xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision 3806950135d2c8633ec0764e8807eacc87cf3e10)
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  * 3. 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_inet.h"
43 
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/resourcevar.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/jail.h>
53 #include <sys/malloc.h>
54 #include <sys/mbuf.h>
55 #include <sys/namei.h>
56 #include <sys/socket.h>
57 #include <sys/vnode.h>
58 #include <sys/dirent.h>
59 #include <sys/fcntl.h>
60 #include <sys/lockf.h>
61 #include <sys/stat.h>
62 #include <sys/sysctl.h>
63 #include <sys/signalvar.h>
64 
65 #include <vm/vm.h>
66 #include <vm/vm_extern.h>
67 #include <vm/vm_object.h>
68 
69 #include <fs/nfs/nfsport.h>
70 #include <fs/nfsclient/nfsnode.h>
71 #include <fs/nfsclient/nfsmount.h>
72 #include <fs/nfsclient/nfs.h>
73 #include <fs/nfsclient/nfs_kdtrace.h>
74 
75 #include <net/if.h>
76 #include <netinet/in.h>
77 #include <netinet/in_var.h>
78 
79 #include <nfs/nfs_lock.h>
80 
81 #ifdef KDTRACE_HOOKS
82 #include <sys/dtrace_bsd.h>
83 
84 dtrace_nfsclient_accesscache_flush_probe_func_t
85 		dtrace_nfscl_accesscache_flush_done_probe;
86 uint32_t	nfscl_accesscache_flush_done_id;
87 
88 dtrace_nfsclient_accesscache_get_probe_func_t
89 		dtrace_nfscl_accesscache_get_hit_probe,
90 		dtrace_nfscl_accesscache_get_miss_probe;
91 uint32_t	nfscl_accesscache_get_hit_id;
92 uint32_t	nfscl_accesscache_get_miss_id;
93 
94 dtrace_nfsclient_accesscache_load_probe_func_t
95 		dtrace_nfscl_accesscache_load_done_probe;
96 uint32_t	nfscl_accesscache_load_done_id;
97 #endif /* !KDTRACE_HOOKS */
98 
99 /* Defs */
100 #define	TRUE	1
101 #define	FALSE	0
102 
103 extern struct nfsstatsv1 nfsstatsv1;
104 extern int nfsrv_useacl;
105 extern int nfscl_debuglevel;
106 MALLOC_DECLARE(M_NEWNFSREQ);
107 
108 static vop_read_t	nfsfifo_read;
109 static vop_write_t	nfsfifo_write;
110 static vop_close_t	nfsfifo_close;
111 static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
112 		    struct thread *);
113 static vop_lookup_t	nfs_lookup;
114 static vop_create_t	nfs_create;
115 static vop_mknod_t	nfs_mknod;
116 static vop_open_t	nfs_open;
117 static vop_pathconf_t	nfs_pathconf;
118 static vop_close_t	nfs_close;
119 static vop_access_t	nfs_access;
120 static vop_getattr_t	nfs_getattr;
121 static vop_setattr_t	nfs_setattr;
122 static vop_read_t	nfs_read;
123 static vop_fsync_t	nfs_fsync;
124 static vop_remove_t	nfs_remove;
125 static vop_link_t	nfs_link;
126 static vop_rename_t	nfs_rename;
127 static vop_mkdir_t	nfs_mkdir;
128 static vop_rmdir_t	nfs_rmdir;
129 static vop_symlink_t	nfs_symlink;
130 static vop_readdir_t	nfs_readdir;
131 static vop_strategy_t	nfs_strategy;
132 static	int	nfs_lookitup(struct vnode *, char *, int,
133 		    struct ucred *, struct thread *, struct nfsnode **);
134 static	int	nfs_sillyrename(struct vnode *, struct vnode *,
135 		    struct componentname *);
136 static vop_access_t	nfsspec_access;
137 static vop_readlink_t	nfs_readlink;
138 static vop_print_t	nfs_print;
139 static vop_advlock_t	nfs_advlock;
140 static vop_advlockasync_t nfs_advlockasync;
141 static vop_getacl_t nfs_getacl;
142 static vop_setacl_t nfs_setacl;
143 static vop_set_text_t nfs_set_text;
144 
145 /*
146  * Global vfs data structures for nfs
147  */
148 struct vop_vector newnfs_vnodeops = {
149 	.vop_default =		&default_vnodeops,
150 	.vop_access =		nfs_access,
151 	.vop_advlock =		nfs_advlock,
152 	.vop_advlockasync =	nfs_advlockasync,
153 	.vop_close =		nfs_close,
154 	.vop_create =		nfs_create,
155 	.vop_fsync =		nfs_fsync,
156 	.vop_getattr =		nfs_getattr,
157 	.vop_getpages =		ncl_getpages,
158 	.vop_putpages =		ncl_putpages,
159 	.vop_inactive =		ncl_inactive,
160 	.vop_link =		nfs_link,
161 	.vop_lookup =		nfs_lookup,
162 	.vop_mkdir =		nfs_mkdir,
163 	.vop_mknod =		nfs_mknod,
164 	.vop_open =		nfs_open,
165 	.vop_pathconf =		nfs_pathconf,
166 	.vop_print =		nfs_print,
167 	.vop_read =		nfs_read,
168 	.vop_readdir =		nfs_readdir,
169 	.vop_readlink =		nfs_readlink,
170 	.vop_reclaim =		ncl_reclaim,
171 	.vop_remove =		nfs_remove,
172 	.vop_rename =		nfs_rename,
173 	.vop_rmdir =		nfs_rmdir,
174 	.vop_setattr =		nfs_setattr,
175 	.vop_strategy =		nfs_strategy,
176 	.vop_symlink =		nfs_symlink,
177 	.vop_write =		ncl_write,
178 	.vop_getacl =		nfs_getacl,
179 	.vop_setacl =		nfs_setacl,
180 	.vop_set_text =		nfs_set_text,
181 };
182 
183 struct vop_vector newnfs_fifoops = {
184 	.vop_default =		&fifo_specops,
185 	.vop_access =		nfsspec_access,
186 	.vop_close =		nfsfifo_close,
187 	.vop_fsync =		nfs_fsync,
188 	.vop_getattr =		nfs_getattr,
189 	.vop_inactive =		ncl_inactive,
190 	.vop_print =		nfs_print,
191 	.vop_read =		nfsfifo_read,
192 	.vop_reclaim =		ncl_reclaim,
193 	.vop_setattr =		nfs_setattr,
194 	.vop_write =		nfsfifo_write,
195 };
196 
197 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
198     struct componentname *cnp, struct vattr *vap);
199 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
200     int namelen, struct ucred *cred, struct thread *td);
201 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
202     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
203     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
204 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
205     struct componentname *scnp, struct sillyrename *sp);
206 
207 /*
208  * Global variables
209  */
210 SYSCTL_DECL(_vfs_nfs);
211 
212 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
213 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
214 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
215 
216 static int	nfs_prime_access_cache = 0;
217 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
218 	   &nfs_prime_access_cache, 0,
219 	   "Prime NFS ACCESS cache when fetching attributes");
220 
221 static int	newnfs_commit_on_close = 0;
222 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
223     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
224 
225 static int	nfs_clean_pages_on_close = 1;
226 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
227 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
228 
229 int newnfs_directio_enable = 0;
230 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
231 	   &newnfs_directio_enable, 0, "Enable NFS directio");
232 
233 int nfs_keep_dirty_on_error;
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
235     &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
236 
237 /*
238  * This sysctl allows other processes to mmap a file that has been opened
239  * O_DIRECT by a process.  In general, having processes mmap the file while
240  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
241  * this by default to prevent DoS attacks - to prevent a malicious user from
242  * opening up files O_DIRECT preventing other users from mmap'ing these
243  * files.  "Protected" environments where stricter consistency guarantees are
244  * required can disable this knob.  The process that opened the file O_DIRECT
245  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
246  * meaningful.
247  */
248 int newnfs_directio_allow_mmap = 1;
249 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
250 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
251 
252 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
253 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
254 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
255 
256 /*
257  * SMP Locking Note :
258  * The list of locks after the description of the lock is the ordering
259  * of other locks acquired with the lock held.
260  * np->n_mtx : Protects the fields in the nfsnode.
261        VM Object Lock
262        VI_MTX (acquired indirectly)
263  * nmp->nm_mtx : Protects the fields in the nfsmount.
264        rep->r_mtx
265  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
266  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
267        nmp->nm_mtx
268        rep->r_mtx
269  * rep->r_mtx : Protects the fields in an nfsreq.
270  */
271 
272 static int
273 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
274     struct ucred *cred, u_int32_t *retmode)
275 {
276 	int error = 0, attrflag, i, lrupos;
277 	u_int32_t rmode;
278 	struct nfsnode *np = VTONFS(vp);
279 	struct nfsvattr nfsva;
280 
281 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
282 	    &rmode, NULL);
283 	if (attrflag)
284 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
285 	if (!error) {
286 		lrupos = 0;
287 		mtx_lock(&np->n_mtx);
288 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
289 			if (np->n_accesscache[i].uid == cred->cr_uid) {
290 				np->n_accesscache[i].mode = rmode;
291 				np->n_accesscache[i].stamp = time_second;
292 				break;
293 			}
294 			if (i > 0 && np->n_accesscache[i].stamp <
295 			    np->n_accesscache[lrupos].stamp)
296 				lrupos = i;
297 		}
298 		if (i == NFS_ACCESSCACHESIZE) {
299 			np->n_accesscache[lrupos].uid = cred->cr_uid;
300 			np->n_accesscache[lrupos].mode = rmode;
301 			np->n_accesscache[lrupos].stamp = time_second;
302 		}
303 		mtx_unlock(&np->n_mtx);
304 		if (retmode != NULL)
305 			*retmode = rmode;
306 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
307 	} else if (NFS_ISV4(vp)) {
308 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
309 	}
310 #ifdef KDTRACE_HOOKS
311 	if (error != 0)
312 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
313 		    error);
314 #endif
315 	return (error);
316 }
317 
318 /*
319  * nfs access vnode op.
320  * For nfs version 2, just return ok. File accesses may fail later.
321  * For nfs version 3, use the access rpc to check accessibility. If file modes
322  * are changed on the server, accesses might still fail later.
323  */
324 static int
325 nfs_access(struct vop_access_args *ap)
326 {
327 	struct vnode *vp = ap->a_vp;
328 	int error = 0, i, gotahit;
329 	u_int32_t mode, wmode, rmode;
330 	int v34 = NFS_ISV34(vp);
331 	struct nfsnode *np = VTONFS(vp);
332 
333 	/*
334 	 * Disallow write attempts on filesystems mounted read-only;
335 	 * unless the file is a socket, fifo, or a block or character
336 	 * device resident on the filesystem.
337 	 */
338 	if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
339 	    VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
340 	    VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
341 		switch (vp->v_type) {
342 		case VREG:
343 		case VDIR:
344 		case VLNK:
345 			return (EROFS);
346 		default:
347 			break;
348 		}
349 	}
350 	/*
351 	 * For nfs v3 or v4, check to see if we have done this recently, and if
352 	 * so return our cached result instead of making an ACCESS call.
353 	 * If not, do an access rpc, otherwise you are stuck emulating
354 	 * ufs_access() locally using the vattr. This may not be correct,
355 	 * since the server may apply other access criteria such as
356 	 * client uid-->server uid mapping that we do not know about.
357 	 */
358 	if (v34) {
359 		if (ap->a_accmode & VREAD)
360 			mode = NFSACCESS_READ;
361 		else
362 			mode = 0;
363 		if (vp->v_type != VDIR) {
364 			if (ap->a_accmode & VWRITE)
365 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
366 			if (ap->a_accmode & VAPPEND)
367 				mode |= NFSACCESS_EXTEND;
368 			if (ap->a_accmode & VEXEC)
369 				mode |= NFSACCESS_EXECUTE;
370 			if (ap->a_accmode & VDELETE)
371 				mode |= NFSACCESS_DELETE;
372 		} else {
373 			if (ap->a_accmode & VWRITE)
374 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
375 			if (ap->a_accmode & VAPPEND)
376 				mode |= NFSACCESS_EXTEND;
377 			if (ap->a_accmode & VEXEC)
378 				mode |= NFSACCESS_LOOKUP;
379 			if (ap->a_accmode & VDELETE)
380 				mode |= NFSACCESS_DELETE;
381 			if (ap->a_accmode & VDELETE_CHILD)
382 				mode |= NFSACCESS_MODIFY;
383 		}
384 		/* XXX safety belt, only make blanket request if caching */
385 		if (nfsaccess_cache_timeout > 0) {
386 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
387 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
388 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
389 		} else {
390 			wmode = mode;
391 		}
392 
393 		/*
394 		 * Does our cached result allow us to give a definite yes to
395 		 * this request?
396 		 */
397 		gotahit = 0;
398 		mtx_lock(&np->n_mtx);
399 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
400 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
401 			    if (time_second < (np->n_accesscache[i].stamp
402 				+ nfsaccess_cache_timeout) &&
403 				(np->n_accesscache[i].mode & mode) == mode) {
404 				NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
405 				gotahit = 1;
406 			    }
407 			    break;
408 			}
409 		}
410 		mtx_unlock(&np->n_mtx);
411 #ifdef KDTRACE_HOOKS
412 		if (gotahit != 0)
413 			KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
414 			    ap->a_cred->cr_uid, mode);
415 		else
416 			KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
417 			    ap->a_cred->cr_uid, mode);
418 #endif
419 		if (gotahit == 0) {
420 			/*
421 			 * Either a no, or a don't know.  Go to the wire.
422 			 */
423 			NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
424 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
425 			    ap->a_cred, &rmode);
426 			if (!error &&
427 			    (rmode & mode) != mode)
428 				error = EACCES;
429 		}
430 		return (error);
431 	} else {
432 		if ((error = nfsspec_access(ap)) != 0) {
433 			return (error);
434 		}
435 		/*
436 		 * Attempt to prevent a mapped root from accessing a file
437 		 * which it shouldn't.  We try to read a byte from the file
438 		 * if the user is root and the file is not zero length.
439 		 * After calling nfsspec_access, we should have the correct
440 		 * file size cached.
441 		 */
442 		mtx_lock(&np->n_mtx);
443 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
444 		    && VTONFS(vp)->n_size > 0) {
445 			struct iovec aiov;
446 			struct uio auio;
447 			char buf[1];
448 
449 			mtx_unlock(&np->n_mtx);
450 			aiov.iov_base = buf;
451 			aiov.iov_len = 1;
452 			auio.uio_iov = &aiov;
453 			auio.uio_iovcnt = 1;
454 			auio.uio_offset = 0;
455 			auio.uio_resid = 1;
456 			auio.uio_segflg = UIO_SYSSPACE;
457 			auio.uio_rw = UIO_READ;
458 			auio.uio_td = ap->a_td;
459 
460 			if (vp->v_type == VREG)
461 				error = ncl_readrpc(vp, &auio, ap->a_cred);
462 			else if (vp->v_type == VDIR) {
463 				char* bp;
464 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
465 				aiov.iov_base = bp;
466 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
467 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
468 				    ap->a_td);
469 				free(bp, M_TEMP);
470 			} else if (vp->v_type == VLNK)
471 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
472 			else
473 				error = EACCES;
474 		} else
475 			mtx_unlock(&np->n_mtx);
476 		return (error);
477 	}
478 }
479 
480 
481 /*
482  * nfs open vnode op
483  * Check to see if the type is ok
484  * and that deletion is not in progress.
485  * For paged in text files, you will need to flush the page cache
486  * if consistency is lost.
487  */
488 /* ARGSUSED */
489 static int
490 nfs_open(struct vop_open_args *ap)
491 {
492 	struct vnode *vp = ap->a_vp;
493 	struct nfsnode *np = VTONFS(vp);
494 	struct vattr vattr;
495 	int error;
496 	int fmode = ap->a_mode;
497 	struct ucred *cred;
498 
499 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
500 		return (EOPNOTSUPP);
501 
502 	/*
503 	 * For NFSv4, we need to do the Open Op before cache validation,
504 	 * so that we conform to RFC3530 Sec. 9.3.1.
505 	 */
506 	if (NFS_ISV4(vp)) {
507 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
508 		if (error) {
509 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
510 			    (gid_t)0);
511 			return (error);
512 		}
513 	}
514 
515 	/*
516 	 * Now, if this Open will be doing reading, re-validate/flush the
517 	 * cache, so that Close/Open coherency is maintained.
518 	 */
519 	mtx_lock(&np->n_mtx);
520 	if (np->n_flag & NMODIFIED) {
521 		mtx_unlock(&np->n_mtx);
522 		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
523 		if (error == EINTR || error == EIO) {
524 			if (NFS_ISV4(vp))
525 				(void) nfsrpc_close(vp, 0, ap->a_td);
526 			return (error);
527 		}
528 		mtx_lock(&np->n_mtx);
529 		np->n_attrstamp = 0;
530 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
531 		if (vp->v_type == VDIR)
532 			np->n_direofoffset = 0;
533 		mtx_unlock(&np->n_mtx);
534 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
535 		if (error) {
536 			if (NFS_ISV4(vp))
537 				(void) nfsrpc_close(vp, 0, ap->a_td);
538 			return (error);
539 		}
540 		mtx_lock(&np->n_mtx);
541 		np->n_mtime = vattr.va_mtime;
542 		if (NFS_ISV4(vp))
543 			np->n_change = vattr.va_filerev;
544 	} else {
545 		mtx_unlock(&np->n_mtx);
546 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
547 		if (error) {
548 			if (NFS_ISV4(vp))
549 				(void) nfsrpc_close(vp, 0, ap->a_td);
550 			return (error);
551 		}
552 		mtx_lock(&np->n_mtx);
553 		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
554 		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
555 			if (vp->v_type == VDIR)
556 				np->n_direofoffset = 0;
557 			mtx_unlock(&np->n_mtx);
558 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
559 			if (error == EINTR || error == EIO) {
560 				if (NFS_ISV4(vp))
561 					(void) nfsrpc_close(vp, 0, ap->a_td);
562 				return (error);
563 			}
564 			mtx_lock(&np->n_mtx);
565 			np->n_mtime = vattr.va_mtime;
566 			if (NFS_ISV4(vp))
567 				np->n_change = vattr.va_filerev;
568 		}
569 	}
570 
571 	/*
572 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
573 	 */
574 	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
575 	    (vp->v_type == VREG)) {
576 		if (np->n_directio_opens == 0) {
577 			mtx_unlock(&np->n_mtx);
578 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
579 			if (error) {
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_flag |= NNONCACHE;
586 		}
587 		np->n_directio_opens++;
588 	}
589 
590 	/* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
591 	if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
592 		np->n_flag |= NWRITEOPENED;
593 
594 	/*
595 	 * If this is an open for writing, capture a reference to the
596 	 * credentials, so they can be used by ncl_putpages(). Using
597 	 * these write credentials is preferable to the credentials of
598 	 * whatever thread happens to be doing the VOP_PUTPAGES() since
599 	 * the write RPCs are less likely to fail with EACCES.
600 	 */
601 	if ((fmode & FWRITE) != 0) {
602 		cred = np->n_writecred;
603 		np->n_writecred = crhold(ap->a_cred);
604 	} else
605 		cred = NULL;
606 	mtx_unlock(&np->n_mtx);
607 
608 	if (cred != NULL)
609 		crfree(cred);
610 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
611 	return (0);
612 }
613 
614 /*
615  * nfs close vnode op
616  * What an NFS client should do upon close after writing is a debatable issue.
617  * Most NFS clients push delayed writes to the server upon close, basically for
618  * two reasons:
619  * 1 - So that any write errors may be reported back to the client process
620  *     doing the close system call. By far the two most likely errors are
621  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
622  * 2 - To put a worst case upper bound on cache inconsistency between
623  *     multiple clients for the file.
624  * There is also a consistency problem for Version 2 of the protocol w.r.t.
625  * not being able to tell if other clients are writing a file concurrently,
626  * since there is no way of knowing if the changed modify time in the reply
627  * is only due to the write for this client.
628  * (NFS Version 3 provides weak cache consistency data in the reply that
629  *  should be sufficient to detect and handle this case.)
630  *
631  * The current code does the following:
632  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
633  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
634  *                     or commit them (this satisfies 1 and 2 except for the
635  *                     case where the server crashes after this close but
636  *                     before the commit RPC, which is felt to be "good
637  *                     enough". Changing the last argument to ncl_flush() to
638  *                     a 1 would force a commit operation, if it is felt a
639  *                     commit is necessary now.
640  * for NFS Version 4 - flush the dirty buffers and commit them, if
641  *		       nfscl_mustflush() says this is necessary.
642  *                     It is necessary if there is no write delegation held,
643  *                     in order to satisfy open/close coherency.
644  *                     If the file isn't cached on local stable storage,
645  *                     it may be necessary in order to detect "out of space"
646  *                     errors from the server, if the write delegation
647  *                     issued by the server doesn't allow the file to grow.
648  */
649 /* ARGSUSED */
650 static int
651 nfs_close(struct vop_close_args *ap)
652 {
653 	struct vnode *vp = ap->a_vp;
654 	struct nfsnode *np = VTONFS(vp);
655 	struct nfsvattr nfsva;
656 	struct ucred *cred;
657 	int error = 0, ret, localcred = 0;
658 	int fmode = ap->a_fflag;
659 
660 	if (NFSCL_FORCEDISM(vp->v_mount))
661 		return (0);
662 	/*
663 	 * During shutdown, a_cred isn't valid, so just use root.
664 	 */
665 	if (ap->a_cred == NOCRED) {
666 		cred = newnfs_getcred();
667 		localcred = 1;
668 	} else {
669 		cred = ap->a_cred;
670 	}
671 	if (vp->v_type == VREG) {
672 	    /*
673 	     * Examine and clean dirty pages, regardless of NMODIFIED.
674 	     * This closes a major hole in close-to-open consistency.
675 	     * We want to push out all dirty pages (and buffers) on
676 	     * close, regardless of whether they were dirtied by
677 	     * mmap'ed writes or via write().
678 	     */
679 	    if (nfs_clean_pages_on_close && vp->v_object) {
680 		VM_OBJECT_WLOCK(vp->v_object);
681 		vm_object_page_clean(vp->v_object, 0, 0, 0);
682 		VM_OBJECT_WUNLOCK(vp->v_object);
683 	    }
684 	    mtx_lock(&np->n_mtx);
685 	    if (np->n_flag & NMODIFIED) {
686 		mtx_unlock(&np->n_mtx);
687 		if (NFS_ISV3(vp)) {
688 		    /*
689 		     * Under NFSv3 we have dirty buffers to dispose of.  We
690 		     * must flush them to the NFS server.  We have the option
691 		     * of waiting all the way through the commit rpc or just
692 		     * waiting for the initial write.  The default is to only
693 		     * wait through the initial write so the data is in the
694 		     * server's cache, which is roughly similar to the state
695 		     * a standard disk subsystem leaves the file in on close().
696 		     *
697 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
698 		     * potential races with other processes, and certainly
699 		     * cannot clear it if we don't commit.
700 		     * These races occur when there is no longer the old
701 		     * traditional vnode locking implemented for Vnode Ops.
702 		     */
703 		    int cm = newnfs_commit_on_close ? 1 : 0;
704 		    error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
705 		    /* np->n_flag &= ~NMODIFIED; */
706 		} else if (NFS_ISV4(vp)) {
707 			if (nfscl_mustflush(vp) != 0) {
708 				int cm = newnfs_commit_on_close ? 1 : 0;
709 				error = ncl_flush(vp, MNT_WAIT, ap->a_td,
710 				    cm, 0);
711 				/*
712 				 * as above w.r.t races when clearing
713 				 * NMODIFIED.
714 				 * np->n_flag &= ~NMODIFIED;
715 				 */
716 			}
717 		} else {
718 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
719 		}
720 		mtx_lock(&np->n_mtx);
721 	    }
722  	    /*
723  	     * Invalidate the attribute cache in all cases.
724  	     * An open is going to fetch fresh attrs any way, other procs
725  	     * on this node that have file open will be forced to do an
726  	     * otw attr fetch, but this is safe.
727 	     * --> A user found that their RPC count dropped by 20% when
728 	     *     this was commented out and I can't see any requirement
729 	     *     for it, so I've disabled it when negative lookups are
730 	     *     enabled. (What does this have to do with negative lookup
731 	     *     caching? Well nothing, except it was reported by the
732 	     *     same user that needed negative lookup caching and I wanted
733 	     *     there to be a way to disable it to see if it
734 	     *     is the cause of some caching/coherency issue that might
735 	     *     crop up.)
736  	     */
737 	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
738 		    np->n_attrstamp = 0;
739 		    KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
740 	    }
741 	    if (np->n_flag & NWRITEERR) {
742 		np->n_flag &= ~NWRITEERR;
743 		error = np->n_error;
744 	    }
745 	    mtx_unlock(&np->n_mtx);
746 	}
747 
748 	if (NFS_ISV4(vp)) {
749 		/*
750 		 * Get attributes so "change" is up to date.
751 		 */
752 		if (error == 0 && nfscl_mustflush(vp) != 0 &&
753 		    vp->v_type == VREG &&
754 		    (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
755 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
756 			    NULL);
757 			if (!ret) {
758 				np->n_change = nfsva.na_filerev;
759 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
760 				    NULL, 0, 0);
761 			}
762 		}
763 
764 		/*
765 		 * and do the close.
766 		 */
767 		ret = nfsrpc_close(vp, 0, ap->a_td);
768 		if (!error && ret)
769 			error = ret;
770 		if (error)
771 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
772 			    (gid_t)0);
773 	}
774 	if (newnfs_directio_enable)
775 		KASSERT((np->n_directio_asyncwr == 0),
776 			("nfs_close: dirty unflushed (%d) directio buffers\n",
777 			 np->n_directio_asyncwr));
778 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
779 		mtx_lock(&np->n_mtx);
780 		KASSERT((np->n_directio_opens > 0),
781 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
782 		np->n_directio_opens--;
783 		if (np->n_directio_opens == 0)
784 			np->n_flag &= ~NNONCACHE;
785 		mtx_unlock(&np->n_mtx);
786 	}
787 	if (localcred)
788 		NFSFREECRED(cred);
789 	return (error);
790 }
791 
792 /*
793  * nfs getattr call from vfs.
794  */
795 static int
796 nfs_getattr(struct vop_getattr_args *ap)
797 {
798 	struct vnode *vp = ap->a_vp;
799 	struct thread *td = curthread;	/* XXX */
800 	struct nfsnode *np = VTONFS(vp);
801 	int error = 0;
802 	struct nfsvattr nfsva;
803 	struct vattr *vap = ap->a_vap;
804 	struct vattr vattr;
805 
806 	/*
807 	 * Update local times for special files.
808 	 */
809 	mtx_lock(&np->n_mtx);
810 	if (np->n_flag & (NACC | NUPD))
811 		np->n_flag |= NCHG;
812 	mtx_unlock(&np->n_mtx);
813 	/*
814 	 * First look in the cache.
815 	 */
816 	if (ncl_getattrcache(vp, &vattr) == 0) {
817 		vap->va_type = vattr.va_type;
818 		vap->va_mode = vattr.va_mode;
819 		vap->va_nlink = vattr.va_nlink;
820 		vap->va_uid = vattr.va_uid;
821 		vap->va_gid = vattr.va_gid;
822 		vap->va_fsid = vattr.va_fsid;
823 		vap->va_fileid = vattr.va_fileid;
824 		vap->va_size = vattr.va_size;
825 		vap->va_blocksize = vattr.va_blocksize;
826 		vap->va_atime = vattr.va_atime;
827 		vap->va_mtime = vattr.va_mtime;
828 		vap->va_ctime = vattr.va_ctime;
829 		vap->va_gen = vattr.va_gen;
830 		vap->va_flags = vattr.va_flags;
831 		vap->va_rdev = vattr.va_rdev;
832 		vap->va_bytes = vattr.va_bytes;
833 		vap->va_filerev = vattr.va_filerev;
834 		/*
835 		 * Get the local modify time for the case of a write
836 		 * delegation.
837 		 */
838 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
839 		return (0);
840 	}
841 
842 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
843 	    nfsaccess_cache_timeout > 0) {
844 		NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
845 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
846 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
847 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
848 			return (0);
849 		}
850 	}
851 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
852 	if (!error)
853 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
854 	if (!error) {
855 		/*
856 		 * Get the local modify time for the case of a write
857 		 * delegation.
858 		 */
859 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
860 	} else if (NFS_ISV4(vp)) {
861 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
862 	}
863 	return (error);
864 }
865 
866 /*
867  * nfs setattr call.
868  */
869 static int
870 nfs_setattr(struct vop_setattr_args *ap)
871 {
872 	struct vnode *vp = ap->a_vp;
873 	struct nfsnode *np = VTONFS(vp);
874 	struct thread *td = curthread;	/* XXX */
875 	struct vattr *vap = ap->a_vap;
876 	int error = 0;
877 	u_quad_t tsize;
878 
879 #ifndef nolint
880 	tsize = (u_quad_t)0;
881 #endif
882 
883 	/*
884 	 * Setting of flags and marking of atimes are not supported.
885 	 */
886 	if (vap->va_flags != VNOVAL)
887 		return (EOPNOTSUPP);
888 
889 	/*
890 	 * Disallow write attempts if the filesystem is mounted read-only.
891 	 */
892   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
893 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
894 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
895 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
896 		return (EROFS);
897 	if (vap->va_size != VNOVAL) {
898  		switch (vp->v_type) {
899  		case VDIR:
900  			return (EISDIR);
901  		case VCHR:
902  		case VBLK:
903  		case VSOCK:
904  		case VFIFO:
905 			if (vap->va_mtime.tv_sec == VNOVAL &&
906 			    vap->va_atime.tv_sec == VNOVAL &&
907 			    vap->va_mode == (mode_t)VNOVAL &&
908 			    vap->va_uid == (uid_t)VNOVAL &&
909 			    vap->va_gid == (gid_t)VNOVAL)
910 				return (0);
911  			vap->va_size = VNOVAL;
912  			break;
913  		default:
914 			/*
915 			 * Disallow write attempts if the filesystem is
916 			 * mounted read-only.
917 			 */
918 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
919 				return (EROFS);
920 			/*
921 			 *  We run vnode_pager_setsize() early (why?),
922 			 * we must set np->n_size now to avoid vinvalbuf
923 			 * V_SAVE races that might setsize a lower
924 			 * value.
925 			 */
926 			mtx_lock(&np->n_mtx);
927 			tsize = np->n_size;
928 			mtx_unlock(&np->n_mtx);
929 			error = ncl_meta_setsize(vp, ap->a_cred, td,
930 			    vap->va_size);
931 			mtx_lock(&np->n_mtx);
932  			if (np->n_flag & NMODIFIED) {
933 			    tsize = np->n_size;
934 			    mtx_unlock(&np->n_mtx);
935 			    error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
936 			        0 : V_SAVE, td, 1);
937 			    if (error != 0) {
938 				    vnode_pager_setsize(vp, tsize);
939 				    return (error);
940 			    }
941 			    /*
942 			     * Call nfscl_delegmodtime() to set the modify time
943 			     * locally, as required.
944 			     */
945 			    nfscl_delegmodtime(vp);
946  			} else
947 			    mtx_unlock(&np->n_mtx);
948 			/*
949 			 * np->n_size has already been set to vap->va_size
950 			 * in ncl_meta_setsize(). We must set it again since
951 			 * nfs_loadattrcache() could be called through
952 			 * ncl_meta_setsize() and could modify np->n_size.
953 			 */
954 			mtx_lock(&np->n_mtx);
955  			np->n_vattr.na_size = np->n_size = vap->va_size;
956 			mtx_unlock(&np->n_mtx);
957   		}
958   	} else {
959 		mtx_lock(&np->n_mtx);
960 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
961 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
962 			mtx_unlock(&np->n_mtx);
963 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
964 			if (error == EINTR || error == EIO)
965 				return (error);
966 		} else
967 			mtx_unlock(&np->n_mtx);
968 	}
969 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
970 	if (error && vap->va_size != VNOVAL) {
971 		mtx_lock(&np->n_mtx);
972 		np->n_size = np->n_vattr.na_size = tsize;
973 		vnode_pager_setsize(vp, tsize);
974 		mtx_unlock(&np->n_mtx);
975 	}
976 	return (error);
977 }
978 
979 /*
980  * Do an nfs setattr rpc.
981  */
982 static int
983 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
984     struct thread *td)
985 {
986 	struct nfsnode *np = VTONFS(vp);
987 	int error, ret, attrflag, i;
988 	struct nfsvattr nfsva;
989 
990 	if (NFS_ISV34(vp)) {
991 		mtx_lock(&np->n_mtx);
992 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
993 			np->n_accesscache[i].stamp = 0;
994 		np->n_flag |= NDELEGMOD;
995 		mtx_unlock(&np->n_mtx);
996 		KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
997 	}
998 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
999 	    NULL);
1000 	if (attrflag) {
1001 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1002 		if (ret && !error)
1003 			error = ret;
1004 	}
1005 	if (error && NFS_ISV4(vp))
1006 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1007 	return (error);
1008 }
1009 
1010 /*
1011  * nfs lookup call, one step at a time...
1012  * First look in cache
1013  * If not found, unlock the directory nfsnode and do the rpc
1014  */
1015 static int
1016 nfs_lookup(struct vop_lookup_args *ap)
1017 {
1018 	struct componentname *cnp = ap->a_cnp;
1019 	struct vnode *dvp = ap->a_dvp;
1020 	struct vnode **vpp = ap->a_vpp;
1021 	struct mount *mp = dvp->v_mount;
1022 	int flags = cnp->cn_flags;
1023 	struct vnode *newvp;
1024 	struct nfsmount *nmp;
1025 	struct nfsnode *np, *newnp;
1026 	int error = 0, attrflag, dattrflag, ltype, ncticks;
1027 	struct thread *td = cnp->cn_thread;
1028 	struct nfsfh *nfhp;
1029 	struct nfsvattr dnfsva, nfsva;
1030 	struct vattr vattr;
1031 	struct timespec nctime;
1032 
1033 	*vpp = NULLVP;
1034 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1035 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1036 		return (EROFS);
1037 	if (dvp->v_type != VDIR)
1038 		return (ENOTDIR);
1039 	nmp = VFSTONFS(mp);
1040 	np = VTONFS(dvp);
1041 
1042 	/* For NFSv4, wait until any remove is done. */
1043 	mtx_lock(&np->n_mtx);
1044 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1045 		np->n_flag |= NREMOVEWANT;
1046 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1047 	}
1048 	mtx_unlock(&np->n_mtx);
1049 
1050 	if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1051 		return (error);
1052 	error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1053 	if (error > 0 && error != ENOENT)
1054 		return (error);
1055 	if (error == -1) {
1056 		/*
1057 		 * Lookups of "." are special and always return the
1058 		 * current directory.  cache_lookup() already handles
1059 		 * associated locking bookkeeping, etc.
1060 		 */
1061 		if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1062 			/* XXX: Is this really correct? */
1063 			if (cnp->cn_nameiop != LOOKUP &&
1064 			    (flags & ISLASTCN))
1065 				cnp->cn_flags |= SAVENAME;
1066 			return (0);
1067 		}
1068 
1069 		/*
1070 		 * We only accept a positive hit in the cache if the
1071 		 * change time of the file matches our cached copy.
1072 		 * Otherwise, we discard the cache entry and fallback
1073 		 * to doing a lookup RPC.  We also only trust cache
1074 		 * entries for less than nm_nametimeo seconds.
1075 		 *
1076 		 * To better handle stale file handles and attributes,
1077 		 * clear the attribute cache of this node if it is a
1078 		 * leaf component, part of an open() call, and not
1079 		 * locally modified before fetching the attributes.
1080 		 * This should allow stale file handles to be detected
1081 		 * here where we can fall back to a LOOKUP RPC to
1082 		 * recover rather than having nfs_open() detect the
1083 		 * stale file handle and failing open(2) with ESTALE.
1084 		 */
1085 		newvp = *vpp;
1086 		newnp = VTONFS(newvp);
1087 		if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1088 		    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1089 		    !(newnp->n_flag & NMODIFIED)) {
1090 			mtx_lock(&newnp->n_mtx);
1091 			newnp->n_attrstamp = 0;
1092 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1093 			mtx_unlock(&newnp->n_mtx);
1094 		}
1095 		if (nfscl_nodeleg(newvp, 0) == 0 ||
1096 		    ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1097 		    VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1098 		    timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1099 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1100 			if (cnp->cn_nameiop != LOOKUP &&
1101 			    (flags & ISLASTCN))
1102 				cnp->cn_flags |= SAVENAME;
1103 			return (0);
1104 		}
1105 		cache_purge(newvp);
1106 		if (dvp != newvp)
1107 			vput(newvp);
1108 		else
1109 			vrele(newvp);
1110 		*vpp = NULLVP;
1111 	} else if (error == ENOENT) {
1112 		if (dvp->v_iflag & VI_DOOMED)
1113 			return (ENOENT);
1114 		/*
1115 		 * We only accept a negative hit in the cache if the
1116 		 * modification time of the parent directory matches
1117 		 * the cached copy in the name cache entry.
1118 		 * Otherwise, we discard all of the negative cache
1119 		 * entries for this directory.  We also only trust
1120 		 * negative cache entries for up to nm_negnametimeo
1121 		 * seconds.
1122 		 */
1123 		if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1124 		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1125 		    timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1126 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1127 			return (ENOENT);
1128 		}
1129 		cache_purge_negative(dvp);
1130 	}
1131 
1132 	error = 0;
1133 	newvp = NULLVP;
1134 	NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1135 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1136 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1137 	    NULL);
1138 	if (dattrflag)
1139 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1140 	if (error) {
1141 		if (newvp != NULLVP) {
1142 			vput(newvp);
1143 			*vpp = NULLVP;
1144 		}
1145 
1146 		if (error != ENOENT) {
1147 			if (NFS_ISV4(dvp))
1148 				error = nfscl_maperr(td, error, (uid_t)0,
1149 				    (gid_t)0);
1150 			return (error);
1151 		}
1152 
1153 		/* The requested file was not found. */
1154 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1155 		    (flags & ISLASTCN)) {
1156 			/*
1157 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1158 			 * VWRITE) here instead of just checking
1159 			 * MNT_RDONLY.
1160 			 */
1161 			if (mp->mnt_flag & MNT_RDONLY)
1162 				return (EROFS);
1163 			cnp->cn_flags |= SAVENAME;
1164 			return (EJUSTRETURN);
1165 		}
1166 
1167 		if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1168 			/*
1169 			 * Cache the modification time of the parent
1170 			 * directory from the post-op attributes in
1171 			 * the name cache entry.  The negative cache
1172 			 * entry will be ignored once the directory
1173 			 * has changed.  Don't bother adding the entry
1174 			 * if the directory has already changed.
1175 			 */
1176 			mtx_lock(&np->n_mtx);
1177 			if (timespeccmp(&np->n_vattr.na_mtime,
1178 			    &dnfsva.na_mtime, ==)) {
1179 				mtx_unlock(&np->n_mtx);
1180 				cache_enter_time(dvp, NULL, cnp,
1181 				    &dnfsva.na_mtime, NULL);
1182 			} else
1183 				mtx_unlock(&np->n_mtx);
1184 		}
1185 		return (ENOENT);
1186 	}
1187 
1188 	/*
1189 	 * Handle RENAME case...
1190 	 */
1191 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1192 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1193 			FREE((caddr_t)nfhp, M_NFSFH);
1194 			return (EISDIR);
1195 		}
1196 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1197 		    LK_EXCLUSIVE);
1198 		if (error)
1199 			return (error);
1200 		newvp = NFSTOV(np);
1201 		if (attrflag)
1202 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1203 			    0, 1);
1204 		*vpp = newvp;
1205 		cnp->cn_flags |= SAVENAME;
1206 		return (0);
1207 	}
1208 
1209 	if (flags & ISDOTDOT) {
1210 		ltype = NFSVOPISLOCKED(dvp);
1211 		error = vfs_busy(mp, MBF_NOWAIT);
1212 		if (error != 0) {
1213 			vfs_ref(mp);
1214 			NFSVOPUNLOCK(dvp, 0);
1215 			error = vfs_busy(mp, 0);
1216 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1217 			vfs_rel(mp);
1218 			if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1219 				vfs_unbusy(mp);
1220 				error = ENOENT;
1221 			}
1222 			if (error != 0)
1223 				return (error);
1224 		}
1225 		NFSVOPUNLOCK(dvp, 0);
1226 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1227 		    cnp->cn_lkflags);
1228 		if (error == 0)
1229 			newvp = NFSTOV(np);
1230 		vfs_unbusy(mp);
1231 		if (newvp != dvp)
1232 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1233 		if (dvp->v_iflag & VI_DOOMED) {
1234 			if (error == 0) {
1235 				if (newvp == dvp)
1236 					vrele(newvp);
1237 				else
1238 					vput(newvp);
1239 			}
1240 			error = ENOENT;
1241 		}
1242 		if (error != 0)
1243 			return (error);
1244 		if (attrflag)
1245 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1246 			    0, 1);
1247 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1248 		FREE((caddr_t)nfhp, M_NFSFH);
1249 		VREF(dvp);
1250 		newvp = dvp;
1251 		if (attrflag)
1252 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1253 			    0, 1);
1254 	} else {
1255 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1256 		    cnp->cn_lkflags);
1257 		if (error)
1258 			return (error);
1259 		newvp = NFSTOV(np);
1260 		if (attrflag)
1261 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1262 			    0, 1);
1263 		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1264 		    !(np->n_flag & NMODIFIED)) {
1265 			/*
1266 			 * Flush the attribute cache when opening a
1267 			 * leaf node to ensure that fresh attributes
1268 			 * are fetched in nfs_open() since we did not
1269 			 * fetch attributes from the LOOKUP reply.
1270 			 */
1271 			mtx_lock(&np->n_mtx);
1272 			np->n_attrstamp = 0;
1273 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1274 			mtx_unlock(&np->n_mtx);
1275 		}
1276 	}
1277 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1278 		cnp->cn_flags |= SAVENAME;
1279 	if ((cnp->cn_flags & MAKEENTRY) &&
1280 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1281 	    attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1282 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1283 		    newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1284 	*vpp = newvp;
1285 	return (0);
1286 }
1287 
1288 /*
1289  * nfs read call.
1290  * Just call ncl_bioread() to do the work.
1291  */
1292 static int
1293 nfs_read(struct vop_read_args *ap)
1294 {
1295 	struct vnode *vp = ap->a_vp;
1296 
1297 	switch (vp->v_type) {
1298 	case VREG:
1299 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1300 	case VDIR:
1301 		return (EISDIR);
1302 	default:
1303 		return (EOPNOTSUPP);
1304 	}
1305 }
1306 
1307 /*
1308  * nfs readlink call
1309  */
1310 static int
1311 nfs_readlink(struct vop_readlink_args *ap)
1312 {
1313 	struct vnode *vp = ap->a_vp;
1314 
1315 	if (vp->v_type != VLNK)
1316 		return (EINVAL);
1317 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1318 }
1319 
1320 /*
1321  * Do a readlink rpc.
1322  * Called by ncl_doio() from below the buffer cache.
1323  */
1324 int
1325 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1326 {
1327 	int error, ret, attrflag;
1328 	struct nfsvattr nfsva;
1329 
1330 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1331 	    &attrflag, NULL);
1332 	if (attrflag) {
1333 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1334 		if (ret && !error)
1335 			error = ret;
1336 	}
1337 	if (error && NFS_ISV4(vp))
1338 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1339 	return (error);
1340 }
1341 
1342 /*
1343  * nfs read rpc call
1344  * Ditto above
1345  */
1346 int
1347 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1348 {
1349 	int error, ret, attrflag;
1350 	struct nfsvattr nfsva;
1351 	struct nfsmount *nmp;
1352 
1353 	nmp = VFSTONFS(vnode_mount(vp));
1354 	error = EIO;
1355 	attrflag = 0;
1356 	if (NFSHASPNFS(nmp))
1357 		error = nfscl_doiods(vp, uiop, NULL, NULL,
1358 		    NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1359 	NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1360 	if (error != 0)
1361 		error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1362 		    &attrflag, NULL);
1363 	if (attrflag) {
1364 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1365 		if (ret && !error)
1366 			error = ret;
1367 	}
1368 	if (error && NFS_ISV4(vp))
1369 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1370 	return (error);
1371 }
1372 
1373 /*
1374  * nfs write call
1375  */
1376 int
1377 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1378     int *iomode, int *must_commit, int called_from_strategy)
1379 {
1380 	struct nfsvattr nfsva;
1381 	int error, attrflag, ret;
1382 	struct nfsmount *nmp;
1383 
1384 	nmp = VFSTONFS(vnode_mount(vp));
1385 	error = EIO;
1386 	attrflag = 0;
1387 	if (NFSHASPNFS(nmp))
1388 		error = nfscl_doiods(vp, uiop, iomode, must_commit,
1389 		    NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1390 	NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1391 	if (error != 0)
1392 		error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1393 		    uiop->uio_td, &nfsva, &attrflag, NULL,
1394 		    called_from_strategy);
1395 	if (attrflag) {
1396 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1397 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1398 			    1);
1399 		else
1400 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1401 			    1);
1402 		if (ret && !error)
1403 			error = ret;
1404 	}
1405 	if (DOINGASYNC(vp))
1406 		*iomode = NFSWRITE_FILESYNC;
1407 	if (error && NFS_ISV4(vp))
1408 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1409 	return (error);
1410 }
1411 
1412 /*
1413  * nfs mknod rpc
1414  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1415  * mode set to specify the file type and the size field for rdev.
1416  */
1417 static int
1418 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1419     struct vattr *vap)
1420 {
1421 	struct nfsvattr nfsva, dnfsva;
1422 	struct vnode *newvp = NULL;
1423 	struct nfsnode *np = NULL, *dnp;
1424 	struct nfsfh *nfhp;
1425 	struct vattr vattr;
1426 	int error = 0, attrflag, dattrflag;
1427 	u_int32_t rdev;
1428 
1429 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1430 		rdev = vap->va_rdev;
1431 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1432 		rdev = 0xffffffff;
1433 	else
1434 		return (EOPNOTSUPP);
1435 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1436 		return (error);
1437 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1438 	    rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1439 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1440 	if (!error) {
1441 		if (!nfhp)
1442 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1443 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1444 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1445 			    NULL);
1446 		if (nfhp)
1447 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1448 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1449 	}
1450 	if (dattrflag)
1451 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1452 	if (!error) {
1453 		newvp = NFSTOV(np);
1454 		if (attrflag != 0) {
1455 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1456 			    0, 1);
1457 			if (error != 0)
1458 				vput(newvp);
1459 		}
1460 	}
1461 	if (!error) {
1462 		*vpp = newvp;
1463 	} else if (NFS_ISV4(dvp)) {
1464 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1465 		    vap->va_gid);
1466 	}
1467 	dnp = VTONFS(dvp);
1468 	mtx_lock(&dnp->n_mtx);
1469 	dnp->n_flag |= NMODIFIED;
1470 	if (!dattrflag) {
1471 		dnp->n_attrstamp = 0;
1472 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1473 	}
1474 	mtx_unlock(&dnp->n_mtx);
1475 	return (error);
1476 }
1477 
1478 /*
1479  * nfs mknod vop
1480  * just call nfs_mknodrpc() to do the work.
1481  */
1482 /* ARGSUSED */
1483 static int
1484 nfs_mknod(struct vop_mknod_args *ap)
1485 {
1486 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1487 }
1488 
1489 static struct mtx nfs_cverf_mtx;
1490 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1491     MTX_DEF);
1492 
1493 static nfsquad_t
1494 nfs_get_cverf(void)
1495 {
1496 	static nfsquad_t cverf;
1497 	nfsquad_t ret;
1498 	static int cverf_initialized = 0;
1499 
1500 	mtx_lock(&nfs_cverf_mtx);
1501 	if (cverf_initialized == 0) {
1502 		cverf.lval[0] = arc4random();
1503 		cverf.lval[1] = arc4random();
1504 		cverf_initialized = 1;
1505 	} else
1506 		cverf.qval++;
1507 	ret = cverf;
1508 	mtx_unlock(&nfs_cverf_mtx);
1509 
1510 	return (ret);
1511 }
1512 
1513 /*
1514  * nfs file create call
1515  */
1516 static int
1517 nfs_create(struct vop_create_args *ap)
1518 {
1519 	struct vnode *dvp = ap->a_dvp;
1520 	struct vattr *vap = ap->a_vap;
1521 	struct componentname *cnp = ap->a_cnp;
1522 	struct nfsnode *np = NULL, *dnp;
1523 	struct vnode *newvp = NULL;
1524 	struct nfsmount *nmp;
1525 	struct nfsvattr dnfsva, nfsva;
1526 	struct nfsfh *nfhp;
1527 	nfsquad_t cverf;
1528 	int error = 0, attrflag, dattrflag, fmode = 0;
1529 	struct vattr vattr;
1530 
1531 	/*
1532 	 * Oops, not for me..
1533 	 */
1534 	if (vap->va_type == VSOCK)
1535 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1536 
1537 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1538 		return (error);
1539 	if (vap->va_vaflags & VA_EXCLUSIVE)
1540 		fmode |= O_EXCL;
1541 	dnp = VTONFS(dvp);
1542 	nmp = VFSTONFS(vnode_mount(dvp));
1543 again:
1544 	/* For NFSv4, wait until any remove is done. */
1545 	mtx_lock(&dnp->n_mtx);
1546 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1547 		dnp->n_flag |= NREMOVEWANT;
1548 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1549 	}
1550 	mtx_unlock(&dnp->n_mtx);
1551 
1552 	cverf = nfs_get_cverf();
1553 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1554 	    vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1555 	    &nfhp, &attrflag, &dattrflag, NULL);
1556 	if (!error) {
1557 		if (nfhp == NULL)
1558 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1559 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1560 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1561 			    NULL);
1562 		if (nfhp != NULL)
1563 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1564 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1565 	}
1566 	if (dattrflag)
1567 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1568 	if (!error) {
1569 		newvp = NFSTOV(np);
1570 		if (attrflag == 0)
1571 			error = nfsrpc_getattr(newvp, cnp->cn_cred,
1572 			    cnp->cn_thread, &nfsva, NULL);
1573 		if (error == 0)
1574 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1575 			    0, 1);
1576 	}
1577 	if (error) {
1578 		if (newvp != NULL) {
1579 			vput(newvp);
1580 			newvp = NULL;
1581 		}
1582 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1583 		    error == NFSERR_NOTSUPP) {
1584 			fmode &= ~O_EXCL;
1585 			goto again;
1586 		}
1587 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1588 		if (nfscl_checksattr(vap, &nfsva)) {
1589 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1590 			    cnp->cn_thread, &nfsva, &attrflag, NULL);
1591 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1592 			    vap->va_gid != (gid_t)VNOVAL)) {
1593 				/* try again without setting uid/gid */
1594 				vap->va_uid = (uid_t)VNOVAL;
1595 				vap->va_gid = (uid_t)VNOVAL;
1596 				error = nfsrpc_setattr(newvp, vap, NULL,
1597 				    cnp->cn_cred, cnp->cn_thread, &nfsva,
1598 				    &attrflag, NULL);
1599 			}
1600 			if (attrflag)
1601 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1602 				    NULL, 0, 1);
1603 			if (error != 0)
1604 				vput(newvp);
1605 		}
1606 	}
1607 	if (!error) {
1608 		if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1609 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1610 			    NULL);
1611 		*ap->a_vpp = newvp;
1612 	} else if (NFS_ISV4(dvp)) {
1613 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1614 		    vap->va_gid);
1615 	}
1616 	mtx_lock(&dnp->n_mtx);
1617 	dnp->n_flag |= NMODIFIED;
1618 	if (!dattrflag) {
1619 		dnp->n_attrstamp = 0;
1620 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1621 	}
1622 	mtx_unlock(&dnp->n_mtx);
1623 	return (error);
1624 }
1625 
1626 /*
1627  * nfs file remove call
1628  * To try and make nfs semantics closer to ufs semantics, a file that has
1629  * other processes using the vnode is renamed instead of removed and then
1630  * removed later on the last close.
1631  * - If v_usecount > 1
1632  *	  If a rename is not already in the works
1633  *	     call nfs_sillyrename() to set it up
1634  *     else
1635  *	  do the remove rpc
1636  */
1637 static int
1638 nfs_remove(struct vop_remove_args *ap)
1639 {
1640 	struct vnode *vp = ap->a_vp;
1641 	struct vnode *dvp = ap->a_dvp;
1642 	struct componentname *cnp = ap->a_cnp;
1643 	struct nfsnode *np = VTONFS(vp);
1644 	int error = 0;
1645 	struct vattr vattr;
1646 
1647 	KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1648 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1649 	if (vp->v_type == VDIR)
1650 		error = EPERM;
1651 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1652 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1653 	    vattr.va_nlink > 1)) {
1654 		/*
1655 		 * Purge the name cache so that the chance of a lookup for
1656 		 * the name succeeding while the remove is in progress is
1657 		 * minimized. Without node locking it can still happen, such
1658 		 * that an I/O op returns ESTALE, but since you get this if
1659 		 * another host removes the file..
1660 		 */
1661 		cache_purge(vp);
1662 		/*
1663 		 * throw away biocache buffers, mainly to avoid
1664 		 * unnecessary delayed writes later.
1665 		 */
1666 		error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1667 		if (error != EINTR && error != EIO)
1668 			/* Do the rpc */
1669 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1670 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1671 		/*
1672 		 * Kludge City: If the first reply to the remove rpc is lost..
1673 		 *   the reply to the retransmitted request will be ENOENT
1674 		 *   since the file was in fact removed
1675 		 *   Therefore, we cheat and return success.
1676 		 */
1677 		if (error == ENOENT)
1678 			error = 0;
1679 	} else if (!np->n_sillyrename)
1680 		error = nfs_sillyrename(dvp, vp, cnp);
1681 	mtx_lock(&np->n_mtx);
1682 	np->n_attrstamp = 0;
1683 	mtx_unlock(&np->n_mtx);
1684 	KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1685 	return (error);
1686 }
1687 
1688 /*
1689  * nfs file remove rpc called from nfs_inactive
1690  */
1691 int
1692 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1693 {
1694 	/*
1695 	 * Make sure that the directory vnode is still valid.
1696 	 * XXX we should lock sp->s_dvp here.
1697 	 */
1698 	if (sp->s_dvp->v_type == VBAD)
1699 		return (0);
1700 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1701 	    sp->s_cred, NULL));
1702 }
1703 
1704 /*
1705  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1706  */
1707 static int
1708 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1709     int namelen, struct ucred *cred, struct thread *td)
1710 {
1711 	struct nfsvattr dnfsva;
1712 	struct nfsnode *dnp = VTONFS(dvp);
1713 	int error = 0, dattrflag;
1714 
1715 	mtx_lock(&dnp->n_mtx);
1716 	dnp->n_flag |= NREMOVEINPROG;
1717 	mtx_unlock(&dnp->n_mtx);
1718 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1719 	    &dattrflag, NULL);
1720 	mtx_lock(&dnp->n_mtx);
1721 	if ((dnp->n_flag & NREMOVEWANT)) {
1722 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1723 		mtx_unlock(&dnp->n_mtx);
1724 		wakeup((caddr_t)dnp);
1725 	} else {
1726 		dnp->n_flag &= ~NREMOVEINPROG;
1727 		mtx_unlock(&dnp->n_mtx);
1728 	}
1729 	if (dattrflag)
1730 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1731 	mtx_lock(&dnp->n_mtx);
1732 	dnp->n_flag |= NMODIFIED;
1733 	if (!dattrflag) {
1734 		dnp->n_attrstamp = 0;
1735 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1736 	}
1737 	mtx_unlock(&dnp->n_mtx);
1738 	if (error && NFS_ISV4(dvp))
1739 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1740 	return (error);
1741 }
1742 
1743 /*
1744  * nfs file rename call
1745  */
1746 static int
1747 nfs_rename(struct vop_rename_args *ap)
1748 {
1749 	struct vnode *fvp = ap->a_fvp;
1750 	struct vnode *tvp = ap->a_tvp;
1751 	struct vnode *fdvp = ap->a_fdvp;
1752 	struct vnode *tdvp = ap->a_tdvp;
1753 	struct componentname *tcnp = ap->a_tcnp;
1754 	struct componentname *fcnp = ap->a_fcnp;
1755 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1756 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1757 	struct nfsv4node *newv4 = NULL;
1758 	int error;
1759 
1760 	KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1761 	    (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1762 	/* Check for cross-device rename */
1763 	if ((fvp->v_mount != tdvp->v_mount) ||
1764 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1765 		error = EXDEV;
1766 		goto out;
1767 	}
1768 
1769 	if (fvp == tvp) {
1770 		printf("nfs_rename: fvp == tvp (can't happen)\n");
1771 		error = 0;
1772 		goto out;
1773 	}
1774 	if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1775 		goto out;
1776 
1777 	/*
1778 	 * We have to flush B_DELWRI data prior to renaming
1779 	 * the file.  If we don't, the delayed-write buffers
1780 	 * can be flushed out later after the file has gone stale
1781 	 * under NFSV3.  NFSV2 does not have this problem because
1782 	 * ( as far as I can tell ) it flushes dirty buffers more
1783 	 * often.
1784 	 *
1785 	 * Skip the rename operation if the fsync fails, this can happen
1786 	 * due to the server's volume being full, when we pushed out data
1787 	 * that was written back to our cache earlier. Not checking for
1788 	 * this condition can result in potential (silent) data loss.
1789 	 */
1790 	error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1791 	NFSVOPUNLOCK(fvp, 0);
1792 	if (!error && tvp)
1793 		error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1794 	if (error)
1795 		goto out;
1796 
1797 	/*
1798 	 * If the tvp exists and is in use, sillyrename it before doing the
1799 	 * rename of the new file over it.
1800 	 * XXX Can't sillyrename a directory.
1801 	 */
1802 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1803 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1804 		vput(tvp);
1805 		tvp = NULL;
1806 	}
1807 
1808 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1809 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1810 	    tcnp->cn_thread);
1811 
1812 	if (error == 0 && NFS_ISV4(tdvp)) {
1813 		/*
1814 		 * For NFSv4, check to see if it is the same name and
1815 		 * replace the name, if it is different.
1816 		 */
1817 		MALLOC(newv4, struct nfsv4node *,
1818 		    sizeof (struct nfsv4node) +
1819 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1820 		    M_NFSV4NODE, M_WAITOK);
1821 		mtx_lock(&tdnp->n_mtx);
1822 		mtx_lock(&fnp->n_mtx);
1823 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1824 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1825 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1826 		      tcnp->cn_namelen) ||
1827 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1828 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1829 			tdnp->n_fhp->nfh_len))) {
1830 #ifdef notdef
1831 { char nnn[100]; int nnnl;
1832 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1833 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1834 nnn[nnnl] = '\0';
1835 printf("ren replace=%s\n",nnn);
1836 }
1837 #endif
1838 			FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1839 			fnp->n_v4 = newv4;
1840 			newv4 = NULL;
1841 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1842 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1843 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1844 			    tdnp->n_fhp->nfh_len);
1845 			NFSBCOPY(tcnp->cn_nameptr,
1846 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1847 		}
1848 		mtx_unlock(&tdnp->n_mtx);
1849 		mtx_unlock(&fnp->n_mtx);
1850 		if (newv4 != NULL)
1851 			FREE((caddr_t)newv4, M_NFSV4NODE);
1852 	}
1853 
1854 	if (fvp->v_type == VDIR) {
1855 		if (tvp != NULL && tvp->v_type == VDIR)
1856 			cache_purge(tdvp);
1857 		cache_purge(fdvp);
1858 	}
1859 
1860 out:
1861 	if (tdvp == tvp)
1862 		vrele(tdvp);
1863 	else
1864 		vput(tdvp);
1865 	if (tvp)
1866 		vput(tvp);
1867 	vrele(fdvp);
1868 	vrele(fvp);
1869 	/*
1870 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1871 	 */
1872 	if (error == ENOENT)
1873 		error = 0;
1874 	return (error);
1875 }
1876 
1877 /*
1878  * nfs file rename rpc called from nfs_remove() above
1879  */
1880 static int
1881 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1882     struct sillyrename *sp)
1883 {
1884 
1885 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1886 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1887 	    scnp->cn_thread));
1888 }
1889 
1890 /*
1891  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1892  */
1893 static int
1894 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1895     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1896     int tnamelen, struct ucred *cred, struct thread *td)
1897 {
1898 	struct nfsvattr fnfsva, tnfsva;
1899 	struct nfsnode *fdnp = VTONFS(fdvp);
1900 	struct nfsnode *tdnp = VTONFS(tdvp);
1901 	int error = 0, fattrflag, tattrflag;
1902 
1903 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1904 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1905 	    &tattrflag, NULL, NULL);
1906 	mtx_lock(&fdnp->n_mtx);
1907 	fdnp->n_flag |= NMODIFIED;
1908 	if (fattrflag != 0) {
1909 		mtx_unlock(&fdnp->n_mtx);
1910 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1911 	} else {
1912 		fdnp->n_attrstamp = 0;
1913 		mtx_unlock(&fdnp->n_mtx);
1914 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1915 	}
1916 	mtx_lock(&tdnp->n_mtx);
1917 	tdnp->n_flag |= NMODIFIED;
1918 	if (tattrflag != 0) {
1919 		mtx_unlock(&tdnp->n_mtx);
1920 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1921 	} else {
1922 		tdnp->n_attrstamp = 0;
1923 		mtx_unlock(&tdnp->n_mtx);
1924 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1925 	}
1926 	if (error && NFS_ISV4(fdvp))
1927 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1928 	return (error);
1929 }
1930 
1931 /*
1932  * nfs hard link create call
1933  */
1934 static int
1935 nfs_link(struct vop_link_args *ap)
1936 {
1937 	struct vnode *vp = ap->a_vp;
1938 	struct vnode *tdvp = ap->a_tdvp;
1939 	struct componentname *cnp = ap->a_cnp;
1940 	struct nfsnode *np, *tdnp;
1941 	struct nfsvattr nfsva, dnfsva;
1942 	int error = 0, attrflag, dattrflag;
1943 
1944 	/*
1945 	 * Push all writes to the server, so that the attribute cache
1946 	 * doesn't get "out of sync" with the server.
1947 	 * XXX There should be a better way!
1948 	 */
1949 	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1950 
1951 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1952 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1953 	    &dattrflag, NULL);
1954 	tdnp = VTONFS(tdvp);
1955 	mtx_lock(&tdnp->n_mtx);
1956 	tdnp->n_flag |= NMODIFIED;
1957 	if (dattrflag != 0) {
1958 		mtx_unlock(&tdnp->n_mtx);
1959 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1960 	} else {
1961 		tdnp->n_attrstamp = 0;
1962 		mtx_unlock(&tdnp->n_mtx);
1963 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1964 	}
1965 	if (attrflag)
1966 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1967 	else {
1968 		np = VTONFS(vp);
1969 		mtx_lock(&np->n_mtx);
1970 		np->n_attrstamp = 0;
1971 		mtx_unlock(&np->n_mtx);
1972 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1973 	}
1974 	/*
1975 	 * If negative lookup caching is enabled, I might as well
1976 	 * add an entry for this node. Not necessary for correctness,
1977 	 * but if negative caching is enabled, then the system
1978 	 * must care about lookup caching hit rate, so...
1979 	 */
1980 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1981 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
1982 		cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
1983 	}
1984 	if (error && NFS_ISV4(vp))
1985 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1986 		    (gid_t)0);
1987 	return (error);
1988 }
1989 
1990 /*
1991  * nfs symbolic link create call
1992  */
1993 static int
1994 nfs_symlink(struct vop_symlink_args *ap)
1995 {
1996 	struct vnode *dvp = ap->a_dvp;
1997 	struct vattr *vap = ap->a_vap;
1998 	struct componentname *cnp = ap->a_cnp;
1999 	struct nfsvattr nfsva, dnfsva;
2000 	struct nfsfh *nfhp;
2001 	struct nfsnode *np = NULL, *dnp;
2002 	struct vnode *newvp = NULL;
2003 	int error = 0, attrflag, dattrflag, ret;
2004 
2005 	vap->va_type = VLNK;
2006 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2007 	    ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2008 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2009 	if (nfhp) {
2010 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2011 		    &np, NULL, LK_EXCLUSIVE);
2012 		if (!ret)
2013 			newvp = NFSTOV(np);
2014 		else if (!error)
2015 			error = ret;
2016 	}
2017 	if (newvp != NULL) {
2018 		if (attrflag)
2019 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2020 			    0, 1);
2021 	} else if (!error) {
2022 		/*
2023 		 * If we do not have an error and we could not extract the
2024 		 * newvp from the response due to the request being NFSv2, we
2025 		 * have to do a lookup in order to obtain a newvp to return.
2026 		 */
2027 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2028 		    cnp->cn_cred, cnp->cn_thread, &np);
2029 		if (!error)
2030 			newvp = NFSTOV(np);
2031 	}
2032 	if (error) {
2033 		if (newvp)
2034 			vput(newvp);
2035 		if (NFS_ISV4(dvp))
2036 			error = nfscl_maperr(cnp->cn_thread, error,
2037 			    vap->va_uid, vap->va_gid);
2038 	} else {
2039 		*ap->a_vpp = newvp;
2040 	}
2041 
2042 	dnp = VTONFS(dvp);
2043 	mtx_lock(&dnp->n_mtx);
2044 	dnp->n_flag |= NMODIFIED;
2045 	if (dattrflag != 0) {
2046 		mtx_unlock(&dnp->n_mtx);
2047 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2048 	} else {
2049 		dnp->n_attrstamp = 0;
2050 		mtx_unlock(&dnp->n_mtx);
2051 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2052 	}
2053 	/*
2054 	 * If negative lookup caching is enabled, I might as well
2055 	 * add an entry for this node. Not necessary for correctness,
2056 	 * but if negative caching is enabled, then the system
2057 	 * must care about lookup caching hit rate, so...
2058 	 */
2059 	if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2060 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2061 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2062 	}
2063 	return (error);
2064 }
2065 
2066 /*
2067  * nfs make dir call
2068  */
2069 static int
2070 nfs_mkdir(struct vop_mkdir_args *ap)
2071 {
2072 	struct vnode *dvp = ap->a_dvp;
2073 	struct vattr *vap = ap->a_vap;
2074 	struct componentname *cnp = ap->a_cnp;
2075 	struct nfsnode *np = NULL, *dnp;
2076 	struct vnode *newvp = NULL;
2077 	struct vattr vattr;
2078 	struct nfsfh *nfhp;
2079 	struct nfsvattr nfsva, dnfsva;
2080 	int error = 0, attrflag, dattrflag, ret;
2081 
2082 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2083 		return (error);
2084 	vap->va_type = VDIR;
2085 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2086 	    vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2087 	    &attrflag, &dattrflag, NULL);
2088 	dnp = VTONFS(dvp);
2089 	mtx_lock(&dnp->n_mtx);
2090 	dnp->n_flag |= NMODIFIED;
2091 	if (dattrflag != 0) {
2092 		mtx_unlock(&dnp->n_mtx);
2093 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2094 	} else {
2095 		dnp->n_attrstamp = 0;
2096 		mtx_unlock(&dnp->n_mtx);
2097 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2098 	}
2099 	if (nfhp) {
2100 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2101 		    &np, NULL, LK_EXCLUSIVE);
2102 		if (!ret) {
2103 			newvp = NFSTOV(np);
2104 			if (attrflag)
2105 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2106 				NULL, 0, 1);
2107 		} else if (!error)
2108 			error = ret;
2109 	}
2110 	if (!error && newvp == NULL) {
2111 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2112 		    cnp->cn_cred, cnp->cn_thread, &np);
2113 		if (!error) {
2114 			newvp = NFSTOV(np);
2115 			if (newvp->v_type != VDIR)
2116 				error = EEXIST;
2117 		}
2118 	}
2119 	if (error) {
2120 		if (newvp)
2121 			vput(newvp);
2122 		if (NFS_ISV4(dvp))
2123 			error = nfscl_maperr(cnp->cn_thread, error,
2124 			    vap->va_uid, vap->va_gid);
2125 	} else {
2126 		/*
2127 		 * If negative lookup caching is enabled, I might as well
2128 		 * add an entry for this node. Not necessary for correctness,
2129 		 * but if negative caching is enabled, then the system
2130 		 * must care about lookup caching hit rate, so...
2131 		 */
2132 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2133 		    (cnp->cn_flags & MAKEENTRY) &&
2134 		    attrflag != 0 && dattrflag != 0)
2135 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2136 			    &dnfsva.na_ctime);
2137 		*ap->a_vpp = newvp;
2138 	}
2139 	return (error);
2140 }
2141 
2142 /*
2143  * nfs remove directory call
2144  */
2145 static int
2146 nfs_rmdir(struct vop_rmdir_args *ap)
2147 {
2148 	struct vnode *vp = ap->a_vp;
2149 	struct vnode *dvp = ap->a_dvp;
2150 	struct componentname *cnp = ap->a_cnp;
2151 	struct nfsnode *dnp;
2152 	struct nfsvattr dnfsva;
2153 	int error, dattrflag;
2154 
2155 	if (dvp == vp)
2156 		return (EINVAL);
2157 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2158 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2159 	dnp = VTONFS(dvp);
2160 	mtx_lock(&dnp->n_mtx);
2161 	dnp->n_flag |= NMODIFIED;
2162 	if (dattrflag != 0) {
2163 		mtx_unlock(&dnp->n_mtx);
2164 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2165 	} else {
2166 		dnp->n_attrstamp = 0;
2167 		mtx_unlock(&dnp->n_mtx);
2168 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2169 	}
2170 
2171 	cache_purge(dvp);
2172 	cache_purge(vp);
2173 	if (error && NFS_ISV4(dvp))
2174 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2175 		    (gid_t)0);
2176 	/*
2177 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2178 	 */
2179 	if (error == ENOENT)
2180 		error = 0;
2181 	return (error);
2182 }
2183 
2184 /*
2185  * nfs readdir call
2186  */
2187 static int
2188 nfs_readdir(struct vop_readdir_args *ap)
2189 {
2190 	struct vnode *vp = ap->a_vp;
2191 	struct nfsnode *np = VTONFS(vp);
2192 	struct uio *uio = ap->a_uio;
2193 	ssize_t tresid, left;
2194 	int error = 0;
2195 	struct vattr vattr;
2196 
2197 	if (ap->a_eofflag != NULL)
2198 		*ap->a_eofflag = 0;
2199 	if (vp->v_type != VDIR)
2200 		return(EPERM);
2201 
2202 	/*
2203 	 * First, check for hit on the EOF offset cache
2204 	 */
2205 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2206 	    (np->n_flag & NMODIFIED) == 0) {
2207 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2208 			mtx_lock(&np->n_mtx);
2209 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2210 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2211 				mtx_unlock(&np->n_mtx);
2212 				NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2213 				if (ap->a_eofflag != NULL)
2214 					*ap->a_eofflag = 1;
2215 				return (0);
2216 			} else
2217 				mtx_unlock(&np->n_mtx);
2218 		}
2219 	}
2220 
2221 	/*
2222 	 * NFS always guarantees that directory entries don't straddle
2223 	 * DIRBLKSIZ boundaries.  As such, we need to limit the size
2224 	 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2225 	 * directory entry.
2226 	 */
2227 	left = uio->uio_resid % DIRBLKSIZ;
2228 	if (left == uio->uio_resid)
2229 		return (EINVAL);
2230 	uio->uio_resid -= left;
2231 
2232 	/*
2233 	 * Call ncl_bioread() to do the real work.
2234 	 */
2235 	tresid = uio->uio_resid;
2236 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2237 
2238 	if (!error && uio->uio_resid == tresid) {
2239 		NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2240 		if (ap->a_eofflag != NULL)
2241 			*ap->a_eofflag = 1;
2242 	}
2243 
2244 	/* Add the partial DIRBLKSIZ (left) back in. */
2245 	uio->uio_resid += left;
2246 	return (error);
2247 }
2248 
2249 /*
2250  * Readdir rpc call.
2251  * Called from below the buffer cache by ncl_doio().
2252  */
2253 int
2254 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2255     struct thread *td)
2256 {
2257 	struct nfsvattr nfsva;
2258 	nfsuint64 *cookiep, cookie;
2259 	struct nfsnode *dnp = VTONFS(vp);
2260 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2261 	int error = 0, eof, attrflag;
2262 
2263 	KASSERT(uiop->uio_iovcnt == 1 &&
2264 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2265 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2266 	    ("nfs readdirrpc bad uio"));
2267 
2268 	/*
2269 	 * If there is no cookie, assume directory was stale.
2270 	 */
2271 	ncl_dircookie_lock(dnp);
2272 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2273 	if (cookiep) {
2274 		cookie = *cookiep;
2275 		ncl_dircookie_unlock(dnp);
2276 	} else {
2277 		ncl_dircookie_unlock(dnp);
2278 		return (NFSERR_BAD_COOKIE);
2279 	}
2280 
2281 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2282 		(void)ncl_fsinfo(nmp, vp, cred, td);
2283 
2284 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2285 	    &attrflag, &eof, NULL);
2286 	if (attrflag)
2287 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2288 
2289 	if (!error) {
2290 		/*
2291 		 * We are now either at the end of the directory or have filled
2292 		 * the block.
2293 		 */
2294 		if (eof)
2295 			dnp->n_direofoffset = uiop->uio_offset;
2296 		else {
2297 			if (uiop->uio_resid > 0)
2298 				printf("EEK! readdirrpc resid > 0\n");
2299 			ncl_dircookie_lock(dnp);
2300 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2301 			*cookiep = cookie;
2302 			ncl_dircookie_unlock(dnp);
2303 		}
2304 	} else if (NFS_ISV4(vp)) {
2305 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2306 	}
2307 	return (error);
2308 }
2309 
2310 /*
2311  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2312  */
2313 int
2314 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2315     struct thread *td)
2316 {
2317 	struct nfsvattr nfsva;
2318 	nfsuint64 *cookiep, cookie;
2319 	struct nfsnode *dnp = VTONFS(vp);
2320 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2321 	int error = 0, attrflag, eof;
2322 
2323 	KASSERT(uiop->uio_iovcnt == 1 &&
2324 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2325 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2326 	    ("nfs readdirplusrpc bad uio"));
2327 
2328 	/*
2329 	 * If there is no cookie, assume directory was stale.
2330 	 */
2331 	ncl_dircookie_lock(dnp);
2332 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2333 	if (cookiep) {
2334 		cookie = *cookiep;
2335 		ncl_dircookie_unlock(dnp);
2336 	} else {
2337 		ncl_dircookie_unlock(dnp);
2338 		return (NFSERR_BAD_COOKIE);
2339 	}
2340 
2341 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2342 		(void)ncl_fsinfo(nmp, vp, cred, td);
2343 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2344 	    &attrflag, &eof, NULL);
2345 	if (attrflag)
2346 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2347 
2348 	if (!error) {
2349 		/*
2350 		 * We are now either at end of the directory or have filled the
2351 		 * the block.
2352 		 */
2353 		if (eof)
2354 			dnp->n_direofoffset = uiop->uio_offset;
2355 		else {
2356 			if (uiop->uio_resid > 0)
2357 				printf("EEK! readdirplusrpc resid > 0\n");
2358 			ncl_dircookie_lock(dnp);
2359 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2360 			*cookiep = cookie;
2361 			ncl_dircookie_unlock(dnp);
2362 		}
2363 	} else if (NFS_ISV4(vp)) {
2364 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2365 	}
2366 	return (error);
2367 }
2368 
2369 /*
2370  * Silly rename. To make the NFS filesystem that is stateless look a little
2371  * more like the "ufs" a remove of an active vnode is translated to a rename
2372  * to a funny looking filename that is removed by nfs_inactive on the
2373  * nfsnode. There is the potential for another process on a different client
2374  * to create the same funny name between the nfs_lookitup() fails and the
2375  * nfs_rename() completes, but...
2376  */
2377 static int
2378 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2379 {
2380 	struct sillyrename *sp;
2381 	struct nfsnode *np;
2382 	int error;
2383 	short pid;
2384 	unsigned int lticks;
2385 
2386 	cache_purge(dvp);
2387 	np = VTONFS(vp);
2388 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2389 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2390 	    M_NEWNFSREQ, M_WAITOK);
2391 	sp->s_cred = crhold(cnp->cn_cred);
2392 	sp->s_dvp = dvp;
2393 	VREF(dvp);
2394 
2395 	/*
2396 	 * Fudge together a funny name.
2397 	 * Changing the format of the funny name to accommodate more
2398 	 * sillynames per directory.
2399 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2400 	 * CPU ticks since boot.
2401 	 */
2402 	pid = cnp->cn_thread->td_proc->p_pid;
2403 	lticks = (unsigned int)ticks;
2404 	for ( ; ; ) {
2405 		sp->s_namlen = sprintf(sp->s_name,
2406 				       ".nfs.%08x.%04x4.4", lticks,
2407 				       pid);
2408 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2409 				 cnp->cn_thread, NULL))
2410 			break;
2411 		lticks++;
2412 	}
2413 	error = nfs_renameit(dvp, vp, cnp, sp);
2414 	if (error)
2415 		goto bad;
2416 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2417 		cnp->cn_thread, &np);
2418 	np->n_sillyrename = sp;
2419 	return (0);
2420 bad:
2421 	vrele(sp->s_dvp);
2422 	crfree(sp->s_cred);
2423 	free((caddr_t)sp, M_NEWNFSREQ);
2424 	return (error);
2425 }
2426 
2427 /*
2428  * Look up a file name and optionally either update the file handle or
2429  * allocate an nfsnode, depending on the value of npp.
2430  * npp == NULL	--> just do the lookup
2431  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2432  *			handled too
2433  * *npp != NULL --> update the file handle in the vnode
2434  */
2435 static int
2436 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2437     struct thread *td, struct nfsnode **npp)
2438 {
2439 	struct vnode *newvp = NULL, *vp;
2440 	struct nfsnode *np, *dnp = VTONFS(dvp);
2441 	struct nfsfh *nfhp, *onfhp;
2442 	struct nfsvattr nfsva, dnfsva;
2443 	struct componentname cn;
2444 	int error = 0, attrflag, dattrflag;
2445 	u_int hash;
2446 
2447 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2448 	    &nfhp, &attrflag, &dattrflag, NULL);
2449 	if (dattrflag)
2450 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2451 	if (npp && !error) {
2452 		if (*npp != NULL) {
2453 		    np = *npp;
2454 		    vp = NFSTOV(np);
2455 		    /*
2456 		     * For NFSv4, check to see if it is the same name and
2457 		     * replace the name, if it is different.
2458 		     */
2459 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2460 			(np->n_v4->n4_namelen != len ||
2461 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2462 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2463 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2464 			 dnp->n_fhp->nfh_len))) {
2465 #ifdef notdef
2466 { char nnn[100]; int nnnl;
2467 nnnl = (len < 100) ? len : 99;
2468 bcopy(name, nnn, nnnl);
2469 nnn[nnnl] = '\0';
2470 printf("replace=%s\n",nnn);
2471 }
2472 #endif
2473 			    FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2474 			    MALLOC(np->n_v4, struct nfsv4node *,
2475 				sizeof (struct nfsv4node) +
2476 				dnp->n_fhp->nfh_len + len - 1,
2477 				M_NFSV4NODE, M_WAITOK);
2478 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2479 			    np->n_v4->n4_namelen = len;
2480 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2481 				dnp->n_fhp->nfh_len);
2482 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2483 		    }
2484 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2485 			FNV1_32_INIT);
2486 		    onfhp = np->n_fhp;
2487 		    /*
2488 		     * Rehash node for new file handle.
2489 		     */
2490 		    vfs_hash_rehash(vp, hash);
2491 		    np->n_fhp = nfhp;
2492 		    if (onfhp != NULL)
2493 			FREE((caddr_t)onfhp, M_NFSFH);
2494 		    newvp = NFSTOV(np);
2495 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2496 		    FREE((caddr_t)nfhp, M_NFSFH);
2497 		    VREF(dvp);
2498 		    newvp = dvp;
2499 		} else {
2500 		    cn.cn_nameptr = name;
2501 		    cn.cn_namelen = len;
2502 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2503 			&np, NULL, LK_EXCLUSIVE);
2504 		    if (error)
2505 			return (error);
2506 		    newvp = NFSTOV(np);
2507 		}
2508 		if (!attrflag && *npp == NULL) {
2509 			if (newvp == dvp)
2510 				vrele(newvp);
2511 			else
2512 				vput(newvp);
2513 			return (ENOENT);
2514 		}
2515 		if (attrflag)
2516 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2517 			    0, 1);
2518 	}
2519 	if (npp && *npp == NULL) {
2520 		if (error) {
2521 			if (newvp) {
2522 				if (newvp == dvp)
2523 					vrele(newvp);
2524 				else
2525 					vput(newvp);
2526 			}
2527 		} else
2528 			*npp = np;
2529 	}
2530 	if (error && NFS_ISV4(dvp))
2531 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2532 	return (error);
2533 }
2534 
2535 /*
2536  * Nfs Version 3 and 4 commit rpc
2537  */
2538 int
2539 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2540    struct thread *td)
2541 {
2542 	struct nfsvattr nfsva;
2543 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2544 	struct nfsnode *np;
2545 	struct uio uio;
2546 	int error, attrflag;
2547 
2548 	np = VTONFS(vp);
2549 	error = EIO;
2550 	attrflag = 0;
2551 	if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2552 		uio.uio_offset = offset;
2553 		uio.uio_resid = cnt;
2554 		error = nfscl_doiods(vp, &uio, NULL, NULL,
2555 		    NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2556 		if (error != 0) {
2557 			mtx_lock(&np->n_mtx);
2558 			np->n_flag &= ~NDSCOMMIT;
2559 			mtx_unlock(&np->n_mtx);
2560 		}
2561 	}
2562 	if (error != 0) {
2563 		mtx_lock(&nmp->nm_mtx);
2564 		if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2565 			mtx_unlock(&nmp->nm_mtx);
2566 			return (0);
2567 		}
2568 		mtx_unlock(&nmp->nm_mtx);
2569 		error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2570 		    &attrflag, NULL);
2571 	}
2572 	if (attrflag != 0)
2573 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2574 		    0, 1);
2575 	if (error != 0 && NFS_ISV4(vp))
2576 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2577 	return (error);
2578 }
2579 
2580 /*
2581  * Strategy routine.
2582  * For async requests when nfsiod(s) are running, queue the request by
2583  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2584  * request.
2585  */
2586 static int
2587 nfs_strategy(struct vop_strategy_args *ap)
2588 {
2589 	struct buf *bp;
2590 	struct vnode *vp;
2591 	struct ucred *cr;
2592 
2593 	bp = ap->a_bp;
2594 	vp = ap->a_vp;
2595 	KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2596 	KASSERT(!(bp->b_flags & B_DONE),
2597 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2598 	BUF_ASSERT_HELD(bp);
2599 
2600 	if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2601 		bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2602 		    DEV_BSIZE);
2603 	if (bp->b_iocmd == BIO_READ)
2604 		cr = bp->b_rcred;
2605 	else
2606 		cr = bp->b_wcred;
2607 
2608 	/*
2609 	 * If the op is asynchronous and an i/o daemon is waiting
2610 	 * queue the request, wake it up and wait for completion
2611 	 * otherwise just do it ourselves.
2612 	 */
2613 	if ((bp->b_flags & B_ASYNC) == 0 ||
2614 	    ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2615 		(void) ncl_doio(vp, bp, cr, curthread, 1);
2616 	return (0);
2617 }
2618 
2619 /*
2620  * fsync vnode op. Just call ncl_flush() with commit == 1.
2621  */
2622 /* ARGSUSED */
2623 static int
2624 nfs_fsync(struct vop_fsync_args *ap)
2625 {
2626 
2627 	if (ap->a_vp->v_type != VREG) {
2628 		/*
2629 		 * For NFS, metadata is changed synchronously on the server,
2630 		 * so there is nothing to flush. Also, ncl_flush() clears
2631 		 * the NMODIFIED flag and that shouldn't be done here for
2632 		 * directories.
2633 		 */
2634 		return (0);
2635 	}
2636 	return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2637 }
2638 
2639 /*
2640  * Flush all the blocks associated with a vnode.
2641  * 	Walk through the buffer pool and push any dirty pages
2642  *	associated with the vnode.
2643  * If the called_from_renewthread argument is TRUE, it has been called
2644  * from the NFSv4 renew thread and, as such, cannot block indefinitely
2645  * waiting for a buffer write to complete.
2646  */
2647 int
2648 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2649     int commit, int called_from_renewthread)
2650 {
2651 	struct nfsnode *np = VTONFS(vp);
2652 	struct buf *bp;
2653 	int i;
2654 	struct buf *nbp;
2655 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2656 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2657 	int passone = 1, trycnt = 0;
2658 	u_quad_t off, endoff, toff;
2659 	struct ucred* wcred = NULL;
2660 	struct buf **bvec = NULL;
2661 	struct bufobj *bo;
2662 #ifndef NFS_COMMITBVECSIZ
2663 #define	NFS_COMMITBVECSIZ	20
2664 #endif
2665 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2666 	int bvecsize = 0, bveccount;
2667 
2668 	if (called_from_renewthread != 0)
2669 		slptimeo = hz;
2670 	if (nmp->nm_flag & NFSMNT_INT)
2671 		slpflag = PCATCH;
2672 	if (!commit)
2673 		passone = 0;
2674 	bo = &vp->v_bufobj;
2675 	/*
2676 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2677 	 * server, but has not been committed to stable storage on the server
2678 	 * yet. On the first pass, the byte range is worked out and the commit
2679 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2680 	 * job.
2681 	 */
2682 again:
2683 	off = (u_quad_t)-1;
2684 	endoff = 0;
2685 	bvecpos = 0;
2686 	if (NFS_ISV34(vp) && commit) {
2687 		if (bvec != NULL && bvec != bvec_on_stack)
2688 			free(bvec, M_TEMP);
2689 		/*
2690 		 * Count up how many buffers waiting for a commit.
2691 		 */
2692 		bveccount = 0;
2693 		BO_LOCK(bo);
2694 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2695 			if (!BUF_ISLOCKED(bp) &&
2696 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2697 				== (B_DELWRI | B_NEEDCOMMIT))
2698 				bveccount++;
2699 		}
2700 		/*
2701 		 * Allocate space to remember the list of bufs to commit.  It is
2702 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2703 		 * If we can't get memory (for whatever reason), we will end up
2704 		 * committing the buffers one-by-one in the loop below.
2705 		 */
2706 		if (bveccount > NFS_COMMITBVECSIZ) {
2707 			/*
2708 			 * Release the vnode interlock to avoid a lock
2709 			 * order reversal.
2710 			 */
2711 			BO_UNLOCK(bo);
2712 			bvec = (struct buf **)
2713 				malloc(bveccount * sizeof(struct buf *),
2714 				       M_TEMP, M_NOWAIT);
2715 			BO_LOCK(bo);
2716 			if (bvec == NULL) {
2717 				bvec = bvec_on_stack;
2718 				bvecsize = NFS_COMMITBVECSIZ;
2719 			} else
2720 				bvecsize = bveccount;
2721 		} else {
2722 			bvec = bvec_on_stack;
2723 			bvecsize = NFS_COMMITBVECSIZ;
2724 		}
2725 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2726 			if (bvecpos >= bvecsize)
2727 				break;
2728 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2729 				nbp = TAILQ_NEXT(bp, b_bobufs);
2730 				continue;
2731 			}
2732 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2733 			    (B_DELWRI | B_NEEDCOMMIT)) {
2734 				BUF_UNLOCK(bp);
2735 				nbp = TAILQ_NEXT(bp, b_bobufs);
2736 				continue;
2737 			}
2738 			BO_UNLOCK(bo);
2739 			bremfree(bp);
2740 			/*
2741 			 * Work out if all buffers are using the same cred
2742 			 * so we can deal with them all with one commit.
2743 			 *
2744 			 * NOTE: we are not clearing B_DONE here, so we have
2745 			 * to do it later on in this routine if we intend to
2746 			 * initiate I/O on the bp.
2747 			 *
2748 			 * Note: to avoid loopback deadlocks, we do not
2749 			 * assign b_runningbufspace.
2750 			 */
2751 			if (wcred == NULL)
2752 				wcred = bp->b_wcred;
2753 			else if (wcred != bp->b_wcred)
2754 				wcred = NOCRED;
2755 			vfs_busy_pages(bp, 1);
2756 
2757 			BO_LOCK(bo);
2758 			/*
2759 			 * bp is protected by being locked, but nbp is not
2760 			 * and vfs_busy_pages() may sleep.  We have to
2761 			 * recalculate nbp.
2762 			 */
2763 			nbp = TAILQ_NEXT(bp, b_bobufs);
2764 
2765 			/*
2766 			 * A list of these buffers is kept so that the
2767 			 * second loop knows which buffers have actually
2768 			 * been committed. This is necessary, since there
2769 			 * may be a race between the commit rpc and new
2770 			 * uncommitted writes on the file.
2771 			 */
2772 			bvec[bvecpos++] = bp;
2773 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2774 				bp->b_dirtyoff;
2775 			if (toff < off)
2776 				off = toff;
2777 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2778 			if (toff > endoff)
2779 				endoff = toff;
2780 		}
2781 		BO_UNLOCK(bo);
2782 	}
2783 	if (bvecpos > 0) {
2784 		/*
2785 		 * Commit data on the server, as required.
2786 		 * If all bufs are using the same wcred, then use that with
2787 		 * one call for all of them, otherwise commit each one
2788 		 * separately.
2789 		 */
2790 		if (wcred != NOCRED)
2791 			retv = ncl_commit(vp, off, (int)(endoff - off),
2792 					  wcred, td);
2793 		else {
2794 			retv = 0;
2795 			for (i = 0; i < bvecpos; i++) {
2796 				off_t off, size;
2797 				bp = bvec[i];
2798 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2799 					bp->b_dirtyoff;
2800 				size = (u_quad_t)(bp->b_dirtyend
2801 						  - bp->b_dirtyoff);
2802 				retv = ncl_commit(vp, off, (int)size,
2803 						  bp->b_wcred, td);
2804 				if (retv) break;
2805 			}
2806 		}
2807 
2808 		if (retv == NFSERR_STALEWRITEVERF)
2809 			ncl_clearcommit(vp->v_mount);
2810 
2811 		/*
2812 		 * Now, either mark the blocks I/O done or mark the
2813 		 * blocks dirty, depending on whether the commit
2814 		 * succeeded.
2815 		 */
2816 		for (i = 0; i < bvecpos; i++) {
2817 			bp = bvec[i];
2818 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2819 			if (retv) {
2820 				/*
2821 				 * Error, leave B_DELWRI intact
2822 				 */
2823 				vfs_unbusy_pages(bp);
2824 				brelse(bp);
2825 			} else {
2826 				/*
2827 				 * Success, remove B_DELWRI ( bundirty() ).
2828 				 *
2829 				 * b_dirtyoff/b_dirtyend seem to be NFS
2830 				 * specific.  We should probably move that
2831 				 * into bundirty(). XXX
2832 				 */
2833 				bufobj_wref(bo);
2834 				bp->b_flags |= B_ASYNC;
2835 				bundirty(bp);
2836 				bp->b_flags &= ~B_DONE;
2837 				bp->b_ioflags &= ~BIO_ERROR;
2838 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2839 				bufdone(bp);
2840 			}
2841 		}
2842 	}
2843 
2844 	/*
2845 	 * Start/do any write(s) that are required.
2846 	 */
2847 loop:
2848 	BO_LOCK(bo);
2849 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2850 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2851 			if (waitfor != MNT_WAIT || passone)
2852 				continue;
2853 
2854 			error = BUF_TIMELOCK(bp,
2855 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2856 			    BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
2857 			if (error == 0) {
2858 				BUF_UNLOCK(bp);
2859 				goto loop;
2860 			}
2861 			if (error == ENOLCK) {
2862 				error = 0;
2863 				goto loop;
2864 			}
2865 			if (called_from_renewthread != 0) {
2866 				/*
2867 				 * Return EIO so the flush will be retried
2868 				 * later.
2869 				 */
2870 				error = EIO;
2871 				goto done;
2872 			}
2873 			if (newnfs_sigintr(nmp, td)) {
2874 				error = EINTR;
2875 				goto done;
2876 			}
2877 			if (slpflag == PCATCH) {
2878 				slpflag = 0;
2879 				slptimeo = 2 * hz;
2880 			}
2881 			goto loop;
2882 		}
2883 		if ((bp->b_flags & B_DELWRI) == 0)
2884 			panic("nfs_fsync: not dirty");
2885 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2886 			BUF_UNLOCK(bp);
2887 			continue;
2888 		}
2889 		BO_UNLOCK(bo);
2890 		bremfree(bp);
2891 		if (passone || !commit)
2892 		    bp->b_flags |= B_ASYNC;
2893 		else
2894 		    bp->b_flags |= B_ASYNC;
2895 		bwrite(bp);
2896 		if (newnfs_sigintr(nmp, td)) {
2897 			error = EINTR;
2898 			goto done;
2899 		}
2900 		goto loop;
2901 	}
2902 	if (passone) {
2903 		passone = 0;
2904 		BO_UNLOCK(bo);
2905 		goto again;
2906 	}
2907 	if (waitfor == MNT_WAIT) {
2908 		while (bo->bo_numoutput) {
2909 			error = bufobj_wwait(bo, slpflag, slptimeo);
2910 			if (error) {
2911 			    BO_UNLOCK(bo);
2912 			    if (called_from_renewthread != 0) {
2913 				/*
2914 				 * Return EIO so that the flush will be
2915 				 * retried later.
2916 				 */
2917 				error = EIO;
2918 				goto done;
2919 			    }
2920 			    error = newnfs_sigintr(nmp, td);
2921 			    if (error)
2922 				goto done;
2923 			    if (slpflag == PCATCH) {
2924 				slpflag = 0;
2925 				slptimeo = 2 * hz;
2926 			    }
2927 			    BO_LOCK(bo);
2928 			}
2929 		}
2930 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
2931 			BO_UNLOCK(bo);
2932 			goto loop;
2933 		}
2934 		/*
2935 		 * Wait for all the async IO requests to drain
2936 		 */
2937 		BO_UNLOCK(bo);
2938 		mtx_lock(&np->n_mtx);
2939 		while (np->n_directio_asyncwr > 0) {
2940 			np->n_flag |= NFSYNCWAIT;
2941 			error = newnfs_msleep(td, &np->n_directio_asyncwr,
2942 			    &np->n_mtx, slpflag | (PRIBIO + 1),
2943 			    "nfsfsync", 0);
2944 			if (error) {
2945 				if (newnfs_sigintr(nmp, td)) {
2946 					mtx_unlock(&np->n_mtx);
2947 					error = EINTR;
2948 					goto done;
2949 				}
2950 			}
2951 		}
2952 		mtx_unlock(&np->n_mtx);
2953 	} else
2954 		BO_UNLOCK(bo);
2955 	if (NFSHASPNFS(nmp)) {
2956 		nfscl_layoutcommit(vp, td);
2957 		/*
2958 		 * Invalidate the attribute cache, since writes to a DS
2959 		 * won't update the size attribute.
2960 		 */
2961 		mtx_lock(&np->n_mtx);
2962 		np->n_attrstamp = 0;
2963 	} else
2964 		mtx_lock(&np->n_mtx);
2965 	if (np->n_flag & NWRITEERR) {
2966 		error = np->n_error;
2967 		np->n_flag &= ~NWRITEERR;
2968 	}
2969   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
2970 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2971   		np->n_flag &= ~NMODIFIED;
2972 	mtx_unlock(&np->n_mtx);
2973 done:
2974 	if (bvec != NULL && bvec != bvec_on_stack)
2975 		free(bvec, M_TEMP);
2976 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2977 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2978 	    np->n_directio_asyncwr != 0)) {
2979 		if (trycnt++ < 5) {
2980 			/* try, try again... */
2981 			passone = 1;
2982 			wcred = NULL;
2983 			bvec = NULL;
2984 			bvecsize = 0;
2985 			goto again;
2986 		}
2987 		vn_printf(vp, "ncl_flush failed");
2988 		error = called_from_renewthread != 0 ? EIO : EBUSY;
2989 	}
2990 	return (error);
2991 }
2992 
2993 /*
2994  * NFS advisory byte-level locks.
2995  */
2996 static int
2997 nfs_advlock(struct vop_advlock_args *ap)
2998 {
2999 	struct vnode *vp = ap->a_vp;
3000 	struct ucred *cred;
3001 	struct nfsnode *np = VTONFS(ap->a_vp);
3002 	struct proc *p = (struct proc *)ap->a_id;
3003 	struct thread *td = curthread;	/* XXX */
3004 	struct vattr va;
3005 	int ret, error = EOPNOTSUPP;
3006 	u_quad_t size;
3007 
3008 	if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3009 		if (vp->v_type != VREG)
3010 			return (EINVAL);
3011 		if ((ap->a_flags & F_POSIX) != 0)
3012 			cred = p->p_ucred;
3013 		else
3014 			cred = td->td_ucred;
3015 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3016 		if (vp->v_iflag & VI_DOOMED) {
3017 			NFSVOPUNLOCK(vp, 0);
3018 			return (EBADF);
3019 		}
3020 
3021 		/*
3022 		 * If this is unlocking a write locked region, flush and
3023 		 * commit them before unlocking. This is required by
3024 		 * RFC3530 Sec. 9.3.2.
3025 		 */
3026 		if (ap->a_op == F_UNLCK &&
3027 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3028 		    ap->a_flags))
3029 			(void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3030 
3031 		/*
3032 		 * Loop around doing the lock op, while a blocking lock
3033 		 * must wait for the lock op to succeed.
3034 		 */
3035 		do {
3036 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3037 			    ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3038 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3039 			    ap->a_op == F_SETLK) {
3040 				NFSVOPUNLOCK(vp, 0);
3041 				error = nfs_catnap(PZERO | PCATCH, ret,
3042 				    "ncladvl");
3043 				if (error)
3044 					return (EINTR);
3045 				NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3046 				if (vp->v_iflag & VI_DOOMED) {
3047 					NFSVOPUNLOCK(vp, 0);
3048 					return (EBADF);
3049 				}
3050 			}
3051 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3052 		     ap->a_op == F_SETLK);
3053 		if (ret == NFSERR_DENIED) {
3054 			NFSVOPUNLOCK(vp, 0);
3055 			return (EAGAIN);
3056 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3057 			NFSVOPUNLOCK(vp, 0);
3058 			return (ret);
3059 		} else if (ret != 0) {
3060 			NFSVOPUNLOCK(vp, 0);
3061 			return (EACCES);
3062 		}
3063 
3064 		/*
3065 		 * Now, if we just got a lock, invalidate data in the buffer
3066 		 * cache, as required, so that the coherency conforms with
3067 		 * RFC3530 Sec. 9.3.2.
3068 		 */
3069 		if (ap->a_op == F_SETLK) {
3070 			if ((np->n_flag & NMODIFIED) == 0) {
3071 				np->n_attrstamp = 0;
3072 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3073 				ret = VOP_GETATTR(vp, &va, cred);
3074 			}
3075 			if ((np->n_flag & NMODIFIED) || ret ||
3076 			    np->n_change != va.va_filerev) {
3077 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3078 				np->n_attrstamp = 0;
3079 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3080 				ret = VOP_GETATTR(vp, &va, cred);
3081 				if (!ret) {
3082 					np->n_mtime = va.va_mtime;
3083 					np->n_change = va.va_filerev;
3084 				}
3085 			}
3086 			/* Mark that a file lock has been acquired. */
3087 			mtx_lock(&np->n_mtx);
3088 			np->n_flag |= NHASBEENLOCKED;
3089 			mtx_unlock(&np->n_mtx);
3090 		}
3091 		NFSVOPUNLOCK(vp, 0);
3092 		return (0);
3093 	} else if (!NFS_ISV4(vp)) {
3094 		error = NFSVOPLOCK(vp, LK_SHARED);
3095 		if (error)
3096 			return (error);
3097 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3098 			size = VTONFS(vp)->n_size;
3099 			NFSVOPUNLOCK(vp, 0);
3100 			error = lf_advlock(ap, &(vp->v_lockf), size);
3101 		} else {
3102 			if (nfs_advlock_p != NULL)
3103 				error = nfs_advlock_p(ap);
3104 			else {
3105 				NFSVOPUNLOCK(vp, 0);
3106 				error = ENOLCK;
3107 			}
3108 		}
3109 		if (error == 0 && ap->a_op == F_SETLK) {
3110 			error = NFSVOPLOCK(vp, LK_SHARED);
3111 			if (error == 0) {
3112 				/* Mark that a file lock has been acquired. */
3113 				mtx_lock(&np->n_mtx);
3114 				np->n_flag |= NHASBEENLOCKED;
3115 				mtx_unlock(&np->n_mtx);
3116 				NFSVOPUNLOCK(vp, 0);
3117 			}
3118 		}
3119 	}
3120 	return (error);
3121 }
3122 
3123 /*
3124  * NFS advisory byte-level locks.
3125  */
3126 static int
3127 nfs_advlockasync(struct vop_advlockasync_args *ap)
3128 {
3129 	struct vnode *vp = ap->a_vp;
3130 	u_quad_t size;
3131 	int error;
3132 
3133 	if (NFS_ISV4(vp))
3134 		return (EOPNOTSUPP);
3135 	error = NFSVOPLOCK(vp, LK_SHARED);
3136 	if (error)
3137 		return (error);
3138 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3139 		size = VTONFS(vp)->n_size;
3140 		NFSVOPUNLOCK(vp, 0);
3141 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
3142 	} else {
3143 		NFSVOPUNLOCK(vp, 0);
3144 		error = EOPNOTSUPP;
3145 	}
3146 	return (error);
3147 }
3148 
3149 /*
3150  * Print out the contents of an nfsnode.
3151  */
3152 static int
3153 nfs_print(struct vop_print_args *ap)
3154 {
3155 	struct vnode *vp = ap->a_vp;
3156 	struct nfsnode *np = VTONFS(vp);
3157 
3158 	printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3159 	    (uintmax_t)np->n_vattr.na_fsid);
3160 	if (vp->v_type == VFIFO)
3161 		fifo_printinfo(vp);
3162 	printf("\n");
3163 	return (0);
3164 }
3165 
3166 /*
3167  * This is the "real" nfs::bwrite(struct buf*).
3168  * We set B_CACHE if this is a VMIO buffer.
3169  */
3170 int
3171 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3172 {
3173 	int oldflags, rtval;
3174 
3175 	BUF_ASSERT_HELD(bp);
3176 
3177 	if (bp->b_flags & B_INVAL) {
3178 		brelse(bp);
3179 		return (0);
3180 	}
3181 
3182 	oldflags = bp->b_flags;
3183 	bp->b_flags |= B_CACHE;
3184 
3185 	/*
3186 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3187 	 */
3188 	bundirty(bp);
3189 	bp->b_flags &= ~B_DONE;
3190 	bp->b_ioflags &= ~BIO_ERROR;
3191 	bp->b_iocmd = BIO_WRITE;
3192 
3193 	bufobj_wref(bp->b_bufobj);
3194 	curthread->td_ru.ru_oublock++;
3195 
3196 	/*
3197 	 * Note: to avoid loopback deadlocks, we do not
3198 	 * assign b_runningbufspace.
3199 	 */
3200 	vfs_busy_pages(bp, 1);
3201 
3202 	BUF_KERNPROC(bp);
3203 	bp->b_iooffset = dbtob(bp->b_blkno);
3204 	bstrategy(bp);
3205 
3206 	if ((oldflags & B_ASYNC) != 0)
3207 		return (0);
3208 
3209 	rtval = bufwait(bp);
3210 	if (oldflags & B_DELWRI)
3211 		reassignbuf(bp);
3212 	brelse(bp);
3213 	return (rtval);
3214 }
3215 
3216 /*
3217  * nfs special file access vnode op.
3218  * Essentially just get vattr and then imitate iaccess() since the device is
3219  * local to the client.
3220  */
3221 static int
3222 nfsspec_access(struct vop_access_args *ap)
3223 {
3224 	struct vattr *vap;
3225 	struct ucred *cred = ap->a_cred;
3226 	struct vnode *vp = ap->a_vp;
3227 	accmode_t accmode = ap->a_accmode;
3228 	struct vattr vattr;
3229 	int error;
3230 
3231 	/*
3232 	 * Disallow write attempts on filesystems mounted read-only;
3233 	 * unless the file is a socket, fifo, or a block or character
3234 	 * device resident on the filesystem.
3235 	 */
3236 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3237 		switch (vp->v_type) {
3238 		case VREG:
3239 		case VDIR:
3240 		case VLNK:
3241 			return (EROFS);
3242 		default:
3243 			break;
3244 		}
3245 	}
3246 	vap = &vattr;
3247 	error = VOP_GETATTR(vp, vap, cred);
3248 	if (error)
3249 		goto out;
3250 	error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3251 	    accmode, cred, NULL);
3252 out:
3253 	return error;
3254 }
3255 
3256 /*
3257  * Read wrapper for fifos.
3258  */
3259 static int
3260 nfsfifo_read(struct vop_read_args *ap)
3261 {
3262 	struct nfsnode *np = VTONFS(ap->a_vp);
3263 	int error;
3264 
3265 	/*
3266 	 * Set access flag.
3267 	 */
3268 	mtx_lock(&np->n_mtx);
3269 	np->n_flag |= NACC;
3270 	vfs_timestamp(&np->n_atim);
3271 	mtx_unlock(&np->n_mtx);
3272 	error = fifo_specops.vop_read(ap);
3273 	return error;
3274 }
3275 
3276 /*
3277  * Write wrapper for fifos.
3278  */
3279 static int
3280 nfsfifo_write(struct vop_write_args *ap)
3281 {
3282 	struct nfsnode *np = VTONFS(ap->a_vp);
3283 
3284 	/*
3285 	 * Set update flag.
3286 	 */
3287 	mtx_lock(&np->n_mtx);
3288 	np->n_flag |= NUPD;
3289 	vfs_timestamp(&np->n_mtim);
3290 	mtx_unlock(&np->n_mtx);
3291 	return(fifo_specops.vop_write(ap));
3292 }
3293 
3294 /*
3295  * Close wrapper for fifos.
3296  *
3297  * Update the times on the nfsnode then do fifo close.
3298  */
3299 static int
3300 nfsfifo_close(struct vop_close_args *ap)
3301 {
3302 	struct vnode *vp = ap->a_vp;
3303 	struct nfsnode *np = VTONFS(vp);
3304 	struct vattr vattr;
3305 	struct timespec ts;
3306 
3307 	mtx_lock(&np->n_mtx);
3308 	if (np->n_flag & (NACC | NUPD)) {
3309 		vfs_timestamp(&ts);
3310 		if (np->n_flag & NACC)
3311 			np->n_atim = ts;
3312 		if (np->n_flag & NUPD)
3313 			np->n_mtim = ts;
3314 		np->n_flag |= NCHG;
3315 		if (vrefcnt(vp) == 1 &&
3316 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3317 			VATTR_NULL(&vattr);
3318 			if (np->n_flag & NACC)
3319 				vattr.va_atime = np->n_atim;
3320 			if (np->n_flag & NUPD)
3321 				vattr.va_mtime = np->n_mtim;
3322 			mtx_unlock(&np->n_mtx);
3323 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3324 			goto out;
3325 		}
3326 	}
3327 	mtx_unlock(&np->n_mtx);
3328 out:
3329 	return (fifo_specops.vop_close(ap));
3330 }
3331 
3332 /*
3333  * Just call ncl_writebp() with the force argument set to 1.
3334  *
3335  * NOTE: B_DONE may or may not be set in a_bp on call.
3336  */
3337 static int
3338 nfs_bwrite(struct buf *bp)
3339 {
3340 
3341 	return (ncl_writebp(bp, 1, curthread));
3342 }
3343 
3344 struct buf_ops buf_ops_newnfs = {
3345 	.bop_name	=	"buf_ops_nfs",
3346 	.bop_write	=	nfs_bwrite,
3347 	.bop_strategy	=	bufstrategy,
3348 	.bop_sync	=	bufsync,
3349 	.bop_bdflush	=	bufbdflush,
3350 };
3351 
3352 static int
3353 nfs_getacl(struct vop_getacl_args *ap)
3354 {
3355 	int error;
3356 
3357 	if (ap->a_type != ACL_TYPE_NFS4)
3358 		return (EOPNOTSUPP);
3359 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3360 	    NULL);
3361 	if (error > NFSERR_STALE) {
3362 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3363 		error = EPERM;
3364 	}
3365 	return (error);
3366 }
3367 
3368 static int
3369 nfs_setacl(struct vop_setacl_args *ap)
3370 {
3371 	int error;
3372 
3373 	if (ap->a_type != ACL_TYPE_NFS4)
3374 		return (EOPNOTSUPP);
3375 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3376 	    NULL);
3377 	if (error > NFSERR_STALE) {
3378 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3379 		error = EPERM;
3380 	}
3381 	return (error);
3382 }
3383 
3384 static int
3385 nfs_set_text(struct vop_set_text_args *ap)
3386 {
3387 	struct vnode *vp = ap->a_vp;
3388 	struct nfsnode *np;
3389 
3390 	/*
3391 	 * If the text file has been mmap'd, flush any dirty pages to the
3392 	 * buffer cache and then...
3393 	 * Make sure all writes are pushed to the NFS server.  If this is not
3394 	 * done, the modify time of the file can change while the text
3395 	 * file is being executed.  This will cause the process that is
3396 	 * executing the text file to be terminated.
3397 	 */
3398 	if (vp->v_object != NULL) {
3399 		VM_OBJECT_WLOCK(vp->v_object);
3400 		vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3401 		VM_OBJECT_WUNLOCK(vp->v_object);
3402 	}
3403 
3404 	/* Now, flush the buffer cache. */
3405 	ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
3406 
3407 	/* And, finally, make sure that n_mtime is up to date. */
3408 	np = VTONFS(vp);
3409 	mtx_lock(&np->n_mtx);
3410 	np->n_mtime = np->n_vattr.na_mtime;
3411 	mtx_unlock(&np->n_mtx);
3412 
3413 	vp->v_vflag |= VV_TEXT;
3414 	return (0);
3415 }
3416 
3417 /*
3418  * Return POSIX pathconf information applicable to nfs filesystems.
3419  */
3420 static int
3421 nfs_pathconf(struct vop_pathconf_args *ap)
3422 {
3423 	struct nfsv3_pathconf pc;
3424 	struct nfsvattr nfsva;
3425 	struct vnode *vp = ap->a_vp;
3426 	struct thread *td = curthread;
3427 	int attrflag, error;
3428 
3429 	if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
3430 	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3431 	    ap->a_name == _PC_NO_TRUNC)) ||
3432 	    (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
3433 		/*
3434 		 * Since only the above 4 a_names are returned by the NFSv3
3435 		 * Pathconf RPC, there is no point in doing it for others.
3436 		 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
3437 		 * be used for _PC_NFS4_ACL as well.
3438 		 */
3439 		error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3440 		    &attrflag, NULL);
3441 		if (attrflag != 0)
3442 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3443 			    1);
3444 		if (error != 0)
3445 			return (error);
3446 	} else {
3447 		/*
3448 		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3449 		 * just fake them.
3450 		 */
3451 		pc.pc_linkmax = LINK_MAX;
3452 		pc.pc_namemax = NFS_MAXNAMLEN;
3453 		pc.pc_notrunc = 1;
3454 		pc.pc_chownrestricted = 1;
3455 		pc.pc_caseinsensitive = 0;
3456 		pc.pc_casepreserving = 1;
3457 		error = 0;
3458 	}
3459 	switch (ap->a_name) {
3460 	case _PC_LINK_MAX:
3461 		*ap->a_retval = pc.pc_linkmax;
3462 		break;
3463 	case _PC_NAME_MAX:
3464 		*ap->a_retval = pc.pc_namemax;
3465 		break;
3466 	case _PC_CHOWN_RESTRICTED:
3467 		*ap->a_retval = pc.pc_chownrestricted;
3468 		break;
3469 	case _PC_NO_TRUNC:
3470 		*ap->a_retval = pc.pc_notrunc;
3471 		break;
3472 	case _PC_ACL_EXTENDED:
3473 		*ap->a_retval = 0;
3474 		break;
3475 	case _PC_ACL_NFS4:
3476 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3477 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3478 			*ap->a_retval = 1;
3479 		else
3480 			*ap->a_retval = 0;
3481 		break;
3482 	case _PC_ACL_PATH_MAX:
3483 		if (NFS_ISV4(vp))
3484 			*ap->a_retval = ACL_MAX_ENTRIES;
3485 		else
3486 			*ap->a_retval = 3;
3487 		break;
3488 	case _PC_MAC_PRESENT:
3489 		*ap->a_retval = 0;
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 = vop_stdpathconf(ap);
3524 		break;
3525 	}
3526 	return (error);
3527 }
3528 
3529