xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision f4f8f02054f3abb6ceb84aefcdecc78d5c8b462f)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	from nfs_vnops.c	8.16 (Berkeley) 5/27/95
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 /*
39  * vnode op calls for Sun NFS version 2, 3 and 4
40  */
41 
42 #include "opt_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/malloc.h>
53 #include <sys/mbuf.h>
54 #include <sys/namei.h>
55 #include <sys/socket.h>
56 #include <sys/vnode.h>
57 #include <sys/dirent.h>
58 #include <sys/fcntl.h>
59 #include <sys/lockf.h>
60 #include <sys/stat.h>
61 #include <sys/sysctl.h>
62 #include <sys/signalvar.h>
63 
64 #include <vm/vm.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_extern.h>
67 #include <vm/vm_object.h>
68 
69 
70 #include <fs/nfs/nfsport.h>
71 #include <fs/nfsclient/nfsnode.h>
72 #include <fs/nfsclient/nfsmount.h>
73 #include <fs/nfsclient/nfs.h>
74 #include <fs/nfsclient/nfs_lock.h>
75 
76 #include <net/if.h>
77 #include <netinet/vinet.h>
78 #include <netinet/in.h>
79 #include <netinet/in_var.h>
80 
81 /* Defs */
82 #define	TRUE	1
83 #define	FALSE	0
84 
85 extern struct nfsstats newnfsstats;
86 MALLOC_DECLARE(M_NEWNFSREQ);
87 vop_advlock_t	*ncl_advlock_p = ncl_dolock;
88 
89 /*
90  * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
91  * calls are not in getblk() and brelse() so that they would not be necessary
92  * here.
93  */
94 #ifndef B_VMIO
95 #define	vfs_busy_pages(bp, f)
96 #endif
97 
98 static vop_read_t	nfsfifo_read;
99 static vop_write_t	nfsfifo_write;
100 static vop_close_t	nfsfifo_close;
101 static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
102 		    struct thread *);
103 static vop_lookup_t	nfs_lookup;
104 static vop_create_t	nfs_create;
105 static vop_mknod_t	nfs_mknod;
106 static vop_open_t	nfs_open;
107 static vop_close_t	nfs_close;
108 static vop_access_t	nfs_access;
109 static vop_getattr_t	nfs_getattr;
110 static vop_setattr_t	nfs_setattr;
111 static vop_read_t	nfs_read;
112 static vop_fsync_t	nfs_fsync;
113 static vop_remove_t	nfs_remove;
114 static vop_link_t	nfs_link;
115 static vop_rename_t	nfs_rename;
116 static vop_mkdir_t	nfs_mkdir;
117 static vop_rmdir_t	nfs_rmdir;
118 static vop_symlink_t	nfs_symlink;
119 static vop_readdir_t	nfs_readdir;
120 static vop_strategy_t	nfs_strategy;
121 static vop_lock1_t	nfs_lock1;
122 static	int	nfs_lookitup(struct vnode *, char *, int,
123 		    struct ucred *, struct thread *, struct nfsnode **);
124 static	int	nfs_sillyrename(struct vnode *, struct vnode *,
125 		    struct componentname *);
126 static vop_access_t	nfsspec_access;
127 static vop_readlink_t	nfs_readlink;
128 static vop_print_t	nfs_print;
129 static vop_advlock_t	nfs_advlock;
130 static vop_advlockasync_t nfs_advlockasync;
131 #ifdef NFS4_ACL_EXTATTR_NAME
132 static vop_getacl_t nfs_getacl;
133 static vop_setacl_t nfs_setacl;
134 #endif
135 
136 /*
137  * Global vfs data structures for nfs
138  */
139 struct vop_vector newnfs_vnodeops = {
140 	.vop_default =		&default_vnodeops,
141 	.vop_access =		nfs_access,
142 	.vop_advlock =		nfs_advlock,
143 	.vop_advlockasync =	nfs_advlockasync,
144 	.vop_close =		nfs_close,
145 	.vop_create =		nfs_create,
146 	.vop_fsync =		nfs_fsync,
147 	.vop_getattr =		nfs_getattr,
148 	.vop_getpages =		ncl_getpages,
149 	.vop_putpages =		ncl_putpages,
150 	.vop_inactive =		ncl_inactive,
151 	.vop_link =		nfs_link,
152 	.vop_lock1 = 		nfs_lock1,
153 	.vop_lookup =		nfs_lookup,
154 	.vop_mkdir =		nfs_mkdir,
155 	.vop_mknod =		nfs_mknod,
156 	.vop_open =		nfs_open,
157 	.vop_print =		nfs_print,
158 	.vop_read =		nfs_read,
159 	.vop_readdir =		nfs_readdir,
160 	.vop_readlink =		nfs_readlink,
161 	.vop_reclaim =		ncl_reclaim,
162 	.vop_remove =		nfs_remove,
163 	.vop_rename =		nfs_rename,
164 	.vop_rmdir =		nfs_rmdir,
165 	.vop_setattr =		nfs_setattr,
166 	.vop_strategy =		nfs_strategy,
167 	.vop_symlink =		nfs_symlink,
168 	.vop_write =		ncl_write,
169 #ifdef NFS4_ACL_EXTATTR_NAME
170 	.vop_getacl =		nfs_getacl,
171 	.vop_setacl =		nfs_setacl,
172 #endif
173 };
174 
175 struct vop_vector newnfs_fifoops = {
176 	.vop_default =		&fifo_specops,
177 	.vop_access =		nfsspec_access,
178 	.vop_close =		nfsfifo_close,
179 	.vop_fsync =		nfs_fsync,
180 	.vop_getattr =		nfs_getattr,
181 	.vop_inactive =		ncl_inactive,
182 	.vop_print =		nfs_print,
183 	.vop_read =		nfsfifo_read,
184 	.vop_reclaim =		ncl_reclaim,
185 	.vop_setattr =		nfs_setattr,
186 	.vop_write =		nfsfifo_write,
187 };
188 
189 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
190     struct componentname *cnp, struct vattr *vap);
191 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
192     int namelen, struct ucred *cred, struct thread *td);
193 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
194     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
195     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
196 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
197     struct componentname *scnp, struct sillyrename *sp);
198 
199 /*
200  * Global variables
201  */
202 #define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
203 
204 SYSCTL_DECL(_vfs_newnfs);
205 
206 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
207 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
208 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
209 
210 static int	nfs_prime_access_cache = 0;
211 SYSCTL_INT(_vfs_newnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
212 	   &nfs_prime_access_cache, 0,
213 	   "Prime NFS ACCESS cache when fetching attributes");
214 
215 static int	newnfs_commit_on_close = 0;
216 SYSCTL_INT(_vfs_newnfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
217     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
218 
219 static int	nfs_clean_pages_on_close = 1;
220 SYSCTL_INT(_vfs_newnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
221 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
222 
223 int newnfs_directio_enable = 0;
224 SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_enable, CTLFLAG_RW,
225 	   &newnfs_directio_enable, 0, "Enable NFS directio");
226 
227 static int newnfs_neglookup_enable = 1;
228 SYSCTL_INT(_vfs_newnfs, OID_AUTO, neglookup_enable, CTLFLAG_RW,
229     &newnfs_neglookup_enable, 0, "Enable NFS negative lookup caching");
230 
231 /*
232  * This sysctl allows other processes to mmap a file that has been opened
233  * O_DIRECT by a process.  In general, having processes mmap the file while
234  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
235  * this by default to prevent DoS attacks - to prevent a malicious user from
236  * opening up files O_DIRECT preventing other users from mmap'ing these
237  * files.  "Protected" environments where stricter consistency guarantees are
238  * required can disable this knob.  The process that opened the file O_DIRECT
239  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
240  * meaningful.
241  */
242 int newnfs_directio_allow_mmap = 1;
243 SYSCTL_INT(_vfs_newnfs, OID_AUTO, directio_allow_mmap, CTLFLAG_RW,
244 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
245 
246 #if 0
247 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
248 	   &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
249 
250 SYSCTL_INT(_vfs_newnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
251 	   &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
252 #endif
253 
254 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
255 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
256 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
257 
258 /*
259  * SMP Locking Note :
260  * The list of locks after the description of the lock is the ordering
261  * of other locks acquired with the lock held.
262  * np->n_mtx : Protects the fields in the nfsnode.
263        VM Object Lock
264        VI_MTX (acquired indirectly)
265  * nmp->nm_mtx : Protects the fields in the nfsmount.
266        rep->r_mtx
267  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
268  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
269        nmp->nm_mtx
270        rep->r_mtx
271  * rep->r_mtx : Protects the fields in an nfsreq.
272  */
273 
274 static int
275 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
276     struct ucred *cred, u_int32_t *retmode)
277 {
278 	int error = 0, attrflag, i, lrupos;
279 	u_int32_t rmode;
280 	struct nfsnode *np = VTONFS(vp);
281 	struct nfsvattr nfsva;
282 
283 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
284 	    &rmode, NULL);
285 	if (attrflag)
286 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
287 	if (!error) {
288 		lrupos = 0;
289 		mtx_lock(&np->n_mtx);
290 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
291 			if (np->n_accesscache[i].uid == cred->cr_uid) {
292 				np->n_accesscache[i].mode = rmode;
293 				np->n_accesscache[i].stamp = time_second;
294 				break;
295 			}
296 			if (i > 0 && np->n_accesscache[i].stamp <
297 			    np->n_accesscache[lrupos].stamp)
298 				lrupos = i;
299 		}
300 		if (i == NFS_ACCESSCACHESIZE) {
301 			np->n_accesscache[lrupos].uid = cred->cr_uid;
302 			np->n_accesscache[lrupos].mode = rmode;
303 			np->n_accesscache[lrupos].stamp = time_second;
304 		}
305 		mtx_unlock(&np->n_mtx);
306 		if (retmode != NULL)
307 			*retmode = rmode;
308 	} else if (NFS_ISV4(vp)) {
309 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
310 	}
311 	return (error);
312 }
313 
314 /*
315  * nfs access vnode op.
316  * For nfs version 2, just return ok. File accesses may fail later.
317  * For nfs version 3, use the access rpc to check accessibility. If file modes
318  * are changed on the server, accesses might still fail later.
319  */
320 static int
321 nfs_access(struct vop_access_args *ap)
322 {
323 	struct vnode *vp = ap->a_vp;
324 	int error = 0, i, gotahit;
325 	u_int32_t mode, wmode, rmode;
326 	int v34 = NFS_ISV34(vp);
327 	struct nfsnode *np = VTONFS(vp);
328 
329 	/*
330 	 * Disallow write attempts on filesystems mounted read-only;
331 	 * unless the file is a socket, fifo, or a block or character
332 	 * device resident on the filesystem.
333 	 */
334 	if ((ap->a_accmode & (VWRITE | VAPPEND
335 #ifdef NFS4_ACL_EXTATTR_NAME
336 	    | VWRITE_NAMED_ATTRS | VDELETE_CHILD | VWRITE_ATTRIBUTES |
337 	    VDELETE | VWRITE_ACL | VWRITE_OWNER
338 #endif
339 	    )) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
340 		switch (vp->v_type) {
341 		case VREG:
342 		case VDIR:
343 		case VLNK:
344 			return (EROFS);
345 		default:
346 			break;
347 		}
348 	}
349 	/*
350 	 * For nfs v3 or v4, check to see if we have done this recently, and if
351 	 * so return our cached result instead of making an ACCESS call.
352 	 * If not, do an access rpc, otherwise you are stuck emulating
353 	 * ufs_access() locally using the vattr. This may not be correct,
354 	 * since the server may apply other access criteria such as
355 	 * client uid-->server uid mapping that we do not know about.
356 	 */
357 	if (v34) {
358 		if (ap->a_accmode & VREAD)
359 			mode = NFSACCESS_READ;
360 		else
361 			mode = 0;
362 		if (vp->v_type != VDIR) {
363 			if (ap->a_accmode & VWRITE)
364 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
365 			if (ap->a_accmode & VAPPEND)
366 				mode |= NFSACCESS_EXTEND;
367 			if (ap->a_accmode & VEXEC)
368 				mode |= NFSACCESS_EXECUTE;
369 #ifdef NFS4_ACL_EXTATTR_NAME
370 			if (ap->a_accmode & VDELETE)
371 				mode |= NFSACCESS_DELETE;
372 #endif
373 		} else {
374 			if (ap->a_accmode & VWRITE)
375 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
376 			if (ap->a_accmode & VAPPEND)
377 				mode |= NFSACCESS_EXTEND;
378 			if (ap->a_accmode & VEXEC)
379 				mode |= NFSACCESS_LOOKUP;
380 #ifdef NFS4_ACL_EXTATTR_NAME
381 			if (ap->a_accmode & VDELETE)
382 				mode |= NFSACCESS_DELETE;
383 			if (ap->a_accmode & VDELETE_CHILD)
384 				mode |= NFSACCESS_MODIFY;
385 #endif
386 		}
387 		/* XXX safety belt, only make blanket request if caching */
388 		if (nfsaccess_cache_timeout > 0) {
389 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
390 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
391 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
392 		} else {
393 			wmode = mode;
394 		}
395 
396 		/*
397 		 * Does our cached result allow us to give a definite yes to
398 		 * this request?
399 		 */
400 		gotahit = 0;
401 		mtx_lock(&np->n_mtx);
402 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
403 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
404 			    if (time_second < (np->n_accesscache[i].stamp
405 				+ nfsaccess_cache_timeout) &&
406 				(np->n_accesscache[i].mode & mode) == mode) {
407 				NFSINCRGLOBAL(newnfsstats.accesscache_hits);
408 				gotahit = 1;
409 			    }
410 			    break;
411 			}
412 		}
413 		mtx_unlock(&np->n_mtx);
414 		if (gotahit == 0) {
415 			/*
416 			 * Either a no, or a don't know.  Go to the wire.
417 			 */
418 			NFSINCRGLOBAL(newnfsstats.accesscache_misses);
419 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
420 			    ap->a_cred, &rmode);
421 			if (!error &&
422 			    (rmode & mode) != mode)
423 				error = EACCES;
424 		}
425 		return (error);
426 	} else {
427 		if ((error = nfsspec_access(ap)) != 0) {
428 			return (error);
429 		}
430 		/*
431 		 * Attempt to prevent a mapped root from accessing a file
432 		 * which it shouldn't.  We try to read a byte from the file
433 		 * if the user is root and the file is not zero length.
434 		 * After calling nfsspec_access, we should have the correct
435 		 * file size cached.
436 		 */
437 		mtx_lock(&np->n_mtx);
438 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
439 		    && VTONFS(vp)->n_size > 0) {
440 			struct iovec aiov;
441 			struct uio auio;
442 			char buf[1];
443 
444 			mtx_unlock(&np->n_mtx);
445 			aiov.iov_base = buf;
446 			aiov.iov_len = 1;
447 			auio.uio_iov = &aiov;
448 			auio.uio_iovcnt = 1;
449 			auio.uio_offset = 0;
450 			auio.uio_resid = 1;
451 			auio.uio_segflg = UIO_SYSSPACE;
452 			auio.uio_rw = UIO_READ;
453 			auio.uio_td = ap->a_td;
454 
455 			if (vp->v_type == VREG)
456 				error = ncl_readrpc(vp, &auio, ap->a_cred);
457 			else if (vp->v_type == VDIR) {
458 				char* bp;
459 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
460 				aiov.iov_base = bp;
461 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
462 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
463 				    ap->a_td);
464 				free(bp, M_TEMP);
465 			} else if (vp->v_type == VLNK)
466 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
467 			else
468 				error = EACCES;
469 		} else
470 			mtx_unlock(&np->n_mtx);
471 		return (error);
472 	}
473 }
474 
475 
476 /*
477  * nfs open vnode op
478  * Check to see if the type is ok
479  * and that deletion is not in progress.
480  * For paged in text files, you will need to flush the page cache
481  * if consistency is lost.
482  */
483 /* ARGSUSED */
484 static int
485 nfs_open(struct vop_open_args *ap)
486 {
487 	struct vnode *vp = ap->a_vp;
488 	struct nfsnode *np = VTONFS(vp);
489 	struct vattr vattr;
490 	int error;
491 	int fmode = ap->a_mode;
492 
493 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
494 		return (EOPNOTSUPP);
495 
496 	/*
497 	 * For NFSv4, we need to do the Open Op before cache validation,
498 	 * so that we conform to RFC3530 Sec. 9.3.1.
499 	 */
500 	if (NFS_ISV4(vp)) {
501 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
502 		if (error) {
503 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
504 			    (gid_t)0);
505 			return (error);
506 		}
507 	}
508 
509 	/*
510 	 * Now, if this Open will be doing reading, re-validate/flush the
511 	 * cache, so that Close/Open coherency is maintained.
512 	 */
513 	if ((fmode & FREAD) && (!NFS_ISV4(vp) || nfscl_mustflush(vp))) {
514 		mtx_lock(&np->n_mtx);
515 		if (np->n_flag & NMODIFIED) {
516 			mtx_unlock(&np->n_mtx);
517 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
518 			if (error == EINTR || error == EIO) {
519 				if (NFS_ISV4(vp))
520 					(void) nfsrpc_close(vp, ap->a_cred,
521 					    ap->a_td);
522 				return (error);
523 			}
524 			np->n_attrstamp = 0;
525 			if (vp->v_type == VDIR)
526 				np->n_direofoffset = 0;
527 			error = VOP_GETATTR(vp, &vattr, ap->a_cred);
528 			if (error) {
529 				if (NFS_ISV4(vp))
530 					(void) nfsrpc_close(vp, ap->a_cred,
531 					    ap->a_td);
532 				return (error);
533 			}
534 			mtx_lock(&np->n_mtx);
535 			np->n_mtime = vattr.va_mtime;
536 			if (NFS_ISV4(vp))
537 				np->n_change = vattr.va_filerev;
538 			mtx_unlock(&np->n_mtx);
539 		} else {
540 			struct thread *td = curthread;
541 
542 			if (np->n_ac_ts_syscalls != td->td_syscalls ||
543 			    np->n_ac_ts_tid != td->td_tid ||
544 			    td->td_proc == NULL ||
545 			    np->n_ac_ts_pid != td->td_proc->p_pid) {
546 				np->n_attrstamp = 0;
547 			}
548 			mtx_unlock(&np->n_mtx);
549 			error = VOP_GETATTR(vp, &vattr, ap->a_cred);
550 			if (error) {
551 				if (NFS_ISV4(vp))
552 					(void) nfsrpc_close(vp, ap->a_cred,
553 					    ap->a_td);
554 				return (error);
555 			}
556 			mtx_lock(&np->n_mtx);
557 			if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
558 			    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
559 				if (vp->v_type == VDIR)
560 					np->n_direofoffset = 0;
561 				mtx_unlock(&np->n_mtx);
562 				error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
563 				if (error == EINTR || error == EIO) {
564 					if (NFS_ISV4(vp))
565 						(void) nfsrpc_close(vp,
566 						    ap->a_cred, ap->a_td);
567 					return (error);
568 				}
569 				mtx_lock(&np->n_mtx);
570 				np->n_mtime = vattr.va_mtime;
571 				if (NFS_ISV4(vp))
572 					np->n_change = vattr.va_filerev;
573 			}
574 			mtx_unlock(&np->n_mtx);
575 		}
576 	}
577 
578 	/*
579 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
580 	 */
581 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
582 		if (np->n_directio_opens == 0) {
583 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
584 			if (error) {
585 				if (NFS_ISV4(vp))
586 					(void) nfsrpc_close(vp, ap->a_cred,
587 					    ap->a_td);
588 				return (error);
589 			}
590 			mtx_lock(&np->n_mtx);
591 			np->n_flag |= NNONCACHE;
592 		} else {
593 			mtx_lock(&np->n_mtx);
594 		}
595 		np->n_directio_opens++;
596 		mtx_unlock(&np->n_mtx);
597 	}
598 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
599 	return (0);
600 }
601 
602 /*
603  * nfs close vnode op
604  * What an NFS client should do upon close after writing is a debatable issue.
605  * Most NFS clients push delayed writes to the server upon close, basically for
606  * two reasons:
607  * 1 - So that any write errors may be reported back to the client process
608  *     doing the close system call. By far the two most likely errors are
609  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
610  * 2 - To put a worst case upper bound on cache inconsistency between
611  *     multiple clients for the file.
612  * There is also a consistency problem for Version 2 of the protocol w.r.t.
613  * not being able to tell if other clients are writing a file concurrently,
614  * since there is no way of knowing if the changed modify time in the reply
615  * is only due to the write for this client.
616  * (NFS Version 3 provides weak cache consistency data in the reply that
617  *  should be sufficient to detect and handle this case.)
618  *
619  * The current code does the following:
620  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
621  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
622  *                     or commit them (this satisfies 1 and 2 except for the
623  *                     case where the server crashes after this close but
624  *                     before the commit RPC, which is felt to be "good
625  *                     enough". Changing the last argument to ncl_flush() to
626  *                     a 1 would force a commit operation, if it is felt a
627  *                     commit is necessary now.
628  * for NFS Version 4 - flush the dirty buffers and commit them, if
629  *		       nfscl_mustflush() says this is necessary.
630  *                     It is necessary if there is no write delegation held,
631  *                     in order to satisfy open/close coherency.
632  *                     If the file isn't cached on local stable storage,
633  *                     it may be necessary in order to detect "out of space"
634  *                     errors from the server, if the write delegation
635  *                     issued by the server doesn't allow the file to grow.
636  */
637 /* ARGSUSED */
638 static int
639 nfs_close(struct vop_close_args *ap)
640 {
641 	struct vnode *vp = ap->a_vp;
642 	struct nfsnode *np = VTONFS(vp);
643 	struct nfsvattr nfsva;
644 	struct ucred *cred;
645 	int error = 0, ret, localcred = 0;
646 	int fmode = ap->a_fflag;
647 
648 	if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
649 		return (0);
650 	/*
651 	 * During shutdown, a_cred isn't valid, so just use root.
652 	 */
653 	if (ap->a_cred == NOCRED) {
654 		cred = newnfs_getcred();
655 		localcred = 1;
656 	} else {
657 		cred = ap->a_cred;
658 	}
659 	if (vp->v_type == VREG) {
660 	    /*
661 	     * Examine and clean dirty pages, regardless of NMODIFIED.
662 	     * This closes a major hole in close-to-open consistency.
663 	     * We want to push out all dirty pages (and buffers) on
664 	     * close, regardless of whether they were dirtied by
665 	     * mmap'ed writes or via write().
666 	     */
667 	    if (nfs_clean_pages_on_close && vp->v_object) {
668 		VM_OBJECT_LOCK(vp->v_object);
669 		vm_object_page_clean(vp->v_object, 0, 0, 0);
670 		VM_OBJECT_UNLOCK(vp->v_object);
671 	    }
672 	    mtx_lock(&np->n_mtx);
673 	    if (np->n_flag & NMODIFIED) {
674 		mtx_unlock(&np->n_mtx);
675 		if (NFS_ISV3(vp)) {
676 		    /*
677 		     * Under NFSv3 we have dirty buffers to dispose of.  We
678 		     * must flush them to the NFS server.  We have the option
679 		     * of waiting all the way through the commit rpc or just
680 		     * waiting for the initial write.  The default is to only
681 		     * wait through the initial write so the data is in the
682 		     * server's cache, which is roughly similar to the state
683 		     * a standard disk subsystem leaves the file in on close().
684 		     *
685 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
686 		     * potential races with other processes, and certainly
687 		     * cannot clear it if we don't commit.
688 		     * These races occur when there is no longer the old
689 		     * traditional vnode locking implemented for Vnode Ops.
690 		     */
691 		    int cm = newnfs_commit_on_close ? 1 : 0;
692 		    error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm);
693 		    /* np->n_flag &= ~NMODIFIED; */
694 		} else if (NFS_ISV4(vp)) {
695 			int cm;
696 			if (newnfs_commit_on_close != 0)
697 				cm = 1;
698 			else
699 				cm = nfscl_mustflush(vp);
700 			error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm);
701 			/* as above w.r.t. races when clearing NMODIFIED */
702 			/* np->n_flag &= ~NMODIFIED; */
703 		} else
704 		    error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
705 		mtx_lock(&np->n_mtx);
706 	    }
707  	    /*
708  	     * Invalidate the attribute cache in all cases.
709  	     * An open is going to fetch fresh attrs any way, other procs
710  	     * on this node that have file open will be forced to do an
711  	     * otw attr fetch, but this is safe.
712 	     * --> A user found that their RPC count dropped by 20% when
713 	     *     this was commented out and I can't see any requirement
714 	     *     for it, so I've disabled it when negative lookups are
715 	     *     enabled. (What does this have to do with negative lookup
716 	     *     caching? Well nothing, except it was reported by the
717 	     *     same user that needed negative lookup caching and I wanted
718 	     *     there to be a way to disable it via sysctl to see if it
719 	     *     is the cause of some caching/coherency issue that might
720 	     *     crop up.)
721  	     */
722 	    if (newnfs_neglookup_enable == 0)
723 		    np->n_attrstamp = 0;
724 	    if (np->n_flag & NWRITEERR) {
725 		np->n_flag &= ~NWRITEERR;
726 		error = np->n_error;
727 	    }
728 	    mtx_unlock(&np->n_mtx);
729 	}
730 
731 	if (NFS_ISV4(vp)) {
732 		/*
733 		 * Get attributes so "change" is up to date.
734 		 */
735 		if (!error) {
736 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
737 			    NULL);
738 			if (!ret) {
739 				np->n_change = nfsva.na_filerev;
740 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
741 				    NULL, 0, 0);
742 			}
743 		}
744 
745 		/*
746 		 * and do the close.
747 		 */
748 		ret = nfsrpc_close(vp, cred, ap->a_td);
749 		if (!error && ret)
750 			error = ret;
751 		if (error)
752 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
753 			    (gid_t)0);
754 	}
755 	if (newnfs_directio_enable)
756 		KASSERT((np->n_directio_asyncwr == 0),
757 			("nfs_close: dirty unflushed (%d) directio buffers\n",
758 			 np->n_directio_asyncwr));
759 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
760 		mtx_lock(&np->n_mtx);
761 		KASSERT((np->n_directio_opens > 0),
762 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
763 		np->n_directio_opens--;
764 		if (np->n_directio_opens == 0)
765 			np->n_flag &= ~NNONCACHE;
766 		mtx_unlock(&np->n_mtx);
767 	}
768 	if (localcred)
769 		NFSFREECRED(cred);
770 	return (error);
771 }
772 
773 /*
774  * nfs getattr call from vfs.
775  */
776 static int
777 nfs_getattr(struct vop_getattr_args *ap)
778 {
779 	struct vnode *vp = ap->a_vp;
780 	struct thread *td = curthread;	/* XXX */
781 	struct nfsnode *np = VTONFS(vp);
782 	int error = 0;
783 	struct nfsvattr nfsva;
784 	struct vattr *vap = ap->a_vap;
785 	struct vattr vattr;
786 
787 	/*
788 	 * Update local times for special files.
789 	 */
790 	mtx_lock(&np->n_mtx);
791 	if (np->n_flag & (NACC | NUPD))
792 		np->n_flag |= NCHG;
793 	mtx_unlock(&np->n_mtx);
794 	/*
795 	 * First look in the cache.
796 	 */
797 	if (ncl_getattrcache(vp, &vattr) == 0) {
798 		vap->va_type = vattr.va_type;
799 		vap->va_mode = vattr.va_mode;
800 		vap->va_nlink = vattr.va_nlink;
801 		vap->va_uid = vattr.va_uid;
802 		vap->va_gid = vattr.va_gid;
803 		vap->va_fsid = vattr.va_fsid;
804 		vap->va_fileid = vattr.va_fileid;
805 		vap->va_size = vattr.va_size;
806 		vap->va_blocksize = vattr.va_blocksize;
807 		vap->va_atime = vattr.va_atime;
808 		vap->va_mtime = vattr.va_mtime;
809 		vap->va_ctime = vattr.va_ctime;
810 		vap->va_gen = vattr.va_gen;
811 		vap->va_flags = vattr.va_flags;
812 		vap->va_rdev = vattr.va_rdev;
813 		vap->va_bytes = vattr.va_bytes;
814 		vap->va_filerev = vattr.va_filerev;
815 		/*
816 		 * Get the local modify time for the case of a write
817 		 * delegation.
818 		 */
819 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
820 		return (0);
821 	}
822 
823 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
824 	    nfsaccess_cache_timeout > 0) {
825 		NFSINCRGLOBAL(newnfsstats.accesscache_misses);
826 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
827 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
828 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
829 			return (0);
830 		}
831 	}
832 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
833 	if (!error)
834 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
835 	if (!error) {
836 		/*
837 		 * Get the local modify time for the case of a write
838 		 * delegation.
839 		 */
840 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
841 	} else if (NFS_ISV4(vp)) {
842 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
843 	}
844 	return (error);
845 }
846 
847 /*
848  * nfs setattr call.
849  */
850 static int
851 nfs_setattr(struct vop_setattr_args *ap)
852 {
853 	struct vnode *vp = ap->a_vp;
854 	struct nfsnode *np = VTONFS(vp);
855 	struct thread *td = curthread;	/* XXX */
856 	struct vattr *vap = ap->a_vap;
857 	int error = 0;
858 	u_quad_t tsize;
859 
860 #ifndef nolint
861 	tsize = (u_quad_t)0;
862 #endif
863 
864 	/*
865 	 * Setting of flags and marking of atimes are not supported.
866 	 */
867 	if (vap->va_flags != VNOVAL)
868 		return (EOPNOTSUPP);
869 
870 	/*
871 	 * Disallow write attempts if the filesystem is mounted read-only.
872 	 */
873   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
874 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
875 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
876 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
877 		return (EROFS);
878 	if (vap->va_size != VNOVAL) {
879  		switch (vp->v_type) {
880  		case VDIR:
881  			return (EISDIR);
882  		case VCHR:
883  		case VBLK:
884  		case VSOCK:
885  		case VFIFO:
886 			if (vap->va_mtime.tv_sec == VNOVAL &&
887 			    vap->va_atime.tv_sec == VNOVAL &&
888 			    vap->va_mode == (mode_t)VNOVAL &&
889 			    vap->va_uid == (uid_t)VNOVAL &&
890 			    vap->va_gid == (gid_t)VNOVAL)
891 				return (0);
892  			vap->va_size = VNOVAL;
893  			break;
894  		default:
895 			/*
896 			 * Disallow write attempts if the filesystem is
897 			 * mounted read-only.
898 			 */
899 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
900 				return (EROFS);
901 			/*
902 			 *  We run vnode_pager_setsize() early (why?),
903 			 * we must set np->n_size now to avoid vinvalbuf
904 			 * V_SAVE races that might setsize a lower
905 			 * value.
906 			 */
907 			mtx_lock(&np->n_mtx);
908 			tsize = np->n_size;
909 			mtx_unlock(&np->n_mtx);
910 			error = ncl_meta_setsize(vp, ap->a_cred, td,
911 			    vap->va_size);
912 			mtx_lock(&np->n_mtx);
913  			if (np->n_flag & NMODIFIED) {
914 			    tsize = np->n_size;
915 			    mtx_unlock(&np->n_mtx);
916  			    if (vap->va_size == 0)
917  				error = ncl_vinvalbuf(vp, 0, td, 1);
918  			    else
919  				error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
920  			    if (error) {
921 				vnode_pager_setsize(vp, tsize);
922 				return (error);
923 			    }
924 			    /*
925 			     * Call nfscl_delegmodtime() to set the modify time
926 			     * locally, as required.
927 			     */
928 			    nfscl_delegmodtime(vp);
929  			} else
930 			    mtx_unlock(&np->n_mtx);
931 			/*
932 			 * np->n_size has already been set to vap->va_size
933 			 * in ncl_meta_setsize(). We must set it again since
934 			 * nfs_loadattrcache() could be called through
935 			 * ncl_meta_setsize() and could modify np->n_size.
936 			 */
937 			mtx_lock(&np->n_mtx);
938  			np->n_vattr.na_size = np->n_size = vap->va_size;
939 			mtx_unlock(&np->n_mtx);
940   		};
941   	} else {
942 		mtx_lock(&np->n_mtx);
943 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
944 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
945 			mtx_unlock(&np->n_mtx);
946 			if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
947 			    (error == EINTR || error == EIO))
948 				return (error);
949 		} else
950 			mtx_unlock(&np->n_mtx);
951 	}
952 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
953 	if (error && vap->va_size != VNOVAL) {
954 		mtx_lock(&np->n_mtx);
955 		np->n_size = np->n_vattr.na_size = tsize;
956 		vnode_pager_setsize(vp, tsize);
957 		mtx_unlock(&np->n_mtx);
958 	}
959 	return (error);
960 }
961 
962 /*
963  * Do an nfs setattr rpc.
964  */
965 static int
966 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
967     struct thread *td)
968 {
969 	struct nfsnode *np = VTONFS(vp);
970 	int error, ret, attrflag, i;
971 	struct nfsvattr nfsva;
972 
973 	if (NFS_ISV34(vp)) {
974 		mtx_lock(&np->n_mtx);
975 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
976 			np->n_accesscache[i].stamp = 0;
977 		np->n_flag |= NDELEGMOD;
978 		mtx_unlock(&np->n_mtx);
979 	}
980 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
981 	    NULL);
982 	if (attrflag) {
983 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
984 		if (ret && !error)
985 			error = ret;
986 	}
987 	if (error && NFS_ISV4(vp))
988 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
989 	return (error);
990 }
991 
992 /*
993  * nfs lookup call, one step at a time...
994  * First look in cache
995  * If not found, unlock the directory nfsnode and do the rpc
996  */
997 static int
998 nfs_lookup(struct vop_lookup_args *ap)
999 {
1000 	struct componentname *cnp = ap->a_cnp;
1001 	struct vnode *dvp = ap->a_dvp;
1002 	struct vnode **vpp = ap->a_vpp;
1003 	int flags = cnp->cn_flags;
1004 	struct vnode *newvp;
1005 	struct nfsmount *nmp;
1006 	struct nfsnode *np;
1007 	int error = 0, attrflag, dattrflag;
1008 	struct thread *td = cnp->cn_thread;
1009 	struct nfsfh *nfhp;
1010 	struct nfsvattr dnfsva, nfsva;
1011 
1012 	*vpp = NULLVP;
1013 	if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
1014 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1015 		return (EROFS);
1016 	if (dvp->v_type != VDIR)
1017 		return (ENOTDIR);
1018 	nmp = VFSTONFS(dvp->v_mount);
1019 	np = VTONFS(dvp);
1020 
1021 	/* For NFSv4, wait until any remove is done. */
1022 	mtx_lock(&np->n_mtx);
1023 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1024 		np->n_flag |= NREMOVEWANT;
1025 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1026 	}
1027 	mtx_unlock(&np->n_mtx);
1028 
1029 	if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1030 		return (error);
1031 	if ((error = cache_lookup(dvp, vpp, cnp)) &&
1032 	    (error != ENOENT || newnfs_neglookup_enable != 0)) {
1033 		struct vattr vattr;
1034 
1035 		if (error == ENOENT) {
1036 			if (!VOP_GETATTR(dvp, &vattr, cnp->cn_cred) &&
1037 			    vattr.va_mtime.tv_sec == np->n_dmtime) {
1038 			     NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1039 				return (ENOENT);
1040 			}
1041 			cache_purge_negative(dvp);
1042 			np->n_dmtime = 0;
1043 		} else {
1044 			newvp = *vpp;
1045 			if (nfscl_nodeleg(newvp, 0) == 0 ||
1046 			    (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred) &&
1047 			     vattr.va_ctime.tv_sec==VTONFS(newvp)->n_ctime)) {
1048 			     NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1049 			     if (cnp->cn_nameiop != LOOKUP &&
1050 				 (flags & ISLASTCN))
1051 				     cnp->cn_flags |= SAVENAME;
1052 			     return (0);
1053 			}
1054 			cache_purge(newvp);
1055 			if (dvp != newvp)
1056 				vput(newvp);
1057 			else
1058 				vrele(newvp);
1059 			*vpp = NULLVP;
1060 		}
1061 	}
1062 	error = 0;
1063 	newvp = NULLVP;
1064 	NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1065 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1066 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1067 	    NULL);
1068 	if (dattrflag)
1069 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1070 	if (error) {
1071 		if (newnfs_neglookup_enable != 0 &&
1072 		    error == ENOENT && (cnp->cn_flags & MAKEENTRY) &&
1073 		    cnp->cn_nameiop != CREATE) {
1074 			if (np->n_dmtime == 0)
1075 				np->n_dmtime = np->n_vattr.na_mtime.tv_sec;
1076 			cache_enter(dvp, NULL, cnp);
1077 		}
1078 		if (newvp != NULLVP) {
1079 			vput(newvp);
1080 			*vpp = NULLVP;
1081 		}
1082 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1083 		    (flags & ISLASTCN) && error == ENOENT) {
1084 			if (dvp->v_mount->mnt_flag & MNT_RDONLY)
1085 				error = EROFS;
1086 			else
1087 				error = EJUSTRETURN;
1088 		}
1089 		if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1090 			cnp->cn_flags |= SAVENAME;
1091 		if (NFS_ISV4(dvp))
1092 			error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1093 		return (error);
1094 	}
1095 
1096 	/*
1097 	 * Handle RENAME case...
1098 	 */
1099 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1100 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1101 			FREE((caddr_t)nfhp, M_NFSFH);
1102 			return (EISDIR);
1103 		}
1104 		error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, td, &np,
1105 		    NULL);
1106 		if (error)
1107 			return (error);
1108 		newvp = NFSTOV(np);
1109 		if (attrflag)
1110 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1111 			    0, 1);
1112 		*vpp = newvp;
1113 		cnp->cn_flags |= SAVENAME;
1114 		return (0);
1115 	}
1116 
1117 	if ((flags & ISDOTDOT)) {
1118 		VOP_UNLOCK(dvp, 0);
1119 		error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, td, &np,
1120 		    NULL);
1121 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1122 		if (error)
1123 			return (error);
1124 		newvp = NFSTOV(np);
1125 		if (attrflag)
1126 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1127 			    0, 1);
1128 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1129 		FREE((caddr_t)nfhp, M_NFSFH);
1130 		VREF(dvp);
1131 		newvp = dvp;
1132 		if (attrflag)
1133 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1134 			    0, 1);
1135 	} else {
1136 		error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, td, &np,
1137 		    NULL);
1138 		if (error)
1139 			return (error);
1140 		newvp = NFSTOV(np);
1141 		if (attrflag)
1142 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1143 			    0, 1);
1144 	}
1145 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1146 		cnp->cn_flags |= SAVENAME;
1147 	if ((cnp->cn_flags & MAKEENTRY) &&
1148 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
1149 		np->n_ctime = np->n_vattr.na_vattr.va_ctime.tv_sec;
1150 		cache_enter(dvp, newvp, cnp);
1151 	}
1152 	*vpp = newvp;
1153 	return (0);
1154 }
1155 
1156 /*
1157  * nfs read call.
1158  * Just call ncl_bioread() to do the work.
1159  */
1160 static int
1161 nfs_read(struct vop_read_args *ap)
1162 {
1163 	struct vnode *vp = ap->a_vp;
1164 
1165 	switch (vp->v_type) {
1166 	case VREG:
1167 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1168 	case VDIR:
1169 		return (EISDIR);
1170 	default:
1171 		return (EOPNOTSUPP);
1172 	}
1173 }
1174 
1175 /*
1176  * nfs readlink call
1177  */
1178 static int
1179 nfs_readlink(struct vop_readlink_args *ap)
1180 {
1181 	struct vnode *vp = ap->a_vp;
1182 
1183 	if (vp->v_type != VLNK)
1184 		return (EINVAL);
1185 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1186 }
1187 
1188 /*
1189  * Do a readlink rpc.
1190  * Called by ncl_doio() from below the buffer cache.
1191  */
1192 int
1193 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1194 {
1195 	int error, ret, attrflag;
1196 	struct nfsvattr nfsva;
1197 
1198 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1199 	    &attrflag, NULL);
1200 	if (attrflag) {
1201 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1202 		if (ret && !error)
1203 			error = ret;
1204 	}
1205 	if (error && NFS_ISV4(vp))
1206 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1207 	return (error);
1208 }
1209 
1210 /*
1211  * nfs read rpc call
1212  * Ditto above
1213  */
1214 int
1215 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1216 {
1217 	int error, ret, attrflag;
1218 	struct nfsvattr nfsva;
1219 
1220 	error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva, &attrflag,
1221 	    NULL);
1222 	if (attrflag) {
1223 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1224 		if (ret && !error)
1225 			error = ret;
1226 	}
1227 	if (error && NFS_ISV4(vp))
1228 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1229 	return (error);
1230 }
1231 
1232 /*
1233  * nfs write call
1234  */
1235 int
1236 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1237 	     int *iomode, int *must_commit)
1238 {
1239 	struct nfsvattr nfsva;
1240 	int error = 0, attrflag, ret;
1241 	u_char verf[NFSX_VERF];
1242 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1243 
1244 	*must_commit = 0;
1245 	error = nfsrpc_write(vp, uiop, iomode, verf, cred,
1246 	    uiop->uio_td, &nfsva, &attrflag, NULL);
1247 	NFSLOCKMNT(nmp);
1248 	if (!error && NFSHASWRITEVERF(nmp) &&
1249 	    NFSBCMP(verf, nmp->nm_verf, NFSX_VERF)) {
1250 		*must_commit = 1;
1251 		NFSBCOPY(verf, nmp->nm_verf, NFSX_VERF);
1252 	}
1253 	NFSUNLOCKMNT(nmp);
1254 	if (attrflag) {
1255 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1256 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1257 			    1);
1258 		else
1259 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1260 			    1);
1261 		if (ret && !error)
1262 			error = ret;
1263 	}
1264 	if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1265 		*iomode = NFSWRITE_FILESYNC;
1266 	if (error && NFS_ISV4(vp))
1267 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1268 	return (error);
1269 }
1270 
1271 /*
1272  * nfs mknod rpc
1273  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1274  * mode set to specify the file type and the size field for rdev.
1275  */
1276 static int
1277 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1278     struct vattr *vap)
1279 {
1280 	struct nfsvattr nfsva, dnfsva;
1281 	struct vnode *newvp = NULL;
1282 	struct nfsnode *np = NULL, *dnp;
1283 	struct nfsfh *nfhp;
1284 	struct vattr vattr;
1285 	int error = 0, attrflag, dattrflag;
1286 	u_int32_t rdev;
1287 
1288 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1289 		rdev = vap->va_rdev;
1290 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1291 		rdev = 0xffffffff;
1292 	else
1293 		return (EOPNOTSUPP);
1294 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1295 		return (error);
1296 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1297 	    rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1298 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1299 	if (!error) {
1300 		if (!nfhp)
1301 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1302 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1303 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1304 			    NULL);
1305 		if (nfhp)
1306 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1307 			    cnp->cn_thread, &np, NULL);
1308 	}
1309 	if (dattrflag)
1310 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1311 	if (!error) {
1312 		newvp = NFSTOV(np);
1313 		if (attrflag)
1314 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1315 			    0, 1);
1316 	}
1317 	if (!error) {
1318 		if ((cnp->cn_flags & MAKEENTRY))
1319 			cache_enter(dvp, newvp, cnp);
1320 		*vpp = newvp;
1321 	} else if (NFS_ISV4(dvp)) {
1322 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1323 		    vap->va_gid);
1324 	}
1325 	dnp = VTONFS(dvp);
1326 	mtx_lock(&dnp->n_mtx);
1327 	dnp->n_flag |= NMODIFIED;
1328 	if (!dattrflag)
1329 		dnp->n_attrstamp = 0;
1330 	mtx_unlock(&dnp->n_mtx);
1331 	return (error);
1332 }
1333 
1334 /*
1335  * nfs mknod vop
1336  * just call nfs_mknodrpc() to do the work.
1337  */
1338 /* ARGSUSED */
1339 static int
1340 nfs_mknod(struct vop_mknod_args *ap)
1341 {
1342 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1343 }
1344 
1345 static u_long create_verf;
1346 /*
1347  * nfs file create call
1348  */
1349 static int
1350 nfs_create(struct vop_create_args *ap)
1351 {
1352 	struct vnode *dvp = ap->a_dvp;
1353 	struct vattr *vap = ap->a_vap;
1354 	struct componentname *cnp = ap->a_cnp;
1355 	struct nfsnode *np = NULL, *dnp;
1356 	struct vnode *newvp = NULL;
1357 	struct nfsmount *nmp;
1358 	struct nfsvattr dnfsva, nfsva;
1359 	struct nfsfh *nfhp;
1360 	nfsquad_t cverf;
1361 	int error = 0, attrflag, dattrflag, fmode = 0;
1362 	struct vattr vattr;
1363 
1364 	/*
1365 	 * Oops, not for me..
1366 	 */
1367 	if (vap->va_type == VSOCK)
1368 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1369 
1370 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1371 		return (error);
1372 	if (vap->va_vaflags & VA_EXCLUSIVE)
1373 		fmode |= O_EXCL;
1374 	dnp = VTONFS(dvp);
1375 	nmp = VFSTONFS(vnode_mount(dvp));
1376 again:
1377 	/* For NFSv4, wait until any remove is done. */
1378 	mtx_lock(&dnp->n_mtx);
1379 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1380 		dnp->n_flag |= NREMOVEWANT;
1381 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1382 	}
1383 	mtx_unlock(&dnp->n_mtx);
1384 
1385 	CURVNET_SET(nmp->nm_sockreq.nr_so->so_vnet);
1386 #ifdef INET
1387 	INIT_VNET_INET(curvnet);
1388 	if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1389 		cverf.lval[0] = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1390 	else
1391 #endif
1392 		cverf.lval[0] = create_verf;
1393 	cverf.lval[1] = ++create_verf;
1394 	CURVNET_RESTORE();
1395 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1396 	    vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1397 	    &nfhp, &attrflag, &dattrflag, NULL);
1398 	if (!error) {
1399 		if (nfhp == NULL)
1400 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1401 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1402 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1403 			    NULL);
1404 		if (nfhp != NULL)
1405 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1406 			    cnp->cn_thread, &np, NULL);
1407 	}
1408 	if (dattrflag)
1409 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1410 	if (!error) {
1411 		newvp = NFSTOV(np);
1412 		if (attrflag)
1413 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1414 			    0, 1);
1415 	}
1416 	if (error) {
1417 		if (newvp != NULL) {
1418 			vrele(newvp);
1419 			newvp = NULL;
1420 		}
1421 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1422 		    error == NFSERR_NOTSUPP) {
1423 			fmode &= ~O_EXCL;
1424 			goto again;
1425 		}
1426 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1427 		if (nfscl_checksattr(vap, &nfsva)) {
1428 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1429 			    cnp->cn_thread, &nfsva, &attrflag, NULL);
1430 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1431 			    vap->va_gid != (gid_t)VNOVAL)) {
1432 				/* try again without setting uid/gid */
1433 				vap->va_uid = (uid_t)VNOVAL;
1434 				vap->va_gid = (uid_t)VNOVAL;
1435 				error = nfsrpc_setattr(newvp, vap, NULL,
1436 				    cnp->cn_cred, cnp->cn_thread, &nfsva,
1437 				    &attrflag, NULL);
1438 			}
1439 			if (attrflag)
1440 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1441 				    NULL, 0, 1);
1442 		}
1443 	}
1444 	if (!error) {
1445 		if (cnp->cn_flags & MAKEENTRY)
1446 			cache_enter(dvp, newvp, cnp);
1447 		*ap->a_vpp = newvp;
1448 	} else if (NFS_ISV4(dvp)) {
1449 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1450 		    vap->va_gid);
1451 	}
1452 	mtx_lock(&dnp->n_mtx);
1453 	dnp->n_flag |= NMODIFIED;
1454 	if (!dattrflag)
1455 		dnp->n_attrstamp = 0;
1456 	mtx_unlock(&dnp->n_mtx);
1457 	return (error);
1458 }
1459 
1460 /*
1461  * nfs file remove call
1462  * To try and make nfs semantics closer to ufs semantics, a file that has
1463  * other processes using the vnode is renamed instead of removed and then
1464  * removed later on the last close.
1465  * - If v_usecount > 1
1466  *	  If a rename is not already in the works
1467  *	     call nfs_sillyrename() to set it up
1468  *     else
1469  *	  do the remove rpc
1470  */
1471 static int
1472 nfs_remove(struct vop_remove_args *ap)
1473 {
1474 	struct vnode *vp = ap->a_vp;
1475 	struct vnode *dvp = ap->a_dvp;
1476 	struct componentname *cnp = ap->a_cnp;
1477 	struct nfsnode *np = VTONFS(vp);
1478 	int error = 0;
1479 	struct vattr vattr;
1480 
1481 #ifndef DIAGNOSTIC
1482 	if ((cnp->cn_flags & HASBUF) == 0)
1483 		panic("nfs_remove: no name");
1484 	if (vrefcnt(vp) < 1)
1485 		panic("nfs_remove: bad v_usecount");
1486 #endif
1487 	if (vp->v_type == VDIR)
1488 		error = EPERM;
1489 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1490 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1491 	    vattr.va_nlink > 1)) {
1492 		/*
1493 		 * Purge the name cache so that the chance of a lookup for
1494 		 * the name succeeding while the remove is in progress is
1495 		 * minimized. Without node locking it can still happen, such
1496 		 * that an I/O op returns ESTALE, but since you get this if
1497 		 * another host removes the file..
1498 		 */
1499 		cache_purge(vp);
1500 		/*
1501 		 * throw away biocache buffers, mainly to avoid
1502 		 * unnecessary delayed writes later.
1503 		 */
1504 		error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1505 		/* Do the rpc */
1506 		if (error != EINTR && error != EIO)
1507 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1508 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1509 		/*
1510 		 * Kludge City: If the first reply to the remove rpc is lost..
1511 		 *   the reply to the retransmitted request will be ENOENT
1512 		 *   since the file was in fact removed
1513 		 *   Therefore, we cheat and return success.
1514 		 */
1515 		if (error == ENOENT)
1516 			error = 0;
1517 	} else if (!np->n_sillyrename)
1518 		error = nfs_sillyrename(dvp, vp, cnp);
1519 	np->n_attrstamp = 0;
1520 	return (error);
1521 }
1522 
1523 /*
1524  * nfs file remove rpc called from nfs_inactive
1525  */
1526 int
1527 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1528 {
1529 	/*
1530 	 * Make sure that the directory vnode is still valid.
1531 	 * XXX we should lock sp->s_dvp here.
1532 	 */
1533 	if (sp->s_dvp->v_type == VBAD)
1534 		return (0);
1535 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1536 	    sp->s_cred, NULL));
1537 }
1538 
1539 /*
1540  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1541  */
1542 static int
1543 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1544     int namelen, struct ucred *cred, struct thread *td)
1545 {
1546 	struct nfsvattr dnfsva;
1547 	struct nfsnode *dnp = VTONFS(dvp);
1548 	int error = 0, dattrflag;
1549 
1550 	mtx_lock(&dnp->n_mtx);
1551 	dnp->n_flag |= NREMOVEINPROG;
1552 	mtx_unlock(&dnp->n_mtx);
1553 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1554 	    &dattrflag, NULL);
1555 	mtx_lock(&dnp->n_mtx);
1556 	if ((dnp->n_flag & NREMOVEWANT)) {
1557 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1558 		mtx_unlock(&dnp->n_mtx);
1559 		wakeup((caddr_t)dnp);
1560 	} else {
1561 		dnp->n_flag &= ~NREMOVEINPROG;
1562 		mtx_unlock(&dnp->n_mtx);
1563 	}
1564 	if (dattrflag)
1565 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1566 	mtx_lock(&dnp->n_mtx);
1567 	dnp->n_flag |= NMODIFIED;
1568 	if (!dattrflag)
1569 		dnp->n_attrstamp = 0;
1570 	mtx_unlock(&dnp->n_mtx);
1571 	if (error && NFS_ISV4(dvp))
1572 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1573 	return (error);
1574 }
1575 
1576 /*
1577  * nfs file rename call
1578  */
1579 static int
1580 nfs_rename(struct vop_rename_args *ap)
1581 {
1582 	struct vnode *fvp = ap->a_fvp;
1583 	struct vnode *tvp = ap->a_tvp;
1584 	struct vnode *fdvp = ap->a_fdvp;
1585 	struct vnode *tdvp = ap->a_tdvp;
1586 	struct componentname *tcnp = ap->a_tcnp;
1587 	struct componentname *fcnp = ap->a_fcnp;
1588 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1589 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1590 	struct nfsv4node *newv4 = NULL;
1591 	int error;
1592 
1593 #ifndef DIAGNOSTIC
1594 	if ((tcnp->cn_flags & HASBUF) == 0 ||
1595 	    (fcnp->cn_flags & HASBUF) == 0)
1596 		panic("nfs_rename: no name");
1597 #endif
1598 	/* Check for cross-device rename */
1599 	if ((fvp->v_mount != tdvp->v_mount) ||
1600 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1601 		error = EXDEV;
1602 		goto out;
1603 	}
1604 
1605 	if (fvp == tvp) {
1606 		ncl_printf("nfs_rename: fvp == tvp (can't happen)\n");
1607 		error = 0;
1608 		goto out;
1609 	}
1610 	if ((error = vn_lock(fvp, LK_EXCLUSIVE)))
1611 		goto out;
1612 
1613 	/*
1614 	 * We have to flush B_DELWRI data prior to renaming
1615 	 * the file.  If we don't, the delayed-write buffers
1616 	 * can be flushed out later after the file has gone stale
1617 	 * under NFSV3.  NFSV2 does not have this problem because
1618 	 * ( as far as I can tell ) it flushes dirty buffers more
1619 	 * often.
1620 	 *
1621 	 * Skip the rename operation if the fsync fails, this can happen
1622 	 * due to the server's volume being full, when we pushed out data
1623 	 * that was written back to our cache earlier. Not checking for
1624 	 * this condition can result in potential (silent) data loss.
1625 	 */
1626 	error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1627 	VOP_UNLOCK(fvp, 0);
1628 	if (!error && tvp)
1629 		error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1630 	if (error)
1631 		goto out;
1632 
1633 	/*
1634 	 * If the tvp exists and is in use, sillyrename it before doing the
1635 	 * rename of the new file over it.
1636 	 * XXX Can't sillyrename a directory.
1637 	 */
1638 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1639 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1640 		vput(tvp);
1641 		tvp = NULL;
1642 	}
1643 
1644 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1645 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1646 	    tcnp->cn_thread);
1647 
1648 	if (!error) {
1649 		/*
1650 		 * For NFSv4, check to see if it is the same name and
1651 		 * replace the name, if it is different.
1652 		 */
1653 		MALLOC(newv4, struct nfsv4node *,
1654 		    sizeof (struct nfsv4node) +
1655 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1656 		    M_NFSV4NODE, M_WAITOK);
1657 		mtx_lock(&tdnp->n_mtx);
1658 		mtx_lock(&fnp->n_mtx);
1659 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1660 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1661 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1662 		      tcnp->cn_namelen) ||
1663 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1664 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1665 			tdnp->n_fhp->nfh_len))) {
1666 #ifdef notdef
1667 { char nnn[100]; int nnnl;
1668 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1669 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1670 nnn[nnnl] = '\0';
1671 printf("ren replace=%s\n",nnn);
1672 }
1673 #endif
1674 			FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1675 			fnp->n_v4 = newv4;
1676 			newv4 = NULL;
1677 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1678 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1679 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1680 			    tdnp->n_fhp->nfh_len);
1681 			NFSBCOPY(tcnp->cn_nameptr,
1682 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1683 		}
1684 		mtx_unlock(&tdnp->n_mtx);
1685 		mtx_unlock(&fnp->n_mtx);
1686 		if (newv4 != NULL)
1687 			FREE((caddr_t)newv4, M_NFSV4NODE);
1688 	}
1689 
1690 	if (fvp->v_type == VDIR) {
1691 		if (tvp != NULL && tvp->v_type == VDIR)
1692 			cache_purge(tdvp);
1693 		cache_purge(fdvp);
1694 	}
1695 
1696 out:
1697 	if (tdvp == tvp)
1698 		vrele(tdvp);
1699 	else
1700 		vput(tdvp);
1701 	if (tvp)
1702 		vput(tvp);
1703 	vrele(fdvp);
1704 	vrele(fvp);
1705 	/*
1706 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1707 	 */
1708 	if (error == ENOENT)
1709 		error = 0;
1710 	return (error);
1711 }
1712 
1713 /*
1714  * nfs file rename rpc called from nfs_remove() above
1715  */
1716 static int
1717 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1718     struct sillyrename *sp)
1719 {
1720 
1721 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1722 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1723 	    scnp->cn_thread));
1724 }
1725 
1726 /*
1727  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1728  */
1729 static int
1730 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1731     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1732     int tnamelen, struct ucred *cred, struct thread *td)
1733 {
1734 	struct nfsvattr fnfsva, tnfsva;
1735 	struct nfsnode *fdnp = VTONFS(fdvp);
1736 	struct nfsnode *tdnp = VTONFS(tdvp);
1737 	int error = 0, fattrflag, tattrflag;
1738 
1739 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1740 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1741 	    &tattrflag, NULL, NULL);
1742 	mtx_lock(&fdnp->n_mtx);
1743 	fdnp->n_flag |= NMODIFIED;
1744 	mtx_unlock(&fdnp->n_mtx);
1745 	mtx_lock(&tdnp->n_mtx);
1746 	tdnp->n_flag |= NMODIFIED;
1747 	mtx_unlock(&tdnp->n_mtx);
1748 	if (fattrflag)
1749 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1750 	else
1751 		fdnp->n_attrstamp = 0;
1752 	if (tattrflag)
1753 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1754 	else
1755 		tdnp->n_attrstamp = 0;
1756 	if (error && NFS_ISV4(fdvp))
1757 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1758 	return (error);
1759 }
1760 
1761 /*
1762  * nfs hard link create call
1763  */
1764 static int
1765 nfs_link(struct vop_link_args *ap)
1766 {
1767 	struct vnode *vp = ap->a_vp;
1768 	struct vnode *tdvp = ap->a_tdvp;
1769 	struct componentname *cnp = ap->a_cnp;
1770 	struct nfsnode *tdnp;
1771 	struct nfsvattr nfsva, dnfsva;
1772 	int error = 0, attrflag, dattrflag;
1773 
1774 	if (vp->v_mount != tdvp->v_mount) {
1775 		return (EXDEV);
1776 	}
1777 
1778 	/*
1779 	 * Push all writes to the server, so that the attribute cache
1780 	 * doesn't get "out of sync" with the server.
1781 	 * XXX There should be a better way!
1782 	 */
1783 	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1784 
1785 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1786 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1787 	    &dattrflag, NULL);
1788 	tdnp = VTONFS(tdvp);
1789 	mtx_lock(&tdnp->n_mtx);
1790 	tdnp->n_flag |= NMODIFIED;
1791 	mtx_unlock(&tdnp->n_mtx);
1792 	if (attrflag)
1793 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1794 	else
1795 		VTONFS(vp)->n_attrstamp = 0;
1796 	if (dattrflag)
1797 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1798 	else
1799 		tdnp->n_attrstamp = 0;
1800 	/*
1801 	 * If negative lookup caching is enabled, I might as well
1802 	 * add an entry for this node. Not necessary for correctness,
1803 	 * but if negative caching is enabled, then the system
1804 	 * must care about lookup caching hit rate, so...
1805 	 */
1806 	if (newnfs_neglookup_enable != 0 &&
1807 	    (cnp->cn_flags & MAKEENTRY))
1808 		cache_enter(tdvp, vp, cnp);
1809 	if (error && NFS_ISV4(vp))
1810 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1811 		    (gid_t)0);
1812 	return (error);
1813 }
1814 
1815 /*
1816  * nfs symbolic link create call
1817  */
1818 static int
1819 nfs_symlink(struct vop_symlink_args *ap)
1820 {
1821 	struct vnode *dvp = ap->a_dvp;
1822 	struct vattr *vap = ap->a_vap;
1823 	struct componentname *cnp = ap->a_cnp;
1824 	struct nfsvattr nfsva, dnfsva;
1825 	struct nfsfh *nfhp;
1826 	struct nfsnode *np = NULL, *dnp;
1827 	struct vnode *newvp = NULL;
1828 	int error = 0, attrflag, dattrflag, ret;
1829 
1830 	vap->va_type = VLNK;
1831 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1832 	    ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1833 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1834 	if (nfhp) {
1835 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
1836 		    &np, NULL);
1837 		if (!ret)
1838 			newvp = NFSTOV(np);
1839 		else if (!error)
1840 			error = ret;
1841 	}
1842 	if (newvp != NULL) {
1843 		if (attrflag)
1844 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1845 			    0, 1);
1846 	} else if (!error) {
1847 		/*
1848 		 * If we do not have an error and we could not extract the
1849 		 * newvp from the response due to the request being NFSv2, we
1850 		 * have to do a lookup in order to obtain a newvp to return.
1851 		 */
1852 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1853 		    cnp->cn_cred, cnp->cn_thread, &np);
1854 		if (!error)
1855 			newvp = NFSTOV(np);
1856 	}
1857 	if (error) {
1858 		if (newvp)
1859 			vput(newvp);
1860 		if (NFS_ISV4(dvp))
1861 			error = nfscl_maperr(cnp->cn_thread, error,
1862 			    vap->va_uid, vap->va_gid);
1863 	} else {
1864 		/*
1865 		 * If negative lookup caching is enabled, I might as well
1866 		 * add an entry for this node. Not necessary for correctness,
1867 		 * but if negative caching is enabled, then the system
1868 		 * must care about lookup caching hit rate, so...
1869 		 */
1870 		if (newnfs_neglookup_enable != 0 &&
1871 		    (cnp->cn_flags & MAKEENTRY))
1872 			cache_enter(dvp, newvp, cnp);
1873 		*ap->a_vpp = newvp;
1874 	}
1875 
1876 	dnp = VTONFS(dvp);
1877 	mtx_lock(&dnp->n_mtx);
1878 	dnp->n_flag |= NMODIFIED;
1879 	mtx_unlock(&dnp->n_mtx);
1880 	if (dattrflag)
1881 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1882 	else
1883 		dnp->n_attrstamp = 0;
1884 	return (error);
1885 }
1886 
1887 /*
1888  * nfs make dir call
1889  */
1890 static int
1891 nfs_mkdir(struct vop_mkdir_args *ap)
1892 {
1893 	struct vnode *dvp = ap->a_dvp;
1894 	struct vattr *vap = ap->a_vap;
1895 	struct componentname *cnp = ap->a_cnp;
1896 	struct nfsnode *np = NULL, *dnp;
1897 	struct vnode *newvp = NULL;
1898 	struct vattr vattr;
1899 	struct nfsfh *nfhp;
1900 	struct nfsvattr nfsva, dnfsva;
1901 	int error = 0, attrflag, dattrflag, ret;
1902 
1903 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1904 		return (error);
1905 	vap->va_type = VDIR;
1906 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1907 	    vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
1908 	    &attrflag, &dattrflag, NULL);
1909 	dnp = VTONFS(dvp);
1910 	mtx_lock(&dnp->n_mtx);
1911 	dnp->n_flag |= NMODIFIED;
1912 	mtx_unlock(&dnp->n_mtx);
1913 	if (dattrflag)
1914 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1915 	else
1916 		dnp->n_attrstamp = 0;
1917 	if (nfhp) {
1918 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
1919 		    &np, NULL);
1920 		if (!ret) {
1921 			newvp = NFSTOV(np);
1922 			if (attrflag)
1923 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1924 				NULL, 0, 1);
1925 		} else if (!error)
1926 			error = ret;
1927 	}
1928 	if (!error && newvp == NULL) {
1929 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1930 		    cnp->cn_cred, cnp->cn_thread, &np);
1931 		if (!error) {
1932 			newvp = NFSTOV(np);
1933 			if (newvp->v_type != VDIR)
1934 				error = EEXIST;
1935 		}
1936 	}
1937 	if (error) {
1938 		if (newvp)
1939 			vput(newvp);
1940 		if (NFS_ISV4(dvp))
1941 			error = nfscl_maperr(cnp->cn_thread, error,
1942 			    vap->va_uid, vap->va_gid);
1943 	} else {
1944 		/*
1945 		 * If negative lookup caching is enabled, I might as well
1946 		 * add an entry for this node. Not necessary for correctness,
1947 		 * but if negative caching is enabled, then the system
1948 		 * must care about lookup caching hit rate, so...
1949 		 */
1950 		if (newnfs_neglookup_enable != 0 &&
1951 		    (cnp->cn_flags & MAKEENTRY))
1952 			cache_enter(dvp, newvp, cnp);
1953 		*ap->a_vpp = newvp;
1954 	}
1955 	return (error);
1956 }
1957 
1958 /*
1959  * nfs remove directory call
1960  */
1961 static int
1962 nfs_rmdir(struct vop_rmdir_args *ap)
1963 {
1964 	struct vnode *vp = ap->a_vp;
1965 	struct vnode *dvp = ap->a_dvp;
1966 	struct componentname *cnp = ap->a_cnp;
1967 	struct nfsnode *dnp;
1968 	struct nfsvattr dnfsva;
1969 	int error, dattrflag;
1970 
1971 	if (dvp == vp)
1972 		return (EINVAL);
1973 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1974 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
1975 	dnp = VTONFS(dvp);
1976 	mtx_lock(&dnp->n_mtx);
1977 	dnp->n_flag |= NMODIFIED;
1978 	mtx_unlock(&dnp->n_mtx);
1979 	if (dattrflag)
1980 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1981 	else
1982 		dnp->n_attrstamp = 0;
1983 
1984 	cache_purge(dvp);
1985 	cache_purge(vp);
1986 	if (error && NFS_ISV4(dvp))
1987 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1988 		    (gid_t)0);
1989 	/*
1990 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
1991 	 */
1992 	if (error == ENOENT)
1993 		error = 0;
1994 	return (error);
1995 }
1996 
1997 /*
1998  * nfs readdir call
1999  */
2000 static int
2001 nfs_readdir(struct vop_readdir_args *ap)
2002 {
2003 	struct vnode *vp = ap->a_vp;
2004 	struct nfsnode *np = VTONFS(vp);
2005 	struct uio *uio = ap->a_uio;
2006 	int tresid, error = 0;
2007 	struct vattr vattr;
2008 
2009 	if (vp->v_type != VDIR)
2010 		return(EPERM);
2011 
2012 	/*
2013 	 * First, check for hit on the EOF offset cache
2014 	 */
2015 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2016 	    (np->n_flag & NMODIFIED) == 0) {
2017 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2018 			mtx_lock(&np->n_mtx);
2019 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2020 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2021 				mtx_unlock(&np->n_mtx);
2022 				NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2023 				return (0);
2024 			} else
2025 				mtx_unlock(&np->n_mtx);
2026 		}
2027 	}
2028 
2029 	/*
2030 	 * Call ncl_bioread() to do the real work.
2031 	 */
2032 	tresid = uio->uio_resid;
2033 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2034 
2035 	if (!error && uio->uio_resid == tresid)
2036 		NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2037 	return (error);
2038 }
2039 
2040 /*
2041  * Readdir rpc call.
2042  * Called from below the buffer cache by ncl_doio().
2043  */
2044 int
2045 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2046     struct thread *td)
2047 {
2048 	struct nfsvattr nfsva;
2049 	nfsuint64 *cookiep, cookie;
2050 	struct nfsnode *dnp = VTONFS(vp);
2051 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2052 	int error = 0, eof, attrflag;
2053 
2054 #ifndef DIAGNOSTIC
2055 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2056 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2057 		panic("nfs readdirrpc bad uio");
2058 #endif
2059 
2060 	/*
2061 	 * If there is no cookie, assume directory was stale.
2062 	 */
2063 	ncl_dircookie_lock(dnp);
2064 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2065 	if (cookiep) {
2066 		cookie = *cookiep;
2067 		ncl_dircookie_unlock(dnp);
2068 	} else {
2069 		ncl_dircookie_unlock(dnp);
2070 		return (NFSERR_BAD_COOKIE);
2071 	}
2072 
2073 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2074 		(void)ncl_fsinfo(nmp, vp, cred, td);
2075 
2076 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2077 	    &attrflag, &eof, NULL);
2078 	if (attrflag)
2079 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2080 
2081 	if (!error) {
2082 		/*
2083 		 * We are now either at the end of the directory or have filled
2084 		 * the block.
2085 		 */
2086 		if (eof)
2087 			dnp->n_direofoffset = uiop->uio_offset;
2088 		else {
2089 			if (uiop->uio_resid > 0)
2090 				ncl_printf("EEK! readdirrpc resid > 0\n");
2091 			ncl_dircookie_lock(dnp);
2092 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2093 			*cookiep = cookie;
2094 			ncl_dircookie_unlock(dnp);
2095 		}
2096 	} else if (NFS_ISV4(vp)) {
2097 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2098 	}
2099 	return (error);
2100 }
2101 
2102 /*
2103  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2104  */
2105 int
2106 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2107     struct thread *td)
2108 {
2109 	struct nfsvattr nfsva;
2110 	nfsuint64 *cookiep, cookie;
2111 	struct nfsnode *dnp = VTONFS(vp);
2112 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2113 	int error = 0, attrflag, eof;
2114 
2115 #ifndef DIAGNOSTIC
2116 	if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2117 		(uiop->uio_resid & (DIRBLKSIZ - 1)))
2118 		panic("nfs readdirplusrpc bad uio");
2119 #endif
2120 
2121 	/*
2122 	 * If there is no cookie, assume directory was stale.
2123 	 */
2124 	ncl_dircookie_lock(dnp);
2125 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2126 	if (cookiep) {
2127 		cookie = *cookiep;
2128 		ncl_dircookie_unlock(dnp);
2129 	} else {
2130 		ncl_dircookie_unlock(dnp);
2131 		return (NFSERR_BAD_COOKIE);
2132 	}
2133 
2134 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2135 		(void)ncl_fsinfo(nmp, vp, cred, td);
2136 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2137 	    &attrflag, &eof, NULL);
2138 	if (attrflag)
2139 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2140 
2141 	if (!error) {
2142 		/*
2143 		 * We are now either at end of the directory or have filled the
2144 		 * the block.
2145 		 */
2146 		if (eof)
2147 			dnp->n_direofoffset = uiop->uio_offset;
2148 		else {
2149 			if (uiop->uio_resid > 0)
2150 				ncl_printf("EEK! readdirplusrpc resid > 0\n");
2151 			ncl_dircookie_lock(dnp);
2152 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2153 			*cookiep = cookie;
2154 			ncl_dircookie_unlock(dnp);
2155 		}
2156 	} else if (NFS_ISV4(vp)) {
2157 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2158 	}
2159 	return (error);
2160 }
2161 
2162 /*
2163  * Silly rename. To make the NFS filesystem that is stateless look a little
2164  * more like the "ufs" a remove of an active vnode is translated to a rename
2165  * to a funny looking filename that is removed by nfs_inactive on the
2166  * nfsnode. There is the potential for another process on a different client
2167  * to create the same funny name between the nfs_lookitup() fails and the
2168  * nfs_rename() completes, but...
2169  */
2170 static int
2171 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2172 {
2173 	struct sillyrename *sp;
2174 	struct nfsnode *np;
2175 	int error;
2176 	short pid;
2177 	unsigned int lticks;
2178 
2179 	cache_purge(dvp);
2180 	np = VTONFS(vp);
2181 #ifndef DIAGNOSTIC
2182 	if (vp->v_type == VDIR)
2183 		panic("nfs: sillyrename dir");
2184 #endif
2185 	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2186 	    M_NEWNFSREQ, M_WAITOK);
2187 	sp->s_cred = crhold(cnp->cn_cred);
2188 	sp->s_dvp = dvp;
2189 	VREF(dvp);
2190 
2191 	/*
2192 	 * Fudge together a funny name.
2193 	 * Changing the format of the funny name to accomodate more
2194 	 * sillynames per directory.
2195 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2196 	 * CPU ticks since boot.
2197 	 */
2198 	pid = cnp->cn_thread->td_proc->p_pid;
2199 	lticks = (unsigned int)ticks;
2200 	for ( ; ; ) {
2201 		sp->s_namlen = sprintf(sp->s_name,
2202 				       ".nfs.%08x.%04x4.4", lticks,
2203 				       pid);
2204 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2205 				 cnp->cn_thread, NULL))
2206 			break;
2207 		lticks++;
2208 	}
2209 	error = nfs_renameit(dvp, vp, cnp, sp);
2210 	if (error)
2211 		goto bad;
2212 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2213 		cnp->cn_thread, &np);
2214 	np->n_sillyrename = sp;
2215 	return (0);
2216 bad:
2217 	vrele(sp->s_dvp);
2218 	crfree(sp->s_cred);
2219 	free((caddr_t)sp, M_NEWNFSREQ);
2220 	return (error);
2221 }
2222 
2223 /*
2224  * Look up a file name and optionally either update the file handle or
2225  * allocate an nfsnode, depending on the value of npp.
2226  * npp == NULL	--> just do the lookup
2227  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2228  *			handled too
2229  * *npp != NULL --> update the file handle in the vnode
2230  */
2231 static int
2232 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2233     struct thread *td, struct nfsnode **npp)
2234 {
2235 	struct vnode *newvp = NULL, *vp;
2236 	struct nfsnode *np, *dnp = VTONFS(dvp);
2237 	struct nfsfh *nfhp, *onfhp;
2238 	struct nfsvattr nfsva, dnfsva;
2239 	struct componentname cn;
2240 	int error = 0, attrflag, dattrflag;
2241 	u_int hash;
2242 
2243 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2244 	    &nfhp, &attrflag, &dattrflag, NULL);
2245 	if (dattrflag)
2246 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2247 	if (npp && !error) {
2248 		if (*npp != NULL) {
2249 		    np = *npp;
2250 		    vp = NFSTOV(np);
2251 		    /*
2252 		     * For NFSv4, check to see if it is the same name and
2253 		     * replace the name, if it is different.
2254 		     */
2255 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2256 			(np->n_v4->n4_namelen != len ||
2257 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2258 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2259 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2260 			 dnp->n_fhp->nfh_len))) {
2261 #ifdef notdef
2262 { char nnn[100]; int nnnl;
2263 nnnl = (len < 100) ? len : 99;
2264 bcopy(name, nnn, nnnl);
2265 nnn[nnnl] = '\0';
2266 printf("replace=%s\n",nnn);
2267 }
2268 #endif
2269 			    FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2270 			    MALLOC(np->n_v4, struct nfsv4node *,
2271 				sizeof (struct nfsv4node) +
2272 				dnp->n_fhp->nfh_len + len - 1,
2273 				M_NFSV4NODE, M_WAITOK);
2274 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2275 			    np->n_v4->n4_namelen = len;
2276 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2277 				dnp->n_fhp->nfh_len);
2278 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2279 		    }
2280 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2281 			FNV1_32_INIT);
2282 		    onfhp = np->n_fhp;
2283 		    /*
2284 		     * Rehash node for new file handle.
2285 		     */
2286 		    vfs_hash_rehash(vp, hash);
2287 		    np->n_fhp = nfhp;
2288 		    if (onfhp != NULL)
2289 			FREE((caddr_t)onfhp, M_NFSFH);
2290 		    newvp = NFSTOV(np);
2291 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2292 		    FREE((caddr_t)nfhp, M_NFSFH);
2293 		    VREF(dvp);
2294 		    newvp = dvp;
2295 		} else {
2296 		    cn.cn_nameptr = name;
2297 		    cn.cn_namelen = len;
2298 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2299 			&np, NULL);
2300 		    if (error)
2301 			return (error);
2302 		    newvp = NFSTOV(np);
2303 		}
2304 		if (!attrflag && *npp == NULL) {
2305 			vrele(newvp);
2306 			return (ENOENT);
2307 		}
2308 		if (attrflag)
2309 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2310 			    0, 1);
2311 	}
2312 	if (npp && *npp == NULL) {
2313 		if (error) {
2314 			if (newvp) {
2315 				if (newvp == dvp)
2316 					vrele(newvp);
2317 				else
2318 					vput(newvp);
2319 			}
2320 		} else
2321 			*npp = np;
2322 	}
2323 	if (error && NFS_ISV4(dvp))
2324 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2325 	return (error);
2326 }
2327 
2328 /*
2329  * Nfs Version 3 and 4 commit rpc
2330  */
2331 int
2332 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2333    struct thread *td)
2334 {
2335 	struct nfsvattr nfsva;
2336 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2337 	int error, attrflag;
2338 	u_char verf[NFSX_VERF];
2339 
2340 	mtx_lock(&nmp->nm_mtx);
2341 	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2342 		mtx_unlock(&nmp->nm_mtx);
2343 		return (0);
2344 	}
2345 	mtx_unlock(&nmp->nm_mtx);
2346 	error = nfsrpc_commit(vp, offset, cnt, cred, td, verf, &nfsva,
2347 	    &attrflag, NULL);
2348 	if (!error) {
2349 		if (NFSBCMP((caddr_t)nmp->nm_verf, verf, NFSX_VERF)) {
2350 			NFSBCOPY(verf, (caddr_t)nmp->nm_verf, NFSX_VERF);
2351 			error = NFSERR_STALEWRITEVERF;
2352 		}
2353 		if (!error && attrflag)
2354 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2355 			    0, 1);
2356 	} else if (NFS_ISV4(vp)) {
2357 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2358 	}
2359 	return (error);
2360 }
2361 
2362 /*
2363  * Strategy routine.
2364  * For async requests when nfsiod(s) are running, queue the request by
2365  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2366  * request.
2367  */
2368 static int
2369 nfs_strategy(struct vop_strategy_args *ap)
2370 {
2371 	struct buf *bp = ap->a_bp;
2372 	struct ucred *cr;
2373 
2374 	KASSERT(!(bp->b_flags & B_DONE),
2375 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2376 	BUF_ASSERT_HELD(bp);
2377 
2378 	if (bp->b_iocmd == BIO_READ)
2379 		cr = bp->b_rcred;
2380 	else
2381 		cr = bp->b_wcred;
2382 
2383 	/*
2384 	 * If the op is asynchronous and an i/o daemon is waiting
2385 	 * queue the request, wake it up and wait for completion
2386 	 * otherwise just do it ourselves.
2387 	 */
2388 	if ((bp->b_flags & B_ASYNC) == 0 ||
2389 	    ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2390 		(void)ncl_doio(ap->a_vp, bp, cr, curthread);
2391 	return (0);
2392 }
2393 
2394 /*
2395  * fsync vnode op. Just call ncl_flush() with commit == 1.
2396  */
2397 /* ARGSUSED */
2398 static int
2399 nfs_fsync(struct vop_fsync_args *ap)
2400 {
2401 	return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1));
2402 }
2403 
2404 /*
2405  * Flush all the blocks associated with a vnode.
2406  * 	Walk through the buffer pool and push any dirty pages
2407  *	associated with the vnode.
2408  */
2409 int
2410 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2411     int commit)
2412 {
2413 	struct nfsnode *np = VTONFS(vp);
2414 	struct buf *bp;
2415 	int i;
2416 	struct buf *nbp;
2417 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2418 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2419 	int passone = 1, trycnt = 0;
2420 	u_quad_t off, endoff, toff;
2421 	struct ucred* wcred = NULL;
2422 	struct buf **bvec = NULL;
2423 	struct bufobj *bo;
2424 #ifndef NFS_COMMITBVECSIZ
2425 #define	NFS_COMMITBVECSIZ	20
2426 #endif
2427 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2428 	int bvecsize = 0, bveccount;
2429 
2430 	if (nmp->nm_flag & NFSMNT_INT)
2431 		slpflag = PCATCH;
2432 	if (!commit)
2433 		passone = 0;
2434 	bo = &vp->v_bufobj;
2435 	/*
2436 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2437 	 * server, but has not been committed to stable storage on the server
2438 	 * yet. On the first pass, the byte range is worked out and the commit
2439 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2440 	 * job.
2441 	 */
2442 again:
2443 	off = (u_quad_t)-1;
2444 	endoff = 0;
2445 	bvecpos = 0;
2446 	if (NFS_ISV34(vp) && commit) {
2447 		if (bvec != NULL && bvec != bvec_on_stack)
2448 			free(bvec, M_TEMP);
2449 		/*
2450 		 * Count up how many buffers waiting for a commit.
2451 		 */
2452 		bveccount = 0;
2453 		BO_LOCK(bo);
2454 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2455 			if (!BUF_ISLOCKED(bp) &&
2456 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2457 				== (B_DELWRI | B_NEEDCOMMIT))
2458 				bveccount++;
2459 		}
2460 		/*
2461 		 * Allocate space to remember the list of bufs to commit.  It is
2462 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2463 		 * If we can't get memory (for whatever reason), we will end up
2464 		 * committing the buffers one-by-one in the loop below.
2465 		 */
2466 		if (bveccount > NFS_COMMITBVECSIZ) {
2467 			/*
2468 			 * Release the vnode interlock to avoid a lock
2469 			 * order reversal.
2470 			 */
2471 			BO_UNLOCK(bo);
2472 			bvec = (struct buf **)
2473 				malloc(bveccount * sizeof(struct buf *),
2474 				       M_TEMP, M_NOWAIT);
2475 			BO_LOCK(bo);
2476 			if (bvec == NULL) {
2477 				bvec = bvec_on_stack;
2478 				bvecsize = NFS_COMMITBVECSIZ;
2479 			} else
2480 				bvecsize = bveccount;
2481 		} else {
2482 			bvec = bvec_on_stack;
2483 			bvecsize = NFS_COMMITBVECSIZ;
2484 		}
2485 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2486 			if (bvecpos >= bvecsize)
2487 				break;
2488 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2489 				nbp = TAILQ_NEXT(bp, b_bobufs);
2490 				continue;
2491 			}
2492 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2493 			    (B_DELWRI | B_NEEDCOMMIT)) {
2494 				BUF_UNLOCK(bp);
2495 				nbp = TAILQ_NEXT(bp, b_bobufs);
2496 				continue;
2497 			}
2498 			BO_UNLOCK(bo);
2499 			bremfree(bp);
2500 			/*
2501 			 * Work out if all buffers are using the same cred
2502 			 * so we can deal with them all with one commit.
2503 			 *
2504 			 * NOTE: we are not clearing B_DONE here, so we have
2505 			 * to do it later on in this routine if we intend to
2506 			 * initiate I/O on the bp.
2507 			 *
2508 			 * Note: to avoid loopback deadlocks, we do not
2509 			 * assign b_runningbufspace.
2510 			 */
2511 			if (wcred == NULL)
2512 				wcred = bp->b_wcred;
2513 			else if (wcred != bp->b_wcred)
2514 				wcred = NOCRED;
2515 			vfs_busy_pages(bp, 1);
2516 
2517 			BO_LOCK(bo);
2518 			/*
2519 			 * bp is protected by being locked, but nbp is not
2520 			 * and vfs_busy_pages() may sleep.  We have to
2521 			 * recalculate nbp.
2522 			 */
2523 			nbp = TAILQ_NEXT(bp, b_bobufs);
2524 
2525 			/*
2526 			 * A list of these buffers is kept so that the
2527 			 * second loop knows which buffers have actually
2528 			 * been committed. This is necessary, since there
2529 			 * may be a race between the commit rpc and new
2530 			 * uncommitted writes on the file.
2531 			 */
2532 			bvec[bvecpos++] = bp;
2533 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2534 				bp->b_dirtyoff;
2535 			if (toff < off)
2536 				off = toff;
2537 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2538 			if (toff > endoff)
2539 				endoff = toff;
2540 		}
2541 		BO_UNLOCK(bo);
2542 	}
2543 	if (bvecpos > 0) {
2544 		/*
2545 		 * Commit data on the server, as required.
2546 		 * If all bufs are using the same wcred, then use that with
2547 		 * one call for all of them, otherwise commit each one
2548 		 * separately.
2549 		 */
2550 		if (wcred != NOCRED)
2551 			retv = ncl_commit(vp, off, (int)(endoff - off),
2552 					  wcred, td);
2553 		else {
2554 			retv = 0;
2555 			for (i = 0; i < bvecpos; i++) {
2556 				off_t off, size;
2557 				bp = bvec[i];
2558 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2559 					bp->b_dirtyoff;
2560 				size = (u_quad_t)(bp->b_dirtyend
2561 						  - bp->b_dirtyoff);
2562 				retv = ncl_commit(vp, off, (int)size,
2563 						  bp->b_wcred, td);
2564 				if (retv) break;
2565 			}
2566 		}
2567 
2568 		if (retv == NFSERR_STALEWRITEVERF)
2569 			ncl_clearcommit(vp->v_mount);
2570 
2571 		/*
2572 		 * Now, either mark the blocks I/O done or mark the
2573 		 * blocks dirty, depending on whether the commit
2574 		 * succeeded.
2575 		 */
2576 		for (i = 0; i < bvecpos; i++) {
2577 			bp = bvec[i];
2578 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2579 			if (retv) {
2580 				/*
2581 				 * Error, leave B_DELWRI intact
2582 				 */
2583 				vfs_unbusy_pages(bp);
2584 				brelse(bp);
2585 			} else {
2586 				/*
2587 				 * Success, remove B_DELWRI ( bundirty() ).
2588 				 *
2589 				 * b_dirtyoff/b_dirtyend seem to be NFS
2590 				 * specific.  We should probably move that
2591 				 * into bundirty(). XXX
2592 				 */
2593 				bufobj_wref(bo);
2594 				bp->b_flags |= B_ASYNC;
2595 				bundirty(bp);
2596 				bp->b_flags &= ~B_DONE;
2597 				bp->b_ioflags &= ~BIO_ERROR;
2598 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2599 				bufdone(bp);
2600 			}
2601 		}
2602 	}
2603 
2604 	/*
2605 	 * Start/do any write(s) that are required.
2606 	 */
2607 loop:
2608 	BO_LOCK(bo);
2609 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2610 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2611 			if (waitfor != MNT_WAIT || passone)
2612 				continue;
2613 
2614 			error = BUF_TIMELOCK(bp,
2615 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2616 			    BO_MTX(bo), "nfsfsync", slpflag, slptimeo);
2617 			if (error == 0) {
2618 				BUF_UNLOCK(bp);
2619 				goto loop;
2620 			}
2621 			if (error == ENOLCK) {
2622 				error = 0;
2623 				goto loop;
2624 			}
2625 			if (newnfs_sigintr(nmp, td)) {
2626 				error = EINTR;
2627 				goto done;
2628 			}
2629 			if (slpflag == PCATCH) {
2630 				slpflag = 0;
2631 				slptimeo = 2 * hz;
2632 			}
2633 			goto loop;
2634 		}
2635 		if ((bp->b_flags & B_DELWRI) == 0)
2636 			panic("nfs_fsync: not dirty");
2637 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2638 			BUF_UNLOCK(bp);
2639 			continue;
2640 		}
2641 		BO_UNLOCK(bo);
2642 		bremfree(bp);
2643 		if (passone || !commit)
2644 		    bp->b_flags |= B_ASYNC;
2645 		else
2646 		    bp->b_flags |= B_ASYNC;
2647 		bwrite(bp);
2648 		if (newnfs_sigintr(nmp, td)) {
2649 			error = EINTR;
2650 			goto done;
2651 		}
2652 		goto loop;
2653 	}
2654 	if (passone) {
2655 		passone = 0;
2656 		BO_UNLOCK(bo);
2657 		goto again;
2658 	}
2659 	if (waitfor == MNT_WAIT) {
2660 		while (bo->bo_numoutput) {
2661 			error = bufobj_wwait(bo, slpflag, slptimeo);
2662 			if (error) {
2663 			    BO_UNLOCK(bo);
2664 			    error = newnfs_sigintr(nmp, td);
2665 			    if (error)
2666 				goto done;
2667 			    if (slpflag == PCATCH) {
2668 				slpflag = 0;
2669 				slptimeo = 2 * hz;
2670 			    }
2671 			    BO_LOCK(bo);
2672 			}
2673 		}
2674 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
2675 			BO_UNLOCK(bo);
2676 			goto loop;
2677 		}
2678 		/*
2679 		 * Wait for all the async IO requests to drain
2680 		 */
2681 		BO_UNLOCK(bo);
2682 		mtx_lock(&np->n_mtx);
2683 		while (np->n_directio_asyncwr > 0) {
2684 			np->n_flag |= NFSYNCWAIT;
2685 			error = ncl_msleep(td, (caddr_t)&np->n_directio_asyncwr,
2686 					   &np->n_mtx, slpflag | (PRIBIO + 1),
2687 					   "nfsfsync", 0);
2688 			if (error) {
2689 				if (newnfs_sigintr(nmp, td)) {
2690 					mtx_unlock(&np->n_mtx);
2691 					error = EINTR;
2692 					goto done;
2693 				}
2694 			}
2695 		}
2696 		mtx_unlock(&np->n_mtx);
2697 	} else
2698 		BO_UNLOCK(bo);
2699 	mtx_lock(&np->n_mtx);
2700 	if (np->n_flag & NWRITEERR) {
2701 		error = np->n_error;
2702 		np->n_flag &= ~NWRITEERR;
2703 	}
2704   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
2705 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2706   		np->n_flag &= ~NMODIFIED;
2707 	mtx_unlock(&np->n_mtx);
2708 done:
2709 	if (bvec != NULL && bvec != bvec_on_stack)
2710 		free(bvec, M_TEMP);
2711 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2712 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2713 	     np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2714 		/* try, try again... */
2715 		passone = 1;
2716 		wcred = NULL;
2717 		bvec = NULL;
2718 		bvecsize = 0;
2719 printf("try%d\n", trycnt);
2720 		goto again;
2721 	}
2722 	return (error);
2723 }
2724 
2725 /*
2726  * NFS advisory byte-level locks.
2727  */
2728 static int
2729 nfs_advlock(struct vop_advlock_args *ap)
2730 {
2731 	struct vnode *vp = ap->a_vp;
2732 	struct ucred *cred;
2733 	struct nfsnode *np = VTONFS(ap->a_vp);
2734 	struct proc *p = (struct proc *)ap->a_id;
2735 	struct thread *td = curthread;	/* XXX */
2736 	struct vattr va;
2737 	int ret, error = EOPNOTSUPP, vlret;
2738 	u_quad_t size;
2739 
2740 	if (NFS_ISV4(vp) && (ap->a_flags & F_POSIX)) {
2741 		cred = p->p_ucred;
2742 		vlret = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2743 		if (vlret)
2744 			return (vlret);
2745 
2746 		/*
2747 		 * If this is unlocking a write locked region, flush and
2748 		 * commit them before unlocking. This is required by
2749 		 * RFC3530 Sec. 9.3.2.
2750 		 */
2751 		if (ap->a_op == F_UNLCK &&
2752 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td))
2753 			(void) ncl_flush(vp, MNT_WAIT, cred, td, 1);
2754 
2755 		/*
2756 		 * Loop around doing the lock op, while a blocking lock
2757 		 * must wait for the lock op to succeed.
2758 		 */
2759 		do {
2760 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
2761 			    ap->a_fl, 0, cred, td);
2762 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2763 			    ap->a_op == F_SETLK) {
2764 				VOP_UNLOCK(vp, 0);
2765 				error = nfs_catnap(PZERO | PCATCH, "ncladvl");
2766 				if (error)
2767 					return (EINTR);
2768 				vlret = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2769 				if (vlret)
2770 					return (vlret);
2771 			}
2772 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
2773 		     ap->a_op == F_SETLK);
2774 		if (ret == NFSERR_DENIED) {
2775 			VOP_UNLOCK(vp, 0);
2776 			return (EAGAIN);
2777 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
2778 			VOP_UNLOCK(vp, 0);
2779 			return (ret);
2780 		} else if (ret != 0) {
2781 			VOP_UNLOCK(vp, 0);
2782 			return (EACCES);
2783 		}
2784 
2785 		/*
2786 		 * Now, if we just got a lock, invalidate data in the buffer
2787 		 * cache, as required, so that the coherency conforms with
2788 		 * RFC3530 Sec. 9.3.2.
2789 		 */
2790 		if (ap->a_op == F_SETLK) {
2791 			if ((np->n_flag & NMODIFIED) == 0) {
2792 				np->n_attrstamp = 0;
2793 				ret = VOP_GETATTR(vp, &va, cred);
2794 			}
2795 			if ((np->n_flag & NMODIFIED) || ret ||
2796 			    np->n_change != va.va_filerev) {
2797 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
2798 				np->n_attrstamp = 0;
2799 				ret = VOP_GETATTR(vp, &va, cred);
2800 				if (!ret) {
2801 					np->n_mtime = va.va_mtime;
2802 					np->n_change = va.va_filerev;
2803 				}
2804 			}
2805 		}
2806 		VOP_UNLOCK(vp, 0);
2807 		return (0);
2808 	} else if (!NFS_ISV4(vp)) {
2809 		error = vn_lock(vp, LK_SHARED);
2810 		if (error)
2811 			return (error);
2812 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2813 			size = VTONFS(vp)->n_size;
2814 			VOP_UNLOCK(vp, 0);
2815 			error = lf_advlock(ap, &(vp->v_lockf), size);
2816 		} else {
2817 			if (ncl_advlock_p)
2818 				error = ncl_advlock_p(ap);
2819 			else
2820 				error = ENOLCK;
2821 		}
2822 	}
2823 	return (error);
2824 }
2825 
2826 /*
2827  * NFS advisory byte-level locks.
2828  */
2829 static int
2830 nfs_advlockasync(struct vop_advlockasync_args *ap)
2831 {
2832 	struct vnode *vp = ap->a_vp;
2833 	u_quad_t size;
2834 	int error;
2835 
2836 	if (NFS_ISV4(vp))
2837 		return (EOPNOTSUPP);
2838 	error = vn_lock(vp, LK_SHARED);
2839 	if (error)
2840 		return (error);
2841 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
2842 		size = VTONFS(vp)->n_size;
2843 		VOP_UNLOCK(vp, 0);
2844 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
2845 	} else {
2846 		VOP_UNLOCK(vp, 0);
2847 		error = EOPNOTSUPP;
2848 	}
2849 	return (error);
2850 }
2851 
2852 /*
2853  * Print out the contents of an nfsnode.
2854  */
2855 static int
2856 nfs_print(struct vop_print_args *ap)
2857 {
2858 	struct vnode *vp = ap->a_vp;
2859 	struct nfsnode *np = VTONFS(vp);
2860 
2861 	ncl_printf("\tfileid %ld fsid 0x%x",
2862 	   np->n_vattr.na_fileid, np->n_vattr.na_fsid);
2863 	if (vp->v_type == VFIFO)
2864 		fifo_printinfo(vp);
2865 	printf("\n");
2866 	return (0);
2867 }
2868 
2869 /*
2870  * This is the "real" nfs::bwrite(struct buf*).
2871  * We set B_CACHE if this is a VMIO buffer.
2872  */
2873 int
2874 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
2875 {
2876 	int s;
2877 	int oldflags = bp->b_flags;
2878 #if 0
2879 	int retv = 1;
2880 	off_t off;
2881 #endif
2882 
2883 	BUF_ASSERT_HELD(bp);
2884 
2885 	if (bp->b_flags & B_INVAL) {
2886 		brelse(bp);
2887 		return(0);
2888 	}
2889 
2890 	bp->b_flags |= B_CACHE;
2891 
2892 	/*
2893 	 * Undirty the bp.  We will redirty it later if the I/O fails.
2894 	 */
2895 
2896 	s = splbio();
2897 	bundirty(bp);
2898 	bp->b_flags &= ~B_DONE;
2899 	bp->b_ioflags &= ~BIO_ERROR;
2900 	bp->b_iocmd = BIO_WRITE;
2901 
2902 	bufobj_wref(bp->b_bufobj);
2903 	curthread->td_ru.ru_oublock++;
2904 	splx(s);
2905 
2906 	/*
2907 	 * Note: to avoid loopback deadlocks, we do not
2908 	 * assign b_runningbufspace.
2909 	 */
2910 	vfs_busy_pages(bp, 1);
2911 
2912 	BUF_KERNPROC(bp);
2913 	bp->b_iooffset = dbtob(bp->b_blkno);
2914 	bstrategy(bp);
2915 
2916 	if( (oldflags & B_ASYNC) == 0) {
2917 		int rtval = bufwait(bp);
2918 
2919 		if (oldflags & B_DELWRI) {
2920 			s = splbio();
2921 			reassignbuf(bp);
2922 			splx(s);
2923 		}
2924 		brelse(bp);
2925 		return (rtval);
2926 	}
2927 
2928 	return (0);
2929 }
2930 
2931 /*
2932  * nfs special file access vnode op.
2933  * Essentially just get vattr and then imitate iaccess() since the device is
2934  * local to the client.
2935  */
2936 static int
2937 nfsspec_access(struct vop_access_args *ap)
2938 {
2939 	struct vattr *vap;
2940 	struct ucred *cred = ap->a_cred;
2941 	struct vnode *vp = ap->a_vp;
2942 	accmode_t accmode = ap->a_accmode;
2943 	struct vattr vattr;
2944 	int error;
2945 
2946 	/*
2947 	 * Disallow write attempts on filesystems mounted read-only;
2948 	 * unless the file is a socket, fifo, or a block or character
2949 	 * device resident on the filesystem.
2950 	 */
2951 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
2952 		switch (vp->v_type) {
2953 		case VREG:
2954 		case VDIR:
2955 		case VLNK:
2956 			return (EROFS);
2957 		default:
2958 			break;
2959 		}
2960 	}
2961 	vap = &vattr;
2962 	error = VOP_GETATTR(vp, vap, cred);
2963 	if (error)
2964 		goto out;
2965 	error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
2966 	    accmode, cred, NULL);
2967 out:
2968 	return error;
2969 }
2970 
2971 /*
2972  * Read wrapper for fifos.
2973  */
2974 static int
2975 nfsfifo_read(struct vop_read_args *ap)
2976 {
2977 	struct nfsnode *np = VTONFS(ap->a_vp);
2978 	int error;
2979 
2980 	/*
2981 	 * Set access flag.
2982 	 */
2983 	mtx_lock(&np->n_mtx);
2984 	np->n_flag |= NACC;
2985 	getnanotime(&np->n_atim);
2986 	mtx_unlock(&np->n_mtx);
2987 	error = fifo_specops.vop_read(ap);
2988 	return error;
2989 }
2990 
2991 /*
2992  * Write wrapper for fifos.
2993  */
2994 static int
2995 nfsfifo_write(struct vop_write_args *ap)
2996 {
2997 	struct nfsnode *np = VTONFS(ap->a_vp);
2998 
2999 	/*
3000 	 * Set update flag.
3001 	 */
3002 	mtx_lock(&np->n_mtx);
3003 	np->n_flag |= NUPD;
3004 	getnanotime(&np->n_mtim);
3005 	mtx_unlock(&np->n_mtx);
3006 	return(fifo_specops.vop_write(ap));
3007 }
3008 
3009 /*
3010  * Close wrapper for fifos.
3011  *
3012  * Update the times on the nfsnode then do fifo close.
3013  */
3014 static int
3015 nfsfifo_close(struct vop_close_args *ap)
3016 {
3017 	struct vnode *vp = ap->a_vp;
3018 	struct nfsnode *np = VTONFS(vp);
3019 	struct vattr vattr;
3020 	struct timespec ts;
3021 
3022 	mtx_lock(&np->n_mtx);
3023 	if (np->n_flag & (NACC | NUPD)) {
3024 		getnanotime(&ts);
3025 		if (np->n_flag & NACC)
3026 			np->n_atim = ts;
3027 		if (np->n_flag & NUPD)
3028 			np->n_mtim = ts;
3029 		np->n_flag |= NCHG;
3030 		if (vrefcnt(vp) == 1 &&
3031 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3032 			VATTR_NULL(&vattr);
3033 			if (np->n_flag & NACC)
3034 				vattr.va_atime = np->n_atim;
3035 			if (np->n_flag & NUPD)
3036 				vattr.va_mtime = np->n_mtim;
3037 			mtx_unlock(&np->n_mtx);
3038 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3039 			goto out;
3040 		}
3041 	}
3042 	mtx_unlock(&np->n_mtx);
3043 out:
3044 	return (fifo_specops.vop_close(ap));
3045 }
3046 
3047 /*
3048  * Just call ncl_writebp() with the force argument set to 1.
3049  *
3050  * NOTE: B_DONE may or may not be set in a_bp on call.
3051  */
3052 static int
3053 nfs_bwrite(struct buf *bp)
3054 {
3055 
3056 	return (ncl_writebp(bp, 1, curthread));
3057 }
3058 
3059 struct buf_ops buf_ops_newnfs = {
3060 	.bop_name	=	"buf_ops_nfs",
3061 	.bop_write	=	nfs_bwrite,
3062 	.bop_strategy	=	bufstrategy,
3063 	.bop_sync	=	bufsync,
3064 	.bop_bdflush	=	bufbdflush,
3065 };
3066 
3067 /*
3068  * Cloned from vop_stdlock(), and then the ugly hack added.
3069  */
3070 static int
3071 nfs_lock1(struct vop_lock1_args *ap)
3072 {
3073 	struct vnode *vp = ap->a_vp;
3074 	int error = 0;
3075 
3076 	/*
3077 	 * Since vfs_hash_get() calls vget() and it will no longer work
3078 	 * for FreeBSD8 with flags == 0, I can only think of this horrible
3079 	 * hack to work around it. I call vfs_hash_get() with LK_EXCLOTHER
3080 	 * and then handle it here. All I want for this case is a v_usecount
3081 	 * on the vnode to use for recovery, while another thread might
3082 	 * hold a lock on the vnode. I have the other threads blocked, so
3083 	 * there isn't any race problem.
3084 	 */
3085 	if ((ap->a_flags & LK_TYPE_MASK) == LK_EXCLOTHER) {
3086 		if ((ap->a_flags & LK_INTERLOCK) == 0)
3087 			panic("ncllock1");
3088 		if ((vp->v_iflag & VI_DOOMED))
3089 			error = ENOENT;
3090 		VI_UNLOCK(vp);
3091 		return (error);
3092 	}
3093 	return (_lockmgr_args(vp->v_vnlock, ap->a_flags, VI_MTX(vp),
3094 	    LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file,
3095 	    ap->a_line));
3096 }
3097 
3098 #ifdef NFS4_ACL_EXTATTR_NAME
3099 static int
3100 nfs_getacl(struct vop_getacl_args *ap)
3101 {
3102 	int error;
3103 
3104 	if (ap->a_type != ACL_TYPE_NFS4)
3105 		return (EOPNOTSUPP);
3106 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3107 	    NULL);
3108 	if (error > NFSERR_STALE) {
3109 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3110 		error = EPERM;
3111 	}
3112 	return (error);
3113 }
3114 
3115 static int
3116 nfs_setacl(struct vop_setacl_args *ap)
3117 {
3118 	int error;
3119 
3120 	if (ap->a_type != ACL_TYPE_NFS4)
3121 		return (EOPNOTSUPP);
3122 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3123 	    NULL);
3124 	if (error > NFSERR_STALE) {
3125 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3126 		error = EPERM;
3127 	}
3128 	return (error);
3129 }
3130 
3131 #endif	/* NFS4_ACL_EXTATTR_NAME */
3132