up.)
927  	     */
928 	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
929 		    np->n_attrstamp = 0;
930 		    KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
931 	    }
932 	    if (np->n_flag & NWRITEERR) {
933 		np->n_flag &= ~NWRITEERR;
934 		error = np->n_error;
935 	    }
936 	    NFSUNLOCKNODE(np);
937 	}
938 
939 	if (NFS_ISV4(vp)) {
940 		/*
941 		 * Get attributes so "change" is up to date.
942 		 */
943 		if (error == 0 && nfscl_nodeleg(vp, 0) != 0 &&
944 		    vp->v_type == VREG &&
945 		    (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
946 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva);
947 			if (!ret) {
948 				np->n_change = nfsva.na_filerev;
949 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
950 				    0, 0);
951 			}
952 		}
953 
954 		/*
955 		 * and do the close.
956 		 */
957 		ret = nfsrpc_close(vp, 0, ap->a_td);
958 		if (!error && ret)
959 			error = ret;
960 		if (error)
961 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
962 			    (gid_t)0);
963 	}
964 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
965 		NFSLOCKNODE(np);
966 		KASSERT((np->n_directio_opens > 0),
967 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
968 		np->n_directio_opens--;
969 		if (np->n_directio_opens == 0)
970 			np->n_flag &= ~NNONCACHE;
971 		NFSUNLOCKNODE(np);
972 	}
973 	if (localcred)
974 		NFSFREECRED(cred);
975 	return (error);
976 }
977 
978 /*
979  * nfs getattr call from vfs.
980  */
981 static int
982 nfs_getattr(struct vop_getattr_args *ap)
983 {
984 	struct vnode *vp = ap->a_vp;
985 	struct thread *td = curthread;	/* XXX */
986 	struct nfsnode *np = VTONFS(vp);
987 	int error = 0;
988 	struct nfsvattr nfsva;
989 	struct vattr *vap = ap->a_vap;
990 	struct vattr vattr;
991 	struct nfsmount *nmp;
992 
993 	nmp = VFSTONFS(vp->v_mount);
994 	/*
995 	 * Update local times for special files.
996 	 */
997 	NFSLOCKNODE(np);
998 	if (np->n_flag & (NACC | NUPD))
999 		np->n_flag |= NCHG;
1000 	NFSUNLOCKNODE(np);
1001 	/*
1002 	 * First look in the cache.
1003 	 * For "syskrb5" mounts, nm_fhsize might still be zero and
1004 	 * cached attributes should be ignored.
1005 	 */
1006 	if (nmp->nm_fhsize > 0 && ncl_getattrcache(vp, &vattr) == 0) {
1007 		ncl_copy_vattr(vap, &vattr);
1008 
1009 		/*
1010 		 * Get the local modify time for the case of a write
1011 		 * delegation.
1012 		 */
1013 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
1014 		return (0);
1015 	}
1016 
1017 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
1018 	    nfsaccess_cache_timeout > 0) {
1019 		NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
1020 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
1021 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
1022 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
1023 			return (0);
1024 		}
1025 	}
1026 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva);
1027 	if (!error)
1028 		error = nfscl_loadattrcache(&vp, &nfsva, vap, 0, 0);
1029 	if (!error) {
1030 		/*
1031 		 * Get the local modify time for the case of a write
1032 		 * delegation.
1033 		 */
1034 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
1035 	} else if (NFS_ISV4(vp)) {
1036 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1037 	}
1038 	return (error);
1039 }
1040 
1041 /*
1042  * nfs setattr call.
1043  */
1044 static int
1045 nfs_setattr(struct vop_setattr_args *ap)
1046 {
1047 	struct vnode *vp = ap->a_vp;
1048 	struct nfsnode *np = VTONFS(vp);
1049 	struct thread *td = curthread;	/* XXX */
1050 	struct vattr *vap = ap->a_vap;
1051 	int error = 0;
1052 	u_quad_t tsize;
1053 	struct timespec ts;
1054 
1055 #ifndef nolint
1056 	tsize = (u_quad_t)0;
1057 #endif
1058 
1059 	/*
1060 	 * Setting of flags and marking of atimes are not supported.
1061 	 */
1062 	if (vap->va_flags != VNOVAL)
1063 		return (EOPNOTSUPP);
1064 
1065 	/*
1066 	 * Disallow write attempts if the filesystem is mounted read-only.
1067 	 */
1068   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
1069 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
1070 	    vap->va_mtime.tv_sec != VNOVAL ||
1071 	    vap->va_birthtime.tv_sec != VNOVAL ||
1072 	    vap->va_mode != (mode_t)VNOVAL) &&
1073 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
1074 		return (EROFS);
1075 	if (vap->va_size != VNOVAL) {
1076  		switch (vp->v_type) {
1077  		case VDIR:
1078  			return (EISDIR);
1079  		case VCHR:
1080  		case VBLK:
1081  		case VSOCK:
1082  		case VFIFO:
1083 			if (vap->va_mtime.tv_sec == VNOVAL &&
1084 			    vap->va_atime.tv_sec == VNOVAL &&
1085 			    vap->va_birthtime.tv_sec == VNOVAL &&
1086 			    vap->va_mode == (mode_t)VNOVAL &&
1087 			    vap->va_uid == (uid_t)VNOVAL &&
1088 			    vap->va_gid == (gid_t)VNOVAL)
1089 				return (0);
1090  			vap->va_size = VNOVAL;
1091  			break;
1092  		default:
1093 			/*
1094 			 * Disallow write attempts if the filesystem is
1095 			 * mounted read-only.
1096 			 */
1097 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
1098 				return (EROFS);
1099 			/*
1100 			 *  We run vnode_pager_setsize() early (why?),
1101 			 * we must set np->n_size now to avoid vinvalbuf
1102 			 * V_SAVE races that might setsize a lower
1103 			 * value.
1104 			 */
1105 			NFSLOCKNODE(np);
1106 			tsize = np->n_size;
1107 			NFSUNLOCKNODE(np);
1108 			error = ncl_meta_setsize(vp, td, vap->va_size);
1109 			NFSLOCKNODE(np);
1110  			if (np->n_flag & NMODIFIED) {
1111 			    tsize = np->n_size;
1112 			    NFSUNLOCKNODE(np);
1113 			    error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
1114 			        0 : V_SAVE, td, 1);
1115 			    if (error != 0) {
1116 				    vnode_pager_setsize(vp, tsize);
1117 				    return (error);
1118 			    }
1119 			    /*
1120 			     * Call nfscl_delegmodtime() to set the modify time
1121 			     * locally, as required.
1122 			     */
1123 			    nfscl_delegmodtime(vp);
1124  			} else
1125 			    NFSUNLOCKNODE(np);
1126 			/*
1127 			 * np->n_size has already been set to vap->va_size
1128 			 * in ncl_meta_setsize(). We must set it again since
1129 			 * nfs_loadattrcache() could be called through
1130 			 * ncl_meta_setsize() and could modify np->n_size.
1131 			 */
1132 			NFSLOCKNODE(np);
1133  			np->n_vattr.na_size = np->n_size = vap->va_size;
1134 			NFSUNLOCKNODE(np);
1135   		}
1136   	} else {
1137 		NFSLOCKNODE(np);
1138 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
1139 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
1140 			NFSUNLOCKNODE(np);
1141 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
1142 			if (error == EINTR || error == EIO)
1143 				return (error);
1144 		} else
1145 			NFSUNLOCKNODE(np);
1146 	}
1147 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
1148 	if (vap->va_size != VNOVAL) {
1149 		if (error == 0) {
1150 			nanouptime(&ts);
1151 			NFSLOCKNODE(np);
1152 			np->n_localmodtime = ts;
1153 			NFSUNLOCKNODE(np);
1154 		} else {
1155 			NFSLOCKNODE(np);
1156 			np->n_size = np->n_vattr.na_size = tsize;
1157 			vnode_pager_setsize(vp, tsize);
1158 			NFSUNLOCKNODE(np);
1159 		}
1160 	}
1161 	return (error);
1162 }
1163 
1164 /*
1165  * Do an nfs setattr rpc.
1166  */
1167 static int
1168 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1169     struct thread *td)
1170 {
1171 	struct nfsnode *np = VTONFS(vp);
1172 	int error, ret, attrflag, i;
1173 	struct nfsvattr nfsva;
1174 
1175 	if (NFS_ISV34(vp)) {
1176 		NFSLOCKNODE(np);
1177 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1178 			np->n_accesscache[i].stamp = 0;
1179 		np->n_flag |= NDELEGMOD;
1180 		NFSUNLOCKNODE(np);
1181 		KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1182 	}
1183 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag);
1184 	if (attrflag) {
1185 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
1186 		if (ret && !error)
1187 			error = ret;
1188 	}
1189 	if (error && NFS_ISV4(vp))
1190 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1191 	return (error);
1192 }
1193 
1194 /*
1195  * nfs lookup call, one step at a time...
1196  * First look in cache
1197  * If not found, unlock the directory nfsnode and do the rpc
1198  */
1199 static int
1200 nfs_lookup(struct vop_lookup_args *ap)
1201 {
1202 	struct componentname *cnp = ap->a_cnp;
1203 	struct vnode *dvp = ap->a_dvp;
1204 	struct vnode **vpp = ap->a_vpp;
1205 	struct mount *mp = dvp->v_mount;
1206 	int flags = cnp->cn_flags;
1207 	struct vnode *newvp;
1208 	struct nfsmount *nmp;
1209 	struct nfsnode *np, *newnp;
1210 	int error = 0, attrflag, dattrflag, ltype, ncticks;
1211 	struct thread *td = curthread;
1212 	struct nfsfh *nfhp;
1213 	struct nfsvattr dnfsva, nfsva;
1214 	struct vattr vattr;
1215 	struct timespec nctime, ts;
1216 	uint32_t openmode;
1217 
1218 	*vpp = NULLVP;
1219 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1220 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1221 		return (EROFS);
1222 	if (dvp->v_type != VDIR)
1223 		return (ENOTDIR);
1224 	nmp = VFSTONFS(mp);
1225 	np = VTONFS(dvp);
1226 
1227 	/* For NFSv4, wait until any remove is done. */
1228 	NFSLOCKNODE(np);
1229 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1230 		np->n_flag |= NREMOVEWANT;
1231 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1232 	}
1233 	NFSUNLOCKNODE(np);
1234 
1235 	error = vn_dir_check_exec(dvp, cnp);
1236 	if (error != 0)
1237 		return (error);
1238 	error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1239 	if (error > 0 && error != ENOENT)
1240 		return (error);
1241 	if (error == -1) {
1242 		/*
1243 		 * Lookups of "." are special and always return the
1244 		 * current directory.  cache_lookup() already handles
1245 		 * associated locking bookkeeping, etc.
1246 		 */
1247 		if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1248 			return (0);
1249 		}
1250 
1251 		/*
1252 		 * We only accept a positive hit in the cache if the
1253 		 * change time of the file matches our cached copy.
1254 		 * Otherwise, we discard the cache entry and fallback
1255 		 * to doing a lookup RPC.  We also only trust cache
1256 		 * entries for less than nm_nametimeo seconds.
1257 		 *
1258 		 * To better handle stale file handles and attributes,
1259 		 * clear the attribute cache of this node if it is a
1260 		 * leaf component, part of an open() call, and not
1261 		 * locally modified before fetching the attributes.
1262 		 * This should allow stale file handles to be detected
1263 		 * here where we can fall back to a LOOKUP RPC to
1264 		 * recover rather than having nfs_open() detect the
1265 		 * stale file handle and failing open(2) with ESTALE.
1266 		 */
1267 		newvp = *vpp;
1268 		newnp = VTONFS(newvp);
1269 		if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1270 		    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1271 		    !(newnp->n_flag & NMODIFIED)) {
1272 			NFSLOCKNODE(newnp);
1273 			newnp->n_attrstamp = 0;
1274 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1275 			NFSUNLOCKNODE(newnp);
1276 		}
1277 		if (nfscl_nodeleg(newvp, 0) == 0 ||
1278 		    ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1279 		    VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1280 		    timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1281 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1282 			return (0);
1283 		}
1284 		cache_purge(newvp);
1285 		if (dvp != newvp)
1286 			vput(newvp);
1287 		else
1288 			vrele(newvp);
1289 		*vpp = NULLVP;
1290 	} else if (error == ENOENT) {
1291 		if (VN_IS_DOOMED(dvp))
1292 			return (ENOENT);
1293 		/*
1294 		 * We only accept a negative hit in the cache if the
1295 		 * modification time of the parent directory matches
1296 		 * the cached copy in the name cache entry.
1297 		 * Otherwise, we discard all of the negative cache
1298 		 * entries for this directory.  We also only trust
1299 		 * negative cache entries for up to nm_negnametimeo
1300 		 * seconds.
1301 		 */
1302 		if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1303 		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1304 		    timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1305 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1306 			return (ENOENT);
1307 		}
1308 		cache_purge_negative(dvp);
1309 	}
1310 
1311 	openmode = 0;
1312 	/*
1313 	 * If this an NFSv4.1/4.2 mount using the "oneopenown" mount
1314 	 * option, it is possible to do the Open operation in the same
1315 	 * compound as Lookup, so long as delegations are not being
1316 	 * issued.  This saves doing a separate RPC for Open.
1317 	 * For pnfs, do not do this, since the Open+LayoutGet will
1318 	 * be needed as a separate RPC.
1319 	 */
1320 	NFSLOCKMNT(nmp);
1321 	if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) && !NFSHASPNFS(nmp) &&
1322 	    (nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 &&
1323 	    (!NFSMNT_RDONLY(mp) || (flags & OPENWRITE) == 0) &&
1324 	    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN)) {
1325 		if ((flags & OPENREAD) != 0)
1326 			openmode |= NFSV4OPEN_ACCESSREAD;
1327 		if ((flags & OPENWRITE) != 0)
1328 			openmode |= NFSV4OPEN_ACCESSWRITE;
1329 	}
1330 	NFSUNLOCKMNT(nmp);
1331 
1332 	newvp = NULLVP;
1333 	NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1334 	nanouptime(&ts);
1335 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1336 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1337 	    openmode);
1338 	if (dattrflag)
1339 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
1340 	if (error) {
1341 		if (newvp != NULLVP) {
1342 			vput(newvp);
1343 			*vpp = NULLVP;
1344 		}
1345 
1346 		if (error != ENOENT) {
1347 			if (NFS_ISV4(dvp))
1348 				error = nfscl_maperr(td, error, (uid_t)0,
1349 				    (gid_t)0);
1350 			return (error);
1351 		}
1352 
1353 		/* The requested file was not found. */
1354 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1355 		    (flags & ISLASTCN)) {
1356 			/*
1357 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1358 			 * VWRITE) here instead of just checking
1359 			 * MNT_RDONLY.
1360 			 */
1361 			if (mp->mnt_flag & MNT_RDONLY)
1362 				return (EROFS);
1363 			return (EJUSTRETURN);
1364 		}
1365 
1366 		if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1367 			/*
1368 			 * Cache the modification time of the parent
1369 			 * directory from the post-op attributes in
1370 			 * the name cache entry.  The negative cache
1371 			 * entry will be ignored once the directory
1372 			 * has changed.  Don't bother adding the entry
1373 			 * if the directory has already changed.
1374 			 */
1375 			NFSLOCKNODE(np);
1376 			if (timespeccmp(&np->n_vattr.na_mtime,
1377 			    &dnfsva.na_mtime, ==)) {
1378 				NFSUNLOCKNODE(np);
1379 				cache_enter_time(dvp, NULL, cnp,
1380 				    &dnfsva.na_mtime, NULL);
1381 			} else
1382 				NFSUNLOCKNODE(np);
1383 		}
1384 		return (ENOENT);
1385 	}
1386 
1387 	/*
1388 	 * Handle RENAME case...
1389 	 */
1390 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1391 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1392 			free(nfhp, M_NFSFH);
1393 			return (EISDIR);
1394 		}
1395 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, LK_EXCLUSIVE);
1396 		if (error)
1397 			return (error);
1398 		newvp = NFSTOV(np);
1399 		/*
1400 		 * If n_localmodtime >= time before RPC, then
1401 		 * a file modification operation, such as
1402 		 * VOP_SETATTR() of size, has occurred while
1403 		 * the Lookup RPC and acquisition of the vnode
1404 		 * happened.  As such, the attributes might
1405 		 * be stale, with possibly an incorrect size.
1406 		 */
1407 		NFSLOCKNODE(np);
1408 		if (timespecisset(&np->n_localmodtime) &&
1409 		    timespeccmp(&np->n_localmodtime, &ts, >=)) {
1410 			NFSCL_DEBUG(4, "nfs_lookup: rename localmod "
1411 			    "stale attributes\n");
1412 			attrflag = 0;
1413 		}
1414 		NFSUNLOCKNODE(np);
1415 		if (attrflag)
1416 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1417 		*vpp = newvp;
1418 		return (0);
1419 	}
1420 
1421 	if (flags & ISDOTDOT) {
1422 		ltype = NFSVOPISLOCKED(dvp);
1423 		error = vfs_busy(mp, MBF_NOWAIT);
1424 		if (error != 0) {
1425 			vfs_ref(mp);
1426 			NFSVOPUNLOCK(dvp);
1427 			error = vfs_busy(mp, 0);
1428 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1429 			vfs_rel(mp);
1430 			if (error == 0 && VN_IS_DOOMED(dvp)) {
1431 				vfs_unbusy(mp);
1432 				error = ENOENT;
1433 			}
1434 			if (error != 0)
1435 				return (error);
1436 		}
1437 		NFSVOPUNLOCK(dvp);
1438 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np,
1439 		    cnp->cn_lkflags);
1440 		if (error == 0)
1441 			newvp = NFSTOV(np);
1442 		vfs_unbusy(mp);
1443 		if (newvp != dvp)
1444 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1445 		if (VN_IS_DOOMED(dvp)) {
1446 			if (error == 0) {
1447 				if (newvp == dvp)
1448 					vrele(newvp);
1449 				else
1450 					vput(newvp);
1451 			}
1452 			error = ENOENT;
1453 		}
1454 		if (error != 0)
1455 			return (error);
1456 		if (attrflag)
1457 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1458 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1459 		free(nfhp, M_NFSFH);
1460 		VREF(dvp);
1461 		newvp = dvp;
1462 		if (attrflag)
1463 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1464 	} else {
1465 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np,
1466 		    cnp->cn_lkflags);
1467 		if (error)
1468 			return (error);
1469 		newvp = NFSTOV(np);
1470 		/*
1471 		 * If n_localmodtime >= time before RPC, then
1472 		 * a file modification operation, such as
1473 		 * VOP_SETATTR() of size, has occurred while
1474 		 * the Lookup RPC and acquisition of the vnode
1475 		 * happened.  As such, the attributes might
1476 		 * be stale, with possibly an incorrect size.
1477 		 */
1478 		NFSLOCKNODE(np);
1479 		if (timespecisset(&np->n_localmodtime) &&
1480 		    timespeccmp(&np->n_localmodtime, &ts, >=)) {
1481 			NFSCL_DEBUG(4, "nfs_lookup: localmod "
1482 			    "stale attributes\n");
1483 			attrflag = 0;
1484 		}
1485 		NFSUNLOCKNODE(np);
1486 		if (attrflag)
1487 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1488 		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1489 		    !(np->n_flag & NMODIFIED)) {
1490 			/*
1491 			 * Flush the attribute cache when opening a
1492 			 * leaf node to ensure that fresh attributes
1493 			 * are fetched in nfs_open() since we did not
1494 			 * fetch attributes from the LOOKUP reply.
1495 			 */
1496 			NFSLOCKNODE(np);
1497 			np->n_attrstamp = 0;
1498 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1499 			NFSUNLOCKNODE(np);
1500 		}
1501 	}
1502 	if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp &&
1503 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1504 	    attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1505 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1506 		    newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1507 	*vpp = newvp;
1508 	return (0);
1509 }
1510 
1511 /*
1512  * nfs read call.
1513  * Just call ncl_bioread() to do the work.
1514  */
1515 static int
1516 nfs_read(struct vop_read_args *ap)
1517 {
1518 	struct vnode *vp = ap->a_vp;
1519 
1520 	switch (vp->v_type) {
1521 	case VREG:
1522 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1523 	case VDIR:
1524 		return (EISDIR);
1525 	default:
1526 		return (EOPNOTSUPP);
1527 	}
1528 }
1529 
1530 /*
1531  * nfs readlink call
1532  */
1533 static int
1534 nfs_readlink(struct vop_readlink_args *ap)
1535 {
1536 	struct vnode *vp = ap->a_vp;
1537 
1538 	if (vp->v_type != VLNK)
1539 		return (EINVAL);
1540 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1541 }
1542 
1543 /*
1544  * Do a readlink rpc.
1545  * Called by ncl_doio() from below the buffer cache.
1546  */
1547 int
1548 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1549 {
1550 	int error, ret, attrflag;
1551 	struct nfsvattr nfsva;
1552 
1553 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1554 	    &attrflag);
1555 	if (attrflag) {
1556 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
1557 		if (ret && !error)
1558 			error = ret;
1559 	}
1560 	if (error && NFS_ISV4(vp))
1561 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1562 	return (error);
1563 }
1564 
1565 /*
1566  * nfs read rpc call
1567  * Ditto above
1568  */
1569 int
1570 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1571 {
1572 	int error, ret, attrflag;
1573 	struct nfsvattr nfsva;
1574 	struct nfsmount *nmp;
1575 
1576 	nmp = VFSTONFS(vp->v_mount);
1577 	error = EIO;
1578 	attrflag = 0;
1579 	if (NFSHASPNFS(nmp))
1580 		error = nfscl_doiods(vp, uiop, NULL, NULL,
1581 		    NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1582 	NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1583 	if (error != 0 && error != EFAULT)
1584 		error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1585 		    &attrflag);
1586 	if (attrflag) {
1587 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
1588 		if (ret && !error)
1589 			error = ret;
1590 	}
1591 	if (error && NFS_ISV4(vp))
1592 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1593 	return (error);
1594 }
1595 
1596 /*
1597  * nfs write call
1598  */
1599 int
1600 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1601     int *iomode, int *must_commit, int called_from_strategy, int ioflag)
1602 {
1603 	struct nfsvattr nfsva;
1604 	int error, attrflag, ret;
1605 	struct nfsmount *nmp;
1606 
1607 	nmp = VFSTONFS(vp->v_mount);
1608 	error = EIO;
1609 	attrflag = 0;
1610 	if (NFSHASPNFS(nmp))
1611 		error = nfscl_doiods(vp, uiop, iomode, must_commit,
1612 		    NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1613 	NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1614 	if (error != 0 && error != EFAULT)
1615 		error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1616 		    uiop->uio_td, &nfsva, &attrflag, called_from_strategy,
1617 		    ioflag);
1618 	if (attrflag) {
1619 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1620 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 1, 1);
1621 		else
1622 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
1623 		if (ret && !error)
1624 			error = ret;
1625 	}
1626 	if (DOINGASYNC(vp))
1627 		*iomode = NFSWRITE_FILESYNC;
1628 	if (error && NFS_ISV4(vp))
1629 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1630 	return (error);
1631 }
1632 
1633 /*
1634  * nfs mknod rpc
1635  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1636  * mode set to specify the file type and the size field for rdev.
1637  */
1638 static int
1639 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1640     struct vattr *vap)
1641 {
1642 	struct nfsvattr nfsva, dnfsva;
1643 	struct vnode *newvp = NULL;
1644 	struct nfsnode *np = NULL, *dnp;
1645 	struct nfsfh *nfhp;
1646 	struct vattr vattr;
1647 	int error = 0, attrflag, dattrflag;
1648 	u_int32_t rdev;
1649 
1650 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1651 		rdev = vap->va_rdev;
1652 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1653 		rdev = 0xffffffff;
1654 	else
1655 		return (EOPNOTSUPP);
1656 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1657 		return (error);
1658 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1659 	    rdev, vap->va_type, cnp->cn_cred, curthread, &dnfsva,
1660 	    &nfsva, &nfhp, &attrflag, &dattrflag);
1661 	if (!error) {
1662 		if (!nfhp)
1663 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1664 			    cnp->cn_namelen, cnp->cn_cred, curthread,
1665 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0);
1666 		if (nfhp)
1667 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1668 			    curthread, &np, LK_EXCLUSIVE);
1669 	}
1670 	if (dattrflag)
1671 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
1672 	if (!error) {
1673 		newvp = NFSTOV(np);
1674 		if (attrflag != 0) {
1675 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1676 			if (error != 0)
1677 				vput(newvp);
1678 		}
1679 	}
1680 	if (!error) {
1681 		*vpp = newvp;
1682 	} else if (NFS_ISV4(dvp)) {
1683 		error = nfscl_maperr(curthread, error, vap->va_uid,
1684 		    vap->va_gid);
1685 	}
1686 	dnp = VTONFS(dvp);
1687 	NFSLOCKNODE(dnp);
1688 	dnp->n_flag |= NMODIFIED;
1689 	if (!dattrflag) {
1690 		dnp->n_attrstamp = 0;
1691 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1692 	}
1693 	NFSUNLOCKNODE(dnp);
1694 	return (error);
1695 }
1696 
1697 /*
1698  * nfs mknod vop
1699  * just call nfs_mknodrpc() to do the work.
1700  */
1701 /* ARGSUSED */
1702 static int
1703 nfs_mknod(struct vop_mknod_args *ap)
1704 {
1705 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1706 }
1707 
1708 static struct mtx nfs_cverf_mtx;
1709 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1710     MTX_DEF);
1711 
1712 static nfsquad_t
1713 nfs_get_cverf(void)
1714 {
1715 	static nfsquad_t cverf;
1716 	nfsquad_t ret;
1717 	static int cverf_initialized = 0;
1718 
1719 	mtx_lock(&nfs_cverf_mtx);
1720 	if (cverf_initialized == 0) {
1721 		cverf.lval[0] = arc4random();
1722 		cverf.lval[1] = arc4random();
1723 		cverf_initialized = 1;
1724 	} else
1725 		cverf.qval++;
1726 	ret = cverf;
1727 	mtx_unlock(&nfs_cverf_mtx);
1728 
1729 	return (ret);
1730 }
1731 
1732 /*
1733  * nfs file create call
1734  */
1735 static int
1736 nfs_create(struct vop_create_args *ap)
1737 {
1738 	struct vnode *dvp = ap->a_dvp;
1739 	struct vattr *vap = ap->a_vap;
1740 	struct componentname *cnp = ap->a_cnp;
1741 	struct nfsnode *np = NULL, *dnp;
1742 	struct vnode *newvp = NULL;
1743 	struct nfsmount *nmp;
1744 	struct nfsvattr dnfsva, nfsva;
1745 	struct nfsfh *nfhp;
1746 	nfsquad_t cverf;
1747 	int error = 0, attrflag, dattrflag, fmode = 0;
1748 	struct vattr vattr;
1749 
1750 	/*
1751 	 * Oops, not for me..
1752 	 */
1753 	if (vap->va_type == VSOCK)
1754 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1755 
1756 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1757 		return (error);
1758 	if (vap->va_vaflags & VA_EXCLUSIVE)
1759 		fmode |= O_EXCL;
1760 	dnp = VTONFS(dvp);
1761 	nmp = VFSTONFS(dvp->v_mount);
1762 again:
1763 	/* For NFSv4, wait until any remove is done. */
1764 	NFSLOCKNODE(dnp);
1765 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1766 		dnp->n_flag |= NREMOVEWANT;
1767 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1768 	}
1769 	NFSUNLOCKNODE(dnp);
1770 
1771 	cverf = nfs_get_cverf();
1772 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1773 	    vap, cverf, fmode, cnp->cn_cred, curthread, &dnfsva, &nfsva,
1774 	    &nfhp, &attrflag, &dattrflag);
1775 	if (!error) {
1776 		if (nfhp == NULL)
1777 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1778 			    cnp->cn_namelen, cnp->cn_cred, curthread,
1779 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag, 0);
1780 		if (nfhp != NULL)
1781 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1782 			    curthread, &np, LK_EXCLUSIVE);
1783 	}
1784 	if (dattrflag)
1785 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
1786 	if (!error) {
1787 		newvp = NFSTOV(np);
1788 		if (attrflag == 0)
1789 			error = nfsrpc_getattr(newvp, cnp->cn_cred, curthread,
1790 			    &nfsva);
1791 		if (error == 0)
1792 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
1793 	}
1794 	if (error) {
1795 		if (newvp != NULL) {
1796 			vput(newvp);
1797 			newvp = NULL;
1798 		}
1799 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1800 		    error == NFSERR_NOTSUPP) {
1801 			fmode &= ~O_EXCL;
1802 			goto again;
1803 		}
1804 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1805 		if (nfscl_checksattr(vap, &nfsva)) {
1806 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1807 			    curthread, &nfsva, &attrflag);
1808 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1809 			    vap->va_gid != (gid_t)VNOVAL)) {
1810 				/* try again without setting uid/gid */
1811 				vap->va_uid = (uid_t)VNOVAL;
1812 				vap->va_gid = (uid_t)VNOVAL;
1813 				error = nfsrpc_setattr(newvp, vap, NULL,
1814 				    cnp->cn_cred, curthread, &nfsva, &attrflag);
1815 			}
1816 			if (attrflag)
1817 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1818 				    0, 1);
1819 			if (error != 0)
1820 				vput(newvp);
1821 		}
1822 	}
1823 	if (!error) {
1824 		if ((cnp->cn_flags & MAKEENTRY) && attrflag) {
1825 			if (dvp != newvp)
1826 				cache_enter_time(dvp, newvp, cnp,
1827 				    &nfsva.na_ctime, NULL);
1828 			else
1829 				printf("nfs_create: bogus NFS server returned "
1830 				    "the directory as the new file object\n");
1831 		}
1832 		*ap->a_vpp = newvp;
1833 	} else if (NFS_ISV4(dvp)) {
1834 		error = nfscl_maperr(curthread, error, vap->va_uid,
1835 		    vap->va_gid);
1836 	}
1837 	NFSLOCKNODE(dnp);
1838 	dnp->n_flag |= NMODIFIED;
1839 	if (!dattrflag) {
1840 		dnp->n_attrstamp = 0;
1841 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1842 	}
1843 	NFSUNLOCKNODE(dnp);
1844 	return (error);
1845 }
1846 
1847 /*
1848  * nfs file remove call
1849  * To try and make nfs semantics closer to ufs semantics, a file that has
1850  * other processes using the vnode is renamed instead of removed and then
1851  * removed later on the last close.
1852  * - If v_usecount > 1
1853  *	  If a rename is not already in the works
1854  *	     call nfs_sillyrename() to set it up
1855  *     else
1856  *	  do the remove rpc
1857  */
1858 static int
1859 nfs_remove(struct vop_remove_args *ap)
1860 {
1861 	struct vnode *vp = ap->a_vp;
1862 	struct vnode *dvp = ap->a_dvp;
1863 	struct componentname *cnp = ap->a_cnp;
1864 	struct nfsnode *np = VTONFS(vp);
1865 	int error = 0;
1866 	struct vattr vattr;
1867 
1868 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1869 	if (vp->v_type == VDIR)
1870 		error = EPERM;
1871 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1872 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1873 	    vattr.va_nlink > 1)) {
1874 		/*
1875 		 * Purge the name cache so that the chance of a lookup for
1876 		 * the name succeeding while the remove is in progress is
1877 		 * minimized. Without node locking it can still happen, such
1878 		 * that an I/O op returns ESTALE, but since you get this if
1879 		 * another host removes the file..
1880 		 */
1881 		cache_purge(vp);
1882 		/*
1883 		 * throw away biocache buffers, mainly to avoid
1884 		 * unnecessary delayed writes later.
1885 		 */
1886 		error = ncl_vinvalbuf(vp, 0, curthread, 1);
1887 		if (error != EINTR && error != EIO)
1888 			/* Do the rpc */
1889 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1890 			    cnp->cn_namelen, cnp->cn_cred, curthread);
1891 		/*
1892 		 * Kludge City: If the first reply to the remove rpc is lost..
1893 		 *   the reply to the retransmitted request will be ENOENT
1894 		 *   since the file was in fact removed
1895 		 *   Therefore, we cheat and return success.
1896 		 */
1897 		if (error == ENOENT)
1898 			error = 0;
1899 	} else if (!np->n_sillyrename)
1900 		error = nfs_sillyrename(dvp, vp, cnp);
1901 	NFSLOCKNODE(np);
1902 	np->n_attrstamp = 0;
1903 	NFSUNLOCKNODE(np);
1904 	KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1905 	return (error);
1906 }
1907 
1908 /*
1909  * nfs file remove rpc called from nfs_inactive
1910  */
1911 int
1912 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1913 {
1914 	/*
1915 	 * Make sure that the directory vnode is still valid.
1916 	 * XXX we should lock sp->s_dvp here.
1917 	 */
1918 	if (sp->s_dvp->v_type == VBAD)
1919 		return (0);
1920 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1921 	    sp->s_cred, NULL));
1922 }
1923 
1924 /*
1925  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1926  */
1927 static int
1928 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1929     int namelen, struct ucred *cred, struct thread *td)
1930 {
1931 	struct nfsvattr dnfsva;
1932 	struct nfsnode *dnp = VTONFS(dvp);
1933 	int error = 0, dattrflag;
1934 
1935 	NFSLOCKNODE(dnp);
1936 	dnp->n_flag |= NREMOVEINPROG;
1937 	NFSUNLOCKNODE(dnp);
1938 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1939 	    &dattrflag);
1940 	NFSLOCKNODE(dnp);
1941 	if ((dnp->n_flag & NREMOVEWANT)) {
1942 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1943 		NFSUNLOCKNODE(dnp);
1944 		wakeup((caddr_t)dnp);
1945 	} else {
1946 		dnp->n_flag &= ~NREMOVEINPROG;
1947 		NFSUNLOCKNODE(dnp);
1948 	}
1949 	if (dattrflag)
1950 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
1951 	NFSLOCKNODE(dnp);
1952 	dnp->n_flag |= NMODIFIED;
1953 	if (!dattrflag) {
1954 		dnp->n_attrstamp = 0;
1955 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1956 	}
1957 	NFSUNLOCKNODE(dnp);
1958 	if (error && NFS_ISV4(dvp))
1959 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1960 	return (error);
1961 }
1962 
1963 /*
1964  * nfs file rename call
1965  */
1966 static int
1967 nfs_rename(struct vop_rename_args *ap)
1968 {
1969 	struct vnode *fvp = ap->a_fvp;
1970 	struct vnode *tvp = ap->a_tvp;
1971 	struct vnode *fdvp = ap->a_fdvp;
1972 	struct vnode *tdvp = ap->a_tdvp;
1973 	struct componentname *tcnp = ap->a_tcnp;
1974 	struct componentname *fcnp = ap->a_fcnp;
1975 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1976 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1977 	struct nfsv4node *newv4 = NULL;
1978 	int error;
1979 
1980 	/* Check for cross-device rename */
1981 	if ((fvp->v_mount != tdvp->v_mount) ||
1982 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1983 		error = EXDEV;
1984 		goto out;
1985 	}
1986 
1987 	if (fvp == tvp) {
1988 		printf("nfs_rename: fvp == tvp (can't happen)\n");
1989 		error = 0;
1990 		goto out;
1991 	}
1992 	if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1993 		goto out;
1994 
1995 	/*
1996 	 * We have to flush B_DELWRI data prior to renaming
1997 	 * the file.  If we don't, the delayed-write buffers
1998 	 * can be flushed out later after the file has gone stale
1999 	 * under NFSV3.  NFSV2 does not have this problem because
2000 	 * ( as far as I can tell ) it flushes dirty buffers more
2001 	 * often.
2002 	 *
2003 	 * Skip the rename operation if the fsync fails, this can happen
2004 	 * due to the server's volume being full, when we pushed out data
2005 	 * that was written back to our cache earlier. Not checking for
2006 	 * this condition can result in potential (silent) data loss.
2007 	 */
2008 	error = VOP_FSYNC(fvp, MNT_WAIT, curthread);
2009 	NFSVOPUNLOCK(fvp);
2010 	if (!error && tvp)
2011 		error = VOP_FSYNC(tvp, MNT_WAIT, curthread);
2012 	if (error)
2013 		goto out;
2014 
2015 	/*
2016 	 * If the tvp exists and is in use, sillyrename it before doing the
2017 	 * rename of the new file over it.
2018 	 * XXX Can't sillyrename a directory.
2019 	 */
2020 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
2021 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
2022 		vput(tvp);
2023 		tvp = NULL;
2024 	}
2025 
2026 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
2027 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
2028 	    curthread);
2029 
2030 	if (error == 0 && NFS_ISV4(tdvp)) {
2031 		/*
2032 		 * For NFSv4, check to see if it is the same name and
2033 		 * replace the name, if it is different.
2034 		 */
2035 		newv4 = malloc(
2036 		    sizeof (struct nfsv4node) +
2037 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
2038 		    M_NFSV4NODE, M_WAITOK);
2039 		NFSLOCKNODE(tdnp);
2040 		NFSLOCKNODE(fnp);
2041 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
2042 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
2043 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
2044 		      tcnp->cn_namelen) ||
2045 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
2046 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
2047 			tdnp->n_fhp->nfh_len))) {
2048 			free(fnp->n_v4, M_NFSV4NODE);
2049 			fnp->n_v4 = newv4;
2050 			newv4 = NULL;
2051 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
2052 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
2053 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
2054 			    tdnp->n_fhp->nfh_len);
2055 			NFSBCOPY(tcnp->cn_nameptr,
2056 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
2057 		}
2058 		NFSUNLOCKNODE(tdnp);
2059 		NFSUNLOCKNODE(fnp);
2060 		if (newv4 != NULL)
2061 			free(newv4, M_NFSV4NODE);
2062 	}
2063 
2064 	if (fvp->v_type == VDIR) {
2065 		if (tvp != NULL && tvp->v_type == VDIR)
2066 			cache_purge(tdvp);
2067 		cache_purge(fdvp);
2068 	}
2069 
2070 out:
2071 	if (tdvp == tvp)
2072 		vrele(tdvp);
2073 	else
2074 		vput(tdvp);
2075 	if (tvp)
2076 		vput(tvp);
2077 	vrele(fdvp);
2078 	vrele(fvp);
2079 	/*
2080 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2081 	 */
2082 	if (error == ENOENT)
2083 		error = 0;
2084 	return (error);
2085 }
2086 
2087 /*
2088  * nfs file rename rpc called from nfs_remove() above
2089  */
2090 static int
2091 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
2092     struct sillyrename *sp)
2093 {
2094 
2095 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
2096 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
2097 	    curthread));
2098 }
2099 
2100 /*
2101  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2102  */
2103 static int
2104 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
2105     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
2106     int tnamelen, struct ucred *cred, struct thread *td)
2107 {
2108 	struct nfsvattr fnfsva, tnfsva;
2109 	struct nfsnode *fdnp = VTONFS(fdvp);
2110 	struct nfsnode *tdnp = VTONFS(tdvp);
2111 	int error = 0, fattrflag, tattrflag;
2112 
2113 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
2114 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
2115 	    &tattrflag);
2116 	NFSLOCKNODE(fdnp);
2117 	fdnp->n_flag |= NMODIFIED;
2118 	if (fattrflag != 0) {
2119 		NFSUNLOCKNODE(fdnp);
2120 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, 0, 1);
2121 	} else {
2122 		fdnp->n_attrstamp = 0;
2123 		NFSUNLOCKNODE(fdnp);
2124 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
2125 	}
2126 	NFSLOCKNODE(tdnp);
2127 	tdnp->n_flag |= NMODIFIED;
2128 	if (tattrflag != 0) {
2129 		NFSUNLOCKNODE(tdnp);
2130 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, 0, 1);
2131 	} else {
2132 		tdnp->n_attrstamp = 0;
2133 		NFSUNLOCKNODE(tdnp);
2134 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2135 	}
2136 	if (error && NFS_ISV4(fdvp))
2137 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2138 	return (error);
2139 }
2140 
2141 /*
2142  * nfs hard link create call
2143  */
2144 static int
2145 nfs_link(struct vop_link_args *ap)
2146 {
2147 	struct vnode *vp = ap->a_vp;
2148 	struct vnode *tdvp = ap->a_tdvp;
2149 	struct componentname *cnp = ap->a_cnp;
2150 	struct nfsnode *np, *tdnp;
2151 	struct nfsvattr nfsva, dnfsva;
2152 	int error = 0, attrflag, dattrflag;
2153 
2154 	/*
2155 	 * Push all writes to the server, so that the attribute cache
2156 	 * doesn't get "out of sync" with the server.
2157 	 * XXX There should be a better way!
2158 	 */
2159 	VOP_FSYNC(vp, MNT_WAIT, curthread);
2160 
2161 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2162 	    cnp->cn_cred, curthread, &dnfsva, &nfsva, &attrflag, &dattrflag);
2163 	tdnp = VTONFS(tdvp);
2164 	NFSLOCKNODE(tdnp);
2165 	tdnp->n_flag |= NMODIFIED;
2166 	if (dattrflag != 0) {
2167 		NFSUNLOCKNODE(tdnp);
2168 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, 0, 1);
2169 	} else {
2170 		tdnp->n_attrstamp = 0;
2171 		NFSUNLOCKNODE(tdnp);
2172 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2173 	}
2174 	if (attrflag)
2175 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
2176 	else {
2177 		np = VTONFS(vp);
2178 		NFSLOCKNODE(np);
2179 		np->n_attrstamp = 0;
2180 		NFSUNLOCKNODE(np);
2181 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2182 	}
2183 	/*
2184 	 * If negative lookup caching is enabled, I might as well
2185 	 * add an entry for this node. Not necessary for correctness,
2186 	 * but if negative caching is enabled, then the system
2187 	 * must care about lookup caching hit rate, so...
2188 	 */
2189 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2190 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2191 		if (tdvp != vp)
2192 			cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2193 		else
2194 			printf("nfs_link: bogus NFS server returned "
2195 			    "the directory as the new link\n");
2196 	}
2197 	if (error && NFS_ISV4(vp))
2198 		error = nfscl_maperr(curthread, error, (uid_t)0,
2199 		    (gid_t)0);
2200 	return (error);
2201 }
2202 
2203 /*
2204  * nfs symbolic link create call
2205  */
2206 static int
2207 nfs_symlink(struct vop_symlink_args *ap)
2208 {
2209 	struct vnode *dvp = ap->a_dvp;
2210 	struct vattr *vap = ap->a_vap;
2211 	struct componentname *cnp = ap->a_cnp;
2212 	struct nfsvattr nfsva, dnfsva;
2213 	struct nfsfh *nfhp;
2214 	struct nfsnode *np = NULL, *dnp;
2215 	struct vnode *newvp = NULL;
2216 	int error = 0, attrflag, dattrflag, ret;
2217 
2218 	vap->va_type = VLNK;
2219 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2220 	    ap->a_target, vap, cnp->cn_cred, curthread, &dnfsva,
2221 	    &nfsva, &nfhp, &attrflag, &dattrflag);
2222 	if (nfhp) {
2223 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread,
2224 		    &np, LK_EXCLUSIVE);
2225 		if (!ret)
2226 			newvp = NFSTOV(np);
2227 		else if (!error)
2228 			error = ret;
2229 	}
2230 	if (newvp != NULL) {
2231 		if (attrflag)
2232 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
2233 	} else if (!error) {
2234 		/*
2235 		 * If we do not have an error and we could not extract the
2236 		 * newvp from the response due to the request being NFSv2, we
2237 		 * have to do a lookup in order to obtain a newvp to return.
2238 		 */
2239 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2240 		    cnp->cn_cred, curthread, &np);
2241 		if (!error)
2242 			newvp = NFSTOV(np);
2243 	}
2244 	if (error) {
2245 		if (newvp)
2246 			vput(newvp);
2247 		if (NFS_ISV4(dvp))
2248 			error = nfscl_maperr(curthread, error,
2249 			    vap->va_uid, vap->va_gid);
2250 	} else {
2251 		*ap->a_vpp = newvp;
2252 	}
2253 
2254 	dnp = VTONFS(dvp);
2255 	NFSLOCKNODE(dnp);
2256 	dnp->n_flag |= NMODIFIED;
2257 	if (dattrflag != 0) {
2258 		NFSUNLOCKNODE(dnp);
2259 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
2260 	} else {
2261 		dnp->n_attrstamp = 0;
2262 		NFSUNLOCKNODE(dnp);
2263 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2264 	}
2265 	/*
2266 	 * If negative lookup caching is enabled, I might as well
2267 	 * add an entry for this node. Not necessary for correctness,
2268 	 * but if negative caching is enabled, then the system
2269 	 * must care about lookup caching hit rate, so...
2270 	 */
2271 	if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2272 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2273 		if (dvp != newvp)
2274 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2275 			    NULL);
2276 		else
2277 			printf("nfs_symlink: bogus NFS server returned "
2278 			    "the directory as the new file object\n");
2279 	}
2280 	return (error);
2281 }
2282 
2283 /*
2284  * nfs make dir call
2285  */
2286 static int
2287 nfs_mkdir(struct vop_mkdir_args *ap)
2288 {
2289 	struct vnode *dvp = ap->a_dvp;
2290 	struct vattr *vap = ap->a_vap;
2291 	struct componentname *cnp = ap->a_cnp;
2292 	struct nfsnode *np = NULL, *dnp;
2293 	struct vnode *newvp = NULL;
2294 	struct vattr vattr;
2295 	struct nfsfh *nfhp;
2296 	struct nfsvattr nfsva, dnfsva;
2297 	int error = 0, attrflag, dattrflag, ret;
2298 
2299 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2300 		return (error);
2301 	vap->va_type = VDIR;
2302 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2303 	    vap, cnp->cn_cred, curthread, &dnfsva, &nfsva, &nfhp,
2304 	    &attrflag, &dattrflag);
2305 	dnp = VTONFS(dvp);
2306 	NFSLOCKNODE(dnp);
2307 	dnp->n_flag |= NMODIFIED;
2308 	if (dattrflag != 0) {
2309 		NFSUNLOCKNODE(dnp);
2310 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
2311 	} else {
2312 		dnp->n_attrstamp = 0;
2313 		NFSUNLOCKNODE(dnp);
2314 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2315 	}
2316 	if (nfhp) {
2317 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, curthread,
2318 		    &np, LK_EXCLUSIVE);
2319 		if (!ret) {
2320 			newvp = NFSTOV(np);
2321 			if (attrflag)
2322 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2323 				0, 1);
2324 		} else if (!error)
2325 			error = ret;
2326 	}
2327 	if (!error && newvp == NULL) {
2328 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2329 		    cnp->cn_cred, curthread, &np);
2330 		if (!error) {
2331 			newvp = NFSTOV(np);
2332 			if (newvp->v_type != VDIR)
2333 				error = EEXIST;
2334 		}
2335 	}
2336 	if (error) {
2337 		if (newvp)
2338 			vput(newvp);
2339 		if (NFS_ISV4(dvp))
2340 			error = nfscl_maperr(curthread, error,
2341 			    vap->va_uid, vap->va_gid);
2342 	} else {
2343 		/*
2344 		 * If negative lookup caching is enabled, I might as well
2345 		 * add an entry for this node. Not necessary for correctness,
2346 		 * but if negative caching is enabled, then the system
2347 		 * must care about lookup caching hit rate, so...
2348 		 */
2349 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2350 		    (cnp->cn_flags & MAKEENTRY) &&
2351 		    attrflag != 0 && dattrflag != 0) {
2352 			if (dvp != newvp)
2353 				cache_enter_time(dvp, newvp, cnp,
2354 				    &nfsva.na_ctime, &dnfsva.na_ctime);
2355 			else
2356 				printf("nfs_mkdir: bogus NFS server returned "
2357 				    "the directory that the directory was "
2358 				    "created in as the new file object\n");
2359 		}
2360 		*ap->a_vpp = newvp;
2361 	}
2362 	return (error);
2363 }
2364 
2365 /*
2366  * nfs remove directory call
2367  */
2368 static int
2369 nfs_rmdir(struct vop_rmdir_args *ap)
2370 {
2371 	struct vnode *vp = ap->a_vp;
2372 	struct vnode *dvp = ap->a_dvp;
2373 	struct componentname *cnp = ap->a_cnp;
2374 	struct nfsnode *dnp;
2375 	struct nfsvattr dnfsva;
2376 	int error, dattrflag;
2377 
2378 	if (dvp == vp)
2379 		return (EINVAL);
2380 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2381 	    cnp->cn_cred, curthread, &dnfsva, &dattrflag);
2382 	dnp = VTONFS(dvp);
2383 	NFSLOCKNODE(dnp);
2384 	dnp->n_flag |= NMODIFIED;
2385 	if (dattrflag != 0) {
2386 		NFSUNLOCKNODE(dnp);
2387 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
2388 	} else {
2389 		dnp->n_attrstamp = 0;
2390 		NFSUNLOCKNODE(dnp);
2391 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2392 	}
2393 
2394 	cache_purge(dvp);
2395 	cache_purge(vp);
2396 	if (error && NFS_ISV4(dvp))
2397 		error = nfscl_maperr(curthread, error, (uid_t)0,
2398 		    (gid_t)0);
2399 	/*
2400 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2401 	 */
2402 	if (error == ENOENT)
2403 		error = 0;
2404 	return (error);
2405 }
2406 
2407 /*
2408  * nfs readdir call
2409  */
2410 static int
2411 nfs_readdir(struct vop_readdir_args *ap)
2412 {
2413 	struct vnode *vp = ap->a_vp;
2414 	struct nfsnode *np = VTONFS(vp);
2415 	struct uio *uio = ap->a_uio;
2416 	ssize_t tresid, left;
2417 	int error = 0;
2418 	struct vattr vattr;
2419 
2420 	if (ap->a_eofflag != NULL)
2421 		*ap->a_eofflag = 0;
2422 	if (vp->v_type != VDIR)
2423 		return(EPERM);
2424 
2425 	/*
2426 	 * First, check for hit on the EOF offset cache
2427 	 */
2428 	NFSLOCKNODE(np);
2429 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2430 	    (np->n_flag & NMODIFIED) == 0) {
2431 		NFSUNLOCKNODE(np);
2432 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2433 			NFSLOCKNODE(np);
2434 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2435 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2436 				NFSUNLOCKNODE(np);
2437 				NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2438 				if (ap->a_eofflag != NULL)
2439 					*ap->a_eofflag = 1;
2440 				return (0);
2441 			} else
2442 				NFSUNLOCKNODE(np);
2443 		}
2444 	} else
2445 		NFSUNLOCKNODE(np);
2446 
2447 	/*
2448 	 * NFS always guarantees that directory entries don't straddle
2449 	 * DIRBLKSIZ boundaries.  As such, we need to limit the size
2450 	 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2451 	 * directory entry.
2452 	 */
2453 	left = uio->uio_resid % DIRBLKSIZ;
2454 	if (left == uio->uio_resid)
2455 		return (EINVAL);
2456 	uio->uio_resid -= left;
2457 
2458 	/*
2459 	 * For readdirplus, if starting to read the directory,
2460 	 * purge the name cache, since it will be reloaded by
2461 	 * this directory read.
2462 	 * This removes potentially stale name cache entries.
2463 	 */
2464 	if (uio->uio_offset == 0 &&
2465 	    (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_RDIRPLUS) != 0)
2466 		cache_purge(vp);
2467 
2468 	/*
2469 	 * Call ncl_bioread() to do the real work.
2470 	 */
2471 	tresid = uio->uio_resid;
2472 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2473 
2474 	if (!error && uio->uio_resid == tresid) {
2475 		NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2476 		if (ap->a_eofflag != NULL)
2477 			*ap->a_eofflag = 1;
2478 	}
2479 
2480 	/* Add the partial DIRBLKSIZ (left) back in. */
2481 	uio->uio_resid += left;
2482 	return (error);
2483 }
2484 
2485 /*
2486  * Readdir rpc call.
2487  * Called from below the buffer cache by ncl_doio().
2488  */
2489 int
2490 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2491     struct thread *td)
2492 {
2493 	struct nfsvattr nfsva;
2494 	nfsuint64 *cookiep, cookie;
2495 	struct nfsnode *dnp = VTONFS(vp);
2496 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2497 	int error = 0, eof, attrflag;
2498 
2499 	KASSERT(uiop->uio_iovcnt == 1 &&
2500 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2501 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2502 	    ("nfs readdirrpc bad uio"));
2503 
2504 	/*
2505 	 * If there is no cookie, assume directory was stale.
2506 	 */
2507 	ncl_dircookie_lock(dnp);
2508 	NFSUNLOCKNODE(dnp);
2509 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2510 	if (cookiep) {
2511 		cookie = *cookiep;
2512 		ncl_dircookie_unlock(dnp);
2513 	} else {
2514 		ncl_dircookie_unlock(dnp);
2515 		return (NFSERR_BAD_COOKIE);
2516 	}
2517 
2518 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2519 		(void)ncl_fsinfo(nmp, vp, cred, td);
2520 
2521 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2522 	    &attrflag, &eof);
2523 	if (attrflag)
2524 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
2525 
2526 	if (!error) {
2527 		/*
2528 		 * We are now either at the end of the directory or have filled
2529 		 * the block.
2530 		 */
2531 		if (eof) {
2532 			NFSLOCKNODE(dnp);
2533 			dnp->n_direofoffset = uiop->uio_offset;
2534 			NFSUNLOCKNODE(dnp);
2535 		} else {
2536 			if (uiop->uio_resid > 0)
2537 				printf("EEK! readdirrpc resid > 0\n");
2538 			ncl_dircookie_lock(dnp);
2539 			NFSUNLOCKNODE(dnp);
2540 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2541 			*cookiep = cookie;
2542 			ncl_dircookie_unlock(dnp);
2543 		}
2544 	} else if (NFS_ISV4(vp)) {
2545 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2546 	}
2547 	return (error);
2548 }
2549 
2550 /*
2551  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2552  */
2553 int
2554 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2555     struct thread *td)
2556 {
2557 	struct nfsvattr nfsva;
2558 	nfsuint64 *cookiep, cookie;
2559 	struct nfsnode *dnp = VTONFS(vp);
2560 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2561 	int error = 0, attrflag, eof;
2562 
2563 	KASSERT(uiop->uio_iovcnt == 1 &&
2564 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2565 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2566 	    ("nfs readdirplusrpc bad uio"));
2567 
2568 	/*
2569 	 * If there is no cookie, assume directory was stale.
2570 	 */
2571 	ncl_dircookie_lock(dnp);
2572 	NFSUNLOCKNODE(dnp);
2573 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2574 	if (cookiep) {
2575 		cookie = *cookiep;
2576 		ncl_dircookie_unlock(dnp);
2577 	} else {
2578 		ncl_dircookie_unlock(dnp);
2579 		return (NFSERR_BAD_COOKIE);
2580 	}
2581 
2582 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2583 		(void)ncl_fsinfo(nmp, vp, cred, td);
2584 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2585 	    &attrflag, &eof);
2586 	if (attrflag)
2587 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
2588 
2589 	if (!error) {
2590 		/*
2591 		 * We are now either at end of the directory or have filled the
2592 		 * the block.
2593 		 */
2594 		if (eof) {
2595 			NFSLOCKNODE(dnp);
2596 			dnp->n_direofoffset = uiop->uio_offset;
2597 			NFSUNLOCKNODE(dnp);
2598 		} else {
2599 			if (uiop->uio_resid > 0)
2600 				printf("EEK! readdirplusrpc resid > 0\n");
2601 			ncl_dircookie_lock(dnp);
2602 			NFSUNLOCKNODE(dnp);
2603 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2604 			*cookiep = cookie;
2605 			ncl_dircookie_unlock(dnp);
2606 		}
2607 	} else if (NFS_ISV4(vp)) {
2608 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2609 	}
2610 	return (error);
2611 }
2612 
2613 /*
2614  * Silly rename. To make the NFS filesystem that is stateless look a little
2615  * more like the "ufs" a remove of an active vnode is translated to a rename
2616  * to a funny looking filename that is removed by nfs_inactive on the
2617  * nfsnode. There is the potential for another process on a different client
2618  * to create the same funny name between the nfs_lookitup() fails and the
2619  * nfs_rename() completes, but...
2620  */
2621 static int
2622 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2623 {
2624 	struct sillyrename *sp;
2625 	struct nfsnode *np;
2626 	int error;
2627 	short pid;
2628 	unsigned int lticks;
2629 
2630 	cache_purge(dvp);
2631 	np = VTONFS(vp);
2632 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2633 	sp = malloc(sizeof (struct sillyrename),
2634 	    M_NEWNFSREQ, M_WAITOK);
2635 	sp->s_cred = crhold(cnp->cn_cred);
2636 	sp->s_dvp = dvp;
2637 	VREF(dvp);
2638 
2639 	/*
2640 	 * Fudge together a funny name.
2641 	 * Changing the format of the funny name to accommodate more
2642 	 * sillynames per directory.
2643 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2644 	 * CPU ticks since boot.
2645 	 */
2646 	pid = curthread->td_proc->p_pid;
2647 	lticks = (unsigned int)ticks;
2648 	for ( ; ; ) {
2649 		sp->s_namlen = sprintf(sp->s_name,
2650 				       ".nfs.%08x.%04x4.4", lticks,
2651 				       pid);
2652 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2653 				 curthread, NULL))
2654 			break;
2655 		lticks++;
2656 	}
2657 	error = nfs_renameit(dvp, vp, cnp, sp);
2658 	if (error)
2659 		goto bad;
2660 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2661 		curthread, &np);
2662 	np->n_sillyrename = sp;
2663 	return (0);
2664 bad:
2665 	vrele(sp->s_dvp);
2666 	crfree(sp->s_cred);
2667 	free(sp, M_NEWNFSREQ);
2668 	return (error);
2669 }
2670 
2671 /*
2672  * Look up a file name and optionally either update the file handle or
2673  * allocate an nfsnode, depending on the value of npp.
2674  * npp == NULL	--> just do the lookup
2675  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2676  *			handled too
2677  * *npp != NULL --> update the file handle in the vnode
2678  */
2679 static int
2680 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2681     struct thread *td, struct nfsnode **npp)
2682 {
2683 	struct vnode *newvp = NULL, *vp;
2684 	struct nfsnode *np, *dnp = VTONFS(dvp);
2685 	struct nfsfh *nfhp, *onfhp;
2686 	struct nfsvattr nfsva, dnfsva;
2687 	struct componentname cn;
2688 	int error = 0, attrflag, dattrflag;
2689 	u_int hash;
2690 	struct timespec ts;
2691 
2692 	nanouptime(&ts);
2693 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2694 	    &nfhp, &attrflag, &dattrflag, 0);
2695 	if (dattrflag)
2696 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, 0, 1);
2697 	if (npp && !error) {
2698 		if (*npp != NULL) {
2699 		    np = *npp;
2700 		    vp = NFSTOV(np);
2701 		    /*
2702 		     * For NFSv4, check to see if it is the same name and
2703 		     * replace the name, if it is different.
2704 		     */
2705 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2706 			(np->n_v4->n4_namelen != len ||
2707 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2708 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2709 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2710 			 dnp->n_fhp->nfh_len))) {
2711 			    free(np->n_v4, M_NFSV4NODE);
2712 			    np->n_v4 = malloc(
2713 				sizeof (struct nfsv4node) +
2714 				dnp->n_fhp->nfh_len + len - 1,
2715 				M_NFSV4NODE, M_WAITOK);
2716 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2717 			    np->n_v4->n4_namelen = len;
2718 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2719 				dnp->n_fhp->nfh_len);
2720 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2721 		    }
2722 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2723 			FNV1_32_INIT);
2724 		    onfhp = np->n_fhp;
2725 		    /*
2726 		     * Rehash node for new file handle.
2727 		     */
2728 		    vfs_hash_rehash(vp, hash);
2729 		    np->n_fhp = nfhp;
2730 		    if (onfhp != NULL)
2731 			free(onfhp, M_NFSFH);
2732 		    newvp = NFSTOV(np);
2733 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2734 		    free(nfhp, M_NFSFH);
2735 		    VREF(dvp);
2736 		    newvp = dvp;
2737 		} else {
2738 		    cn.cn_nameptr = name;
2739 		    cn.cn_namelen = len;
2740 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2741 			&np, LK_EXCLUSIVE);
2742 		    if (error)
2743 			return (error);
2744 		    newvp = NFSTOV(np);
2745 		    /*
2746 		     * If n_localmodtime >= time before RPC, then
2747 		     * a file modification operation, such as
2748 		     * VOP_SETATTR() of size, has occurred while
2749 		     * the Lookup RPC and acquisition of the vnode
2750 		     * happened.  As such, the attributes might
2751 		     * be stale, with possibly an incorrect size.
2752 		     */
2753 		    NFSLOCKNODE(np);
2754 		    if (timespecisset(&np->n_localmodtime) &&
2755 			timespeccmp(&np->n_localmodtime, &ts, >=)) {
2756 			NFSCL_DEBUG(4, "nfs_lookitup: localmod "
2757 			    "stale attributes\n");
2758 			attrflag = 0;
2759 		    }
2760 		    NFSUNLOCKNODE(np);
2761 		}
2762 		if (!attrflag && *npp == NULL) {
2763 			if (newvp == dvp)
2764 				vrele(newvp);
2765 			else
2766 				vput(newvp);
2767 			return (ENOENT);
2768 		}
2769 		if (attrflag)
2770 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, 0, 1);
2771 	}
2772 	if (npp && *npp == NULL) {
2773 		if (error) {
2774 			if (newvp) {
2775 				if (newvp == dvp)
2776 					vrele(newvp);
2777 				else
2778 					vput(newvp);
2779 			}
2780 		} else
2781 			*npp = np;
2782 	}
2783 	if (error && NFS_ISV4(dvp))
2784 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2785 	return (error);
2786 }
2787 
2788 /*
2789  * Nfs Version 3 and 4 commit rpc
2790  */
2791 int
2792 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2793    struct thread *td)
2794 {
2795 	struct nfsvattr nfsva;
2796 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2797 	struct nfsnode *np;
2798 	struct uio uio;
2799 	int error, attrflag;
2800 
2801 	np = VTONFS(vp);
2802 	error = EIO;
2803 	attrflag = 0;
2804 	if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2805 		uio.uio_offset = offset;
2806 		uio.uio_resid = cnt;
2807 		error = nfscl_doiods(vp, &uio, NULL, NULL,
2808 		    NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2809 		if (error != 0) {
2810 			NFSLOCKNODE(np);
2811 			np->n_flag &= ~NDSCOMMIT;
2812 			NFSUNLOCKNODE(np);
2813 		}
2814 	}
2815 	if (error != 0) {
2816 		mtx_lock(&nmp->nm_mtx);
2817 		if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2818 			mtx_unlock(&nmp->nm_mtx);
2819 			return (0);
2820 		}
2821 		mtx_unlock(&nmp->nm_mtx);
2822 		error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2823 		    &attrflag);
2824 	}
2825 	if (attrflag != 0)
2826 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
2827 	if (error != 0 && NFS_ISV4(vp))
2828 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2829 	return (error);
2830 }
2831 
2832 /*
2833  * Strategy routine.
2834  * For async requests when nfsiod(s) are running, queue the request by
2835  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2836  * request.
2837  */
2838 static int
2839 nfs_strategy(struct vop_strategy_args *ap)
2840 {
2841 	struct buf *bp;
2842 	struct vnode *vp;
2843 	struct ucred *cr;
2844 
2845 	bp = ap->a_bp;
2846 	vp = ap->a_vp;
2847 	KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2848 	KASSERT(!(bp->b_flags & B_DONE),
2849 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2850 
2851 	if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2852 		bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2853 		    DEV_BSIZE);
2854 	if (bp->b_iocmd == BIO_READ)
2855 		cr = bp->b_rcred;
2856 	else
2857 		cr = bp->b_wcred;
2858 
2859 	/*
2860 	 * If the op is asynchronous and an i/o daemon is waiting
2861 	 * queue the request, wake it up and wait for completion
2862 	 * otherwise just do it ourselves.
2863 	 */
2864 	if ((bp->b_flags & B_ASYNC) == 0 ||
2865 	    ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2866 		(void) ncl_doio(vp, bp, cr, curthread, 1);
2867 	return (0);
2868 }
2869 
2870 /*
2871  * fsync vnode op. Just call ncl_flush() with commit == 1.
2872  */
2873 /* ARGSUSED */
2874 static int
2875 nfs_fsync(struct vop_fsync_args *ap)
2876 {
2877 
2878 	if (ap->a_vp->v_type != VREG) {
2879 		/*
2880 		 * For NFS, metadata is changed synchronously on the server,
2881 		 * so there is nothing to flush. Also, ncl_flush() clears
2882 		 * the NMODIFIED flag and that shouldn't be done here for
2883 		 * directories.
2884 		 */
2885 		return (0);
2886 	}
2887 	return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2888 }
2889 
2890 /*
2891  * Flush all the blocks associated with a vnode.
2892  * 	Walk through the buffer pool and push any dirty pages
2893  *	associated with the vnode.
2894  * If the called_from_renewthread argument is TRUE, it has been called
2895  * from the NFSv4 renew thread and, as such, cannot block indefinitely
2896  * waiting for a buffer write to complete.
2897  */
2898 int
2899 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2900     int commit, int called_from_renewthread)
2901 {
2902 	struct nfsnode *np = VTONFS(vp);
2903 	struct buf *bp;
2904 	int i;
2905 	struct buf *nbp;
2906 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2907 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2908 	int passone = 1, trycnt = 0;
2909 	u_quad_t off, endoff, toff;
2910 	struct ucred* wcred = NULL;
2911 	struct buf **bvec = NULL;
2912 	struct bufobj *bo;
2913 #ifndef NFS_COMMITBVECSIZ
2914 #define	NFS_COMMITBVECSIZ	20
2915 #endif
2916 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2917 	u_int bvecsize = 0, bveccount;
2918 	struct timespec ts;
2919 
2920 	if (called_from_renewthread != 0)
2921 		slptimeo = hz;
2922 	if (nmp->nm_flag & NFSMNT_INT)
2923 		slpflag = PCATCH;
2924 	if (!commit)
2925 		passone = 0;
2926 	bo = &vp->v_bufobj;
2927 	/*
2928 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2929 	 * server, but has not been committed to stable storage on the server
2930 	 * yet. On the first pass, the byte range is worked out and the commit
2931 	 * rpc is done. On the second pass, bwrite() is called to do the
2932 	 * job.
2933 	 */
2934 again:
2935 	off = (u_quad_t)-1;
2936 	endoff = 0;
2937 	bvecpos = 0;
2938 	if (NFS_ISV34(vp) && commit) {
2939 		if (bvec != NULL && bvec != bvec_on_stack)
2940 			free(bvec, M_TEMP);
2941 		/*
2942 		 * Count up how many buffers waiting for a commit.
2943 		 */
2944 		bveccount = 0;
2945 		BO_LOCK(bo);
2946 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2947 			if (!BUF_ISLOCKED(bp) &&
2948 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2949 				== (B_DELWRI | B_NEEDCOMMIT))
2950 				bveccount++;
2951 		}
2952 		/*
2953 		 * Allocate space to remember the list of bufs to commit.  It is
2954 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2955 		 * If we can't get memory (for whatever reason), we will end up
2956 		 * committing the buffers one-by-one in the loop below.
2957 		 */
2958 		if (bveccount > NFS_COMMITBVECSIZ) {
2959 			/*
2960 			 * Release the vnode interlock to avoid a lock
2961 			 * order reversal.
2962 			 */
2963 			BO_UNLOCK(bo);
2964 			bvec = (struct buf **)
2965 				malloc(bveccount * sizeof(struct buf *),
2966 				       M_TEMP, M_NOWAIT);
2967 			BO_LOCK(bo);
2968 			if (bvec == NULL) {
2969 				bvec = bvec_on_stack;
2970 				bvecsize = NFS_COMMITBVECSIZ;
2971 			} else
2972 				bvecsize = bveccount;
2973 		} else {
2974 			bvec = bvec_on_stack;
2975 			bvecsize = NFS_COMMITBVECSIZ;
2976 		}
2977 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2978 			if (bvecpos >= bvecsize)
2979 				break;
2980 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2981 				nbp = TAILQ_NEXT(bp, b_bobufs);
2982 				continue;
2983 			}
2984 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2985 			    (B_DELWRI | B_NEEDCOMMIT)) {
2986 				BUF_UNLOCK(bp);
2987 				nbp = TAILQ_NEXT(bp, b_bobufs);
2988 				continue;
2989 			}
2990 			BO_UNLOCK(bo);
2991 			bremfree(bp);
2992 			/*
2993 			 * Work out if all buffers are using the same cred
2994 			 * so we can deal with them all with one commit.
2995 			 *
2996 			 * NOTE: we are not clearing B_DONE here, so we have
2997 			 * to do it later on in this routine if we intend to
2998 			 * initiate I/O on the bp.
2999 			 *
3000 			 * Note: to avoid loopback deadlocks, we do not
3001 			 * assign b_runningbufspace.
3002 			 */
3003 			if (wcred == NULL)
3004 				wcred = bp->b_wcred;
3005 			else if (wcred != bp->b_wcred)
3006 				wcred = NOCRED;
3007 			vfs_busy_pages(bp, 0);
3008 
3009 			BO_LOCK(bo);
3010 			/*
3011 			 * bp is protected by being locked, but nbp is not
3012 			 * and vfs_busy_pages() may sleep.  We have to
3013 			 * recalculate nbp.
3014 			 */
3015 			nbp = TAILQ_NEXT(bp, b_bobufs);
3016 
3017 			/*
3018 			 * A list of these buffers is kept so that the
3019 			 * second loop knows which buffers have actually
3020 			 * been committed. This is necessary, since there
3021 			 * may be a race between the commit rpc and new
3022 			 * uncommitted writes on the file.
3023 			 */
3024 			bvec[bvecpos++] = bp;
3025 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3026 				bp->b_dirtyoff;
3027 			if (toff < off)
3028 				off = toff;
3029 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3030 			if (toff > endoff)
3031 				endoff = toff;
3032 		}
3033 		BO_UNLOCK(bo);
3034 	}
3035 	if (bvecpos > 0) {
3036 		/*
3037 		 * Commit data on the server, as required.
3038 		 * If all bufs are using the same wcred, then use that with
3039 		 * one call for all of them, otherwise commit each one
3040 		 * separately.
3041 		 */
3042 		if (wcred != NOCRED)
3043 			retv = ncl_commit(vp, off, (int)(endoff - off),
3044 					  wcred, td);
3045 		else {
3046 			retv = 0;
3047 			for (i = 0; i < bvecpos; i++) {
3048 				off_t off, size;
3049 				bp = bvec[i];
3050 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3051 					bp->b_dirtyoff;
3052 				size = (u_quad_t)(bp->b_dirtyend
3053 						  - bp->b_dirtyoff);
3054 				retv = ncl_commit(vp, off, (int)size,
3055 						  bp->b_wcred, td);
3056 				if (retv) break;
3057 			}
3058 		}
3059 
3060 		if (retv == NFSERR_STALEWRITEVERF)
3061 			ncl_clearcommit(vp->v_mount);
3062 
3063 		/*
3064 		 * Now, either mark the blocks I/O done or mark the
3065 		 * blocks dirty, depending on whether the commit
3066 		 * succeeded.
3067 		 */
3068 		for (i = 0; i < bvecpos; i++) {
3069 			bp = bvec[i];
3070 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3071 			if (!NFSCL_FORCEDISM(vp->v_mount) && retv) {
3072 				/*
3073 				 * Error, leave B_DELWRI intact
3074 				 */
3075 				vfs_unbusy_pages(bp);
3076 				brelse(bp);
3077 			} else {
3078 				/*
3079 				 * Success, remove B_DELWRI ( bundirty() ).
3080 				 *
3081 				 * b_dirtyoff/b_dirtyend seem to be NFS
3082 				 * specific.  We should probably move that
3083 				 * into bundirty(). XXX
3084 				 */
3085 				bufobj_wref(bo);
3086 				bp->b_flags |= B_ASYNC;
3087 				bundirty(bp);
3088 				bp->b_flags &= ~B_DONE;
3089 				bp->b_ioflags &= ~BIO_ERROR;
3090 				bp->b_dirtyoff = bp->b_dirtyend = 0;
3091 				bufdone(bp);
3092 			}
3093 		}
3094 	}
3095 
3096 	/*
3097 	 * Start/do any write(s) that are required.
3098 	 */
3099 loop:
3100 	BO_LOCK(bo);
3101 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3102 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3103 			if (waitfor != MNT_WAIT || passone)
3104 				continue;
3105 
3106 			error = BUF_TIMELOCK(bp,
3107 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3108 			    BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
3109 			if (error == 0) {
3110 				BUF_UNLOCK(bp);
3111 				goto loop;
3112 			}
3113 			if (error == ENOLCK) {
3114 				error = 0;
3115 				goto loop;
3116 			}
3117 			if (called_from_renewthread != 0) {
3118 				/*
3119 				 * Return EIO so the flush will be retried
3120 				 * later.
3121 				 */
3122 				error = EIO;
3123 				goto done;
3124 			}
3125 			if (newnfs_sigintr(nmp, td)) {
3126 				error = EINTR;
3127 				goto done;
3128 			}
3129 			if (slpflag == PCATCH) {
3130 				slpflag = 0;
3131 				slptimeo = 2 * hz;
3132 			}
3133 			goto loop;
3134 		}
3135 		if ((bp->b_flags & B_DELWRI) == 0)
3136 			panic("nfs_fsync: not dirty");
3137 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3138 			BUF_UNLOCK(bp);
3139 			continue;
3140 		}
3141 		BO_UNLOCK(bo);
3142 		bremfree(bp);
3143 		bp->b_flags |= B_ASYNC;
3144 		bwrite(bp);
3145 		if (newnfs_sigintr(nmp, td)) {
3146 			error = EINTR;
3147 			goto done;
3148 		}
3149 		goto loop;
3150 	}
3151 	if (passone) {
3152 		passone = 0;
3153 		BO_UNLOCK(bo);
3154 		goto again;
3155 	}
3156 	if (waitfor == MNT_WAIT) {
3157 		while (bo->bo_numoutput) {
3158 			error = bufobj_wwait(bo, slpflag, slptimeo);
3159 			if (error) {
3160 			    BO_UNLOCK(bo);
3161 			    if (called_from_renewthread != 0) {
3162 				/*
3163 				 * Return EIO so that the flush will be
3164 				 * retried later.
3165 				 */
3166 				error = EIO;
3167 				goto done;
3168 			    }
3169 			    error = newnfs_sigintr(nmp, td);
3170 			    if (error)
3171 				goto done;
3172 			    if (slpflag == PCATCH) {
3173 				slpflag = 0;
3174 				slptimeo = 2 * hz;
3175 			    }
3176 			    BO_LOCK(bo);
3177 			}
3178 		}
3179 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
3180 			BO_UNLOCK(bo);
3181 			goto loop;
3182 		}
3183 		/*
3184 		 * Wait for all the async IO requests to drain
3185 		 */
3186 		BO_UNLOCK(bo);
3187 	} else
3188 		BO_UNLOCK(bo);
3189 	if (NFSHASPNFS(nmp)) {
3190 		nfscl_layoutcommit(vp, td);
3191 		/*
3192 		 * Invalidate the attribute cache, since writes to a DS
3193 		 * won't update the size attribute.
3194 		 */
3195 		NFSLOCKNODE(np);
3196 		np->n_attrstamp = 0;
3197 	} else
3198 		NFSLOCKNODE(np);
3199 	if (np->n_flag & NWRITEERR) {
3200 		error = np->n_error;
3201 		np->n_flag &= ~NWRITEERR;
3202 	}
3203   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
3204 	    bo->bo_numoutput == 0)
3205   		np->n_flag &= ~NMODIFIED;
3206 	NFSUNLOCKNODE(np);
3207 done:
3208 	if (bvec != NULL && bvec != bvec_on_stack)
3209 		free(bvec, M_TEMP);
3210 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
3211 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0)) {
3212 		if (trycnt++ < 5) {
3213 			/* try, try again... */
3214 			passone = 1;
3215 			wcred = NULL;
3216 			bvec = NULL;
3217 			bvecsize = 0;
3218 			goto again;
3219 		}
3220 		vn_printf(vp, "ncl_flush failed");
3221 		error = called_from_renewthread != 0 ? EIO : EBUSY;
3222 	}
3223 	if (error == 0) {
3224 		nanouptime(&ts);
3225 		NFSLOCKNODE(np);
3226 		np->n_localmodtime = ts;
3227 		NFSUNLOCKNODE(np);
3228 	}
3229 	return (error);
3230 }
3231 
3232 /*
3233  * NFS advisory byte-level locks.
3234  */
3235 static int
3236 nfs_advlock(struct vop_advlock_args *ap)
3237 {
3238 	struct vnode *vp = ap->a_vp;
3239 	struct ucred *cred;
3240 	struct nfsnode *np = VTONFS(ap->a_vp);
3241 	struct proc *p = (struct proc *)ap->a_id;
3242 	struct thread *td = curthread;	/* XXX */
3243 	struct vattr va;
3244 	int ret, error;
3245 	u_quad_t size;
3246 	struct nfsmount *nmp;
3247 
3248 	error = NFSVOPLOCK(vp, LK_SHARED);
3249 	if (error != 0)
3250 		return (EBADF);
3251 	nmp = VFSTONFS(vp->v_mount);
3252 	if (!NFS_ISV4(vp) || (nmp->nm_flag & NFSMNT_NOLOCKD) != 0) {
3253 		if ((nmp->nm_flag & NFSMNT_NOLOCKD) != 0) {
3254 			size = np->n_size;
3255 			NFSVOPUNLOCK(vp);
3256 			error = lf_advlock(ap, &(vp->v_lockf), size);
3257 		} else {
3258 			if (nfs_advlock_p != NULL)
3259 				error = nfs_advlock_p(ap);
3260 			else {
3261 				NFSVOPUNLOCK(vp);
3262 				error = ENOLCK;
3263 			}
3264 		}
3265 		if (error == 0 && ap->a_op == F_SETLK) {
3266 			error = NFSVOPLOCK(vp, LK_SHARED);
3267 			if (error == 0) {
3268 				/* Mark that a file lock has been acquired. */
3269 				NFSLOCKNODE(np);
3270 				np->n_flag |= NHASBEENLOCKED;
3271 				NFSUNLOCKNODE(np);
3272 				NFSVOPUNLOCK(vp);
3273 			}
3274 		}
3275 		return (error);
3276 	} else if ((ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3277 		if (vp->v_type != VREG) {
3278 			error = EINVAL;
3279 			goto out;
3280 		}
3281 		if ((ap->a_flags & F_POSIX) != 0)
3282 			cred = p->p_ucred;
3283 		else
3284 			cred = td->td_ucred;
3285 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
3286 		if (VN_IS_DOOMED(vp)) {
3287 			error = EBADF;
3288 			goto out;
3289 		}
3290 
3291 		/*
3292 		 * If this is unlocking a write locked region, flush and
3293 		 * commit them before unlocking. This is required by
3294 		 * RFC3530 Sec. 9.3.2.
3295 		 */
3296 		if (ap->a_op == F_UNLCK &&
3297 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3298 		    ap->a_flags))
3299 			(void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3300 
3301 		/*
3302 		 * Mark NFS node as might have acquired a lock.
3303 		 * This is separate from NHASBEENLOCKED, because it must
3304 		 * be done before the nfsrpc_advlock() call, which might
3305 		 * add a nfscllock structure to the client state.
3306 		 * It is used to check for the case where a nfscllock
3307 		 * state structure cannot exist for the file.
3308 		 * Only done for "oneopenown" NFSv4.1/4.2 mounts.
3309 		 */
3310 		if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) {
3311 			NFSLOCKNODE(np);
3312 			np->n_flag |= NMIGHTBELOCKED;
3313 			NFSUNLOCKNODE(np);
3314 		}
3315 
3316 		/*
3317 		 * Loop around doing the lock op, while a blocking lock
3318 		 * must wait for the lock op to succeed.
3319 		 */
3320 		do {
3321 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3322 			    ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3323 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3324 			    ap->a_op == F_SETLK) {
3325 				NFSVOPUNLOCK(vp);
3326 				error = nfs_catnap(PZERO | PCATCH, ret,
3327 				    "ncladvl");
3328 				if (error)
3329 					return (EINTR);
3330 				NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3331 				if (VN_IS_DOOMED(vp)) {
3332 					error = EBADF;
3333 					goto out;
3334 				}
3335 			}
3336 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3337 		     ap->a_op == F_SETLK);
3338 		if (ret == NFSERR_DENIED) {
3339 			error = EAGAIN;
3340 			goto out;
3341 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3342 			error = ret;
3343 			goto out;
3344 		} else if (ret != 0) {
3345 			error = EACCES;
3346 			goto out;
3347 		}
3348 
3349 		/*
3350 		 * Now, if we just got a lock, invalidate data in the buffer
3351 		 * cache, as required, so that the coherency conforms with
3352 		 * RFC3530 Sec. 9.3.2.
3353 		 */
3354 		if (ap->a_op == F_SETLK) {
3355 			if ((np->n_flag & NMODIFIED) == 0) {
3356 				np->n_attrstamp = 0;
3357 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3358 				ret = VOP_GETATTR(vp, &va, cred);
3359 			}
3360 			if ((np->n_flag & NMODIFIED) || ret ||
3361 			    np->n_change != va.va_filerev) {
3362 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3363 				np->n_attrstamp = 0;
3364 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3365 				ret = VOP_GETATTR(vp, &va, cred);
3366 				if (!ret) {
3367 					np->n_mtime = va.va_mtime;
3368 					np->n_change = va.va_filerev;
3369 				}
3370 			}
3371 			/* Mark that a file lock has been acquired. */
3372 			NFSLOCKNODE(np);
3373 			np->n_flag |= NHASBEENLOCKED;
3374 			NFSUNLOCKNODE(np);
3375 		}
3376 	} else
3377 		error = EOPNOTSUPP;
3378 out:
3379 	NFSVOPUNLOCK(vp);
3380 	return (error);
3381 }
3382 
3383 /*
3384  * NFS advisory byte-level locks.
3385  */
3386 static int
3387 nfs_advlockasync(struct vop_advlockasync_args *ap)
3388 {
3389 	struct vnode *vp = ap->a_vp;
3390 	u_quad_t size;
3391 	int error;
3392 
3393 	error = NFSVOPLOCK(vp, LK_SHARED);
3394 	if (error)
3395 		return (error);
3396 	if (NFS_ISV4(vp)) {
3397 		NFSVOPUNLOCK(vp);
3398 		return (EOPNOTSUPP);
3399 	}
3400 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3401 		size = VTONFS(vp)->n_size;
3402 		NFSVOPUNLOCK(vp);
3403 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
3404 	} else {
3405 		NFSVOPUNLOCK(vp);
3406 		error = EOPNOTSUPP;
3407 	}
3408 	return (error);
3409 }
3410 
3411 /*
3412  * Print out the contents of an nfsnode.
3413  */
3414 static int
3415 nfs_print(struct vop_print_args *ap)
3416 {
3417 	struct vnode *vp = ap->a_vp;
3418 	struct nfsnode *np = VTONFS(vp);
3419 
3420 	printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3421 	    (uintmax_t)np->n_vattr.na_fsid);
3422 	if (vp->v_type == VFIFO)
3423 		fifo_printinfo(vp);
3424 	printf("\n");
3425 	return (0);
3426 }
3427 
3428 /*
3429  * nfs special file access vnode op.
3430  * Essentially just get vattr and then imitate iaccess() since the device is
3431  * local to the client.
3432  */
3433 static int
3434 nfsspec_access(struct vop_access_args *ap)
3435 {
3436 	struct vattr *vap;
3437 	struct ucred *cred = ap->a_cred;
3438 	struct vnode *vp = ap->a_vp;
3439 	accmode_t accmode = ap->a_accmode;
3440 	struct vattr vattr;
3441 	int error;
3442 
3443 	/*
3444 	 * Disallow write attempts on filesystems mounted read-only;
3445 	 * unless the file is a socket, fifo, or a block or character
3446 	 * device resident on the filesystem.
3447 	 */
3448 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3449 		switch (vp->v_type) {
3450 		case VREG:
3451 		case VDIR:
3452 		case VLNK:
3453 			return (EROFS);
3454 		default:
3455 			break;
3456 		}
3457 	}
3458 	vap = &vattr;
3459 	error = VOP_GETATTR(vp, vap, cred);
3460 	if (error)
3461 		goto out;
3462 	error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3463 	    accmode, cred);
3464 out:
3465 	return error;
3466 }
3467 
3468 /*
3469  * Read wrapper for fifos.
3470  */
3471 static int
3472 nfsfifo_read(struct vop_read_args *ap)
3473 {
3474 	struct nfsnode *np = VTONFS(ap->a_vp);
3475 	int error;
3476 
3477 	/*
3478 	 * Set access flag.
3479 	 */
3480 	NFSLOCKNODE(np);
3481 	np->n_flag |= NACC;
3482 	vfs_timestamp(&np->n_atim);
3483 	NFSUNLOCKNODE(np);
3484 	error = fifo_specops.vop_read(ap);
3485 	return error;
3486 }
3487 
3488 /*
3489  * Write wrapper for fifos.
3490  */
3491 static int
3492 nfsfifo_write(struct vop_write_args *ap)
3493 {
3494 	struct nfsnode *np = VTONFS(ap->a_vp);
3495 
3496 	/*
3497 	 * Set update flag.
3498 	 */
3499 	NFSLOCKNODE(np);
3500 	np->n_flag |= NUPD;
3501 	vfs_timestamp(&np->n_mtim);
3502 	NFSUNLOCKNODE(np);
3503 	return(fifo_specops.vop_write(ap));
3504 }
3505 
3506 /*
3507  * Close wrapper for fifos.
3508  *
3509  * Update the times on the nfsnode then do fifo close.
3510  */
3511 static int
3512 nfsfifo_close(struct vop_close_args *ap)
3513 {
3514 	struct vnode *vp = ap->a_vp;
3515 	struct nfsnode *np = VTONFS(vp);
3516 	struct vattr vattr;
3517 	struct timespec ts;
3518 
3519 	NFSLOCKNODE(np);
3520 	if (np->n_flag & (NACC | NUPD)) {
3521 		vfs_timestamp(&ts);
3522 		if (np->n_flag & NACC)
3523 			np->n_atim = ts;
3524 		if (np->n_flag & NUPD)
3525 			np->n_mtim = ts;
3526 		np->n_flag |= NCHG;
3527 		if (vrefcnt(vp) == 1 &&
3528 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3529 			VATTR_NULL(&vattr);
3530 			if (np->n_flag & NACC)
3531 				vattr.va_atime = np->n_atim;
3532 			if (np->n_flag & NUPD)
3533 				vattr.va_mtime = np->n_mtim;
3534 			NFSUNLOCKNODE(np);
3535 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3536 			goto out;
3537 		}
3538 	}
3539 	NFSUNLOCKNODE(np);
3540 out:
3541 	return (fifo_specops.vop_close(ap));
3542 }
3543 
3544 static int
3545 nfs_getacl(struct vop_getacl_args *ap)
3546 {
3547 	int error;
3548 
3549 	if (ap->a_type != ACL_TYPE_NFS4)
3550 		return (EOPNOTSUPP);
3551 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp);
3552 	if (error > NFSERR_STALE) {
3553 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3554 		error = EPERM;
3555 	}
3556 	return (error);
3557 }
3558 
3559 static int
3560 nfs_setacl(struct vop_setacl_args *ap)
3561 {
3562 	int error;
3563 
3564 	if (ap->a_type != ACL_TYPE_NFS4)
3565 		return (EOPNOTSUPP);
3566 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp);
3567 	if (error > NFSERR_STALE) {
3568 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3569 		error = EPERM;
3570 	}
3571 	return (error);
3572 }
3573 
3574 /*
3575  * VOP_ADVISE for NFS.
3576  * Just return 0 for any errors, since it is just a hint.
3577  */
3578 static int
3579 nfs_advise(struct vop_advise_args *ap)
3580 {
3581 	struct thread *td = curthread;
3582 	struct nfsmount *nmp;
3583 	uint64_t len;
3584 	int error;
3585 
3586 	/*
3587 	 * First do vop_stdadvise() to handle the buffer cache.
3588 	 */
3589 	error = vop_stdadvise(ap);
3590 	if (error != 0)
3591 		return (error);
3592 	if (ap->a_start < 0 || ap->a_end < 0)
3593 		return (0);
3594 	if (ap->a_end == OFF_MAX)
3595 		len = 0;
3596 	else if (ap->a_end < ap->a_start)
3597 		return (0);
3598 	else
3599 		len = ap->a_end - ap->a_start + 1;
3600 	nmp = VFSTONFS(ap->a_vp->v_mount);
3601 	mtx_lock(&nmp->nm_mtx);
3602 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3603 	    (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) ==
3604 	    0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) {
3605 		mtx_unlock(&nmp->nm_mtx);
3606 		return (0);
3607 	}
3608 	mtx_unlock(&nmp->nm_mtx);
3609 	error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice,
3610 	    td->td_ucred, td);
3611 	if (error == NFSERR_NOTSUPP) {
3612 		mtx_lock(&nmp->nm_mtx);
3613 		nmp->nm_privflag |= NFSMNTP_NOADVISE;
3614 		mtx_unlock(&nmp->nm_mtx);
3615 	}
3616 	return (0);
3617 }
3618 
3619 /*
3620  * nfs allocate call
3621  */
3622 static int
3623 nfs_allocate(struct vop_allocate_args *ap)
3624 {
3625 	struct vnode *vp = ap->a_vp;
3626 	struct thread *td = curthread;
3627 	struct nfsvattr nfsva;
3628 	struct nfsmount *nmp;
3629 	struct nfsnode *np;
3630 	off_t alen;
3631 	int attrflag, error, ret;
3632 	struct timespec ts;
3633 	struct uio io;
3634 
3635 	attrflag = 0;
3636 	nmp = VFSTONFS(vp->v_mount);
3637 	np = VTONFS(vp);
3638 	mtx_lock(&nmp->nm_mtx);
3639 	if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3640 	    (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) {
3641 		mtx_unlock(&nmp->nm_mtx);
3642 		alen = *ap->a_len;
3643 		if ((uint64_t)alen > nfs_maxalloclen)
3644 			alen = nfs_maxalloclen;
3645 
3646 		/* Check the file size limit. */
3647 		io.uio_offset = *ap->a_offset;
3648 		io.uio_resid = alen;
3649 		error = vn_rlimit_fsize(vp, &io, td);
3650 
3651 		/*
3652 		 * Flush first to ensure that the allocate adds to the
3653 		 * file's allocation on the server.
3654 		 */
3655 		if (error == 0) {
3656 			vnode_pager_clean_sync(vp);
3657 			error = ncl_flush(vp, MNT_WAIT, td, 1, 0);
3658 		}
3659 		if (error == 0)
3660 			error = nfsrpc_allocate(vp, *ap->a_offset, alen,
3661 			    &nfsva, &attrflag, ap->a_cred, td);
3662 		if (error == 0) {
3663 			*ap->a_offset += alen;
3664 			*ap->a_len -= alen;
3665 			nanouptime(&ts);
3666 			NFSLOCKNODE(np);
3667 			np->n_localmodtime = ts;
3668 			NFSUNLOCKNODE(np);
3669 		} else if (error == NFSERR_NOTSUPP) {
3670 			mtx_lock(&nmp->nm_mtx);
3671 			nmp->nm_privflag |= NFSMNTP_NOALLOCATE;
3672 			mtx_unlock(&nmp->nm_mtx);
3673 			error = EINVAL;
3674 		}
3675 	} else {
3676 		mtx_unlock(&nmp->nm_mtx);
3677 		error = EINVAL;
3678 	}
3679 	if (attrflag != 0) {
3680 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
3681 		if (error == 0 && ret != 0)
3682 			error = ret;
3683 	}
3684 	if (error != 0)
3685 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3686 	return (error);
3687 }
3688 
3689 /*
3690  * nfs deallocate call
3691  */
3692 static int
3693 nfs_deallocate(struct vop_deallocate_args *ap)
3694 {
3695 	struct vnode *vp = ap->a_vp;
3696 	struct thread *td = curthread;
3697 	struct nfsvattr nfsva;
3698 	struct nfsmount *nmp;
3699 	struct nfsnode *np;
3700 	off_t tlen, mlen;
3701 	int attrflag, error, ret;
3702 	bool clipped;
3703 	struct timespec ts;
3704 
3705 	error = 0;
3706 	attrflag = 0;
3707 	nmp = VFSTONFS(vp->v_mount);
3708 	np = VTONFS(vp);
3709 	mtx_lock(&nmp->nm_mtx);
3710 	if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3711 	    (nmp->nm_privflag & NFSMNTP_NODEALLOCATE) == 0) {
3712 		mtx_unlock(&nmp->nm_mtx);
3713 		tlen = omin(OFF_MAX - *ap->a_offset, *ap->a_len);
3714 		NFSCL_DEBUG(4, "dealloc: off=%jd len=%jd maxfilesize=%ju\n",
3715 		    (intmax_t)*ap->a_offset, (intmax_t)tlen,
3716 		    (uintmax_t)nmp->nm_maxfilesize);
3717 		if ((uint64_t)*ap->a_offset >= nmp->nm_maxfilesize) {
3718 			/* Avoid EFBIG error return from the NFSv4.2 server. */
3719 			*ap->a_len = 0;
3720 			return (0);
3721 		}
3722 		clipped = false;
3723 		if ((uint64_t)*ap->a_offset + tlen > nmp->nm_maxfilesize)
3724 			tlen = nmp->nm_maxfilesize - *ap->a_offset;
3725 		if ((uint64_t)*ap->a_offset < np->n_size) {
3726 			/* Limit the len to nfs_maxalloclen before EOF. */
3727 			mlen = omin((off_t)np->n_size - *ap->a_offset, tlen);
3728 			if ((uint64_t)mlen > nfs_maxalloclen) {
3729 				NFSCL_DEBUG(4, "dealloc: tlen maxalloclen\n");
3730 				tlen = nfs_maxalloclen;
3731 				clipped = true;
3732 			}
3733 		}
3734 		if (error == 0)
3735 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
3736 		if (error == 0) {
3737 			vnode_pager_purge_range(vp, *ap->a_offset,
3738 			    *ap->a_offset + tlen);
3739 			error = nfsrpc_deallocate(vp, *ap->a_offset, tlen,
3740 			    &nfsva, &attrflag, ap->a_cred, td);
3741 			NFSCL_DEBUG(4, "dealloc: rpc=%d\n", error);
3742 		}
3743 		if (error == 0) {
3744 			NFSCL_DEBUG(4, "dealloc: attrflag=%d na_size=%ju\n",
3745 			    attrflag, (uintmax_t)nfsva.na_size);
3746 			nanouptime(&ts);
3747 			NFSLOCKNODE(np);
3748 			np->n_localmodtime = ts;
3749 			NFSUNLOCKNODE(np);
3750 			if (attrflag != 0) {
3751 				if ((uint64_t)*ap->a_offset < nfsva.na_size)
3752 					*ap->a_offset += omin((off_t)
3753 					    nfsva.na_size - *ap->a_offset,
3754 					    tlen);
3755 			}
3756 			if (clipped && tlen < *ap->a_len)
3757 				*ap->a_len -= tlen;
3758 			else
3759 				*ap->a_len = 0;
3760 		} else if (error == NFSERR_NOTSUPP) {
3761 			mtx_lock(&nmp->nm_mtx);
3762 			nmp->nm_privflag |= NFSMNTP_NODEALLOCATE;
3763 			mtx_unlock(&nmp->nm_mtx);
3764 		}
3765 	} else {
3766 		mtx_unlock(&nmp->nm_mtx);
3767 		error = EIO;
3768 	}
3769 	/*
3770 	 * If the NFS server cannot perform the Deallocate operation, just call
3771 	 * vop_stddeallocate() to perform it.
3772 	 */
3773 	if (error != 0 && error != NFSERR_FBIG && error != NFSERR_INVAL) {
3774 		error = vop_stddeallocate(ap);
3775 		NFSCL_DEBUG(4, "dealloc: stddeallocate=%d\n", error);
3776 	}
3777 	if (attrflag != 0) {
3778 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
3779 		if (error == 0 && ret != 0)
3780 			error = ret;
3781 	}
3782 	if (error != 0)
3783 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3784 	return (error);
3785 }
3786 
3787 /*
3788  * nfs copy_file_range call
3789  */
3790 static int
3791 nfs_copy_file_range(struct vop_copy_file_range_args *ap)
3792 {
3793 	struct vnode *invp = ap->a_invp;
3794 	struct vnode *outvp = ap->a_outvp;
3795 	struct mount *mp;
3796 	vm_object_t invp_obj;
3797 	struct nfsvattr innfsva, outnfsva;
3798 	struct vattr va, *vap;
3799 	struct uio io;
3800 	struct nfsmount *nmp;
3801 	size_t len, len2;
3802 	ssize_t r;
3803 	int error, inattrflag, outattrflag, ret, ret2, invp_lock;
3804 	off_t inoff, outoff;
3805 	bool consecutive, must_commit, tryoutcred;
3806 
3807 	/*
3808 	 * NFSv4.2 Copy is not permitted for infile == outfile.
3809 	 * TODO: copy_file_range() between multiple NFS mountpoints
3810 	 */
3811 	if (invp == outvp || invp->v_mount != outvp->v_mount) {
3812 generic_copy:
3813 		return (ENOSYS);
3814 	}
3815 
3816 	invp_lock = LK_SHARED;
3817 relock:
3818 
3819 	/* Lock both vnodes, avoiding risk of deadlock. */
3820 	do {
3821 		mp = NULL;
3822 		error = vn_start_write(outvp, &mp, V_WAIT);
3823 		if (error == 0) {
3824 			error = vn_lock(outvp, LK_EXCLUSIVE);
3825 			if (error == 0) {
3826 				error = vn_lock(invp, invp_lock | LK_NOWAIT);
3827 				if (error == 0)
3828 					break;
3829 				VOP_UNLOCK(outvp);
3830 				if (mp != NULL)
3831 					vn_finished_write(mp);
3832 				mp = NULL;
3833 				error = vn_lock(invp, invp_lock);
3834 				if (error == 0)
3835 					VOP_UNLOCK(invp);
3836 			}
3837 		}
3838 		if (mp != NULL)
3839 			vn_finished_write(mp);
3840 	} while (error == 0);
3841 	if (error != 0)
3842 		return (error);
3843 
3844 	/*
3845 	 * More reasons to avoid nfs copy: not NFSv4.2, or explicitly
3846 	 * disabled.
3847 	 */
3848 	nmp = VFSTONFS(invp->v_mount);
3849 	mtx_lock(&nmp->nm_mtx);
3850 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3851 	    (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0) {
3852 		mtx_unlock(&nmp->nm_mtx);
3853 		VOP_UNLOCK(invp);
3854 		VOP_UNLOCK(outvp);
3855 		if (mp != NULL)
3856 			vn_finished_write(mp);
3857 		goto generic_copy;
3858 	}
3859 	mtx_unlock(&nmp->nm_mtx);
3860 
3861 	/*
3862 	 * Do the vn_rlimit_fsize() check.  Should this be above the VOP layer?
3863 	 */
3864 	io.uio_offset = *ap->a_outoffp;
3865 	io.uio_resid = *ap->a_lenp;
3866 	error = vn_rlimit_fsizex(outvp, &io, 0, &r, ap->a_fsizetd);
3867 	*ap->a_lenp = io.uio_resid;
3868 	/*
3869 	 * No need to call vn_rlimit_fsizex_res before return, since the uio is
3870 	 * local.
3871 	 */
3872 
3873 	/*
3874 	 * Flush the input file so that the data is up to date before
3875 	 * the copy.  Flush writes for the output file so that they
3876 	 * do not overwrite the data copied to the output file by the Copy.
3877 	 * Set the commit argument for both flushes so that the data is on
3878 	 * stable storage before the Copy RPC.  This is done in case the
3879 	 * server reboots during the Copy and needs to be redone.
3880 	 */
3881 	if (error == 0) {
3882 		invp_obj = invp->v_object;
3883 		if (invp_obj != NULL && vm_object_mightbedirty(invp_obj)) {
3884 			if (invp_lock != LK_EXCLUSIVE) {
3885 				invp_lock = LK_EXCLUSIVE;
3886 				VOP_UNLOCK(invp);
3887 				VOP_UNLOCK(outvp);
3888 				if (mp != NULL)
3889 					vn_finished_write(mp);
3890 				goto relock;
3891 			}
3892 			vnode_pager_clean_sync(invp);
3893 		}
3894 		error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0);
3895 	}
3896 	if (error == 0)
3897 		error = ncl_vinvalbuf(outvp, V_SAVE, curthread, 0);
3898 
3899 	/* Do the actual NFSv4.2 RPC. */
3900 	ret = ret2 = 0;
3901 	len = *ap->a_lenp;
3902 	mtx_lock(&nmp->nm_mtx);
3903 	if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0)
3904 		consecutive = true;
3905 	else
3906 		consecutive = false;
3907 	mtx_unlock(&nmp->nm_mtx);
3908 	inoff = *ap->a_inoffp;
3909 	outoff = *ap->a_outoffp;
3910 	tryoutcred = true;
3911 	must_commit = false;
3912 	if (error == 0) {
3913 		vap = &VTONFS(invp)->n_vattr.na_vattr;
3914 		error = VOP_GETATTR(invp, vap, ap->a_incred);
3915 		if (error == 0) {
3916 			/*
3917 			 * Clip "len" at va_size so that RFC compliant servers
3918 			 * will not reply NFSERR_INVAL.
3919 			 * Setting "len == 0" for the RPC would be preferred,
3920 			 * but some Linux servers do not support that.
3921 			 * If the len is being set to 0, do a Setattr RPC to
3922 			 * set the server's atime.  This behaviour was the
3923 			 * preferred one for the FreeBSD "collective".
3924 			 */
3925 			if (inoff >= vap->va_size) {
3926 				*ap->a_lenp = len = 0;
3927 				if ((nmp->nm_mountp->mnt_flag & MNT_NOATIME) ==
3928 				    0) {
3929 					VATTR_NULL(&va);
3930 					va.va_atime.tv_sec = 0;
3931 					va.va_atime.tv_nsec = 0;
3932 					va.va_vaflags = VA_UTIMES_NULL;
3933 					inattrflag = 0;
3934 					error = nfsrpc_setattr(invp, &va, NULL,
3935 					    ap->a_incred, curthread, &innfsva,
3936 					    &inattrflag);
3937 					if (inattrflag != 0)
3938 						ret = nfscl_loadattrcache(&invp,
3939 						    &innfsva, NULL, 0, 1);
3940 					if (error == 0 && ret != 0)
3941 						error = ret;
3942 				}
3943 			} else if (inoff + len > vap->va_size)
3944 				*ap->a_lenp = len = vap->va_size - inoff;
3945 		} else
3946 			error = 0;
3947 	}
3948 
3949 	/*
3950 	 * len will be set to 0 upon a successful Copy RPC.
3951 	 * As such, this only loops when the Copy RPC needs to be retried.
3952 	 */
3953 	while (len > 0 && error == 0) {
3954 		inattrflag = outattrflag = 0;
3955 		len2 = len;
3956 		if (tryoutcred)
3957 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3958 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3959 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3960 			    ap->a_outcred, consecutive, &must_commit);
3961 		else
3962 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3963 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3964 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3965 			    ap->a_incred, consecutive, &must_commit);
3966 		if (inattrflag != 0)
3967 			ret = nfscl_loadattrcache(&invp, &innfsva, NULL, 0, 1);
3968 		if (outattrflag != 0)
3969 			ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL,
3970 			    1, 1);
3971 		if (error == 0) {
3972 			if (consecutive == false) {
3973 				if (len2 == len) {
3974 					mtx_lock(&nmp->nm_mtx);
3975 					nmp->nm_privflag |=
3976 					    NFSMNTP_NOCONSECUTIVE;
3977 					mtx_unlock(&nmp->nm_mtx);
3978 				} else
3979 					error = NFSERR_OFFLOADNOREQS;
3980 			}
3981 			*ap->a_lenp = len2;
3982 			len = 0;
3983 			if (len2 > 0 && must_commit && error == 0)
3984 				error = ncl_commit(outvp, outoff, *ap->a_lenp,
3985 				    ap->a_outcred, curthread);
3986 			if (error == 0 && ret != 0)
3987 				error = ret;
3988 			if (error == 0 && ret2 != 0)
3989 				error = ret2;
3990 		} else if (error == NFSERR_OFFLOADNOREQS && consecutive) {
3991 			/*
3992 			 * Try consecutive == false, which is ok only if all
3993 			 * bytes are copied.
3994 			 * If only some bytes were copied when consecutive
3995 			 * is false, there is no way to know which bytes
3996 			 * still need to be written.
3997 			 */
3998 			consecutive = false;
3999 			error = 0;
4000 		} else if (error == NFSERR_ACCES && tryoutcred) {
4001 			/* Try again with incred. */
4002 			tryoutcred = false;
4003 			error = 0;
4004 		}
4005 		if (error == NFSERR_STALEWRITEVERF) {
4006 			/*
4007 			 * Server rebooted, so do it all again.
4008 			 */
4009 			*ap->a_inoffp = inoff;
4010 			*ap->a_outoffp = outoff;
4011 			len = *ap->a_lenp;
4012 			must_commit = false;
4013 			error = 0;
4014 		}
4015 	}
4016 	VOP_UNLOCK(invp);
4017 	VOP_UNLOCK(outvp);
4018 	if (mp != NULL)
4019 		vn_finished_write(mp);
4020 	if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS ||
4021 	    error == NFSERR_ACCES) {
4022 		/*
4023 		 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can
4024 		 * use a_incred for the read and a_outcred for the write, so
4025 		 * try this for NFSERR_ACCES failures for the Copy.
4026 		 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can
4027 		 * never succeed, so disable it.
4028 		 */
4029 		if (error != NFSERR_ACCES) {
4030 			/* Can never do Copy on this mount. */
4031 			mtx_lock(&nmp->nm_mtx);
4032 			nmp->nm_privflag |= NFSMNTP_NOCOPY;
4033 			mtx_unlock(&nmp->nm_mtx);
4034 		}
4035 		*ap->a_inoffp = inoff;
4036 		*ap->a_outoffp = outoff;
4037 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
4038 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
4039 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
4040 	} else if (error != 0)
4041 		*ap->a_lenp = 0;
4042 
4043 	if (error != 0)
4044 		error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0);
4045 	return (error);
4046 }
4047 
4048 /*
4049  * nfs ioctl call
4050  */
4051 static int
4052 nfs_ioctl(struct vop_ioctl_args *ap)
4053 {
4054 	struct vnode *vp = ap->a_vp;
4055 	struct nfsvattr nfsva;
4056 	struct nfsmount *nmp;
4057 	int attrflag, content, error, ret;
4058 	bool eof = false;			/* shut up compiler. */
4059 
4060 	/* Do the actual NFSv4.2 RPC. */
4061 	switch (ap->a_command) {
4062 	case FIOSEEKDATA:
4063 		content = NFSV4CONTENT_DATA;
4064 		break;
4065 	case FIOSEEKHOLE:
4066 		content = NFSV4CONTENT_HOLE;
4067 		break;
4068 	default:
4069 		return (ENOTTY);
4070 	}
4071 
4072 	error = vn_lock(vp, LK_EXCLUSIVE);
4073 	if (error != 0)
4074 		return (EBADF);
4075 
4076 	if (vp->v_type != VREG) {
4077 		VOP_UNLOCK(vp);
4078 		return (ENOTTY);
4079 	}
4080 	nmp = VFSTONFS(vp->v_mount);
4081 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) {
4082 		VOP_UNLOCK(vp);
4083 		error = vop_stdioctl(ap);
4084 		return (error);
4085 	}
4086 
4087 	attrflag = 0;
4088 	if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size)
4089 		error = ENXIO;
4090 	else {
4091 		/*
4092 		 * Flush all writes, so that the server is up to date.
4093 		 * Although a Commit is not required, the commit argument
4094 		 * is set so that, for a pNFS File/Flexible File Layout
4095 		 * server, the LayoutCommit will be done to ensure the file
4096 		 * size is up to date on the Metadata Server.
4097 		 */
4098 
4099 		vnode_pager_clean_sync(vp);
4100 		error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0);
4101 		if (error == 0)
4102 			error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof,
4103 			    content, ap->a_cred, &nfsva, &attrflag);
4104 		/* If at eof for FIOSEEKDATA, return ENXIO. */
4105 		if (eof && error == 0 && content == NFSV4CONTENT_DATA)
4106 			error = ENXIO;
4107 	}
4108 	if (attrflag != 0) {
4109 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4110 		if (error == 0 && ret != 0)
4111 			error = ret;
4112 	}
4113 	NFSVOPUNLOCK(vp);
4114 
4115 	if (error != 0)
4116 		error = ENXIO;
4117 	return (error);
4118 }
4119 
4120 /*
4121  * nfs getextattr call
4122  */
4123 static int
4124 nfs_getextattr(struct vop_getextattr_args *ap)
4125 {
4126 	struct vnode *vp = ap->a_vp;
4127 	struct nfsmount *nmp;
4128 	struct ucred *cred;
4129 	struct thread *td = ap->a_td;
4130 	struct nfsvattr nfsva;
4131 	ssize_t len;
4132 	int attrflag, error, ret;
4133 
4134 	nmp = VFSTONFS(vp->v_mount);
4135 	mtx_lock(&nmp->nm_mtx);
4136 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4137 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4138 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4139 		mtx_unlock(&nmp->nm_mtx);
4140 		return (EOPNOTSUPP);
4141 	}
4142 	mtx_unlock(&nmp->nm_mtx);
4143 
4144 	cred = ap->a_cred;
4145 	if (cred == NULL)
4146 		cred = td->td_ucred;
4147 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4148 	attrflag = 0;
4149 	error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva,
4150 	    &attrflag, cred, td);
4151 	if (attrflag != 0) {
4152 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4153 		if (error == 0 && ret != 0)
4154 			error = ret;
4155 	}
4156 	if (error == 0 && ap->a_size != NULL)
4157 		*ap->a_size = len;
4158 
4159 	switch (error) {
4160 	case NFSERR_NOTSUPP:
4161 	case NFSERR_OPILLEGAL:
4162 		mtx_lock(&nmp->nm_mtx);
4163 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4164 		mtx_unlock(&nmp->nm_mtx);
4165 		error = EOPNOTSUPP;
4166 		break;
4167 	case NFSERR_NOXATTR:
4168 	case NFSERR_XATTR2BIG:
4169 		error = ENOATTR;
4170 		break;
4171 	default:
4172 		error = nfscl_maperr(td, error, 0, 0);
4173 		break;
4174 	}
4175 	return (error);
4176 }
4177 
4178 /*
4179  * nfs setextattr call
4180  */
4181 static int
4182 nfs_setextattr(struct vop_setextattr_args *ap)
4183 {
4184 	struct vnode *vp = ap->a_vp;
4185 	struct nfsmount *nmp;
4186 	struct ucred *cred;
4187 	struct thread *td = ap->a_td;
4188 	struct nfsvattr nfsva;
4189 	int attrflag, error, ret;
4190 
4191 	nmp = VFSTONFS(vp->v_mount);
4192 	mtx_lock(&nmp->nm_mtx);
4193 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4194 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4195 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4196 		mtx_unlock(&nmp->nm_mtx);
4197 		return (EOPNOTSUPP);
4198 	}
4199 	mtx_unlock(&nmp->nm_mtx);
4200 
4201 	if (ap->a_uio->uio_resid < 0)
4202 		return (EINVAL);
4203 	cred = ap->a_cred;
4204 	if (cred == NULL)
4205 		cred = td->td_ucred;
4206 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4207 	attrflag = 0;
4208 	error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva,
4209 	    &attrflag, cred, td);
4210 	if (attrflag != 0) {
4211 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4212 		if (error == 0 && ret != 0)
4213 			error = ret;
4214 	}
4215 
4216 	switch (error) {
4217 	case NFSERR_NOTSUPP:
4218 	case NFSERR_OPILLEGAL:
4219 		mtx_lock(&nmp->nm_mtx);
4220 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4221 		mtx_unlock(&nmp->nm_mtx);
4222 		error = EOPNOTSUPP;
4223 		break;
4224 	case NFSERR_NOXATTR:
4225 	case NFSERR_XATTR2BIG:
4226 		error = ENOATTR;
4227 		break;
4228 	default:
4229 		error = nfscl_maperr(td, error, 0, 0);
4230 		break;
4231 	}
4232 	return (error);
4233 }
4234 
4235 /*
4236  * nfs listextattr call
4237  */
4238 static int
4239 nfs_listextattr(struct vop_listextattr_args *ap)
4240 {
4241 	struct vnode *vp = ap->a_vp;
4242 	struct nfsmount *nmp;
4243 	struct ucred *cred;
4244 	struct thread *td = ap->a_td;
4245 	struct nfsvattr nfsva;
4246 	size_t len, len2;
4247 	uint64_t cookie;
4248 	int attrflag, error, ret;
4249 	bool eof;
4250 
4251 	nmp = VFSTONFS(vp->v_mount);
4252 	mtx_lock(&nmp->nm_mtx);
4253 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4254 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4255 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4256 		mtx_unlock(&nmp->nm_mtx);
4257 		return (EOPNOTSUPP);
4258 	}
4259 	mtx_unlock(&nmp->nm_mtx);
4260 
4261 	cred = ap->a_cred;
4262 	if (cred == NULL)
4263 		cred = td->td_ucred;
4264 
4265 	/* Loop around doing List Extended Attribute RPCs. */
4266 	eof = false;
4267 	cookie = 0;
4268 	len2 = 0;
4269 	error = 0;
4270 	while (!eof && error == 0) {
4271 		len = nmp->nm_rsize;
4272 		attrflag = 0;
4273 		error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof,
4274 		    &nfsva, &attrflag, cred, td);
4275 		if (attrflag != 0) {
4276 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4277 			if (error == 0 && ret != 0)
4278 				error = ret;
4279 		}
4280 		if (error == 0) {
4281 			len2 += len;
4282 			if (len2 > SSIZE_MAX)
4283 				error = ENOATTR;
4284 		}
4285 	}
4286 	if (error == 0 && ap->a_size != NULL)
4287 		*ap->a_size = len2;
4288 
4289 	switch (error) {
4290 	case NFSERR_NOTSUPP:
4291 	case NFSERR_OPILLEGAL:
4292 		mtx_lock(&nmp->nm_mtx);
4293 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4294 		mtx_unlock(&nmp->nm_mtx);
4295 		error = EOPNOTSUPP;
4296 		break;
4297 	case NFSERR_NOXATTR:
4298 	case NFSERR_XATTR2BIG:
4299 		error = ENOATTR;
4300 		break;
4301 	default:
4302 		error = nfscl_maperr(td, error, 0, 0);
4303 		break;
4304 	}
4305 	return (error);
4306 }
4307 
4308 /*
4309  * nfs setextattr call
4310  */
4311 static int
4312 nfs_deleteextattr(struct vop_deleteextattr_args *ap)
4313 {
4314 	struct vnode *vp = ap->a_vp;
4315 	struct nfsmount *nmp;
4316 	struct nfsvattr nfsva;
4317 	int attrflag, error, ret;
4318 
4319 	nmp = VFSTONFS(vp->v_mount);
4320 	mtx_lock(&nmp->nm_mtx);
4321 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4322 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4323 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4324 		mtx_unlock(&nmp->nm_mtx);
4325 		return (EOPNOTSUPP);
4326 	}
4327 	mtx_unlock(&nmp->nm_mtx);
4328 
4329 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4330 	attrflag = 0;
4331 	error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred,
4332 	    ap->a_td);
4333 	if (attrflag != 0) {
4334 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4335 		if (error == 0 && ret != 0)
4336 			error = ret;
4337 	}
4338 
4339 	switch (error) {
4340 	case NFSERR_NOTSUPP:
4341 	case NFSERR_OPILLEGAL:
4342 		mtx_lock(&nmp->nm_mtx);
4343 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4344 		mtx_unlock(&nmp->nm_mtx);
4345 		error = EOPNOTSUPP;
4346 		break;
4347 	case NFSERR_NOXATTR:
4348 	case NFSERR_XATTR2BIG:
4349 		error = ENOATTR;
4350 		break;
4351 	default:
4352 		error = nfscl_maperr(ap->a_td, error, 0, 0);
4353 		break;
4354 	}
4355 	return (error);
4356 }
4357 
4358 /*
4359  * Return POSIX pathconf information applicable to nfs filesystems.
4360  */
4361 static int
4362 nfs_pathconf(struct vop_pathconf_args *ap)
4363 {
4364 	struct nfsv3_pathconf pc;
4365 	struct nfsvattr nfsva;
4366 	struct vnode *vp = ap->a_vp;
4367 	struct nfsmount *nmp;
4368 	struct thread *td = curthread;
4369 	off_t off;
4370 	bool eof;
4371 	int attrflag, error;
4372 
4373 	if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
4374 	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
4375 	    ap->a_name == _PC_NO_TRUNC)) ||
4376 	    (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
4377 		/*
4378 		 * Since only the above 4 a_names are returned by the NFSv3
4379 		 * Pathconf RPC, there is no point in doing it for others.
4380 		 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
4381 		 * be used for _PC_NFS4_ACL as well.
4382 		 */
4383 		error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
4384 		    &attrflag);
4385 		if (attrflag != 0)
4386 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, 0, 1);
4387 		if (error != 0)
4388 			return (error);
4389 	} else {
4390 		/*
4391 		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
4392 		 * just fake them.
4393 		 */
4394 		pc.pc_linkmax = NFS_LINK_MAX;
4395 		pc.pc_namemax = NFS_MAXNAMLEN;
4396 		pc.pc_notrunc = 1;
4397 		pc.pc_chownrestricted = 1;
4398 		pc.pc_caseinsensitive = 0;
4399 		pc.pc_casepreserving = 1;
4400 		error = 0;
4401 	}
4402 	switch (ap->a_name) {
4403 	case _PC_LINK_MAX:
4404 #ifdef _LP64
4405 		*ap->a_retval = pc.pc_linkmax;
4406 #else
4407 		*ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
4408 #endif
4409 		break;
4410 	case _PC_NAME_MAX:
4411 		*ap->a_retval = pc.pc_namemax;
4412 		break;
4413 	case _PC_PIPE_BUF:
4414 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
4415 			*ap->a_retval = PIPE_BUF;
4416 		else
4417 			error = EINVAL;
4418 		break;
4419 	case _PC_CHOWN_RESTRICTED:
4420 		*ap->a_retval = pc.pc_chownrestricted;
4421 		break;
4422 	case _PC_NO_TRUNC:
4423 		*ap->a_retval = pc.pc_notrunc;
4424 		break;
4425 	case _PC_ACL_NFS4:
4426 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
4427 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
4428 			*ap->a_retval = 1;
4429 		else
4430 			*ap->a_retval = 0;
4431 		break;
4432 	case _PC_ACL_PATH_MAX:
4433 		if (NFS_ISV4(vp))
4434 			*ap->a_retval = ACL_MAX_ENTRIES;
4435 		else
4436 			*ap->a_retval = 3;
4437 		break;
4438 	case _PC_PRIO_IO:
4439 		*ap->a_retval = 0;
4440 		break;
4441 	case _PC_SYNC_IO:
4442 		*ap->a_retval = 0;
4443 		break;
4444 	case _PC_ALLOC_SIZE_MIN:
4445 		*ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
4446 		break;
4447 	case _PC_FILESIZEBITS:
4448 		if (NFS_ISV34(vp))
4449 			*ap->a_retval = 64;
4450 		else
4451 			*ap->a_retval = 32;
4452 		break;
4453 	case _PC_REC_INCR_XFER_SIZE:
4454 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4455 		break;
4456 	case _PC_REC_MAX_XFER_SIZE:
4457 		*ap->a_retval = -1; /* means ``unlimited'' */
4458 		break;
4459 	case _PC_REC_MIN_XFER_SIZE:
4460 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4461 		break;
4462 	case _PC_REC_XFER_ALIGN:
4463 		*ap->a_retval = PAGE_SIZE;
4464 		break;
4465 	case _PC_SYMLINK_MAX:
4466 		*ap->a_retval = NFS_MAXPATHLEN;
4467 		break;
4468 	case _PC_MIN_HOLE_SIZE:
4469 		/* Only some NFSv4.2 servers support Seek for Holes. */
4470 		*ap->a_retval = 0;
4471 		nmp = VFSTONFS(vp->v_mount);
4472 		if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) {
4473 			/*
4474 			 * NFSv4.2 doesn't have an attribute for hole size,
4475 			 * so all we can do is see if the Seek operation is
4476 			 * supported and then use f_iosize as a "best guess".
4477 			 */
4478 			mtx_lock(&nmp->nm_mtx);
4479 			if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) {
4480 				mtx_unlock(&nmp->nm_mtx);
4481 				off = 0;
4482 				attrflag = 0;
4483 				error = nfsrpc_seek(vp, &off, &eof,
4484 				    NFSV4CONTENT_HOLE, td->td_ucred, &nfsva,
4485 				    &attrflag);
4486 				if (attrflag != 0)
4487 					(void) nfscl_loadattrcache(&vp, &nfsva,
4488 					    NULL, 0, 1);
4489 				mtx_lock(&nmp->nm_mtx);
4490 				if (error == NFSERR_NOTSUPP)
4491 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED;
4492 				else
4493 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED |
4494 					    NFSMNTP_SEEK;
4495 				error = 0;
4496 			}
4497 			if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0)
4498 				*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4499 			mtx_unlock(&nmp->nm_mtx);
4500 		}
4501 		break;
4502 
4503 	default:
4504 		error = vop_stdpathconf(ap);
4505 		break;
4506 	}
4507 	return (error);
4508 }
4509