xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision 0a10f22a30d61a6f32777a236a82d461129538cc)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	from nfs_vnops.c	8.16 (Berkeley) 5/27/95
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 /*
39  * vnode op calls for Sun NFS version 2, 3 and 4
40  */
41 
42 #include "opt_kdtrace.h"
43 #include "opt_inet.h"
44 
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/systm.h>
48 #include <sys/resourcevar.h>
49 #include <sys/proc.h>
50 #include <sys/mount.h>
51 #include <sys/bio.h>
52 #include <sys/buf.h>
53 #include <sys/jail.h>
54 #include <sys/malloc.h>
55 #include <sys/mbuf.h>
56 #include <sys/namei.h>
57 #include <sys/socket.h>
58 #include <sys/vnode.h>
59 #include <sys/dirent.h>
60 #include <sys/fcntl.h>
61 #include <sys/lockf.h>
62 #include <sys/stat.h>
63 #include <sys/sysctl.h>
64 #include <sys/signalvar.h>
65 
66 #include <vm/vm.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_object.h>
69 
70 #include <fs/nfs/nfsport.h>
71 #include <fs/nfsclient/nfsnode.h>
72 #include <fs/nfsclient/nfsmount.h>
73 #include <fs/nfsclient/nfs.h>
74 #include <fs/nfsclient/nfs_kdtrace.h>
75 
76 #include <net/if.h>
77 #include <netinet/in.h>
78 #include <netinet/in_var.h>
79 
80 #include <nfs/nfs_lock.h>
81 
82 #ifdef KDTRACE_HOOKS
83 #include <sys/dtrace_bsd.h>
84 
85 dtrace_nfsclient_accesscache_flush_probe_func_t
86 		dtrace_nfscl_accesscache_flush_done_probe;
87 uint32_t	nfscl_accesscache_flush_done_id;
88 
89 dtrace_nfsclient_accesscache_get_probe_func_t
90 		dtrace_nfscl_accesscache_get_hit_probe,
91 		dtrace_nfscl_accesscache_get_miss_probe;
92 uint32_t	nfscl_accesscache_get_hit_id;
93 uint32_t	nfscl_accesscache_get_miss_id;
94 
95 dtrace_nfsclient_accesscache_load_probe_func_t
96 		dtrace_nfscl_accesscache_load_done_probe;
97 uint32_t	nfscl_accesscache_load_done_id;
98 #endif /* !KDTRACE_HOOKS */
99 
100 /* Defs */
101 #define	TRUE	1
102 #define	FALSE	0
103 
104 extern struct nfsstats newnfsstats;
105 extern int nfsrv_useacl;
106 extern int nfscl_debuglevel;
107 MALLOC_DECLARE(M_NEWNFSREQ);
108 
109 /*
110  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
111  * calls are not in getblk() and brelse() so that they would not be necessary
112  * here.
113  */
114 #ifndef B_VMIO
115 #define	vfs_busy_pages(bp, f)
116 #endif
117 
118 static vop_read_t	nfsfifo_read;
119 static vop_write_t	nfsfifo_write;
120 static vop_close_t	nfsfifo_close;
121 static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
122 		    struct thread *);
123 static vop_lookup_t	nfs_lookup;
124 static vop_create_t	nfs_create;
125 static vop_mknod_t	nfs_mknod;
126 static vop_open_t	nfs_open;
127 static vop_pathconf_t	nfs_pathconf;
128 static vop_close_t	nfs_close;
129 static vop_access_t	nfs_access;
130 static vop_getattr_t	nfs_getattr;
131 static vop_setattr_t	nfs_setattr;
132 static vop_read_t	nfs_read;
133 static vop_fsync_t	nfs_fsync;
134 static vop_remove_t	nfs_remove;
135 static vop_link_t	nfs_link;
136 static vop_rename_t	nfs_rename;
137 static vop_mkdir_t	nfs_mkdir;
138 static vop_rmdir_t	nfs_rmdir;
139 static vop_symlink_t	nfs_symlink;
140 static vop_readdir_t	nfs_readdir;
141 static vop_strategy_t	nfs_strategy;
142 static vop_lock1_t	nfs_lock1;
143 static	int	nfs_lookitup(struct vnode *, char *, int,
144 		    struct ucred *, struct thread *, struct nfsnode **);
145 static	int	nfs_sillyrename(struct vnode *, struct vnode *,
146 		    struct componentname *);
147 static vop_access_t	nfsspec_access;
148 static vop_readlink_t	nfs_readlink;
149 static vop_print_t	nfs_print;
150 static vop_advlock_t	nfs_advlock;
151 static vop_advlockasync_t nfs_advlockasync;
152 static vop_getacl_t nfs_getacl;
153 static vop_setacl_t nfs_setacl;
154 
155 /*
156  * Global vfs data structures for nfs
157  */
158 struct vop_vector newnfs_vnodeops = {
159 	.vop_default =		&default_vnodeops,
160 	.vop_access =		nfs_access,
161 	.vop_advlock =		nfs_advlock,
162 	.vop_advlockasync =	nfs_advlockasync,
163 	.vop_close =		nfs_close,
164 	.vop_create =		nfs_create,
165 	.vop_fsync =		nfs_fsync,
166 	.vop_getattr =		nfs_getattr,
167 	.vop_getpages =		ncl_getpages,
168 	.vop_putpages =		ncl_putpages,
169 	.vop_inactive =		ncl_inactive,
170 	.vop_link =		nfs_link,
171 	.vop_lock1 = 		nfs_lock1,
172 	.vop_lookup =		nfs_lookup,
173 	.vop_mkdir =		nfs_mkdir,
174 	.vop_mknod =		nfs_mknod,
175 	.vop_open =		nfs_open,
176 	.vop_pathconf =		nfs_pathconf,
177 	.vop_print =		nfs_print,
178 	.vop_read =		nfs_read,
179 	.vop_readdir =		nfs_readdir,
180 	.vop_readlink =		nfs_readlink,
181 	.vop_reclaim =		ncl_reclaim,
182 	.vop_remove =		nfs_remove,
183 	.vop_rename =		nfs_rename,
184 	.vop_rmdir =		nfs_rmdir,
185 	.vop_setattr =		nfs_setattr,
186 	.vop_strategy =		nfs_strategy,
187 	.vop_symlink =		nfs_symlink,
188 	.vop_write =		ncl_write,
189 	.vop_getacl =		nfs_getacl,
190 	.vop_setacl =		nfs_setacl,
191 };
192 
193 struct vop_vector newnfs_fifoops = {
194 	.vop_default =		&fifo_specops,
195 	.vop_access =		nfsspec_access,
196 	.vop_close =		nfsfifo_close,
197 	.vop_fsync =		nfs_fsync,
198 	.vop_getattr =		nfs_getattr,
199 	.vop_inactive =		ncl_inactive,
200 	.vop_print =		nfs_print,
201 	.vop_read =		nfsfifo_read,
202 	.vop_reclaim =		ncl_reclaim,
203 	.vop_setattr =		nfs_setattr,
204 	.vop_write =		nfsfifo_write,
205 };
206 
207 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
208     struct componentname *cnp, struct vattr *vap);
209 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
210     int namelen, struct ucred *cred, struct thread *td);
211 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
212     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
213     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
214 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
215     struct componentname *scnp, struct sillyrename *sp);
216 
217 /*
218  * Global variables
219  */
220 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
221 
222 SYSCTL_DECL(_vfs_nfs);
223 
224 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
225 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
226 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
227 
228 static int	nfs_prime_access_cache = 0;
229 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
230 	   &nfs_prime_access_cache, 0,
231 	   "Prime NFS ACCESS cache when fetching attributes");
232 
233 static int	newnfs_commit_on_close = 0;
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
235     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
236 
237 static int	nfs_clean_pages_on_close = 1;
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
239 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
240 
241 int newnfs_directio_enable = 0;
242 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
243 	   &newnfs_directio_enable, 0, "Enable NFS directio");
244 
245 int nfs_keep_dirty_on_error;
246 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
247     &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
248 
249 /*
250  * This sysctl allows other processes to mmap a file that has been opened
251  * O_DIRECT by a process.  In general, having processes mmap the file while
252  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
253  * this by default to prevent DoS attacks - to prevent a malicious user from
254  * opening up files O_DIRECT preventing other users from mmap'ing these
255  * files.  "Protected" environments where stricter consistency guarantees are
256  * required can disable this knob.  The process that opened the file O_DIRECT
257  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
258  * meaningful.
259  */
260 int newnfs_directio_allow_mmap = 1;
261 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
262 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
263 
264 #if 0
265 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
266 	   &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
267 
268 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
269 	   &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
270 #endif
271 
272 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
273 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
274 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
275 
276 /*
277  * SMP Locking Note :
278  * The list of locks after the description of the lock is the ordering
279  * of other locks acquired with the lock held.
280  * np->n_mtx : Protects the fields in the nfsnode.
281        VM Object Lock
282        VI_MTX (acquired indirectly)
283  * nmp->nm_mtx : Protects the fields in the nfsmount.
284        rep->r_mtx
285  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
286  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
287        nmp->nm_mtx
288        rep->r_mtx
289  * rep->r_mtx : Protects the fields in an nfsreq.
290  */
291 
292 static int
293 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
294     struct ucred *cred, u_int32_t *retmode)
295 {
296 	int error = 0, attrflag, i, lrupos;
297 	u_int32_t rmode;
298 	struct nfsnode *np = VTONFS(vp);
299 	struct nfsvattr nfsva;
300 
301 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
302 	    &rmode, NULL);
303 	if (attrflag)
304 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
305 	if (!error) {
306 		lrupos = 0;
307 		mtx_lock(&np->n_mtx);
308 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
309 			if (np->n_accesscache[i].uid == cred->cr_uid) {
310 				np->n_accesscache[i].mode = rmode;
311 				np->n_accesscache[i].stamp = time_second;
312 				break;
313 			}
314 			if (i > 0 && np->n_accesscache[i].stamp <
315 			    np->n_accesscache[lrupos].stamp)
316 				lrupos = i;
317 		}
318 		if (i == NFS_ACCESSCACHESIZE) {
319 			np->n_accesscache[lrupos].uid = cred->cr_uid;
320 			np->n_accesscache[lrupos].mode = rmode;
321 			np->n_accesscache[lrupos].stamp = time_second;
322 		}
323 		mtx_unlock(&np->n_mtx);
324 		if (retmode != NULL)
325 			*retmode = rmode;
326 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
327 	} else if (NFS_ISV4(vp)) {
328 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
329 	}
330 #ifdef KDTRACE_HOOKS
331 	if (error != 0)
332 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
333 		    error);
334 #endif
335 	return (error);
336 }
337 
338 /*
339  * nfs access vnode op.
340  * For nfs version 2, just return ok. File accesses may fail later.
341  * For nfs version 3, use the access rpc to check accessibility. If file modes
342  * are changed on the server, accesses might still fail later.
343  */
344 static int
345 nfs_access(struct vop_access_args *ap)
346 {
347 	struct vnode *vp = ap->a_vp;
348 	int error = 0, i, gotahit;
349 	u_int32_t mode, wmode, rmode;
350 	int v34 = NFS_ISV34(vp);
351 	struct nfsnode *np = VTONFS(vp);
352 
353 	/*
354 	 * Disallow write attempts on filesystems mounted read-only;
355 	 * unless the file is a socket, fifo, or a block or character
356 	 * device resident on the filesystem.
357 	 */
358 	if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
359 	    VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
360 	    VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
361 		switch (vp->v_type) {
362 		case VREG:
363 		case VDIR:
364 		case VLNK:
365 			return (EROFS);
366 		default:
367 			break;
368 		}
369 	}
370 	/*
371 	 * For nfs v3 or v4, check to see if we have done this recently, and if
372 	 * so return our cached result instead of making an ACCESS call.
373 	 * If not, do an access rpc, otherwise you are stuck emulating
374 	 * ufs_access() locally using the vattr. This may not be correct,
375 	 * since the server may apply other access criteria such as
376 	 * client uid-->server uid mapping that we do not know about.
377 	 */
378 	if (v34) {
379 		if (ap->a_accmode & VREAD)
380 			mode = NFSACCESS_READ;
381 		else
382 			mode = 0;
383 		if (vp->v_type != VDIR) {
384 			if (ap->a_accmode & VWRITE)
385 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
386 			if (ap->a_accmode & VAPPEND)
387 				mode |= NFSACCESS_EXTEND;
388 			if (ap->a_accmode & VEXEC)
389 				mode |= NFSACCESS_EXECUTE;
390 			if (ap->a_accmode & VDELETE)
391 				mode |= NFSACCESS_DELETE;
392 		} else {
393 			if (ap->a_accmode & VWRITE)
394 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
395 			if (ap->a_accmode & VAPPEND)
396 				mode |= NFSACCESS_EXTEND;
397 			if (ap->a_accmode & VEXEC)
398 				mode |= NFSACCESS_LOOKUP;
399 			if (ap->a_accmode & VDELETE)
400 				mode |= NFSACCESS_DELETE;
401 			if (ap->a_accmode & VDELETE_CHILD)
402 				mode |= NFSACCESS_MODIFY;
403 		}
404 		/* XXX safety belt, only make blanket request if caching */
405 		if (nfsaccess_cache_timeout > 0) {
406 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
407 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
408 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
409 		} else {
410 			wmode = mode;
411 		}
412 
413 		/*
414 		 * Does our cached result allow us to give a definite yes to
415 		 * this request?
416 		 */
417 		gotahit = 0;
418 		mtx_lock(&np->n_mtx);
419 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
420 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
421 			    if (time_second < (np->n_accesscache[i].stamp
422 				+ nfsaccess_cache_timeout) &&
423 				(np->n_accesscache[i].mode & mode) == mode) {
424 				NFSINCRGLOBAL(newnfsstats.accesscache_hits);
425 				gotahit = 1;
426 			    }
427 			    break;
428 			}
429 		}
430 		mtx_unlock(&np->n_mtx);
431 #ifdef KDTRACE_HOOKS
432 		if (gotahit != 0)
433 			KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
434 			    ap->a_cred->cr_uid, mode);
435 		else
436 			KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
437 			    ap->a_cred->cr_uid, mode);
438 #endif
439 		if (gotahit == 0) {
440 			/*
441 			 * Either a no, or a don't know.  Go to the wire.
442 			 */
443 			NFSINCRGLOBAL(newnfsstats.accesscache_misses);
444 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
445 			    ap->a_cred, &rmode);
446 			if (!error &&
447 			    (rmode & mode) != mode)
448 				error = EACCES;
449 		}
450 		return (error);
451 	} else {
452 		if ((error = nfsspec_access(ap)) != 0) {
453 			return (error);
454 		}
455 		/*
456 		 * Attempt to prevent a mapped root from accessing a file
457 		 * which it shouldn't.  We try to read a byte from the file
458 		 * if the user is root and the file is not zero length.
459 		 * After calling nfsspec_access, we should have the correct
460 		 * file size cached.
461 		 */
462 		mtx_lock(&np->n_mtx);
463 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
464 		    && VTONFS(vp)->n_size > 0) {
465 			struct iovec aiov;
466 			struct uio auio;
467 			char buf[1];
468 
469 			mtx_unlock(&np->n_mtx);
470 			aiov.iov_base = buf;
471 			aiov.iov_len = 1;
472 			auio.uio_iov = &aiov;
473 			auio.uio_iovcnt = 1;
474 			auio.uio_offset = 0;
475 			auio.uio_resid = 1;
476 			auio.uio_segflg = UIO_SYSSPACE;
477 			auio.uio_rw = UIO_READ;
478 			auio.uio_td = ap->a_td;
479 
480 			if (vp->v_type == VREG)
481 				error = ncl_readrpc(vp, &auio, ap->a_cred);
482 			else if (vp->v_type == VDIR) {
483 				char* bp;
484 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
485 				aiov.iov_base = bp;
486 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
487 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
488 				    ap->a_td);
489 				free(bp, M_TEMP);
490 			} else if (vp->v_type == VLNK)
491 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
492 			else
493 				error = EACCES;
494 		} else
495 			mtx_unlock(&np->n_mtx);
496 		return (error);
497 	}
498 }
499 
500 
501 /*
502  * nfs open vnode op
503  * Check to see if the type is ok
504  * and that deletion is not in progress.
505  * For paged in text files, you will need to flush the page cache
506  * if consistency is lost.
507  */
508 /* ARGSUSED */
509 static int
510 nfs_open(struct vop_open_args *ap)
511 {
512 	struct vnode *vp = ap->a_vp;
513 	struct nfsnode *np = VTONFS(vp);
514 	struct vattr vattr;
515 	int error;
516 	int fmode = ap->a_mode;
517 	struct ucred *cred;
518 
519 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
520 		return (EOPNOTSUPP);
521 
522 	/*
523 	 * For NFSv4, we need to do the Open Op before cache validation,
524 	 * so that we conform to RFC3530 Sec. 9.3.1.
525 	 */
526 	if (NFS_ISV4(vp)) {
527 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
528 		if (error) {
529 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
530 			    (gid_t)0);
531 			return (error);
532 		}
533 	}
534 
535 	/*
536 	 * Now, if this Open will be doing reading, re-validate/flush the
537 	 * cache, so that Close/Open coherency is maintained.
538 	 */
539 	mtx_lock(&np->n_mtx);
540 	if (np->n_flag & NMODIFIED) {
541 		mtx_unlock(&np->n_mtx);
542 		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
543 		if (error == EINTR || error == EIO) {
544 			if (NFS_ISV4(vp))
545 				(void) nfsrpc_close(vp, 0, ap->a_td);
546 			return (error);
547 		}
548 		mtx_lock(&np->n_mtx);
549 		np->n_attrstamp = 0;
550 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
551 		if (vp->v_type == VDIR)
552 			np->n_direofoffset = 0;
553 		mtx_unlock(&np->n_mtx);
554 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
555 		if (error) {
556 			if (NFS_ISV4(vp))
557 				(void) nfsrpc_close(vp, 0, ap->a_td);
558 			return (error);
559 		}
560 		mtx_lock(&np->n_mtx);
561 		np->n_mtime = vattr.va_mtime;
562 		if (NFS_ISV4(vp))
563 			np->n_change = vattr.va_filerev;
564 	} else {
565 		mtx_unlock(&np->n_mtx);
566 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
567 		if (error) {
568 			if (NFS_ISV4(vp))
569 				(void) nfsrpc_close(vp, 0, ap->a_td);
570 			return (error);
571 		}
572 		mtx_lock(&np->n_mtx);
573 		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
574 		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
575 			if (vp->v_type == VDIR)
576 				np->n_direofoffset = 0;
577 			mtx_unlock(&np->n_mtx);
578 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
579 			if (error == EINTR || error == EIO) {
580 				if (NFS_ISV4(vp))
581 					(void) nfsrpc_close(vp, 0, ap->a_td);
582 				return (error);
583 			}
584 			mtx_lock(&np->n_mtx);
585 			np->n_mtime = vattr.va_mtime;
586 			if (NFS_ISV4(vp))
587 				np->n_change = vattr.va_filerev;
588 		}
589 	}
590 
591 	/*
592 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
593 	 */
594 	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
595 	    (vp->v_type == VREG)) {
596 		if (np->n_directio_opens == 0) {
597 			mtx_unlock(&np->n_mtx);
598 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
599 			if (error) {
600 				if (NFS_ISV4(vp))
601 					(void) nfsrpc_close(vp, 0, ap->a_td);
602 				return (error);
603 			}
604 			mtx_lock(&np->n_mtx);
605 			np->n_flag |= NNONCACHE;
606 		}
607 		np->n_directio_opens++;
608 	}
609 
610 	/* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
611 	if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
612 		np->n_flag |= NWRITEOPENED;
613 
614 	/*
615 	 * If this is an open for writing, capture a reference to the
616 	 * credentials, so they can be used by ncl_putpages(). Using
617 	 * these write credentials is preferable to the credentials of
618 	 * whatever thread happens to be doing the VOP_PUTPAGES() since
619 	 * the write RPCs are less likely to fail with EACCES.
620 	 */
621 	if ((fmode & FWRITE) != 0) {
622 		cred = np->n_writecred;
623 		np->n_writecred = crhold(ap->a_cred);
624 	} else
625 		cred = NULL;
626 	mtx_unlock(&np->n_mtx);
627 
628 	if (cred != NULL)
629 		crfree(cred);
630 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
631 	return (0);
632 }
633 
634 /*
635  * nfs close vnode op
636  * What an NFS client should do upon close after writing is a debatable issue.
637  * Most NFS clients push delayed writes to the server upon close, basically for
638  * two reasons:
639  * 1 - So that any write errors may be reported back to the client process
640  *     doing the close system call. By far the two most likely errors are
641  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
642  * 2 - To put a worst case upper bound on cache inconsistency between
643  *     multiple clients for the file.
644  * There is also a consistency problem for Version 2 of the protocol w.r.t.
645  * not being able to tell if other clients are writing a file concurrently,
646  * since there is no way of knowing if the changed modify time in the reply
647  * is only due to the write for this client.
648  * (NFS Version 3 provides weak cache consistency data in the reply that
649  *  should be sufficient to detect and handle this case.)
650  *
651  * The current code does the following:
652  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
653  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
654  *                     or commit them (this satisfies 1 and 2 except for the
655  *                     case where the server crashes after this close but
656  *                     before the commit RPC, which is felt to be "good
657  *                     enough". Changing the last argument to ncl_flush() to
658  *                     a 1 would force a commit operation, if it is felt a
659  *                     commit is necessary now.
660  * for NFS Version 4 - flush the dirty buffers and commit them, if
661  *		       nfscl_mustflush() says this is necessary.
662  *                     It is necessary if there is no write delegation held,
663  *                     in order to satisfy open/close coherency.
664  *                     If the file isn't cached on local stable storage,
665  *                     it may be necessary in order to detect "out of space"
666  *                     errors from the server, if the write delegation
667  *                     issued by the server doesn't allow the file to grow.
668  */
669 /* ARGSUSED */
670 static int
671 nfs_close(struct vop_close_args *ap)
672 {
673 	struct vnode *vp = ap->a_vp;
674 	struct nfsnode *np = VTONFS(vp);
675 	struct nfsvattr nfsva;
676 	struct ucred *cred;
677 	int error = 0, ret, localcred = 0;
678 	int fmode = ap->a_fflag;
679 
680 	if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
681 		return (0);
682 	/*
683 	 * During shutdown, a_cred isn't valid, so just use root.
684 	 */
685 	if (ap->a_cred == NOCRED) {
686 		cred = newnfs_getcred();
687 		localcred = 1;
688 	} else {
689 		cred = ap->a_cred;
690 	}
691 	if (vp->v_type == VREG) {
692 	    /*
693 	     * Examine and clean dirty pages, regardless of NMODIFIED.
694 	     * This closes a major hole in close-to-open consistency.
695 	     * We want to push out all dirty pages (and buffers) on
696 	     * close, regardless of whether they were dirtied by
697 	     * mmap'ed writes or via write().
698 	     */
699 	    if (nfs_clean_pages_on_close && vp->v_object) {
700 		VM_OBJECT_WLOCK(vp->v_object);
701 		vm_object_page_clean(vp->v_object, 0, 0, 0);
702 		VM_OBJECT_WUNLOCK(vp->v_object);
703 	    }
704 	    mtx_lock(&np->n_mtx);
705 	    if (np->n_flag & NMODIFIED) {
706 		mtx_unlock(&np->n_mtx);
707 		if (NFS_ISV3(vp)) {
708 		    /*
709 		     * Under NFSv3 we have dirty buffers to dispose of.  We
710 		     * must flush them to the NFS server.  We have the option
711 		     * of waiting all the way through the commit rpc or just
712 		     * waiting for the initial write.  The default is to only
713 		     * wait through the initial write so the data is in the
714 		     * server's cache, which is roughly similar to the state
715 		     * a standard disk subsystem leaves the file in on close().
716 		     *
717 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
718 		     * potential races with other processes, and certainly
719 		     * cannot clear it if we don't commit.
720 		     * These races occur when there is no longer the old
721 		     * traditional vnode locking implemented for Vnode Ops.
722 		     */
723 		    int cm = newnfs_commit_on_close ? 1 : 0;
724 		    error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
725 		    /* np->n_flag &= ~NMODIFIED; */
726 		} else if (NFS_ISV4(vp)) {
727 			if (nfscl_mustflush(vp) != 0) {
728 				int cm = newnfs_commit_on_close ? 1 : 0;
729 				error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
730 				    cm, 0);
731 				/*
732 				 * as above w.r.t races when clearing
733 				 * NMODIFIED.
734 				 * np->n_flag &= ~NMODIFIED;
735 				 */
736 			}
737 		} else
738 		    error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
739 		mtx_lock(&np->n_mtx);
740 	    }
741  	    /*
742  	     * Invalidate the attribute cache in all cases.
743  	     * An open is going to fetch fresh attrs any way, other procs
744  	     * on this node that have file open will be forced to do an
745  	     * otw attr fetch, but this is safe.
746 	     * --> A user found that their RPC count dropped by 20% when
747 	     *     this was commented out and I can't see any requirement
748 	     *     for it, so I've disabled it when negative lookups are
749 	     *     enabled. (What does this have to do with negative lookup
750 	     *     caching? Well nothing, except it was reported by the
751 	     *     same user that needed negative lookup caching and I wanted
752 	     *     there to be a way to disable it to see if it
753 	     *     is the cause of some caching/coherency issue that might
754 	     *     crop up.)
755  	     */
756 	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
757 		    np->n_attrstamp = 0;
758 		    KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
759 	    }
760 	    if (np->n_flag & NWRITEERR) {
761 		np->n_flag &= ~NWRITEERR;
762 		error = np->n_error;
763 	    }
764 	    mtx_unlock(&np->n_mtx);
765 	}
766 
767 	if (NFS_ISV4(vp)) {
768 		/*
769 		 * Get attributes so "change" is up to date.
770 		 */
771 		if (error == 0 && nfscl_mustflush(vp) != 0) {
772 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
773 			    NULL);
774 			if (!ret) {
775 				np->n_change = nfsva.na_filerev;
776 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
777 				    NULL, 0, 0);
778 			}
779 		}
780 
781 		/*
782 		 * and do the close.
783 		 */
784 		ret = nfsrpc_close(vp, 0, ap->a_td);
785 		if (!error && ret)
786 			error = ret;
787 		if (error)
788 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
789 			    (gid_t)0);
790 	}
791 	if (newnfs_directio_enable)
792 		KASSERT((np->n_directio_asyncwr == 0),
793 			("nfs_close: dirty unflushed (%d) directio buffers\n",
794 			 np->n_directio_asyncwr));
795 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
796 		mtx_lock(&np->n_mtx);
797 		KASSERT((np->n_directio_opens > 0),
798 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
799 		np->n_directio_opens--;
800 		if (np->n_directio_opens == 0)
801 			np->n_flag &= ~NNONCACHE;
802 		mtx_unlock(&np->n_mtx);
803 	}
804 	if (localcred)
805 		NFSFREECRED(cred);
806 	return (error);
807 }
808 
809 /*
810  * nfs getattr call from vfs.
811  */
812 static int
813 nfs_getattr(struct vop_getattr_args *ap)
814 {
815 	struct vnode *vp = ap->a_vp;
816 	struct thread *td = curthread;	/* XXX */
817 	struct nfsnode *np = VTONFS(vp);
818 	int error = 0;
819 	struct nfsvattr nfsva;
820 	struct vattr *vap = ap->a_vap;
821 	struct vattr vattr;
822 
823 	/*
824 	 * Update local times for special files.
825 	 */
826 	mtx_lock(&np->n_mtx);
827 	if (np->n_flag & (NACC | NUPD))
828 		np->n_flag |= NCHG;
829 	mtx_unlock(&np->n_mtx);
830 	/*
831 	 * First look in the cache.
832 	 */
833 	if (ncl_getattrcache(vp, &vattr) == 0) {
834 		vap->va_type = vattr.va_type;
835 		vap->va_mode = vattr.va_mode;
836 		vap->va_nlink = vattr.va_nlink;
837 		vap->va_uid = vattr.va_uid;
838 		vap->va_gid = vattr.va_gid;
839 		vap->va_fsid = vattr.va_fsid;
840 		vap->va_fileid = vattr.va_fileid;
841 		vap->va_size = vattr.va_size;
842 		vap->va_blocksize = vattr.va_blocksize;
843 		vap->va_atime = vattr.va_atime;
844 		vap->va_mtime = vattr.va_mtime;
845 		vap->va_ctime = vattr.va_ctime;
846 		vap->va_gen = vattr.va_gen;
847 		vap->va_flags = vattr.va_flags;
848 		vap->va_rdev = vattr.va_rdev;
849 		vap->va_bytes = vattr.va_bytes;
850 		vap->va_filerev = vattr.va_filerev;
851 		/*
852 		 * Get the local modify time for the case of a write
853 		 * delegation.
854 		 */
855 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
856 		return (0);
857 	}
858 
859 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
860 	    nfsaccess_cache_timeout > 0) {
861 		NFSINCRGLOBAL(newnfsstats.accesscache_misses);
862 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
863 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
864 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
865 			return (0);
866 		}
867 	}
868 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
869 	if (!error)
870 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
871 	if (!error) {
872 		/*
873 		 * Get the local modify time for the case of a write
874 		 * delegation.
875 		 */
876 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
877 	} else if (NFS_ISV4(vp)) {
878 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
879 	}
880 	return (error);
881 }
882 
883 /*
884  * nfs setattr call.
885  */
886 static int
887 nfs_setattr(struct vop_setattr_args *ap)
888 {
889 	struct vnode *vp = ap->a_vp;
890 	struct nfsnode *np = VTONFS(vp);
891 	struct thread *td = curthread;	/* XXX */
892 	struct vattr *vap = ap->a_vap;
893 	int error = 0;
894 	u_quad_t tsize;
895 
896 #ifndef nolint
897 	tsize = (u_quad_t)0;
898 #endif
899 
900 	/*
901 	 * Setting of flags and marking of atimes are not supported.
902 	 */
903 	if (vap->va_flags != VNOVAL)
904 		return (EOPNOTSUPP);
905 
906 	/*
907 	 * Disallow write attempts if the filesystem is mounted read-only.
908 	 */
909   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
910 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
911 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
912 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
913 		return (EROFS);
914 	if (vap->va_size != VNOVAL) {
915  		switch (vp->v_type) {
916  		case VDIR:
917  			return (EISDIR);
918  		case VCHR:
919  		case VBLK:
920  		case VSOCK:
921  		case VFIFO:
922 			if (vap->va_mtime.tv_sec == VNOVAL &&
923 			    vap->va_atime.tv_sec == VNOVAL &&
924 			    vap->va_mode == (mode_t)VNOVAL &&
925 			    vap->va_uid == (uid_t)VNOVAL &&
926 			    vap->va_gid == (gid_t)VNOVAL)
927 				return (0);
928  			vap->va_size = VNOVAL;
929  			break;
930  		default:
931 			/*
932 			 * Disallow write attempts if the filesystem is
933 			 * mounted read-only.
934 			 */
935 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
936 				return (EROFS);
937 			/*
938 			 *  We run vnode_pager_setsize() early (why?),
939 			 * we must set np->n_size now to avoid vinvalbuf
940 			 * V_SAVE races that might setsize a lower
941 			 * value.
942 			 */
943 			mtx_lock(&np->n_mtx);
944 			tsize = np->n_size;
945 			mtx_unlock(&np->n_mtx);
946 			error = ncl_meta_setsize(vp, ap->a_cred, td,
947 			    vap->va_size);
948 			mtx_lock(&np->n_mtx);
949  			if (np->n_flag & NMODIFIED) {
950 			    tsize = np->n_size;
951 			    mtx_unlock(&np->n_mtx);
952  			    if (vap->va_size == 0)
953  				error = ncl_vinvalbuf(vp, 0, td, 1);
954  			    else
955  				error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
956  			    if (error) {
957 				vnode_pager_setsize(vp, tsize);
958 				return (error);
959 			    }
960 			    /*
961 			     * Call nfscl_delegmodtime() to set the modify time
962 			     * locally, as required.
963 			     */
964 			    nfscl_delegmodtime(vp);
965  			} else
966 			    mtx_unlock(&np->n_mtx);
967 			/*
968 			 * np->n_size has already been set to vap->va_size
969 			 * in ncl_meta_setsize(). We must set it again since
970 			 * nfs_loadattrcache() could be called through
971 			 * ncl_meta_setsize() and could modify np->n_size.
972 			 */
973 			mtx_lock(&np->n_mtx);
974  			np->n_vattr.na_size = np->n_size = vap->va_size;
975 			mtx_unlock(&np->n_mtx);
976   		};
977   	} else {
978 		mtx_lock(&np->n_mtx);
979 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
980 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
981 			mtx_unlock(&np->n_mtx);
982 			if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
983 			    (error == EINTR || error == EIO))
984 				return (error);
985 		} else
986 			mtx_unlock(&np->n_mtx);
987 	}
988 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
989 	if (error && vap->va_size != VNOVAL) {
990 		mtx_lock(&np->n_mtx);
991 		np->n_size = np->n_vattr.na_size = tsize;
992 		vnode_pager_setsize(vp, tsize);
993 		mtx_unlock(&np->n_mtx);
994 	}
995 	return (error);
996 }
997 
998 /*
999  * Do an nfs setattr rpc.
1000  */
1001 static int
1002 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1003     struct thread *td)
1004 {
1005 	struct nfsnode *np = VTONFS(vp);
1006 	int error, ret, attrflag, i;
1007 	struct nfsvattr nfsva;
1008 
1009 	if (NFS_ISV34(vp)) {
1010 		mtx_lock(&np->n_mtx);
1011 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1012 			np->n_accesscache[i].stamp = 0;
1013 		np->n_flag |= NDELEGMOD;
1014 		mtx_unlock(&np->n_mtx);
1015 		KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1016 	}
1017 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1018 	    NULL);
1019 	if (attrflag) {
1020 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1021 		if (ret && !error)
1022 			error = ret;
1023 	}
1024 	if (error && NFS_ISV4(vp))
1025 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1026 	return (error);
1027 }
1028 
1029 /*
1030  * nfs lookup call, one step at a time...
1031  * First look in cache
1032  * If not found, unlock the directory nfsnode and do the rpc
1033  */
1034 static int
1035 nfs_lookup(struct vop_lookup_args *ap)
1036 {
1037 	struct componentname *cnp = ap->a_cnp;
1038 	struct vnode *dvp = ap->a_dvp;
1039 	struct vnode **vpp = ap->a_vpp;
1040 	struct mount *mp = dvp->v_mount;
1041 	int flags = cnp->cn_flags;
1042 	struct vnode *newvp;
1043 	struct nfsmount *nmp;
1044 	struct nfsnode *np, *newnp;
1045 	int error = 0, attrflag, dattrflag, ltype, ncticks;
1046 	struct thread *td = cnp->cn_thread;
1047 	struct nfsfh *nfhp;
1048 	struct nfsvattr dnfsva, nfsva;
1049 	struct vattr vattr;
1050 	struct timespec nctime;
1051 
1052 	*vpp = NULLVP;
1053 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1054 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1055 		return (EROFS);
1056 	if (dvp->v_type != VDIR)
1057 		return (ENOTDIR);
1058 	nmp = VFSTONFS(mp);
1059 	np = VTONFS(dvp);
1060 
1061 	/* For NFSv4, wait until any remove is done. */
1062 	mtx_lock(&np->n_mtx);
1063 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1064 		np->n_flag |= NREMOVEWANT;
1065 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1066 	}
1067 	mtx_unlock(&np->n_mtx);
1068 
1069 	if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1070 		return (error);
1071 	error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1072 	if (error > 0 && error != ENOENT)
1073 		return (error);
1074 	if (error == -1) {
1075 		/*
1076 		 * Lookups of "." are special and always return the
1077 		 * current directory.  cache_lookup() already handles
1078 		 * associated locking bookkeeping, etc.
1079 		 */
1080 		if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1081 			/* XXX: Is this really correct? */
1082 			if (cnp->cn_nameiop != LOOKUP &&
1083 			    (flags & ISLASTCN))
1084 				cnp->cn_flags |= SAVENAME;
1085 			return (0);
1086 		}
1087 
1088 		/*
1089 		 * We only accept a positive hit in the cache if the
1090 		 * change time of the file matches our cached copy.
1091 		 * Otherwise, we discard the cache entry and fallback
1092 		 * to doing a lookup RPC.  We also only trust cache
1093 		 * entries for less than nm_nametimeo seconds.
1094 		 *
1095 		 * To better handle stale file handles and attributes,
1096 		 * clear the attribute cache of this node if it is a
1097 		 * leaf component, part of an open() call, and not
1098 		 * locally modified before fetching the attributes.
1099 		 * This should allow stale file handles to be detected
1100 		 * here where we can fall back to a LOOKUP RPC to
1101 		 * recover rather than having nfs_open() detect the
1102 		 * stale file handle and failing open(2) with ESTALE.
1103 		 */
1104 		newvp = *vpp;
1105 		newnp = VTONFS(newvp);
1106 		if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1107 		    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1108 		    !(newnp->n_flag & NMODIFIED)) {
1109 			mtx_lock(&newnp->n_mtx);
1110 			newnp->n_attrstamp = 0;
1111 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1112 			mtx_unlock(&newnp->n_mtx);
1113 		}
1114 		if (nfscl_nodeleg(newvp, 0) == 0 ||
1115 		    ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1116 		    VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1117 		    timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1118 			NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1119 			if (cnp->cn_nameiop != LOOKUP &&
1120 			    (flags & ISLASTCN))
1121 				cnp->cn_flags |= SAVENAME;
1122 			return (0);
1123 		}
1124 		cache_purge(newvp);
1125 		if (dvp != newvp)
1126 			vput(newvp);
1127 		else
1128 			vrele(newvp);
1129 		*vpp = NULLVP;
1130 	} else if (error == ENOENT) {
1131 		if (dvp->v_iflag & VI_DOOMED)
1132 			return (ENOENT);
1133 		/*
1134 		 * We only accept a negative hit in the cache if the
1135 		 * modification time of the parent directory matches
1136 		 * the cached copy in the name cache entry.
1137 		 * Otherwise, we discard all of the negative cache
1138 		 * entries for this directory.  We also only trust
1139 		 * negative cache entries for up to nm_negnametimeo
1140 		 * seconds.
1141 		 */
1142 		if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1143 		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1144 		    timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1145 			NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1146 			return (ENOENT);
1147 		}
1148 		cache_purge_negative(dvp);
1149 	}
1150 
1151 	error = 0;
1152 	newvp = NULLVP;
1153 	NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1154 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1155 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1156 	    NULL);
1157 	if (dattrflag)
1158 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1159 	if (error) {
1160 		if (newvp != NULLVP) {
1161 			vput(newvp);
1162 			*vpp = NULLVP;
1163 		}
1164 
1165 		if (error != ENOENT) {
1166 			if (NFS_ISV4(dvp))
1167 				error = nfscl_maperr(td, error, (uid_t)0,
1168 				    (gid_t)0);
1169 			return (error);
1170 		}
1171 
1172 		/* The requested file was not found. */
1173 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1174 		    (flags & ISLASTCN)) {
1175 			/*
1176 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1177 			 * VWRITE) here instead of just checking
1178 			 * MNT_RDONLY.
1179 			 */
1180 			if (mp->mnt_flag & MNT_RDONLY)
1181 				return (EROFS);
1182 			cnp->cn_flags |= SAVENAME;
1183 			return (EJUSTRETURN);
1184 		}
1185 
1186 		if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
1187 		    dattrflag) {
1188 			/*
1189 			 * Cache the modification time of the parent
1190 			 * directory from the post-op attributes in
1191 			 * the name cache entry.  The negative cache
1192 			 * entry will be ignored once the directory
1193 			 * has changed.  Don't bother adding the entry
1194 			 * if the directory has already changed.
1195 			 */
1196 			mtx_lock(&np->n_mtx);
1197 			if (timespeccmp(&np->n_vattr.na_mtime,
1198 			    &dnfsva.na_mtime, ==)) {
1199 				mtx_unlock(&np->n_mtx);
1200 				cache_enter_time(dvp, NULL, cnp,
1201 				    &dnfsva.na_mtime, NULL);
1202 			} else
1203 				mtx_unlock(&np->n_mtx);
1204 		}
1205 		return (ENOENT);
1206 	}
1207 
1208 	/*
1209 	 * Handle RENAME case...
1210 	 */
1211 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1212 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1213 			FREE((caddr_t)nfhp, M_NFSFH);
1214 			return (EISDIR);
1215 		}
1216 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1217 		    LK_EXCLUSIVE);
1218 		if (error)
1219 			return (error);
1220 		newvp = NFSTOV(np);
1221 		if (attrflag)
1222 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1223 			    0, 1);
1224 		*vpp = newvp;
1225 		cnp->cn_flags |= SAVENAME;
1226 		return (0);
1227 	}
1228 
1229 	if (flags & ISDOTDOT) {
1230 		ltype = NFSVOPISLOCKED(dvp);
1231 		error = vfs_busy(mp, MBF_NOWAIT);
1232 		if (error != 0) {
1233 			vfs_ref(mp);
1234 			NFSVOPUNLOCK(dvp, 0);
1235 			error = vfs_busy(mp, 0);
1236 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1237 			vfs_rel(mp);
1238 			if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1239 				vfs_unbusy(mp);
1240 				error = ENOENT;
1241 			}
1242 			if (error != 0)
1243 				return (error);
1244 		}
1245 		NFSVOPUNLOCK(dvp, 0);
1246 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1247 		    cnp->cn_lkflags);
1248 		if (error == 0)
1249 			newvp = NFSTOV(np);
1250 		vfs_unbusy(mp);
1251 		if (newvp != dvp)
1252 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1253 		if (dvp->v_iflag & VI_DOOMED) {
1254 			if (error == 0) {
1255 				if (newvp == dvp)
1256 					vrele(newvp);
1257 				else
1258 					vput(newvp);
1259 			}
1260 			error = ENOENT;
1261 		}
1262 		if (error != 0)
1263 			return (error);
1264 		if (attrflag)
1265 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1266 			    0, 1);
1267 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1268 		FREE((caddr_t)nfhp, M_NFSFH);
1269 		VREF(dvp);
1270 		newvp = dvp;
1271 		if (attrflag)
1272 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1273 			    0, 1);
1274 	} else {
1275 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1276 		    cnp->cn_lkflags);
1277 		if (error)
1278 			return (error);
1279 		newvp = NFSTOV(np);
1280 		if (attrflag)
1281 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1282 			    0, 1);
1283 		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1284 		    !(np->n_flag & NMODIFIED)) {
1285 			/*
1286 			 * Flush the attribute cache when opening a
1287 			 * leaf node to ensure that fresh attributes
1288 			 * are fetched in nfs_open() since we did not
1289 			 * fetch attributes from the LOOKUP reply.
1290 			 */
1291 			mtx_lock(&np->n_mtx);
1292 			np->n_attrstamp = 0;
1293 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1294 			mtx_unlock(&np->n_mtx);
1295 		}
1296 	}
1297 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1298 		cnp->cn_flags |= SAVENAME;
1299 	if ((cnp->cn_flags & MAKEENTRY) &&
1300 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1301 	    attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1302 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1303 		    newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1304 	*vpp = newvp;
1305 	return (0);
1306 }
1307 
1308 /*
1309  * nfs read call.
1310  * Just call ncl_bioread() to do the work.
1311  */
1312 static int
1313 nfs_read(struct vop_read_args *ap)
1314 {
1315 	struct vnode *vp = ap->a_vp;
1316 
1317 	switch (vp->v_type) {
1318 	case VREG:
1319 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1320 	case VDIR:
1321 		return (EISDIR);
1322 	default:
1323 		return (EOPNOTSUPP);
1324 	}
1325 }
1326 
1327 /*
1328  * nfs readlink call
1329  */
1330 static int
1331 nfs_readlink(struct vop_readlink_args *ap)
1332 {
1333 	struct vnode *vp = ap->a_vp;
1334 
1335 	if (vp->v_type != VLNK)
1336 		return (EINVAL);
1337 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1338 }
1339 
1340 /*
1341  * Do a readlink rpc.
1342  * Called by ncl_doio() from below the buffer cache.
1343  */
1344 int
1345 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1346 {
1347 	int error, ret, attrflag;
1348 	struct nfsvattr nfsva;
1349 
1350 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1351 	    &attrflag, NULL);
1352 	if (attrflag) {
1353 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1354 		if (ret && !error)
1355 			error = ret;
1356 	}
1357 	if (error && NFS_ISV4(vp))
1358 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1359 	return (error);
1360 }
1361 
1362 /*
1363  * nfs read rpc call
1364  * Ditto above
1365  */
1366 int
1367 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1368 {
1369 	int error, ret, attrflag;
1370 	struct nfsvattr nfsva;
1371 	struct nfsmount *nmp;
1372 
1373 	nmp = VFSTONFS(vnode_mount(vp));
1374 	error = EIO;
1375 	attrflag = 0;
1376 	if (NFSHASPNFS(nmp))
1377 		error = nfscl_doiods(vp, uiop, NULL, NULL,
1378 		    NFSV4OPEN_ACCESSREAD, cred, uiop->uio_td);
1379 	NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1380 	if (error != 0)
1381 		error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1382 		    &attrflag, NULL);
1383 	if (attrflag) {
1384 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1385 		if (ret && !error)
1386 			error = ret;
1387 	}
1388 	if (error && NFS_ISV4(vp))
1389 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1390 	return (error);
1391 }
1392 
1393 /*
1394  * nfs write call
1395  */
1396 int
1397 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1398     int *iomode, int *must_commit, int called_from_strategy)
1399 {
1400 	struct nfsvattr nfsva;
1401 	int error, attrflag, ret;
1402 	struct nfsmount *nmp;
1403 
1404 	nmp = VFSTONFS(vnode_mount(vp));
1405 	error = EIO;
1406 	attrflag = 0;
1407 	if (NFSHASPNFS(nmp))
1408 		error = nfscl_doiods(vp, uiop, iomode, must_commit,
1409 		    NFSV4OPEN_ACCESSWRITE, cred, uiop->uio_td);
1410 	NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1411 	if (error != 0)
1412 		error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1413 		    uiop->uio_td, &nfsva, &attrflag, NULL,
1414 		    called_from_strategy);
1415 	if (attrflag) {
1416 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1417 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1418 			    1);
1419 		else
1420 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1421 			    1);
1422 		if (ret && !error)
1423 			error = ret;
1424 	}
1425 	if (DOINGASYNC(vp))
1426 		*iomode = NFSWRITE_FILESYNC;
1427 	if (error && NFS_ISV4(vp))
1428 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1429 	return (error);
1430 }
1431 
1432 /*
1433  * nfs mknod rpc
1434  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1435  * mode set to specify the file type and the size field for rdev.
1436  */
1437 static int
1438 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1439     struct vattr *vap)
1440 {
1441 	struct nfsvattr nfsva, dnfsva;
1442 	struct vnode *newvp = NULL;
1443 	struct nfsnode *np = NULL, *dnp;
1444 	struct nfsfh *nfhp;
1445 	struct vattr vattr;
1446 	int error = 0, attrflag, dattrflag;
1447 	u_int32_t rdev;
1448 
1449 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1450 		rdev = vap->va_rdev;
1451 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1452 		rdev = 0xffffffff;
1453 	else
1454 		return (EOPNOTSUPP);
1455 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1456 		return (error);
1457 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1458 	    rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1459 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1460 	if (!error) {
1461 		if (!nfhp)
1462 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1463 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1464 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1465 			    NULL);
1466 		if (nfhp)
1467 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1468 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1469 	}
1470 	if (dattrflag)
1471 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1472 	if (!error) {
1473 		newvp = NFSTOV(np);
1474 		if (attrflag != 0) {
1475 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1476 			    0, 1);
1477 			if (error != 0)
1478 				vput(newvp);
1479 		}
1480 	}
1481 	if (!error) {
1482 		*vpp = newvp;
1483 	} else if (NFS_ISV4(dvp)) {
1484 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1485 		    vap->va_gid);
1486 	}
1487 	dnp = VTONFS(dvp);
1488 	mtx_lock(&dnp->n_mtx);
1489 	dnp->n_flag |= NMODIFIED;
1490 	if (!dattrflag) {
1491 		dnp->n_attrstamp = 0;
1492 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1493 	}
1494 	mtx_unlock(&dnp->n_mtx);
1495 	return (error);
1496 }
1497 
1498 /*
1499  * nfs mknod vop
1500  * just call nfs_mknodrpc() to do the work.
1501  */
1502 /* ARGSUSED */
1503 static int
1504 nfs_mknod(struct vop_mknod_args *ap)
1505 {
1506 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1507 }
1508 
1509 static struct mtx nfs_cverf_mtx;
1510 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1511     MTX_DEF);
1512 
1513 static nfsquad_t
1514 nfs_get_cverf(void)
1515 {
1516 	static nfsquad_t cverf;
1517 	nfsquad_t ret;
1518 	static int cverf_initialized = 0;
1519 
1520 	mtx_lock(&nfs_cverf_mtx);
1521 	if (cverf_initialized == 0) {
1522 		cverf.lval[0] = arc4random();
1523 		cverf.lval[1] = arc4random();
1524 		cverf_initialized = 1;
1525 	} else
1526 		cverf.qval++;
1527 	ret = cverf;
1528 	mtx_unlock(&nfs_cverf_mtx);
1529 
1530 	return (ret);
1531 }
1532 
1533 /*
1534  * nfs file create call
1535  */
1536 static int
1537 nfs_create(struct vop_create_args *ap)
1538 {
1539 	struct vnode *dvp = ap->a_dvp;
1540 	struct vattr *vap = ap->a_vap;
1541 	struct componentname *cnp = ap->a_cnp;
1542 	struct nfsnode *np = NULL, *dnp;
1543 	struct vnode *newvp = NULL;
1544 	struct nfsmount *nmp;
1545 	struct nfsvattr dnfsva, nfsva;
1546 	struct nfsfh *nfhp;
1547 	nfsquad_t cverf;
1548 	int error = 0, attrflag, dattrflag, fmode = 0;
1549 	struct vattr vattr;
1550 
1551 	/*
1552 	 * Oops, not for me..
1553 	 */
1554 	if (vap->va_type == VSOCK)
1555 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1556 
1557 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1558 		return (error);
1559 	if (vap->va_vaflags & VA_EXCLUSIVE)
1560 		fmode |= O_EXCL;
1561 	dnp = VTONFS(dvp);
1562 	nmp = VFSTONFS(vnode_mount(dvp));
1563 again:
1564 	/* For NFSv4, wait until any remove is done. */
1565 	mtx_lock(&dnp->n_mtx);
1566 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1567 		dnp->n_flag |= NREMOVEWANT;
1568 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1569 	}
1570 	mtx_unlock(&dnp->n_mtx);
1571 
1572 	cverf = nfs_get_cverf();
1573 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1574 	    vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1575 	    &nfhp, &attrflag, &dattrflag, NULL);
1576 	if (!error) {
1577 		if (nfhp == NULL)
1578 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1579 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1580 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1581 			    NULL);
1582 		if (nfhp != NULL)
1583 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1584 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1585 	}
1586 	if (dattrflag)
1587 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1588 	if (!error) {
1589 		newvp = NFSTOV(np);
1590 		if (attrflag == 0)
1591 			error = nfsrpc_getattr(newvp, cnp->cn_cred,
1592 			    cnp->cn_thread, &nfsva, NULL);
1593 		if (error == 0)
1594 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1595 			    0, 1);
1596 	}
1597 	if (error) {
1598 		if (newvp != NULL) {
1599 			vput(newvp);
1600 			newvp = NULL;
1601 		}
1602 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1603 		    error == NFSERR_NOTSUPP) {
1604 			fmode &= ~O_EXCL;
1605 			goto again;
1606 		}
1607 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1608 		if (nfscl_checksattr(vap, &nfsva)) {
1609 			/*
1610 			 * We are normally called with only a partially
1611 			 * initialized VAP. Since the NFSv3 spec says that
1612 			 * the server may use the file attributes to
1613 			 * store the verifier, the spec requires us to do a
1614 			 * SETATTR RPC. FreeBSD servers store the verifier in
1615 			 * atime, but we can't really assume that all servers
1616 			 * will so we ensure that our SETATTR sets both atime
1617 			 * and mtime.
1618 			 */
1619 			if (vap->va_mtime.tv_sec == VNOVAL)
1620 				vfs_timestamp(&vap->va_mtime);
1621 			if (vap->va_atime.tv_sec == VNOVAL)
1622 				vap->va_atime = vap->va_mtime;
1623 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1624 			    cnp->cn_thread, &nfsva, &attrflag, NULL);
1625 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1626 			    vap->va_gid != (gid_t)VNOVAL)) {
1627 				/* try again without setting uid/gid */
1628 				vap->va_uid = (uid_t)VNOVAL;
1629 				vap->va_gid = (uid_t)VNOVAL;
1630 				error = nfsrpc_setattr(newvp, vap, NULL,
1631 				    cnp->cn_cred, cnp->cn_thread, &nfsva,
1632 				    &attrflag, NULL);
1633 			}
1634 			if (attrflag)
1635 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1636 				    NULL, 0, 1);
1637 			if (error != 0)
1638 				vput(newvp);
1639 		}
1640 	}
1641 	if (!error) {
1642 		if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1643 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1644 			    NULL);
1645 		*ap->a_vpp = newvp;
1646 	} else if (NFS_ISV4(dvp)) {
1647 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1648 		    vap->va_gid);
1649 	}
1650 	mtx_lock(&dnp->n_mtx);
1651 	dnp->n_flag |= NMODIFIED;
1652 	if (!dattrflag) {
1653 		dnp->n_attrstamp = 0;
1654 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1655 	}
1656 	mtx_unlock(&dnp->n_mtx);
1657 	return (error);
1658 }
1659 
1660 /*
1661  * nfs file remove call
1662  * To try and make nfs semantics closer to ufs semantics, a file that has
1663  * other processes using the vnode is renamed instead of removed and then
1664  * removed later on the last close.
1665  * - If v_usecount > 1
1666  *	  If a rename is not already in the works
1667  *	     call nfs_sillyrename() to set it up
1668  *     else
1669  *	  do the remove rpc
1670  */
1671 static int
1672 nfs_remove(struct vop_remove_args *ap)
1673 {
1674 	struct vnode *vp = ap->a_vp;
1675 	struct vnode *dvp = ap->a_dvp;
1676 	struct componentname *cnp = ap->a_cnp;
1677 	struct nfsnode *np = VTONFS(vp);
1678 	int error = 0;
1679 	struct vattr vattr;
1680 
1681 	KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1682 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1683 	if (vp->v_type == VDIR)
1684 		error = EPERM;
1685 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1686 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1687 	    vattr.va_nlink > 1)) {
1688 		/*
1689 		 * Purge the name cache so that the chance of a lookup for
1690 		 * the name succeeding while the remove is in progress is
1691 		 * minimized. Without node locking it can still happen, such
1692 		 * that an I/O op returns ESTALE, but since you get this if
1693 		 * another host removes the file..
1694 		 */
1695 		cache_purge(vp);
1696 		/*
1697 		 * throw away biocache buffers, mainly to avoid
1698 		 * unnecessary delayed writes later.
1699 		 */
1700 		error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1701 		/* Do the rpc */
1702 		if (error != EINTR && error != EIO)
1703 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1704 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1705 		/*
1706 		 * Kludge City: If the first reply to the remove rpc is lost..
1707 		 *   the reply to the retransmitted request will be ENOENT
1708 		 *   since the file was in fact removed
1709 		 *   Therefore, we cheat and return success.
1710 		 */
1711 		if (error == ENOENT)
1712 			error = 0;
1713 	} else if (!np->n_sillyrename)
1714 		error = nfs_sillyrename(dvp, vp, cnp);
1715 	mtx_lock(&np->n_mtx);
1716 	np->n_attrstamp = 0;
1717 	mtx_unlock(&np->n_mtx);
1718 	KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1719 	return (error);
1720 }
1721 
1722 /*
1723  * nfs file remove rpc called from nfs_inactive
1724  */
1725 int
1726 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1727 {
1728 	/*
1729 	 * Make sure that the directory vnode is still valid.
1730 	 * XXX we should lock sp->s_dvp here.
1731 	 */
1732 	if (sp->s_dvp->v_type == VBAD)
1733 		return (0);
1734 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1735 	    sp->s_cred, NULL));
1736 }
1737 
1738 /*
1739  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1740  */
1741 static int
1742 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1743     int namelen, struct ucred *cred, struct thread *td)
1744 {
1745 	struct nfsvattr dnfsva;
1746 	struct nfsnode *dnp = VTONFS(dvp);
1747 	int error = 0, dattrflag;
1748 
1749 	mtx_lock(&dnp->n_mtx);
1750 	dnp->n_flag |= NREMOVEINPROG;
1751 	mtx_unlock(&dnp->n_mtx);
1752 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1753 	    &dattrflag, NULL);
1754 	mtx_lock(&dnp->n_mtx);
1755 	if ((dnp->n_flag & NREMOVEWANT)) {
1756 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1757 		mtx_unlock(&dnp->n_mtx);
1758 		wakeup((caddr_t)dnp);
1759 	} else {
1760 		dnp->n_flag &= ~NREMOVEINPROG;
1761 		mtx_unlock(&dnp->n_mtx);
1762 	}
1763 	if (dattrflag)
1764 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1765 	mtx_lock(&dnp->n_mtx);
1766 	dnp->n_flag |= NMODIFIED;
1767 	if (!dattrflag) {
1768 		dnp->n_attrstamp = 0;
1769 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1770 	}
1771 	mtx_unlock(&dnp->n_mtx);
1772 	if (error && NFS_ISV4(dvp))
1773 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1774 	return (error);
1775 }
1776 
1777 /*
1778  * nfs file rename call
1779  */
1780 static int
1781 nfs_rename(struct vop_rename_args *ap)
1782 {
1783 	struct vnode *fvp = ap->a_fvp;
1784 	struct vnode *tvp = ap->a_tvp;
1785 	struct vnode *fdvp = ap->a_fdvp;
1786 	struct vnode *tdvp = ap->a_tdvp;
1787 	struct componentname *tcnp = ap->a_tcnp;
1788 	struct componentname *fcnp = ap->a_fcnp;
1789 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1790 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1791 	struct nfsv4node *newv4 = NULL;
1792 	int error;
1793 
1794 	KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1795 	    (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1796 	/* Check for cross-device rename */
1797 	if ((fvp->v_mount != tdvp->v_mount) ||
1798 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1799 		error = EXDEV;
1800 		goto out;
1801 	}
1802 
1803 	if (fvp == tvp) {
1804 		ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1805 		error = 0;
1806 		goto out;
1807 	}
1808 	if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1809 		goto out;
1810 
1811 	/*
1812 	 * We have to flush B_DELWRI data prior to renaming
1813 	 * the file.  If we don't, the delayed-write buffers
1814 	 * can be flushed out later after the file has gone stale
1815 	 * under NFSV3.  NFSV2 does not have this problem because
1816 	 * ( as far as I can tell ) it flushes dirty buffers more
1817 	 * often.
1818 	 *
1819 	 * Skip the rename operation if the fsync fails, this can happen
1820 	 * due to the server's volume being full, when we pushed out data
1821 	 * that was written back to our cache earlier. Not checking for
1822 	 * this condition can result in potential (silent) data loss.
1823 	 */
1824 	error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1825 	NFSVOPUNLOCK(fvp, 0);
1826 	if (!error && tvp)
1827 		error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1828 	if (error)
1829 		goto out;
1830 
1831 	/*
1832 	 * If the tvp exists and is in use, sillyrename it before doing the
1833 	 * rename of the new file over it.
1834 	 * XXX Can't sillyrename a directory.
1835 	 */
1836 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1837 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1838 		vput(tvp);
1839 		tvp = NULL;
1840 	}
1841 
1842 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1843 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1844 	    tcnp->cn_thread);
1845 
1846 	if (error == 0 && NFS_ISV4(tdvp)) {
1847 		/*
1848 		 * For NFSv4, check to see if it is the same name and
1849 		 * replace the name, if it is different.
1850 		 */
1851 		MALLOC(newv4, struct nfsv4node *,
1852 		    sizeof (struct nfsv4node) +
1853 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1854 		    M_NFSV4NODE, M_WAITOK);
1855 		mtx_lock(&tdnp->n_mtx);
1856 		mtx_lock(&fnp->n_mtx);
1857 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1858 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1859 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1860 		      tcnp->cn_namelen) ||
1861 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1862 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1863 			tdnp->n_fhp->nfh_len))) {
1864 #ifdef notdef
1865 { char nnn[100]; int nnnl;
1866 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1867 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1868 nnn[nnnl] = '\0';
1869 printf("ren replace=%s\n",nnn);
1870 }
1871 #endif
1872 			FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1873 			fnp->n_v4 = newv4;
1874 			newv4 = NULL;
1875 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1876 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1877 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1878 			    tdnp->n_fhp->nfh_len);
1879 			NFSBCOPY(tcnp->cn_nameptr,
1880 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1881 		}
1882 		mtx_unlock(&tdnp->n_mtx);
1883 		mtx_unlock(&fnp->n_mtx);
1884 		if (newv4 != NULL)
1885 			FREE((caddr_t)newv4, M_NFSV4NODE);
1886 	}
1887 
1888 	if (fvp->v_type == VDIR) {
1889 		if (tvp != NULL && tvp->v_type == VDIR)
1890 			cache_purge(tdvp);
1891 		cache_purge(fdvp);
1892 	}
1893 
1894 out:
1895 	if (tdvp == tvp)
1896 		vrele(tdvp);
1897 	else
1898 		vput(tdvp);
1899 	if (tvp)
1900 		vput(tvp);
1901 	vrele(fdvp);
1902 	vrele(fvp);
1903 	/*
1904 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1905 	 */
1906 	if (error == ENOENT)
1907 		error = 0;
1908 	return (error);
1909 }
1910 
1911 /*
1912  * nfs file rename rpc called from nfs_remove() above
1913  */
1914 static int
1915 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1916     struct sillyrename *sp)
1917 {
1918 
1919 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1920 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1921 	    scnp->cn_thread));
1922 }
1923 
1924 /*
1925  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1926  */
1927 static int
1928 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1929     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1930     int tnamelen, struct ucred *cred, struct thread *td)
1931 {
1932 	struct nfsvattr fnfsva, tnfsva;
1933 	struct nfsnode *fdnp = VTONFS(fdvp);
1934 	struct nfsnode *tdnp = VTONFS(tdvp);
1935 	int error = 0, fattrflag, tattrflag;
1936 
1937 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1938 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1939 	    &tattrflag, NULL, NULL);
1940 	mtx_lock(&fdnp->n_mtx);
1941 	fdnp->n_flag |= NMODIFIED;
1942 	if (fattrflag != 0) {
1943 		mtx_unlock(&fdnp->n_mtx);
1944 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1945 	} else {
1946 		fdnp->n_attrstamp = 0;
1947 		mtx_unlock(&fdnp->n_mtx);
1948 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1949 	}
1950 	mtx_lock(&tdnp->n_mtx);
1951 	tdnp->n_flag |= NMODIFIED;
1952 	if (tattrflag != 0) {
1953 		mtx_unlock(&tdnp->n_mtx);
1954 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1955 	} else {
1956 		tdnp->n_attrstamp = 0;
1957 		mtx_unlock(&tdnp->n_mtx);
1958 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1959 	}
1960 	if (error && NFS_ISV4(fdvp))
1961 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1962 	return (error);
1963 }
1964 
1965 /*
1966  * nfs hard link create call
1967  */
1968 static int
1969 nfs_link(struct vop_link_args *ap)
1970 {
1971 	struct vnode *vp = ap->a_vp;
1972 	struct vnode *tdvp = ap->a_tdvp;
1973 	struct componentname *cnp = ap->a_cnp;
1974 	struct nfsnode *np, *tdnp;
1975 	struct nfsvattr nfsva, dnfsva;
1976 	int error = 0, attrflag, dattrflag;
1977 
1978 	if (vp->v_mount != tdvp->v_mount) {
1979 		return (EXDEV);
1980 	}
1981 
1982 	/*
1983 	 * Push all writes to the server, so that the attribute cache
1984 	 * doesn't get "out of sync" with the server.
1985 	 * XXX There should be a better way!
1986 	 */
1987 	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1988 
1989 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1990 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1991 	    &dattrflag, NULL);
1992 	tdnp = VTONFS(tdvp);
1993 	mtx_lock(&tdnp->n_mtx);
1994 	tdnp->n_flag |= NMODIFIED;
1995 	if (dattrflag != 0) {
1996 		mtx_unlock(&tdnp->n_mtx);
1997 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1998 	} else {
1999 		tdnp->n_attrstamp = 0;
2000 		mtx_unlock(&tdnp->n_mtx);
2001 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2002 	}
2003 	if (attrflag)
2004 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2005 	else {
2006 		np = VTONFS(vp);
2007 		mtx_lock(&np->n_mtx);
2008 		np->n_attrstamp = 0;
2009 		mtx_unlock(&np->n_mtx);
2010 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2011 	}
2012 	/*
2013 	 * If negative lookup caching is enabled, I might as well
2014 	 * add an entry for this node. Not necessary for correctness,
2015 	 * but if negative caching is enabled, then the system
2016 	 * must care about lookup caching hit rate, so...
2017 	 */
2018 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2019 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2020 		cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2021 	}
2022 	if (error && NFS_ISV4(vp))
2023 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2024 		    (gid_t)0);
2025 	return (error);
2026 }
2027 
2028 /*
2029  * nfs symbolic link create call
2030  */
2031 static int
2032 nfs_symlink(struct vop_symlink_args *ap)
2033 {
2034 	struct vnode *dvp = ap->a_dvp;
2035 	struct vattr *vap = ap->a_vap;
2036 	struct componentname *cnp = ap->a_cnp;
2037 	struct nfsvattr nfsva, dnfsva;
2038 	struct nfsfh *nfhp;
2039 	struct nfsnode *np = NULL, *dnp;
2040 	struct vnode *newvp = NULL;
2041 	int error = 0, attrflag, dattrflag, ret;
2042 
2043 	vap->va_type = VLNK;
2044 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2045 	    ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2046 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2047 	if (nfhp) {
2048 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2049 		    &np, NULL, LK_EXCLUSIVE);
2050 		if (!ret)
2051 			newvp = NFSTOV(np);
2052 		else if (!error)
2053 			error = ret;
2054 	}
2055 	if (newvp != NULL) {
2056 		if (attrflag)
2057 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2058 			    0, 1);
2059 	} else if (!error) {
2060 		/*
2061 		 * If we do not have an error and we could not extract the
2062 		 * newvp from the response due to the request being NFSv2, we
2063 		 * have to do a lookup in order to obtain a newvp to return.
2064 		 */
2065 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2066 		    cnp->cn_cred, cnp->cn_thread, &np);
2067 		if (!error)
2068 			newvp = NFSTOV(np);
2069 	}
2070 	if (error) {
2071 		if (newvp)
2072 			vput(newvp);
2073 		if (NFS_ISV4(dvp))
2074 			error = nfscl_maperr(cnp->cn_thread, error,
2075 			    vap->va_uid, vap->va_gid);
2076 	} else {
2077 		*ap->a_vpp = newvp;
2078 	}
2079 
2080 	dnp = VTONFS(dvp);
2081 	mtx_lock(&dnp->n_mtx);
2082 	dnp->n_flag |= NMODIFIED;
2083 	if (dattrflag != 0) {
2084 		mtx_unlock(&dnp->n_mtx);
2085 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2086 	} else {
2087 		dnp->n_attrstamp = 0;
2088 		mtx_unlock(&dnp->n_mtx);
2089 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2090 	}
2091 	/*
2092 	 * If negative lookup caching is enabled, I might as well
2093 	 * add an entry for this node. Not necessary for correctness,
2094 	 * but if negative caching is enabled, then the system
2095 	 * must care about lookup caching hit rate, so...
2096 	 */
2097 	if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2098 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2099 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2100 	}
2101 	return (error);
2102 }
2103 
2104 /*
2105  * nfs make dir call
2106  */
2107 static int
2108 nfs_mkdir(struct vop_mkdir_args *ap)
2109 {
2110 	struct vnode *dvp = ap->a_dvp;
2111 	struct vattr *vap = ap->a_vap;
2112 	struct componentname *cnp = ap->a_cnp;
2113 	struct nfsnode *np = NULL, *dnp;
2114 	struct vnode *newvp = NULL;
2115 	struct vattr vattr;
2116 	struct nfsfh *nfhp;
2117 	struct nfsvattr nfsva, dnfsva;
2118 	int error = 0, attrflag, dattrflag, ret;
2119 
2120 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2121 		return (error);
2122 	vap->va_type = VDIR;
2123 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2124 	    vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2125 	    &attrflag, &dattrflag, NULL);
2126 	dnp = VTONFS(dvp);
2127 	mtx_lock(&dnp->n_mtx);
2128 	dnp->n_flag |= NMODIFIED;
2129 	if (dattrflag != 0) {
2130 		mtx_unlock(&dnp->n_mtx);
2131 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2132 	} else {
2133 		dnp->n_attrstamp = 0;
2134 		mtx_unlock(&dnp->n_mtx);
2135 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2136 	}
2137 	if (nfhp) {
2138 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2139 		    &np, NULL, LK_EXCLUSIVE);
2140 		if (!ret) {
2141 			newvp = NFSTOV(np);
2142 			if (attrflag)
2143 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2144 				NULL, 0, 1);
2145 		} else if (!error)
2146 			error = ret;
2147 	}
2148 	if (!error && newvp == NULL) {
2149 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2150 		    cnp->cn_cred, cnp->cn_thread, &np);
2151 		if (!error) {
2152 			newvp = NFSTOV(np);
2153 			if (newvp->v_type != VDIR)
2154 				error = EEXIST;
2155 		}
2156 	}
2157 	if (error) {
2158 		if (newvp)
2159 			vput(newvp);
2160 		if (NFS_ISV4(dvp))
2161 			error = nfscl_maperr(cnp->cn_thread, error,
2162 			    vap->va_uid, vap->va_gid);
2163 	} else {
2164 		/*
2165 		 * If negative lookup caching is enabled, I might as well
2166 		 * add an entry for this node. Not necessary for correctness,
2167 		 * but if negative caching is enabled, then the system
2168 		 * must care about lookup caching hit rate, so...
2169 		 */
2170 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2171 		    (cnp->cn_flags & MAKEENTRY) &&
2172 		    attrflag != 0 && dattrflag != 0)
2173 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2174 			    &dnfsva.na_ctime);
2175 		*ap->a_vpp = newvp;
2176 	}
2177 	return (error);
2178 }
2179 
2180 /*
2181  * nfs remove directory call
2182  */
2183 static int
2184 nfs_rmdir(struct vop_rmdir_args *ap)
2185 {
2186 	struct vnode *vp = ap->a_vp;
2187 	struct vnode *dvp = ap->a_dvp;
2188 	struct componentname *cnp = ap->a_cnp;
2189 	struct nfsnode *dnp;
2190 	struct nfsvattr dnfsva;
2191 	int error, dattrflag;
2192 
2193 	if (dvp == vp)
2194 		return (EINVAL);
2195 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2196 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2197 	dnp = VTONFS(dvp);
2198 	mtx_lock(&dnp->n_mtx);
2199 	dnp->n_flag |= NMODIFIED;
2200 	if (dattrflag != 0) {
2201 		mtx_unlock(&dnp->n_mtx);
2202 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2203 	} else {
2204 		dnp->n_attrstamp = 0;
2205 		mtx_unlock(&dnp->n_mtx);
2206 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2207 	}
2208 
2209 	cache_purge(dvp);
2210 	cache_purge(vp);
2211 	if (error && NFS_ISV4(dvp))
2212 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2213 		    (gid_t)0);
2214 	/*
2215 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2216 	 */
2217 	if (error == ENOENT)
2218 		error = 0;
2219 	return (error);
2220 }
2221 
2222 /*
2223  * nfs readdir call
2224  */
2225 static int
2226 nfs_readdir(struct vop_readdir_args *ap)
2227 {
2228 	struct vnode *vp = ap->a_vp;
2229 	struct nfsnode *np = VTONFS(vp);
2230 	struct uio *uio = ap->a_uio;
2231 	ssize_t tresid;
2232 	int error = 0;
2233 	struct vattr vattr;
2234 
2235 	if (ap->a_eofflag != NULL)
2236 		*ap->a_eofflag = 0;
2237 	if (vp->v_type != VDIR)
2238 		return(EPERM);
2239 
2240 	/*
2241 	 * First, check for hit on the EOF offset cache
2242 	 */
2243 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2244 	    (np->n_flag & NMODIFIED) == 0) {
2245 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2246 			mtx_lock(&np->n_mtx);
2247 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2248 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2249 				mtx_unlock(&np->n_mtx);
2250 				NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2251 				if (ap->a_eofflag != NULL)
2252 					*ap->a_eofflag = 1;
2253 				return (0);
2254 			} else
2255 				mtx_unlock(&np->n_mtx);
2256 		}
2257 	}
2258 
2259 	/*
2260 	 * Call ncl_bioread() to do the real work.
2261 	 */
2262 	tresid = uio->uio_resid;
2263 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2264 
2265 	if (!error && uio->uio_resid == tresid) {
2266 		NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2267 		if (ap->a_eofflag != NULL)
2268 			*ap->a_eofflag = 1;
2269 	}
2270 	return (error);
2271 }
2272 
2273 /*
2274  * Readdir rpc call.
2275  * Called from below the buffer cache by ncl_doio().
2276  */
2277 int
2278 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2279     struct thread *td)
2280 {
2281 	struct nfsvattr nfsva;
2282 	nfsuint64 *cookiep, cookie;
2283 	struct nfsnode *dnp = VTONFS(vp);
2284 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2285 	int error = 0, eof, attrflag;
2286 
2287 	KASSERT(uiop->uio_iovcnt == 1 &&
2288 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2289 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2290 	    ("nfs readdirrpc bad uio"));
2291 
2292 	/*
2293 	 * If there is no cookie, assume directory was stale.
2294 	 */
2295 	ncl_dircookie_lock(dnp);
2296 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2297 	if (cookiep) {
2298 		cookie = *cookiep;
2299 		ncl_dircookie_unlock(dnp);
2300 	} else {
2301 		ncl_dircookie_unlock(dnp);
2302 		return (NFSERR_BAD_COOKIE);
2303 	}
2304 
2305 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2306 		(void)ncl_fsinfo(nmp, vp, cred, td);
2307 
2308 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2309 	    &attrflag, &eof, NULL);
2310 	if (attrflag)
2311 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2312 
2313 	if (!error) {
2314 		/*
2315 		 * We are now either at the end of the directory or have filled
2316 		 * the block.
2317 		 */
2318 		if (eof)
2319 			dnp->n_direofoffset = uiop->uio_offset;
2320 		else {
2321 			if (uiop->uio_resid > 0)
2322 				ncl_printf("EEK! readdirrpc resid > 0\n");
2323 			ncl_dircookie_lock(dnp);
2324 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2325 			*cookiep = cookie;
2326 			ncl_dircookie_unlock(dnp);
2327 		}
2328 	} else if (NFS_ISV4(vp)) {
2329 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2330 	}
2331 	return (error);
2332 }
2333 
2334 /*
2335  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2336  */
2337 int
2338 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2339     struct thread *td)
2340 {
2341 	struct nfsvattr nfsva;
2342 	nfsuint64 *cookiep, cookie;
2343 	struct nfsnode *dnp = VTONFS(vp);
2344 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2345 	int error = 0, attrflag, eof;
2346 
2347 	KASSERT(uiop->uio_iovcnt == 1 &&
2348 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2349 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2350 	    ("nfs readdirplusrpc bad uio"));
2351 
2352 	/*
2353 	 * If there is no cookie, assume directory was stale.
2354 	 */
2355 	ncl_dircookie_lock(dnp);
2356 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2357 	if (cookiep) {
2358 		cookie = *cookiep;
2359 		ncl_dircookie_unlock(dnp);
2360 	} else {
2361 		ncl_dircookie_unlock(dnp);
2362 		return (NFSERR_BAD_COOKIE);
2363 	}
2364 
2365 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2366 		(void)ncl_fsinfo(nmp, vp, cred, td);
2367 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2368 	    &attrflag, &eof, NULL);
2369 	if (attrflag)
2370 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2371 
2372 	if (!error) {
2373 		/*
2374 		 * We are now either at end of the directory or have filled the
2375 		 * the block.
2376 		 */
2377 		if (eof)
2378 			dnp->n_direofoffset = uiop->uio_offset;
2379 		else {
2380 			if (uiop->uio_resid > 0)
2381 				ncl_printf("EEK! readdirplusrpc resid > 0\n");
2382 			ncl_dircookie_lock(dnp);
2383 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2384 			*cookiep = cookie;
2385 			ncl_dircookie_unlock(dnp);
2386 		}
2387 	} else if (NFS_ISV4(vp)) {
2388 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2389 	}
2390 	return (error);
2391 }
2392 
2393 /*
2394  * Silly rename. To make the NFS filesystem that is stateless look a little
2395  * more like the "ufs" a remove of an active vnode is translated to a rename
2396  * to a funny looking filename that is removed by nfs_inactive on the
2397  * nfsnode. There is the potential for another process on a different client
2398  * to create the same funny name between the nfs_lookitup() fails and the
2399  * nfs_rename() completes, but...
2400  */
2401 static int
2402 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2403 {
2404 	struct sillyrename *sp;
2405 	struct nfsnode *np;
2406 	int error;
2407 	short pid;
2408 	unsigned int lticks;
2409 
2410 	cache_purge(dvp);
2411 	np = VTONFS(vp);
2412 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2413 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2414 	    M_NEWNFSREQ, M_WAITOK);
2415 	sp->s_cred = crhold(cnp->cn_cred);
2416 	sp->s_dvp = dvp;
2417 	VREF(dvp);
2418 
2419 	/*
2420 	 * Fudge together a funny name.
2421 	 * Changing the format of the funny name to accomodate more
2422 	 * sillynames per directory.
2423 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2424 	 * CPU ticks since boot.
2425 	 */
2426 	pid = cnp->cn_thread->td_proc->p_pid;
2427 	lticks = (unsigned int)ticks;
2428 	for ( ; ; ) {
2429 		sp->s_namlen = sprintf(sp->s_name,
2430 				       ".nfs.%08x.%04x4.4", lticks,
2431 				       pid);
2432 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2433 				 cnp->cn_thread, NULL))
2434 			break;
2435 		lticks++;
2436 	}
2437 	error = nfs_renameit(dvp, vp, cnp, sp);
2438 	if (error)
2439 		goto bad;
2440 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2441 		cnp->cn_thread, &np);
2442 	np->n_sillyrename = sp;
2443 	return (0);
2444 bad:
2445 	vrele(sp->s_dvp);
2446 	crfree(sp->s_cred);
2447 	free((caddr_t)sp, M_NEWNFSREQ);
2448 	return (error);
2449 }
2450 
2451 /*
2452  * Look up a file name and optionally either update the file handle or
2453  * allocate an nfsnode, depending on the value of npp.
2454  * npp == NULL	--> just do the lookup
2455  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2456  *			handled too
2457  * *npp != NULL --> update the file handle in the vnode
2458  */
2459 static int
2460 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2461     struct thread *td, struct nfsnode **npp)
2462 {
2463 	struct vnode *newvp = NULL, *vp;
2464 	struct nfsnode *np, *dnp = VTONFS(dvp);
2465 	struct nfsfh *nfhp, *onfhp;
2466 	struct nfsvattr nfsva, dnfsva;
2467 	struct componentname cn;
2468 	int error = 0, attrflag, dattrflag;
2469 	u_int hash;
2470 
2471 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2472 	    &nfhp, &attrflag, &dattrflag, NULL);
2473 	if (dattrflag)
2474 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2475 	if (npp && !error) {
2476 		if (*npp != NULL) {
2477 		    np = *npp;
2478 		    vp = NFSTOV(np);
2479 		    /*
2480 		     * For NFSv4, check to see if it is the same name and
2481 		     * replace the name, if it is different.
2482 		     */
2483 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2484 			(np->n_v4->n4_namelen != len ||
2485 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2486 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2487 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2488 			 dnp->n_fhp->nfh_len))) {
2489 #ifdef notdef
2490 { char nnn[100]; int nnnl;
2491 nnnl = (len < 100) ? len : 99;
2492 bcopy(name, nnn, nnnl);
2493 nnn[nnnl] = '\0';
2494 printf("replace=%s\n",nnn);
2495 }
2496 #endif
2497 			    FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2498 			    MALLOC(np->n_v4, struct nfsv4node *,
2499 				sizeof (struct nfsv4node) +
2500 				dnp->n_fhp->nfh_len + len - 1,
2501 				M_NFSV4NODE, M_WAITOK);
2502 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2503 			    np->n_v4->n4_namelen = len;
2504 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2505 				dnp->n_fhp->nfh_len);
2506 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2507 		    }
2508 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2509 			FNV1_32_INIT);
2510 		    onfhp = np->n_fhp;
2511 		    /*
2512 		     * Rehash node for new file handle.
2513 		     */
2514 		    vfs_hash_rehash(vp, hash);
2515 		    np->n_fhp = nfhp;
2516 		    if (onfhp != NULL)
2517 			FREE((caddr_t)onfhp, M_NFSFH);
2518 		    newvp = NFSTOV(np);
2519 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2520 		    FREE((caddr_t)nfhp, M_NFSFH);
2521 		    VREF(dvp);
2522 		    newvp = dvp;
2523 		} else {
2524 		    cn.cn_nameptr = name;
2525 		    cn.cn_namelen = len;
2526 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2527 			&np, NULL, LK_EXCLUSIVE);
2528 		    if (error)
2529 			return (error);
2530 		    newvp = NFSTOV(np);
2531 		}
2532 		if (!attrflag && *npp == NULL) {
2533 			if (newvp == dvp)
2534 				vrele(newvp);
2535 			else
2536 				vput(newvp);
2537 			return (ENOENT);
2538 		}
2539 		if (attrflag)
2540 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2541 			    0, 1);
2542 	}
2543 	if (npp && *npp == NULL) {
2544 		if (error) {
2545 			if (newvp) {
2546 				if (newvp == dvp)
2547 					vrele(newvp);
2548 				else
2549 					vput(newvp);
2550 			}
2551 		} else
2552 			*npp = np;
2553 	}
2554 	if (error && NFS_ISV4(dvp))
2555 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2556 	return (error);
2557 }
2558 
2559 /*
2560  * Nfs Version 3 and 4 commit rpc
2561  */
2562 int
2563 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2564    struct thread *td)
2565 {
2566 	struct nfsvattr nfsva;
2567 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2568 	int error, attrflag;
2569 
2570 	mtx_lock(&nmp->nm_mtx);
2571 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2572 		mtx_unlock(&nmp->nm_mtx);
2573 		return (0);
2574 	}
2575 	mtx_unlock(&nmp->nm_mtx);
2576 	error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2577 	    &attrflag, NULL);
2578 	if (attrflag != 0)
2579 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2580 		    0, 1);
2581 	if (error != 0 && NFS_ISV4(vp))
2582 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2583 	return (error);
2584 }
2585 
2586 /*
2587  * Strategy routine.
2588  * For async requests when nfsiod(s) are running, queue the request by
2589  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2590  * request.
2591  */
2592 static int
2593 nfs_strategy(struct vop_strategy_args *ap)
2594 {
2595 	struct buf *bp = ap->a_bp;
2596 	struct ucred *cr;
2597 
2598 	KASSERT(!(bp->b_flags & B_DONE),
2599 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2600 	BUF_ASSERT_HELD(bp);
2601 
2602 	if (bp->b_iocmd == BIO_READ)
2603 		cr = bp->b_rcred;
2604 	else
2605 		cr = bp->b_wcred;
2606 
2607 	/*
2608 	 * If the op is asynchronous and an i/o daemon is waiting
2609 	 * queue the request, wake it up and wait for completion
2610 	 * otherwise just do it ourselves.
2611 	 */
2612 	if ((bp->b_flags & B_ASYNC) == 0 ||
2613 	    ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2614 		(void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2615 	return (0);
2616 }
2617 
2618 /*
2619  * fsync vnode op. Just call ncl_flush() with commit == 1.
2620  */
2621 /* ARGSUSED */
2622 static int
2623 nfs_fsync(struct vop_fsync_args *ap)
2624 {
2625 
2626 	if (ap->a_vp->v_type != VREG) {
2627 		/*
2628 		 * For NFS, metadata is changed synchronously on the server,
2629 		 * so there is nothing to flush. Also, ncl_flush() clears
2630 		 * the NMODIFIED flag and that shouldn't be done here for
2631 		 * directories.
2632 		 */
2633 		return (0);
2634 	}
2635 	return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2636 }
2637 
2638 /*
2639  * Flush all the blocks associated with a vnode.
2640  * 	Walk through the buffer pool and push any dirty pages
2641  *	associated with the vnode.
2642  * If the called_from_renewthread argument is TRUE, it has been called
2643  * from the NFSv4 renew thread and, as such, cannot block indefinitely
2644  * waiting for a buffer write to complete.
2645  */
2646 int
2647 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2648     int commit, int called_from_renewthread)
2649 {
2650 	struct nfsnode *np = VTONFS(vp);
2651 	struct buf *bp;
2652 	int i;
2653 	struct buf *nbp;
2654 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2655 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2656 	int passone = 1, trycnt = 0;
2657 	u_quad_t off, endoff, toff;
2658 	struct ucred* wcred = NULL;
2659 	struct buf **bvec = NULL;
2660 	struct bufobj *bo;
2661 #ifndef NFS_COMMITBVECSIZ
2662 #define	NFS_COMMITBVECSIZ	20
2663 #endif
2664 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2665 	int bvecsize = 0, bveccount;
2666 
2667 	if (called_from_renewthread != 0)
2668 		slptimeo = hz;
2669 	if (nmp->nm_flag & NFSMNT_INT)
2670 		slpflag = PCATCH;
2671 	if (!commit)
2672 		passone = 0;
2673 	bo = &vp->v_bufobj;
2674 	/*
2675 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2676 	 * server, but has not been committed to stable storage on the server
2677 	 * yet. On the first pass, the byte range is worked out and the commit
2678 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2679 	 * job.
2680 	 */
2681 again:
2682 	off = (u_quad_t)-1;
2683 	endoff = 0;
2684 	bvecpos = 0;
2685 	if (NFS_ISV34(vp) && commit) {
2686 		if (bvec != NULL && bvec != bvec_on_stack)
2687 			free(bvec, M_TEMP);
2688 		/*
2689 		 * Count up how many buffers waiting for a commit.
2690 		 */
2691 		bveccount = 0;
2692 		BO_LOCK(bo);
2693 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2694 			if (!BUF_ISLOCKED(bp) &&
2695 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2696 				== (B_DELWRI | B_NEEDCOMMIT))
2697 				bveccount++;
2698 		}
2699 		/*
2700 		 * Allocate space to remember the list of bufs to commit.  It is
2701 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2702 		 * If we can't get memory (for whatever reason), we will end up
2703 		 * committing the buffers one-by-one in the loop below.
2704 		 */
2705 		if (bveccount > NFS_COMMITBVECSIZ) {
2706 			/*
2707 			 * Release the vnode interlock to avoid a lock
2708 			 * order reversal.
2709 			 */
2710 			BO_UNLOCK(bo);
2711 			bvec = (struct buf **)
2712 				malloc(bveccount * sizeof(struct buf *),
2713 				       M_TEMP, M_NOWAIT);
2714 			BO_LOCK(bo);
2715 			if (bvec == NULL) {
2716 				bvec = bvec_on_stack;
2717 				bvecsize = NFS_COMMITBVECSIZ;
2718 			} else
2719 				bvecsize = bveccount;
2720 		} else {
2721 			bvec = bvec_on_stack;
2722 			bvecsize = NFS_COMMITBVECSIZ;
2723 		}
2724 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2725 			if (bvecpos >= bvecsize)
2726 				break;
2727 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2728 				nbp = TAILQ_NEXT(bp, b_bobufs);
2729 				continue;
2730 			}
2731 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2732 			    (B_DELWRI | B_NEEDCOMMIT)) {
2733 				BUF_UNLOCK(bp);
2734 				nbp = TAILQ_NEXT(bp, b_bobufs);
2735 				continue;
2736 			}
2737 			BO_UNLOCK(bo);
2738 			bremfree(bp);
2739 			/*
2740 			 * Work out if all buffers are using the same cred
2741 			 * so we can deal with them all with one commit.
2742 			 *
2743 			 * NOTE: we are not clearing B_DONE here, so we have
2744 			 * to do it later on in this routine if we intend to
2745 			 * initiate I/O on the bp.
2746 			 *
2747 			 * Note: to avoid loopback deadlocks, we do not
2748 			 * assign b_runningbufspace.
2749 			 */
2750 			if (wcred == NULL)
2751 				wcred = bp->b_wcred;
2752 			else if (wcred != bp->b_wcred)
2753 				wcred = NOCRED;
2754 			vfs_busy_pages(bp, 1);
2755 
2756 			BO_LOCK(bo);
2757 			/*
2758 			 * bp is protected by being locked, but nbp is not
2759 			 * and vfs_busy_pages() may sleep.  We have to
2760 			 * recalculate nbp.
2761 			 */
2762 			nbp = TAILQ_NEXT(bp, b_bobufs);
2763 
2764 			/*
2765 			 * A list of these buffers is kept so that the
2766 			 * second loop knows which buffers have actually
2767 			 * been committed. This is necessary, since there
2768 			 * may be a race between the commit rpc and new
2769 			 * uncommitted writes on the file.
2770 			 */
2771 			bvec[bvecpos++] = bp;
2772 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2773 				bp->b_dirtyoff;
2774 			if (toff < off)
2775 				off = toff;
2776 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2777 			if (toff > endoff)
2778 				endoff = toff;
2779 		}
2780 		BO_UNLOCK(bo);
2781 	}
2782 	if (bvecpos > 0) {
2783 		/*
2784 		 * Commit data on the server, as required.
2785 		 * If all bufs are using the same wcred, then use that with
2786 		 * one call for all of them, otherwise commit each one
2787 		 * separately.
2788 		 */
2789 		if (wcred != NOCRED)
2790 			retv = ncl_commit(vp, off, (int)(endoff - off),
2791 					  wcred, td);
2792 		else {
2793 			retv = 0;
2794 			for (i = 0; i < bvecpos; i++) {
2795 				off_t off, size;
2796 				bp = bvec[i];
2797 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2798 					bp->b_dirtyoff;
2799 				size = (u_quad_t)(bp->b_dirtyend
2800 						  - bp->b_dirtyoff);
2801 				retv = ncl_commit(vp, off, (int)size,
2802 						  bp->b_wcred, td);
2803 				if (retv) break;
2804 			}
2805 		}
2806 
2807 		if (retv == NFSERR_STALEWRITEVERF)
2808 			ncl_clearcommit(vp->v_mount);
2809 
2810 		/*
2811 		 * Now, either mark the blocks I/O done or mark the
2812 		 * blocks dirty, depending on whether the commit
2813 		 * succeeded.
2814 		 */
2815 		for (i = 0; i < bvecpos; i++) {
2816 			bp = bvec[i];
2817 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2818 			if (retv) {
2819 				/*
2820 				 * Error, leave B_DELWRI intact
2821 				 */
2822 				vfs_unbusy_pages(bp);
2823 				brelse(bp);
2824 			} else {
2825 				/*
2826 				 * Success, remove B_DELWRI ( bundirty() ).
2827 				 *
2828 				 * b_dirtyoff/b_dirtyend seem to be NFS
2829 				 * specific.  We should probably move that
2830 				 * into bundirty(). XXX
2831 				 */
2832 				bufobj_wref(bo);
2833 				bp->b_flags |= B_ASYNC;
2834 				bundirty(bp);
2835 				bp->b_flags &= ~B_DONE;
2836 				bp->b_ioflags &= ~BIO_ERROR;
2837 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2838 				bufdone(bp);
2839 			}
2840 		}
2841 	}
2842 
2843 	/*
2844 	 * Start/do any write(s) that are required.
2845 	 */
2846 loop:
2847 	BO_LOCK(bo);
2848 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2849 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2850 			if (waitfor != MNT_WAIT || passone)
2851 				continue;
2852 
2853 			error = BUF_TIMELOCK(bp,
2854 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2855 			    BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
2856 			if (error == 0) {
2857 				BUF_UNLOCK(bp);
2858 				goto loop;
2859 			}
2860 			if (error == ENOLCK) {
2861 				error = 0;
2862 				goto loop;
2863 			}
2864 			if (called_from_renewthread != 0) {
2865 				/*
2866 				 * Return EIO so the flush will be retried
2867 				 * later.
2868 				 */
2869 				error = EIO;
2870 				goto done;
2871 			}
2872 			if (newnfs_sigintr(nmp, td)) {
2873 				error = EINTR;
2874 				goto done;
2875 			}
2876 			if (slpflag == PCATCH) {
2877 				slpflag = 0;
2878 				slptimeo = 2 * hz;
2879 			}
2880 			goto loop;
2881 		}
2882 		if ((bp->b_flags & B_DELWRI) == 0)
2883 			panic("nfs_fsync: not dirty");
2884 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2885 			BUF_UNLOCK(bp);
2886 			continue;
2887 		}
2888 		BO_UNLOCK(bo);
2889 		bremfree(bp);
2890 		if (passone || !commit)
2891 		    bp->b_flags |= B_ASYNC;
2892 		else
2893 		    bp->b_flags |= B_ASYNC;
2894 		bwrite(bp);
2895 		if (newnfs_sigintr(nmp, td)) {
2896 			error = EINTR;
2897 			goto done;
2898 		}
2899 		goto loop;
2900 	}
2901 	if (passone) {
2902 		passone = 0;
2903 		BO_UNLOCK(bo);
2904 		goto again;
2905 	}
2906 	if (waitfor == MNT_WAIT) {
2907 		while (bo->bo_numoutput) {
2908 			error = bufobj_wwait(bo, slpflag, slptimeo);
2909 			if (error) {
2910 			    BO_UNLOCK(bo);
2911 			    if (called_from_renewthread != 0) {
2912 				/*
2913 				 * Return EIO so that the flush will be
2914 				 * retried later.
2915 				 */
2916 				error = EIO;
2917 				goto done;
2918 			    }
2919 			    error = newnfs_sigintr(nmp, td);
2920 			    if (error)
2921 				goto done;
2922 			    if (slpflag == PCATCH) {
2923 				slpflag = 0;
2924 				slptimeo = 2 * hz;
2925 			    }
2926 			    BO_LOCK(bo);
2927 			}
2928 		}
2929 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
2930 			BO_UNLOCK(bo);
2931 			goto loop;
2932 		}
2933 		/*
2934 		 * Wait for all the async IO requests to drain
2935 		 */
2936 		BO_UNLOCK(bo);
2937 		mtx_lock(&np->n_mtx);
2938 		while (np->n_directio_asyncwr > 0) {
2939 			np->n_flag |= NFSYNCWAIT;
2940 			error = newnfs_msleep(td, &np->n_directio_asyncwr,
2941 			    &np->n_mtx, slpflag | (PRIBIO + 1),
2942 			    "nfsfsync", 0);
2943 			if (error) {
2944 				if (newnfs_sigintr(nmp, td)) {
2945 					mtx_unlock(&np->n_mtx);
2946 					error = EINTR;
2947 					goto done;
2948 				}
2949 			}
2950 		}
2951 		mtx_unlock(&np->n_mtx);
2952 	} else
2953 		BO_UNLOCK(bo);
2954 	if (NFSHASPNFS(nmp)) {
2955 		nfscl_layoutcommit(vp, td);
2956 		/*
2957 		 * Invalidate the attribute cache, since writes to a DS
2958 		 * won't update the size attribute.
2959 		 */
2960 		mtx_lock(&np->n_mtx);
2961 		np->n_attrstamp = 0;
2962 	} else
2963 		mtx_lock(&np->n_mtx);
2964 	if (np->n_flag & NWRITEERR) {
2965 		error = np->n_error;
2966 		np->n_flag &= ~NWRITEERR;
2967 	}
2968   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
2969 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2970   		np->n_flag &= ~NMODIFIED;
2971 	mtx_unlock(&np->n_mtx);
2972 done:
2973 	if (bvec != NULL && bvec != bvec_on_stack)
2974 		free(bvec, M_TEMP);
2975 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2976 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2977 	     np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2978 		/* try, try again... */
2979 		passone = 1;
2980 		wcred = NULL;
2981 		bvec = NULL;
2982 		bvecsize = 0;
2983 printf("try%d\n", trycnt);
2984 		goto again;
2985 	}
2986 	return (error);
2987 }
2988 
2989 /*
2990  * NFS advisory byte-level locks.
2991  */
2992 static int
2993 nfs_advlock(struct vop_advlock_args *ap)
2994 {
2995 	struct vnode *vp = ap->a_vp;
2996 	struct ucred *cred;
2997 	struct nfsnode *np = VTONFS(ap->a_vp);
2998 	struct proc *p = (struct proc *)ap->a_id;
2999 	struct thread *td = curthread;	/* XXX */
3000 	struct vattr va;
3001 	int ret, error = EOPNOTSUPP;
3002 	u_quad_t size;
3003 
3004 	if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3005 		if (vp->v_type != VREG)
3006 			return (EINVAL);
3007 		if ((ap->a_flags & F_POSIX) != 0)
3008 			cred = p->p_ucred;
3009 		else
3010 			cred = td->td_ucred;
3011 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3012 		if (vp->v_iflag & VI_DOOMED) {
3013 			NFSVOPUNLOCK(vp, 0);
3014 			return (EBADF);
3015 		}
3016 
3017 		/*
3018 		 * If this is unlocking a write locked region, flush and
3019 		 * commit them before unlocking. This is required by
3020 		 * RFC3530 Sec. 9.3.2.
3021 		 */
3022 		if (ap->a_op == F_UNLCK &&
3023 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3024 		    ap->a_flags))
3025 			(void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
3026 
3027 		/*
3028 		 * Loop around doing the lock op, while a blocking lock
3029 		 * must wait for the lock op to succeed.
3030 		 */
3031 		do {
3032 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3033 			    ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3034 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3035 			    ap->a_op == F_SETLK) {
3036 				NFSVOPUNLOCK(vp, 0);
3037 				error = nfs_catnap(PZERO | PCATCH, ret,
3038 				    "ncladvl");
3039 				if (error)
3040 					return (EINTR);
3041 				NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3042 				if (vp->v_iflag & VI_DOOMED) {
3043 					NFSVOPUNLOCK(vp, 0);
3044 					return (EBADF);
3045 				}
3046 			}
3047 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3048 		     ap->a_op == F_SETLK);
3049 		if (ret == NFSERR_DENIED) {
3050 			NFSVOPUNLOCK(vp, 0);
3051 			return (EAGAIN);
3052 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3053 			NFSVOPUNLOCK(vp, 0);
3054 			return (ret);
3055 		} else if (ret != 0) {
3056 			NFSVOPUNLOCK(vp, 0);
3057 			return (EACCES);
3058 		}
3059 
3060 		/*
3061 		 * Now, if we just got a lock, invalidate data in the buffer
3062 		 * cache, as required, so that the coherency conforms with
3063 		 * RFC3530 Sec. 9.3.2.
3064 		 */
3065 		if (ap->a_op == F_SETLK) {
3066 			if ((np->n_flag & NMODIFIED) == 0) {
3067 				np->n_attrstamp = 0;
3068 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3069 				ret = VOP_GETATTR(vp, &va, cred);
3070 			}
3071 			if ((np->n_flag & NMODIFIED) || ret ||
3072 			    np->n_change != va.va_filerev) {
3073 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3074 				np->n_attrstamp = 0;
3075 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3076 				ret = VOP_GETATTR(vp, &va, cred);
3077 				if (!ret) {
3078 					np->n_mtime = va.va_mtime;
3079 					np->n_change = va.va_filerev;
3080 				}
3081 			}
3082 		}
3083 		NFSVOPUNLOCK(vp, 0);
3084 		return (0);
3085 	} else if (!NFS_ISV4(vp)) {
3086 		error = NFSVOPLOCK(vp, LK_SHARED);
3087 		if (error)
3088 			return (error);
3089 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3090 			size = VTONFS(vp)->n_size;
3091 			NFSVOPUNLOCK(vp, 0);
3092 			error = lf_advlock(ap, &(vp->v_lockf), size);
3093 		} else {
3094 			if (nfs_advlock_p != NULL)
3095 				error = nfs_advlock_p(ap);
3096 			else {
3097 				NFSVOPUNLOCK(vp, 0);
3098 				error = ENOLCK;
3099 			}
3100 		}
3101 	}
3102 	return (error);
3103 }
3104 
3105 /*
3106  * NFS advisory byte-level locks.
3107  */
3108 static int
3109 nfs_advlockasync(struct vop_advlockasync_args *ap)
3110 {
3111 	struct vnode *vp = ap->a_vp;
3112 	u_quad_t size;
3113 	int error;
3114 
3115 	if (NFS_ISV4(vp))
3116 		return (EOPNOTSUPP);
3117 	error = NFSVOPLOCK(vp, LK_SHARED);
3118 	if (error)
3119 		return (error);
3120 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3121 		size = VTONFS(vp)->n_size;
3122 		NFSVOPUNLOCK(vp, 0);
3123 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
3124 	} else {
3125 		NFSVOPUNLOCK(vp, 0);
3126 		error = EOPNOTSUPP;
3127 	}
3128 	return (error);
3129 }
3130 
3131 /*
3132  * Print out the contents of an nfsnode.
3133  */
3134 static int
3135 nfs_print(struct vop_print_args *ap)
3136 {
3137 	struct vnode *vp = ap->a_vp;
3138 	struct nfsnode *np = VTONFS(vp);
3139 
3140 	ncl_printf("\tfileid %ld fsid 0x%x",
3141 	   np->n_vattr.na_fileid, np->n_vattr.na_fsid);
3142 	if (vp->v_type == VFIFO)
3143 		fifo_printinfo(vp);
3144 	printf("\n");
3145 	return (0);
3146 }
3147 
3148 /*
3149  * This is the "real" nfs::bwrite(struct buf*).
3150  * We set B_CACHE if this is a VMIO buffer.
3151  */
3152 int
3153 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3154 {
3155 	int s;
3156 	int oldflags = bp->b_flags;
3157 #if 0
3158 	int retv = 1;
3159 	off_t off;
3160 #endif
3161 
3162 	BUF_ASSERT_HELD(bp);
3163 
3164 	if (bp->b_flags & B_INVAL) {
3165 		brelse(bp);
3166 		return(0);
3167 	}
3168 
3169 	bp->b_flags |= B_CACHE;
3170 
3171 	/*
3172 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3173 	 */
3174 
3175 	s = splbio();
3176 	bundirty(bp);
3177 	bp->b_flags &= ~B_DONE;
3178 	bp->b_ioflags &= ~BIO_ERROR;
3179 	bp->b_iocmd = BIO_WRITE;
3180 
3181 	bufobj_wref(bp->b_bufobj);
3182 	curthread->td_ru.ru_oublock++;
3183 	splx(s);
3184 
3185 	/*
3186 	 * Note: to avoid loopback deadlocks, we do not
3187 	 * assign b_runningbufspace.
3188 	 */
3189 	vfs_busy_pages(bp, 1);
3190 
3191 	BUF_KERNPROC(bp);
3192 	bp->b_iooffset = dbtob(bp->b_blkno);
3193 	bstrategy(bp);
3194 
3195 	if( (oldflags & B_ASYNC) == 0) {
3196 		int rtval = bufwait(bp);
3197 
3198 		if (oldflags & B_DELWRI) {
3199 			s = splbio();
3200 			reassignbuf(bp);
3201 			splx(s);
3202 		}
3203 		brelse(bp);
3204 		return (rtval);
3205 	}
3206 
3207 	return (0);
3208 }
3209 
3210 /*
3211  * nfs special file access vnode op.
3212  * Essentially just get vattr and then imitate iaccess() since the device is
3213  * local to the client.
3214  */
3215 static int
3216 nfsspec_access(struct vop_access_args *ap)
3217 {
3218 	struct vattr *vap;
3219 	struct ucred *cred = ap->a_cred;
3220 	struct vnode *vp = ap->a_vp;
3221 	accmode_t accmode = ap->a_accmode;
3222 	struct vattr vattr;
3223 	int error;
3224 
3225 	/*
3226 	 * Disallow write attempts on filesystems mounted read-only;
3227 	 * unless the file is a socket, fifo, or a block or character
3228 	 * device resident on the filesystem.
3229 	 */
3230 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3231 		switch (vp->v_type) {
3232 		case VREG:
3233 		case VDIR:
3234 		case VLNK:
3235 			return (EROFS);
3236 		default:
3237 			break;
3238 		}
3239 	}
3240 	vap = &vattr;
3241 	error = VOP_GETATTR(vp, vap, cred);
3242 	if (error)
3243 		goto out;
3244 	error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3245 	    accmode, cred, NULL);
3246 out:
3247 	return error;
3248 }
3249 
3250 /*
3251  * Read wrapper for fifos.
3252  */
3253 static int
3254 nfsfifo_read(struct vop_read_args *ap)
3255 {
3256 	struct nfsnode *np = VTONFS(ap->a_vp);
3257 	int error;
3258 
3259 	/*
3260 	 * Set access flag.
3261 	 */
3262 	mtx_lock(&np->n_mtx);
3263 	np->n_flag |= NACC;
3264 	vfs_timestamp(&np->n_atim);
3265 	mtx_unlock(&np->n_mtx);
3266 	error = fifo_specops.vop_read(ap);
3267 	return error;
3268 }
3269 
3270 /*
3271  * Write wrapper for fifos.
3272  */
3273 static int
3274 nfsfifo_write(struct vop_write_args *ap)
3275 {
3276 	struct nfsnode *np = VTONFS(ap->a_vp);
3277 
3278 	/*
3279 	 * Set update flag.
3280 	 */
3281 	mtx_lock(&np->n_mtx);
3282 	np->n_flag |= NUPD;
3283 	vfs_timestamp(&np->n_mtim);
3284 	mtx_unlock(&np->n_mtx);
3285 	return(fifo_specops.vop_write(ap));
3286 }
3287 
3288 /*
3289  * Close wrapper for fifos.
3290  *
3291  * Update the times on the nfsnode then do fifo close.
3292  */
3293 static int
3294 nfsfifo_close(struct vop_close_args *ap)
3295 {
3296 	struct vnode *vp = ap->a_vp;
3297 	struct nfsnode *np = VTONFS(vp);
3298 	struct vattr vattr;
3299 	struct timespec ts;
3300 
3301 	mtx_lock(&np->n_mtx);
3302 	if (np->n_flag & (NACC | NUPD)) {
3303 		vfs_timestamp(&ts);
3304 		if (np->n_flag & NACC)
3305 			np->n_atim = ts;
3306 		if (np->n_flag & NUPD)
3307 			np->n_mtim = ts;
3308 		np->n_flag |= NCHG;
3309 		if (vrefcnt(vp) == 1 &&
3310 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3311 			VATTR_NULL(&vattr);
3312 			if (np->n_flag & NACC)
3313 				vattr.va_atime = np->n_atim;
3314 			if (np->n_flag & NUPD)
3315 				vattr.va_mtime = np->n_mtim;
3316 			mtx_unlock(&np->n_mtx);
3317 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3318 			goto out;
3319 		}
3320 	}
3321 	mtx_unlock(&np->n_mtx);
3322 out:
3323 	return (fifo_specops.vop_close(ap));
3324 }
3325 
3326 /*
3327  * Just call ncl_writebp() with the force argument set to 1.
3328  *
3329  * NOTE: B_DONE may or may not be set in a_bp on call.
3330  */
3331 static int
3332 nfs_bwrite(struct buf *bp)
3333 {
3334 
3335 	return (ncl_writebp(bp, 1, curthread));
3336 }
3337 
3338 struct buf_ops buf_ops_newnfs = {
3339 	.bop_name	=	"buf_ops_nfs",
3340 	.bop_write	=	nfs_bwrite,
3341 	.bop_strategy	=	bufstrategy,
3342 	.bop_sync	=	bufsync,
3343 	.bop_bdflush	=	bufbdflush,
3344 };
3345 
3346 /*
3347  * Cloned from vop_stdlock(), and then the ugly hack added.
3348  */
3349 static int
3350 nfs_lock1(struct vop_lock1_args *ap)
3351 {
3352 	struct vnode *vp = ap->a_vp;
3353 	int error = 0;
3354 
3355 	/*
3356 	 * Since vfs_hash_get() calls vget() and it will no longer work
3357 	 * for FreeBSD8 with flags == 0, I can only think of this horrible
3358 	 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3359 	 * and then handle it here. All I want for this case is a v_usecount
3360 	 * on the vnode to use for recovery, while another thread might
3361 	 * hold a lock on the vnode. I have the other threads blocked, so
3362 	 * there isn't any race problem.
3363 	 */
3364 	if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3365 		if ((ap->a_flags & LK_INTERLOCK) == 0)
3366 			panic("ncllock1");
3367 		if ((vp->v_iflag & VI_DOOMED))
3368 			error = ENOENT;
3369 		VI_UNLOCK(vp);
3370 		return (error);
3371 	}
3372 	return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3373 	    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3374 	    ap->a_line));
3375 }
3376 
3377 static int
3378 nfs_getacl(struct vop_getacl_args *ap)
3379 {
3380 	int error;
3381 
3382 	if (ap->a_type != ACL_TYPE_NFS4)
3383 		return (EOPNOTSUPP);
3384 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3385 	    NULL);
3386 	if (error > NFSERR_STALE) {
3387 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3388 		error = EPERM;
3389 	}
3390 	return (error);
3391 }
3392 
3393 static int
3394 nfs_setacl(struct vop_setacl_args *ap)
3395 {
3396 	int error;
3397 
3398 	if (ap->a_type != ACL_TYPE_NFS4)
3399 		return (EOPNOTSUPP);
3400 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3401 	    NULL);
3402 	if (error > NFSERR_STALE) {
3403 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3404 		error = EPERM;
3405 	}
3406 	return (error);
3407 }
3408 
3409 /*
3410  * Return POSIX pathconf information applicable to nfs filesystems.
3411  */
3412 static int
3413 nfs_pathconf(struct vop_pathconf_args *ap)
3414 {
3415 	struct nfsv3_pathconf pc;
3416 	struct nfsvattr nfsva;
3417 	struct vnode *vp = ap->a_vp;
3418 	struct thread *td = curthread;
3419 	int attrflag, error;
3420 
3421 	if (NFS_ISV4(vp) || (NFS_ISV3(vp) && (ap->a_name == _PC_LINK_MAX ||
3422 	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3423 	    ap->a_name == _PC_NO_TRUNC))) {
3424 		/*
3425 		 * Since only the above 4 a_names are returned by the NFSv3
3426 		 * Pathconf RPC, there is no point in doing it for others.
3427 		 */
3428 		error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3429 		    &attrflag, NULL);
3430 		if (attrflag != 0)
3431 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3432 			    1);
3433 		if (error != 0)
3434 			return (error);
3435 	} else {
3436 		/*
3437 		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3438 		 * just fake them.
3439 		 */
3440 		pc.pc_linkmax = LINK_MAX;
3441 		pc.pc_namemax = NFS_MAXNAMLEN;
3442 		pc.pc_notrunc = 1;
3443 		pc.pc_chownrestricted = 1;
3444 		pc.pc_caseinsensitive = 0;
3445 		pc.pc_casepreserving = 1;
3446 		error = 0;
3447 	}
3448 	switch (ap->a_name) {
3449 	case _PC_LINK_MAX:
3450 		*ap->a_retval = pc.pc_linkmax;
3451 		break;
3452 	case _PC_NAME_MAX:
3453 		*ap->a_retval = pc.pc_namemax;
3454 		break;
3455 	case _PC_PATH_MAX:
3456 		*ap->a_retval = PATH_MAX;
3457 		break;
3458 	case _PC_PIPE_BUF:
3459 		*ap->a_retval = PIPE_BUF;
3460 		break;
3461 	case _PC_CHOWN_RESTRICTED:
3462 		*ap->a_retval = pc.pc_chownrestricted;
3463 		break;
3464 	case _PC_NO_TRUNC:
3465 		*ap->a_retval = pc.pc_notrunc;
3466 		break;
3467 	case _PC_ACL_EXTENDED:
3468 		*ap->a_retval = 0;
3469 		break;
3470 	case _PC_ACL_NFS4:
3471 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3472 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3473 			*ap->a_retval = 1;
3474 		else
3475 			*ap->a_retval = 0;
3476 		break;
3477 	case _PC_ACL_PATH_MAX:
3478 		if (NFS_ISV4(vp))
3479 			*ap->a_retval = ACL_MAX_ENTRIES;
3480 		else
3481 			*ap->a_retval = 3;
3482 		break;
3483 	case _PC_MAC_PRESENT:
3484 		*ap->a_retval = 0;
3485 		break;
3486 	case _PC_ASYNC_IO:
3487 		/* _PC_ASYNC_IO should have been handled by upper layers. */
3488 		KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3489 		error = EINVAL;
3490 		break;
3491 	case _PC_PRIO_IO:
3492 		*ap->a_retval = 0;
3493 		break;
3494 	case _PC_SYNC_IO:
3495 		*ap->a_retval = 0;
3496 		break;
3497 	case _PC_ALLOC_SIZE_MIN:
3498 		*ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3499 		break;
3500 	case _PC_FILESIZEBITS:
3501 		if (NFS_ISV34(vp))
3502 			*ap->a_retval = 64;
3503 		else
3504 			*ap->a_retval = 32;
3505 		break;
3506 	case _PC_REC_INCR_XFER_SIZE:
3507 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3508 		break;
3509 	case _PC_REC_MAX_XFER_SIZE:
3510 		*ap->a_retval = -1; /* means ``unlimited'' */
3511 		break;
3512 	case _PC_REC_MIN_XFER_SIZE:
3513 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3514 		break;
3515 	case _PC_REC_XFER_ALIGN:
3516 		*ap->a_retval = PAGE_SIZE;
3517 		break;
3518 	case _PC_SYMLINK_MAX:
3519 		*ap->a_retval = NFS_MAXPATHLEN;
3520 		break;
3521 
3522 	default:
3523 		error = EINVAL;
3524 		break;
3525 	}
3526 	return (error);
3527 }
3528 
3529